UbiCC Journal - Volume 4 Number 3 - Ubiquitous Computing and Communication Journal

UUBBIICCCC JJoouurrnnaall Ubiquitous Computing and Communication Journal

2009 Volume 4 . 08-15-2009 . ISSN 1992-8424

UBICC Publishers © 2009

Ubiquitous Computing and Communication Journal

Co-Editor Dr. Thanos Vasilakos

Ubiquitous Computing and

Communication Journal

Book: 2009 Volume 4

Publishing Date: 08-15-2009

Proceedings

ISSN 1992-8424

This work is subjected to copyright. All rights are reserved whether the whole or part of the material is

concerned, specifically the rights of translation, reprinting, re-use of illusions, recitation, broadcasting,

reproduction on microfilms or in any other way, and storage in data banks. Duplication of this publication of

parts thereof is permitted only under the provision of the copyright law 1965, in its current version, and

permission of use must always be obtained from UBICC Publishers. Violations are liable to prosecution under

the copy right law.

UBICC Journal is a part of UBICC Publishers

www.ubicc.org

© UBICC Journal

Printed in South Korea

Typesetting: Camera-ready by author, data conversation by UBICC Publishing Services, South Korea

UBICC Publishers

UUUBBBIIICCCCCC JJJooouuurrrnnnaaalll

Volume 4, Number 3, August 2009

746 Variable step size algorithms for network echo cancellation

O.O. Oyerinde, S.H. Mneney

758 A framework for automatic reconfigurations of protocol stacks in

ubiquitous computing systems

Mahdi Niamanesh, Rasool Jalili

771 A novel strategy to provide secure channel over wireless to wire

communication

Alaa Hussain Al- Hamami, Mohammad Alaa Al- Hamami

775 Securing route discovery in MAODV for wireless sensor networks

R.Shyamala, S.Valli

784 A solution for backward-compatible reconfiguration of running protocol

components in protocol stacks

Mahdi Niamanesh, Rasool Jalili

794 A modified image water marking using scalar quantization in wavelet

domain

Mohiy Mohammed Hadhoud, Abdalhameed Shaalan, Hanaa Abdalaziz Abdallah

801 Managing unstructured data using agent technology

Amit Kumar Goel, Ritu Sindhu , Monica Mehrotra ,G.N. Purohit

807 Multi-layer fiber for dispersion compensating and wide band

amplification

A. S. Samra, H. A. M. Harb

813 Performance analysis of a novel OFDM system based on dual – tree

complex wavelet transform (DT-CWT)

Mohamed H. M. Nerma, Nidal S. Kamel, Varun Jeoti

823 Ant colony optimization algorithm

Nada M. A. Al Salami

827 A survey of MAC protocols for wireless sensor networks

Rajesh Yadav, Shirshu Varma, N. Malaviya

834 Adaptive call admission control in TDD-CDMA cellular wireless networks

Dhananjay Kumar, Chellappan C

VARIABLE STEP SIZE ALGORITHMS FOR NETWORK

ECHO CANCELLATION

O.O. Oyerinde and S.H. Mneney

School of Electrical, Electronic and Computer Engineering, University of KwaZulu-Natal, King George V Avenue, Glenwood, Durban, 4041, South Africa

[email protected] and [email protected]

ABSTRACT Convergence rate of an algorithm is an important factor that determines the deployment of such algorithm in a real time application. In this paper, we propose improved versions of normalized least mean square (NLMS) algorithm: single and multiple -variable step size normalized least mean square (VSSNLMS) algorithms for echo cancellation. The presented algorithms exhibit faster convergence rate in comparison to NLMS algorithm. Simulation results employing standard figure of merits show how the algorithms perform better than NLMS algorithm based echo canceller. The good performance exhibit by these algorithms in terms of convergence rate as indicated by Means Squared Error (MSE) and Echo Return Loss Enhancement (ERLE) will lend them to deployment in the real-time network echo cancellation applications.

. Keywords: Echo cancellation, double talk, normalized least mean square (NLMS), single variable step size normalized least mean square (SVSSNLMS), multiple variable step size normalized least mean square (MVSSNLMS).

1 INTRODUCTION Echo cancellation in communication system has been deployed in telephone networks for voice quality enhancement for several decades. Echo, a delayed or distorted version of the transmitted signal reflected back to the source is caused by the four-wire to two-wire impedance mismatch in telephone networks. Distinct echoes are perceived when an un-attenuated reflection’s round-trip delay exceeds few tens of a millisecond. If the echo’s round-trip delay approaches a quarter of a second and there is little or no attenuation of the echo, most people cannot carry on with a conversation without stuttering. Consequently, there is a need for network echo cancellers for echo paths with long impulse responses such as 32ms or more.

In [1, 2] Adaptive Electrical Echo canceller for Telephone Network based on a combination of a Normalized Least Mean Square (NLMS) and Geigel double-talk detector (DTD) algorithms was presented. The improvement of the canceller as a result of the combination of the speech detector algorithm with NLMS algorithm was obvious in the results presented, but this was with a penalty of a slow convergence rate for longer impulse responses. In [3] a new NLMS adaptation scheme for echo cancellation was presented. The scheme combines the advantages of the Geigel algorithm with some initiative ideas. A new architecture that was

introduced makes the Geigel DTD algorithm to be more sensitive to the double talk condition, thus improving the echo canceller performance during the double talk condition but the problem of slow convergence rate was not addressed. In a bid to address the convergence rate exhibited by the echo canceller based on NLMS algorithm, various algorithms have been proposed with varied performances. Among these algorithms are proportionate normalized least mean squares (PNLMS) and PNLMS++ proposed in [4] and [5] respectively.

This paper focuses on improving the convergence rate of the echo canceller based on NLMS algorithm by employing variable step size instead of fixed step size for NLMS adaptive algorithm. This work is an improvement on the work presented in [1, 2].

Throughout this paper bold small letters such as x denote column vectors and dependency on time

index n are denoted as nx . { }E x is the expectation

of x . Superscript T denotes transpose. The paper is organized as follows. The system

model is described in Section II. In Section III, the NLMS adaptive algorithm is presented while in Section IV the proposed Single and Multiple-VSSNLMS, and DTD algorithms are presented. Figure of merits used to establish the performance of the algorithms are discussed in section V and the simulation processes are discussed in Section VI,

UbiCC Journal, Volume 4, Number 3, August 2009 746

while the conclusion is drawn in Section VII. 2 SYSTEM MODEL

The system model for echo canceller and double-talk detector considered in this paper is illustrated in Fig.1. The echo path impulse response vector is represented by vector

0 1[ , ... , ]TLep h h −=h and its model in the canceller

is represented by the vector, 0, 1,[ , ... , ]ˆ Tn L nn

h h −=h ,

where L is the adaptive filter length. The signal nx is the sampled far-end signal. The response of the model ˆny is subtracted from the combination of the echo and the speech of the near-end speaker ny leaving only the sampled speech of the near-end speaker nv to be sent to the far-end user. The problem, of course, is in building (and maintaining) the model and, to some extent, in obtaining the response of the model to the excitation signal.

Echo cancellers as in Fig.1 are predominantly used to terminate long-distance 4-wire circuits on a per call basis, each circuit having a different impulse response. Also, during the call, variations in the echo path may occur.

Therefore, the echo path model ˆnh must have the

ability to learn and adapt to the new echo path impulse response at the beginning of each call. To accomplish this, the echo canceller uses an adaptive filter to construct the echo impulse response model. The adaptive filter is generally based on mathematical algorithm(s). The adaptive filter attempts to build the echo impulse response model by adjusting its filter coefficients (or tap-weights) in such a way as to drive ne to zero. This is fine if ny consists only of the echo of the far-end speech. In that case, the correlation of nx and

ny contains valuable information about the echo impulse response. If, on the other hand, ny also contains significant amounts of the summation of near-end signal, nv and background noise, then the echo impulse response information is corrupted by any extraneous correlation between nx and nv . For this reason, practical echo cancellers need to inhibit adaptation of the filter taps when significant near-end signal is present and this is made possible by the presence of DTD. 3 NLMS ADAPTIVE ALGORITHM

The simplest and most popular adaptive iterative algorithm is the list mean square (LMS) algorithm given by the following equation [6]:

Figure 1: System model for echo canceller and double-talk detector

1ˆ ˆ n nn n

eµ+

= + xh h , (1)

ˆTn n nn

e y= − xh , (2)

where µ is the fixed step-size. LMS algorithm adjust the estimated impulse

response so as to minimize the cost function,

{ }2nE e , i.e., the mean square error. Each iteration

updates the current estimate of ˆnh by n neµ x ,

which is a step in the direction of a stochastic

approximation to the gradient { }2nE e . The

algorithm, though widely used because of its simplicity of implementation, suffers from relatively slow and data-dependent convergence behaviour. In order to make LMS algorithm insensitive to changes of the level of input signal, nx , the fixed step-size µ is normalized, resulting in the NLMS adaptive algorithm given as [6]:

21ˆ ˆ n

nn nn

eµ+

= + xh h

x, (3)

where 2

nx denote the Euclidean norm of the input

vector nx .

4 PROPOSED VARIABLE STEP SIZE NLMS

(VSSNLMS) ALGORITHMS AND DTD 4.1 Single-VSSNLMS Algorithm

The NLMS algorithm is given more attention in real-time applications because it exhibits a good balance between computational cost and performance. However, a very serious problem associated with both the LMS and NLMS algorithms is the choice of the step-size (µ) parameters. A small step size (small compared to the reciprocal of the input signal strength) will ensure small


misadjustments in the steady state, but the algorithms will converge slowly and may not track the nonstationary behaviour of the operating environment very well. On the other hand a large step size will in general provide faster convergence and better tracking capabilities at the cost of higher misadjustment. Any selection of the step-size must therefore be a trade-off between the steady-state misadjustment and the speed of adaptation.

Several studies [7, 8, 9] have thus presented the idea of variable step-size LMS algorithms in order to eliminate the “guesswork” involved in selection of the step-size parameter and at the same time ensuring that the speed of convergence is fast. When operating in stationary environment, the steady-state misadjustment values is very small, and when operating in non-stationary environment the algorithm should be able to sense the rate at which the optimal coefficients are changing and select a step-size that can result in estimates that are close to the best possible in the mean-squared-error sense. The variable step-size expression for Single-VSSNLMS algorithm employed in this paper is obtained by extending the approach used in [7] to derive similar variable step-size expression for the LMS algorithm. This is done by adapting the step-size sequence using a gradient descent algorithm so as to reduce the squared-estimation error at each time index. The Single-VSSNLMS algorithm is then given as:

21ˆ ˆ n

n nn nn

eµ+

= + xh h

x . (4)

The variable step-size nµ is updated as [10]:

2

11

ˆ2

nn n

n

eρµ µµ−

−

∂= −

∂ (5a)

2

11

.2

ˆˆ

Tn n

nnn

eρµµ−

−

∂∂= −

∂∂h

h (5b)

1 11 2

1

Tn n n n

n

n

e eρµ − −

−

−

= + x xx

. (6)

In Eq. (6), ρ is a small positive constant that controls the adaptive behavior of the step-size sequence nµ . Deriving conditions on ρ so that convergence of the adaptive system can be guaranteed appears to be a difficult task. However, the convergence of the adaptive filter can be guaranteed by restricting nµ to always stay within the range that would ensure convergence. Therefore the step size obtained from Eq. (6) would not be used for coefficient adaptation at any particular time index if it falls outside the values that guarantee convergence of the NLMS algorithm with a fixed step-size. As a result the step-

size sequence nµ will be restricted to within the range 0 2nµ< < [11]. The variable step size nµ is then restricted as follow:

max max

min min

ˆˆ

ˆ

n

n n

n

if

if

otherwise

µ µ µµ µ µ µ

µ

>��= <��

(7) where min max0 2µ µ< < < .

In [12] the order of coefficient update of NLMS is given as O(ML) where L is the filter length and M is the echo path maximum delay. However, the VSSNLMS algorithm only requires L extra additions and (L+4) extra multiplications (divisions) compared with NLMS algorithm, the value which is more or less negligible. 4.2 Multiple-VSSNLMS Algorithm

In Multiple-VSSNLMS algorithm rather than using a single variable step size for the adaptation of all the echo canceller’s coefficients in the coefficient vector, ˆ

nh , each coefficient is adapted with unique

variable step size resulting in multiple- VSSNLMS algorithm. As a result, the variable step-size nµ in Eq.

(4) becomes a vector 0, 1,, ... ,T

n L nnµ µ −� �= � �µµµµ and

is derived following Eq. (5) and Eq. (6) as :

1 121

1

ˆ ˆT

n n n nn n

n

e eρ − −−

−

= + x xx

µ µµ µµ µµ µ . (8)

Similarly, each of the variable step size, nµ in the

multiple-variable step size vector nµµµµ is restricted

within the range as given in Eq. (7). 4.3 Geigel Double Talk Detection (DTD)

During double talk, the period where there is presence of the far- and near- end speech simultaneously, double-talk detector is needed to inhibit taps adaptation. A very efficient and simple way of detecting double-talk is to compare the magnitude of the far-end and near-end signals and declare double -talk if the near-end signal is lager in magnitude than a value set by the far-end speech. Geigel DTD algorithm [13], attributed to A. A. Geigel is a proven algorithm in general use for this purpose and is given by Eq. (9) through which a double talk is declared if

{ }1 1max , , ... ,n n n n Ly x x xξ − − +≥ , (9)

where ξ is the detector threshold factor normally set to 0.5 if the network hybrid attenuation, Echo Return Loss (ERL), is assumed to be 6dB and to 0.71 if the ERL is assumed to be 3dB. Beside this threshold factor, a hangover time, holdτ , is also specified such


that if double-talk is detected, then the adaptation is inhibited for this specified duration beyond the detected end of double-talk. 5 FIGURE OF MERITS

There are two figures of merit employed in this simulation. One of these figure of merit used to establish the performances of the proposed echo canceller algorithms is a quantity called Echo Return Loss Enhancement (ERLE). This is a comparison of the echoes before and after cancellation. It is calculated as:

1010 logpower of the echo signal

ERLE dBpower of the residual echo

= � �

�

,

{ }( )

10 210 log

ˆ

2n

n n

EdB

E

y

y y

� �� =� �� −� ��

� � �

,

{ }{ }10 2

10 log

2n

n

EdB

E e

y � �

= � ��

.(10)

The ERLE therefore is the amount of attenuation of the echo signal introduced by the echo canceller. It does not include any further reduction in the residual echo by any extra nonlinear processing after the basic echo cancellation. The ERLE provides a figure of merit for determining how effective the echo cancellation process is. It assumes that there is always a certain amount of loss incurred by the echo and then shows the rate of improvement after echo cancellation. It reflects both the convergence rate and the steady-state residual echo. The plot of ERLE versus time or sample shows the rate of change in the enhancement: it shows the rate of convergence of the algorithm to the steady-state error value. The ERLE gives a good indication of the performance of the echo canceller. Over time the ERLE changes, initially it may be quite small but as the algorithm converges towards the optimum tap-weight values it increases. Theoretically the steady state ERLE could be very large and an ideal echo canceller with a perfectly linear echo signal would output an infinite ERLE in a very short period of time. Practically however, there are limiting factors to this result; the echo path always contains some non-linearities introduced by various components in the transmission path; the devices that generate the echo produce a certain amount of echo loss that little can be done about and the use of finite-precision devices limit the accuracy of the computations. Therefore the ERLE will not reach its theoretical steady-state

maximum value. Nevertheless, a good performing echo canceller will output a very large steady-state ERLE in a very short convergence time [14].

Another important figure of merit used is the MSE which shows the adaptation curves of the algorithm employed. It is given mathematically as the expectation of the norms of the square error as follow:

{ }( )210MSE 10 log ( ) (dB)E e n= ×

{ }( )*1010 log ( ) ( ) (dB)E e n e n= × (11)

6 SIMULATION

The performances of both Single and Multiple-VSSNLMS algorithms have been compared with that of NLMS algorithm with fixed step size. The echo path is modeled with an impulse response, g(n) of a linear digital filter. In order to account for the delay experienced by the echo signal and the ERL of hybrid transformer in a telephone network, g(n) is chosen as a delayed and attenuated version of the excitation signal according to ITU G.168 standard for testing network echo canceller performance [15]. The mathematical expression for g(n) is given as:

( ) 10exp( ) ( )20 i

ERLg n M n δ= − × − , (12)

where the sequence ( )iM n denotes the echo paths with varied time-delay, and δ represents the total delay experienced by the echo signal. For all the results presented in this paper, ERL value of 6dB is used because this is a typical worst case value encountered for most networks, and most current networks even have typical ERL values better than 6dB.

Two types of excitation signals are employed for the simulation of the results presented in this paper: the type of random signal used in [1, 2], as well as sampled speech signal as shown in Fig. 2 and Fig. 7 respectively. For each of the excitation signals, maximum echo delay was set to 16ms (128 samples) and 32ms (256 samples), while the echo canceller length (adaptive filter length) was set to 256(32ms) and 512 (64ms) respectively. For effective performance of any echo canceller the length of the echo canceller is always selected such as to be longer than the maximum possible echo delay in the network. The following other parameters were used in the simulation:µ =0.02 for NLMS algorithm and also to initialize the Single-VSSNLMS and Multiple-VSSNLMS algorithms, ρ = 2×10-4 . The performances of the proposed algorithms based on the figure of merits discussed in section V are as shown in Fig. 3 to Fig. 6, and Fig. 8 to Fig. 11 for random signal and sampled speech signal as excitation signals respectively.


0 1000 2000 3000 4000 5000 6000 7000 8000-4

-3

-2

-1

0

1

2

3

4

Samples

Mag

nitu

de

Input Signal

Figure 2: Random signal as the excitation signal

0 1000 2000 3000 4000 5000 6000 7000 8000-50

-45

-40

-35

-30

-25

-20

-15

-10

-5

0

MS

E(d

B)

Samples

MSE of the Echo Canceller

NLMS AlgorithmVSSNLMS AlgorithmMultiple-VSSNLMS Algorithm

Figure 3: MSE for the algorithms with random signal as the excitation signal, L =256(32ms), M =128 (16ms).


0 1000 2000 3000 4000 5000 6000 7000 8000-50

-45

-40

-35

-30

-25

-20

-15

-10

-5

0

MS

E(d

B)

Samples



Figure 4: MSE for the algorithms with random signal as the excitation signal, L =512(64ms), M =256 (32ms).

0 1000 2000 3000 4000 5000 6000 7000 80000

5

10

15

20

25

30

35

40

45

50

ER

LE(d

B)

Samples

ERLE of the Echo Canceller


Figure 5: ERLE for the algorithms with random signal as the excitation signal, L =256(32ms), M =128 (16ms).


0 1000 2000 3000 4000 5000 6000 7000 80000

5

10

15

20

25

30

35

40

45

50

ER

LE(d

B)

Samples



Figure 6: ERLE for the algorithms with random signal as the excitation signal, L =512(64ms), M =256 (32ms).

0 1000 2000 3000 4000 5000 6000 7000 8000-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1

Samples

Mag

nitu

de

Input Signal

Figure 7: Sampled speech signal as the excitation signal


0 1000 2000 3000 4000 5000 6000 7000 8000-50

-45

-40

-35

-30

-25

-20

-15

-10

-5

0

MS

E(d

B)

Samples



Figure 8: MSE for the algorithms with sampled speech signal as the excitation signal, L =256(32ms), M =128 (16ms).

0 1000 2000 3000 4000 5000 6000 7000 8000-50

-45

-40

-35

-30

-25

-20

-15

-10

-5

0

MS

E(d

B)

Samples



Figure 9: MSE for the algorithms with sampled speech signal as the excitation signal, L =512(64ms), M =256 (32ms).


0 1000 2000 3000 4000 5000 6000 7000 80000

5

10

15

20

25

30

35

40

45

50

ER

LE(d

B)

Samples



Figure 10: ERLE for the algorithms with sampled speech signal as the excitation signal, L =256(32ms), M =128 (16ms).

0 1000 2000 3000 4000 5000 6000 7000 80000

5

10

15

20

25

30

35

40

45

50

ER

LE(d

B)

Samples



Figure 11: ERLE for the algorithms with sampled speech signal as the excitation signal, L =512(64ms), M =256 (32ms).


0 1000 2000 3000 4000 5000 6000 7000 8000-4

-3

-2

-1

0

1

2

3

4

Samples

Mag

nitu

de

Input Signal

Figure 12: Reference signal (far-end signal) for double-talk condition testing

0 1000 2000 3000 4000 5000 6000 7000 8000

-0.1

-0.05

0

0.05

0.1

0.15

Samples

Mag

nitu

de

Near-end Signal

Figure 13: Near-end signal for double-talk condition testing


0 1000 2000 3000 4000 5000 6000 7000 8000-50

-45

-40

-35

-30

-25

-20

-15

-10

-5

0

MS

E(d

B)

Samples

MSE of the Echo Canceller during double-talk condition


Figure 14: MSE for the combination DTD and proposed algorithms during the double-talk condition, L =512(64ms), M =256 (32ms).

0 1000 2000 3000 4000 5000 6000 7000 80000

5

10

15

20

25

30

35

40

45

50

ER

LE(d

B)

Samples

ERLE of the Echo Canceller during double-talk condition


Figure 10: ERLE for the combination DTD and proposed algorithms during the double-talk condition, L =512(64ms), M =256 (32ms).


In order to establish the robustness of the combination of Geigel DTD algorithm with the proposed echo canceller algorithms during double talk condition, random signal of different magnitude compared with the reference signal was added with the echo signal to serve as the near-end signal after about half of the period of the cancellation process has elapsed. The simulation was run for the echo canceller of length 512. Fig. 14 and Fig.15 show how effective the combination of Geigel DTD and the proposed algorithms performed in comparison with the combination with NLMS algorithm during the double talk condition. Although the performances of the algorithms were reduced compared with the situation where there was no double-talk, the results still show that there was effective cancellation during this condition with the help of Geigel DTD algorithm.

It could be observed from these results that both Single-VSSNLMS and Multiple-VSSNLMS algorithms outperformed the NLMS algorithm as a result of the variability of the step-size, but the differences in the performance of Single-VSSNLMS and Multiple-VSSNLMS algorithms are negligible. This shows that assignment of a unique variable step size for the adaptation of each of the coefficients of the echo canceller makes no or little difference compared with adapting all the coefficients with the same variable step size.

7 CONCLUSION

In this paper we have presented Single-VSSNLMS and Multiple-VSSNLMS algorithms for the network echo cancellation. These algorithms use variable step sizes instead of fixed step size used by NLMS algorithm. As a result, the convergence rates of these algorithms are significantly faster than that of NLMS algorithm. These algorithms also exhibit high performance during double-talk condition. As a result of the negligible difference in the performance of the Single-VSSNLMS and Multiple-VSSNLMS algorithms, it could be concluded that Single-VSSNLMS algorithm which is less complex than Multiple-VSSNLMS algorithm should be employed in the real-time network echo cancellation applications. 8 REFERENCES [1] O. O. Oyerinde, and T. K. Yesufu: Adaptive

Electrical Echo Canceller for Telephone Networks, CD-ROM Proc. IEEE Military Communication Conference, MILCOM 2005, Atlantic City NJ, USA, Vol. xxii+3341, pp.1-5, Oct. 17-20, (2005).

[2] O. O. Oyerinde, and T. K. Yesufu: Adaptive Electrical Echo Canceller Algorithm, Proc.

Intelligent Engineering System through Artificial Neural Network Conference, ANNIE 2005, Missouri-Rolla , USA, Vol. 15, pp.613-622, Nov. 6-9, 2005.

[3] J. F. Liu: A Novel Adaptation Scheme in the NLMS Algorithm for Echo Cancellation, IEEE Signal Processing Letter., Vol. 8, No. 1 pp. 20– 22, January, (2001).

[4] D. L. Duttweiler: Proportionate normalized least mean square adaptation in echo cancellers, IEEE Trans. Speech Audio Processing, Vol. 8, pp. 508–518, Sept.., (2002).

[5] S. L. Gay: An efficient fast convergence adaptive filter for network echo cancellation, Proc. Assilomar Conf., Nov., (1998).

[6] B. Widrow, J. R. Glover, J.M. McCool Jr., J. Kaunitz, C. S. Williams, R. H. Hearn, J. R. Zeidler Jr., E. Dong, R. C. Goodlin: Adaptive noise canceling: principles and applications, Proc. IEEE 63 (12), pp. 1692-1716, Dec. (1975).

[7] V.J. Mathews, Z. Xie: A stochastic gradient adaptive filter with gradient adaptive step-size, IEEE Trans. Signal Process., Vol.41, no.6, pp. 2075–2087 June (1993).

[8] T. Aboulnasr: A Robust variable step-size LMS-Type Algorithm: Analysis and Simulation, IEEE Trans. Signal Process. Vol.45, no.3, pp. 631–639 March, (1997).

[9] D.I. Pazaitis, A.G. Constantinides: A novel kurtosis driven variable step-size adaptive algorithm, IEEE Trans. Signal Proc. Vol.47, no.3 pp.864–872 March (1999).

[10] Y.K. Shin, J.G. Lee: A study on the fast convergence algorithm for the LMS adaptive 3lter design, Proc. KIEE, Vol.19, no. 5, pp. 12–19, October (1985).

[11] M. Tarrab, A. Feuer: Convergence and performance analysis of the normalized LMS algorithm with uncorrelated Gaussian data, IEEE Trans. Inform. Theory, Vol.34, no.4, pp.680– 691, July (1988).

[12] Dieter Schafhuber, and Gerald Matz: MMSE and Adaptive Prediction of Time- Varying Channels for OFDM Systems, IEEE Transactions on Wireless Communications, vol. 4, no. 2, pp. 593-602, March (2005).

[13] D. L. Duttweiler: A twelve-channel digital echo canceller, IEEE Trans. Commun., Vol. COM-26, pp. 647-653, May (1978).

[14] Y. Lu and J.M. Morris: Gabor Expansion for Adaptive Echo Cancellation, IEEE Signal Processing Magazine, pp. 68-80, March (1999).

[15] ITU G.168, Recommendations: Digital Echo Canceller, (2002)


A Framework for Automatic Reconfigurations of Protocol Stacks inUbiquitous Computing Systems

Mahdi Niamanesh, Rasool JaliliDepartment of Computer Engineering

Sharif University of Technology, Tehran, [email protected], [email protected]

Abstract

Ubiquitous computing environment includes wireless networks, autonomic networked systems and devices withheterogeneous standards and protocols for different contexts and situations. Autonomic networked systems requiremechanisms for automatic on-the-fly reconfigurations in their protocol stacks to be able to operate in differentsituations and contexts. Automatic reconfigurations allow the system to continue its normal operation withoutany need for human intervention. Since every protocol in a protocol stack has at least one peer protocol in anothersystem, a dynamic reconfiguration of a protocol may raise the need for a reconfiguration in the peer stack. Dependingon the reconfiguration type, the reconfiguration maybe carried out in a single or in both peer systems. For an assureddynamic (run-time) reconfiguration of a protocol, the execution of the protocol component is stopped in a safe state,a new protocol is initialized, and stacks switch to the new protocol at the same time. This paper proposes a protocolstack framework to support automatic and assured reconfigurations of protocol components. Our evaluation andperformance measurement show the feasibility of automatic reconfigurations and an acceptable level of transparencyin dynamic reconfigurations.

Keywords: Protocol Stack, Automatic Reconfigurations, Dynamic Reconfiguration


1 Introduction

Ubiquitous computing environment includes wire-less networks, autonomic networked systems and de-vices with heterogeneous standards and protocols fordifferent contexts and situations [20]. The SoftwareRadio [3, 8] offers dynamic reconfigurability for pro-tocol stacks of such systems and devices in order tofacilitate applications such as changing routing algo-rithms of switches, changing security modules in pro-tocol stacks, bug fixing, and customizing the protocolstack of a device for better performance.

In general, in a software system, dynamic reconfig-uration of a component to a new one includes suchphases as freeze (stopping the current execution of thecomponent), change (adding/binding a new componentand unbinding/removing the unnecessary old compo-nent from the system), state transfer (finding and ini-tializing a proper state in the new component in orderto resume the execution), and re-execution (resumingthe execution from a non-initial state in the new com-ponent) [15]. In order to have assured reconfiguration,the old component should be frozen in a “safe state”and the new component should resume the executionfrom a “reachable” state [6].

In the context of protocol stack reconfigurations,since each protocol is defined at least between twopeer components, reconfiguration of a running proto-col component may require a corresponding reconfig-uration in the peer component(s). For example, con-sider a reconfiguration that changes TCP component ina TCP/IP protocol stack into SCTP component [22].Both TCP and SCTP have the same role, as trans-port protocol. For this reconfiguration, both peersshould be synchronized in terms of the change (e.g.,from TCP to SCTP) and freeze state. However, inreconfiguring TCP component into TCP-Nice compo-nent [24], since TCP-Nice is backward-compatible withTCP, the reconfiguration can be performed in a senderstack transparently and independent of its peer stack.As a result, reconfigurations can be categorized intosingle reconfigurations (involving a single stack) anddistributed reconfigurations (involving changes in twoor more stacks).

Autonomic systems in ubiquitous computing envi-ronment should have dynamic protocol stacks that canautomatically detect reconfiguration types and performthe assured reconfigurations. Related work for dy-namic protocol stacks only address single or distributedreconfigurations. For example [9, 11] suppose a runningprotocol stack as a stand-alone system (not in a net-work) and perform a single reconfiguration. In order toprovide an assured reconfiguration, they have defined

the safe state of a running component as a state wherethe component has no data and is not in any interac-tion with the other components [7, 11]. In the SoftwareRadio and Cognitive Radio domains, related work hasmainly concentrated on device protocol stack reconfig-urations [8] and cognitive engine development for soft-ware radio (e.g., [14]). In the Internet domain, relatedwork such as [19, 23] have concentrated on changing thecommunication protocol between two systems. Theseworks limit safe states to states in which the old proto-col execution has been completed and the new protocolshould start a new execution from an initial state.

In this paper, the reconfiguration problem is definedas changing the protocol between two peer protocol com-ponents at run-time automatically. The reconfigura-tion maybe performed in a single peer (single recon-figuration) or in both peers (distributed reconfigura-tion) Our idea is to keep enough information aboutprotocols to perform automatic and assured reconfigu-rations. Based on our knowledge about two communi-cating protocol components, the reconfiguration typeis detected, and is carried out in a safe state. Wedo not limit safe states to the end of protocol execu-tions. For such reconfigurations, we propose a softwareframework including mechanisms for dynamic reconfig-uration management and control for both types of re-configuration. Through the framework, each stack canrequest a reconfiguration from its peer while they arecommunicating. Our framework for automatic recon-figurations guarantees smooth change of the old proto-col to the new one autonomously.

The rest of this paper is organized as follows. Sec-tion 2 describes background about protocol executions,protocol reconfigurations, and reconfiguration assur-ance in protocol stacks. In Section 3, we explain theproposed framework. Mechanisms for reconfigurationmanagement and control for single and distributed re-configurations are presented. In Section 4, we describeimplementation and evaluation of the proposed frame-work. In Section 5, we discuss related work and Section6 concludes the paper.

2 Background

In this section, we describe a simple model for proto-col execution. Afterwards, two types of protocol stackreconfigurations are explained and finally, we state as-surance in reconfigurations of two communicating stacks.

2.1 Protocol Execution

We consider a simple model for describing the recon-figuration problem in communicating protocol stacks.

2


Figure 1 shows two communicating protocol stacks,Stack1 and Stack2. In each layer of the stacks, onecomponent, which we refer to as protocol component,implements functionality of its corresponding protocol.For instance, in the figure, components P1 and P2 pro-vide functionality of protocol P . A protocol compo-nent is a set of rules as well as a processing entity thatgoverns communication between communicating peers.The protocol components interact with each other byexchanging protocol messages.

Figure 1. An example of two communicatingprotocol stacks

State of a protocol component describes all usefulinformation about the protocol in a point of execu-tion. Such information includes values of all variablesand contents of all input/output buffers, related to thecomponent. For simplicity, we split states of a protocolcomponent into macro-states and micro-states. Macro-states describe states of the protocols’ finite state ma-chines. Examples of macro-states in the TCP protocolare CLOSED and ESTABLISHED. Micro-states de-scribe states of protocols at run-time, to maintain in-formation for operations such as reliability, error han-dling, and congestion control. We use macro-states ofcomponents in specification levels and micro-states inexecution levels.

2.2 Reconfigurations in Protocol Stacks

With two communicating protocol stacks, a dynamicreconfiguration in a component may lead to a corre-sponding reconfiguration in its peer component. In ourperspective, protocol stack reconfigurations are cate-gorized into two types, single reconfiguration and dis-tributed reconfiguration. Figure 2 depicts scenarios forsingle and distributed reconfigurations. Figure 2-(a)depicts a situation in which in one of the stacks (e.g.,Stack1), a protocol changes into another variation thatis supported by the peer protocol component (in Stack2).According to the figure, the old protocol component,

P1, is replaced by a new component P3, which can“safely” interoperate with the old peer component, P2,in the peer stack. In this case, the new component, P3,is “backward-compatible” with the old one. Therefore,the reconfiguration can be carried out in the changingstack without disrupting the peer stack.

Figure 2-(b) depicts a scenario in which two peercomponents in the two stacks are synchronously re-placed with new protocol components. For two commu-nicating TCP/IP protocol stacks, changing TCP pro-tocol of two stacks into SCTP protocol [22] can be anexample of this type.

Figure 2. Types of dynamic reconfigurationsfor protocol stacks: (a) single reconfigura-tion, (b) distributed reconfiguration

For single reconfigurations, the goal is to reconfigurea running protocol component independent of its peer,transparently (in point of its peer component view).However, the goal of distributed reconfigurations is toreconfigure two peer components synchronously.

2.3 Reconfiguration Assurance

Intuitively, a dynamic reconfiguration that changesa running component into a new one is assured if afterthe changing phase, the new component can be exe-cuted just as if it has been executed from its initialstate [6]. Based on this notion, if the new component

3


starts re-execution from a reachable state1, then the re-configuration will be assured. Based on [6], to achievesuch a reachable state, the running component shouldbe frozen in a safe state, and its state should be trans-ferred into the new component. Considering two com-municating components as a distributed component,its state is the global state2 of the two components andthe reconfiguration of two communicating componentsis assured if after the reconfiguration, the executionresumes from a reachable global state.

We explain requirements for assured reconfigurations(both single and distributed reconfigurations) in thefollowing:

Safe state A safe state for a reconfiguration has beendefined as a state that has no interaction withthe other components [7, 11]. A component ina safe state does not accept new requests, doesnot initiate new operations, and all its initiatedoperations have been completed [18]. However,for two communicating stacks, the global stateshould be safe. Accordingly, two communicat-ing components should be frozen in a safe globalstate, and two new components should resumethe execution from a reachable global state.

State transfer Execution of the new component mayberesumed from a non-initial state, which we referto as the restarting state. The old component’sstate should be transferred to the restarting state.In a reconfiguration of two peer components, therestarting states of both peers should be initial-ized. An important point in the state transferis the possible dependency of the states of twopeers upon each other. For example, a UDPsender component should know the port numberof the peer UDP receiver. For the state transfer,firstly, it is necessary to find the restarting statein the new component; secondly, the new restart-ing state should be initialized to resume the ex-ecution; thirdly, some parts of the new compo-nent’s state may require to be initialized basedon its peer component.

3 DRAPS Framework

DRAPS (Dynamic-Reconfigurable Architecture forProtocol Stack) is an extendable framework that presentsassured and synchronous dynamic reconfiguration for

1State s in component C is said reachable state, if and only ifan execution of component C starting from an initial state canreach s at some time for some inputs.

