Upload
avishek-patra
View
27
Download
1
Embed Size (px)
DESCRIPTION
Packet Scheduling for Real-Time Communication Over LTE Systems
Citation preview
Packet Scheduling for Real-Time Communication
over LTE Systems A Comparative Study of Dynamic and Semi-Persistent Scheduling Schemes
Avishek Patra
RWTH Aachen University
Aachen, Germany
Volker Pauli, Lang Yu
Nomor Research GmbH
Munich, Germany
{pauli, yu}@nomor.de
Abstract— With various packet-switched networks coming to
the fore, real-time services like voice and video, transmitted
traditionally using circuit-switched bearers, can have limited
capacity due to the limited availability of resource-granting
control channels. Such packets are frequent and require more
grants compared to other services like FTP. To compound the
issue, often these packets are large in size compared to available
resources for allocation. To improve the capacity of real-time
communication over LTE (-A), various scheduling methods are
being studied. However, often the packet sizes are unaccounted
for by these studies. This work deals with the development of
semi-persistent scheduling (SPS) algorithms based on wide-band
time-average SINR information for resource allocation to voice
traffic users, with a focus on large packets. A comparative study
between dynamic scheduling (DS) and developed SPS algorithms
is done to determine the suitable scheduling mechanism for voice
packets transmission over LTE (-A) systems in the downlink.
Keywords—semi-persistent scheduling; dynamic scheduling;
VoIP; Voice over LTE; radio resource management
I. INTRODUCTION
Over the past decade, there has been a path-breaking shift in the methods of communication, from just voice–based communication to voice– and data–based communication in cellular networks with the global mobile data usage doubling every year since 2006 [1]. To cater to the ever-increasing demand for higher data rates, various data-centric standards such as Long Term Evolution (LTE) and Long Term Evolution – Advanced (LTE-A) standards from 3GPP have come up. To meet the required capacity and coverage, various Radio Resource Management (RRM) algorithms such as Hybrid ARQ (HARQ), Link Adaptation (LA), Channel Quality Indication (CQI) and Packet Scheduling (PS) are implemented in LTE (-A). One of the main feature of LTE (-A) is that it is supported by packet switched bearers instead of circuit switched bearers. Hence, transmission of voice requires a VoIP-based solution in LTE (-A) that would have at least same coverage and capacity as 2G and 3G networks [2].
For data packets in LTE (-A), allocation of resources are done by schedulers in a dynamic fashion. Based on the immediate requirement of the user, allocation of resources is done per Transmission Time Interval (TTI). Physical
Downlink Control Channels (PDCCH) plays the major role in signalling resource allocation. Dynamic allocation works well for data packets, which are infrequent and non-periodic. Compared to this, voice and video traffic are bursty in nature and consists of periodically repeating packets and silence periods. Problem arises with respect to real-time packet scheduling when multiple users try to access resources in a single TTI as the allocations of new grants are limited to the number of available PDCCHs. To compound the problem, real-time packets are often large and available Physical Resource Blocks (PRBs) may not be able to accommodate them. Although to overcome the scheduling problem, different persistent, semi-persistent and modified dynamic resource scheduling algorithms have been studied [3, 4], often the packet sizes are not taken into consideration.
In this work, we aim to study the capacity achievable by full dynamic scheduling and semi-persistent scheduling for the downlink case in multi-user LTE (-A) scenario for voice communication. For this, two novel semi-persistent scheduling algorithms have been proposed, which partition large packets before transmitting them. Based on the partitioning method, these leftovers are transmitted dynamically or semi-persistently. Also, contrary to previous works, which use instantaneous CQI information, this work is based on time-averaged wideband SINR information.
The work is organised as follows: Section II briefly discusses the different scheduling schemes and importance of packet size consideration in scheduling. Section III elaborates the semi-persistent scheduling algorithms developed. In Section IV, the different simulation specifications and scenarios are explained and the results obtained from the comparative study between semi-persistent and dynamic scheduling methods are presented. Finally, we conclude the work in Section V.
