ECE (Electrical, Communications, and Computer Engineering)  

Symbol Timing Tracking Algorithm for WAVE Systems

Presenter: Dae-Ki Hong, PhD, Professor Sangmyung University, Cheonan City, Chungnam, Korea, hongdk@smu.ac.kr

Authors: Dae-Ki Hong, PhD, Professor, Sangmyung University

SUMMARY

In this paper, tracking algorithms that synchronize

the symbol timing is proposed to continually

compensate additional timing delay. To track the

additional delay, we modify a conventional WAVE

frame with known symbol sequences.

I. PROPOSED TRACKING ALGORITHM

We assume that the initial timing synchronization is perfect. Firstly, 4-times oversampled mid-ambles are received and stored in shift register with additional flip-flop elements. The additional elements are required for the windowing push/pull operation. The received samples are correlated with the known early/late mid-amble samples. If the initial time synchronization is perfect, the correlation peak is observed at on-time correlator output However, if Ts time delay occurs in received signal, the correlation peak is observed at 1-sample late sample correlator. The maximum selector selects the index of the maximum correlation value. The index at maximum correlation value is accumulated at the maximum index accumulator. The accumulated value controls (pushes/pulls) the FFT and correlation windows. In the symbol timing estimation, the estimation performance can be

Fig. 1 Proposed MC Tracking Algorithm

more and more improved as the gap between the correlation value at on-time correlator and the correlation value at adjacent correlator increases. Therefore we propose the multiple correlation (MC) tracking algorithm which uses sum of the consecutive 3 correlation values. (see Fig. 1)

II. RESULTS

Fig. 2 Tracking Performance Comparisons

Fig. 2 depicts BER performance of the tracking algorithms under AWGN channels. In the figure, with time delay and no tracking case represents the worst performance. The three curves (with time delay and no tracking, with time delay and ideal tracking, with time delay and multiple correlation tracking) is overlapped. Therefore, the proposed MC tracking algorithm approaches the ideal performance. The general tracking algorithm does not approach the ideal performance.

REFERENCES

1. IEEE P802.11p/D1.0, “Wireless LAN Medium

Access control (MAC) and Physical PLayer(PHY)

Specification - Amendment 3: Wireless Access in

Vehiclular Environments (WAVE).”

mailto:hongdk@smu.ac.kr

Recovery of a 3D Signal using a Structured Light System

Deokwoo Lee North Carolina State University

890 Oval Drive, Raleigh, NC27606 dlee4@ncsu.edu

SUMMARY

Reconstructing 3D geometric information (e.g.

coordinates) from a 2D image has been a widely

studied issue as well as one of the most important

problems in image processing. In this paper, we

propose an approach to reconstructing 3D

coordinates of a surface from a 2D image taken by

a camera using structured light systems. Light

source or projectors generate multiple co-centric

circular patterns. The relationships between the

known information (i.e. intrinsic and extrinsic

parameters of the camera, the structure of the

circular patterns, a fixed optical center of the

camera and the location of the reference plane of

the surface) and the deformed patterns provides a

sufficient information to establish a mathematical

model for 3D coordinates extraction. The

reconstruction is based on a geometrical

relationship between a given pattern projected onto

a 3D surface and a pattern captured in a 2D image

plane from a viewpoint. This paper chiefly deals

with a mathematical proof of concept for the

reconstruction problem.

I. INTRODUCTION

3D image reconstruction from a 2D projected image

on the basis of structured light patterns has been of

much interest and an important topic in image

processing. Applications are in the areas of object

recognition, medical technologies, robotic visions

and inspections of properties of target images, etc

[1]. There are two main approaches to the

reconstruction problem : active and passive. The

general principle involved in the passive method is

triangulation using 2 or more cameras(usually 2

cameras) [2]. The relative positions of two cameras

and an object in each image plane provides the

necessary information to reconstruct 3D coordinate

information [2,3]. The alternative approach to the

reconstruction problem is an active stereo vision, or

an active method, using a structured light system

which is used widely [4]. Generally one camera is

replaced by a light source such as an LED or a

laser beam that projects a known pattern.

II. MOTIVATION / IDEA

Circular patterns have very good characteristics such as a closed and continuous form, and a symmetric shape leading to a lower computational complexity. Moreover, only a single point(i.e. the center of a circle) and the radii of the circles are needed. This advantage can improve the algorithm efficiency. Our approach is based on analyzing a deformed circular pattern acquired by the camera providing the 3D coordinate information of an object. (Fig. 1). This paper gives an account of proof of concept for a reconstruction and some simple simulated results.

Figure 1. Ideal and deformed circles.

REFERENCES

1. Zhengzhong Wei, Fuqiang Zhou and

Guangjun Zhang, 3D coordinates

measurement based on structured light

sensor, Sensors and Actuators A : Physical,

Volume 120, Issue 2, 17 May 2005, Page

527-535.

2. Umesh R. Dhond and J.K. Aggarwal,

Structure from Stereo - A Review, IEEE

Transaction on systems, man, and

cybernetics, Vol. 19, No. 6, November /

December 1989.

3. J. Batlle, E. Mouaddib and J. Salvi., Recent

Progress in Coded Structured Light as a

Technique to solve the Correspondence

Problem : A Survey, Pattern Recognition,

Vol. 31, No. 7, pp963 - 982, 1998.

4. P.M.Will and K.S.Pennington, Grid coding :

a preprocessing technique for robot and

machine vision, Proc. Int. Joint Conf. on

Artificial Intelligence, pp. 66-70(1971).

Biometrics with Hardware Security Module for User Authentication

Myung Geun Chun Chungbuk National University

CheongJu, Chungbuk Republic of Korea

mgchun@chungbuk.ac.kr

Yong Nyuo Shin Hanyang Cyber University

Seoul Republic of Korea ynshin@hycu.ac.kr

Seong Gon Kong Temple University

1700 North Broad Street Philadelphia, PA, USA

e-mail address

SUMMARY

This paper describes an user authentication

scheme using biometric hardware security module

(BHSM) for proving owner of X.509 certificate

registered individual at RA (Register Authority).

I. INTRODUCTION

Biometric technology can prevent various illegal

acts related to identification through its positive

function of authenticating individuals. However,

biometric technology can also be used to trace

personal profiles, related transactions, or changes

in the database without the consent of the person in

question. There have been studies to overcome

these threats[1]. Here, we introduce a privacy

sympathetic user authentication method based on

the biometrics with the HSM.

II. USER AUTHENTICATION WITH BHSM

A. Enrollment Procedures

(1) Alice(an user) presents his biometric

characteristics under control of the RA. And

the extracted biometric feature is stored at

Alice’s BHSM.

(2) RA generates a random number R to make

a PBID(= h(h(BT, R)) with Alice’s biometric

template. Here, BT is the biometric

template and h() is a hashing function.