2Global state for communicating components is defined in thesame manner as that for the global state in distributed systems.

two communicating protocol stacks Without losing thegenerality, we assume that each protocol in DRAPS, isimplemented as a distinct component. A protocol com-ponent is a set of rules as well as a processing entitythat governs the communication between communicat-ing peers.

Architectural components of DRAPS are shown inFigure 3. As shown, DRAPS is built out from a coreframework and some plug-in components. The coreframework is responsible to perform assured dynamicreconfigurations and consists of two components, namely,Reconfiguration Management and Control (RMC), andProtocol Knowledge Base (PKB). RMC provides mech-anisms for automatic reconfiguration management oftwo peer stacks. PKB is a knowledge base component,responsible for initializing protocols specifically whenthey are resumed from non-initial states.

Plug-in components in DRAPS present extendabil-ity for the framework by providing supplementary re-quirements of dynamic reconfigurations. For instance,a repository of reconfigurable components is not an es-sential requirement in a reconfiguration; however, it isa supplementary requirement in every reconfigurationframework. DRAPS provides a plug-in type for such acomponent repository.

Figure 3. Architecture of DRAPS includingcore framework and plug-ins

3.1 The Protocol Knowledge Base

We introduce a knowledge base component, calledProtocol Knowledge Base (PKB), to support assuredand dynamic reconfigurations for protocol stacks of au-tonomic systems. Theses systems should detect re-configuration types and perform assured reconfigura-tions without any need for human intervention. Forthis reason, we keep a protocol knowledge in the PKBcomponent. Specifically, PKB contains three typesof knowledge, namely, peer-compatibility of protocols,

4


role-compatibility of protocols, and protocol state in-formation (Figure 4) all explained in the following.

Figure 4. PKB consists of 1) Knowledge ofrole-compatibility of protocols, 2) Knowledgeof peer-compatibility of protocols, and 3)Knowledge of protocols states

3.1.1 Peer-Compatible Protocols

Every protocol reconfiguration requires commitment ofits endpoint components on a change and a freeze state.Commitment on the change means the two compo-nents change into compatible protocols. For example,when a peer component changes into the TCP pro-tocol, the other peer can not choose UDP protocolfor its transport protocol. For this reason, we definepeer-compatibility relationship between two protocolcomponents as follows. Two components have peer-compatibility relationship (are peer-compatible) if theycan inter-operate with each other as two peer compo-nents. We introduce two types of peer-compatibility.i.e., Strong ly peer-compatible if they can inter-operatein such a way that all their functionalities can be poten-tially used in the communication, Weak ly peer-compatibleIf parts of functionalities of the components can notbe used, due to problems such as version mismatch-ing. For example, TCP and TCP-NewReno [4] (orTCP-SACK [10]) components, as two communicatingpeers, have the weak peer-compatibility, whereas TCPand TCP-Nice possess strong peer-compatibility [17].The PKB component keeps all types of peer-compatibleprotocols.

The commitment on the freeze state necessitates twopeer components to start the reconfiguration in a safeglobal state. For this purpose, the PKB componentkeeps all peer-compatible macro-states for each pair ofpeer-compatible protocols. By peer-compatible macro-states, we mean two “compatible” macro-states from

two peer-compatible protocols3. For example, when aTCP sender component is in the ESTABLISHED state,its peer component, which is a TCP receiver, can notbe in the CLOSED state.

3.1.2 Role-Compatible Protocols

In a reconfiguration of an existing component into anew component, the new one should play the same roleas the existing one. For example, we may reconfigureTCP into SCTP or UDP, but it is not possible to re-configure TCP into IP. In fact, the role of TCP andIP differs completely. We define role-compatibility re-lationship between two protocol components that canbe reconfigured to each other. A pair of protocol andits backward-compatible version is a well-known exam-ple of role-compatible protocols. We define three typesof role-compatibility relationship between two protocolcomponents, which are determined based on the peer-compatibility of new component and the peer compo-nent. Considering a component, P1, and its peer, P2,which are strongly peer-compatible components, we de-scribe types of role-compatibility as follows.

We call a component, P3, as fully backward-compatible(FBC, in short) with P1, if P3 can be strongly peer-compatible with P2. For two communicating peer com-ponents, we can perform a single reconfiguration bychanging a component to one of its FBC components.For a protocol component, its FBC protocols are thosethat inherently require changes to only one endpointcomponent. For example, both the Fast Recovery mod-ification to the sender-side TCP [1] and TCP-Nice aretransparent to the receiver.

Not Backward-Compatible role-compatibility (NBC,in short) relationship is held between P1 and P3, ifP3 can not have any peer-compatibility with P2. Inthis case, the reconfiguration necessitates a distributedreconfiguration. For example, SCTP can be regardedas a role-compatible protocol with TCP and UDP. Butchanging SCTP into TCP or UDP should be synchronouslycarried out in both peer components to be of value.

Partially Backward-Compatible role-compatibility (PBC,in short) relationship is hold between P1 and P3, ifP3 can have weak peer-compatibility with P2. In thiscase, any reconfiguration from P1 into P3 can be car-ried out either single or distributed. For a protocol,PBC protocols are those that have the potential to beeither more effective if both peer components could bereconfigured and the new functionality could be opera-

3In [16], we have formally defined two compatible states. Wehave proposed protocol automata specifying the communicationof two communicating components. Each state of protocol au-tomaton is composed of two compatible states from the two com-ponents.

5


tional between the sender and receiver. If we reconfig-ure only one peer, we will have partial or no new valuein the communication. For example, for two commu-nicating TCP components, changing one of them intoTCP-NewReno, which uses a heuristic interpretationof duplicate acknowledgments to avoid timeouts, doesprove useful. However, by changing the peer compo-nent into TCP-NewReno, the communication will havemore useful. Through changing one of the componentsinto TCP-SACK, which provides better recovery fromlosses, the communication will serve no benefit unlessthe peer component is TCP-SACK enabled or we re-configure the peer component into TCP-SACK as well.In these cases, we can perform either a single or dis-tributed reconfiguration.

For each pair of role-compatible protocols, PKB con-tains their role-compatibility type. Moreover, it con-tains role-compatible macro-states of the two proto-cols. A pair of role-compatible macro-states of tworole-compatible protocols consists of two correspondingstates that can play the same role in the two protocols4.This helps reconfigurations as follows; when a protocolstops in a state (freeze state), a corresponding state(restarting state) is found through its role-compatiblemacro-states to resume the execution.

3.1.3 Protocol State Information

All required knowledge for finding safe states and ini-tializing restarting states is kept in PKB.

We present a model for state representation of pro-tocol components by introducing protocol control block(PCB) that contains state information. This informa-tion includes addressing parameters, buffers (sent, re-ceived, un-acknowledged, etc.), protocol segment vari-ables, counters, timers, send variables, receive vari-ables, and the other required parameters. Protocol de-velopers should implement one PCB data structure foreach protocol component.

PCB handlers are used to set values of PCB pa-rameters during a state transfer. They are also usedin distributed reconfigurations to set values of the re-mote PCBs in peer systems. In this case, we call themremote PCB handlers. As an example of the appli-cation of remote PCB handlers, consider a reconfigu-ration in UDP protocol. In UDP protocol, the UDPsender should know the port number of the UDP re-ceiver; therefore, the receiver should send its port num-ber to the sender. PKB contains PCB parameters andformats for different protocols.

In the state transfer, the old PCB is used to valuate4Based on [16], we define two role-compatible states in two

substitutable components as two bisimilar states.

the same parameters in the new PCB. Then, the PCBhandlers are used to complete the state transfer andfinally the remote PCB handlers from the peer (onlyin distributed reconfigurations) are applied.

3.2 Reconfiguration supports for ProtocolComponents

Reconfigurable protocol components should providesome extra functionalities to support dynamic recon-figurations. For each protocol component, protocoldevelopers should implement reconfiguration interfaceincluding saveState() and restoreState() methodsfor state transfer, start() and stop() methods tostart and stop execution of the component. Finally,the semiFreeze() method should be implemented foreach component in order to support freezing the com-ponent in a proper state.

Moreover, dynamic reconfigurations of protocol com-ponents require indirect communication between twoadjacent protocol components (layers) in order to presenttransparent run-time reconfigurations [5]. For this rea-son, a wrapper component is used between two adja-cent protocol components. For a protocol component,the wrapper presents interface of the protocol compo-nent and manages the incoming requests to the com-ponent, by buffering them during the freeze periods.

3.2.1 Finding a Safe State

Due to the independence of protocol stack components(layers), finding a safe state for a dynamic reconfigu-ration can be carried out in a simple way. We definea safe reconfiguration point (SRP) as a point of exe-cution of a protocol component where the componentis at the beginning or at the end of the processing ofan input packet. In other words, the states just afterwriting a packet into the output buffer or before read-ing a packet from the input buffer are SRPs. In suchstates, either no operation is started on the packet or allthe operations are completed. Usually, macro states ofprotocols are like this and we use them as SRPs. Thisdefinition is consistent with [11], where all operationsof the component in SRP should be completed.

Generally, finding a SRP in an execution of a com-ponent is a reachability problem, which is undecidable[6]. To cope with this problem, we ask protocol de-velopers to “mark” some SRPs (macro states) in thesource code of protocols. Therefore, protocol develop-ers are asked to put some pieces of code in the sourcecode of protocol components to enable them to reportmacro-states during the execution if requested. In anormal execution mode, a component does not report

6


SRPs. However, we introduce a semi-freeze mode inwhich the component reports every SRP (like in debugmode) upon reaching its execution. The RMC com-ponent monitors the reported states of the componentin the semi-freeze mode and stops its execution uponreaching the proper SRP for a dynamic reconfiguration.

For distributed reconfigurations, the global state oftwo peer components should be safe. A global state(s, r) for two communicating components is a safe globalstate if s and r are locally SRP and their role-compatiblestates, denoted as s′ and r′ respectively, are peer-compatiblewith each other. Figure 5 depicts the scenario of findinga safe global state for a distributed reconfiguration. Foreach SRP state, say s, RMC checks whether its role-compatible state in the new component, e.g., s′, has apeer-compatible state in the new peer component. Ifso, the pair of reported state and its peer-compatiblestate in the peer component is the safe global state forthe reconfiguration.

Figure 5. The sequence of finding a safeglobal state

3.3 Automatic Reconfiguration Managementand Control

We offer automatic reconfiguration management andcontrol (RMC) through a software component (RMCcomponent) for protocol stacks. The RMC componentcan receive reconfiguration commands from differentsources including system administrators, monitoringcomponents in the system, or peer system RMCs. A re-configuration command asks replacing an old protocolcomponent with a new one.

We provide automatic reconfigurations in three phases;in the reconfiguration preparation phase, conditions foran assured reconfiguration are checked. In reconfigura-tion synchronization, two peer stacks synchronize on adynamic reconfiguration. In reconfiguration execution

the new component is installed and executed. In thefollowing subsections, we explain automatic reconfigu-ration phases.

3.3.1 Reconfiguration Preparation

Considering two communicating protocol stacks, wewould like to reconfigure one of the stacks by sending areconfiguration command indicating the currently run-ning protocol component (old component) and a newcomponent. We describe how RMC and PKB com-ponents can manage an automatic reconfiguration forthese protocol stacks. Each protocol stack contains itsown RMC and PKB components. Moreover, we havethe following assumptions to simplify the problem:

• The communication channel between the two com-ponents is FIFO (First-In, First-Out), error-free,and having bounded communication delay.

• There is only one reconfiguration at a time. Noconcurrent reconfigurations are started.

The reconfiguration preparation is started upon re-ceiving a reconfiguration command by a RMC. Thegoal is to check preconditions for assured reconfigura-tions. Therefore, RMC checks SRPs for the currentlyrunning component. The component should have atleast one SRP and its role-compatible states, with re-spect to the reconfiguration, in the PKB component.Moreover, RMC checks the existence of new compo-nent’s PCB and PCB handlers.

3.3.2 Reconfiguration Synchronization

The goal of reconfiguration synchronization is to deter-mine a compatible reconfiguration for two peer stacks.Thus, the initiator RMC detects the received recon-figuration type, as depicted in Figure 6. If the role-compatibility between the old and new components isFBC then the reconfiguration type is single, and is car-ried out in the initiator stack. If the role-compatibilityis PBC and the peer-compatibility of the new compo-nent and the peer component is strong then, the re-configuration will be single in the initiator stack. Oth-erwise, the reconfiguration can be either single or dis-tributed.

Three types of messages (START, RECON, and Eo-REC) are exchanged by two RMCs to determine andperform a compatible reconfiguration. The STARTmessage is a command for reconfiguration start andindicates the current (old) component and a set of allpossible options for the peer stack. Each option in-dicates an alternative component and a proper freeze

7


state for the peer. Alternative components are deter-mined through the peer-compatibility relation with thenew component in the initiator stack. The freeze stateis determined based on safe global state. In addition,the START message includes a list of PCB values forremote PCB initialization. During the synchroniza-tion, the initiator RMC sends a START message tothe peer RMC. The peer RMC, based on the receivedoptions, prepares a corresponding START message (in-cluding its running protocol name and options for thepeer) and sends back to the initiator RMC. Now, theinitiator RMC can decide on the reconfiguration type.

For distributed reconfigurations, the initiator RMCselects two peer-compatible protocols for the initia-tor and the peer stacks. The peer’s reconfigurationis determined in a RECON (RECONfigure) message.Each RMC sends the EoREC (End of REConfigura-tion) message to its peer RMC after finishing a recon-figuration. It only indicates that the reconfigurationhas been successfully finished and that the new pro-tocol is operational. Therefore, the EoREC messagedoes not play any critical role in the reconfiguration ofthe two peers. The sequence of exchanging messagesbetween two stack during a synchronization is depictedin Figure 7. Note that RECON and EoREC messagesare not used in single reconfigurations.

An important point in dynamic reconfigurations oftwo communicating protocol stacks is that, every re-configuration causes communication inside the stacksto become limited or even blocked. For a synchronousreconfiguration of two peer stacks, this can lead tothe problem in communication of their RMCs duringthe synchronization. However, as explained, the RMCcomponents in the two stacks are synchronized in sucha way that peer RMCs do not require any communica-tion in the blocked period.

For single reconfigurations, the initiator RMC startsthe reconfiguration execution, as described in the nextsubsection.

3.3.3 Reconfiguration Execution

The reconfiguration execution includes installation, freeze,and re-execution steps. Figure 8 depicts the steps fora component replacement in the reconfiguration exe-cution phase. In the case of satisfaction of all pre-conditions for an assured reconfiguration, the RMCcomponent installs the new component (Figure 8-(b))and then, invokes the semiFreeze() method of thecurrently running component to start its semi-freezemode (Figure 8-(c)). In this mode, RMC monitorsthe component execution and freezes it by invoking itsstop() method in a SRP determined in the RECON

Figure 6. Reconfiguration type detection

Figure 7. The reconfiguration synchroniza-tion

8


message. Afterwards, the RMC uses saveState() andrestoreState() methods to transfer the state valuesin the old component’s PCB into the new component’sPCB (Figure 8-(d)). Later, the new component is re-executed through its start() method (Figure 8-(e)).Now the new component is available for its user com-ponents and the application can send data through thenew protocol.

Figure 8. Component replacement in the re-configuration execution

It is worth noting that when two peers enter intothe freeze mode, there maybe some packets inside thecommunication channel (packets that have been sentby a component but not delivered to the peer compo-nent yet). In this case, although the macro-states oftwo peers are compatible, their micro-states (run-timestates) are not consistent. If the old protocol is reli-able, after changing into another reliable protocol, thenew one can resend these packets. If the old or newprotocol is unreliable, these packets maybe lost in thecommunication.

4 Implementation and Evaluation

In this section, we evaluate the proposed frameworkfor automatic dynamic reconfigurations. Overhead ofusing wrappers in protocol stack performance has beenevaluated in related work such as [5, 19].

Java is chosen as the programming language dueto its platform independence. To load and reconfig-

ure protocol components at run-time, we use dynamicclass loading. The configuration of the experimentalenvironment includes two Centrino 1.5 and 1.7 GHzIBM personal computers with 256 and 512 MB mem-ory. Linux (Debian 3.1 distribution) is used as theoperating system.

Figure 9. The experimentation environment

Two communicating peer protocol stacks are consid-ered, Stack1 and Stack2 (Figure 9). Both stacks arebased on the DRAPS framework. Applications on thetop of stacks exchange data with each other. Bothapplications use a light-weight version of TCP pro-tocol, which we have implemented for the transportlayer. The IP layer is simulated using two Linux FI-FOs, one for outgoing data and the other for incomingdata. Wrappers for TCP and IP layers have also beenimplemented.

In our experiments, reconfiguration commands aresent to the RMC in Stack1 to initiate the reconfigu-rations. Each stack contains 5 role-compatible compo-nents with the running TCP, namely, TCP-CA (TCPwith Congestion Avoidance), SecureTCP (TCP withencryption capability), UDP, TCP-Nice, TCP-SACK.They have been implemented such that all their macro-states are SRPs. We have tested scenarios for FBC,PBC, and NBC role-compatibilities to demonstrate allcases of reconfiguration type detection.

Figure 10 shows duration of reconfiguration phases.Numbers in the chart are averaged over a large num-ber of iterations. The preparation phase takes 71 mil-liseconds; the synchronization phase, which includesthe times for sending and receiving START and RE-CON messages, takes 112 milliseconds; it depends onthe round trip time. The execution phase takes 56 mil-liseconds. The maximum total time for an automaticreconfiguration takes 239 milliseconds in average.

Figure 11 shows the performance of execution phaseof TCP (with empty buffer) reconfigurations accord-ing to the steps explained in the reconfiguration execu-tion. The measured times for component installation,PKB operations, semi-freeze, and freeze are shown in

9


Figure 10. Time spent in preparation, syn-chronization, and execution phases

Figure 11. Time spent in execution

the chart. PKB operations include the time for find-ing SRPs and the peer-compatible macro-state. Inaddition, it includes the time to verify assurance ofthe reconfiguration through examining the possibilityof complete initialization of new PCB. In our exper-iments, it takes around 3 milliseconds. According tothe definition of SRP, the semi-freeze time depends onthe processing time for input/output data in the TCPcomponent. For the implemented TCP, in which allmacro-states are SRPs, the semi-freeze time is 3 mil-liseconds.

The freeze time includes the time for state trans-fer, which takes up 17 milliseconds (5 milliseconds forcopying the old PCB into the new one, and 12 millisec-onds for invoking PCB-handlers). It is important tonote that, the freeze time is the only period in whichboth old and new protocol components are unavail-able. The total reconfiguration execution time, whichincludes installation, PKB operations, semi-freeze, andfreeze times, is 56 milliseconds in DRAPS.

Table 1 shows a comparison of DRAPS and two re-lated frameworks, which are implemented and evalu-ated in a similar environment with DRAPS. To be com-parable, we do not include preparation time, in whichwe check preconditions for an assured reconfigurationin calculating the time for a single or distributed re-

configuration. In DRAPS, the single reconfigurationtime is lower than the others because of using PCBand PCB handlers in state transfer. In [9], for statetransfer, authors use Java serialization technique [12],which takes around 190 milliseconds. In DRAPS, dis-tributed reconfigurations takes up 112 milliseconds forstacks synchronization and reconfiguration type detec-tion. Whereas, DPF [2] does not support reconfigura-tion type detection.

Table 1. Comparison of reconfiguration timesin the related frameworks (”-” = Not Sup-ported)

Framework/Time(ms) single distributedDPF [2] 77 200

Yueh-Feng Lee et. al. [9] 214 -DRAPS 56 168

For the TCP protocol, since the timeout of TCPsockets is usually more than ten seconds on differentmachines, the reconfiguration is performed transpar-ently from the application point of view. In general,it is important to demonstrate that, our frameworkcan transparently reconfigure other protocols (besidesTCP) as well. Therefore, the duration of the freezeperiod in a protocol should be less than the timeout ofthe protocol users (generally the upper layer protocolusers). As the timeout value for most of the commu-nication protocols is much more than a second, we canexpect that the presented framework can transparentlyreconfigure other protocols as well. In our experimentsfor the TCP protocol, the freeze period takes 5 mil-liseconds for the default size of “send buffer”, which ismuch less than the TCP socket timeout in the applica-tion layer.

5 Related Work

Ioana Sora et al. propose the “protocol buildingblock” description and protocol selection algorithmsin [21]. They provide automatic stack compositionthrough an algorithm to select building blocks in caseall specified features are provided and all dependenciesof selected components are satisfied. This paper offersprotocol stack compositions at deployment-time.

In [13], a framework called OPtIMA is introducedfor protocol stacks of software-radio based systems.The goal is to reconfigure protocol stacks built withinthe framework. OPtIMA presents the definition andprovision of a library of classes, which can be usedto build reconfigurable protocol stacks. However, the

10


framework does not support dynamic reconfigurationof protocols. It only presents customizibility of proto-col stacks at deployment-time.

The DiPS/CuPS framework [11] is intended for thedevelopment of customizable system software. This re-search mainly concentrates on a component model forprotocol components. Reflection points in DiPS/CuPS,which perform packet switching within a protocol stack,can only support run-time reconfigurations. Safe statesfor reconfigurations are defined based on protocol trans-actions; when the protocol execution is completed andbuffers are empty, the state is safe for a reconfiguration.Specifically, DiPS/CuPS presents dynamic reconfigu-rations for protocol stacks with idle protocols. It doesnot provide flexible enough mechanisms for reconfig-urations of running protocols (reconfiguration duringprotocol transactions).

Dynamic Protocol Framework (DPF) [2] mainly con-centrates on building an adaptive protocol stack byautomatic discovery and selection of protocol compo-nents. It supports on-the-fly reconfigurations of proto-col stacks and provides a synchronization mechanism toensure the compatibility of protocol stacks on commu-nication peers. In DPF, safe states for reconfigurationsare restricted to be at the end of protocol transactions.Moreover, supporting mechanisms for state transfer arenot clearly defined and addressed. While the automa-tion in DPF helps to select proper protocol compo-nents for a stack reconfiguration, DRAPS detects thereconfiguration types automatically. Based on the pro-vided services of protocol components, DPF selects andcomposes a proper protocol stack. In contrast, DRAPSuses PKB to select proper protocol components. More-over, through the synchronization, DRAPS detects thereconfiguration type and proposes proper componentsfor the peer stack as well. Whereas DPF does not sup-port protocol selection for the peer stack.

In [9], authors propose a Java-based framework thatallows programmers to create, remove, and replace pro-tocol modules at run-time. Programmers implementtheir components using the component framework. Theframework can dynamically reconfigure components insafe states. To find a safe state, the framework useslock management techniques to access a protocol mod-ule. When a reconfiguration command is received, theframework attempts to access the “write” lock to startthe reconfiguration. Lock management adds complex-ity in programming protocol modules and also in theapplication layer. In contrast, we have implementedlock management in wrappers and so there is no ex-tra overhead for the protocol programmer. For thestate transfer, authors use the Java synchronizationtechnique as a direct state transfer mechanism. How-

ever, we have used the PCB model for the indirect statetransfer that causes a short freeze time.

In the context of distributed reconfiguration of com-municating peers, there are quite a number of imple-mentations that support distributed reconfigurations.In [19], authors propose an algorithm and a model fordistributed reconfigurations of peers. A replacementmodule is responsible for reconfiguration control andmanagement. It provides indirect access to the chang-ing module, like the wrappers in DRAPS. In this pa-per, authors identify two properties of dynamically up-dateable systems, i.e. stack well-formedness and proto-col operationability. Preserving these properties dur-ing a dynamic update guarantees the transparent up-dates. An algorithm, which can switch between dif-ferent distributed agreement protocols (e.g., consensusand atomic broadcast), is proposed in that paper. Assafe states are at the end of protocol transactions, theyprovide no mechanisms for finding global states or statetransfer.

Maestro [23] supports only the replacement of wholeprotocol stacks; that is, in order to replace a protocolcomponent, the whole stack containing the componenthas to be replaced. They propose a stack switchingmodule installed on each machine to dynamically re-place stacks. The main role of the module is to syn-chronize the start of the new stack. Protocols are de-veloped within the Ensemble framework. They are de-composed into micro-protocols, each specialized to doa specific task. Since Maestro performs reconfigura-tions for the whole stack, the applications on top ofthe stack are blocked. In Maestro, a dynamic recon-figuration is started when the transaction of the oldprotocol has been completed and no data is exchangedbetween peers.

In DRAPS, dynamic reconfigurations can take placeduring the protocol transactions. Therefore, for longrunning servers DRAPS does not wait until the end ofprotocols executions. Distributed reconfigurations inDRAPS require freezing of both peer components ina safe global state; however, in related frameworks re-configurations take place at the end of protocols trans-actions and there is no need for safe global state. Theshortening of the freeze time in DRAPS is achievedthrough the PCB model, which incurs one-time de-velopment overhead for each protocol and presents anecessarily short time for the freeze period.

6 Conclusion

Ubiquitous computing environment includes auto-nomic networked systems requiring automatic dynamicreconfigurations in their protocol stacks. This paper

11


proposes a framework for automatic reconfigurationsof protocol stacks. Although dynamic reconfigurationis not new, the proposed framework is novel in that itsupports automatic reconfigurations of protocol stacks.We introduce two types of relationship between proto-cols, role-compatibility and peer-compatibility. Basedon these relationships, the type of reconfiguration isdetected and an assured reconfiguration is performedwithout the need for human intervention. Test sce-narios for automatic single and distributed reconfig-urations of TCP protocol in TCP/IP protocol stackare realized through the framework. The experimentalresults show that an acceptable transparency can bemaintained using the proposed framework.

References

[1] M. Allman, V. Paxson, and W. Stevens. Tcp conges-tion control. RFC 2581, 99(7):1–100, April 1999.

[2] L. An, H. K. Pung, and L. Zhou. Design and imple-mentation of a dynamic protocol framework. Journalof Computer Communications, 29(9):1309–1315, May2006.

[3] V. G. Bose, A. B. Shah, and M. Ismert. Softwareradios for wireless networking. In Proc. of INFOCOM1998, 3:1030–1036, 1998.

[4] S. Floyd. The newreno modification to tcps fast re-covery algorithm. RFC 2582, April 1999.

[5] N. Georganopoulos, T. Farnham, R. Burgess, T. S. J.Sessler, P. Warr, Z. Golubicic, F. Platbrood, B. Sou-ville, and S. Buljore. Terminal-centric view of soft-ware reconfigurable system architecture and enablingcomponents and technologies. IEEE CommunicationsMagazine, 42(4):100–110, May 2004.

[6] D. Gupta. On-line software version change. PhD the-sis, Department of Computer Science and Engineer-ing, Indian Institute of Technology, 1994.

[7] C. Hofmeister and J. M. Purtilo. Dynamic reconfigura-tion in distributed systems: Adapting software mod-ules for replacement. In Intl. Conf. on DistributedComputing Systems, (7):101–110, January 1993.

[8] M. Laddomada. Reconfiguration issues of future mo-bile software radio platforms. WIRELESS COMMU-NICATIONS AND MOBILE COMPUTING, 2:815–826, 2002.

[9] Y. Lee and R. Chang. Developing dynamic-reconfigurable communication protocol stacks usingjava. Software Practice Experience, 6(35):601–620,January 2005.

[10] M. Mathis, J. Mahdavi, S. Floyd, and A. Romanow.Tcp selective acknowledgment options. RFC 2018,Sun Microsystems, October 1996.

[11] S. Michiels, F. Matthijs, D. Walravens, and P. Ver-baeten. Dips: A unifying approach for developing sys-tem software. In A. D. Williams, editor, Proceedings -The Eight Workshop on Hot Topics in Operating Sys-tems, IEEE Computer Society, 2001.

[12] S. Microsystems. Java object serialization specifica-tion. 2001.

[13] K. Moessner, S. Vahid, and R. Tafazolli. Termi-nal reconfiguration: The optima framework. SecondInternational Conference on 3G Mobile Communica-tion Technologies, IEE-3G2001, pages 241–246, March2001.

[14] T. R. Newman, B. A. Barker, A. M. Wyglinski,A. Agah, J. B. Evans, and G. J. Minden. Cogni-tive engine implementation for wireless multicarriertransceivers. Wiley Journal on Wireless Communica-tions and Mobile Computing, 7(9):1129–1142, Novem-ber 2007.

[15] M. Niamanesh, F. H. Dehkordi, N. F. Nobakht, andR. Jalili. On validity assurance of dynamic reconfig-uration in component-based program. In IPM Inter-national Workshop on Foundations of Software Engi-neering (Theory and Practice) FSEN 2005, ENTCS,159(7):227–239, May 2006.

[16] M. Niamanesh and R. Jalili. Formalizing compatibilityand substitutability in communication protocols usingi/o-constraint automata. In Arbab, F., Sirjani, M.(eds.) FSEN07, LNCS, pages 49–64, April 2007.

[17] P. Patel, A. Whitaker, D. Wetherall, J. Lepreau, andT. Stack. Upgrading transport protocols using un-trusted mobile code. In Proc. of the 19th ACM Sympo-sium on Operating Systems Principles, pages 684–693,2003.

[18] J. Paula, A. Almeida, M. Wegdam, M. van Sinderen,and L. Nieuwenhuis. Transparent dynamic reconfig-uration for corba. In Proceedings of the 3rd Interna-tional Symposium on Distributed Objects and Applica-tions, 2001.

[19] O. Rutti, P. T. Wojciechowski, and A. Schiper.Structural and algorithmic issues of dynamic proto-col update. In the Proc. of the 20th IEEE Interna-tional Parallel and Distributed Processing Symposium(IPDPS06), April 2006.

[20] M. Satyanarayanan. Pervasive computing: Vision andchallenges. IEEE PCM, pages 10–17, January 2001.

[21] I. Sora, S. Michiels, and F. Matthijs. Policies fordynamic stack composition. Technical Report, Dept.Computer Science, Leuven, Belgium, November 2000.

[22] R. Stewart, Q. Xie, K. Morneault, C. Sharp,H. Schwarzbauer, T. Taylor, I. Rytina, M. Kalla,L. Zhang, and V. Paxson. Stream control transmis-sion protocol. RFC 2960, October 2000.

[23] R. van Renesse, K. Birman, M. Hayden, A. Vaysburd,and D. Karr. Building adaptive systems using ensem-ble. Software Practice and Experience, 28(9):963–979,February 1998.

[24] A. Venkataramani, R. Kokku, and M. Dahlin. Tcpnice: A mechanism for background transfers. In Proc.of the Fifth Symposium on Operating Systems Designand Implementation, December 2002.

12


A NOVEL STRATEGY TO PROVIDE SECURE CHANNEL OVER WIRELESS TO WIRE COMMUNICATION

Prof. Dr. Alaa Hussain Al- Hamami, Dr. Mohammad Alaa Al- Hamami Amman Arab University for Delmon University for Graduate Studies Science and Technology [email protected] [email protected]

ABSTRACT This research aims to provide secure channel for the communication over mobile network to access the internet. This could be done by divide the security between mobile phone and WAP. Mobile phone takes the plaintext from user then compresses it by Huffman compression method. After getting the compressed text then will be encrypted by elliptic curve taking in account the limitations of mobile phone. After transmitting the encrypted compressed text to WAP, WAP will translate the wireless communication to wire and will do the following procedures: decrypt and decompress the cipher text then get the plaintext for translating to internet. If the message will be transferred from the WAP to a mobile, the following process will be taken place: recompress and re encrypt the message by using elliptic curve. Finally the cipher text will be encrypted by either blowfish or twofish to strength channel security over the internet. The encrypted message will be transmitted with image hidden in it that contains information and the key of the used encryption algorithm (twofish or blowfish).

Keywords: Mobile devices, Wireless network, Encryption, Compression, and WAP.

1 INTRODUCTION

Mobile devices and wireless network is being used widely now days, wireless networks are available in most public places, this encourage the unauthorized used for those networks and devices. Wireless Access Protocol (WAP) is the protocol that allows Internet access from wireless devices and as more subscribers demand WAP services, the need for wireless Internet security will continue to grow. 1.1- WAP Philosophy WAP stands for Wireless Access Protocol, a general term used to describe the multi-layered protocol and related technologies that bring Internet content to mobile devices such as PDAs and cell phones [1]. Such devices are

referred to as thin clients because they have one or more constraints in the form of display, input, memory, CPU, or other hardware or usability limitations. The platform constraints and the slower (and more expensive) bandwidth of cellular and related networks make standard Internet protocols difficult to utilize. Using the growing set of WAP tools and protocols, however, the mobile Internet is quite capable tool. As previously stated, WAP refers to a wide range of technologies and protocols, all related to mobile Internet functionality. Many articles focus on the delivery of Wireless Markup Language (WML) content to mobile devices over a cellular or related technology network. However, the delivery of many protocols and technologies takes the same route-namely, through a WAP proxy server that bridges the gap between the wired Internet and the


mailto:[email protected]

wireless service provider's network as shown in Figure 1.

Figure (1): The WAP Gateway provides wireless networks with Internet access and optional content translation and filtering.

1.2- Steganography Philosophy [2]

Embedding information, which is to be hidden, into media requires two files. The first is the innocent-looking image that will hold the hidden information, called the cover media. The second file is the secret message that the information to be hidden as shown in Figure 2.

Figure 2: The Hidden Process in an Image.

A message may be plaintext, cipher

text, images, or anything that can be embedded in a bit streams. The most media use is an image for hiding information. When combined, the cover media and the embedded message this will form a stego-object product. A stego-key (a type of password) may also be used in hiding process and then later may be used to decode the message.

1.3- Elliptic Curve Philosophy

It unlike earlier cryptosystem, an elliptic curve works with a finite Abelian group formed by the points on an elliptic curve defined over a finite field. The point addition operation in Elliptic Curve Cryptography (ECC) is the counterpart of modular multiplication in RSA and multiple addition of point (scalar multiplication) is the counterpart of the modular exponentiation. Menezes-Vanstone Elliptic Curve Cryptography (MVEC) is a cryptosystem that has no analogues for Discrete Logarithm Problem (DLP). Once one has a curve and a point on it, one is sure to succeed in embedding data into the system. That is not true for the elliptic curve analogues of DLP. In this system the finite field Fp, the elliptic curve E, and the “base point” B Є E (preferably, but not necessarily a generator of the curve) are public information. Bob randomly chooses secret integer d (1<d<N, where N is the number of points of E) and publishes the point dB. If Alice wants to send the message M (as any two number) to Bob, she will choose a secret random integer e (1<e<N) and sends the pair [(c1, c2), eB], where (k1, k2) = edB , (m1, m2) = M and c1 = m1*k1 mod q , c2 = m2*k2 mod q [3, 4]. Bob will then multiply the second point in the pair by d to find d(eB) ((k1, k2) = deB) and computes the inverse of each number in this point ( i.e k1*k1’ = 1 mod q, and k2*k2’ = 1 mod q), and find the original message M = (m1, m2) as follows: m1 = c1*k1’ mod q, m2 = c2*k2’ mod q. 1.4- Blowfish and Twofish Philosophy [5, 6] Blowfish, a new secret-key block cipher, is proposed. It is a Feistel network, iterating a simple encryption function 16 times. The block size is 64 bits, and the key can be any length up to 448 bits. Although there is a


complex initialization phase required before any encryption can take place, the actual encryption of data is very efficient on large microprocessors. Blowfish key expansion converts a variable-length key of at most 56 byte (448 bits) into several subkey arrays totaling 4168 bytes. Twofish is a 128-bit block cipher that accepts a variable-length key up to 256 bits. The cipher is a 16-round Feistel network with a bijective F function made up of four key-dependent 8-by-8-bit S-boxes, a mixed 4-by-4 maximum distance separable matrix over GF(28), a pseudo-Hadamard transform, bitwise rotations, and a carefully designed key schedule. Twofish can be implemented in hardware in 14000 gates. 2 THE PROPOSED SYSTEM The proposed system presents a strategy to provide secure channel over mobile communication to the internet. This strategy will be explained in detail by the following steps: First Step: design the suitable WAP architecture for the strategy. See Figure 3 .