II. BACKGROUND CONCEPTS
A. Voice over LTE
In voice traffic, the voice packets are periodic in nature with a period of inter-packet arrival interval. Although voice data rates are low compared to data traffic, being real-time, transmission of voice over LTE (VoLTE) system is highly
978-1-4799-0543-0/13/$31.00 ©2013 IEEE
sensitive to transmission latency and loss of packets. Thus, even though packet switching may increase resource usage efficiency through multiplexing, it has the limited capability of guaranteeing the required QoS. E-UTRAN is expected to meet the capacity limit set by the previous 2G and 3G standards. In this context, a satisfied user is defined as the user receiving 98% of the packets within the acceptable end-to-end delay – also known as the delay budget [5]. For a VoLTE system, the capacity is defined as the number of users served in the cell such that 95% of the users are satisfied [8].
To meet these requirements, robust packet scheduling is required. PDCCH is important in scheduling as it provides user equipments (UEs) information about the scheduled channels – primarily, the allocated Physical Resource Blocks (PRBs), and modulation and coding schemes (MCS) – for both the uplink and downlink. Allocation of resources is normally done using Dynamic Scheduling (DS) algorithm for the data packets. However, as voice packet transmission is periodic with regular voice packets, with the possibility of large number of users vying for resource in a TTI, the number of available PDCCHs for granting new resources can limit the possible allocation. With only 1-3 symbols of each carrier in each PRB allocated for control signal, this may not only increase call drop rate but also may fail in providing the required QoS to the existing users. To overcome this limitation, various resource scheduling algorithms other than dynamic scheduling are studied.
B. Scheduling Schemes
In this section, various scheduling schemes such as dynamic, persistent, semi-persistent and talk-spurt based persistent scheduling are briefly explained.
1) Dynamic Scheduling: In DS, queued packets of the
users are scheduled every TTI by allocating the required PRBs
and the transmission format combination (TFC) to the users
(based on SINR information). As these resource grants are
sent via PDCCHs, VoIP capacity using DS may be limited by
the PDCCH limit. A variation of DS is by using the concept of
packet bundling, where consecutive packets of the same user
are queued and bundled together before being transmitted.
Although this may increase capacity [6], packet bundling may
increase transmission delay.
2) Persistent Scheduling: Persistent scheduling is the
process of allocation of fixed time and frequency resources to
the user along with fixed TFC for the entire call duration or
duration of burst [4]. This is advantageous in comparison to
DS as the capacity is not limited by the available PDCCH.
However, it is highly inefficient in resource utilization as the
resources are dedicated for long durations even though there is
no transmission of VoIP packets. Also, the capacity is limited
to the bandwidth available as maximum capacity possible C =
(1000 x N)/B, where B – bandwidth/user in KHz and N –
Total bandwidth in MHz. Also, lack of link adaptation may
reduce the user experience.
3) Semi-Persistent Scheduling: SPS takes the advantage
of both dynamic and persistent scheduling. In this scheme, the
users are allocated resources for an extended period for
transmission of voice packets. At the end of the burst, the user
resources are deallocated and are allocated to another user.
The TFC may be changed for the duration of the burst based
on the channel state information. Thus, PDCCH are required
only in case of new resource allocation, for changing of the
TFC or transmission power within a burst. For the
retransmission of the packets, they are dynamically scheduled
and this also requires the PDCCH resources.
4) Talk-Spurts based Scheduling: As silence periods
consume half the duration of the talk burst or voice call, talk-
spurt based scheduling [4] aims at allocating PRBs every talk-
spurt and deallocating PRBs at the end of the spurt. TFC
remains same for each spurt. The Silence Insertion Descriptors
(SIDs) are transmitted in a dynamic basis, which consumes
PDCCH but as they are less frequent compared to voice
packets, talk-spurt based scheduling perform better than full
dynamic scheduling. The resources for voice packets can
either be allocated for both transmission and retransmission or
only transmission, with retransmission being dynamically
scheduled.
C. Large VoIP Packets and concept of leftovers
For scheduling algorithm, there are three main processes -
initial allocation, periodic allocation and retransmission of
VoIP packets and SIDs - that have been considered in most of
the literatures [3-7]. An important issue often ignored is
regarding the size of the VoIP packets compared to the
available PRBs for allocation. Conflict in the schedulers
would arise if the required number of PRBs for a user at a
given TTI is greater than the number of available PRBs. It is
important to consider the large packets and the possibility of
transmission of the leftovers any real-time communication
where the QoS depends on the continuous uninterrupted
transmission and reception. Fig. 1 shows the ratio of VoIP
packets requiring leftover to total VoIP packets in the
downlink scenario to underline the importance of considering
leftover transmission.