(3) RA stores the random number R in the

Alice’s BHSM with a predefined format.

(4) RA sends the generated PBID to the CA.

Alice generated his private and public keys

with the BHSM.

(5) Alice sends a requesting message for

X.509 certificate to CA.

(6) CA generates Alice’s X.509 certificate

involving the PBID in the extension field.

(7) CA sends Alice’s X.509 certificate and it

will be securely stored at Alice’s BHSM.

B. User Authentication

Figure 1: User Authentication Process

(1) Alice performs biometric authentication

using the biometric module in the BHSM.

(2) After successful authentication of Alice

herself, she sends the encrypted hashing

of the biometric template and random

number R stored in the BHSM to the Bob

with Bob’s public key. Here, the X.509

certificate is also transmitted.

(3) Bob authenticates the ownership of the

X.509 certificate by comparing the PBID

in the X.509 certificate with the decrypted

hashing value of the received value.

Acknowledgement

This research was supported by Basic Science Research

Program through the National Research Foundation of

Korea(KRF) funded by the Ministry of Education, Science

and Technology (2011-0026223)

REFERENCES

1. Dae-Jong Lee, Yong-Nyuo Shin, Seon-Hong

Park and Myung Geun Chun, “RN-ECC Based

Fuzzy Vault for Protecting Fingerprint

Templates,” International Journal of Fuzzy Logic

and Intelligent Systems, Vol 11, No. 4, pp. 286-

292, December, 2011.

Design of a Planar UWB Antenna

with Notched WLAN Band by Using Slot and Slit

Chang Joo Lee

Engineering research institute, Gyeong Sang National

University

Jinju, South Korea

e-mail: mynameiscj@gnu.ac.kr

Mun Soo Lee

Engineering research institute, Gyeong Sang National

University

Jinju, South Korea

e-mail: munslee@gnu.ac.kr

SUMMARY

In this paper, a planar UWB antenna with notched

WLAN band by using slot and slit is proposed. The

rectangular slits at the corner of the patch and ground plane

are used to achieve wideband characteristic. By embedding U

type slot on the patch and rectangular slit on the ground plane,

a notched band can be realized. By adjusting the width and

length of U type slot and rectangular slits, the return loss

bandwidth and notched bandwidth of the proposed antenna

can be optimized. The optimized antenna has the return loss

bandwidth of about 8 GHz (3 GHz ∼ 11 GHz), its radiation pattern in H-plane is omni directional radiation pattern which

is satisfied with required condition.

I. INTRODUCTION

It is a well-known fact that planar monopole antennas present really appealing physical features, such as simple structure, small size, lightweight and low cost. Additionally, planar monopoles are compact broadband omni directional antennas [1], and are also non-dispersive [2]. Due to all these interesting characteristics, planar monopoles are extremely attractive to be used in emerging UWB applications, and growing research activity is being focused on them. However, the wideband 4.8

∼ 7.1 GHz, should be notched to avoid the interferences with WLAN and GPS and so on.

In this paper, a planar UWB antenna with notched WLAN band by using slot and slit is designed. Single-feeding by a coaxial probe is used for the simple structure. In order to obtain a wideband characteristic, the slits are inserted at corner of the patch and the ground plane. The U type slot and rectangular slit are used to notch WLAN band. The width, length and distance of slot and slits are optimized by using CST MWS(CST ver. 2006). Through a number of iterative simulations, the optimum width and the length of the slits and U type slot are obtained.

II. ANTENNA DESIGN

Fig. 1 shows the geometry of the proposed planar UWB antenna with notched WLAN band, which consists of a rectangular patch with inserted U type slot, a partially modified ground plane with rectangular slit. Each of antenna parameters: W1 and L1 are width and length of U type slot, distance D is between U type slot and low edge of the patch, W2 and L2 are width and length of rectangular slit on the ground plane. Fig. 2 shows the return loss of the antenna.

D

W1

L1

W2

L2

(a) (b)

Figure 1. Geometry of antenna of the planar UWB antenna with notched WLAN band (a) Front view (b) Back view

Figure 2. Return loss of the antenna (a) Simulation (b) measurement

III. CONCLUSION

The measured 10dB return loss bandwidth of the proposed

antenna has 7.98GHz (3.02 GHz ∼ 10.6 GHz) with notched band (5.26 GHz ∼ 6.08 GHz). The notched center frequency is slightly shifted against simulated results. The notched characteristic is observed in the return loss and gain.

The proposed antenna not only has very simple structure but also is very easily constructed with a very low cost. It is suitable for avoiding interference with WLAN communication.

REFERENCES

[1] Constantine A. Balanis. Antenna Theory. John Wiley & Sons, INC. 1997. Chapter 2,14.

[2] J.R. James & P.S. Hall. Handbook of Microstrip Antennas. Short Run Press LTD. 1989. pp. 219-232.

[3] Hisao Iwasaki. A Circularly Polarized Small-Size Microstrip antenna with a Cross Slot. IEEE Transactions on Antennas and Propagation, Vol. 44, NO. 10. October 1996. pp. 1399-1401.

Control of Power Converters for Electric Vehicle applications

Bradley University 1501 W.Bradley ave

Peoria,IL, USA e-mail : wkna@bradley.edu

I. ABSTRACT

Lately, Fuel Cell Vehicles(FCV), Hybrid Electric

Vehicles(HEV), and Plug-in-HEVs(PHEV) make

significant progress in research and even

commercialization by improving of their fuel

economy and producing of less pollutants

compared to conventional vehicles[1]. However,

since these Electric Vehicle (FCV,HEV and PHEV))

systems have inherent control issues, e.g., control

of the power converters has to be considered for

releasing and saving energy from the energy

storages during transients[2]. Thus, in this paper,

control methods of power converters for Electric

Vehicle applications are presented. The simulation

results based on PSIM, and experimental data for

these power converters are shown in the paper.

II. ILLUSTRATIONS

Fig.1 shows the overall power circuit of the

system which consists of a Power Factor

Corrected(PFC) boost converter and bidirectional

converter.

Fig. 1: Power Circuit of the system

Fig. 2 shows the experimental setup for the boost

and bidirectional converters. The most of the control

algorithm for the power converters are implemented

by a Texas Instrument 32 bit DSP board, TMS

320F2812 shown in Fig. 3.