Figure 3 WAP infrastructures with specified

WAP location. Figure 3 explains the proposed architecture. This architecture consists of mobile phone in mobile communication and WAP which will be put in the nearest point to mobile since the security with mobile

communication is not strong because of the different limitations with mobile phones.

Second Step: mobile phone role. Mobile phone will take the plaintext which represents credit card information or any other specified account information, putting them in file and then compress the file before encryption, also because of the mobile phone limitation. See Figure 4 which displays the compression process using Huffman method.

Figure 4 Huffman compressions with mobile

phone. Then take the compressed file to encrypt it by elliptic curve, see Figure 5.

Figure 5: elliptic curve with mobile phone.

Using elliptic curve rather than RSA for the following reasons: The principle of attractive of ECC compared to RSA is that it appears to offer equal security for a far smaller bit size, thereby


reducing processing overhead. On the other hand, although the theory of ECC has been around for some time, it is only recently that products have begun to appear and that there has been sustained cryptographic interest in probing for weakness. Finally transmit the compressed and encrypted information to WAP server. Third Step: WAP server role. After WAP received the encrypted and compressed information, it will decrypt and decompress the information. Then WAP converts the information from the mobile communication protocol to wire protocol and then compress the information (Huffman) and encrypt the information by using elliptic curve. Until now the security is pure over the internet so the proposed system supports the secure channel (Huffman and elliptic curve) over internet by strong encryption algorithm like blowfish or twofish. Figure 6 represents the blowfish implementation.

Figure 6: the complete proposed

implementation of blowfish.

This means encrypting the compressed information by either twofish or blowfish. The encrypted information will be transmitted over the internet with an image (stego) which hides the type of encryption algorithm and 1000-bit. According to specific schema, it will take 448 bit for blowfish and 128 bit for twofish. This schema knows both WAP server and protected server. See Figure 7 which represents the file which contains the encryption information and

the image that contains the 600 bit which represents the pool of the blowfish or twofish keys.

Figure 7: WAP will send the encryption file and image hiding the keys of blowfish and

twofish The specific schema of the key: The 1000 bit will be selected randomly. For blowfish key 448:

• Divide these 1000 bit to ten 100 bit. • First 100 bit takes from 30 to 80 bit

and inverse them (50 bit). • Second 100 bit takes from 5 to 55 and

from 30 to 80 then exoring (XOR) them (50 bit).

• Third 100 bit takes first 10 bit, fourth 10 bits, sixth 10 bits, eighth 10 bit and the last 8 bit (48 bit).

• Fourth 100 bit take from 10 to 60 and from 40 to 90 anding (AND) them (50 bit)

• Fifth 100 bit takes from 20 to 70 bit and inverse them (50 bit).

• Sixth 100 bit take from 20 to 70 and from 50 to 100 then oring (OR) them (50 bit).

• Seventh 100 bit takes first 10 bit, fourth 10 bits, sixth 10 bits, eighth 10 bit and the last 10 bit. Each xored (XOR) with 0111001100 (50 bit).


• Eighth 100 bit take from 10 to 60 (50 bit)

• Ninth 100 bit take from 80 to 100 (20 bit)

• Tenth 100 bit take from 5 to 35 (30 bit)

The 1000 bit will be selected randomly. For twofish key 128:

• First 100 bit takes from 30 to 40 bit and inverse them (10 bit).

• Second 100 bit takes from 5 to 15 and from 10 to 20 then exoring (XOR) them (10 bit).

• Third 100 bit takes first 10 bit, fourth 10 bits and the last 8 bit (28 bit).

• Fourth 100 bit take from 10 to 30 and from 40 to 60 anding (AND) them (20 bit)

• Fifth 100 bit takes from 20 to 30 bit and inverse them (10 bit).

• Sixth 100 bit takes from 20 to 30 and from 25 to 35 then oring (OR) them (10 bit).

• Seventh 100 bit takes first 10 bit exoring (XOR) with eighth 10 bit (10 bit).

• Eighth 100 bit take from 72 to 82 (10 bit)

• Ninth 100 bit take from 82 to 92 (10 bit)

• Tenth 100 bit take from 34 to 44 (10 bit)

Forth Step: protected server role. Now the protected server will receive the encryption file and image that contains the hidden information. First extract hidden information which represents type of encryption algorithm and the 1000 bit to extract the key of blowfish or twofish, then decrypt the encryption by the blow or two and decrypt the elliptic curve. Finally decompress the resulted decrypted file by using Huffman method.

3 CONCLUSIONS • Placing a WAP gateway in the

mobile communication network that due to mobile limitations which make the mobile phone unable to use what is efficient of security algorithms.

• Adding blowfish or twofish will strength the secure channel over the internet.

• Using the keys of blow and two encryptions online by creating schema for extracting them from 1000 bit. These bits which are hidden in image by using steganography will make the proposed system more immune against the attackers.

4 REFERENCES

1. Korhonen J.; “Introduction to 3G mobile communication”, second edition, Artech house, INC., 2003.

2. Nile F. Johnson and Suhil Jajodia, " Steganalysis of Images Created Using Current Steganographic Software ", in Proceeding of Information Hiding - Second International Workshop, Springer -Verlage, April 1998.

3. “Standard for efficient cryptography; SEC1: elliptic curve cryptography”, certicom crop, [email protected], 2000.

4. Murari A.; “Software implementation of EEC”, Oregon state university, 2003.

5. Schneier B., and Kelsey J.; “Unbalanced fiestel network and block-cipher design”, {Schneier,Kelsey}@counterpane.com, 2000.

6. Schneier B.; “The blowfish encryption algorithm”, counterpane-internet-security, Inc., 2000.




mailto:Kelsey%[email protected]

SECURING ROUTE DISCOVERY IN MAODV FOR WIRELESS SENSOR NETWORKS

R.Shyamala*, Dr.S.Valli*** Lecturer, Department of Computer Science and Engineering,

University College of Engineering, [email protected]

** Assitant Professor, Department of Computer Science and Engineering, Anna University, Chennai-25.

[email protected]

ABSTRACT

The deployment of sensor networks for security and safety related environments requires securing communication primitives such as broadcast, multicast and point to point communication. Recent research shows that multicast provides high energy saving in IP supported wireless sensor networks (WSN). In this work, we investigate how to secure route discovery of the Multicast Ad-hoc on Demand Distance Vector (MAODV) protocol for WSNs. For the ad-hoc networks ARIADNE, SEAD, ARAN are some of the solutions, which cannot be directly applied to energy constrained WSNs. Here, we propose a secure route discovery of MAODV based on TESLA and one-way hash function. We have simulated our work using NS-2. The performance of the secured MAODV and the unsecured MAODV is compared.

Keywords – MAODV, TESLA, Multicast, packet delivery ratio.

1 INTRODUCTION

Wireless sensor networks [1] (WSN) consist of a large set of multi-functional, low cost, wirelessly networked sensor nodes. These sensor nodes have control components and communication functionality. Sensor nodes co-operate with each other and are deployed in environment monitoring, habitat monitoring, healthcare, home automation, traffic control and industrial system automation. Recent work [2][26][25] shows that rather than broadcast, multicast in sensor network saves power. Among various routing protocols [11] MAODV has a very good performance.

For the application of MAODV, a platform that supports IP functionality in WSNs is required. Since sensor networks have power consumption restriction, it is too complex to apply the full TCP/IP protocol stack with IPV6 functionality to them. The authors [8][25] describe a mechanism, which enables IP

addressing in sensor nodes in a 4G-environment. The IETE group has also created, LOWPANWG, which addresses IPV6 functionality for Personnel Area Networks (PAN) including WSNs. The SICS research group developed a new microprocessor, a WSN-oriented operating system called Contiki [7], which describes IP connectivity. The authors [25] developed a test bed with IP and multicast support in WSNs and concluded that it is feasible to apply source specific multicast to WSNs.

In sensor networks, adverse nodes can freely join the network, listen to and/or interfere with network traffic, which leads to network failures. So, it is necessary to design a security mechanism that can prevent attacks by insiders or outsiders for the tree based routing protocol, namely MAODV.

2 RELATED WORK


Recently, several works have addressed securing unicast routing protocols for ad-hoc networks. ARIADNE [15], Secure Efficient Ad hoc Distance vector (SEAD) [14], Authenticated Routing for Ad hoc Networks (ARAN) [18], Secure Ad hoc On-demand Distance Vector (SAODV)[9], and Byzantine-Resilient Secure Multicast Routing in Multi-hop Wireless Networks (BSMR) [6] are the works, which secure routing protocol.

ARIADNE [15] is based on Dynamic Source Routing (DSR); it describes end-to-end security for ad-hoc network routing. It makes use of the message authentication code (MAC) to authenticate routing table entries. It requires clock synchronization in the network since the security of DSR is based on TESLA. SEAD [14] secures the Destination-Sequenced Distance-Vector (DSDV) routing protocol and provides authentication-using TESLA. It requires a shared secret key between each pair of nodes. ARAN [18] uses limited time certificates. So, it requires a trusted certificate authority. This protocol increases latency and discovers optimal shortest paths. SAODV [9] combines digital signature and hash chain and gives a secure Ad hoc On-demand Distance Vector (AODV). It splits the message into mutable and non-mutable fields. Cryptographic signatures are used to authenticate non-mutable fields. A One-way hash chain is created for every route discovery process to secure a hop count field, a mutable field of the AODV message. This protocol requires a key management mechanism. In any on-demand routing protocol, route discovery plays an important role. So, we consider how to secure route discovery with minimal cryptographic primitives, such as hash function and symmetric encryption for wireless sensor networks.

The rest of the paper is organized as follows; section 3 describes the network model and system assumption used in the implementation. Section 4 describes how to secure route discovery of MAODV. Simulation results are given in section 5 and section 6 concludes the paper and suggests future work.

3 NETWORK MODEL AND SYSTEM ASSUMPTION

We assume that a network has a central base station and sensor nodes are densely deployed.

The base station has abundant computation resources. The base station has a private/public key pair; the public key of the base station is known to all the sensor nodes. For our system we require a trusted key distribution center (KDC). The base station acts as the KDC. It is trustworthy and cannot be compromised. The sensor nodes communicate using unreliable radio links. The nodes in the sensor network are resource constrained. Each node has a unique ID. The nodes communicate with each other by multihops. Since we use the TESLA-based certificate for secure route discovery, all the nodes have loosely synchronized clocks. The clock difference between any two nodes does not exceed the error rate (∆). All the nodes know the value of ∆. The time synchronization is maintained by the hardware. The network links are bi-directional.

We assume that an attacker attempts to tamper with some links to reduce routing information to its neighbor. An intruder can jam outgoing packets. An attacker may be a compromised node. If so, it will make use of all the cryptographic keys and it may cooperate with other attackers. We make no assumption on the number of intruders, their location and their radio range.

3.1 TeslaPerring et al [23] present an efficient source

authentication technique called TESLA. It uses pure symmetric cryptographic function and ensures asymmetric property. TESLA divides time into ‘n’ equal intervals. Each ‘n’ is assigned a corresponding key Kn. The sender appends MAC to the data generated at time interval n, using the secret key Kn. The receiver simply buffers the packets. After some ‘d’ time slot, the sender discloses the corresponding key seed Sn. Using a public one way function F’, the key Kn can be derived from Sn.

To create a key seed chain the sender chooses a terminal seed St and generates St-1 using a one-way function F. Thus, F (St) → St-1, F(St-1) → St-2, F(St-2) → St-3, F(St-3) ….→ S0. For key generation the key seed chain is used in the reverse order (i.e. it starts from S0). Therefore, F’ (Sn) → Kn . Once the user receives a TESLA seed Sn , he guarantees the authenticity by checking F(Sn )→ Sn-1. The authenticating key Kn

is generated using the one way hash function F’.


In our approach, TESLA is used in creating group membership and node membership certificate, which supports source authentication in the MAODV secure route discovery.

4 SECURE MAODV

The Multicast Ad-hoc On Demand Distance Vector (MAODV) routing protocol [27] constructs a shared multicast tree, which connects the group members. In addition, the MAODV helps the group members to get connected to the multicast tree through the forwarding nodes. Members can join or leave the group at any time. One of the group members is the Group Leader. It initiates a route request at first and initializes the group sequence to one. The Group Leader is responsible for maintaining the multicast sequence number. At regular intervals it generates Group Hello (GH) messages across the network to maintain the multicast sequence number. The multicast sequence number is used to check the freshness of the multicast group.

The Wireless Sensor Network [2] is deployed in a hostile environment. The communication and deployment mode of WSNs makes it insecure. Sensor networks consist of resource constrained devices, implementing a security mechanism. Thus, our proposed authentication framework has the following objectives

• An unauthorized node should not be able to participate in the MAODV routing.

• Non-Group member node should not be able to act as a group member.

• Non-tree node should not be able to impersonate a tree node.To achieve the above goals, this work

proposes an authentication mechanism in which the sensor nodes and base station need to have necessary credentials to participate in the MAODV protocol. All the four types of nodes are used in the implementation. The nodes can be

• A Group Member

• A Group Leader from one among the Group Members.

• A Multicast Tree Member

• A Non-tree Member

The given group certificate and node certificate are the elements of the authentication frame work;4.1 Group Certificate

Every group member has the privilege to initiate a route request to join a multicast tree. It has a Group Membership Certificate, which justifies that it belongs to a particular multicast group. To reduce the communication overhead

Group ID

Group Private Key

Life Time

SIGN Group Public

Key(….)2 Bytes 2 Bytes 2 Byte 2 Byte

Figure 1: Group Membership Certificate

in WSNs, we assume that a node can belong to only one group. The format of the group membership certificate [28] is shown in Fig. 1. It contains the group ID, Group Private Key, Lifetime of the certificate and the Group Public Key used for signing the certificate.

4.2 Node CertificateAll the sensor nodes, whether a group

member or non-group member in the network have a certificate called the node certificate. This certificate is issued by a trusted certificate authority (CA). In WSNs, the base station acts as a certification authority. The format of the node certificate is shown in Fig 2. Only nodes possessing node certificates are eligible to participate in routing. The base station pre-distributes the node certificate off line to all the sensor nodes in the network, before deployment. All the nodes know the public key of the Certificate Authority (CA). Thus, there is no need to contact the CA for verification of a node.

Node ID

Node Private Key

Life Time

SIGN Node’s

Public Key(….)2 Bytes 2 Bytes 2 Byte 2 Byte

Figure 2: Node Membership Certificate

4.3 Group LeaderThe Group Leader broadcasts the Group

Hello (GH) packets to all the valid group members. The previous credentials of the group leader are used to authenticate the GH messages.


Join Req

Request ID

ID destination

group

Group Membership Cert

Source Node Certificate

Start Time (ts)

SIGN source private

key (….)

Figure 3: Route Request Packet

RouteRepl

y

Request ID

ID group

ID source

node

Group sequence number

Destination Node certificate

SIGN destination

private key (……)E Source private key

( Dest ID, Tree key)

Figure 4: Route Reply Packet

Request ID

In Hop Node

Out Hop Node

Life time

Figure 5: Request table

4.4 Multicast tree MemberWhen a non-member joins the Multicast

tree, a shared tree is constructed if it is needed. The Group Leader is the root of the Multicast tree. It generates and securely distributes the tree key to the entire Multicast tree members. A node that has the valid tree key will reply the route request.

4.5 Secure Route DiscoverySecure route discovery has two stages. At

first, a source node broadcasts the route request and then it receives a route reply. The format of the RREQ Message is given in Fig 3. RREQ has seven fields; RREQ: (Join Req , Request ID, ID destination group ,

Group Cert, Source Node Certificate, Start Time, SIGN source private key (….) )The request ID is used to check the freshness of the request. The entire packet is signed using the source private key. A node initiates the Route Request Message [RREQ] in two situations;

Case i - If a group member ‘i’ wants to join a multicast group, then it prepares the RREQ packet and broadcasts it to all its one hop neighbors. If the receiver is a multicast tree node, then it applies validation of the source algorithm and sends a reply to the sender by checking its routing table. The routing table contains the control information of the destination. If the receiver is a non-multicast tree node, then it forwards the request by replacing the outer signature with its own signature and forwards the request to the neighbors. The nodes of the tree

Network link

Reply route

Multicast tree node

Forwarding node Node wanting to join the Multicast tree node

Figure 6: Secure Route Discovery in MAODV

maintain a dynamic request table. The format of the request table is shown in Fig 5.

Case ii - if a group member ‘i’ wants to send data to a destination group by creating the multicast tree with valid group members of that session, then it prepares the RREQ and replaces the Join Request field by creating a multicast tree field and the rest of the process is the same as in Fig 7.

k

J2

1

J1

i


RREQ : (Create Tree Request, Request ID, ID destination group , Group Cert, Source Node Certificate, Start Time, SIGN source private key (….) ). If a node ‘i’ does not receive a RREP message then it rebroadcasts the request ‘m’ times. Then, the node ‘i’ declares itself as the group leader by sending Group Hello Packets to all the group members.

4.6 Working of Secure Route DiscoveryIf a node ‘i’ wants to join a multicast tree,

the node will broadcast the route request packet to all its one hop neighbors. The non-group member forwarding the RREQ is the forwarding node. The RREQ has two signatures. The signature of the source node certificate is called the inner signature. The entire RREQ packet is signed by the forwarding node and it is called the outer signature.

Algorithm: Validation of the Source Node // tr denotes the time when the RREQ is received // ts denotes the issue time of the RREQ// D- propagation delay1. Assume node ‘k’ receives a RREQ at tr .

2. if ts +D<tr then 3. Discard the RREQ4. else 5. if ( forwarding node is a trusted node) then6. Buffer the RREQ packet and store forward node ID in request table.7. Wait (for group authenticating key to be disclosed.)8. if (source node ‘i’ has a valid group member certificate) then9. Send the RREP packet with the following details;10. Append the tree key and reply node ID 11. Encrypt with the source node public key.12. else13. Discard the RREQ14. end if15. else16. Report the unauthorized node to CA.17. endif18. endif

If a forwarding node ‘j1 ‘ gets the RREQ, then its checks the inner signature and stores the

sender’s ID as in-hop ID in its request table. The node ‘j1’ appends its certificate to the RREQ and broadcasts it. Suppose a forwarding node ‘j2‘gets the RREQ packet from node ‘j1‘then it checks the outer signature, inner signature and updates the request table. The node ‘j2’ adds its certificate by replacing the certificate of node ‘j1’ and broadcasts the RREQ. If a multicast tree member ‘k’ receives the RREQ then it removes the outer and inner signature. The node ‘k’ checks the routing table and sends the RREP packet. The format of the RREP is shown in Fig 4. The working of secure route discovery is given in Fig 7.

i→ * : [ RREQ ] ; // node ‘i’ broadcasts RREQ. j1 : [ RREQ ] ; // node ‘j’ receives the RREQ. // it checks inner signature and appends // the forwarding Node Certificate.j1 → * : [ RREQ , Node Certificate of j1 ]; j2 : [ RREQ , Node Certificate of j1 ]; // validates & replaces certificate of node j1.

j2 →* : [ RREQ , Node Certificate of j2 ]; k : Executes validation of the source node algorithm and sends RREP packet.k→ j2 : [ RREP ] j2 : [ RREP ];increments the hop count;j2 → j1 : [ RREP , Node Certificate j2 ] ; j1 : [ RREP , Node Certificate j1 ] ; j1 → i : [ RREP , Node Certificate ji ] ; i : // validates inner signature. Checks the //group sequence no. updates multicast // tree table, and request table

Figure 7: RREQ initiated by node ‘i’ and its subsequent actions to establish a secure route.

5 NETWORK PERFORMANCE

The network simulator NS-2 (http://www.isi.edu/nanam/ns/) has been used in evaluating the performance of the secure route discovery. Figure 8 shows the topology of the simulation. Table 1 gives the simulation parameters and Table 2 gives the default MAODV parameters.

The simulated scenario consists of a network with 30 nodes and two multicast groups A and B with 10 members each. All the group members


join the multicast group at the beginning of the simulation leading to the construction of the

Figure 8 : Topology of the Simulated Wireless Sensor network

multicast tree. The field ID is used in identifying the nodes. The sensor with ID zero is the base station, which is responsible for the distribution of certificates. The base station plays the role of the central TESLA certificate authority (CA). Table 2 lists the default values of the MAODV parameters used in the simulation. The node and group certificates in our simulation use a 512 bit public key and 16-byte signature as in [15].

Table 1: Simulation Environment Parameters

Parameters ValuesNumber of Nodes Pause Time Simulation Grid Size Radio Transmission Range Transmitting powerV Receiving powerSensing powerV Idle powerInitial energy

3050 sec500mX500m50 m0.175 mW0.145 mW0.00000175 mW0.0 mW5.0 Joule

Each simulation was run for 100 seconds. After 30 seconds, one group member (node 6) joins the group securely. For each set of experiments, 50 runs were performed. The average values are used in plotting the graphs

and these values are compared with the MAODV values [25].

Table 2: Default MAODV Parameters

Parameters ValuesNumber of Allowed Hello Loss Group Hello Interval Hello Interval Lifetime of Route Table Entries Max. No. of RREQ Retransmissions Max Time to Wait for a RREP

3 5 secs 1 secs 3 secs 35 secs

5.1 Performance MetricsThe packet delivery ratio, packet overhead,

average delay and average energy in the network are used as the metrics for evaluating the performance.a) Packet Delivery Ratio (PDR) is defined as the ratio of the total number of data packets received by the multicast group members to the product of the number of data packets sent and the number of group members.

The analysis shows that PDR is 40% high for secure MAODV, because RREQ message collision takes place while broadcasting. The certificate of the sender is appended along with the message. So, the packet size increases. b) Average delay in the network: In the simulated environment group A is distributed and group B is densely packed. So, the average delay in the network is 15 % more for group A when compared to group B. In real time scenario the WSNs are densely packed and hence cause less delay.

0.45

0.5

0.55

0.6

0.65

0.7

0.75

0.8

0.85

0.9

0.95

1

50 100 150 200 250

Time (ms)

Pa

ck

et_

Se

nd

x 1

03

S- MAODV

MAODV

Figure 8: Packet delivery ratio of Secure MAODV and MAODV.


0

50

100

150

200

250

300

350

400

450

500

50 100 150 200 250

Time (ms)

Av

era

ge

En

erg

y

S- MAODV

MAODV

Figure 9: Energy analysis of Secure MAODV and MAODV.

c) Energy Analysis: The average energy of the network is calculated as the sum of the energy of the sensor nodes divided by the number of nodes. The simulation shows that the average energy drops to 20% with an increase of 30 numbers of nodes.

Figure 10: Average delay of Secure MAODV and MAODV.

The disadvantage is that, it requires clock synchronization and the existence of a private/public key pair between base station and sensor nodes. To authenticate the TESLA key for each node, a one-way hash chain has to be maintained. TESLA keys are distributed to the participating nodes via the offline key distribution center.

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1.1

1.2

50 100 150 200 250

Time (ms)

Ave

rag

e E

ner

gy

x 10

3

group A group B

Figure 11: Average delay for member join in group A and B.

6 CONCLUSION

In this work, we have reduced the group membership and node membership certificate. We have introduced a request table, which stores the next hop and previous hop IDs. We have also reduced authentication overhead. All these features are well-suited for low-power devices such as sensor networks.

In this work, we have applied a TESLA-based certificate to secure MAODV for low power Wireless sensor networks. It is proven that we can apply Secure MAODV, which has a negligible impact on MAODV performance during the normal operation of the protocol.

Future EnhancementThe simulation study shows that the

TESLA-based Certificate with MAODV is possible with wireless sensor networks. Further, if we create a cluster-based multicast, the performance will greatly improve. The tradeoff between delay and overhead is a critical issue for further study.

REFERENCES

[1] I. Akyildiz W. Su, Y. Sankarasubramaniam and E. Cayirci: A Survey on Sensor Networks, IEEE Communications Magazine,” Vol. 40, No. 8, pp. 102-116, (2002).

[2] I. Akyildiz and I. Kasimoglu: “Wireless sensor and actor networks: research

0

50

100

150

200

250

300

350

400

450

50 100 150 200 250

Time (ms)

S- MAODV

MAODV

Average Energy x 10

3


challenges” Ad Hoc Networks , Vol 2(4): pp. 351-367, (2004).

[3] Chenyang Lu, Guoliang Xing, Octav Chipara, Chien-Liang Fok, Sangeeta Bhattacharya : A Spatiotemporal Query Service for Mobile Users in Sensor Networks, pp. 381-390, 25th IEEE International Conference on Distributed Computing Systems (ICDCS'05), pp. 381-390(2005)

[4] Perrig, A., Szewczyk, R., Tygar, J. D., Wen, V., and Culler, D. E: SPINS: security protocols for sensor networks. Wirel. Netw. 8, 5 (Sep. 2002), pp 521-534 (2002).

[5] Lidong Zhou; Haas, Z.J: Securing ad hoc networks, Network, IEEE, Vol 13, Issue 6, pp 24 – 30(1999).

[6] F. Stajano and R. J. Anderson, “The resurrecting duckling: Security issues for ad-hoc wireless networks,” in Seventh International Security Protocols Workshop, Vol 19,No 21, pp. 172–194 (1999).

[7] Contiki Operating System, http://www.sics.se/~adam/contiki/, (2005).

[8] A. Dunkels, T. Voigt and J. Alonso. Making TCP/IP Viable for Wireless Sensor Networks. In Proceedings of the First European Workshop on Wireless Sensor Networks (EWSN'04), work-in-progress session, Berlin, Germany, pp. (2004).

[9] Zapata, M. G.: Secure ad hoc on-demand distance vector routing. SIGMOBILE Mob. Comput. Commun. Rev. 6, 3 pp 106-107 (Jun. 2002).

[10] H. Luo, P. Zefros, J. Kong, S. Lu, and L. Zhang, “Self-securing ad hoc wireless networks,” in Seventh IEEE Symposium on Computers and Communications (ISCC ’02), (2002).

[11] S. Bhattacharyya, "An Overview of Source-Specific Multicast", RFC3569, (2003).

[12] B. Dahill, B. N. Levine, E. Royer, and C. Shields, “A secure routing protocol for ad-hoc networks,” Electrical Engineering and Computer Science, University of Michigan, Tech. Rep. UM-CS-2001-037, August 2001.

[13] J. Kong, H. Luo, K. Xu, D. L. Gu, M. Gerla, and S. Lu, “Adaptive security for multi-layer ad-hoc networks,” Special Issue of Wireless Communications and Mobile Computing, Wiley Interscience Press,( 2002).

[14] Y.-C. Hu, D. B. Johnson, and A. Perrig, “SEAD: Secure efficient distance vector routing for mobile wireless ad hoc networks,” in Proceedings of the 4th IEEE Workshop on Mobile Computing Systems and Applications (WMCSA 2002), pp. 3–13(2002).

[15] Hu, Y., Perrig, A., and Johnson, D. B. 2005. Ariadne: a secure on-demand routing protocol for ad hoc networks. Wirel. Netw. Vol 11, No 1-2 pp 21-38 (2005).

[16] S. Basagni, K. Herrin, E. Rosti, and D. Bruschi, “Secure pebblenets,” in ACM International Symposium on Mobile Ad Hoc Networking and Computing, pp. 156–163(2001)

[17] P. Papadimitratos and Z. Haas, “Secure routing for mobile ad hoc networks,” in SCS Communication Networks and Distributed Systems Modeling and Simulation Conference (CNDS 2002), (2002).

[18] Sanzgiri, K., Dahill, B., Levine, B. N., Shields, C., and Belding-Royer, E. M.: A Secure Routing Protocol for Ad Hoc Networks. In Proceedings of the 10th IEEE international Conference on Network Protocols, pp 78-89(2002).

[19] S. Buchegger and J.-Y. L. Boudec, “Nodes bearing grudges: Towards routing security, fairness, and robustness in mobile ad hoc networks,” in Proceedings of the Tenth Euromicro Workshop on Parallel, Distributed and Network-based Processing. Canary Islands, Spain: IEEE Computer Society, pp. 403–410(2002)

[20] Zhu Yufang, Kunz T.: MAODV Implementation for NS-2.26 (SCE-04-01). Ottawa:Department of Systems and Computing Engineering . Carleton University, (2004).

[21] NS-2: http://www.isi.edu/nsnam/ns/[22] SensorSim: NRL's Sensor Network

Extension to NS-2, Naval Research Laboratory.

[23] A. Perrig, R. Canetti, J. Tygar, and D. Song, “The TESLA Broadcast Authentication Protocol,” RSA CryptoBytes, 5, 2002.

[24] YANG Mingxi, LI Layuan, FANG Yiwei. “Securing Multicast Route Discovery for Mobile Ad Hoc Networks”, Journal of natural science,Vol.12, 189-192(2007).


http://www.isi.edu/nsnam/ns/

http://www.sics.se/~adam/contiki/

[25] Jorge Sá Silva, Tiago Camilo, Pedro Pinto, Ricardo Ruivo, André Rodrigues, Filipa Gaudêncio, Fernando Boavida. “Multicast and IP Multicast support in Wireless Sensor Networks”, JOURNAL OF NETWORKS, VOL. 3, NO. 3, pp. 19-26(2008).

[26] Camilo, T., Sá Silva, J.,Boavida, F., “IPv6 in Wireless Sensor Networks, a New Challenges”, First International Workshop on Convergence of Heterogeneous Wireless Networks, Julho 2005.

[27] E. M. Royer and C. E. Perkins. Multicast operation of the ad-hoc on-demand distance vector routing protocol. In Fifth Annual ACM/IEEE International Conference on Mobile Computing and Networking, Mobicom ’99, pp. 207–218 (1999).

[28] Bohge, M. and Trappe, W.: An authentication framework for hierarchical ad hoc sensor networks. In Proceedings of the 2nd ACM Workshop on Wireless Security, pp ,79 – 87(2003).

[29] David A. Maltz, Josh Broch, Jorjeta Jetcheva, and David B. Johnson. The Effects of On-Demand Behavior in Routing Protocols for Multi-Hop Wireless Ad Hoc Networks. IEEE Journal on Selected Areas in Communications, Vol. 17(8), pp. 1439–1453(1999).

[30] R. Szewczyk, E. Osterweil, J. Polastre, M. Hamilton, A. Mainwaring, and D. Estrin. Habitat monitoring with sensor networks. Communications of the ACM, Special Issue: Wireless sensor networks, Vol. 47(6), pp. 34–40(2004).

[31] Lorincz, K., Malan, D. J., Fulford-Jones, T. R., Nawoj, A., Clavel, A., Shnayder, V., Mainland, G., Welsh, M., and Moulton, S. : Sensor Networks for Emergency Response: Challenges and Opportunities. IEEE Pervasive Computing,Vol 3, No 4, pp 16-23(2004).


A Solution for Backward-Compatible Reconfigurations of RunningProtocol Components in Protocol Stacks

Mahdi [email protected]

Department of Computer EngineeringSharif University of Technology, Tehran, Iran

Rasool [email protected]

Department of Computer EngineeringSharif University of Technology, Tehran, Iran

Abstract

Forthcoming networked systems require mechanisms for on-the-fly reconfigurations in their protocol stacks tobe able to operate in different situations and networks. Backward-compatible reconfigurations of protocols arefast and easy ways for new protocol distributions. However, performing such reconfigurations at run-time andfor running protocol components, without disrupting peer components, is more desirable. This paper proposes asolution for dynamic reconfiguration management that can transparently reconfigure running protocol componentsin the middle of their protocol transaction. Mechanisms for the reconfiguration management including finding safestates as well as state transfer are proposed. For demonstration, we have implemented a prototype of the solutionto reconfigure a running TCP component. Our experimental results on dynamic reconfigurations show that anacceptable transparency (through providing a short time for the freeze period) can be maintained using the proposedsolution.


1 Introduction

Future communication and computation world, knownas pervasive computing environment, includes wirelessnetworks, networked systems and devices with hetero-geneous standards and protocols for different contextsand situations [25]. The Software Radio technology [3]offers dynamic reconfigurability for protocol stacks ofsuch systems and devices in order to facilitate applica-tions such as changing routing algorithms of switches,changing security modules in protocol stacks, bug fix-ing, and customizing protocol stack of a device for bet-ter performance.

In general, in a software system, dynamic reconfig-uration of a component to a new one includes suchphases as freeze (stopping the current execution of thecomponent), change (adding/binding a new componentand unbinding/removing the unnecessary old compo-nent from the system), state transfer (finding and ini-tializing a proper state in the new component in orderto resume the execution), and re-execution (resumingthe execution from a non-initial state in the new com-ponent) [19]. In order to have assured reconfiguration,the old component should be frozen in a “safe state”and the new component should resume the executionfrom a “reachable” state [10].

In the context of protocol stack reconfiguration, sinceeach protocol is defined at least between two peer com-ponents, reconfiguration of a running protocol compo-nent may require a corresponding reconfiguration inthe peer component(s). However, by backward-compatiblechanges of a protocol component, the protocol reconfig-uration can be carried out in the component indepen-dently without disrupting its peer component. As anexample, consider a reconfiguration that changes TCPcomponent in a running TCP/IP protocol stack intoTCP-Vegas component [4]. TCP-Vegas is backward-compatible with TCP and therefore the reconfigurationcan be performed in the stack independent of its peerstack.

There are some research activities for presenting dy-namic protocol stacks. Some of them, such as [18,29], provide reconfigurability at deployment-time (cus-tomizability) for protocol stacks. Some others includ-ing [16, 1] support reconfigurability at run-time for“idle” protocol stacks. In these works, safe states arepoints of protocols’ executions in which transactions ofthe corresponding protocol have been completed. How-ever, in long-running servers, having long and impor-tant connections (e.g., TCP connections), it is unfa-vorable to wait until the end of protocol transactions.A few approaches, such as [13], support dynamic re-configuration of “running” (not idle) protocol stacks.

They can reconfigure running protocols in safe statesthat can be in the middle of protocol transactions. Inthese approaches, reconfiguration of a protocol com-ponent should be transparent in the peer component’spoint of view.

In this paper, the reconfiguration problem is definedas changing one of the peer stacks at run-time transpar-ently. Unlike the related work, we can reconfigure run-ning protocol components in the middle of their proto-col transaction. For such a reconfiguration, we proposea procedure for reconfiguration management and con-trol. The procedure employs two ideas; we proposeevery protocol has a data structure for representingits state (called PCB); and protocol developers marksome states as safe states for staring possible reconfig-urations.

The rest of this paper is organized as follows. Sec-tion 2 describes backgrounds about protocol executionsand reconfiguration assurance in protocol stacks. InSection 3, we explain the proposed solution for backward-compatible reconfigurations in protocol stacks. Mech-anisms for assurance and also the reconfiguration pro-cedure are presented. In Section 4, we describe imple-mentation and evaluation of the solution. In Section5, we discuss related work and Section 6 concludes thepaper.

2 Background

In this section, we describe a simple model for proto-col execution and explain assurance in reconfigurationsof protocol stacks.