III. SEMI-PERSISTENT SCHEDULING FOR VOLTE
With the basic idea of DS scheme explained, in this section, we look towards the SPS algorithms for scheduling of
90 100 110 120 130 140 150 160 170 180 190 2000
0.5
1
Tota
l V
oIP
Pac
ket
s an
d
Lef
tov
er V
oIP
Pack
ets
Number of Users
VoIP Packets with Leftovers in Downlink
80 100 120 140 160 180 2005
10
15
Lef
tover
Per
centa
ge
Total Packets
Packets withLeftovers
Percentage ofPackets with Leftovers
10
0
15
19090 130 150 160 170 180 200140120110100
1.0
0.5
0.0
Fig. 1. Ratio of VoIP packets requiring leftover to total VoIP packets
VoIP packets, designed to take into account the effects of large packets. To resolve the problem, two strategies of allocating PRBs have been adopted with variations based on partitioning and transmission of leftovers.
A. Non-Segmentation based Semi-Persistent Scheduling
In Non-Segmentation based Semi-Persistent Scheduling (NS-SPS) scheme, when a packet arrives at the transmission buffer initially, users are allocated resources based on availability and feasibility of PRBs for the initial packet. This is referred as Initial Allocation. For this, new TFCs are chosen and hence, PDCCH is required to signal the grant for initial allocation. For further periodic arrival of packets, PRBs are automatically reserved for the users and allocated periodically (with the repetition interval known as SPS Interval) till the end of the talk burst, with the same TFC as used for previous transmission being used. This is termed as Periodic Allocation. In case a packet requires less PRBs compared to the reserved ones, then the extra PRBs are released with the new allocation being recorded for future reservation. If the required PRBs are more, then extra PRBs are added to the previous reservation or a completely new set of PRBs are chosen depending on feasibility. Any change of reserved PRBs, TFC or transmission power requires new grant allocation through PDCCH. If no packets are received after the SPS Interval, then the PRBs are dynamically allocated to other users. In case no further packet is received even after the pre-determined active period of the user, the user is assumed to be dormant, after which the PRBs remain no longer reserved for the user. For further transmission, the user is treated as a ‘new’ user.
For either of the two transmission processes, the packets may require more PRBs than the available for the TTI. As real-time communication is delay sensitive, queuing of such packets until PRBs are available may reduce the QoS. Hence, for such large packets, the available PRBs are allocated for a part of the packet while the leftover is transmitted in the subsequent TTIs on a dynamic basis in a process known as the Leftover Allocation. The dynamic transmission of leftover requires availability of PDCCH for grant allocation.
For any of the allocation process, in case of packet loss, Retransmission of packets takes place in a dynamic fashion. Thus, the developed NS-SPS algorithm can be separated into four parts with the following order of priority of allocation: (a) Periodic Allocation, (b) Leftover Allocation, (c) Retransmission, and (d) Initial Allocation. Fig. 2. below shows the basic block diagram of the NS-SPS algorithm.
B. Segmentation based Semi-Persistent Scheduling
The basic process of Segmentation based Semi-Persistent Scheduling (S-SPS) algorithm is same as NS-SPS. The variation is only in term of allocation of PRBs to large packets. When packets are large and require more PRBs than the available or feasible, they are divided into multiple segments. The number of segments is calculated by dividing the packet size by the available/ feasible PRBs, such that the obtained number is a factor of the SPS Interval. For example, using 20 ms as SPS Interval, the possible numbers of segments are 1, 2, 4, 5, 10 and 20. The number of PRBs for these
segments equals the number of available/ feasible PRBs at the present TTI (that is, the TTI when the packet is segmented). These segments are transmitted periodically within the SPS Interval of 20 ms before the arrival of the next regular packet. Effectively, by segmentation, the SPS Interval is reduced, such that no new grants are required for the allocations of segments 2, 3… Thus, for each segment, reservation of PRBs is done for the duration of SPS Interval and these segments use the same TFC and transmission power for the SPS Interval. As this algorithm does not contain a separate Leftover Allocation step, the three allocation steps in order of priority are as follows: (a) Periodic Allocation, (b) Retransmission, and (c) Initial Allocation. Fig. 3. below shows the basic block diagram of the S-SPS algorithm while Fig. 4. shows the difference in allocation process of large packets in NS-SPS and S-SPS. �
C. Additional Points
Some of the additional points are elaborated below:
1) Non–Availability of PDCCH: In case of PDCCH non-availability for Periodic Allocation of resources, as allocation of new grants cannot be done for this TTI, the grant for the previous transmission is used – even if it is not optimal for the present situation. For other allocation processes, no further allocation is possible for the given TTI.