Fig. 2: Experiment Setup

Fig. 3: TMS 320F2812 DSP board

REFERENCES

1. W.D. Jones, “ Take this car and plug it,” IEEE Spectr. Vol. 42, no.7, pp10-13,Jul. 2005

2. A.F. Burke, “ Batteries and Ultracapacitors for Electric, Hybrid, and Fuel Cell Vehicles,” Proceedings of the IEEE, Vol. 95, no 4, pp 806-820, 2007

Woonki Na, Renee Kohl, Matthew Daly and Peter Burrmann

Balancing and Driving Control of a Bicycle Robot

Sukin Lee and Jangmyung Lee*

Pusan National University

Abstract: This paper proposes a balancing and driving control system for a bicycle robot. A reaction wheel pendulum control method

is adopted to maintain the balance while the bicycle robot is driving. For the speed control of the bicycle robot, PD control algorithm

with a variable gain adjustment has been developed in this paper, where the gains are heuristically adjusted during the experiments.

To measure the angles of the wheels the encoders are used. For the balancing control, a roll controller is designed with a non-model

based algorithm to make the shortest cycle. The tilt angle is measured by the fusion of the acceleration and gyroscope sensors, which

is used to generate the control input of the roll controller to make the tilt angle zero. The performance of the designed control system

has been verified through the real experiments with the developed bicycle robot.

Keywords: Bicycle Robot, Balancing control, Reaction Wheel Pendulum, PID control

I. Introduction Currently mobile robots are well applied for the industrial

applications with the new development of sensors and algorithms.

Unmanned navigation systems are very useful for the special

environment such as nuclear plant or mine areas to carry our

surveillance and monitoring tasks. Mostly four wheeled or six

wheeled systems are developed and applied in real applications.

However these are not effective in the view of energy consumption

and require a large working space because of the large rotation radius.

Especially when the working space is narrow and long, the four

wheeled or six wheeled vehicles are not suitable for prompt and

flexible motions. To overcome these difficulties, two wheeled

autonomous systems are getting interest. The two-wheeled system is

well-suited in a narrow space than the four wheeled system with a

critical shortcoming. That is, the system is very unstable and

susceptible to external disturbances by its structure. Especially for the

bicycle type two-wheeled robot, the roll control is necessary to keep

the stable posture since the robot body and two wheels are aligned in a

line. There are several research results on this stable standing and

driving[1]. Based on the dynamic model of the bicycle robot, the front

wheel angle is controlled[2,3], the center of the mass is changed to

keep the balance, and the Gyroscope and reaction wheel pendulum

have been used to maintain the balance.

In this research, the complex dynamics are ignored to shorten the

control cycle as short as possible with a cheap microprocessor. For the

roll control, the reaction wheel pendulum is used. The reaction wheel

pendulum rotates the disk to the tilting direction of the bicycle robot

such that the reaction force to the inertia of the disk pushes the robot

body to the opposite direction. For the purpose, the angle of the robot

body is measured by the gyroscope and acceleration sensors whose

data are fused properly to get the accurate angle. There are two DC

motors for driving and disk rotation and a servo motor for direction

control.

This paper consists of five sections including this introduction. The

system composition of the bicycle robot is described in section 2.

The roll controller and its operations are presented in section 3. The

designed bicycle robot is checked for the control performance of roll

and speed control accuracy in section 4. Conclusions of this research

are added in section 5.

II. Conclusions The balance and driving control of a bicycle robot has been

performed in this research. To shorten the balance control cycle, a

simple controlled is designed without deriving dynamics of the

bicycle robot using the body angle as a control input to the controller

and the velocity data of the disk. The control performance is verified

by the real bicycle robot developed for this research. For the roll

control, the center of mass location data are utilized, which should be

kept very accurate not to cause a very high speed rotation of the disk

or heavy vibrations with poor data. For the speed control, PI controller

is designed with variable gains during the experiments. To check the

velocity control performance, a unit step and trapezoidal velocity

profiles are applied for the desired velocities. Through the driving test,

it is verified that the balance control and velocity control loops can be

controlled independently for the straight line motions. For the S-

curved path driving, the driving errors become big, which is

compensated by commends generated external C# program and sent

through the blue-tooth. As a future research, obstacle avoidance

problems need to be handled with the setting of the center of mass.

References

[1] Y. Tanaka, T. Murakami, ”Self sustaining bicycle robot with

steering controller,” The 8th IEEE International Workshop on

Advanced Motion Control, AMC’04, March 2004, pp. 193-197.

[2] T. Takei, M. Tsuruta, J. Okumura, T. Nakabayashi, T. Tsu

bouchi, “Stabilized Motion of a Small Sized Bike Robot o

nly by Steering Control,” The 11th IEEE International Work

shop on Advanced Motion Control, March 2010, 466-471.

[3] Yonghua Huang, Qizheng Liao, Shimin Wei, Lei Guo, “D

ynamic Modeling of a Bicycle Robot with Front-Wheel Dr

ive Based on Kane's Method,” Proceedings of the 2010 IE

EE International Conference on Information and Automatio

n, June 2010, 758-764.

* Corresponding Author: jmlee@pusan.ac.kr

※ This research was supported by the MKE(The Ministry of

Knowledge Economy), Korea, under the Human resource

development program for robotics support program supervised by

the NIPA(National IT Industry Promotion Agency). (NIPA-2010-

(C7000-1001-0009))

An Efficient DVFS Energy Reduction Algorithm based on Frame Decoding Complexity Estimation for Battery-powered Video Decoders

Heejune Ahn

Univ. Virginia, SeoulTech Rice Hall, 85 Engineer’s Way

Charlottesville, VA 22904, USA heejune@{virginia.edu,seoultech.ac.kr}

Seungho Jeung SeoulTech

Mirae Hall, Gongreng-dong, Nowon-gu, Seoul, South Korea

snghojeong@gmail.com

SUMMARY

Our previous work [1] has proposed a static

optimal DVFS (Dynamic Voltage Frequency Scaling)

algorithm for energy reduction in video decoders.

Based a frame complexity estimation, this paper

presents a dynamic extension of the static optimal

algorithm.

I. INTRODUCTION

Power consumption is crucial in battery-

powered embedded systems. Because the dynamic

power consumption of a CMOS device is

proportional to its operating clock frequency and

voltage, i.e., ,2 clkeffdyn fVCP Dynamic Voltage

and Frequency Scaling (DVFS) technique has

become a key system-level energy reduction

technique ([1] and the references therein). Our

previous work [1] proposed a ‘static’ optimal DVFS

scheduling algorithm with the assumption of known

frame decoding complexity in advance. In this

paper, we propose a ‘dynamic’ extension algorithm

that combines the static optimal scheduling with an

accurate frame complexity estimation method.

II. METHOLOGY

A. Video Decoder System Model

In Figure 1, DVFS controller determines the

operating frequency and voltage in every frame,

based on the current decoded frame buffer

occupancy and estimated complexity so that the

decoder buffer does not reach buffer underflow (no

frame to display) or buffer overflow (loss of results).

B. Frame Complexity Estimation

We observed that the linear model of its frame

size can estimate the frame decoding complexity

accurately, especially with classified picture types,

less than 5% in general.