2.1 Protocol Execution

We consider a simple layered protocol stack modelto describe the reconfiguration problem for communi-cating protocol components. Fig. 1 shows two com-municating protocol stacks, (Stack1 and Stack2). Ineach layer of each stack, one independent entity whichwe refer to as protocol components (or in short compo-nents), provides functionality of its corresponding pro-tocol. For instance, components P1 and P2 in the fig-ure, provide functionality of the protocol P . We sup-pose components as an independent run-time entity.The protocol components interact with each other byexchanging protocol messages.

State of a protocol component describes all infor-mation related to the protocol in a point of execu-tion. Such information includes values of all variablesand contents of all input/output buffers, related to thecomponent. For simplicity, we split states of a pro-tocol component into macro-states and micro-states

2


Figure 1. An example of two communicatingprotocol stacks

[6]. Macro-states describe states of the protocols’ finitestate machines. Examples of macro-states in the TCPprotocol are CLOSED and ESTABLISHED. Micro-statesdescribe states of protocols at run-time, to maintain in-formation for operations such as reliability, error han-dling, and congestion control. We use macro-states ofcomponents in specification levels and micro-states inexecution levels.

2.2 Reconfiguration Assurance

One of the important reasons for the lack of practicaluse of reconfigurable component-based systems is deal-ing with assurance of reconfigurations [28]. Intuitively,a dynamic reconfiguration that changes a running com-ponent into a new one is assured if after the changingphase, the new component can be executed just as if ithas been executed from its initial state [10]. Based onthis notion, if the new component starts re-executionfrom a reachable state1, then the reconfiguration willbe assured. Based on [10], to achieve such a reachablestate, the running component should be frozen in a safestate, and its state should be transferred into the newcomponent. Requirements for such a safe state andstate transfer are described below.

Safe state A safe state for a reconfiguration has beendefined as a state having no interaction with theother components [11, 16]. A component in asafe state does not accept new requests, does notinitiate new operations, and all its initiated op-erations have been completed [23].

There are two types of algorithms to find safestates in an execution. Static algorithms, such

1State s in component C is said reachable state, if and only ifan execution of component C starting from an initial state canreach s at some time for some inputs.

as [12, 33], use knowledge of the system struc-ture to identify safe states. They always identifythe same set of safe states for a particular re-configuration. Dynamic algorithms, for example[9, 23, 5], use run-time knowledge such as com-ponents interactions to find safe states. Usually,dynamic algorithms disrupt small parts of a sys-tem than static algorithms.

State transfer Execution of the new component maybe resumed from a non-initial state, which we re-fer to as the restarting state. Two approaches ex-ist to transfer a state between two components,direct state transfer and indirect state transfer[31]. In the former approach, the new componentuses the implementation of the old component tointerpret and convert the state from the old com-ponent. In the latter approach, the old compo-nent exports its state in an abstract representa-tion form which is used by the new component.For the state transfer, firstly, it is necessary tofind the restarting state in the new component;secondly, the restarting state should be initializedto resume the execution.

In the following section, we explain the proposedsolution for backward-compatible reconfigurations ofrunning protocol components.

3 The Solution

In this section, we propose our solution for transpar-ent reconfigurations of running protocol components.First, we describe reconfiguration supports for proto-col components; then we explain required knowledgefor such reconfigurations; finally, we state the proposedsolution for the reconfiguration management and con-trol.

3.1 Reconfiguration Support for ProtocolComponents

Reconfigurable protocol components should providesome extra functionalities to support dynamic reconfig-urations. Every protocol component should implementthe ReconfigurableComponent interface. Fig. 2 showsthe required methods; saveState() and restoreState()methods are used for state transfer and start() andstop() methods are used to start and stop execution ofthe component. The semiFreeze() method should beimplemented for each component in order to supportfreezing the component in a proper state.

Moreover, dynamic reconfigurations of protocol com-ponents require indirect communication between two

3


public interface ReconfigurableComponent {void saveState(String path);void restoreState(String path);void start();void stop();void semiFreeze();

}

Figure 2. Reconfiguration interface for a pro-tocol component

Figure 3. A protocol component

adjacent protocol components (layers) in order to presenttransparent run-time reconfigurations [8]. For this rea-son, a wrapper component is used between two adja-cent protocol components. For a protocol component,the wrapper presents interface of the protocol compo-nent and manages the incoming requests to the compo-nent, by buffering them during the freeze periods (Fig.3.

3.2 Protocol Knowledge Base

As stated in the Subsection 2.2, an assured reconfig-uration of a protocol component necessitates to freezethe component in a safe state and to initialize the newcomponent completely. In order to achieve this goal,our idea is to keep enough information about reconfig-urable protocols in a Protocol Knowledge Base (PKB)component. Specifically, PKB contains two types ofknowledge, namely, role-compatible protocols, and pro-tocol state information (Fig. 4), as explained in thefollowing subsections.

3.2.1 Role-Compatible Protocols

In a reconfiguration of an existing component into anew component, the new one should play the same roleas the existing one. For example, we may reconfigureTCP into SCTP or UDP, but it is not possible to re-configure TCP into IP. In fact, the role of TCP and IPdiffers completely. Two protocols that can be reconfig-

ured to each other are called role-compatible protocols.A pair of protocol and its backward-compatible versionis a well-known example for role-compatible protocols.

For each pair of role-compatible protocols, PKB con-tains a state mapping function that maps correspond-ing macro-states of the two protocols. This helps recon-figurations as follows; when a protocol stops in a state(freeze state), through the state mapping function, acorresponding state (restarting state) can be found toresume the execution.

3.2.2 Protocol State Information

We present a model for state (micro-state) represen-tation of protocol components by introducing protocolcontrol block (PCB) that contains state information atrun-time. This information includes addressing param-eters, buffers (sent, received, un-acknowledged, etc.),protocol segment variables, counters, timers, send vari-ables, receive variables, and the other required param-eters. Protocol developers should implement one PCBdata structure for each protocol component in DRAPS.

Moreover, for protocols in one category (a proto-col and its extensions), we define basic PCB (BPCB)as an abstract PCB for the base protocol specifiedin standards and RFCs. For example, Fig. 5 showsthe parameters for BPCB in original TCP protocol(RFC793). PKB contains a set of PCBs’ and BPCBs’parameters and formats for different protocols.

PCB handlers are provided to set values of PCB pa-rameters. In state transfer, the old PCB is used tovaluate the same parameters in the new PCB. After-wards, the PCB handlers are used to complete the statetransfer. PKB includes PCB handlers for different re-configurations. Through PCBs and PCB-handlers weoffer indirect state transfer.

Figure 4. PKB includes PCBs and BPCBs for-mats, and knowledge of role-compatible pro-tocols

4


Variables (V)A. Addressing Variables

SRC: source portDST: destination port

B. Segment VariablesSEG.SEQ: segment sequence numberSEG.ACK: segment acknowledgment numberSEG.LEN: segment lengthSEG.WND: segment window (Receiver Advertised Window)SEG.CTL: control bits (ACK, RST, SYN, FIN)

C. Send Sequence VariablesSND.UNA: send unacknowledgedSND.NXT: send nextSND.WND: send windowISS: initial send sequence number

D. Receive Sequence VariablesRCV.NXT: receive nextRCV.WND: receive windowIRS: initial receive sequence number

E. TimersREXMT: Retransmission TimerTIMEWAIT: Time-wait TimerUSERTIME: User Timer

F. CountersdACK: duplicate ACK counterExpBoff: exponential backoff counter

G. OtherCurrState: Current StatePrevState: Previous StateRTO: Retransmission Timer Out valueRTT: Round Trip Time, used to calculate RTOSRTT: Smoothed RTT, used to calculate RTOCWND: Congestion windowMSS: Maximum Segment SizeSSthresh: Slow Start ThresholdMSL: Maximum Segment Lifetime

H. BuffersSND.Buff: Send BufferRCV.Buff: Receive BufferOO.RCV.Buff: Out of Order Receive BufferSND.UNAQ: Holds sent but unacknowledged segmentsUCallQ: Holds outstanding user calls(e.g., SEND, RECEIVE, CLOSE)

Figure 5. TCP parameters in its BPCB

3.2.3 TCP-SACK State Initialization

We give an example for building a PCB. TCP-SACK[15] is a selective acknowledgment extension for TCP2.It uses two TCP options, namely, SACK-PERMIT andSACK. These options are added to the TCP header.The first is an enabling option, which may be sentin a SYN packet to indicate the SACK option can beused once a connection is established. The second maybe sent over an established connection, once permis-sion has been given by SACK-permitted. The SACKoption is sent by a data receiver to inform the datasender that non-contiguous blocks of data have beenreceived and queued. The data receiver awaits the datareception in order to fill the gaps in the sequence spacebetween received blocks. When missing parts are re-ceived, the data receiver acknowledges the data by ad-vancing the left window edge in the AcknowledgementNumber Field of the TCP header.

PCB for TCP-SACK can be built by adding twoparameters, SACK-PERMIT and SACK, to the TCPBPCB. For state initialization in TCP-SACK, the TCPPCB is copied into the TCP-SACK PCB and the SACK-PERMIT parameter in the PCB is set through a PCBhandler. Since the SACK parameter is calculated basedon the received data, it is not necessary to be initial-ized.

3.3 Reconfiguration Management and Con-trol

Reconfiguration management and control is providedthrough a component, called RMC. It can receive re-configuration commands from different sources includ-ing system administrators, monitoring components inthe system, or peer systems’ RMC. A reconfigurationcommand asks replacing an old protocol componentwith a new one. Accordingly, RMC freezes the old com-ponent in a safe state, installs a new component, andinitializes it for re-execution from a restarting state.

3.3.1 Finding a Safe State

Due to the independence of protocol stack components(layers), finding a safe state for a dynamic reconfigura-tion can be carried out in a simple way. We define a safereconfiguration point (SRP) as a point of componentexecution where the component is at the beginning orat the end of the processing of an input packet. Inother words, the states just after writing a packet intothe output buffer or before reading a packet from the

2Although TCP-SACK requires both endpoint protocol com-ponents to change in order to be of value, this example onlyexplains state initialization.

5


input buffer are SRPs. In such points, either no opera-tion has been started on the packet or all the operationshave been completed. Usually, macro states of proto-cols are like this and we use them as SRPs. This defi-nition is consistent with the definitions in [11, 34, 16],where all operations of the component in SRP shouldbe completed. However, we do not restrict the com-ponent’s buffers to become free in SRPs. This impliesgreat flexibility in finding SRPs, in comparison withthe others such as [1]. In our solution, in order to finda SRP, it is not necessary to wait for the buffers tobecome empty.

Generally, finding a SRP in an execution of a com-ponent is a reachability problem, which is undecidable[10]. To cope with this problem, we ask protocol de-velopers to “mark” some SRPs in the source code ofprotocols. Therefore, protocol developers are asked toput some pieces of code in the source code of proto-col components to enable them to report macro-statesduring the execution if requested. In normal executionmode, a component does not report SRPs. However,we introduce semi-freeze mode in which the componentreports every SRP upon reaching (like in debug mode).The RMC component monitors the reported states ofthe component in the semi-freeze mode and stops itsexecution upon reaching the proper SRP for a dynamicreconfiguration.

3.3.2 The Reconfiguration Procedure

To simplify the problem of transparent reconfigurationsof a running protocol component, we assume there isonly one reconfiguration at a time and no concurrentreconfigurations can be initiated.

The flowchart of the reconfiguration procedure is de-picted in Fig. 6. There are three preconditions to starta reconfiguration. First, the old component shouldhave at least one SRP; second, mapping functions forthe SRPs of the old component should have been de-fined in PKB; third, new PCB can be completely ini-tialized through the old PCB and PCB handlers.

In the case of satisfaction of all preconditions, theRMC component starts installation and execution ofa new component. It installs the new component andinvokes the semiFreeze() method of the currently run-ning component to start its semi-freeze mode. In thismode, RMC monitors the component execution andfreezes it by invoking its stop() method in a deter-mined SRP. Afterwards, the RMC uses saveState(),and restoreState() methods to transfer the state val-ues in the old component’s PCB into the new compo-nent’s PCB in order to transfer the protocol state. Af-ter that, the new component is re-executed through its

start() method. From now, the new component isavailable for its user components and the applicationlayer can send data through the new protocol.

It is worth noting that, when the old protocol com-ponent enters into the freeze mode, there maybe somepackets inside the communication protocol stacks (pack-ets that have been sent by the peer component but notdelivered by the old component yet). These packets arebuffered in upper/lower layers of the frozen component.

Figure 6. Flowchart for the reconfigurationprocedure

4 Implementation and Evaluation

We have implemented a prototype of the RMC com-ponent3 to demonstrate feasibility of our solution intransparent reconfigurations of running protocol com-ponents. Overhead of using wrappers in protocol stackperformance has been evaluated in related works suchas [8, 24]. In this section, we describe the evaluation ofthe proposed mechanisms for assured and transparentdynamic reconfigurations. The evaluation includes thereconfiguration procedure and the overhead of findingsafe states and using PCBs in state transfer.

Java is chosen as the programming language dueto its platform independence. To load and reconfig-ure protocol components at run-time, we use dynamicclass loading. The configuration of the experimental

3An implemented prototype of the RMC is available athttp://mehr.sharif.edu/∼niamanesh/RG.htm.

6


environment includes two Centrino 1.5 and 1.7 GHzIBM personal computers with 256 and 512 MB mem-ory. Linux (Debian 3.1 distribution) is used as theoperating system.

Figure 7. The experimentation environment

Two communicating peer protocol stacks are con-sidered, Stack1 and Stack2 (Fig. 7). Stack1 is imple-mented using the proposed solution. Applications onthe top of stacks exchange data with each other. Bothapplications use a light-weight version of TCP protocol,which have been implemented for the transport layer.The IP layer is simulated using two Linux FIFOs4, onefor outgoing data and the other for incoming data. InStack1, wrappers for TCP and IP layers have also beenimplemented.

In the experiments, a reconfiguration command issent to the RMC in Stack1 to initiate a transparentreconfiguration. The command is to replace runningTCP with TCP-CA (TCP with Congestion Avoidance).TCP and TCP-CA are implemented such that all theirmacro-states are SRPs.

The TCP-CA PCB has two more parameters thatare congestion window size (CWND) and slow startthreshold (SSthresh). These parameters are evaluatedusing two PCB handlers to set their default values(CWND = 512 bytes, SSthresh = 65536 bytes)

Table 1. The performance of replacing TCPTime(ms)

Install 33PKB Operations 3

Semi-Freeze 3Freeze 17

Reconfiguration 56

Table 1 shows the performance of TCP reconfigu-4Using FIFO guarantees in-order sent or received packets;

however, this is not our point.

rations (with empty buffer) according to the steps de-signed in our reconfiguration procedure. All numbersare averaged over a large number of iterations. Themeasured times for component installation, PKB op-erations, semi-freeze, freeze, and the total reconfigu-ration time are shown in the table. PKB operationsincludes the time of finding SRPs, state mapping func-tions through PKBs, and the time to verify assuranceof the reconfiguration (through examining the possibil-ity of complete initialization of new PCB). In our ex-periments, it takes around 3 milliseconds. Accordingto the definition of SRP, the semi-freeze time dependson the processing time for an input/output data in theTCP component. For the implemented TCP, which allmacro-states are SRPs, the semi-freeze time is 3 mil-liseconds5.

The freeze time includes the time for state transfer.In the state transfer TCP-CA PCB is initialized basedon TCP PCB and two PCB-handlers. Restricting theTCP “send buffer” to become empty for starting a re-configuration, the state transfer takes up 17 millisec-onds (5 milliseconds for copying the old PCB into thenew one, and 12 milliseconds for two PCB-handlers).It is important to note that, the freeze time is the onlyperiod that both old and new protocol components areunavailable. The total reconfiguration time, which in-cludes installation, PKB operations, semi-freeze, andfreeze times, is 56 milliseconds.

We have not restricted buffers to become empty forstarting a reconfiguration; therefore, the overall statetransfer time depends on the amount of data in TCP“send buffer”. Based on our experiments state transferof TCP with full buffers of the size 91KB and 133KBtakes 43.5 and 71 milliseconds correspondingly.

Table 2 shows comparisons of our solution (RMC)and the two related works, which are implemented andevaluated in the similar environment. As shown inthe table, the freeze period in our solution is muchlower than the others. This is due to performing statetransfer through PCB and PCB-handlers. In compar-ison, [13] uses Java serialization technique [17] whichtakes around 190 milliseconds. This shortening of thefreeze time in our solution is achieved through the PCBmodel, which forces one-time development overhead foreach protocol and presents a short time for the freezeperiod, which is critical to be short.

For the TCP protocol, since the timeout of TCPsockets is usually more than tens of seconds on differ-ent machines6, the reconfiguration is performed trans-

5If a protocol component does not consider all its macro-statesas SRP, the semi-freeze period will be longer

6The default timeout of TCP socket is 5 days; however, ap-plications set a lower timeout value.

7


Table 2. The comparison of times of reconfig-uration steps in the related solutions (NA =Not Available)

Solution/ Finding State Recon. FreezeTime(ms) SRP Transfer PeriodDPF [1] NA NA 200 ≥ 200

[13] 3 190 214 ≥ 190RMC 3 5 56 5

parently in application point of view. In general, it isimportant to demonstrate that, our solution can trans-parently reconfigure other protocols (besides TCP) aswell. For this reason, the duration of the freeze periodin a protocol should be less than the timeout of the pro-tocol users (generally the upper layer protocol users).As the timeout value for the most of communicationprotocols is much more than a second, we can expectthat the presented solution can transparently reconfig-ure other protocols as well. In our experiments for theTCP protocol, the freeze period takes 5 millisecondsfor the default size of “send buffer”, which is much lessthan the TCP socket timeout in the application layer.

5 Related Work

Developing dynamic reconfigurable systems have beenreported in the literature, such as CONIC[14], ARGUS[2], and POLYLITH [11]; some others such as [7, 16]provide reconfigurable component models. In the con-text of reconfigurable protocol stacks, related work aremainly focused on implementing frameworks to sup-port reconfigurable protocol stacks. Reconfigurationsin protocol components can be either at deployment-time, at run-time with idle protocols, or at run-timewith running protocols.

Ioana Sora et al. propose the “protocol buildingblock” description and protocol selection algorithms in[26]. They use an algorithm to select building blocksin case that all specified features are provided and alldependencies of selected components are satisfied. Thispaper offers protocol stack compositions at deployment-time.

In [18], a framework called OPtIMA is introducedfor protocol stacks of software-radio based systems.The goal is to reconfigure protocol stacks built withinthe framework. OPtIMA presents the definition andprovision of a library of classes, which can be usedto build reconfigurable protocol stacks. However, theframework does not support dynamic reconfigurationof protocols. It only presents customizibility of proto-

col stacks at deployment-time.The DiPS/CuPS framework [16] is intended for the

development of customizable system software. This re-search mainly concentrate on a component model forprotocol components. Reflection points in DiPS/CuPS,which performs packet switching within a protocol stack,can only support run-time reconfigurations. Safe statesfor reconfigurations are defined based on protocol trans-actions; when the protocol execution is completed andbuffers are empty, the state is safe for a reconfiguration.Specifically, DiPS/CuPS presents dynamic reconfigu-rations for protocol stacks with idle protocols. It doesnot provide flexible enough mechanisms for reconfig-urations of running protocols (reconfiguration duringprotocol transactions). DiPS/CuPS supports both di-rect and indirect state transfer mechanisms.

Dynamic Protocol Framework (DPF) [1] mainly con-centrates on building an adaptive protocol stack byautomatic discovery and selection of protocol compo-nents. It supports on-the-fly reconfigurations of proto-col stacks and provides a synchronization mechanism toensure the compatibility of protocol stacks on commu-nication peers. In DPF, safe states for reconfigurationsare restricted to be at the end of protocol transactions.Moreover, supporting mechanisms for state transfer arenot clearly defined and addressed.

In [13], authors propose a Java-based frameworkthat allows programmers to create, remove, and re-place protocol modules at run-time. Programmers im-plement their components using the component frame-work. The framework can dynamically reconfigure com-ponents in safe states. To find a safe state, the frame-work uses lock management techniques to access a pro-tocol module. When a reconfiguration command is re-ceived, the framework attempts to access the “write”lock to start the reconfiguration. Lock managementadds complexity in programming protocol modules andalso in the application layer. In contrast, we have im-plemented lock management in wrappers and so thereis no extra overhead for the protocol programmer. Forthe state transfer, authors use the Java synchroniza-tion technique as a direct state transfer mechanism.However, we have used the PCB model for the indirectstate transfer that causes short freeze time.

6 Concluding Remark

This paper proposes a solution for dynamic recon-figurations of protocol stacks. Although dynamic re-configuration for protocol stacks is not new, our workis novel in that it supports assured and transparentdynamic reconfigurations for running protocol compo-nents. Unlike related work, we can reconfigure protocol

8


components during their transactions. Through model-ing protocols’ states by PCBs, we have achieved a veryshort time for the freeze period. This presents enoughtransparency of reconfigurations in the peer compo-nent’s point of view. We believe, PCBs and BPCBscan be added to protocols standards and RFCs; there-fore protocol developers can share PCBs and BPCBseasily.

For validation of the solution, we have implementeda prototype. Test scenarios for assured and transpar-ent reconfigurations of the TCP protocol in a TCP/IPprotocol stack are realized through the RMC compo-nent. The experimental results show maintaining of anacceptable transparency.

Future work can be carried out in several directions.Firstly, managing knowledge in PKB and specificationof protocol components require proper tools. Secondly,automatic generating of SRPs in the source code ofprotocol components is feasible. Lastly, providing au-tomatic detection and execution of transparent recon-figurations has a great importance in self-configuringof autonomic systems.

References

[1] L. An, H. K. Pung, and L. Zhou. Design and imple-mentation of a dynamic protocol framework. Journalof Computer Communications, 29(9):1309–1315, May2006.

[2] T. Bloom and M. Day. Reconfiguration and modulereplacement in argus: Theory and practice. IEE Soft-ware Engineering Journal, 2(8):102–108, 1993.

[3] V. G. Bose, A. B. Shah, and M. Ismert. Softwareradios for wireless networking. In Proc. of INFOCOM1998, 3:1030–1036, 1998.

[4] L. S. Brakmo and L. L. Peterson. Tcp vegas: end toend congestion avoidance on a global internet. IEEEJournal on Selected Areas in Communications, 8(13),October 1995.

[5] X. Chen and M. Simmons. Extending rmi to supportdynamic reconfiguration of distributed systems. In 22nd International Conference on Distributed Comput-ing Systems (ICDCS’02), 2002.

[6] D. Comer and D. Stevens. Internetworking with tcp/ipv2. Prentice-Hall, NJ, 1991.

[7] J. Dowling and V. Cahill. The k-component archi-tecture meta-model for self-adaptive software. In Lec-ture Notes In Computer Science, 2192:81–88, February2001.

[8] N. Georganopoulos, T. Farnham, R. Burgess, T. S. J.Sessler, P. Warr, Z. Golubicic, F. Platbrood, B. Sou-ville, and S. Buljore. Terminal-centric view of soft-ware reconfigurable system architecture and enablingcomponents and technologies. IEEE CommunicationsMagazine, 42(4):100–110, May 2004.

[9] K. M. Goudarzi. Consistency preserving dynamic re-configuration of distributed systems. PhD Thesis, Im-perial College, London, 1999.

[10] D. Gupta. On-line software version change. PhD the-sis, Department of Computer Science and Engineer-ing, Indian Institute of Technology, 1994.

[11] C. Hofmeister and J. M. Purtilo. Dynamic reconfigura-tion in distributed systems: Adapting software mod-ules for replacement. In Intl. Conf. on DistributedComputing Systems, (7):101–110, January 1993.

[12] J. Kramer and J. Magee. The evolving philosophersproblem: Dynamic change management. IEEE Trans-actions on Software Engineering, 11(16):1293–1306,1990.

[13] Y. Lee and R. Chang. Developing dynamic-reconfigurable communication protocol stacks usingjava. Software Practice Experience, 6(35):601–620,January 2005.

[14] J. Magee, J. Kramer, and M. Sloman. Constructingdistributed systems in conic. IEEE Transactions onSoftware Engineering, 6(15):663–675, 1989.

[15] M. Mathis, J. Mahdavi, S. Floyd, and A. Romanow.Tcp selective acknowledgment options. RFC 2018,Sun Microsystems, October 1996.

[16] S. Michiels, F. Matthijs, D. Walravens, and P. Ver-baeten. Dips: A unifying approach for developing sys-tem software. In A. D. Williams, editor, Proceedings -The Eight Workshop on Hot Topics in Operating Sys-tems, IEEE Computer Society, 2001.

[17] S. Microsystems. Java object serialization specifica-tion. 2001.

[18] K. Moessner, S. Vahid, and R. Tafazolli. Termi-nal reconfiguration: The optima framework. SecondInternational Conference on 3G Mobile Communica-tion Technologies, IEE-3G2001, pages 241–246, March2001.

[19] M. Niamanesh, F. H. Dehkordi, N. F. Nobakht, andR. Jalili. On validity assurance of dynamic reconfig-uration in component-based program. In IPM Inter-national Workshop on Foundations of Software Engi-neering (Theory and Practice) FSEN 2005, ENTCS,159(7):227–239, May 2006.

[20] M. Niamanesh and R. Jalili. A dynamic-reconfigurablearchitecture for protocol stacks of networked systems.31st Annual IEEE International Computer Softwareand Applications Conference (COMPSAC07), 1:609–612, July 2007.

[21] M. Niamanesh and R. Jalili. Formalizing compatibilityand substitutability in communication protocols usingi/o-constraint automata. In Arbab, F., Sirjani, M.(eds.) FSEN07, LNCS, pages 49–64, April 2007.

[22] M. Niamanesh, S. Sabetghadam, R. Y. Rahaghi, andR. Jalili. Design and implementation of a dynamic-reconfigurable architecture for protocol stack. InIPM International Workshop on Foundations of Soft-ware Engineering (Theory and Practice) FSEN07, InArbab, F., Sirjani, M. (eds.) FSEN07, LNCS, pages396–403, 2007.

9


[23] J. Paula, A. Almeida, M. Wegdam, M. van Sinderen,and L. Nieuwenhuis. Transparent dynamic reconfig-uration for corba. In Proceedings of the 3rd Interna-tional Symposium on Distributed Objects and Applica-tions, 2001.

[24] O. Rutti, P. T. Wojciechowski, and A. Schiper.Structural and algorithmic issues of dynamic proto-col update. In the Proc. of the 20th IEEE Interna-tional Parallel and Distributed Processing Symposium(IPDPS06), April 2006.

[25] M. Satyanarayanan. Pervasive computing: Vision andchallenges. IEEE PCM, pages 10–17, January 2001.

[26] I. Sora, S. Michiels, and F. Matthijs. Policies fordynamic stack composition. Technical Report, Dept.Computer Science, Leuven, Belgium, November 2000.

[27] R. Stewart, Q. Xie, K. Morneault, C. Sharp,H. Schwarzbauer, T. Taylor, I. Rytina, M. Kalla,L. Zhang, and V. Paxson. Stream control transmis-sion protocol. RFC 2960, October 2000.

[28] C. Szyperski. Component technology: what, where,and how? In Proc. of the Int. Conf. on Software En-gineering(ICSE), pages 684–693, 2003.

[29] S. K. Tan, Y. Ge, K. S. Tan, C. W. Ang, andN. Ghosh. Dynamically loadable protocol stacks - amessage parser-generator implementation. IEEE In-ternet Computing, 8(2):19–25, March 2004.

[30] R. van Renesse, K. Birman, M. Hayden, A. Vaysburd,and D. Karr. Building adaptive systems using ensem-ble. Software Practice and Experience, 28(9):963–979,February 1998.

[31] Y. Vandewoude and Y. Berbers. Component statemapping for runtime evolution. In Proceedings of the2005 International Conference on Programming Lan-guages and Compilers, pages 230–236, June 2005.

[32] A. Venkataramani, R. Kokku, and M. Dahlin. Tcpnice: A mechanism for background transfers. In Proc.of the Fifth Symposium on Operating Systems Designand Implementation, December 2002.

[33] I. Warren. A model for dynamic configuration whichpreserves application integrity. PhD thesis, LancasterUniversity, 2000.

[34] M. A. Wermelinger. Specification of software architec-ture reconfiguration. PhD thesis, Universidade Novade Lisboa, September 1999.

10


AMODIFIED IMAGE WATERMARKING USING SCALAR

QUANTIZATION IN WAVELET DOMAIN

Prof.Dr/ Mohiy Mohammed hadhoud #, Dr.Abdalhameed shaalan

*, hanaa abdalaziz abdallah*

# Faculty of computers and information, Menofia University, Shebin Elkom, EGYPT

*faculty of engineering, zagazig university, zagazig, EGYPT

E-mails:

[email protected], [email protected], [email protected]

ABSTRACT

A quantization based method for watermarking digital images is presented here.

This scheme inserting a watermark bit into wavelet coefficients of high magnitude

(LH3-HL3). Also, this watermarking technique is blind (i.e., neither the original

image nor any side information is required in the recovery process) as well as

being very computationally efficient. This watermarking algorithm combines and

adapts various aspects from more than one existing watermarking method. Results

show that the newly presented method improves upon the existing techniques.

Keywords: Watermarking Techniques, Digital Image Watermarking, Wavelet.

1 INTRODUCTION

This paper introduces a new quantization

based, blind watermarking algorithm operating

within the wavelet domain. The motivation for

this new algorithm was based upon various

aspects from more than watermarking

schemes .For example the new algorithm

improves upon the [1] algorithm in that it can

survive the same malicious attacks whilst

producing marked images of greater visual

quality. An improvement is made upon the semi-

blind used in [2] scheme as the new method does

not require a file containing the positions of the

marked coefficients (i.e., the new watermarking

scheme is blind).

2 BACKGROUNDS

Previously, algorithm in [1] presented an

additive watermarking method operating in the

wavelet domain. A three level wavelet transform

with a Daubechies 8-tap filter was used; no

watermark was inserted into the low-pass

subband. Unlike some non-blind watermarking

schemes [3, 4], this scheme allowed a watermark

to be detected without requiring access to the

original image (i.e., it is a blind watermarking

system).this scheme also performed implicit

visual masking as only wavelet coefficients with

a large enough magnitude were selected for

watermark insertion. Wavelet coefficients of

large magnitude correspond to regions of texture

and edges within an image. This has the effect of

making it difficult for a human viewer to

perceive any degradation to an image marked via

this scheme. Also, because wavelet coefficients

of large magnitude are perceptually significant, it

is difficult to remove the watermark without

severely distorting the marked image. The most

novel aspect of this scheme was the introduction

of an image sized watermark consisting of

pseudorandom real numbers. However, only a

few of these watermark values are added to the

host image. Using an image sized watermark

fixes the locations of the watermark values; thus,

there is no dependence on the ordering of

significant coefficients in the correlation process

for watermark detection. This is advantageous as

the correlation process is extremely sensitive to

the ordering of significant coefficients and any

change in this ordering (via image

manipulations) can result in a poor detector

response. Another watermarking algorithm

operating upon significant coefficients within the

wavelet domain (implemented via 5/3 taps

symmetric short kernel filters) was presented by

algorithm in. [2]. This method takes a three level

wavelet transform of the image to be

watermarked and inserts the watermark into the

detail coefficients at the coarsest scales (LH3,

HL3); the low pass component LL3 and diagonal

HH3 are excluded).The [2] scheme is a

quantization based watermarking technique

which aims to modify wavelet coefficients of

high magnitude thus embedding the watermark

into edge and textured regions of an image. The

quantization process used by this scheme is very

straightforward and simple to implement as it

requires a file to be saved detailing the locations


of where the watermark bits were embedded. It is

thus a semi-blind scheme as opposed to a blind

scheme.

3 ADVANTAGES AND DISADVANTAGES OF

THE PREVIOUS WATERMARKING

ALGORITHM

The algorithm in [1] has three main

advantages:

1. It is a blind algorithm,

2. It incorporates implicit visual masking,

thus, the watermark is inserted into the

perceptually significant areas of an image via a

simple and straightforward process.

3. It uses an image sized watermark to

negate the order dependence of significant

coefficients in the detection process.

There are two main disadvantages to the

algorithm:

(1) It embeds the watermark in an additive

fashion. This is a drawback as blind detectors for

additive watermarking schemes must correlate

the possibly watermarked image coefficients with

the known watermark in order to determine if the

image has or has not been marked. Thus, the

image itself must be treated as noise which makes

detection of the watermark exceedingly difficult

[5]. In order to overcome this, it is necessary to

correlate a very high number of coefficients

(which in turn requires the watermark to be

embedded into many image coefficients at the

insertion stage). This has the effect of increasing

the amount of degradation to the marked image.

(2) The detector can only tell if the watermark is

present or absent. It cannot recover the actual

watermark.

4 PROPOSED TECHNIQUE

It is possible to use the advantages of the [1]

watermarking schemes whilst removing the

disadvantages. This can be achieved by using its

idea of an image sized watermark in conjunction

with adapted versions of scalar quantization

insertion/detection techniques. The resultant

watermarking system will be blind and

quantization based. It will employ a watermark

equal in size to the detail subbands from the

coarsest wavelet level and only perceptually

significant coefficients will be used to embed

watermark bits. In summary, this new technique

improves upon the previous method by using

quantization and replaces insertion process

(rather than an additive insertion process). Thus,

for comparable robustness performance, the new

method will produce watermarked images with

less degradation than the [1] scheme. It improves

upon the [2] scheme by having no need for a

position file in the recovery process. A flow

diagram detailing the necessary steps is shown in

Figure 1.

4.1 Embedding

The watermark embedding process:

1. Transform the host image into the

wavelet domain (3rd

level Daubechies

wavelets of length 4).

2. The coefficients in the third wavelet level

(excluding the LL and HH subbands) with

magnitude greater than T1 and magnitude

less than T2 are selected.

3. Let fmax is the maximum absolute wavelet

coefficient of a set of subbands to adjust

the significant threshold T= α. fmax Where

.01 <α <0.1. And T2 > T1 >T.

4. A binary watermark the same size as the

entire third level of the wavelet transform

is created using a secret key (which is a

seed to a random number generator).

5. The selected wavelet coefficients are then

quantized in order to embed a watermark

bit. The value that the selected

coefficients are quantized to depends upon

whether they are embedding a 1 or a 0.

6. A selected wavelet coefficient,𝑤𝑖𝑗𝑠 will

embed a 1 if the value in the watermark

file at the same location 𝑥𝑖𝑗 , is 1.

Alternatively, 𝑤𝑖𝑗𝑠 will embed a 0 if 𝑥𝑖𝑗 is 0.

Thus, The quantization method is similar

to that used by[2]:

If xij = 1 and wijs > 0, then wij

s = T2 – X1,

If xij = 0 and wijs > 0, then wij

s = T1+ X1,

If xij = 1 and wijs < 0, then wij

s = -T2+ X1,

If xij = 0 and wijs < 0, then wij

s = -T1- X1,

7. The X1 parameter narrows the range

between the two quantization values

of T1 and T2 in order to aid robust

oblivious detection.

8. After all the selected coefficients have

been quantized, the inverse wavelet

transform is applied to all the wavelet

coefficients and the watermarked image is

obtained.

4.2 Detection

For oblivious detection:

1. The wavelet transform of a possibly

corrupted watermark image is taken.

2. Then all the wavelet coefficients of

magnitude greater than or equal to T1 + X2

and less than or equal to T2 – X2 are

selected; these shall be denoted via 𝑤𝑖𝑗′𝑠 .

Note that X2 should be less than X1.


3. In the insertion process, all wavelet

coefficients with a magnitude greater than

T1 and less than T2 are selected and then

quantized to either T1 + X1 or T2 – X1.