�
Fig.2. Block Diagram - Non-Segmentation based Semi-Persistent Scheduling
�
Fig.3. Block Diagram of the Segmentation based Semi-Persistent Scheduling
PERIODIC ALLOCATION
FOR NS - SPS USERS
LEFTOVER ALLOCATION
FOR NS - SPS USERS
RETRANSMISSION FOR
NS-SPS USERS
INITIAL ALLOCATION
FOR NEW NS-SPS
USERS
TFC SELECTION AND PRB
RESERVATION FROM
PREVIOUS PERIODIC
ALLOCATION
STORE TFC SELECTION
AND PRB RESERVATION
FROM INITIAL
ALLOCATION
DYNAMICALLY SELECT PRBs & TFC
SEMI-PERSISTENT
SCHEDULING
DYNAMICALLY SELECT PRBs & TFC
DYNAMICALLY SELECT PRBs & TFC
DYNAMIC
SCHEDULING
DYNAMIC
SCHEDULING
SEMI-PERSISTENT
SCHEDULING
PERIODIC ALLOCATION OF
SEGMENTS FOR S -SPS
USERS
PERIODIC ALLOCATION OF
PACKETS FOR S - SPS
USERS
RETRANSMISSION FOR
S-SPS USERS
INITIAL ALLOCATION FOR
NEW S-SPS USERS
TFC & PRBs FROM
PREVIOUS PERIODIC
PACKET ALLOCATION
STORE TFC SELECTION
AND PRB RESERVATION
FROM INITIAL PACKET /
SEGMENT ALLOCATION
SEMI-PERSISTENT
SCHEDULING
DYNAMICALLY SELECT PRBs & TFC
DYNAMICALLY SELECT PRBs & TFC
SEMI-PERSISTENT
SCHEDULING
DYNAMIC
SCHEDULING
SEMI-PERSISTENT
SCHEDULING
TFC & PRBs FROM
PREVIOUS PERIODIC
SEGMENT ALLOCATION
TTIs 01 02 03 04 05 06 07 08 09 … 16
PRBs
SPS Interval
(a) Large Packets requiring PRBs greater than available/feasible
TTIs 01 02 03 04 05 06 07 08 09 … 16
PRBs
(b) Non-Segmentation based allocation of Large Packets
TTIs 01 02 03 04 05 06 07 08 09 … 16
PRBs S1 S2 S3 S4
(c) Segmentation based allocation of Large Packets
Fig.4. Difference in allocation process of Large Packets in Non-Segmentation based and Segmentation based Semi-Persistent Scheduling Algorithm
2) Information Carry-forward: For both the algorithms, information such as the TFC used, reserved PRBs, last scheduled time, transmission power and number of segments (only in case of S-SPS) are recorded in a database. This database enables the successful semi-persistent allocation of resources for further TTIs.
IV. SIMULATION SCENARIO AND RESULTS
All the simulation cases were done in a three tier diamond-pattern macro scenario with 3-sector sites. The deployment of the users is on a random basis. The traffic model used for the VoIP capacity analysis is Adaptive Multi-Rate (AMR) audio codec with bit rate of 12.2 kbps. In this simulation, the active period of user is considered to be 20 ms. If no packet is received from user within 20 ms, the user is assumed to be dormant. The simulations were done for the downlink scenario in a single cell with randomly distributed users. For the simulation scenarios, the number of users varied from 90 to 200, with an increment of 10 users in each scenario. In each scenario, simulation was done for 15 x 10
5 TTIs or 1500
seconds, with the position of the users being randomly shuffled every 30 seconds. The measurements are obtained for delay budgets varying from 40 ms to 100 ms (with a step of 10 ms). The main simulation parameters are given in Table I whereas the parameters related to VoIP traffic are listed in Table II. The performance of the three algorithms for different scenarios is shown in Fig. 5. The comparison of performance of the three different algorithms for a delay budget of 70 ms and 90 ms are shown in Fig. 6 and Fig. 7 respectively.