TTT iSiC )()(

. (T= I, P, B) (1)

C. Dynamic DVFS Algorithm

The proposed dynamic DVFS algorithm

consists of two parts; the optimal decoding

frequency decision part (it is same as the static

algorithm in [1]) and the complexity estimation part.

The outline of algorithm is as follows: STEP 0. Set the default estimation parameters with system parameters and variables STEP 1. Get the next frame size, STEP 2. Estimate the complexity of the next frame to decode. (using eqn. (1)) STEP 3. Calculate optimal freq. and the corresponding voltage. STEP 4. Decode one frame: measure the decoding clock count and decoding time that frame, and the result buffer occupancy. STEP 5. Update the complexity estimation parameters.

III. EXPERINMENT RESULTS

Table 1 show the energy reduction performance

of dynamic algorithm is similar to static version and

5% to 10% higher than the previously proposed

algorithms.

Figure 1: The system model for DVFS Vid. Decoder

REFERENCES

1. S. Jeong, and H. Ahn, “Optimal Power Reduction

based on DVFS Algorithm for Video Decoders,”

in Proc. ACM RACS, USA ,November 2-5 2012.

Table 1: Energy Reduction Perf. Comparison

(power consumption relative with optimal)

traces

hori. resol.

static

optimal

DZ

Alg. panic

Alg.

dynamic

(prop.)

SF-1080 1.00 1.03 1.03 1.02

SF-720 1.00 1.04 1.04 1.02

DR-720 1.00 1.04 1.05 1.02

DR-480 1.00 1.05 1.06 1.03

Refining Preoperative Surgical Models for Individual Hospitals through Transfer Learning

Gyemin Lee, PhD

Department of Electrical Engineering and Computer Science University of Michigan, Ann Arbor, MI

gyemin@umich.edu

SUMMARY

Clinical models for surgical outcomes are

typically developed on multi-hospital registry data.

This approach fails to reflect the variations in

patients and clinical services across hospitals. Our

study explores the feasibility of predicting surgical

mortality by adaptively transferring a multi-hospital

model trained on a large national registry to an

individual hospital. Our results show that

transferring models trained on multi-hospital data to

an individual hospital significantly improves

discrimination for surgical mortality at the individual

provider level.

I. INTRODUCTION

Preoperative models to assess surgical

mortality are important clinical tools in determining

optimal patient care. The traditional approach to

develop these models has been primarily

centralized, i.e., it uses surgical case records

aggregated across multiple hospitals. While this

approach of pooling greatly increases the data size,

the resulting models fail to reflect individual

variations across hospitals in terms of patients and

the delivery of care. In this work, we improve this

process by adapting the multi-hospital data model

to an individual hospital. This approach simul-

taneously leverages the large multi-hospital data

and the patient-and-case mix at individual hospitals.

We explore transfer learning to refine surgical

models for individual hospitals in the framework of

support vector machine by using data from both the

National Surgical Quality Improvement Program

(NSQIP) and a single hospital.

II. METHODS

The following models were developed and

compared for predicting mortality within 30 days of

surgery using 48 preoperative variables:

(i) a single-hospital model trained on data from

5,000 patients at a participating hospital within

NSQIP (# mortality=65);

(ii) a multi-hospital model trained on 161,869

patient records (# mortality=1,860) from the

NSQIP 2008;

(iii) a merged model trained on a combination of

the data for both the single- and multi-hospital

models; and

(iv) a transferred model that adapted the multi-

hospital model in (ii) using the training data in (i)

through transfer learning.

All models were evaluated on data from the

participating hospital for another 2,053 patients

using the area under the ROC curve (AUROC).

III. RESULTS

The AUROCs for all models are presented in

Table 1. The transfer learning approach achieved

an improvement in discrimination relative to all the

other approaches.

IV. CONCLUSIONS

Surgical models can be individualized for

single-hospital use through transfer learning. This

approach improves discrimination relative to models

developed on large amounts of multi-hospital data,

as well as models developed on small amounts of

historical data drawn from the single hospital where

use is intended.

Model Single Multi Merged Transfer

AUROC 0.881 (+/- 0.002) 0.902 (+/- 0.002) 0.903 (+/- 0.002) 0.915 (+/- 0.002)

Table1: Performance Comparison of the Preoparative Models.

Biologically-Inspired Deceptive Behavior for a Robot

Jaeeun Shim Georgia Institute of Technology

Electrical and Computer Engineering Atlanta GA 30308, USA

jaeeun.shim@gatech.edu

SUMMARY

A common behavior in animals or human

beings is deception. We focus on deceptive

behavior in robotics because the appropriate use of

deception is beneficial in several domains ranging

from the military to a more everyday context.

I. INTRODUCTION

A common and essential behavior for survival in

a variety of intelligent systems ranging from insects

to human beings is deception. Animals act

deceptively to gain benefits from others. It appears

also in higher-level primates to involve the theory of

mind mechanism. We argue that robots can also

potentially gain advantage over adversaries by

possessing deceptive behaviors. We further posit

that to achieve more socially intelligent robots

operating in the presence of humans, we must

develop robots that interpret, generate, and

respond to deceptive behavior. In this paper, we

present a novel approach for deceptive behavior by

a robot, inspired by observations of squirrels in

cache protection strategies [1].

II. Algorithms and Implementation

A. Squirrel’s Deceptive Behaviors

We approach this problem inspired by animal

behavior. One interesting deception behavior

occurs in the food protecting strategies of squirrels

[1]. After hoarding food items, squirrels begin to

protect the resources from pilfering by patrolling the

caching areas. First of all, squirrels move around

the caching areas. However, if the potential

competitors are present nearby, they visit several

empty cache locations (false signals) to confuse

competitors about the food's location. Therefore,

social context (i.e., presence or absence of

competitors) appears to be pivotal to the expression

of cache protection behaviors.

B. Computational Model and implementation

A bio-inspired behavior based model of squirrel

caching and protecting behaviors for application to

robotic systems is developed using MissionLab [2].

Fig. 1 shows the model of deceptive behaviors of a

robot using a finite state acceptor (FSA).

Figure 1: (a) FSA: caching behaviors, (b) sub-FSA : food

hoarding (c) sub-FSA : food patrolling

III. Simulation Results

A simple scenario of the squirrel-like deceptive

behavior was simulated in MissionLab [2]. To

evaluate the approach, we measured the time

duration until the competitor robot detects the exact

caching places and begins pilfering. The same

scenarios without deceptive behaviors formed the

baseline. In two, the average time to successful

pilferage when the squirrel robot includes deceptive

behavior is 10.4 minutes (std: 3.04), compared to

the average time duration without deception is 7.69

minutes (std: 2.91). As a result, it can be concluded

that with the squirrel robot with deceptive behaviors

protects resources longer and performs significantly

better than the one without deceptive behaviors.