4. In the recovery process, all the wavelet

coefficients of magnitude greater than or

equal to T1 + X2 and less than or equal to

T2 – X2 are selected to be dequantized.

This helps ensure that all the marked

coefficients are recovered and dequantized

after being attacked.

5. Unmarked coefficients are unlikely to

drift into the range of selected coefficients

after an attack. The introduction of the X1

and X2 parameters to the watermarking

algorithm gives a degree of tolerance to

the system against attacks, i.e., they

collaborate to give a noise margin

watermark bit is decoded for each of the

selected wavelet coefficients via the same

process described by [2]:

If 𝑤𝑖𝑗′𝑠 < (T1 + T2)/2, then the recovered

watermark bit is a 0.

If 𝑤𝑖𝑗′𝑠 ≥ (T1 + T2)/2, then the recovered

watermark bit is a 1

6. The recovered watermark is then

correlated with the original copy of the

watermark file (obtained via the secret

key) only in the locations of the selected

coefficients. This allows a confidence

measure to be ascertained for the presence

or non-presence of a watermark in an

image.

4.3 Perceptual quality metrics

The aforementioned limitations of pixel based

image quality metrics helps argue the case for

quality metrics based upon the HVS. In recent

years, there has been an increase in the amount of

these metrics published. Two such metrics were

presented by Lambrecht et al. [6] and Watson [7].

The Lambrecht metric was described by Kutter et

al as a fair and viable method for determining the

amount of degradation suffered by a watermarked

image. It makes use of coarse image

segmentation and alters banks to examine

contrast sensitivity as well as the masking

phenomena of the HVS. This scheme then returns

an overall measure of distortion for the

watermarked (modified) image compared to the

un-watermarked (original) image. The Watson

metric was incorporated into the Checkmark

package [8] in order to help determine the visual

quality of a watermarked image. It operates

within the DCT domain and utilizes contrast

sensitivity, luminance masking and contrast

masking in order to calculate a Total Perceptual

Error (TPE) value between the watermarked and

un-watermarked images.

4 RESULTS

This section outlines the results obtained by

[1], scheme, [2] scheme and the newly proposed

scheme. It is the aim of the new scheme to be as

robust as the [1] scheme without degrading the

marked images to the same extent. This newly

proposed blind scheme improves upon the semi-

blind [2] scheme as it does not require a file

containing the locations of the coefficients that

were marked. In order to measure the degradation

suffered by host images after watermark insertion,

the Peak Signal to Noise Ratio (PSNR) metric

and the Watson Metric [9, 10] are used. The

Watson Metric computes the Total Perceptual

Error (TPE) which is an image quality metric

based upon the Human Visual System (HVS). It

takes contrast sensitivity, luminance masking and

contrast masking into account when calculating a

perceptual error value. Unlike the PSNR, this

merely measures the differences between pixels

without considering the HVS. The higher the

TPE value, the more degraded an image would

appear to a human viewer. The Checkmark

package [11] was used to determine the TPE. The

original and recovered messages were compared

by computing the Normalized Correlation (NC):

𝑁𝐶 =𝑚 ∗.𝑚

𝑚∗ . 𝑚 (1)

Where m is the original message and 𝑚∗ is the

recovered message (convert unipolar vectors, m €

{0, 1}, to bipolar vectors, m € {−1, 1}, in this

equation).For all the tests in this paper,

MATLAB 6.5 was used. JPEG compression was

carried out via the imwrite function which uses

the Independent JPEG Group’s. All tests were

performed upon an 8-bit (grayscale), 256 × 256

baboon image.

5.1 Results of technique used in [1]: T1 = 40, T2 = 50 and α = 0.2 (the same

parameters that were used in our paper). The

watermarked image was then attacked with JPEG

quality5 (Q5), quality 10 (Q10) and quality 15

(Q15), Gaussian noise addition (σ^2 = 375) and

cropping (from rows 60 to 190 and from columns

60 to 190) on average, it can be seen that scheme

is surviving all the attacks. However, for the

JPEG quality 5 and half sizing attacks, the

watermark was not always detected. . Results are

shown in Table 1.

5.2 Results of technique used in [2]: This algorithm was encoded and the

following parameters were set: T1 =50, T2 = 120,

we find the results are better than that of [1]

because it is semi-blind technique. Results are

shown in Table 2.


Figure 1. The blind quantization based watermarking scheme. The top part shows the insertion process whereas

the bottom part shows the detection process.

Table 1. Results of [1], T1 = 40, T2 = 50 and

α = 0.2. Each test was upon the baboon image.

Table 2. Results of [2]. T1 = 50 and T2 = 120.

Each test was done upon the baboon image.

NC PSNR(dB) TPE

No attacks 1 39.57 0.0259

JPEG Q5 0.526882 22.13 0.3432

JPEG Q10 0.942652 23.82 0.3017

JPEG Q15 1 24.82 0.2692

Gaussian 0.935484 22.36 0.3758

Salt&pepper 0.978495 23.71 0.2665

Half sizing 0.706093 21.57 0.3785

(a) (b)

Figure.2 (a) Baboon image marked using watermarking

scheme in [1] (b) Baboon image marked using

watermarking scheme in [2].

5. 3 proposed technique results

The same attacks were used to test the new

algorithm.T1 = 115, T2 = 200, X1 = 20 and X2 =

10 were the parametric values used; Figure 3

shows this watermarked image and the effect of

attacking this watermarked image with various

attacks. Table 3 presents the quantitative results

for these various attacks. An analysis of the

probability of obtaining a false positive detector

response is studied in Table 4. In this study, the

recorded normalized correlation (NC) value and

the recorded recovered watermark length were

saved. Using these values, it is possible to

calculate the probability of obtaining a false

positive reading [12]. The probability of

obtaining a false positive Pfp reading is calculated

via:

𝑃𝑓𝑝 = 𝑁𝑤𝑛

0.5𝑁𝑤

𝑁𝑤

𝑛=𝑁𝑤(𝑇+1)/2 (2)

Where Nw is the length of the recovered

watermark and T is the chosen detector threshold

value. This is a worst case scenario of obtaining a

false positive detection as the NC value and the

recovered watermark length are being used in the

calculation. The detector threshold value of 0.4

was selected to determine the presence or non-

presence of a watermark. This value means that

the new algorithm is highly robust to JPEG

quality 10 attacks, Gaussian noise ( 𝜎2 = 375)

attacks and half sizing attacks. Also, the chance

of obtaining a false positive reading after

suffering one of these attacks is extremely remote.

However, the “cropping" attack poses a problem

NC

Chosen

threshold

Det.

PSNR

(dB)

TPE

No attacks 0.429889 7.157312 36.68 0.057164

JPEG Q5 0.145309 8.20676 22.079 0.344422

JPEG Q10 0.22643 7.91128 23.76 0.2992

JPEG Q15 0.264691 7.91188 24.75274 0.26493

Gaussian 0.189755 7.1900 22.3604 0.37091

Salt& pepper

0.213674 7.440845 23.82075 0.2453117

cropping 0.187535 9.275418 7.365388 0.673658

Half sizing 0.118576 7.15731 21.46714 0.382016

Recovered watermark

N x N

Input image

Pick all high pass coefficients

in the third wavelet level

(LH3-HL3) of magnitude greater than T1 AND less than

T2

Embed watermark at these locations via quantization N x N

Watermarked image

N x N Binary watermark (equal

in size to wavelet level 3(LH3-

HL3)

N x N

Watermarked image

Pick all high pass coefficients

watermark

In the third wavelet level of magnitude greater than T1+X2

and less than T2−X2

Dequantize the coefficients at these locations to obtain the

recovered watermark.

Dwt (3levels)

Owner seed

Dwt

IDWT


in that, only 47 out of a possible 91 watermark

bits were used by the detector, thus decreasing

the reliability of the scheme. This lower number

of recovered watermark bits leads to a greater

chance of a false positive reading than the other

survived attacks (see Table 4). The scheme is not

robust to JPEG quality 5 attacks (just like the [1]

method).Thus, while surviving the same attacks

as the [1] scheme, the new scheme does not

degrade the watermarked image to the same

extent. From Table1, PSNR value is 36.68dB and

the TPE is 0.057164. The PSNR (39.57dB) and

the TPE 0.0259) values recorded for the [2]

scheme and PSNR value is (46.60dB) and the

TPE is (0.00771) values recorded for the new

scheme which is the better one.

Table 3. Results for the proposed scheme (with

T1 = 115, T2 = 200, X1 = 20 and X2 = 10).

Table 4. Probability of false positive detector

response for the new scheme (with T1 = 115, T2 =

200, X1 = 20 and X2 = 10).

6 CONCLUSIONS

The proposed watermarking scheme using the

method in [1] of determining the positions of

marked confidents (via an image sized/subband

sized watermark) in collaboration with adapted

versions of the [2] insertion and detection

techniques (using the notion of noise margins)

has been presented. The new method is superior

to the [1]method in that it can survive the same

attacks with higher detection (NC) and

producing marked images of higher visual quality

(measured via the PSNR and Watson metric

quantitative techniques). Although the robustness

of this new scheme is not quite as strong as that

presented by [2], this can be attributed to its blind

nature compared to the semi-blind nature of the

[2] method but it gives higher visual quality.

7 REFERENCES

[1] R. Dugad, K. Ratakonda and N. Ahuja, A new

wavelet-based scheme for watermarking

images,Proc. IEEE Intl. Conf. on Image

Processing, ICIP’98,Chicago, IL, USA, Oct.

1998, 419-423.

[2] H., A. Miyazaki, A. Yamamoto and T.

Katsura,A digital watermarking technique based

on the wavelet transform and its robustness on

image compression and transformation, IEICE

Trans., Special Section on Cryptography and

Information Security,E82-A, No. 1, Jan. 1999, 2-

10.

[3] I. J. Cox, F. T. Leighton and T. Shamoon,

Secure spread spectrum watermarking for

multimedia, IEEE Trans. on Image Processing,

Vol. 6, Dec. 1997, 1673-1678.

[4] M. Corvi and G. Nicchiotti, Wavelet-based

image watermarking for copyright protection,

Scandinavian Conference on Image Analysis,

SCIA ’97, Lappeenranta,Finland, June 1997, 157-

163.

[5] P. Meerwald, Digital image watermarking in

the wavelet transform domain, Master’s thesis,

Department of Scientific Computing, University

of Salzburg, Austria, 2001.

http://www.cosy.sbg.ac.at/˜pmeerw/Watermarkin

g/

[6] C. J. van den B. Lambrecht and J. E. Farrell.

Perceptual quality metric for digitally coded color

images. In Proceedings of EUSIPCO, Trieste,

Italy, September 1996.

[7] A. B. Watson. DCT quantization matrices

visually optimized for individual images. In J. P.

Allebach and B. E. Rogowitz, editors, Human

Vision, Visual Processing,and Digital Display IV,

volume 1913, pages 202{206, San Jose, CA,

USA, February 1993. SPIE.

[8] Checkmark benchmarking project [online].

Available from World Wide Web (date accessed:

December, 2004):

http://watermarking.unige.ch/Checkmark/.

[9] A. B.Watson, DCT quantizationmatrices

visually optimized for individual images, Human

Vision, Visual Processing and Digital Display IV,

Proc. SPIE, Vol.1913, San Jose, CA, USA, Feb.

NC

WM

length

in

WM

lengt

h out

PSNR

(dB)

TPE

No

attacks

1 91 91 46.60 0.0077

1

JPEG Q5 0.146667 91 75 22.14 0.3421

JPEG

Q10

0.480519 91 77 23.81 0.2981

JPEG Q15

0.853659 91 82 24.80 0.26432

Gaussian 0.544304 91 79 22.37 0.3707

55

Salt&pepper

0.794872 91 78 23.88 0.253507

cropping 0.489362 91 47 7.37 0.6730

94

Half

sizing

0.395 91 77 21.58 0.378

NC

WM

length out

Worst case Pfp

No attacks 1 91 000000

JPEG Q5 0.146667 75 0.1240228

JPEG Q10 0.480519 77 0.000014

JPEG Q15 0.853659 82 0.00000

Gaussian 0.544304 79 0.00000063484

Salt&pepper 0.794872 78 0.00000000000008724523

cropping 0.489362 47 0.00054426

Half sizing 0.395 77 0.24719


1993, 202-216.

[10] A.Mayache, T. Eude and H. Cherefi, A

comparison of image quality models and metrics

based on human visual sensitivity, Proc. IEEE

Intl. Conf. on Image Processing,ICIP’98, Chicago,

IL, USA, Oct. 1998, 409-413.

[11] S. Pereira, S. Voloshynovskiy, M. Madueo,

S.Marchand-Maillet and T. Pun, Second

generation benchmarking and application oriented

evaluation, Information Hiding Workshop,

Pittsburgh, PA, USA, April 2001, 340-353.

[12] D. Kundur and D. Hatzinakos, Digital

watermarking using multiresolution wavelet

decomposition, IEEE ICASSP’98, Volume 5,

Seattle,WA, USA, May 1998,2659-2662

(a) (b)

© (d)


(e) (f)

(g) (h)

(i)

Figure 3. (a)original baboon image (b) baboon image marked via our watermarking scheme with T 1 = 115, T 2 =

200, X1 = 20 and X2 = 10 and attacked with: (c) JPEG quality 5, (d) JPEG quality 10, (e) JPEG quality 15, (f)

Gaussian noise (σ^2 = 375), (g) impulse noise (normalized density of 0.015), (h) cropping and (j) half sizing

(followed by resizing back to the original size).


Managing Unstructured Data Using Agent Technology

Amit Kumar Goel1, Ritu Sindhu2 , Monica Mehrotra3 ,G.N. Purohit4

([email protected], [email protected], [email protected] , [email protected] )

ABSTRACTIn today’s scenario, all the companies are storing their data electronically, due to reducing cost of Media, Servers etc. So now a big problem arises to organize unstructured data properly. Mainly source of unstructured data are documents, spreadsheets and Emails. Most of the information companies generate—more than 70 percent, according to experts—won’t fit into the cells of a traditional relational database. So the problem is how to handle unstructured data.This paper raises the issues related to Information Lifecycle Management (ILM) and gives the powerful approach to manage such type of data. This paper includes issues surrounding Information Lifecycle Management (ILM) and resolves the problem of Information Retrieval through agents. Keywords: Information Lifecycle Management (ILM), Extraction Agent, Retrieval Agent, Categorization Agent, Customer Relationship Management, Ontology

1.0 INTRODUCTION

Corporate information stored on file servers and network attached storage (NAS) devices is in danger of compromise because IT governance policies and access rules in many companies are incapable of dealing with a massive growth of unstructured data. A Ponemon survey of 870 IT professionals found that only 23% believe unstructured data stored by their companies is properly secured and protected. A wide majority - 84% -- of respondents said that too many workers at their companies can access critical corporate unstructured data. About 76% said their companies have no process in place to control which employees can access specific unstructured data. Such unchecked access could expose internal security gaps and increase the potential for misuse of data, the study notes [5][14]. Larry Ponemon, chairman of the Traverse City, Mich.-based research firm, noted that IT managers say that it's difficult to find automated access control processes that can determine the importance of information the moment it's created. About 61% of respondents said they cannot keep track of which user’s access specific unstructured data, and 91% said their organizations lack the ability to determine data ownership because of faulty governance policies and a lack of available storage tools that can remedy the problem. While IT managers continue to spend significant sums of money on storage technology to hold rapidly increasing amounts of structured data, many admit that the complexity of unstructured data still makes it difficult to secure it. "What we find is not that they won't spend money on it, but they really don't know how to resolve the issue because of the complexity; it's a knowledge issue,” The respondents

said that without adequate controls for unstructured data, the top potential problems are insider negligence and deliberate misuse or theft of information from within an organization. Unstructured data is defined as electronic information residing on file servers and NAS devices that is not stored in a database or in a document/content management system. He said it can include: e-mail, instant messages, Microsoft Word documents; PowerPoint files; electronic spreadsheets; and source code. 2.0 ILM ILM is used to manage data from the beginning of its creation to end. ILM is comprised of the policies, processes, practices, and tools used to align the business value of information with the most appropriate and cost-effective. Since many organizations have no formal Records Management Policies that have been transferred to electronic content this may mean that final disposition is never reached. Other organizations have even worse Electronic Records Management Policies that are based on questionable analysis of existing rules and regulations, and call for the destruction of possibly valuable data by fiat directive. E.g. All e-mails older than 90 days will be deleted from company systems. ILM uses a number of technologies and business methodologies, including the following:

Assessment Socialization Classification






http://www.computerworld.com/action/inform.do?command=search&searchTerms=Larry+Ponemon

http://www.computerworld.com/action/article.do?command=viewArticleBasic&taxonomyName=san&articleId=9076799&taxonomyId=147

http://www.computerworld.com/action/inform.do?command=search&searchTerms=Microsoft+Word

http://www.computerworld.com/action/inform.do?command=search&searchTerms=Microsoft+PowerPoint

Automation Review

In the assessment phase of ILM, storage administrators can take advantage of storage resource management (SRM) technologies. SRM solutions help IT administrators figure out what data resides on the storage assets in their environment. Most SRM tools can generate reports for IT that outline data usage patterns. Once the IT department understands what data it has and where this data lives, it can begin the next steps of the ILM process: generating reports from the SRM tools, presenting them to the company's department heads and explaining the breakdown of storage asset utilization and the costs involved. This process is known as the socialization phase of ILM [14] [15]. Once IT meets with the department heads, and the groups collaborate to understand data usage patterns, department heads must determine how this data is used and how critical it is to the business at any given point in time. The ability to prioritize data based on business requirements (that is, mission-critical, business-sensitive and departmental) will allow IT to determine where data should live through its lifecycle and assist in creating policies to migrate data to the proper storage "class" over time. IT must work with department heads to set up a classification schema for the company. Data can be classified in the following ways: Data type Data "Organization" Data age Data "Value" IT will use all data collected at this point to establish policies to automate the data's migration through the environment, with a minimum amount of hands-on data management. SRM solutions should be employed throughout the ILM process, not merely for an initial assessment. SRM technology can monitor the storage environment constantly, revealing where excess capacity, duplicate files, "unnecessary files" or aged files exist. This information is very important in the ILM process, for it is essential in understanding which data should be migrated, archived or purged [5]. 2.1 AUTOMATION PROCESS There are various elements of most ILM systems. Tiered Infrastructure Data Management Layer Application Specific Interfaces Tiered infrastructure is essentially the same as HSM (Hierarchical Storage Management) without using Data Management Layer. The idea behind this is to use different storage solution, moving from expensive Tier I to Tier IV. Tier I: fast access, high-performance primary disk. Tier II: low-cost disk such as SATA [serial advanced technology architecture] disk.

Tier III: tape technology where data must be retained but is unlikely to be referenced again. Tier IV: offline tape in a secure facility, possibly offsite, which can be manually reintroduced into a tape library in the very unlikely event that it needs to be recalled,”. The emergence of SATA disks has done much to boost ILM efforts, he says. SATA arrays can store data at a fraction of the cost of high-performance disk. When shifting point-in-time copies of data, for example, SATA is often the obvious choice. That is not to say, however, that tape technology is becoming redundant. "Despite advances in disk, there are still huge advantages to tape technology. Tape technology that can hold around 1.5 terabytes on an [pounds sterling] 80 tape. The costs involved in storing large volumes of data that will probably not be accessed again on tape are now staggeringly low, and disk - even low-cost disk -- still cannot match them." Example: One customer is used ILM to balance availability and cost by automating payroll data management and migration. Payroll processing is a mission-critical application, so it made sense to store the data on high-performance disk during the processing cycle and replicate it every two hours. Once the pay cycle is complete, the automated management system now moves payroll data to mid-range SATA disk arrays. At this stage, users can access payroll data from the company's web site for a period of three months. After three months, the data is written to a tape library, which is on the same campus as the data archive. For disaster recovery protection, the data is replicated to a remote location, where it is stored on a back-up tape library. ILM is an ongoing process - data storage administrators will need to continually maintain a balance between data performance needs and storage options. "The struggle is to get the client to realise that getting benefit out of ILM is only 20% about technology and 80% about business processes. It's that kind of housework that drives the biggest savings," he says. The Data Management Layers are the tools responsible for performing the "aging" process. Rudimentary retention policies can be applied to the data as it passes through the tier storage layers. Application Specific Interface: Most systems do not provide direct interfaces into Specific Applications Interfaces, without having some sort of helper application. They may have an Application Programming Interface (API) for addressing their specific methods of applying retention policies. 2.2 FUNCTIONALITY OF ILM There are five phases identified as part of ILM.

Creation and Receipt Distribution


Use Maintenance Disposition Exception

Creation and Receipt: deals with records from their point of origination. This could include their creation by a member of an organization at varying levels or receipt of information from an external source. It includes correspondence, forms, reports, drawings, computer input/output, or other sources. Distribution is the process of managing the information once it has been created or received. This includes both internal and external distribution, as information that leaves an organization becomes a record of a transaction with others. Use takes place after information is distributed internally, and can generate business decisions, document further actions, or serve other purposes. Maintenance is the management of information. This can include processes such as filing, retrieval and transfers. While the connotation of 'filing' presumes the placing of information in a prescribed container and leaving it there, there is much more involved. Filing is actually the process of arranging information in a predetermined sequence and creating a system to manage it for its useful existence within an organization. Failure to establish a sound method for filing information makes its retrieval and use nearly impossible. Transferring information refers to the process of responding to requests, retrieval from files and providing access to users authorized by the organization to have access to the information. While removed from the files, the information is tracked by the use of various processes to ensure it is returned and/or available to others who may need access to it. Disposition is the practice of handling information that is less frequently accessed or has met its assigned retention periods. Less frequently accessed records may be considered for relocation to an 'inactive records facility' until they have met their assigned retention period. Retention periods are based on the creation of an organization-specific retention schedule, based on research of the regulatory, statutory and legal requirements for management of information for the industry in which the organization operates. Additional items to consider when establishing a retention period are any business needs that may exceed those requirements and consideration of the potential historic, intrinsic or enduring value of the information. If the information has met all of these needs and is no longer considered to be valuable, it should be disposed of by means appropriate for the content. This may include ensuring that others cannot obtain access to outdated or obsolete information as well as measures for protection privacy and confidentiality. Long-term records are those that are identified to have a continuing value to an organization. Based on the period assigned in the retention schedule, these may be held for periods of 25 years or longer, or may even be assigned a

retention period of "indefinite" or "permanent". The term "permanent" is used much less frequently outside of the Federal Government, as it is impossible to establish a requirement for such a retention period. There is a need to ensure records of a continuing value are managed using methods that ensure they remain persistently accessible for length of the time they are retained. While this is relatively easy to accomplishing with paper or microfilm based records by providing appropriate environmental conditions and adequate protection from potential hazards, it is less simple for electronic format records. There are unique concerns related to ensuring the format they are generated/captured in remains viable and the media they are stored on remains accessible. Media is subject to both degradation and obsolescence over its lifespan, and therefore, policies and procedures must be established for the periodic conversion and migration of information stored electronically to ensure it remains accessible for its required retention periods. Exceptions occur with non-recurring issues outside the normal day to day operations. One example of this is a legal hold, litigation hold or legal freeze is requested by an attorney. What follows is that the records manager will place a legal hold inside the records management application which will stop the files from being in queued for disposition. 3.0 AN AGENT APPROACH TO MANAGE UNSTRUCTURED DATA Three Agents can be formed to manage unstructured data. 3.1.1. Extraction Agent 3.1.2. Categorization Agent 3.1.3. Retrieval Agent Extraction Extraction Agent: This agent is used for examining the semantics of document. This agent extract document before categorizing them. Categorization Agent: This agent is responsible to categorize document & consider the way in which document is subdivided. Retrieval Agent: This agent is responsible for retrieving information from the collection of documents efficiently and effectively. Before applying information retrieval technique the document should be categorized . To determine whether or a document is pertinent to a particular retrievel process for retrieval agent. In Artificial Intelligence ontologies are developed by humans as models. Ontology serves as a representation vocabulary that provides a set of terms with which to describe the facts in some domain. Concepts represented by an ontology can usually be clearly depicted through natural language because the ontology and natural language function similiarly.. Depending on the construction of the ontology, the meaning of each


world could remain the same as in natural language.In a computer system; context may be represented and constrained by ontology. In other words ontology provides a context for the vocabulary it contains [1][2]. Categorization agent is responsible to categorize data because manually categorize information is highly inefficient and often impractical. Once awareness of the issue is raised, the next step is to identify the unstructured data in the organization. In content-management systems, such as those from Interwoven, Web pages are typically considered unstructured data even though essentially all Web pages are defined by the HTML markup language, which has a rich structure. This is because Web pages also contain links and references to external, often unstructured content such as images, XML files, animations and databases (see Figure 1).

Unstructured data is also prevalent in customer relationship management (CRM) systems, specifically when customer- service representatives and call-center staff create notes. However, once again the verbatim text in call-center and customer-service notes is embedded within a form that is

both highly structured and easily represented in a database format. In sum, unstructured data nearly always occurs within documents. Even though many documents follow a defined format, they may also contain unstructured parts. This is another reason why it's more accurate to talk about the problem of semi-structured documents. A basic requirement for semi-structured documents is that they be searchable. Prior to the emergence of the Web, full-text and other text-search techniques were widely implemented within library, document- management and database management systems. However, with the growth of the Internet, the Web browser quickly became the standard tool for information searching. Indeed, office workers now spend an average of 9.5 hours each week searching, gathering and analyzing information, according to market-research firm Outsell Inc.; and nearly 60 percent of that time, or 5.5 hours a week, is spent on the Internet, at an average cost of $13,182 per worker per year. Is all this searching efficient? Not really. Current Web search engines operate similarly to traditional information-retrieval systems: They create indexes of keywords within documents and then return a ranked list of documents in response to a user query. Several studies have shown that the average length of search terms used on the public Web is only 1.5 to 2.5 words and that the average search contains efficient Boolean operators (such as and, or and not) fewer than 10 percent of the time. With such short queries and so little use of advanced search techniques, the results are predictably poor. In fact, a performance assessment of the top five Web search engines, conducted by the U.S. National Institute of Standards and Technology, showed that when 2.5 search words are used, only 23 to 30 percent of the first 20 documents returned are actually relevant to the query. In recognition of the weakness of basic, keyword search, the search-engine vendors have continued to improve their technology. For example, Verity has added techniques such as stemming and spelling correction to its K2 arsenal, while newcomer phrase employs natural language processing. Information Retrieval through Agents: Due to the popularity of www, has created bulk of unstructured data in the form of documents, spreadsheets, Emails and PDF. So a great issue is to extract information from online documents. There is a information retrieval agent (IR agent) and information extraction agent (IE agent) for same [4] [6].

Web Page (consisting Images & Graphics, DataBase, XML

document, Flash Animation)

Data Base Content

Flash Animation

Images and Graphics

XML

Figure1: Web Page Extraction


Agents are intelligent because they can adapt their behavior according to the user’s instructions and the feedback they get from their environments. In other words, they are learning agents the user, an intelligent agent that use neural networks to store and modify their knowledge [8][9][11].

Behavior

Advice

Reformulated

Advice

Figure 3 illustrates the interaction between the user, an intelligent agent, and the agent’s environment. The user observes the agent’s behavior , and provides helpful instructions to the agent. We refer to users instructions as advice, si nce this name emphasizes that the agent does not blindly follow the user-provided instructions, but instead refines the advice based on its experiences. The user inputs his/her advice into a user-friendly advice interface. The given advice is then processed and mapped into the agent’s knowledge base (i.e.,i ts two neural networks),where it gets refined based on the agent’s experiences. Hence, the agent is able to represent the user model in its neural networks, which is used for effective learning

ENVIRONMENT

USE R

4.0 CONCLUSION The Information Lifecycle Management is the complete automation of the entire unstructured data management process. The ideal system will monitor the network, automatically enforcing policies on file naming and storage availability based on how valuable the content is. Intelligent analysis tools will suggest which files should be imported into structured data system, and which should be downgraded to low cost storage or deleted. This article presented concept of linking unstructured information. Solution is presented by using agent oriented approach with emphasis on cooperation with business user while searching for information and exploiting navigational support [12][13]. We envision future research to focus in the area of integrating user’s context when retrieving information from unstructured documents. The semantic web is one possible approach, in which pages can be given well defined meaning. Software agents can also assist web users by using this information to search , filter and prepare information in new ways. This approach allows better integration between machine and people and assists the evolution of human knowledge. In addition, future technologies must have the capability to automatically extract the meaning of unstructured documents with reference to the context of the users with minimal human intervention.

5.0 REFERENCES

[1] Albers M, Jonker CM, Karami M,Treur J (2004) Agents models and different user ontology’s for an electronic market place, Knowl Inf Syst 6 (1): 1:41.

[2] Alexander Smirnov & Nikolay Shilov (2007), Ontology-driven intelligent service for

IR Subsystem

IE subsystem

IE AG

ENT

IR A

GEN

T

Fig 2: An Agent Overview for Information Extraction

User

Environment

Action Advice

Interface

Advice Processor

Network Mapper

Figure 3: The interaction between a user, an intelligent agent, and Environment agent’s

i t


configuration support in networked organization. Springer-Verlag London Limited.

[3] Belew, R. K.: 2000, Finding Out About: A Cognitive Perspective on Search Engine Technology and the WWW. New York, NY: Cambridge University Press.

[4] Croft, W., Turtle, H. and Lewis, D.: 1991, ‘The use of phrases and structured queries in information retrieval’ In: proceedings of the Fourteenth International ACMSIGIR Conference on R & D in Information Retrieval. Chicago, IL, pp. 32-45.

[5] Ching Kang Cheng and Xiao Shan Pan, Using Perception in Managing Unstructured Documents. ACM Student Magazine.

[6] Dejan & Viljan, Intelligent agent aided use of unstructured information in decision support.

[7] David A Maluf & Peter B Taran, Managing Unstructured Data with Structured Legacy Systems. NASA Aims Research Center Intelligent System Divisions.

[8] Eliassi-Rad, T.: 2001, ‘Building Intelligent Agents that Learn to Retrieve and Extract Information’ Ph.D. thesis, Computer Sciences Department, University of Wisconsin, Madison, WI. (Also appears as UW Technical Report CS-TR-01-1431)..

[9] Maes, P.Agents that reduce work and Information overload. Communications of the ACM, 37(7), 1994, pp. 31-40.

[10] Sebastiani, F.Machine Learning in Automated Text Categorization. ACM Computing Surveys (CSUR). Volume 34 Issue 1, March 2002

[11] Soderland, S.: 1999, ‘Learning information extraction rules for semi-structured and free text’ Machine Learning: Special Issue on Natural Language Learning 34 (1/3), 233-272.

[12] Seymore,K.,McCallum,A.and Rosenfeld, R.: 1999, ‘Learning Hidden Markov Model Structure for Information Extraction’ In: Proceedings of the Sixteenth National Conference on Arti¢cial Intelligence Workshop on Machine Learning for Information Extraction. Orlando, FL, pp. 37-42.

[13] Shavlik, J. and Eliassi-Rad,T.: 1998a, ‘Building intelligent agents for web-based tasks: A theory-refinement approach’ In: Proceedings of the Conference on Automated Learning and Discovery Workshop on Learning from Text and the Web. Pittsburgh, PA.

[14] Tony Pfitzner & Tyson Lloyd Thwaites, Unstructured data: A management overview. www.allianza.com.au

[15] Vinita Gupta, Managing unstructured data, www.expresscomputeronline.com dated 21/01/2008


http://www.expresscomputeronline.com/

Multi-layer Fiber for Dispersion Compensating And Wide Band Amplification

Asso. Prof. A. S. Samra, Eng. H. A. M. Harb Department of Electronics and Communications

Faculty of Engineering, Mansoura University, Egypt (e-mail: [email protected], [email protected])

ABSTRACT

This paper investigates the dispersion compensating performance in multi-layer fiber. We show that very large negative dispersion value can be obtained, depending on the geometrical parameters. The splice loss between the standard G.652 fiber and the multi-layer fiber is calculated. Raman amplifier using multi-layer fiber as a gain medium is investigated using one pump as well as 2 pumps, ASE is calculated. Keywords: Dispersion compensation, multi clad fiber, Raman amplification

1 INTRODUCTION

In recent years DWDM optical communication is

seeing a steady mitigate from 2.5 to 40 Gbps over each wavelength achieving higher spectral efficiency, which is defined as the ratio of average transmission rate to channel spacing. Amplification and dispersion compensation/management have assumed great importance as there are the main impairing factors for achieving repeater less transmission distance in excess of 100km over standard single mode fibers.

One of the earliest techniques suggested to reduce the dispersion at 1550nm band was to tailor the refractive index profile of a single mode fiber in such a way that its zero dispersion wavelength is shifted from the conventional 1310nm window to a round 1550nm[1]. These fibers, called dispersion shifted fibers (DSF) through appeared promising for a while, but, were found to be unusable in DWDM link due to the fact that operating a fiber with near zero dispersion is known to introduce nonlinear effects like FWM[2]. It is known that FWM effect can be greatly reduced by allowing a small but finite local dispersion all along a DWDM link. This task could be fulfilled either through dispersion management (i.e. by combing alternate lengths of positive and negative dispersion fibers [3]) or by employing so called nonzero dispersion shifted fibers. Which is designed to leave a small residual average dispersion of 2.6ps/km.nm to omit nonlinear propagation effects in the single mode fiber.

Chromatic dispersion is a linear effect and

inserting a component with opposite sign could greatly reduce its detrimental effect in G.652 fibers at the 1550nm band. Out of the several different technique that have been proposed in the literature, the ones which seem to hold immediate promise could be classified as dispersion compensating fiber

(DCF)[4], chirped fiber Bragg grating (FBG) [5],[6], high order mode (HOM) fibers[7].

In chirped grating the optical pitch (product between the grating period and the mode effective index) varies along length of the FBG. As a result, resonant reflection frequency of the FBG becomes a function of position along length of FBG. Thus, each frequency component of a propagating pulse is reflected from a different point along length of chirped FBG. This is depending on the sign of the chirp; a chirped FBG could impart either a positive or negative dispersion to a propagating pulse [8]. Since, dispersion compensation is achieved or reflection to access the dispersion corrected pulse. And optical circulator or a fiber coupler is required as an additional component with associated insertion loss. Further more, errors in the chirped phase mask periodicity could lead to ripples in group delay with wavelength.

The HOM technique exploits large negative dispersion slop, which are characteristics of higher order modes of a fiber relative to the fundamental mode. Thus one requires a fiber, which supports more than one mode at the operating wavelengths. Further, conversion of power from the fundamental to a higher order mode and reconverting the same back to fundamental mode have not been as easy task through lately there have been a number of promising studies reported on the technique[9],[10].Out of all these, by far the DCF technique has been the most widely used technique. One of the main advantages of this technical solution is that when appropriately designed it can provide a passive system, in principle, with negative chromatic dispersion coefficient D as high as -5000ps/km.nm. Such a scheme should compensate the positive chromatic dispersion over a relatively short length of the DCF and having low sensitivity to environmental influence (temperature,




23

vibertional, etc) like that of the signal carrying fiber [11].