It must be noted that while for DS, the TFC selection is based on the CQI reporting, in NS-SPS and S-SPS, TFCs depend on time-averaged wideband SINR measurements. As can be observed from Fig. 5, the percentage of satisfied users is directly proportional to the increase in delay budget and is inversely proportional to the decrease in the number of users in the cell. For DS in Fig. 5 (a), the rate of increase in user satisfaction percentage with respect to the increase of delay budget and decrease of user per cell are similar. For the delay
budget of 50 ms, 95% users are satisfied for 110 users per cell. With the increase of delay budget to 70 ms, the capacity increases to 125. In NS-SPS, as users are semi-persistently scheduled, there is no (or rather reduced) dynamic access and hence, PRBs are allocated periodically to all the users that can be accommodated. Fig. 5 (b) shows that for NS-SPS, percentage of satisfied users in downlink gradually decreases with the increase of the number of users in the cell till a threshold value after which the percentage of satisfied users falls drastically. If the number of users goes beyond the maximum user threshold, no more users can be accommodated due to unavailability of unreserved PRBs – even if the delay budget is increased. In the given results, the threshold of maximum users is approximately 175 users. The increase in the percentage of satisfied users with respect to increasing delay budget is gradual. Although at a lesser rate, it occurs as with more time, the chances of successful transmission of leftover packets and retransmissions increases. For the delay budget of 50 ms, the percentage of satisfied users is 95% for a cell capacity of 120. For an increased delay budget of 70 ms, 95% users are satisfied for a cell capacity of 160 users per cell. The downlink behavior of the S-SPS algorithm is illustrated in Fig. 5 (c). As can be observed from the figure, the variation of percentage of satisfied users for S-SPS with respect to increase of delay budget and with respect to decrease of user per cell is similar. From the results, it can be seen that for a delay budget of 50 ms, 95% users are satisfied for a cell capacity of 90 users whereas for a delay budget of 70 ms, the capacity increases to 110 such that 95% users are satisfied.
TABLE I. MAIN SIMULATION PARAMETERS
Parameter Value
Cellular Layout Hexagonal grid with 3-sector sites
Inter-site distance 500 m
Shadow Model 2D uncorrelated grid with bilinear
interpolation
Shadow Standard Deviation 8 dB Thermal Noise Density -174 dBm/Hz
Noise Figure at UE 9 dB
System Bandwidth 5 MHz
Carrier Frequency 2GHz
Sub-frame duration 1 ms
Duplexing FDD Carrier per PRB 12
Frequency Reuse 1
DRX Enabled
No. of PDCCH 4 for Uplink; 4 for Downlink
Bundling No
Link Adaptation Fast OLLA
Max. eNB Transmission Power 46 dBm
eNB Height 32 m
Max. UE Transmission Power 23 dBm UE Height/ Mobility 1.5 m/ 3 kmph
Downlink Antenna Configuration 2 x Transmitter; 2 x Receiver
Downlink Antenna Gain 14 dBi
HARQ Scheme Incremental Redundancy
No. of HARQ Processes 8
Max. HARQ Retransmission 4 CQI Delay (only for DS) 4 TTI
CQI Report Rate (only for DS) 20 ms
CQI Resolution (only for DS) 2 PRBs
CQI Report Rate (only for SPS) Beginning of talk-spurt
CQI Resolution (only for SPS) Wideband CQI
Simulation duration 1500 sec
Leftover Allocation
Segments
TABLE II. VOIP TRAFFIC PARAMETERS
Parameter Value
Call length 30 sec
Average talk-spurt duration/ Voice activity 3 sec/ 50% AMR Voice Codec Rate (burst rate) 12.2 kbps
SID Rate (during silence periods) 0.24375 kbps
Voice packet inter-arrival time (SPS Interval)/ size 20 ms/ 40 bytes
SID inter-arrival time/ size 160 ms/ 15 bytes
(a) Dynamic scheduling
(b) Non-segmentation based semi-persistent scheduling
(c) Segmentation based semi-persistent scheduling
Fig. 5. Capacity analysis w.r.t. different delay budgets and users per cell for different algorithms (Blue grid represents 95% users level)