IV. Conclusions and Future Work

In this paper, a novel approach was presented

for deception in robots, focusing on how to preserve

resource gains inspired from biological findings. We

are currently applying our simulations to real robot

experiments. Even though this research focuses on

deceptive behaviors of robots in the military domain,

we will potentially extend our research more

towards human-friendly environments.

REFERENCES

1. M. a. Steele, et al., Cache protection strategies of

a scatter hoarding rodent: do tree squirrels

engage in behavioural deception? Animal

Behaviour (2008)

2. D. MacKenzie, R. Arkin, and J. Cameron.

Multiagent Mission Specification and Execution.

Autonomous Robotics (1997)

Wireless Microfluidic Pressure Meter for Monitoring of Cerebrospinal Fluid Shunt

Suk-Heung Song* and Marcel Utz

School of Applied Science and Engineering, University of Virginia, Charlottesville, VA 22904

*no27indy@gmail.com

SUMMARY

A microfluidic pressure sensor with inductively coupled, wireless readout capability has been developed for integration into cerebrospinal fluid shunt valve implants. The sensor consists of a deformable PDMS film that is bonded over a microfluidic reservoir, forming a fluidic capacitor. Deflection of the capacitor membrane is detected remotely through a shift in the resonance frequency

of a micro-fabricated LC circuit. The sensor transduces pressure into a change in resonant

frequency with sensitivity > 3.4ppm/Pa and responsivity 4.6 kHz/Pa, over a dynamic range

of 0 ~ 3 kPa.

I. INTRODUCTION

Our elastomer pressure sensors are fabricated by

combined MEMS and direct pattern transfer in a

simple and cost-effective manner. The performance

of the sensor membranes is documented for the

clinically relevant pressure range from 0.3 to 2.8

kPa, and flow metering in the of range

0~2400μl/min is demonstrated in an on-bench test.

Finally, preliminary data is presented that

demonstrates wireless pressure sensing which will

allow for non-invasive measurements of pressure

and flow in a next generation prototype device.

II. CONCLUSION

In this paper, we have been developed a

microfluidic flow meter consists of telemetry

pressure sensors that provided the dynamic

responses by fluidic capacitors using the pressure

induced deformation of PDMS film. The fluidic

capacitor has a top and bottom half of the capacitor

with embedded inductor. The sensors are

fabricated by combined conventional MEMS

technologies and rapid soft lithography. Especially,

the bottom half of the capacitor is micro-fabricated

using the gold coated PDMS film by direct pattern

transfer technique. The developed sensors can be

measured the pressure by the wireless sensing,

responsivity 4.6 kHz/Pa, sensitivity > 3400 ppm/kPa

over the pressure range of 0~ 3 kPa. They achieve

a good pressure sensing by highly compliant

deformable PDMS film capacitor. By given the

demonstrated engineering results, these devices

will applied to substantial improvement in CSF

shunts. Besides developing the devices, there are

some important issues for the next round are

remained. In-vitro test will be aimed to perform to

quantify the effect of real CSF including proteins

that are accumulated on shunt catheters as a form

of thin film, and realized long-time monitoring of the

functions of the sensor under varying physiological

conditions. Future work will be focused on

improving the sensing for practical applications.

III. ILLUSTRATIONS

The micro-fabricated gold coated on PDMS film-

based pressure sensor for CSF.

Figure 1: Images of the micro-fabricated devices

REFERENCES

1. Akar, O., Akin, T., and Najafi, K, Sens. Actu. A,

95, 29–38, (2001).

2.Terman, F, Radio engineers’ handbook. McGraw-

Hill New York, (1943).

3. Kim, J., Takama, N., Kim, B., and Fujita, H, J.

Micromech. Microeng, 19, 055017, (2009).

Handling Randomness of Multi-Class Random Access Loads in LTE-Advanced Network

Ki-Dong Lee

LG Electronics U.S.A. 920 Sylvan Ave.

Englewood Cliffs, NJ 07632 (kidong.lee@lge.com)

Ki-Dong Lee, Sang Kim LG Electronics U.S.A.

920 Sylvan Ave. Englewood Cliffs, NJ 07632

(iLoveLTE@gmail.com)

3rd Author Name Affiliation

Address line1 Address line2 e-mail address

SUMMARY

Although the Access Barring in LTE-Advanced

provides means to reduce Random Access (RA)

load, the reduction is merely in average sense due

to the randomness nature of the RA mechanism.

We identify this issue from probabilistic behavior

perspectives and propose a new means to improve

load control performance.

I. SUMMARY OF THIS WORK

Long Term Evolution (LTE) is commercially

available in many countries, such as South Korea

and the United States, as of March 2012. Now the

telecommunication technology is being evolved

towards LTE-Advanced, which is expected to allow

people to use more diversified application services.

One of the most attractive services is machine-to-

machine (M2M), which will further provide means to

ubiquitous healthcare, fleet management, and so on.

Since a large number of user devices are

expected in LTE-Advanced services, it is imperative

to study efficient methods to accommodate the load

from those devices. The type of load is categorized

as: (1) RA load (i.e., load required to establish data

connection), (2) user data (i.e., data traffic from/to

user devices). Due to the unique characteristics of

M2M and related emerging applications services,

RA load is expected to be more challenging as user

devices will generate a huge number of small data

packets per unit time, which cause RA attempts to

happen so frequently.

In order for LTE-Advanced system to

efficiently accommodate such a heavy RA load, so-

called several “Access Barring” methods have been

proposed, such as Extended Access Barring.

However, the most typical effect of Access Barring

results from random-draw performed by individual

user devices based on System Information (SI).

That is, user devices shall draw a random number;

if the number is greater than the value indicated in

SI, the user device is prevented from performing

RA; otherwise, it can perform RA. Therefore, the

number of user devices passing through Access

Barring can be seen as a Binomial random variable

providing that the user devices performing RA are

independent in their behavior (“identical” property is

already satisfied).

pro

ba

bili

tyLo

w co

llision

prob

abilit

y

no of UE’s passing through Access Barring

High

collis

ion pr

obab

ility

Figure 1: Effect of user-device population size on

overall collision probability (LTE-Advanced case):

normal load (Green), heavy load (Violet).

In this paper, we formally identify and describe this

behavior. With mathematical formulation of RA

behavior for parameter control, we develop an

efficient method that can improve RA performance,

such as collision probability and RA channel

throughput. The proposed method can be utilized in

future Release of LTE-Advanced after a minor

modification of signaling protocol.

REFERENCES

1. 3GPP TS 36.300, Evolved Universal Terrestrial Radio Access (E-UTRA) and Evolved Universal Terrestrial Radio Access Network (E-UTRAN); Overall description; Stage 2, v.11.1.0, March, 2012.