After a brief presentation of many possible solutions usable to compensate dispersion, we will discuses the multi-layer fiber design in the next section. Different variations in refractive index profiles with several layers and different shapes had been simulated in previous work [11],[12], but, step index profiles composed of two concentric spatially separated cores appears to give the best performances according to the fabrication constraints. The third section focuses on the splicing losses between the multi-layer fiber and the standard G.652 fiber. The use of multi-layer fiber with Raman amplifier will be discussed in the fourth section. Finally, conclusions are presented in section 5. 2 Multi-layer Fiber

Fig. 1 shows the refractive index profile of the multi-layer fiber which has a dual core design. It consists of two concentric cores: the inner core with a large ∆ and the outer core with a small ∆. Here ∆ is defined as .where ∆2 2

3( ) / 2i in n n∆ = − 1=0.02 and ∆2=0.003, n3 is calculated with the well known Sellmeier equation:

2323 2 2

1

.1

( )i

j j

An

Bλ

λ=

= +−∑ (1)

with A1=0.6981388, A2=0.40865177, A3=0.89374039, B1=0.070555513, B2=0.11765660, B3=9.8754039 are the Sellmeier constants [13].

As seen from Fig. 1 multi-layer fiber has four distinct regions: rod (0<r< r1), gap (r1<r<r2), barrier (r2<r<r3), and clad (r>r3), and can be thought as composed of two substructures namely rod and tube, as indicated in the fig. (1).

It is seen from Fig.2 that at wavelengths shorter than 1550nm the field is essentially confined to the inner core and for which the guide essentially functions like a step index single mode fiber, the effect of the outer core being negligible.

Around 1550nm optical coupling takes place between the inner and the outer core modes. At wavelengths longer than 1550nm however, most of the power of the fundamental mode spreads to the outer core and is effectively guided in the outer core. The fractional power in the second supermode of the fiber, which is orthogonal to the first supermode, is maximal in the outer core for wavelengths longer than 1550nm. This phenomenon induces a rapid change in the slope of the effective index (ne) versus wavelength around 1550nm [11],[12]. Fig.3 depicts the variation of ne as a function of the wavelength for the fundamental supermode.

The resulting chromatic dispersion coefficient of the fiber is then computed through the following formula:

2

2ed n

Dc dλ

λ−

= (2)

A sample result corresponding to the fiber parameters in Fig.1 shown in Fig.4. Since such a profile is easily attainable with common perform fabrication systems.

n1

n2

M

Multi-layer fiber n3

0

5

10

15

20

0

5

10

15

20

0.20.40.60.81 A

mplitude

r1

λ(µm) Radius (µm) 1.55

1.6

1.5

-20

0

20

Fig.(2) Evolution of the mode amplitude of the fundamental supermode versus wavelength.

Rod fiber

Tube fiber r2

r3

∆1

∆2

Fig.(1) Refractive index profile of the multi-layer fiber.


The parameters of the two cores are chosen that each of these supports a single azimuthally symmetric mode in the operating range of wavelengths. The fiber parameters are so optimized that the two individual modes (corresponding to the inner and the outer core) are nearly phase matched at 1550nm. In such a case, because the non-supermode, this mode is expected to have a large dispersion, the magnitude and spectral variation of which can be optimized by varying the separation (r2-r1).the behavior of the dispersion curve is sensitive to the variation of the rod radius as see from Fig. 5. 3 Splice Losses

To calculate the splice loss between the standard G.652 fiber and the designed multi-layer fiber, we have used the analysis given in [14]. According to this analysis the fractional power coupled from a G.652 fiber to a multi-layer fiber is given by the equation (4), where subscripts in ψ (which is the mode field shape) correspond to the fiber type. Thus, the total splice loss, including both input and output splices, is given by Total splice loss= (3) 1020log T− where,

.652

22

.6520 0

2 2

.6520 0 0 0

.G Coaxial

G Coaxial

G Coa

rdrdT

rdrd rdrd

π

π π

ψ ψ ϕ

xialψ ψ ϕ ψ ψ

∞

∞ ∞∗ ∗

=∫ ∫

∫ ∫ ∫ ∫ ϕ

(4)

To reduce the splice loss and its spectral variation, we have considered the G.652 fiber to be tapered by 40%. The effect of tapering has been modeled as scaling of the fiber dimensions. This results in the spreading of the model field of the G.652 fiber, leading to a better overlap with the fundamental supermode field of the multi-layer fiber in addition.

The spectral variation in the operating range of

wavelengths, assuming 40 % tapering of the G.652 fiber, including the effect of these wavelength dependant losses in our analysis, we have iteratively tuned the fiber parameters such that the net output gain spectrum is flat. Fig.7 shows the spectral variation of the total splice loss at the input and the output splice f x( )

4 Amplification

Fiber Raman amplifier (FRA) is considered to be a key component to realize a next generation photonic networks because of its features of the noise reduction, flexible gain bandwidth, and simple configuration. The Raman amplifier configuration with forward pumping is shown in Fig.8, the pumping signals are launched into fiber through an optical coupler and propagate a long with the information signals that are fed at the fiber input [15]. A typical Raman gain spectrum for pure silica fiber is shown in Fig. 9, for the pump wavelength of 1450nm. The optical signal gain strongly depends on the Raman gain coefficient, which is a function of the wavelength. The total amplified power over all signal band of the optical fiber Raman amplifier with one pumping source is ( ) ( , )sig sigP z P z dυ υ= ∫ , where υ is the stokes frequency, and the signal spectral power density Psig(υ,z) is the stokes power Ps(υ,z) in z point along the fiber per unit frequency range.

Multi-layer fiber shows high Raman amplification. Multi-layer fiber can be designed in a manner that in the wavelength range where Raman gain coefficient (gR) decreases, the effective area of interaction Aeff also, decreases in almost the same manner. Such that the effective Raman gain (gR/Aeff) is reasonably flat on a large wavelength range. Using multi-layer fiber with only one pump will achieve flat gain Raman amplification over any band [16].

d x( )

x1.2 1.26 1.32 1.38 1.44 1.5 1.56 1.62 1.68 1.74 1.8

1.445

1.45

10000

Ef

fect

ive

inde

x

Wavelength (µm)

Fig. 3. Evolution of effective index of the fundamental supermode with wavelength.

x1.4 1.5 1.6

1000

500

0

D

ispe

rsio

n ps

/km

.nm

Wavelength (µm)

Fig. 4. Evolution of the chromatic dispersion versus wavelength.


This fiber has a unique property that the effective Raman gain spectrum is inherently flat over a large wavelength range and the effective Raman gain spectrum can be tuned by the fiber parameters.The parameters of the two cores are chosen so that each of them supports a single azimuthally symmetric mode in the operating range of wavelength. These parameters are optimized such that the two individual modes (corresponding to the inner and the outer core) are nearly phase matched at 1550 nm [17], [18].

The signal wavelengths below PWM will be tightly confined to the inner core, leading to a high pump-signal overlap, and thus a small Aeff. However, as the signal wavelength approaches the phase matching wavelength, the fractional power of the fundamental mode will gradually increase in the outer core.

M

0 1 2 3 4 5 6 7 8 9 10 11 12

01.534.567.5

910.51213.515

0

750

1500

2250

3000

00

Hence, the overlap between the pump and the

signal fields starts to decrease, increasing the effective area. Thus, by suitably choosing the fiber parameters, phase matching wavelength and the pump wavelength, one can ensure that the decrease in Aeff almost compensates for the decrease in the Raman gain coefficient, such that a flat effective gain spectrum is achieved.

Hence, the mode field at the pump wavelength 1450nm and the signal wavelengths will be tightly confined to the inner core and thus the pump and the signal overlap will be high, giving a small Aeff. However, as the signal wavelength approaches and crosses the phase wavelength, the fractional power of the fundamental mode will gradually increase in the outer core.

0

0

0

1000

-Dis

pers

ion

ps/k

m.n

m

750

500

250

1.56 1.55

1.54 1.53

1.52

0.9 λ (µm) 1

1.1 r1 (µm) 1.2

Fig. 5. Wavelength dependent dispersion curves for different r1

Standard SMF G.652

Standard SMF G.652

Multi-layer fiber

Coupler Output signal

Input signal

Forward pumping Splice Splice

f x( )

x15.2 15.4 15.6 15.8 16

1.6

1.7

1.8

1.9

Tota

l spl

ice

loss

(dB

)

Wavelength (µm)

Fig. 7. Variation of the total spices loss at the input and output splices.

Fig. 6. Input and output splices between SMF(G.652) and multi-layer fiber.

PUMP

Figure 8. Configuration of Raman amplifier with forward pumping.

g υ( )122

1.423 υ

10000

1.45 1.5 1.55 1.6 1.65 1.70

0.2

0.4

0.6

0.8

1

Ram

an G

ain

coef

ficie

nt (m

/wat

t)

Wavelength (µm)

Figure 9. Raman gain spectrum in a pure silica optical fiber for pump wavelength 1450nm.


Assuming the same parameters for all refractive indices do not affect the general trend of the results presented here. The effective index ne is shown in Fig.3, for a multi-layer fiber with r1=1µm, r2 =15µm, r3=22µm. Fig.10, shows the effective Raman gain spectrum using multi-layer fiber for the band model, it is obvious that a flat effective gain spectrum with -3dB is achievable over the range (1450-1550 nm), corresponding ASE is shown in Fig. 12. Fig. 11 shows the effective Raman gain spectrum using inherently flat gain with two pumps, such Raman amplifier can serve as broadband amplifier (for C and L bands) as well as dispersion compensating module. It is also cost-effective since it uses only two pumps (1426, 1486nm). The ASE curve is shown in Fig. 12, and by compare it with the multipumping scheme, we can conclude that decreasing the number of pumps, also decreases the ASE. 5 CONCLUSION

We briefly review the most usable solutions to compensate dispersion. We also, study the effect of chromatic dispersion of the multi-layer fiber, showing the effect of varying the geometric parameters (rod radius). The splice loss between the multi-layer fiber and SMF(G.652) is also presented.

We show that using multi-layer fiber as gain medium in Raman amplifier show that multi-layer fiber is inherently flat gain (IFGF). The Raman amplifier with single pump using multi-layer fiber as well as with two pumps is introduced; also, the ASE is estimated. It is obvious that, decreasing the number of pumps also decreases the ASE. Here, we conclude that Raman amplifier with two pumps using inherently flat gain fiber can cover the C and L bands for DWDM communication systems. MATHCAD is used as an analytical programming tool. REFERENCES [1] M.A. Saifi, S. J. Lang, L. G. Cohen, and J. Stone,” Triangular

profile single mode fiber,” Opt. Lett., vol.7, No. 43 (1982). [2] G. P. Agrawal, “Nonlinear fiber optics” Third ed. Academic,

San Diego, Ca. 2001. ISBN: 0-12-045143-3. [3] I.P. Kaminow, “Optical fiber telecommunications,” Elsevier

Academic Press IV 2002. ISBN: 0-12-395172-0. [4]A. Huttunen,"Optimization of dual-core and microstructure

fiber geometries for dispersion compensation and large mode area", OPTICS EXPRESS, Vol.13, No.2, Jan. 2005.

[5] Ruchti, Randy , ”Performance of multiclad scintillating and waveguide optical fibers readout with visible light photon counters” Proc. SPIE Vol. 2007, p. 78-94,2007.

[6] B.J. Eggleton et al., ”Recompression of pulse broadened by transmission through 10 km of non dispersion shifted fiber at 1.55m using 40 mm long optical fiber Bragg grating with tunable chirp and central wavelength,” IEEE Photon. Technol. Lett. Vol.7, no. 5, 1995.

[7] G. P. Agrawal, “Raman Amplification in fiber optical communication systems” 1st ed. Elsevier Academic Press, 2005. ISBN: 0-12-044506-9.

[8] B.P. Pal, “All fiber components, “, in Electromagnetic field unconventional structures and material A. Lakhtakia and O. N. Singh, Eds. Wiley, New York, 2000.

[9] S. Ramachandran et al., “All fiber grating based higher order dispersion compensator for broadband compensation and 1000km transmission at 40Gbps,” In Proc. ECOC2000, Paper PD-25, 2000.

[10] A.H.Gnauck, L. D. Garret, Y. Danziger, L. Levy and M.Shur, “Dispersion and dispersion slop compensation of NZ-DSF for 40 Gbps operation over the entire C band.” In Proc. OFC 2000, Paper PD-8, 2000.

[11] K. Thyagarajan, R. K. Varshney, and P. Palai, “A novel design of a dispersion compensating fiber,” IEEE. Photon. Techno. Lett, Vol.8, no.11 1996.

[12] P. Palai, R. K. Varshney, and K. Thyagarajan, “A dispersion flattening dispersion compensating fiber design for broadband dispersion compensation,” Fiber Integr. Opt., Vol.20, 2001, pp.21-27.

e υ( )

2 υ( )3

e

e1 υ( )5

c1 υ( )

12 log s1 υ 100( )( ).

1.45 3υ

10000.

1.45 1.5 1.55 1.6 1.6520

15

10

5

0

Effe

ctiv

e R

aman

gai

n

Wavelength (µm)

Figure 10. Effective Raman gain with inherently flat fiber with single pump.

Band model With IFGF

c υ( )

12 log s1 υ 100( )( ).

1.45 3υ.

1.45 1.5 1.55 1.6 1.6520

15

10

5

0

Wavelength (µm)

Effe

ctiv

e R

aman

gai

n

Band model 2 pump with IFGF

Figure 11. Effective Raman gain with inherently flat fiber with two pumps.

1.07 3. υ

10000.

1.5 1.55 1.6 1.658

6.4

4.8

3.2

1.6

0

A

SE P

ower

(dB

)

Single pump with IFGF 2 pumps with IFGF Multipumping

Wavelength (µm)

Figure 12. ASE power of the inherently flat gain fiber with single pump and two pumps.


http://adsabs.harvard.edu/cgi-bin/author_form?author=Ruchti,+R&fullauthor=Ruchti,%20Randy%20C.&charset=UTF-8&db_key=PHY

[13] M. J. Adams, “An Introduction to optical waveguides” John Willy & Sons, pp.213, 1981

[14] A. Ghatak and K. Thyagarajan, “Introduction to fiber optics.” Cambridge Univ. Press, 2008.

[15] M.N. Islam,“Raman Amplifiers for Telecommunications” IEEE J. Sel. Top. Quant. Elect., Vol.8, No. 3, 2002, pp.548-559.

[16] S. P. Singh and N. Singh, "Nonlinear Effects in Optical Fiber: Origin, Management and Applications", Progress In Electromagnetic Research, PIER 73, 2007, pp.249–275.

[17] I.P. Kaminow, “Optical fiber telecommunications,” Elsevier Academic Press IV 2002. ISBN: 0-12-395172.

Ahmed Shaban Samra was born in Mansoura ,Egypt 1954. He received the B.Sc. and the M.Sc degree in communications engineering from Menoufia University 1977, 1982 respectively, and the Ph.D. degree in optical communications and integrated optics from ENSEG, Greroble, France in 1988. He is now an associate professor at the faculty of engineering, Mansoura University. His research interests are in the field of optical communications and optical measurement technique. Hani Ali Mahmoud Harb was born in Mansoura, Egypt 1976. He received the B.Sc. in electronics engineering and the M.Sc degree in communications engineering, both from Mansoura University, Egypt, in 1999 and 2003, respectively. He is currently working toward the Ph.D. degree in communications at Mansoura University. His research activities have been devoted to optical communication systems, optical CDMA, DWDM, and Raman fiber amplifiers.


Performance Analysis of a Novel OFDM System Based on Dual – Tree Complex Wavelet Transform (DT-CWT)

Mohamed H. M. Nerma1, Nidal S. Kamel2 and Varun Jeoti3 Electrical & Electronic Engineering Department, University Technology PETRONAS, Malayisa

1E-mail: [email protected] , [email protected] , and [email protected]

ABSTRACT As demand for higher data rates is continuously rising, there is always a need to develop more efficient wireless communication systems. The work described in this paper is an effort in this direction. We have proposed a novel OFDM system base on DT-CWT. In the proposed scheme, DT-CWT is used in the place of FFT. The proposed scheme offers the best PAPR performance than the conventional OFDM and wavelet packet modulation (WPM) systems at the expense of acceptable computational complexity without using any pruning techniques. The complementary cumulative distribution function (CCDF) of PAPR for the proposed scheme signal achieves about 3 dB improvement in PAPR over the traditional OFDM and WPM signals at 0.1% of CCDF. Also the proposed scheme achieves excellent improvements in BER over conventional OFDM and WPM systems. The need for CP is eliminated in the system design due to the good orthogonality and time frequency localization proprieties of the wavelet. Keywords: OFDM, WPT, CWT, DT-CWT, FFT, MCM, BER, PAPR.

1 INTRODUCTION

Orthogonal frequency division multiplexing (OFDM) and wavelet packet modulation (WPM) have emerged as an efficient multicarrier modulation schemes for wireless, frequency selective, communication channels. Ease of implementation, high spectral efficiency, resilience to impulse noise and multipath are a few advantages of OFDM and WPM systems. However, a major drawback in the signals of these two systems are their large envelope function, which limits the efficiency of the non-linear power amplifiers specific to wireless communications by forcing them to operate at lower average power. This problem is quantified by the Peak – to – average power ratio (PAPR) and results from the superposition of a large number of usually statistically independent sub-channels that can constructively sum up to high peaks. Also from the central limit theorem (CLT) [1], this causes the OFDM and WPM signals to have complex Gaussian process behavior and the instantaneous power is chi-square distributed. Various schemes have been developed to reduce high PAPR in OFDM [2] [3], and WPM signals [4] [5].

1.1 Wavelet Modulation Wavelet Transform (WT) is a relatively new

transform compared to the discrete Fourier transform (DFT). WT provides the time-frequency representation of signals, whereas DFT gives only the frequency representation. The properties of wavelet, such as localization in time and frequency, orthogonality across scale and translation presents to a new perspective in digital communication. It is used as a modulation technique in many communication fields including multicarrier modulation (MCM) and wireless communication [6]. The conventional multicarrier systems are DFT based systems. In MCM, the broadband channel splits into larger number of sub-channels. The high data rate bit-streams are divided into parallel sub streams with lower data rate [7].

However, DFT based MCM systems suffer from the high side lobes due to rectangular shaped DFT window and also these systems waste precious bandwidth due to the redundant cyclic prefix (CP). Moreover, the pulse shaping function used to modulate each subcarrier extends to infinity in the frequency domain this leads to high interference and lower performance levels. WT based MCM systems, namely wavelet based OFDM systems (WOFDM) can help to mitigate these problems. Therefore, conventional DFT based orthogonal systems are being replaced by wavelet based transceivers. The


wavelet based transceiver uses quadrature mirror filters (QMF) in the synthesis and analysis filter banks (FBs). The wavelet filters posses the advantages of having greater side-lobe attenuation and requires no CP [8]. WPT modulation in wireless communication has been proposed in [9]. The characteristics of multicarrier modulated signal are dependent on the basis functions being used in a modulation scheme. Therefore, using WPT as basis functions, the sensitivity to multipath channel distortion, inter symbol interference (ISI), inter carrier interference (ICI), and synchronization can be reduced as compared to traditional OFDM and the system performance of WOFDM with reference to ISI, ICI and signal-to-noise ratio (SNR) is shown to be far better than the conventional OFDM (DFT-OFDM) [9], [10]. 1.2 Peak – to – Average Power Ratio (PAPR)

The PAPR of the baseband transmitted signal is defined as the ratio of the peak power

( max | |2 ); i.e., the maximum power of the transmitted signal over the average power ( E | |2 ). In digital implementations of communications transceivers, rather than using the continuous time signal in PAPR computation, we instead work with , the discrete time samples of x(t), provided that an oversampling factor of at least 4 is used. PAPR is then expressed as [11]:

max | |

E | | 1

where . denotes ensemble average calculated over the duration of the OFDM or WPM symbols.

In both OFDM and WPM systems, the signal going into the channel is a sum of random symbols modulating orthogonal basis functions. Based on the CLT, it is claimed that is complex Gaussian and its envelope follows a Rayleigh distribution. This implies a large PAPR. A high PAPR of the transmitted signals demands a very linear transmission path and limits the practical deployment of low-cost non-linear power amplifiers forcing them to operate with reduced power efficiency. Driving amplifiers operating close to saturation with signals of high PAPR results in the generation of unwanted spectral energy both in-band and out-of-band, which in turn reduces the systems performance and gives rise to adjacent channel interference (ACI).

In this work, the performance of the OFDM based on DT-CWT in PAPR reduction is demonstrated through the CCDF of PAPR, which is a performance metric independent of the transmitter amplifier. Given the reference level PAPR0 > 0, the probability of a PAPR being higher than the reference value is the CCDF and is expressed as follows [12]:

2

To reduce the PAPR in OFDM and WPM

systems, several techniques have been proposed, which basically can be divided in three categories. First, there are signal distortion techniques, which reduce the peak amplitudes simply by nonlinearly distorting the OFDM signal at or around the peaks. Examples of distortion techniques are clipping, peak windowing and peak cancellation. The second category is coding techniques that use a special forward-error correcting code set that excludes OFDM symbols with a large PAPR. The third technique is based on scrambling each OFDM symbol with different scrambling sequences and selecting that sequence that gives the smallest PAPR [13].

This paper is organized as follows: In section II we discuss the dual-tree complex wavelet transform (DT-CWT); in section III we discuss the half-sample delay condition; in section IV we make a comparisons between traditional OFDM and WPM systems; in section V we discuss the OFDM based on DT-CWT; in section VI we discuss the PAPR in OFDM based on DT-CWT; the BER results are presented in section VII; and we conclude this paper in section VIII by discussion the simulation results. 2 THE DUAL – TREE COMPLEX WAVELET

TRANSFORM (DT-CWT)

Since the early 1990s the WT and WPT have received more and more attention in modern communications and have been widely used in wireless communication [14]. A number of modulation schemes based on wavelets have been proposed [15 - 23]. In fact, complex wavelet transform (CWT) is applied perfectly to digital image processing. Kingsbury [24 - 28] introduced and made a concrete description of DT-CWT. The DT-CWT employs two real discrete WT (DWT); the upper part of the FB gives the real part of the transform while the lower one gives the imaginary part. This transform uses the pair of the filters ( ,

the low-pass/high-pass filter pair for the upper FB respectively) and ( , the low-pass/high-pass filter pair for the lower FB respectively) that are used to define the sequence of wavelet function and scaling function as follows

√2 ∑ 2 3

√2 ∑ 2 4

Where 1 1 0 , the wavelet function , the scaling function and the high-pass filter for the imaginary part are


defined similarly. The two real wavelets associated with each of the two real transform are and . To satisfying the perfect reconstruction (PR) conditions, the filters are designed so that the complex wavelet is approximately analytic. Equivalently, they are designed so that is approximately the Hilbert transform of .

5

The analysis (decomposition or demodulation) and the synthesis (reconstruction or modulation) FBs used to implement the DT-CWT and their inverses are illustrated in fig. 1 and fig. 2 respectively. The inverse of DT-CWT is as simple as the forward transform. To invert the transform, the real part and the imaginary part are each inverted. 3 THE HALF-SAMPLE DELAY CONDITION

The two low pass filters should satisfy a very simple property: one of them should be approximately a half-sample shift of the other [29]

0.5 6

Since and are defined only on the integers, this statement is somewhat informal. However, we can make the statement rigorous using FT. In [5] it is shown that, if

.

, then . The converse has been proven in [30] [31], making the condition necessary and sufficient. The necessary and sufficient conditions for biorthogonal case were proven in [32].

Figure 1: The dual tree discrete CWT (DT-DCWT) Analysis (demodulation) FB. We can rewrite the half-sample delay condition in terms of magnitude and phase function separately as shown in (7) and (8) as follows.

Figure 2: The Inverse dual tree discrete CWT (IDT-DCWT) Synthesis (modulation) FB.

7

0.5 8

In practical implementation of the DT-CWT, the delay condition (7) and (8) will be satisfied only approximately; the wavelets and will form only an approximate Hilbert pair; and the complex wavelet will be only approximately analytic. While the FT is based on complex valued oscillating cosine and sine components form a complete Hilbert transform pairs; i.e., they are 90 out of phase with each other. Together they constitute an analytic signal that is supported on only one-half of the frequency axis

0 [24]. 3.1 Filter Design for the DT-CWT

There are various approaches to the design of filters for the DT-CWT, such as linear-phase biorthogonal method, quarter shift method, and common factor method. These filters are satisfied the following desired properties:

1- Approximately half-sample delay property. 2- PR (orthogonal or biorthogonal). 3- Finite support (finite impulse response (FIR)

filters). 4- Vanishing moments/good stop-band. 5- Linear phase filters.

It turns out that the implementation of the DT-CWT requires that the first stage of the dual-tree (DT) FB requires a different set of filters from the succeeding stages, and the succeeding stages can be used same sets of filters. See fig. (1). If the same PR filters are used for each stage, then the first several stages of the FB will not be approximately analytic [24]. 4 OFDM AND WPM SYSTEMS

Traditionally, OFDM is implemented using FFT. This transform however has the drawback that is uses a rectangular window, which creates rather high sidelobes. Moreover, the pulse shaping function used to modulate each subcarrier extends to infinity in the


frequency domain (FD) this leads to high interference and lower performance levels. ISI and ICI can be avoided by adding a CP to the head of OFDM symbol. Adding CP can largely reduce the spectrum efficiency. The WPM system has a higher spectral efficiency and providing robustness with regard to inter-channel interference than the conventional OFDM system, because of the out-of-band energy (low sidelobes). Moreover WPM is able to decompose T-F plane flexibly by arranging FB constructions. In addition, using FFT in the traditional OFDM gives resolution only in the FD while using WPT in WPM system gives resolution in both FD and time domain (TD) [16].

WPM system do not required CP, thereby enhancing the spectrum efficiency. According to the IEEE broadband wireless standard 802.16.3, avoiding CP gives wavelet OFDM an advantage of roughly 20% in bandwidth (BW) efficiency. Moreover as pilot tones are not necessary for wavelet based OFDM system, they perform better in comparison to existing OFDM systems like 802.11a or HiperLAN, where 4 out of 52 sub-bands are used for pilots. This gives wavelet based OFDM system another 8% advantage over typical OFDM implementations [39]. We expected the OFDM based on DT-CWT will take all the advantages of WPM system.

However, a major problem of the common discrete WPT (DWPT) is its lack of shift invariance; this means that on shifts of the input signal, the wavelet coefficients vary substantially. The signal information may even not be stationary in the sub-bands so that the energy distribution across the sub-bands may change [15]. To overcome the problem of shift dependence, one possible approach is to simply omit the sub-sampling causing the shift dependence. Techniques that omit or partially omit sub-sampling are also known as cycle spinning, oversampled FBs or undecimated WT. However, these transforms are redundant [33], which is not desirable in multicarrier modulation. As an alternative, we used a non-redundant WT that achieves approximately shift invariance [34], this transform yields to complex wavelet coefficients that modulate the data stream in the same way that WPM do. 5 OFDM BASED ON DT-CWT

In this study, simulations are focusing on using DT-CWT in the OFDM system as a non-redundant WT that can achieves approximately shift invariance [35 – 38]. Similar to the conventional OFDM and WPM systems, a functional block diagram of OFDM based on DT-CWT is shown in fig. (3). At the transmitter an inverse DT-CWT (IDT-CWT) block is used in place of inverse FFT (IFFT) block in conventional OFDM system or in place of inverse DWPT (IDWPT) block in WPM system. At the

receiver side a DT-CWT is used in place of FFT block in conventional OFDM system or in place of DWPT block in WPM system. Data to be transmitted are typically in the form of a serial data stream. PSK or QAM modulations can be implemented in the proposed system the choice depends on various factors like the bit rate and sensitivity to errors. The transmitter accepts modulated data (in this paper we use 16 QAM). This stream is passed through a serial to parallel (S/P) converter, giving N lower bit rate data stream, and then this stream is modulated through an IDT-CWT matrix realized by an N-band synthesis FB. Before the receiver can demodulate the subcarriers, it has to perform the synchronization. For the proposed system, known data interleaved among unknown data are used for channel estimation. Then, the signal is down sampled by N and demodulated using elements of the DT-CWT matrix realized by an N-band analysis FB. The signal is equalized after DT-CWT stage.

Figure 3: DT-CWT modulation (DT-CWTM) functional block diagram.

IDT-CWT works in a similar fashion to an IFFT

or IDWPT. It takes as the input QAM symbols and outputs them in parallel time-frequency “subcarriers”. In fig. (2) as the synthesis process, it can be shown that the transmitted signal, is constructed as the sum of M waveform individually modulated with the QAM or PSK symbols [34], φ t is the scaling function, ψ t is the wavelet function, and a is kth symbol, i 1, 2, … , N as follows:

. 9

, , , , ./

10

/ , / , , , ./

11

Where , is a constellation encoded data symbol modulating the DT-CWT function. The IDT-CWT synthesis a discrete representation of the transmitted signal as sum of M waveforms


shifted in time that embed information about data symbols. Those waveforms are built by successive iterations of ^, ^ and ^, ^. The DT-CWT at the receiver recovers the transmitted symbols , through the analysis formula exploiting orthogonality properties of DT-CWT and schematically represented in fig. (1). In the baseband equivalent OFDM transmitter with

frame of N QAM symbols, , 0,1, … , 1, the OFDM frame is given by:

/ 12

While for WPM system, the transmitted signal is constructed as the sum of M wavelet packet function individually modulated with the QAM symbols as the case in this paper‡.

, 13

The construction of discrete versions of

transmitted waveforms for the conventional OFDM and WPM systems using (12) and (13) is quite similar. For any time index n, both waveforms are sum of random symbols a or a , .

Figure 4: PSD of the Conventional OFDM, WPM and OFDM based on DT-CWT.

In order to achieve fair comparisons, same

simulation parameters are used. The simulations were carried out for conventional OFDM with a 64 subcarriers using a 16 QAM modulation. The Daubechies-1 (Daub-1) wavelet packet bases were used to construct the wavelet packet trees in WPM system. For the proposed scheme, near symmetric 13,19 tap filters and quarter sample shift orthogonal 14 tap filters were used to construct the real and the imaginary part of DT-CWT respectively. From the CLT, x n is complex Gaussian distributed, and the

‡ We use φ n for the wavelet packet function and ψ n for the DT-CWT function to avoid any confusion.

sequence x n has high PAPR. Furthermore, to demonstrate the similarities between power spectrum density (PSD) characteristic of conventional OFDM, WPM and the proposed scheme, the simulated PSD characteristics are presented in fig. 4. This fig. was shows that the proposed scheme performs better than the other two systems.

The simulation parameters are documented as follows: Modulation type is 16-QAM; the number of subcarriers is 64 subcarriers; a wavelet packet base is Daubechies-1 (DAUB-1); maximum tree depth (D = 7); PAPR threshold is 2dB; shaping filter is Raised Cosine (rolloff factor 0.001, upsampler = 4); DT-CWT using different filters (LeGall 5,3 tap filters (leg), Antonini 9,7 tap filters (anto), Near Symmetric 5,7 tap filters (n-sym-a), and Near Symmetric 13,19 tap filters (n-sym-b)) for the first stage of the FB and (Quarter Sample Shift Orthogonal 10,10 tap filters (q-sh-06) only 6,6 non-zero taps, Quarter Sample Shift Orthogonal 10,10 tap filters (q-sh-a) with 10,10 non-zero taps, unlike q-sh-06, Quarter Sample Shift Orthogonal 14,14 tap filters (q-sh-b), Quarter Sample Shift Orthogonal 16,16 tap filters (q-sh-c), and Quarter Sample Shift Orthogonal 18,18 tap filters (q-sh-d)) for the succeeding stages of the FB. 6 PAPR ANALYSIS RESULTS

Consider the simple case where all the sub-symbols are independently and identically distributed (i.i.d). then, by the CLT, the real and imaginary parts of the N-point IFFT output have mutually independent Gaussian probability distribution function with zero mean and standard deviation to σ . The instantaneous power of baseband signal , is defined by

14

We can characterize the instantaneous power as a

chi-square distribution with two degrees of freedom [31]

1

, 0 15

If E | | is normalized to unity, then the CCDF of the PAPR is given by:

Pr 1 1 16

Where N is the number of subcarriers.

In order to analyze PAPR, we generate the transmitted waveforms using 16 QAM modulation with 64 subcarriers for all these systems. Fig. 5 shows, in the TD, the envelope of the proposed system. For the comparison, we also plotted the envelope of the conventional OFDM and WPM


waveforms corresponding to the same information symbol pattern. The transmitted envelopes for the conventional OFDM and WPM systems illustrate approximately similar behavior, where the peak is about 2.25, while the transmitted envelope for the proposed system demonstrates better behavior than the other two systems, where the peak is only about 1.25 and this is the reason for that the proposed system gives better result for PAPR than the other two systems.

Figure 5. The Envelope of the Conventional OFDM, WPM and OFDM based on DT-CWT.

Figure 6: CCDF the Conventional OFDM, WPM and OFDM based on DT-CWT.

The results for PAPR are best quantified using

CCDF. In fig. 6, for 64 subcarriers with 16 QAM modulation, the CCDF plots for the proposed system, conventional OFDM and WPM system are shown. This figure shows that the OFDM based on DT-CWT offers the best PAPR performance without using any reduction techniques. The proposed scheme signal achieves about 3 dB improvement in PAPR over the traditional OFDM and WPM signals at 0.1% of CCDF while the other two systems, are approximately given same results of PAPR.

The simulation results for PAPR were also repeated with 16 QAM modulation and 64 subcarriers (in fig. 7) using different set of filters. OFDM DT-CWT_1 illustrate the PAPR when using near-symmetric (n-sym) 13,19 tap filters in the first

stage of the FB and quarter sample shift orthogonal (q-sh) 14 tap filters in the succeeding stages, OFDM DT-CWT_2 using (n-sym 13,19 with q-sha 10 (10 non zero taps) filters), OFDM DT-CWT_3 using (antonini (anto) 9,7 tap filters with q-sh0 10 (only 6 non zero taps) filters), OFDM DT-CWT_4 using (anto 9,7 with q-sh 14 filters), OFDM DT-CWT_5 using (n-sym 5,7 with q-sh 14 filters), OFDM DT-CWT_6 using (LeGall (leg) 5,3 tap filters with q-sh 14 filters), OFDM DT-CWT_7 using (n-sym 5,7 with q-sh 16 filters) and OFDM DT-CWT_8 using (leg 5,3 with q-sh 18 filters).

The results in fig. 7 show that there is no observed degradation as a result of using different set of mismatching filters in the design of the proposed scheme.

Figure 7: The effect of using different set of filters in design of the OFDM based on DT-CWT.

Figure 8: CCDF of PAPR for 16-QAM modulated conventional OFDM symbol with various values of subcarriers (N).

Again, the results for PAPR were repeated in

both fig. 8 (for the conventional OFDM system) and fig. 9 (for the proposed system) using different numbers of subcarriers (64, 128, 256, 512, and 1024) with 16QAM modulation. We observe from these figures that the PAPR increases as the number of subcarriers numbers (N) increases. As shown in fig. 6, 7 and in fig. 8, 9 the PAPR performance of DT-CWT based OFDM systems is better than the


conventional OFDM system or even WPM systems. As we saw in figure 5 showing the time-domain signal envelop of all the systems, this improvement in PAPR performance is explained from lower peaks of DTCWT systems observed in the fig. 5.

Figure 9: CCDF of PAPR for 16-QAM modulated OFDM based on DT-CWT symbol with various values of subcarriers (N). 7 BER ANALYSIS RESULTS

The results given in this section compare the BER in the OFDM based on DT-CWT, with that for traditional OFDM, and WPM. Also, same simulation parameters are used to achieve a fair comparison. The results of BER in OFDM based on DT-CWT using different set of filters are also shown in this section.

The results for BER are shown that the proposed scheme gives excellent improvements in BER over conventional OFDM and WPM systems. At the same time the conventional OFDM outperform the WPM system in term of BER. As shown in fig. 10.