As evident from the Fig. 6, the fitted curve for NS-SPS algorithm can support a capacity of nearly 160 users. Compared to NS-SPS algorithm, DS algorithm can support about 125 users and S-SPS algorithm can support only 110 users per cell. Thus, NS-SPS algorithm can sustain a capacity 22% and 31% more compared to DS and S-SPS algorithms respectively. The capacity gain for NS-SPS over DS occurs due to the semi-persistent scheduling of the users, as they do not have to compete for the available PRBs. This illustrates the benefit of reservation of packets with periodic arrival. However, this does not ensure that reservation of resources always performs better than dynamic allocation. Even though resources are reserved for S-SPS, the performance of S-SPS is poorer compared to NS-SPS and DS for delay budget of 70 ms. This occurs mainly as the segmentation of packets in S-SPS require more reservations within the SPS interval. Compared to NS-SPS, the frequent reservation to some users causes the number of allowable users for semi-persistent allocation to reduce. Also, due to the segmentation process, the loss of the segments can cause retransmission to start with a delay – in worst case equaling SPS Interval in case of loss of the last segment (since the last segment is transmitted after a period equaling SPS Interval after the transmission of the first segment). This wait period may cause late delivery of the packets beyond the acceptable delay. However, it must be noted that the rate of decrease in user satisfaction in S-SPS is very less compared to DS and NS-SPS. As can be seen in Fig. 7, with the increase in delay budget, S-SPS out performs both NS-SPS and DS and support a higher capacity for delay budget of 80 ms or higher. With a delay budget of 90 ms, a capacity of 200 can be reached using S-SPS whereas the capacity of DS and NS-SPS with 100 ms delay budget is about 155 (22% less) and 165 (17% less) respectively. Therefore, for downlink at low delay budget, it can be concluded that the performance of NS-SPS algorithm is better than DS and S-SPS algorithms. However, with better optimization, S-SPS can be a good candidate for downlink scheduling.
In terms of the efficiency of resource utilization, the results can be seen in Fig. 8. The three sets of bar graphs compare the percentage of resources used with each set showing the percentage of resources used when the cell contained 90, 140 and 190 users respectively. The resources used by NS-SPS and S-SPS are relatively similar and much better compared to DS. The low resource utilization in DS can be attributed to the lack of PDCCH available to allocate new resources to users in queue every TTI. Compared to DS, in NS-SPS and S-SPS, PDCCH is used only when certain changes are required and not used for every transmission by the active users in every TTI. Hence, this may reduce the capacity achievable using DS. Comparing results for the 140 user scenario, the resources used by NS-SPS is about 83% and for S-SPS, about 86% PRBs are allocated. Compared to these, DS is only able to allocate 35% of the resources available. For capacity of 90 and 190 users respectively, NS-SPS uses about 53% and 95%, S-SPS uses 52% and 94%, while DS users only 25% and 38% of the available PRBs. Also, it can be intuitively understood that with the increase of number of users per cell, the PRB requirement increases and this can be clearly seen in the variation shown in the graph. However, it must be noted that though NS-SPS and S-SPS definitely has advantage of consu-
100 120 140 160 180 200
60
70
80
90
100
Number of Users
Per
cen
tag
e of
Sat
isfi
ed U
sers
Percentage of Satisfied Users in Downlink
DS Curve-Fit
DS Markers
S-SPS Marker
S-SPS Curve-Fit
NS-SPS Marker
NS-SPS Curve-Fit
Target Level
Fig. 6. Comparative analysis of the performance of dynamic, segmentation based semi-persistent and non-segmentation based semi-persistent scheduling algorithms w.r.t. the percentage of satisfied users for delay budget of 70 ms
100 120 140 160 180 20075
80
85
90
95
100
Number of Users
Per
centa
ge
of
Sat
isfi
ed U
sers
Percentage of Satisfied Users in Downlink
DS Markers
DS Curve-Fit
NS-SPS Markers
NS-SPS Curve-Fit
S-SPS Markers
S-SPS Curve-Fit
Target Level
Fig. 7. Comparative analysis of the performance of dynamic, segmentation based semi-persistent and non-segmentation based semi-persistent scheduling algorithms w.r.t. the percentage of satisfied users for delay budget of 90 ms
1 2 30
20
40
60
80
100
Per
cen
tag
e of
PR
Bs
use
d
Comparision of Resource Utilization
90 Users
140 Users
190 Users
NS-SPS S-SPS DS
Fig. 8. Percentage of resource allocated using different scheduling algorithm
for 90, 140 and 190 users per cell
ming lesser PDCCH, in terms of PDSCH and amount of resource usage, DS performs better than NS-SPS and S-SPS.