A Programmable Processing Array Architecture Supporting Dynamic Task Scheduling and Module-Level Prefetching

Junghee Lee*, Hyung Gyu Lee*, Soonhoi Ha†, Jongman Kim* and Chrysostomos Nicopoulos‡ *Georgia Institute of Technology

777 Atlantic Dr NW Atlanta, GA 30332, USA

{junghee.lee, hyunggyu, jkim}@gatech.edu

†Seoul National University

1 Gwanak-ro, Gwanak-gu Seoul 151-742, Korea

sha@iris.snu.ac.kr

‡University of Cyprus

75 Kallipoleos Avenue P.O. Box 20537, 1678 Nicosia, Cyprus

nicopoulos@ucy.ac.cy

SUMMARY

This paper introduces a novel massively

parallel processing array (MPPA) architecture that

supports dynamic task scheduling and module-level

prefetching. The dynamic task scheduling makes it

easier to handle dependencies among tasks and

the prefetching hides the run-time overhead of the

dynamic task scheduling.

I. INTRODUCTION

General-Purpose Graphics Processing Unit

(GPGPU) is a well-known programmable hardware

accelerator. The programming model of GPU is

single instruction multiple data (SIMD) model where

a same instruction is executed on multiple data.

Although SIMD model works well for applications

with regular computation kernels like graphics

applications, it exhibits poor performance on other

applications that are dominated by irregular

computation kernels.

II. PROPOSED MPPA ARCHITECTURE

A. Execution Model

The execution model of the proposed MPPA is

an extension of the even-driven execution model. It

consists of modules and their interconnections. The

module is triggered whenever any registered input

changes.

B. Hardware Architecture

The top view of the proposed MPPA

architecture is a tiled architecture. It consists of

multiple identical core tiles. Each tile has one

processing element, a scratch pad memory and

several peripherals that support the execution

model. There are a dedicated tile for

communication with the host CPU and a tile for

accessing the external memory.

C. Supporting Execution Model

Although every core tile is identical, one or

more core tiles should be designated for the

execution engine. The execution engine is software

running on the core tile. It consists of scheduler,

signal storage and interconnection directory. It

supports the execution model and the prefetching.

III. EVALUATION

We developed a cycle-level simulator. It has 32

core tiles. Its memory access time is one cycle for

the scratch-pad memory and 100 cycles for the

external memory. Each core tile has 8KB scratch-

pad memory.

The recognition, mining and synthesis (RMS)

benchmarks are used for the evaluation. Since the

chosen RMS benchmarks have dependency among

tasks, they would exhibit poor utilization of

processing element if they are implemented with

SIMD model.

Figure 1: Average access times of the scheduler,

and average utilization of the processing elements

(i.e., the core tiles).

In contrast, they can exploit the processing

elements of the proposed MPPA as shown in

Figure 1. It also shows that the module-level

prefetching enhances the utilization.

Overview of High Efficiency Video Coding (HEVC)

Jaehong Chon Qualcomm Inc.

5775 Morehouse Drive San Diego CA 92121

jaehong.chon@gmail.com

SUMMARY

This paper will briefly describe essential parts of

new video coding known as HEVC [1], which will be

finalized in January 2013 aiming to reduce bitrate

by half with H.264 AVC video standard.

In order to achieve this goal, the HEVC uses

variable block size from 8x8 to 64x64 for coding unit,

quadtree structured transform block size from 4x4

to 32x32, NSQT, advanced motion vector prediction,

ALF. The HEVC provides a bit rate savings for

equal PSNR when compared to AVC.

I. INTRODUCTION

The demands on the codecs have grown

severely in recent years with the increased use of

digital video with higher resolution and quality. A

new generation of video compression technology

known as High Efficiency Video Coding (HEVC)

that has substantially higher compression capability

than the existing H.264 AVC standard is being

developed jointly by ISO/IEC MPEG and ITU-

T/VCEG with the goal to reduce bit rate by 50%.

II. ESSENTIAL PARTS OF HEVC

A. CU, PU and TU

The HEVC uses nested and pre-configurable

quadtree structures for flexible picture partitioning.

There are three different units: Coding Unit (CU),

Prediction Unit (PU), and Transfrom Unit (TU). The

CU block sizes are from 8x8 to 64x64 in tree

structure and it can be further recursively

subdivided into the PU for prediction and TU for

DCT-based residual coding.

B. Motion Prediction

There are three inter-prediction modes which

are Inter, Skip, and Merge. The Inter mode

performs motion compensated prediction with

motion vectors (MVs), while the Skip and Merge

modes utilize motion inference methods to obtain

the motion information from spatially neighboring

blocks or a temporal block located in a co-located

picture. Based on the rate-distortion optimization

(RDO) decision, the encoder selects one final MVP

within a given candidate set of MVPs for Inter, Skip,

or Merge modes and transmits the index of the

selected MVP to the decoder.

C. Non Square Quadtree Transform (NSQT)

When a transform block cross the boundary of

motion block, high frequency coefficients are

generated which will impose negative in�uence on

coding performance. The NSQT is used for this

non-square TU.

D. Adaptive Loop Filter (ALF)

The ALF is applied within the prediction loop

prior to copying the frame into a reference decoded

picture bu�er to provide improved objective and

subjective quality.

III. RESULTS

Figure 1 shows the bit rate saving of HEVC for

a WVGA test sequence against H.264.

Figure 1: PSNR for H.264 and HEVC

REFERENCES

1. B. Bross, W.-J. Han, J.-R. Ohm, G. J. Sullivan, T.

Wiegand, “High efficiency video coding (HEVC)

text specification draft 6,” Output Document of

JCT-VC, JCTVC-H1003, Nov. 2011.

An Automatic GPIO Block Generator for a Reliable SoC Architecture

Myoungseo Kim Department of EECS

University of California, Irvine myoungseo.kim@ieee.org

Jean-Luc Gaudiot Department of EECS

University of California, Irvine gaudiot@uci.edu

SUMMARY

The complexity of general-purpose Input/Output

(GPIO) block which implements I/O paths for off-

chip communication has been increased

significantly. GPIO has a many productivity issues

because multi-I/O paths are usually fixed at

relatively late stage of design activities. This feature

makes various human errors when we use the

traditional RTL description. In our case study, we

succeeded on auto-generating a GPIO block in a

system-on-a-chip (SoC) platform which has more

than 200 GPIO pins and 700 PAD pins. We are

finally reduced the amount of manual description for

generating GPIO block by 95%.

I. INTRODUCTION

The pace of the nanotechnology innovation is

excessively fast, it has been required to develop the

new design methodology for the improvement of

design productivity. Especially, automation of

platform integration and verification process has

been proposed under the platform integration

methodology. Moreover, commercial solutions of

design automation for signal multiplexer and PAD

block have been presented [1]. However, many

architectural parameters, such as the maximum

number of I/O pins, PAD control signals, and

various address regions, made it difficult to reuse

the GPIO block. Therefore, RTL designer used to

depend on manual design work in very demanding

conditions which is time-consuming and error-prone.