The simulation results for BER were repeated using different set of filters (OFDM DT-CWT-1, OFDM DT-CWT-2, …, OFDM DT-CWT-8) and is shown in fig. 11. We see there is degradation of BER as a result of using different set of mismatching filters in the design of the proposed scheme.

Figure 10: BER performance of conventional OFDM, WPM and OFDM based on DT-CWT.

Figure 11: The effect of using different type of filters in BER for OFDM based on DT-CWT. 8 CONCLUSION

In this paper a new OFDM scheme that is based on DT-CWT is proposed. Comparing the proposed scheme in terms of PAPR and BER with the traditional OFDM and WPM systems we see that the proposed scheme offers 3dB better PAPR performance over the conventional OFDM and WPM systems at 0.1% of CCDF. While the conventional OFDM and WPM systems shows similar behavior. Simulation results shows that there is no observed PAPR and BER degradation as a result of using different set of mismatching filters in DT-CWT based system. The proposed scheme outperforms the traditional OFDM and WPM systems in term of BER. Also we found that the conventional OFDM system gives better results of BER than WPM system.

ACKNOWLEDGMENTS Authors wish to thanks Prof. Nick G. Kingsbury

(University of Cambridge, United Kingdom) and Dr. Mohan Baro (Dalhousie University, Canada) whose comments and suggestions have largely contributed to improve this paper.

APPENDIXES A. CCDF Approximations

We can characterize the instantaneous power as a chi-square distribution with two degrees of freedom [40]

1

, 0 17

As a result, the cumulative distribution function (CDF) is defined as:


1 18

If E | | is normalized to unity, then the CCDF of the PAPR is given by:

1 1 19

Where N is the number of subcarriers, however, this approximation is not close to the experimental results as the assumption made in deriving CCDF that the samples should be mutually uncorrelated is no longer valid when oversampling is employed [40]. There has been several attempts to determine the closed for approximations for the distribution of PAPR. Some of the approximations are shown in Table 1. Table 1: Approximation to CCDF of PAPR.

CCDF Remarks

= 2.8 and N 64

[40]

[41]

[42]

√

/ N and are large [43]

/

[44]

B. DT-CWT and IDT-CWT If the two real DWTs are represented by the

square matrices for the upper part and for the lower part, then the DT-CWT can be represented by the following form.

1

√2. 20

And the IDT-CWT is given by

1

√2. 21

Note that the complex sum/difference matrix in (20) is unitary (its conjugate transpose is its inverse).

1√2

.1

√2 22

Note that the identity matrix on the right-hand side of (22) is twice the size of those on the left-hand side. Therefore if the two real DWTs are orthonormal transforms, then the DT-CWT satisfies . , where * denotes the conjugate transpose. 9 REFERENCES

[1] Papoulis A., Pillai S. U., “Probability Random Variables and Stovhastic Processes,” New York: McGraw-Hill Inc., 2002.

[2] M. Breiling, S. H. muller, and J. B. Huber, “SLM peak power reduction without explicit side information”, IEEE Commun. Lett., vol. 5, no. 6, pp. 239-241, 2001.

[3] J. Tellado, Ed., Multicarrier Modulation with low PAPR Application to DSL and wireless. New York: Kluwer Academic Publishers, 2002.

[4] M. Baro and J. Ilow, “PAPR reduction in wavelet packet modulation using tree pruning”, in 2007 IEEE 65th Vehicular Technology Conference VTC 2007 – spring, Dubin, Ireland, Apr. 2007.

[5] M. Baro and J. Ilow, “PAPR reduction in OFDM using wavelet packet pre-processing”, Consumer Communications and Networking Conference, 2008. CCNC 2008. 5th IEEE.

[6] M K Lakshmanan, H Nikookar, "A review of Wavelets for Digital Communication," Wireless Personal Communication (2006) 37: 387- OFDM* 420, Springer 2006.

[7] J. A. C. Bingham, "Multicarrier modulation for data transmission: An Idea Whose Time Has Come," IEEE Communications Magazine, vol. 28, no.5, pp. 5-14, 1990.

[8] H. M. Newlin, "Developments in the use of wavelets in communication systems," TRW Systems & Information Technology Group, Sunnyvale, California.

[9] A. Jamin, P. Mahonen, "Wavelet packet modulation for wireless communications," Wireless Communications and Mobile Computing 5 (2): 123-137 (2005).

[10] C. V. Bouwel, J. Potemans, S. Schepers, B. Nauwelaers and A.V. de Capelle, "Wavelet Packet Based Multicarrier Modulation," Proc. IEEE Benelux Symposium on Communications and Vehicular Technology, Leuven, Belgium, 19 October 2000.

[11] C. Schurgers and M. B. Srivastava, “A systematic approach to peak – to – average power ratio in OFDM,” in SPIE’s 47th Annual Meeting, San Diego, CA, 2001, pp. 454-464.

[12] S. H. Han and J. H. Lee, “An Overview of to peak – to – average power ratio reduction techniques for multimedia transmission,” IEEE Wireless Communications, vol. 12, no. 2, pp. 56-65, 2005.

[13] Richard Van Nee and Ramjee Prasal, “OFDM for Wireless Multimedia


Communications”, P. cm Artech House Universal Personal Communications Series, Inc. Boston. London, 2000.

[14] Panchamkumar D Shukla, “Complex wavelet Transforms and Their Applications” Master Thesis 2003. Signal Processing Division. University of Strathclyde Department of Electronic and Electrical Engineering.

[15] M. Guatier, J. Lienard, and M. Arndt, “Efficient Wavelet Packet Modulation for Wireless Communication”, AICT’07 IEEE Computer Society, 2007.

[16] A. Jamin, and P. Mahonen, “Wavelet Packet Modulation for Wireless Communications”, Wiley Wireless Communications and networking, Journal, vol. 5, no. 2, pp. 123-137, Mar. 2005.

[17] Xiaodong Zhang and Guangguo Bi, “OFDM Scheme Based on Complex Orthogonal Wavelet Packet”,http://ieeexplore.ieee.org/iel5/7636/20844/00965270.pdf.

[18] M. Guatier, and J. Lienard, “Performance of Complex Wavelet packet Based Multicarrier Transmission through Double Dispersive Channel”, NORSIG 06, IEEE Nordic Signal Processing Symposium (Iceland), June 2006.

[19] C. J. Mtika and R. Nunna, “A wavelet-based multicarrier modulation scheme,” in Proceedings of the 40th Midwest Symposium on Circuits and Systems, vol. 2, August 1997, pp. 869–872.

[20] N. Erdol, F. Bao, and Z. Chen, “Wavelet modulation: a prototype for digital communication systems,” in IEEE Southcon Conference, 1995, pp. 168–171.

[21] A. R. Lindsey and J. C. Dill, “Wavelet packet modulation: a generalized method for orthogonally multiplexed communications,” in IEEE 27th Southeastern Symposium on System Theory, 1995, pp. 392–396.

[22] A. R. Lindsey, “Wavelet packet modulation for orthogonally multiplexed communication,” IEEE Transaction on Signal Processing, vol. 45, no. 5, pp. 1336–1339, May 1997.

[23] C. V. Bouwel, J. Potemans, S. schepers, B. Nauwelaers, and A. Van Caelle, “wavelet packet Based Multicarrier Modulation”, IEEE Communication and Vehicular Technology, SCVT 2000, pp. 131-138, 2000.

[24] Ivan W. Selesnick, Richard G. Baraniuk, and Nick G. Kingsbury, “The Dual-Tree Complex Wavelet Transform,” IEEE Signal Processing Mag, pp. 1053-5888, Nov 2005.

[25] N.G. Kingsbury, “The dual-tree complex wavelet transform: A new technique for shift invariance and directional filters,” in Proc. 8th IEEE DSP Workshop, Utah, Aug. 9–12, 1998, paper no. 86.

[26] N.G. Kingsbury, “Image processing with complex wavelets,” Philos. Trans. R. Soc. London A, Math. Phys. Sci., vol. 357, no. 1760, pp. 2543–2560, Sept. 1999.

[27] N.G. Kingsbury, “A dual-tree complex wavelet transform with improved orthogonality and symmetry properties,” in Proc. IEEE Int. Conf. Image Processing,

Vancouver, BC, Canada, Sept. 10–13, 2000, vol. 2, pp. 375–378.

[28] N.G. Kingsbury, “Complex wavelets for shift invariant analysis and filtering of signals,” Appl. Comput. Harmon. Anal., vol. 10, no. 3, pp. 234–253, May 2001.

[29] I W. Selenick, “Hilbert transform pairs of wavelet bases,” IEEE Signal Processing Lett., vol. 8, no. 6, pp. 170-173, June 2001.

[30] H. Ozkaramanli and R. Yu, “on the phase condition and its solution for Hilbert transform pairs of wavelet bases,” IEEE Trans. Signal Processing, vol. 51, no. 12, pp. 2393–3294, Dec. 2003.

[31] R. Yu and H. Ozkaramanli, “Hilbert transform pairs of orthogonal wavelet bases: Necessary and sufficient conditions”, IEEE Trans. Signal Processing IEEE Transactions on Signal Processing, vol. 53, no. 12, Dec. 2005.

[32] R. Yu and H. Ozkaramanli, “Hilbert transform pairs of biorthogonal wavelet bases”, IEEE Transactions on Signal Processing, VOL. 54, NO. 6, JUNE 2006.

[33] I. W. Selesnick, “the Double Density Dual-Tree DWT”, IEEE Transactions on Signal Processing, 52(5): 1304 – 1315, May 2004.

[34] J. M. Lina,”Complex Daubechies Wavelets: Filter Design and Applications”, ISAAC Conference, June 1997.

[35] Mohamed H. M. Nerma, Nidal S. Kamel, and Varun jeoti, “PAPR Analysis for OFDM Based on DT-CWT” Proceeding of 2008 Student Conference on Research and Development (SCOReD 2008), 26-27 Nov. 2008, Johor, Malaysia.

[36] Mohamed H. M. Nerma, Nidal S. Kamel, and Varun jeoti, “On DT-CWT Based OFDM: PAPR Analysis” Multi-Carrier Systems & Solutions 2009, V. 41, pp. 207-217, SpringerLink April 26, 2009- Dordrecht, Netherlands.

[37] Mohamed H. M. Nerma, Nidal S. Kamel, and Varun jeoti, “BER Performance Analysis of OFDM System Based on Dual – Tree Complex Wavelet Transform in AWGN Channel” Proceeding of 8th WSEAS International Conference on SIGNAL PROCESSING (SIP '09), Istanbul, Turkey, May 30 - June 1, 2009.

[38] Mohamed H. M. Nerma, Nidal S. Kamel, and Varun jeoti, “An OFDM System Based on Dual Tree Complex Wavelet Transform (DT-CWT)” Journal: Signal Processing: An International Journal, Volume: 3, Issue: 2, Pages: 14-21, May 2009.

[39] M. K. Lakshmanan and H. Nikookar, “A Review of Wavelets for Digital Wireless Communication”, Wireless Personal Communications Springer, 37: 387-420, Jan. 2006.

[40] R. Van Nee and A. De Wild, “reduction the peak-to average power ratio of OFDM” 48th IEEE Vehicular Technology Conference, Ottawa, Canada, May 18-21, 1998; 2072-2076, IEEE New York, USA.

[41] H. Ochiai and H. Imai, “On the disitribution of the peak-to-average power ratio in OFDM signals”, IEEE Trans. Commun. 2001; 49, 282-289.

[42] S. wei, D. L. Goeckel, and P. E. Kelly “A modern extreme value theory approach to


calculate the distribution of the peak-to-average power ratio in OFDM systems”, IEEE International Conference on Communications, New York, Apr. 28- May 2, 2002; vol. 3, 1686-1690; IEEE New York, USA.

[43] H. Yu, M. Chen and G. wei “Distribution of PAPR in DMT systems”, Electron. Lett. 2003; 39, 799-801.

[44] R. Rajbanshi, A. M. Wyglinski and G. J. Minden, “Multicarrier Transceivers: Peak-to-average power ratio reduction”, IEEE Commun. Society. WCNC 2008 proceeding.

Mohamed H. M. Nerma, was born in Khartoum, Sudan. He received the B.Sc. degree in Electrical Engineering (Control) from Sudan University of Science and Technology (SUST), Sudan in 1999. He received the M.Sc. degree in Communication

Engineering from Karary Academy of Technology, Sudan in 2002. From 2002 to 2006 he was lecturer in the Sudan University of Science and Technology. He is currently PhD. Student in University Technology PETRONAS, Malaysia.

Nidal S. Kamel, received his Ph.D degree (Hons) in telecommunication and statistical signal processing from the Technical University of Gdansk, Poland, in 1993. His research is focused on linear estimation, noise reduction,

pattern recognition, optimal filtering, and telecommunications. Currently he is working for Universiti Teknologi PETRONAS, Malaysia. He is senior member of IEEE.

Varun Jeoti Jagadish, received his Ph.D. degree from Indian Institute of Technology Delhi India in 1992. He worked on several sponsored R&D projects in IIT Delhi and IIT Madras during 1980 to 1989 developing Surface Acoustic

Wave Pulse Compression filters, underwater optical receivers etc.. He was a Visiting Faculty in Electronics department in Madras Institute of Technology for about 1 year during 1989 to 1990 and joined Delhi Institute of Technology for next 5 years till 1995. He moved to Electrical & Electronic Engineering (E&E Engg) department of Universiti Sains Malaysia in 1995 and joined E&E Engg of Universiti Teknologi PETRONAS in 2001. His research interests are in Wireless LAN and MAN technologies, DSL technology and related signal processing.


Ant Colony Optimization Algorithm

Nada M. A. Al Salami

[email protected]

ABSTRACT Hybrid algorithm is proposed to solve combinatorial optimization problem by using Ant Colony and Genetic programming algorithms. Evolutionary process of Ant Colony Optimization algorithm adapts genetic operations to enhance ant movement towards solution state. The algorithm converges to the optimal final solution, by accumulating the most effective sub-solutions. Keywords:, ACO, Genetic algorithm, System theory.

1. Background and Related Work

Genetic Algorithms (GA) have been used to evolve computer programs for specific tasks, and to design other computational structures. The recent resurgence of interest in AP with GA has been spurred by the work on Genetic Programming (GP). GP paradigm provides a way to do program induction by searching the space of possible computer programs for an individual computer program that is highly fit in solving or approximately solving the problem at hand[1][2]. The genetic programming paradigm permits the evolution of computer programs which can perform alternative computations conditioned on the outcome of intermediate calculations, which can perform computations on variables of many different types, which can perform iterations and recursions to achieve the desired result, which can define and subsequently use computed values and sub-programs, and whose size, shape, and complexity is not specified in advance. GP use relatively low-level primitives, which are defined separately rather than combined a priori into high-level primitives, since such mechanism generate hierarchical structures that would facilitate the creation of new high-level primitives from built-in low-level primitives [3] [4] [5]. Unfortunately, since every real life problem are dynamic problem, thus their behaviors are much complex, GP suffers from serious weaknesses. . random systems. Chaos is important, in part, because it helps us to cope with unstable system by improving our ability to describe, to understand, perhaps even to forecast them. Ant Colony Optimization (ACO) is the result of research on computational intelligence approaches to combinatorial optimization originally conducted by Dr. Marco Dorigo, in collaboration with Alberto Colorni and Vittorio Maniezzo [6]. The fundamental approach underlying ACO is an iterative process in which a population of simple agents repeatedly construct candidate solutions; this construction

process is probabilistically guided by heuristic information on the given problem instance as well as by a shared memory containing experience gathered by the ants in previous iteration. ACO has been applied to a broad range of hard combinatorial problems. Problems are defined in terms of components and states, which are sequences of components. Ant Colony Optimization incrementally generates solutions paths in the space of such components, adding new components to a state. Memory is kept of all the observed transitions between pairs of solution components and a degree of desirability is associated to each transition depending on the quality of the solutions in which it occurred so far. While a new solution is generated, a component y is included in a state, with a probability that is proportional to the desirability of the transition between the last component included in the state, and y itself [7]. The main idea is to use the self-organizing principles to coordinate populations of artificial agents that collaborate to solve computational problems. Self-organization is a set of dynamical mechanisms whereby structures appear at the global level of a system from interactions among its lower-level components. The rules specifying the interactions among the system’s constituent units are executed on the basis of purely local information, without reference to the global pattern, which is an emergent property of the system rather than a property imposed upon the system by an external ordering influence. For example, the emerging structures in the case of foraging in ants include spatiotemporally organized networks of pheromone trails [8][9][10]. The aim of this work is to enhance the ability of ACO by using GP technique, as we describe in the next secttion. 2.Genetic Programming Some specific advantages of genetic programming are that no analytical knowledge is needed and still could get accurate results. GP approach does scale


with the problem size. GP does impose restrictions on how the structure of solutions should be formulated. There are Several variants of GP, some of them are: Linear Genetic Programming (LGP), Gene Expression Programming (GEP), Multi Expression Programming (MEP), Cartesian Genetic Programming (CGP), Traceless Genetic Programming (TGP) and Genetic Algorithm for Deriving Software (GADS). In the next section we shall concentrate on CGP, since it is the most near to our proposed method [4][5].Cartesian Genetic Programming was originally developed by Miller and Thomson [11][12] for the purpose of evolving digital circuits and represents a program as a directed graph. One of the benefits of this type of representation is the implicit re-use of nodes in the directed graph. Originally CGP used a program topology defined by a rectangular grid of nodes with a user defined number of rows and columns. In CGP, the genotype is a fixed-length representation and consists of a list of integers which encode the function and connections of each node in the directed graph. The genotype is then mapped to an indexed graph that can be executed as a program. In CGP there are very large numbers of genotypes that map to identical genotypes due to the presence of a large amount of redundancy. Firstly there is node redundancy that is caused by genes associated with nodes that are not part of the connected graph representing the program. Another form of redundancy in CGP, also present in all other forms of GP is, functional redundancy. Simon Harding, and Ltd introduce computational development using a form of Cartesian Genetic Programming that includes self-modification operations..The interesting characteristic of CGP are : 1. More powerful program encoding using graphs, than using conventional GP tree-like representations, the population of strings are of fixed length, whereas their corresponding graphs are of variable length depending on the number of genes in use. 2, Efficient evaluation derived from the intrinsic feature of subgraph-reuse exhibited by graphs. 3. Less complicated graph recombination via the crossover and mutation genetic operators. . 3. Proposed ACO Genetic Algorithm (ACOG) A combinatorial optimization problem is a problem defined over a set C = c1, ......, cn of basic components. A subset S of components represents a solution of the problem; F ⊆ 2C is the subset of feasible solutions, thus a solution S is feasible if and only if S ∈ F. A cost function z is defined over the solution domain, z : 2C R, the objective being to find a minimum cost feasible solution S*, i.e., to find S*: S* ∈ F and z(S*) ≤ z(S), ∀S∈F [8].. They move by applying a stochastic local decision policy based

on two parameters, called trails and attractiveness. By moving, each ant incrementally constructs a solution to the problem. The ACO system contains two rules:

1. Local pheromone update rule, which applied whilst constructing solutions.

2. Global pheromone updating rule, which applied after all ants construct a solution.

Furthermore, an ACO algorithm includes two more mechanisms: trail evaporation and, optionally, daemon actions. Trail evaporation decreases all trail values over time, in order to avoid unlimited accumulation of trails over some component. Daemon actions can be used to implement centralized actions which cannot be performed by single ants, such as the invocation of a local optimization procedure, or the update of global information to be used to decide whether to bias the search process from a non-local perspective [6][10]

At each step, each ant computes a set of feasible expansions to its current state, and moves to one of these in probability. The probability distribution is specified as follows. For ant k, the probability of moving from state t to state n depends on the combination of two values [9]:

• the attractiveness of the move, as computed by some heuristic indicating the priori desirability of that move;

• the trail level of the move, indicating how proficient it has been in the past to make that particular move: it represents therefore an a posteriori indication of the desirability of that move.

ACOG Algorithm An ACOG is differ from that algorithm given in refrence [13], it use genetic programming to enhance performance. It consists of two main sections: initialization and a main loop, where Gp is used in the second sections. The main loop runs for a user-defined number of iterations. These are described below: ♦Initialization: a. Set initial parameters that are system: variable, states, function, input, output, input trajectory, output trajectory. b. Set initial pheromone trails value. c. Each ant is individually placed on initial state with empty memory. ♦While termination conditions not meet do a. Construct Ant Solution: Each ant constructs a path by successively applying the transition function the probability of moving from state to state depend on: as the attractiveness of the move, and the trail level of the move. b. Apply Local Search


c. Best Tour check: If there is an improvement, update it. d. Update Trails: - Evaporate a fixed proportion of the pheromone on each road. - For each ant perform the “ant-cycle” pheromone update. - Reinforce the best tour with a set number of “elitist ants” performing the “ant-cycle” d. Create a new population by applying the following operation, based on pheromone trails. The operations are applied to individual(s) selected from the population with a probability based on fitness.

• Darwinian Reproduction • Structure-Preserving Crossover • Structure-Preserving Mutation

End While

4. The Performance of Genetic Process

Genetic generation process involves probabilistic steps, because of these probabilistic steps, non-convergence and premature convergence, i .e. convergence to a globally sub-optimal result, problems become inherent features of genetic generation process. To minimize the effect of these problems, multiple independent runs of a problem must be made . Bes t -o f - run ind iv idua l f rom a l l such mu l t ip le independen t runs can then be designated as the result of the group of runs. If every run of GPG were successful in yielding a solution, the computational effort required to get the solution would depends primarily on four factors: population size, M, number of genera t ion tha t a re run, g , (g must be less than or equa l to the maximum number of generation G) the amount of processing required for fitness measure over all fitness cases, and the amount of processing required for test phase e, we assume that the processing time to measure the fitness of an individual is its run time, P. If success occurs on the same generation of every run, then the computational effort E would be computed as follows:

E= M • g • β • e ……..E q.1

Since the value of e is too small with respect to other factors, we shall not consider it. However, in most cases, success occurs on a different generations in different runs, then the computational effort E would be computed as follows:

E=M•gavr• β …….Eq.2

where: gavr is the average number of executed generations

Since GPG is a probabilistic algorithm, not all runs are successful at yielding a solution to the problem by generation G. Thus, the computational effort is computed in this way, first determining the number of independent runs R needed to yield a success with a certain probability. Second, multiply R by the amount of processing required for each run , tha t is . The number of independent runs R required to sa t is fy the success predicate by generation i with a probability z which depends on both z and P (M, i), where z is the probability of satisfying the success predicate by generation i at least once in R runs defined by:

z = 1 - [ 1- P (M, i)]R ……Eq.3

P (M, i) is the cumulative probability of success for all the generations between generation 0 and generation i . P (M, i) is computed after experimentally obtaining an estimate for the instantaneous probability Y (M, i) that a particular run with a population size M yields, for the first time, on a specified generation i, an individual is satisfying the success predicate fo r the p rob lem ] . Th is expe r imen ta l measuremen t o f Y(M, i ) u sua l ly r equ i r e s a substantial number of runs. After taking logarithms for equation 4, we find:

⎡ ⎤)),(1log()1log(

iMzR ρ−

−=

…..Eq.4

The computational effort E, is the minimal value of the total number of individuals that must be processed to yield a solution for the problem with z probability (ex: z = 99%):

E= M • ( •g + 1) • β • R ..…….Eq.5

Where •g is the first generation at which minimum number of individual evaluation is produced, it is called best generation. •g value is incremented by one since generation •g must also run to reach the solution. From equation (5), computational effort depends on the particular choices of values for M, G, P (M, i), and the effort required for fitness evaluation, hence, the value of E is not necessarily the minimum computational effort possible for the problem.

5. Conclution Since Ant colony algorithm may produce redundant states in the graph, its better to minimize such graphs to enhance the behavior of the inducted system. A colony of ants moves through system states X, by applying Genetic Operations. By moving, each ant incrementally constructs a solution to the problem. .


When an ant complete solution, or during the construction phase, the ant evaluates the solution and modifies the trail value on the components used in its solution. Ants deposit a certain amount of pheromone on the components; that is, either on the vertices or on the edges that they traverse. The amount of pheromone deposited may depend on the quality of the solution found. Subsequent ants use the pheromone information as a guide toward promising regions of the search space. Ants adaptively modify the way the problem is represented and perceived by other ants, but they are not adaptive themselves. The genetic programming paradigm permits the evolution of computer programs which can perform alternative computations conditioned on the outcome of intermediate calculations, which can perform computations on variables of many different types, which can perform iterations and recursions to achieve the desired result, which can define and subsequently use computed values and sub-programs, and whose size, shape, and complexity is not specified in advance. References [1] A. Abraham et al.: Evolutionary Computation: from Genetic Algorithms to Genetic Programming, Studies in Computational Intelligence (SCI) 13, 1–20 (2006), www.springerlink.com c _ Springer-Verlag Berlin Heidelberg 2006. [2] C. Grosan and A. Abraham: Hybrid Evolutionary Algorithms: Methodologies, Architectures, and Reviews, Studies in Computational Intelligence (SCI) 75, 1–17 (2007), www.springerlink.com c_ Springer-Verlag Berlin Heidelberg 2007. [3] Héctor A Montes and Jeremy L Wyatt ,” Cartesian Genetic Programming for Image Processing Tasks”, [4] John R. Koza, Margaret Jacks Hall, “ SURVEY OF GENETIC ALGORITHMS AND GENETIC PROGRAMMING”, http://www-cs-faculty.stanford.edu/~koza/. [5] Riccardo Poli, William B. Langdon , Nicholas F. McPhee, John R. Koza, “Genetic Programming :An Introductory Tutorial and a Survey of Techniques and pplications” , Technical Report CES-475 ISSN: 1744-8050 October 2007. essex.ac.uk/dces/research/publications/.../2007/ces475.pdf Appendex [6] Figures[1] M. Dorigo, M. Birattari, and T. Stitzle, “Ant Colony Optimization: Arificial Ants as a Computational Intelligence Technique, IEEE computational intelligence magazine, November, 2006. [7] M. Dorigo, G. Di Caro, and L.M. Gambardella, “Ant algorithm for discrete optimization”, Artificial Life, vol. 5, no. 2, pp. 137-172, 1999. [8] J. Holland, “Adaptation in Natural and Artificial

Systems”, Ann Arbor: University of Michigan Press, 1975. [9] Nada M. A. AL-salami, Saad Ghaleb Yaseen, “Ant Colony Optimization”, IJCSNS International Journal of Computer Science and Network Security, VOL.8 No.6, pp 351-357, June, 2008. [10] M. Dorigo and G. Di Caro, “The Ant Colony Optimization meta-heuristic”, in New Ideas in Optimization, D. Corne et al., Eds., McGraw Hill, London, UK, pp. 11-32, 1999. [11] J. F. Miller and P. Thomson. Cartesian genetic programming. In R. Poli, W. Banzhaf, W. B. Langdon, J. F. Miller, P. Nordin, and T. C. Fogarty, editors, Genetic Programming, Proceedings of EuroGP’2000, volume 1802 of LNCS, pages 121–132, Edinburgh, 2000. Springer-Verlag. [12] Simon Harding, Julian F. Miller, Wolfgang Banzhaf, “Self-Modifying Cartesian Genetic Programming”, GECCO’07, July 7–11, 2007, ACM 978-1-59593-697-4/07/0007, pp: 1021-1028. [13] Nada M.A. AL-Salami, “System Evolving using Ant Colony Optimization Algorithm “, Journal of Computer Science 5 (5): 380-387, 2009, ISSN 1549-3636.


A SURVEY OF MAC PROTOCOLS FOR WIRELESS SENSOR NETWORKS

Rajesh Yadav Electronis and Radar Development Establishment

Defense R & D Organization, Bangalore, India

Shirshu Varma Indian Institute of Information Technolgy, Allahabad, India

N. Malaviya

Institute of Engineering & Technology, Lucknow, India

ABSTRACT Wireless sensor networks (WSNs) have become an active research area for the researchers. The sensor nodes are generally unattended after their deployment in hazardous, hostile or remote areas. These nodes have to work with their limited and non replenish able energy resources. Energy efficiency is one of the main design objectives for these sensor networks. In this paper, we present the challenges in the design of the energy efficient medium access control (MAC) protocols for the wireless sensor network. We describe several MAC protocols for the WSNs emphasizing their strength and weakness wherever possible. Finally, we discuss the future research directions in the MAC protocol design. Keywords: Energy Efficiency, Medium Access Control, Wireless Sensor Network

1 INTRODUCTION

IRELESS Sensor Networks (WSNs) have emerged as one of the dominant technology trends of this decade (2000-2010) that has

potential usage in defence and scientific applications. These WSNs can be used for different purposes such as target tracking, intrusion detection, wildlife habitat monitoring, climate control and disaster management [1]. A typical node in the WSN consists of a sensor, embedded processor, moderate amount of memory and transmitter/receiver circuitry. These sensor nodes are normally battery powered and they coordinate among them selves to perform a common task. These Wireless Sensor Networks have severe resource constrains and energy conservation is very essential. The sensor node’s radio in the WSNs consumes a significant amount of energy. Substantial research has been done on the design of low power electronic devices in order to reduce energy consumption of these sensor nodes. Because of hardware limitations further energy efficiency can be achieved through the design of energy efficient communication protocols. Medium access control (MAC) is an important technique that ensures the successful operation of the network. One of the main functions of the MAC protocol is to avoid collisions from interfering nodes. The classical IEEE 802.11 MAC protocol for wireless local area network wastes a lot of energy because of idle listening. Designing

power efficient MAC protocol is one of the ways to prolong the life time of the network. In this work we carried the study of the energy efficient MAC protocols for the wireless sensor network. The rest of the paper is organized as follows. Section 2 discusses challenges in the design of the MAC protocol. Section 3 presents the different proposed MAC protocols emphasizing their strength and weakness wherever possible. Section 4 discusses future research directions in the MAC protocol design. Finally, Section 5 concludes the paper. 2 MAC PROTOCOL DESIGN CHALLENGES

The medium access control protocols for the wireless sensor network have to achieve two objectives. The first objective is the creation of the sensor network infrastructure. A large number of sensor nodes are deployed and the MAC scheme must establish the communication link between the sensor nodes. The second objective is to share the communication medium fairly and efficiently. 2.1 Attributes of a Good MAC Protocol

To design a good MAC protocol for the wireless sensor networks, the following attributes are to be considered [2].

W


(i) Energy Efficiency: The first is the energy efficiency. The sensor nodes are battery powered and it is often very difficult to change or recharge batteries for these sensor nodes. Sometimes it is beneficial to replace the sensor node rather than recharging them.

(ii) Latency: The second is latency. Latency requirement basically depends on the application. In the sensor network applications, the detected events must be reported to the sink node in real time so that the appropriate action could be taken immediately.

(iii) Throughput: Throughput requirement also varies with different applications. Some of the sensor network application requires to sample the information with fine temporal resolution. In such sensor applications it is better that sink node receives more data.

(iv) Fairness: In many sensor network applications when bandwidth is limited, it is necessary to ensure that the sink node receives information from all sensor nodes fairly. However among all of the above aspects the energy efficiency and throughput are the major aspects. Energy efficiency can be increased by minimizing the energy wastage.

2.2 Major Sources of Energy Wastes Major sources of energy waste in wireless sensor network are basically of four types [2] [3].

(i) Collision: The first one is the collision. When a transmitted packet is corrupted due to interference, it has to be discarded and the follow on retransmissions increase energy consumption. Collision increases latency also.

(ii) Overhearing: The second is overhearing, meaning that a node picks up packets that are destined to other nodes.

(iii) Packet Overhead: The third source is control packet overhead. Sending and receiving control packets consumes energy too and less useful data packets can be transmitted.

(iv) Idle listening: The last major source of inefficiency is idle listening i.e., listening to receive possible traffic that is not sent. This is especially true in many sensor network applications. If nothing is sensed, the sensor node will be in idle state for most of the time. The main goal of any MAC protocol for sensor network is to minimize the energy waste due to idle listening, overhearing and collision.

2.3 MAC Performance Matrices

In order to evaluate and compare the performance of energy conscious MAC protocols, the following matrices are being used by the research community.

(i) Energy Consumption per bit: - The energy efficiency of the sensor nodes can be defined as the total energy consumed / total bits transmitted. The unit of energy efficiency is joules/bit. The lesser the number, the better is the efficiency of a protocol in transmitting the information in the network. This performance matrices gets affected by all the major sources of energy waste in wireless sensor network such as idle listening, collisions, control packet overhead and overhearing.

(ii) Average Delivery Ratio: - The average packet delivery ratio is the number of packets received to the number of packets sent averaged over all the nodes.

(iii) Average Packet Latency: - The average packet latency is the average time taken by the packets to reach to the sink node.

(iv) Network Throughput:-The network throughput is defined as the total number of packets delivered at the sink node per time unit.

3 PROPOSED MAC PROTOCOLS

The medium access control protocols for the wireless sensor networks can be classified broadly into two categories: Contention based and Schedule based. The schedule based protocol can avoid collisions, overhearing and idle listening by scheduling transmit & listen periods but have strict time synchronization requirements. The contention based protocols on the other hand relax time synchronization requirements and can easily adjust to the topology changes as some new nodes may join and others may die few years after deployment. These protocols are based on Carrier Sense Multiple Access (CSMA) technique and have higher costs for message collisions, overhearing and idle listening. 3.1 IEEE 802.11

The IEEE 802.11 [19] is a well known contention based medium access control protocol which uses carrier sensing and randomized back-offs to avoid collisions of the data packets. The Power Save Mode (PSM) of the IEEE 802.11 protocol reduces the idle listening by periodically entering into the sleep state. This PSM mode is for the single-hop network where the time synchronization is simple and may not be suitable for multi-hop networks because of the problems in clock synchronization, neighbour discovery and network partitioning.


3.2 PAMAS: Power Aware Multi-Access Signaling

PAMAS: Power Aware Multi-Access [15] is one

of the earliest contention based MAC protocol designed with energy efficiency as the main objective. In this protocol nodes which are not transmitting or receiving are turned “OFF” in order to conserve energy. This protocol uses two separate channels for the data and control packets. It requires the use of two radios in the different frequency bands at each sensor node leading to the increase in the sensors cost, size and design complexity. Moreover, there is significant power consumption because of excessive switching between sleep and wakeup states. 3.3 Sensor S-MAC

Sensor S-MAC [2] a contention based MAC protocol is modification of IEEE 802.11 protocol specially designed for the wireless sensor network in 2002. In this medium access control protocol sensor node periodically goes to the fixed listen/sleep cycle. A time frame in S-MAC is divided into to parts: one for a listening session and the other for a sleeping session. Only for a listen period, sensor nodes are able to communicate with other nodes and send some control packets such as SYNC, RTS (Request to Send), CTS (Clear to Send) and ACK (Acknowledgement). By a SYNC packet exchange all neighbouring nodes can synchronize together and using RTS/CTS exchange the two nodes can communicate with each other. The basic S-MAC scheme where node 1 transmits data to node 2 is shown in Fig. 1. A lot of energy is still wasted in this protocol during listen period as the sensor will be awake even if there is no reception/transmission.

Figure 1: Basic S-MAC Scheme, Node 1 Transmits Data to Node 2 3.4 Timeout T-MAC

Timeout T-MAC [3] is the protocol based on the S-MAC protocol in which the Active period is pre-empted and the sensor goes to the sleep period if no

activation event has occurred for a time ‘Ta’ as shown in Fig. 2. The event can be reception of data, start of listen/sleep frame time etc. The time ‘Ta’ is the minimal amount of idle listening per frame. The interval Ta > Tci + Trt + Tta + Tct where Tci is the length of the contention interval, Trt is the length of an RTS packet, Tta is the turn-around time (time between the end of the RTS packet and the beginning of the CTS packet) and Tct is the length of the CTS packet. The energy consumption in the Timeout T-MAC protocol is less than the Sensor S-MAC protocol. But the Timeout T-MAC protocol has high latency as compared to the S-MAC protocol.