V. CONCLUSION
As LTE (-A) is based on packet switching real-time services like voice communication require VoIP based solution. Therefore, scheduling assumes importance as, unlike data communication over a packet switched network, real-time
communication packets arrive more frequently. Therefore it requires efficient scheduling such that the capacity of the system is not limited by the available control channel. Towards this issue, different scheduling mechanisms have been proposed in literature. However, many of the proposals fail to take into account the possibility of large packet sizes for voice transmission as well as other real-time applications. This work deals with the development of semi-persistent scheduling algorithms and comparing the capacity results with the dynamic scheduling mechanism. Two variations of semi-persistent algorithms developed are based on the segmentation and non-segmentation of the large packets. While for segmentation, the leftovers are semi-persistently scheduled, non-segmented leftover packets are dynamically transmitted. Also, unlike previous works, the SPS algorithms are based on time-averaged wideband SINR information and not instantaneous CQI information. Simulation of these algorithms shows that the non-segmentation based semi-persistent algorithm works best amongst the three in the downlink for lower delay budgets. However, with the increase in delay budget, segmentation algorithm works better than both dynamic and non-segmentation based algorithms. Also, the low resource utilization in dynamic scheduling shows that the capacity is restricted due to unavailability of PDCCH to grant new resources. On the other hand, VoIP capacity in SPS is constrained by the PDSCH availability.
This work has focussed only on voice communication in downlink using AMR 12.2 voice codec. As a future scope for this work, simulations can be carried out for uplink scenarios as well and also using video transmission packet using the developed scheduling algorithms to determine the performance for the algorithms with respect to video transmission.
REFERENCES
[1] "Cisco visual networking index: Global mobile data traffic forecast update, 2010-2015," white paper, Cisco Systems Inc., Feb. 2011.
[2] M. Anehill; M. Larsson; G. Strömberg; E. Parsons, "Validating voice over LTE end-to-end," white paper, Ericsson Review, Jan. 2012.
[3] D. Jiang; H. Wang; E. Malkamaki; E. Tuomaala, "Principle and Performance of Semi-Persistent Scheduling for VoIP in LTE System," International Conference on Wireless Communications, Networking and Mobile Computing (WiCom, ‘07), pp.286-2864, 21-25 Sep. 2007.
[4] J. Puttonen; N. Kolehmainen; T. Henttonen; M. Moisio, "Persistent packet scheduling performance for Voice-over-IP in evolved UTRAN downlink," IEEE 19th International Symposium on Personal, Indoor
and Mobile Radio Communications, (PIMRC’08), pp.1-6, 15-18 Sep. 2008.
[5] M. Muhleisen; B. Walke, "Evaluation and improvement of VoIP capacity for LTE," 18th European Wireless Conference (EW‘12), pp.1-7, 18-20 Apr. 2012.
[6] J. Puttonen; T. Henttonen; N. Kolehmainen; K. Aschan; M. Moisio; P. Kela, "Voice-Over-IP Performance in UTRA Long Term Evolution Downlink," IEEE 67th Vehicular Technology Conference (VTC’S08), pp.2502-2506, 11-14 May 2008.
[7] Y. Fan; P. Lunden; M. Kuusela; M. Valkama, "Efficient Semi-Persistent
Scheduling for VoIP on EUTRA Downlink," IEEE 68th Vehicular
Technology Conference (VTC’F08), pp.1-5, 21-24 Sep. 2008. [8] "LTE physical layer framework for performance verification," 3GPP
R1-070674, St Louis, USA, 12-16 Feb. 2007.