II. GPIO SPECIFICATION WITH FORMALIZED

DESCRIPTION

In this abstract, we propose a formalized

description methodology for specification which can

effectively express various structure of GPIO design,

and we also propose design automation of the

GPIO block. By using our approach, RTL designer

can generate RTL GPIO block with reducing design

time and keeping consistency among the generated

outputs.

Formalized Description

Table 1 shows the summary of functionalities

and structural parameters which are essential to

implement the GPIO block. Our goal is to develop a

formalized description method to express all the

functions and parameters described in Table 1.

III. EXPERIMENTAL RESULTS

To demonstrate the feasibility of description

capability for GPIO architecture with proposed input

formalized description text format, we use media

platform in some Samsung projects based on 65nm

and 45nm technology. Also, we chose python script

language to make auto-generator. To evaluate the

efficiency of proposed design methodology, we

applied to a typical media SoC platform. We

measured the efficiency ratio of traditional RTL

design time to proposed RTL auto-generating time

in consideration of the number of wafer revision. As

a result, the amount of manual description for

generating the GPIO block was reduced by 95%,

compared to the traditional RTL description. This

automatic GPIO block generator is applicable to the

different platform with simply changing the input

formalized description text format.

REFERENCES

[1]. GensysIO, http://www.atrenta.com/solutions/

gensys-family/gensys-io.htm

Table 1: GPIO Functions and Parameters

Type Function Parameter

ARM I/O

multiplexer

Data I/O through

BUS interface

Bus interface

type

IP blocks

I/O

multiplexer

Programmable I/O

multiplexer for IP

blocks

Control signal

name and I/O

PAD

control Pull-up/down

PAD control

register

http://www.atrenta.com/solutions/

A study of acoustic-to-articulatory inversion based on generalized smoothness criterion for emotional speech

Jangwon Kim

University of Southern California Los Angeles, CA

jangwon@usc.edu

Sungbok Lee University of Southern California

Los Angeles, CA sungbokl@usc.edu

Shrikanth S. Narayanan University of Southern California

Los Angeles, CA shri@sipi.usc.edu

ABSTRACT

The acoustic-to-articulatory inversion of speech

is a way of estimating the vocal tract shape

configurations or articulatory trajectories from

(acoustic) speech signal. It has gained a lot of

attention from researchers due to its benefits and

potentials: the use of estimated articulatory

parameters for better classification and, further, for

better understanding of speech production.

However, the fact that acoustic signals and

articulatory signals are nonlinear and many-to-one

mapping poses a challenge to it. To address this

challenge, several inversion algorithms have been

proposed. Basic idea of them is to incorporate the

knowledge of articulatory movements (articulatory

positions in the vocal tract vary continuously and

slowly, so articulatory trajectories are smooth) so

that we can constrain the articulatory trajectory

estimations.

Acoustic-to-articulatory inversion studies have

been focused on and evaluated in only neutral

speech so far. Despite of the usefulness and

potential of emotional speech, there is no study to

systematically understand how the inversion

algorithms work with “emotional” speech and to

what extent para-linguistic information, in particular

of emotional speech, in the articulatory domain is

preserved during inversion. The goal of this study is

to examine how well emotional variation in the

articulatory domain is preserved during acoustic-to-

articulatory inversion for emotional speech.

In this study, the inversion algorithm based on

the generalized smoothness criterion (GSC) [1] is

used. GSC works on the principle of constraining

each articulatory trajectory by the corresponding

articulator-specific smoothness (the smoothness is

articulator-dependent). GSC learns configurations

for the smoothness of each articulator from data

and preserves it in the estimated articulatory

trajectories. We expect that emotion may influence

the smoothness of each articulator, because

emotional speech has different articulatory

characteristics, e.g. in terms of articulatory position

and velocity [2].

The list of analyses in this study is following.

Firstly, we examine the degree of similarity between

estimated and original articulatory trajectory pairs

for the same and different emotions. Secondly, we

examine the amount of emotion-specific information

(ESI) of the estimated trajectory and its usefulness

for emotions research. Thirdly, we analyze the

parameters (of the inversion method) learned from

different emotional data and examine their effects

on preserving ESI during inversion. Lastly, for

subject-independent inversion (e.g. as in [3]) we

compare the different representation schemes of

general acoustic space in the sense of how well ESI

is preserved during acoustic mapping.

We use speech waveform and articulatory

trajectory recorded by Electromagnetic

articulography. The data includes five elicited

emotions spoken by three native American

speakers. Experimental results for the first and the

second experiments suggest that the ESI in the

estimated trajectory, although smaller than in the

direct articulatory measurements, is complementary

to that in the prosodic features, hence suggesting

the usefulness of estimated articulatory data for

emotions research. Experiments for the other

analyses are on-going.

REFERENCES

1. Ghosh, P. K., Narayanan, S. S., “Generalized

smoothness criterion for acoustic-to-articulatory

inversion,” J. Acoust. Soc. Am., vol. 128, no. 4, pp.

2162-2172, 2010.

2. Lee. S., Yildirim, S., Kazemzadeh A., Narayanan, S. S.,

“An articulatory study of emotional speech production,”

in Proceedings of Interspeech, pages 497-500, 2005.

3. Ghosh, P. K., Narayanan, S. S., “A subject-

independent acoustic-to-articulatory inversion.” in

Proceedings ICASSP, Prague, Czech Republic, 22-27

May, 2011.

Peak shaving control with adaptive filter to reduce

size of energy storage system Yongduk Lee

University of Connecticut

371 Fairfield Way, Storrs, CT, USA

yongduk@engr.uconn.edu

Sung-Yeul Park University of Connecticut

371 Fairfield Way, Storrs, CT, USA

supark@engr.uconn.edu

SUMMARY

This paper presents a peak shaving control of a battery

energy storage system (BESS) to reduce its size including

volume, power rating, voltage range, and cost of energy storage

system. In order to achieve a function, this approach used an

auto-fit frequency adaptive filter. It will bring the reduced

storage system size comparing with conventional method.

I. INTRODUCTION

Most intermittent renewable energy sources (RES) such as

wind power and solar power generate electricity and send it to

the grid. Its fluctuating behaviors affect to the grid voltage [1].

If power fluctuation of the grid can be reduced, the grid voltage

will be more stable. The main operation of BESS is to reduce a

power fluctuation of grid voltage using the charging /

discharging of battery energy. This function called the peak

shaving. In order to perform this function, BESS has to

separate low frequency and high frequency elements of grid

power. BESS mainly take care of the high frequency

component of the grid power.