Figure 2: Basic T-MAC Scheme

3.5 Optimized MAC

In the Optimized MAC protocol [5], the sensors duty cycle is changed based on the network load. If the traffic is more than the duty cycle will be more and for low traffic the duty cycle will be less. The network load is identified based on the number of messages in the queue pending at a particular sensor. The control packet overhead is minimized by reducing the number and size of the control packets as compared to those used in the S-MAC protocol. This protocol may be suited for applications in which apart from energy efficiency there is need for low latency. 3.6 Traffic Adaptive Medium Access Protocol

(TRAMA)

The traffic adaptive medium access (TRAMA) [6] is a TDMA based protocol that has been designed for energy efficient collision free channel in WSNs. In this protocol the power consumption has been reduced by ensuring collision free transmission and by switching the nodes to low power idle state when they are not transmitting or receiving. This protocol consists of three main parts: a) The Neighbor Protocol is for collecting the information about the neighboring nodes b) The Schedule Exchange Protocol is for exchanging the two-hop neighbor information and their schedule c) The Adaptive Election Algorithm decides the transmitting and receiving nodes for the current time slot using the neighborhood and schedule information. The other nodes in the same time slot are switched to low power mode.


The TRAMA is shown to be more energy efficient and has higher throughput than Sensor S-MAC protocol. However, the latency of TRAMA is more as compared to the other contention based MAC protocol such as S-MAC and IEEE 802.11. The delay performance obtained by the analytical model for TRAMA and NAMA [7] shows that TRAMA has higher delays than NAMA. This protocol may be suitable for applications which are not delay sensitive but require higher energy efficiency and throughput.

3.7 Self Organizing Medium Access Control

for Sensor Networks (SMACS) SMACS [9] is a schedule based medium access

control protocol for the wireless sensor network. This MAC protocol uses a combination of TDMA and FDMA or CDMA for accessing the channel. In this protocol the time slots are wasted if the sensor node does not have data to be sent to the intended receivers. This is one of the drawbacks of this MAC scheme. 3.8 Aloha with Preamble Sampling

Aloha with Preamble Sampling is proposed in [11] where the ALOHA protocol [20] has been combined with the preamble sampling technique. The main draw back of the Carrier Sense Multiple Access (CSMA) is the energy wastage due to idle listening. El-Hoiydi in [11] proposed low power listening technique that efficiently duty cycles the radio (i.e., turns it ON periodically).This approach works at the physical layer based on the PHY Header going to sensor’s radio. The Header starts with the Preamble which intimates the receiver of upcoming messages. The receiver periodically turns radio ON to sample for the incoming messages and if the preamble is detected, it continues listening for the normal message transfer. If the preamble is not detected it turns OFF radio till next sample. This carrier sensing approach as shown in Fig. 3 was combined with ALOHA by El-Hoiydi in [11] and named it Aloha with Preamble Sampling which is suitable for low traffic wireless sensor network applications. This paper also presents the power consumption, delay performance and life time computed by analytical methods.

Figure 3: Low Power Listening and Preamble Sampling

3.9 WiseMAC

The WiseMAC [14] medium access control protocol was developed for the “WiseNET” wireless sensor network. This protocol is similar to Spatial TDMA and CSMA with Preamble Sampling protocol [13] where all the sensor nodes have two communication channels. TDMA is used for accessing data channel and CSMA is used for accessing control channel. However, WiseMAC [14] needs only one channel and uses non-persistent CSMA with preamble sampling technique to reduce power consumption during idle listening. This protocol uses the preamble of minimum size based on the information of the sampling schedule of its direct neighbors. The sleep schedules of the neighboring nodes are updated by the acknowledgement message (ACK) during every data transfer. WiseMAC is adaptive to the traffic loads and provides low power consumption during low traffic and high energy efficiency during high traffic. The simulation results show that WiseMAC performs better than S-MAC protocol.

3.10 Berkeley a Access Control (B-MAC)

The Berkeley Media Access Control (B-MAC)

[10] is a contention based MAC protocol for WSNs. B-MAC is similar to Aloha with Preamble Sampling [11], which duty cycles the radio transceiver i.e. the sensor node turns ON/OFF repeatedly without missing the data packets. However in B-MAC, the preamble length is provided as parameter to the upper layer. This provides optimal trade-off between energy savings and latency or throughput. The paper also presents an analytical model for monitoring application to calculate and set B-MAC parameters in order to optimize the power consumption. The experimental results show B-MAC has better performance in terms of latency, throughput and often energy consumption as compared to S-MAC.

3.11 Energy Aware TDMA Based MAC

Energy Aware TDMA Based MAC [16] protocol

assumes the formation of clusters in the network. Each of the cluster sensor nodes is managed by the Gateway. The Gateways collects the information from the other sensor nodes within its cluster, performs the data fusion, communicates with the other gateways and finally sends the data to the control center. The assignment of the time slots to the sensor nodes within its cluster is performed by Gateways. The Gateways inform to the other nodes about the time slot when it should listen to other nodes and the time slot when it can transmit own data.

This TDMA based MAC protocol consist of four main phases: data transfer, refresh, event triggered-


rerouting and refresh-based rerouting. The data transfer phase is for sending the data in its allocated time slot. During refresh phase, the nodes update its state (energy level, state, position etc) to the gateway. The gateway requires this nodes state information for performing rerouting during event triggered-rerouting. The refresh-based rerouting occurs periodically after the refresh phase. During both these rerouting phases the gateway execute the routing algorithms and sends new routes to the sensor nodes.

The paper presents two approaches for slot assignment based on graph parsing strategy: Breadth First Search (BFS) and Depth First Search (DFS). BFS technique, assigns the time slot numbers starting from outer most sensor node giving them contiguous slots. While DFS technique assigns contiguous time slots for the nodes on the route from outermost sensor node to the gateway.

Simulations have been performed for energy consumption per packet, end-to-end delay, throughput, nodes lifetime etc. against the buffer size for both BFS and DFS techniques. BFS saves the energy consumption in switching between the ON & OFF states and therefore the nodes lifetime is high. This technique requires the nodes to have sufficient buffer capacity. While DFS does not save the energy consumption of switching between the ON & OFF states but avoids buffer overflow problem. However, DFS has low latency and high throughput as compared to BFS.

3.12 Data Gathering MAC (D-MAC)

The Data–Gathering Medium Access Control (D-

MAC) [12] is a schedule based MAC protocol which has been designed and optimized for tree based data gathering (converge cast communication) in wireless sensor network. The main objective of this MAC protocol is to achieve low latency and still maintaining the energy efficiency. In this protocol the time is divided in small slots and runs carrier sensing multiple access (CSMA) with acknowledgement within each slot to transmit/receive one packet. The sensor node periodically executes the basic sequence of ‘1’ transmit, ‘1’ receive and ‘n’ sleep slots. In this approach a single packet from a source node at depth ‘k’ in the tree reaches the sink node with a delay of just ‘k’ time slots. This delay is very small and it is in the order of tens of milliseconds. A data gathering (converge cast) tree with staggered DMAC slots is shown in Fig. 4.

D-MAC includes an overflow mechanism to handle the problem when each single source node has low traffic rate but the aggregate rate at intermediate node is larger than the basic duty cycle. In this mechanism the sensor node will remain

awake for one extra time slot after forwarding the packet.

Figure 4: Data gathering tree in D-MAC scheme Therefore, if two children were contending for

parents receive slot, the loosing child will get a second chance to send its packet. The D-MAC uses a separate control packet named MTS (More to Send) to solve the problem of the interference between nodes on the different branches of the tree. The MTS packet makes all the nodes on the multi-hop path to remain active in case of nodes failure due to interference.

The simulation results shows that the D-MAC protocol outperforms the Sensor S-MAC protocol in terms of energy efficiency, latency and throughput in both multi-hop chain topology and random data gathering tree topology.

4 FUTURE RESEARCH DIRECTIONS

In the recent years a large number of medium

access control (MAC) protocols for the wireless sensor network have been published by the researchers. Most of the work on the MAC focuses primarily on the energy efficiency in the sensor network [8]. However, still a lot of work has to done in the other areas at the MAC layer such as: (i) Network Security: - Sensor network security at

MAC layer to protect against eavesdropping and malicious behavior has to be studied further. Karlof et al. in TinySec [22] have proposed secure MAC protocol based on shared key but still more advanced schemes needs to be developed.

(ii) Nodes Mobility: - The nodes in the wireless sensor network were originally assumed to be static. Recently there has been increasing interest in medical care and disaster response applications where the mobile sensors can be attached to the patient, doctor or first responder. The mobility at the MAC layer has been considered in MMAC [21], still there is a lot of scope for future research in this area.


(iii) Evaluation on Sensor Platforms: - Most of the protocols for the wireless sensor network have been evaluated through the simulations. However, the performance of the MAC protocol needs to be evaluated on the actual sensor system. The researchers should focus on experimenting on the real sensor platforms.

(iv) Real Time Systems: - Energy efficiency is the main design objective of the sensor network but the reliable delivery of data in the real time is essential for certain time critical applications. This is also a promising research area which needs to be studied more extensively.

5 CONCLUSIONS Recently several medium access control protocols

for the wireless sensor network have been proposed by the researchers. However, no protocol is accepted as standard. This is because the MAC protocol in general will be application specific. Therefore, there will not be one standard MAC protocol for the WSNs. The schedule based (TDMA) have collision free access to the medium but the synchronization is critical. Moreover, there is difficulty in adapting to the changes in the network topology because of the addition and deletion of nodes. The contention based (CSMA) have low latency and high throughput. However, it still suffers from the collisions. The Frequency Division Multiple Access (FDMA) scheme also allow collision free access to the media but the extra circuitry required to dynamically communicate with different radio channels increases the cost of the sensor nodes. This contradicts the main objective of the wireless sensor networks (WSNs). The Code Division Multiple Access (CDMA) scheme also offers collision free access to the medium. However, the high computational complexity is the limitation in the lower energy consumption needs of the sensor network.

6 REFRENCES [1] I. Akyildiz, W. Su, Y. Sankarasubramaniam

and E. Cayirci: A Survey on Sensor Networks, IEEE Communication Magazine, pp. 102-114 (August 2002).

[2] Wei Ye, J.Heidemann and D. Estrin: An Energy-Efficient MAC Protocol for Wireless Sensor Networks, IEEE INFOCOM, New York, Vol. 2, pp. 1567-1576 (June 2002).

[3] Tijs van Dam, Koen Langendoen: An Adaptive Energy Efficient MAC Protocol for Wireless Networks, in Proceedings of the First ACM

Conference on Embedded Networked Sensor Systems (November 2003).

[4] Changsu Suh, Young-Mi Song, Young-Bee Ko, and We Duke Cho: Energy Efficient & Delay Optimized MAC for Wireless Sensor Networks, in Proceedings of the Workshop in the Seventh International Conference on Ubiquitous Computing (Ubicomp’05) (September 2005).

[5] Rajesh Yadav, Shirshu Varma and N.Malaviya: Optimized Medium Access Control for Wireless Sensor Network, IJCSNS International Journal of Computer Science and Network Security, Vol. 8, No.2, pp. 334 -338 (February 2008).

[6] V. Rajendran, K. Obraczka and J.J. Gracia-Luna-Aceves: Energy Efficient, Collision Free Medium Access Control for Wireless Sensor Networks, in ACM International Conference on Embedded Networked Sensor Systems (SenSys), pp. 181-192 (November 2003).

[7] L. Bao and J.J. Garcia-Luna-Aceves: A New Approach To Channel Access Scheduling for Ad Hoc Network, in Seventh Annual International Conference on Mobile Computing and Networking, pp. 210-221 (2001).

[8] M. Ali, Saif, A. Dunkels, T. Voigt, K. Romer, K. Langendoen, J. Polastre, Z. A. Uzmi: Medium Access Control Issues in Sensor Networks, ACM SIGCOMM Computer Communication Review, Vol. 36, No. 2 (April 2006).

[9] K. Sohrabi, J.Gao, V.Ailawadhi and G.J.Pottie: Protocols for Self Organization of a Wireless Sensor Network, IEEE Personal Communication, Vol. 7, Issue 5, pp. 16-27 (October 2000).

[10] J. Polastre, J. Hill, D. Culler: Versatile low Power Media Access for Wireless Sensor Networks, Proceedings of the 2nd ACM Conference on Embedded Networked Sensor Systems (SenSys’04), Baltimore, MD, (November 2004).

[11] A. El-Hoiydi: Aloha with Preamble Sampling for Sporadic Traffic in Ad-hoc Wireless Sensor Networks”, in Proceedings of IEEE International Conference on Communications (April 2002).

[12] G. Lu, B. Krishnamachari, C. Raghavendra: An Adaptive Energy Efficient and Low Latency MAC for Data Gathering in Wireless Sensor Networks, Proceedings of 18thth International Parallel and Distributed Processing Symposium (April 2004).

[13] A. El-Hoiydi: Spatial TDMA and CSMA with Preamble Sampling for Low Power Ad-hoc Wireless Sensor Network, Proceedings of ISCC ‘02, Seventh International Symposium on Computers and Communications, pp. 685-692 (July 2002).


[14] C.C. Enz, A. El-Hoiydi, J.-D. Decotignie, V. Peiris: WiseNET: An Ultralow-Power Wireless Sensor Network Solution, IEEE Computer, Vol. 37, Issue 8 (August 2004).

[15] S. Singh and C. Raghavendra: PAMAS: Power Aware Multi-Access Protocol with Signaling for Ad-hoc Network, ACM SIGCOMM Computer Communication Review (July 1998).

[16] K. Arisha, M. Youssef and M. Younis: Energy Aware TDMA based MAC for Sensor Network, in IEEE Workshop on Integrated Management of Power Aware Communications Computing and Networking (IMPACCT’02) (2002).

[17] Zhihui Chen, Ashfaq Khokhar: Self Organisation and Energy Efficient TDMA MAC Protocol by Wakeup for Wireless Sensor Networks, in Proceedings of the IEEE Conference (SECON’04) (August 2004).

[18] M. Ali, Saif, A. Dunkels, T. Voigt, K. Romer, K. Langendoen, J. Polastre, Z. A. Uzmi: Medium Access Control Issues in Sensor Networks, ACM SIGCOMM Computer Communication Review, Vol. 36, No. 2 (April 2006).

[19] IEEE Standard 802.11. Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications (1999).

[20] N. Abramson: The ALOHA System – Another Alternative for Computer Communications, in Proceedings Fall Joint Computer Conference, AFIPS Press, Vol. 37, pp. 281-285 (1970).

[21] M. Ali, T. Suleman, and Z. A. Uzmi: MMAC: A Mobility Adaptive , Collision Free MAC Protocol for Wireless Sensor Networks, in Proceedings 24th IEEE IPCCC'05, Phoenix, Arizona, USA (April 2005).

[22] C. Karlof, N. Sastry, and D. Wagner: TinySec: A Link Layer Security Architecture for Wireless Sensor Networks", in Proceedings SenSys'04, pp. 162-175 (November 2004).

Rajesh Yadav completed his B.Tech (Hons) in Computer Science and Engineering from Bundel-khand Institute of Engineering and Tech-nology, JHANSI (U.P), India in 1993. He obtained his M.Tech from Dayalbagh University AGRA (U.P) in 1998. He

joined Electronics and Radar Development Establishment, Defence R&D Organization (DRDO), Bangalore in 1998. Presently he is working as Scientist ‘D’ and his areas of interest are wireless sensor networks and real time embedded systems for

radar applications. Before joining DRDO, he has worked as Lecturer in Computer Sc.& Engg. Department in Kumaon Engineering College, ALMORA (Uttranchal) from 1995 to 1998.

Shirshu Varma graduated in Electronics and Commu-nication Engineering from Allahabad University and post graduated in Commu-nication Engineering from BIT Mesra Ranchi, India. He completed his Ph.D in Optical Communication from University of Lucknow. He

has served many organizations like BIT Mesra Ranchi, IET Lucknow, C-DAC Noida in the capacity of lecturer, Sr. lecturer & IT Consultant. Presently he is working Assistant Professor in IIIT Allahabad. Dr. Varma has published about 27 papers in international and national journals and conferences of repute. He is a member of IEEE and life member of ISTE. He has been a recipient of many national awards in this area. His areas of interest are intelligent sensor network, wireless sensor network, Optical wireless communication, Wireless communication & network.

N. Malaviya worked as Prof & Head Electronics Department at Institute of Engineering and Tech-nology, Lucknow (U.P), India. He completed his Ph.D and M.Tech from Indian Institute of Technology, Roorkee. He

has over thirty years of teaching and research experience. He has guided 10 Ph.D students and several M.E and B.Tech students. He was also Dean Research in U.P Technical University, Lucknow (U.P).


ADAPTIVE CALL ADMISSION CONTROL IN TDD-CDMA CELLULAR WIRELESS NETWORKS

Dhananjay Kumar 1, Chellappan C2

1Department of Information Technology, Anna University, Chennai 1 [email protected]

2Department of Computer Science & Engineering, Anna University, Chennai 2 [email protected]

ABSTRACT The Code Division Multiple Access system with Time Division Duplex mode (TDD-CDMA), adopting unbalanced slot allocation between uplink and downlink, can meet the asymmetric traffic requirement of multimedia services. Here a call admission control policy is proposed to support different multimedia applications. The scheme operates at the connection-level where the CDMA code of ongoing call can be dynamically changed to provide an acceptable trade off level between connection blocking and dropping probabilities for different traffic class. Although Orthogonal Variable Spreading Factor (OVSF) code is used in simulation here, the paramount interest is on algorithm that allows optimum use of the TDD-CDMA resources i.e. code and time slots. Simulation results shows that on the expense of resource allocated to non real-time services, the call dropping and blocking rate for high priority (real-time) services can be minimized.

Keywords: TDD-CDMA; Interference; Multimedia; Call Blocking/Dropping.

1 INTRODUCTION In future wireless networking environments, the data traffic for Internet, real-time voice, and multimedia traffic will coexist. For the multimedia application such as streaming audio/video or web services, the downlink traffic will be the bottleneck of the system. On the other hand, the uplink traffic may be bursty and irregular when mobile users use the application like file uploading services. The dynamic change of the traffic asymmetry between uplink and downlink makes the resource allocation of the future wireless system difficult. The code division multiple access system with time division duplex mode (TDD-CDMA) is a promising solution to cope with the traffic asymmetry problem [1, 2].

The TDD-CDMA mode of 3GPP, named UTRA-TDD (Also called UMTS-TDD), is based on TD-CDMA technology, which is a mixture of TDMA and CDMA [3]. In Frequency Division Duplex (FDD) mode, a pair of frequency band is chosen for uplink and downlink communication, but in TDD same frequency is used for both directions (Fig.1). Although FDD is dominant candidate in W-CDMA (Originally developed by NTT DoCoMo, Japan), which uses a pair of 5 MHz-wide radio channels, the TDD mode is more popular in the case of micro/pico

Fig.1 FDD and TDD mode of communication

cell environment because of its capacity and flexibility to support asymmetric traffic [4,5].

Even though TDD-CDMA is not directly compatible with UMTS (Which is based on W-CDMA), it is closely related to W-CDMA, and

Upstream Frequency Band

Downstream

Time Division Duplexing (TDD)

Frequency Division Duplexing (FDD)

Same Upstream & Downstream Frequency band


provides the same types of channels where possible. In this paper, we investigate and analyze an adaptive algorithm for resource allocation in TDD-CDMA system which is one of the candidates for multiple access technique for the fourth generation (4G) systems [2, 6].

The resource of TDD-CDMA is divided in both time and code domain. In a cell of a TDD-CDMA system, the base station and all mobiles occupy a single band employing a direct sequence spreading waveform. The bidirectional communication between base and mobiles is accomplished by a TDD scheme [7]. Each slot in TDD carries traffic for different mobile station separated by CDMA code.

Figure2 shows an example of the TDD frame structure. A TDD frame consists of a fixed number of time slots. At least one slot is dedicated to uplink (from mobile to base) and at least another one slot is dedicated to downlink (from base to mobile). The number of uplink and downlink slot can be adaptively controlled by the base station.

Fig.2 Uplink and downlink in a TDD frame

In this paper, we characterize services into two

types: adaptive, and non-adaptive. In the case of a non-adaptive service, the bandwidth of a call is fixed throughout its lifetime and needs strict bandwidth guarantees, else the call will be dropped. But in the case of an adaptive service, the call will not be dropped, but will suffer bandwidth degradation. Many real-time multimedia services are adaptive in nature and can operate over a wide range of bandwidth. Adaptive Code Allocation (ACA) algorithm is called to support the bandwidth requirement of a new high priority calls in the system. The ACA algorithm decreases assigned bandwidth of ongoing connections in the cell depending on the network load situation, by allocating dynamic OVSF code.

N. Nasser and H. Hassanein [8,9], has proposed an adaptive framework to support multimedia applications, but not with respect to any existing systems. Further it does not consider the multiple access techniques which inherently govern the resource allocation & hence call control mechanism. N. Nasser again in a similar paper [10] talks about adaptability enhancement framework, but once again without considering under laying techniques. Zhihua Zheng [11] has proposed an efficient dynamic channel assignment with channel locked for TDD-CDMA communication considering three sectored cell. Ioannis Spyropoulos et al.[1] has proposed a

decentralized scheme for TDD-CDMA systems, which combines an interference-aware dynamic channel allocation algorithm with space–time linear minimum-mean-square-error (LMMSE) joint detection at the base and mobile stations. Their analysis includes outage and average throughput via analytical approximations.

The proposed algorithm here, aims to support multimedia calls while optimizing the resource allocation in TDD-CDMA system, thereby reducing both the blocking and dropping rates of real-time multimedia calls. The proposed Call Admission Control (CAC) algorithm is capable of offering services even when there is insufficient number of codes available by intelligently swapping the OVSF codes of ongoing calls. This results in reduced volume of the blocking and dropping rates; thus striking a proper balance between Quality of Service (QoS) fulfilment and code/bandwidth utilization.

This paper is organized as follows. In Section2 interference pattern for TDD-CDMA system is presented. An introduction to OVSF and CAC algorithm is explained in Section3. The simulation environment for the proposed adaptive algorithm is presented in Section4. Simulated result is discussed in Section5. We conclude with an overview of the simulation carried out in Section6. 2 INTERFERENCE PATTERN IN TDD-CDMA

In TDD-CDMA each slot will carry many user

data on different channel separated and identified by pseudo noise (PN) code. Considering intra-cell and inter-cell interference separately in a multi-cell environment, the bit energy to noise ratio can be modelled as

WNII

SFPNEext

rb .

./0int

0 ++= (1)

Where Pr is the received power, SF is the spreading factor, Iint is the internal noise within the cell, Iext is external noise coming from other cells, No is the noise power spectral density, and W is the total transmission bandwidth. 2.1 Interference in Uplink channels

Suppose mk be the number of MS served by a channel, where k = 1,2,3,…..K, represent the kth channel to support K type of services in a cell. Let Pki denote the transmit power of ith MS to maintain certain quality of service (QoS) for a kth service, and Gki the gain between ith MS and it’s BS. The internal interference Iint in uplink for kth channel carrying data of Ith MS may be given by

(2)

DL DLUL

TDD Frame

DL DL DL DL DL ULUL

∑ ∑≠= =

≠

=K

Ikk

m

iki ki

k

Ii

P GI1 1

int


A multi-user detection (MUD) factor (1-β), where β is MUD efficiency, is multiplied with the external interference Iint to achieve 3GPP requirements for CDMA-TDD [12]. In other words, β is an interference reduction factor. For example, MUD in uplink, β = l is a case of ideal MUD, while β = 0 represents absence of interference cancellation technique and hence employing a rake filter.

To compute external interference Iext , let ml be the number of MS served on lth channel , where l = 1,2,…..L represents L channels in the neighbour cell j supporting L type of multimedia application. Now Iext may be expressed as

∑ ∑ ∑= = =

=J

j

L

l

m

i

ijl

ijlext

l

PGI1 1 1

(3)

Where J is the number of interfering neighbour cells, Gi

jl the link gain between ith MS in neighbour cell and the tagged BS, and Pi

jl the transmit power of ith MS to support its QoS requirement in its cell. 2.2 Interference in Downlink Channels

Data in downlink channels (for example in W-CDMA) are transmitted with orthogonal codes; in other word, they are coded such that mutual interference is minimal. Assuming perfect time synchronization between MS and BS, and if the type of channel is flat fading i.e. if the orthogonality is preserved during downlink slot, then the internal noise Iint is absent. But the multipath propagation destroys some of this orthogonality in downlink. An orthogonality factor (α) which is the percentage of downlink orthogonality remaining at the mobile receiver, is introduce to compute Iint . Now, the internal interference Iint arising due to non-orthogonality of the received signals is given by

∑=

=L

lltli PGI

1int α (4)

Where Plt is the total base station power allocated to signals using the same scrambling code for lth channel , Gli thelink gain between ith MS and tagged BS for the same lth channel.

To compute the external interference Iext , we take the advantages of similarity with (3) and can be represented as

∑ ∑ ∑= = =

=J

j

L

l

m

i

ijl

ijlext

l

PGI1 1 1

(5)

Where Gijl is the link gain between ith tagged MS

and a MS in neighbor Jth cell, ml the number of MS in a jth cell, and Pi

jl the transmit power of a MS in cell J.

3 OVSF CODE AND CAC ALGORITHM The use of OVSF codes (Fig.3) to support wide

variety of multimedia calls has been widely advocated [13,14]. Different spreading factor (SF) means different code length. The requirement is to combine different messages with different spreading factors and keep the orthogonality between them. We therefore need codes of different length that are still orthogonal. It is assumed that voice traffic will require constant bandwidth but other components of multimedia like images, audio/video streaming will demand higher variable data rate support. So, the resource pool maintains OVSF code representing different data rate R, 2R, 3R…. corresponding to SF = 1, 2, 3,…. 3.1 CAC Algorithm

When a new or a handoff call of a class arrives in a cell, the base station (BS) calls the CAC algorithm (Fig.4). CAC selects a code form resource corresponding to the types service requested. After allocating a code, the Eb/No is computed for the current slot. If SNR falls below the threshold (γm), next code with higher spreading factor is selected. If no higher SF codes are available in resource pool, and if the call has higher priority, then ACA is called. The ACA finds whether an existing low priority non-real time call can be bandwidth degraded, and hence the existing code is swapped with a low SF code. On failure of that, the subsequent slot is declare the same status, and the new call is accommodated in the newly declared slot. If the subsequent slot cannot be declared same (UL or DL) because of the existing traffic pattern, then the new call is rejected. A new call may be a handoff calls which have higher priority than the new calls originating from mobile user. 4 SIMULATION ENVIRONMENT

Simulation environment to represent TDD-CDMA was implemented using Network Simulator-2 (NS-2). A Universal Mobile Telecommunications System (UMTS) patch was incorporated in existing NS-2 to create the cellular environment. A single cell environment with a BS and radio network controller (RNC) in UMTS is created, but interference pattern were implemented corresponding to the multi-cell scenario. As per UMTS specification in a frame length of 10ms, 15 time slots were considered. In the simulation, 150 user equipment (UE) were created, allowing them to move randomly across the cell and make random request for calls according to Poisson rate. To communicate with the BS, UE uses dedicated channels assigned to them.


Fig.3 Orthogonal variable spreading factor code

Following are the major functions implemented in

NS-2. 4.1 Request for Connection by UE

New call arrival rate is assumed to follow Poisson Process with rates λ. i.e. λ calls per second. In the simulation, λ = 1 call/sec to 16 calls/sec were considered and each runs for about 10 seconds. A UE sends request for call connection to BS by sending a request packet. The request packet contains the class type, call type and link required by the call. 4.2 Processing Request by BS

The BS receives random call request from the users according to Poisson distribution. This call request may be for a new call or a handoff call. The BS on receiving request packet for a new call, try to allocate available code (bandwidth) efficiently among users, which is done by calling the CAC algorithm. Also, the BS maintains details of currently ongoing calls for each class in linked lists.

4.3 Termination of Connection by UE

We have also considered call holding time, with an upper bound, which is the waiting time between the period, when the call gets connected and the resource gets allocated. Once the timer corresponding to call holding expires, a terminate packet is sent to BS. On receiving this packet BS calls the terminate procedure to free the resources allocated for that call. 4.4 Recording by BS

For each λ value for call arriving, the simulation is run for 10 seconds and the BS maintains the number of calls blocked and dropped in the process. These will be recorded after the entire simulation and will be used for plotting graphs. 4.5 Segregation of Calls

We consider two types of calls, namely class1 and class2. Time insensitive calls are considered in class1, e.g. Web Browsing, Data upload/download, email services etc. Time sensitive calls are considered in class2, e.g. Audio/Video streaming, Telephonic etc. Handoff calls considered here falls under class2 i.e. it has higher priority than class1 service.

C1 = (1, 1, 1, 1, 1, 1, 1, 1)

C1 = (1, 1, 1, 1, ‐1, ‐1, ‐1, ‐1)

C1 = (1, 1, ‐1, ‐1, 1, 1, ‐1, ‐1)

C1 = (1, 1, ‐1, ‐1, ‐1, ‐1, 1, 1)

C1,1 = (1)

C2,1 = (1, 1)

C2,1 = (1, ‐1)

C2,1 = (1, 1, 1, 1 )

C2,1 = (1, 1, ‐1, ‐1 )

C2,1 = (1, ‐1, 1, ‐1 )

C2,1 = (1, ‐1, ‐1, 1 )

C1 = (1, ‐1, 1, ‐1, 1, ‐1, 1, ‐1)

C1 = (1, ‐1, 1, ‐1, ‐1, 1, ‐1, 1)

C1 = (1, ‐1, ‐1, 1, 1, ‐1, ‐1, 1)

C1 = (1, ‐1, ‐1, 1, ‐1, 1, 1, ‐1)

SF = 1 SF = 2 SF = 4 SF = 8


Fig.4 Flow chart of CAC algorithm 5 RESULTS

The record created by BS after the simulation is

used to plot connection dropping and blocking rate. 5.1 Call Dropping

Fig.5 shows call dropping rate vs. call arrival rate (Poisson Rate). As expected, as the call arrival rate increases the call dropping rate increases. Also

the call dropping rate for class1 is greater than class2. This is because class2 has higher priority than class1. We also note that, for a call arrival rate λ = 6, for both class1 and class2, the call dropping rate is less than 0.5, In particular for class2, it is less than 0.2.

Yes

No

No

No

No

Allocate the code for ith call

Update code resource pool

Pick‐up next unallocated call

No

Yes

All call accommodate

End

Yes

Call ACA procedure

Code available

YesNo

Is it class2 call Yes

Find if it can be accommodated in next

slot

Next slot

Reject ith call

Declare next slot in same direction UL/DL

Yes

Is SF > Am

Choose a code with next higher SF

Start

Is SNRi > γm

Select a code corresponding to the BWi

requirement

Estimate SNRi in the current slot


5.2 Call Blocking The plot in Figure6 corresponds to call blocking

rate vs. call arrival rate (Poisson Rate). Call blocking rate gradually increases as arrival rate increases, and for Class2 that includes handoff calls has higher priority, blocking rate is low. Although for a call arrival rate λ = 6, call blocking rate is around 0.55 for class 1, it is less than 0.3 for class 2.

Fig.5 Call dropping rate vs. arrival rate

Fig.6 Call blocking vs. arrival rate

6 CONCLUSION

A call admission control algorithm for multimedia

call is implemented in TDD-CDMA system. The algorithm considers three parameters namely SNR, available OVSF code, and time slot. The framework achieves a better connection dropping and blocking rate for higher priority calls on the expense of non-real time lower priority calls. Class2 calls include handoff calls which are of higher priority in any systems. Simulation result shows that connection dropping rate for class2 is less than 0.2 for an arrival rate λ = 6. For the same parameter (λ = 6) although connection blocking rate for class 1 is more than 0.5, for class2, it is below 0.3

Although we have categorized only two classes of calls, there could be more than two, and exhaustive simulation needs to be carried out to study the connection dropping and blocking rate separately.

During system operation, and because of the statistical nature of the arrival and departure processes, the occupied codes will be randomly scattered across the code tree, so countermeasures need to be taken. 7 REFERENCES [1] Ioannis Spyropoulos, and James R. Zeidler:

Supporting Asymmetric Traffic in a TDD/CDMA Cellular Network via Interference-Aware Dynamic Channel Allocation and Space–Time LMMSE Joint Detection, IEEE Transaction on Vehicular Technology Vol.58, No.2, February (2009).

[2] Ming Yang, and Peter Han Joo Chong: Uplink Capacity Analysis for Multihop TDD-CDMA Cellular System, IEEE Transaction on Communication Vol.57, No.2, February (2009).

[3] 3GPP Technical Specification for CDMA-TDD:TS25.221, TS25.222, TS25.223, and TS25.224. http://www.3gpp.org.

[4] Overview of The Universal Mobile Telecommunication System, http://www.umtsworld.com/technology/overview.htm.

[5] Maryam Arabshahi and Peter Han Joo Chong: High-Speed Multimedia Services for TDD-CDMA Multihop Cellular Networks, International Conference on Wireless Communications and Mobile Computing, 2008, IEEE xplore (2008).

[6] R. Esmailzadeh, M. Nakagawa, and Alan Jones: TDD-CDMA for the 4th Generation of wireless communications”, IEEE Wireless Communication Magazine, pp.8-15, August (2003).

[7] Ki-Dong Lee; Yi, B.K.; Kun-Nyeong Chang; Leung, V: Information Exchange Based Low-Complexity Slot Allocation in TDD-CDMA Cellular Systems, International Conference on Wireless Communications and Mobile Computing, 2008, IEEE xplore (2008).

[8] N. Nasser and H. Hassanein: Connection-level Performance Analysis for Adaptive Bandwidth Allocation in Multimedia Wireless Cellular Networks, Proceedings of the IEEE International Performance Computing and Communications Conference (IPCCC)-2004, Phoenix, Arizona, pp. 61- 68 (2004).

[9] N. Nasser and H. Hassanein: Prioritized Multi-class Adaptive Framework for Multimedia Wireless Networks, Proceedings of the IEEE International Conference on Communications (ICC)-2004, Paris, France, Vol. 7, pp. 4295-4300 (2004).


[10] Nidal Nasser: Adaptability Enhanced Framework for Provisioning Connection-level QoS in Multimedia Wireless Networks, IEEE and IFIP International Conference on Wireless and Optical Communications Networks (WOCN)-2005, Dubai, UAE, pp. 275-279 (2005).

[11] Zhihua Zheng: An Efficient Dynamic Channel Assignment with Channel Locked for TDD-CDMA Systems, 4th IEEE International Conference on Circuits and Systems for Communications, May 2008. ICCSC (2008).

[12] T. Ojanpera, R. Prasad: Wideband CDMA for Third Generation Mobile Communication”, Artech House Publication (1998).

[13] Sun-Ho Lee and Dong-Ho Cho: OVSF Code Assignment Method Considering Traffic characterisrics in W-CDMA Systems, http://whitepapers.zdnet.co.uk/0,1000000651,260235555p,00.htm.

[14] Angelos N. Rouskas and Dimitrios N. Skoutas: OVSF Codes Assignemnt and Reassignement at the Forward Link of W-CDMA 3G Systems, IEEE PIMRC (2002).