This high frequency element of the grid power is generated

by RES. It will be the power reference of BESS to compensate

the voltage fluctuation. Generally, low pass filter (LPF) is used

for separating high frequency [1]. However, high frequency

component of the grid power varies widely because of

intermittent characteristics of RES. Therefore, usually, the

fixed cut off frequency of LPF is factor to increase the energy

storage size, because low frequency bandwidth needs larger

reference for charging / discharging of BESS.

This paper presents to control the peak shaving using the

auto-fit adaptive filter. It will reduce the size comparing with

conventional method up to28%.

II. BESS control algorithm for the peak shaving

The proposed control scheme is to combine LPF and

adaptive filter mechanism. The way of getting suitable cut-off

frequency is to use a phase detector and reflect to the digital

controllable LPF. The phase detector of the adaptive filter

carries out the low frequency detection to adjust the cutoff

frequency. The fit cut-off frequency is applied to LPF. This

filter tracks the fit frequency to reduce BESS peak power.

III. Simulation

Figure 1 (a) ~ (c) show the fluctuated power and

compensated power using conventional method. Figure 1 (d) ~

(f) show results of proposed idea. Both results show smoothing

power results and reducing voltage fluctuation. However,

BESS power of conventional method is higher than that of the

proposed idea. Through these results, the power rating of the

energy storage can reduce up to 28%. It will turn out to be

reduced the overall size. Also, the voltage fluctuation is

reduced comparing with the conventional method.

Figure 1 Simulation results of peak shaving: (a) Grid power and RES power by conventional method, (b) BESS power by conventional

method, (c) the voltage fluctuation in conventional method, (d) Grid power and RES power by proposed idea, (e) BESS power by proposed

idea, and (f) the voltage fluctuation in proposed idea.

IV. Conclusion

This paper presents a peak shaving control of a BESS to

reduce size of energy storage system. As results, the fluctuated

power is compensated by BESS and voltage is stable. Also this

paper shows to reduce energy storage system size using the

proposed idea.

REFERENCES

1. Kai, T. and Tanaka, A., “A new smooth scheme for power

fluctuation using inverter of wind power generation with

doubly fed induction generator,” Electrical Machines and

Systems, 2008. ICEMS 2008. International Conference on,

pp. 2390 - 2395, 2008.

Specialized Array Architecture using Block-level Approach for H.264/AVC Intra-Picture Coding

Myoungseo Kim

Department of EECS University of California, Irvine

myoungseo.kim@ieee.org

Jean-Luc Gaudiot Department of EECS

University of California, Irvine gaudiot@uci.edu

SUMMARY

Dedicated hardware architecture for intra-

prediction with reorganizing neighbor pixels is

proposed. The core architecture is based on a

conventional 1-D systolic array for inter-prediction

and additional adders for intra-prediction at the

prediction stage. Thus, the proposed architecture

can decrease computational complexity significantly

without any degradation of the signal to noise ratio.

I. INTRODUCTION

In H.264/AVC [1], inter-prediction part is

growing more in complexity and intra-prediction is

enhanced with different methods.

Dedicated hardware architecture for intra-

prediction, called a neighboring pixel pre-

processing unit (NPPU), presents in this abstract.

The basic architecture is a processing element (PE)

array running in parallel with a conventional 1-D

systolic array exploited by motion estimation of

inter-prediction. Input neighboring pixels are

reorganized effectively for a pipelined simultaneous

data flow..

II. PROPOSED ARCHITECTURE

The core of proposed architecture is to

reconfigure a conventional 1-D systolic array that

has been used for accelerating the motion

estimation in the previous standards. This is

feasible because intra-prediction also computes the

sum of absolute differences (SAD) to determine the

best mode, like inter-prediction. Here NPPU, which

can produce predictors in advance, is attached.

Outputs of NPPU become the inputs of t he systolic

array. NPPU is constructed as 3 inputs and 1 output

per one processing element (PE). Result values of

NPPU flow into inputs of the systolic array and

those 16 pixels belonging to the current macro

block (MB) flow into other inputs of it.

A. Operations

For 4x4, original neighboring pixels of a given

current block are arranged in the frame buffer.

Arranged pixels enter into NPPU based on the data

flow schedule and are computed as the predictor.

The systolic array produces the SAD by using these

predictors and 16 pixels of a current block flowing

into the array every cycle. For the first three cycle,s

of every period, the front-end multiplexer (Mux)

near to PE selects the original value to calculate the

mean of 8 pixels for DC mode. For this time,

accumulator (Acc) starts operation and adds the

results of PE. At the third cycle it stops, its value is

shifted right by three and then is broadcasted to all

back-end Mux. Except 0 and 1 mode, the one mode

checking for 4x4 block can be achieved every cycle

after first 15 cycles. We can obtain the best mode

every 9 cycle for one 4x4 block and make minimal

cost for 4x4 intra-prediction at the 144th cycle. If the

minimal SAD for 4x4 intra-prediction is produced,

16x16 mode checking is executed subsequently.

The process is similar to 4x4 intra-prediction except

the data arrangement.

III. SIMULATION RESULTS

The simulation using JVT software JM8.2 [2] is

performed on the QCIF sequences, i.e., Foreman,

and Highway. We eliminate the rounding operation

in all intra-prediction calculations and encode

sequences as only I-picture for showing the effect

of rounding certainly. In this case, the PSNR is

almost not affected although a few best modes for

each block are changed. A little change of total size

depending on the best modes with the rounding

removal is not so significant, considering a huge

total size.

REFERENCES

[1]. ISO/IEC 14496-10, “Advanced video coding for

genetic audiovisual service”, 2005

[2]. Joint Video Team (JVT) reference software

JM8.2, May, 2004.

Sensor Fusion for Advanced Driver Assistant System

H. Kim, E. Park, Y. Yun, and J. Lee Dept. of Information & Communication Engineering

Inha University, Incheon, Korea hikim@inha.ac.kr, {espark, yjyun, jhlee}@vision.inha.ac.kr

SUMMARY

This talk briefly presents the state-of-the-art

advanced driver assistant system (ADAS) for

intelligent vehicles. In particular, a lane departure

warning system is implemented in a smartphone

utilizing both camera and GPS sensor.

I. INTRODUCTION

Recently, commercial vehicles have become

more intelligent by adopting information and

communication technology (ICT) for the purpose of

avoiding traffic accidents [1]. As depicted in Fig. 1,

this presentation introduces a lane detection system

using a smartphone, a system of generating rear-

side panorama video using four cameras, and

performance comparison of three pedestrian

detection algorithms.

Figure 1: Proposed ADAS.

II. ADVANCED DRIVER ASSISTANT FUNCTIONS

A. Lane Detection using Smartphone

Existing lane detection methods are depending

on visual features (color, geometric) acquired by a

vision sensor [2]. They adopt the hypothesis and

verification mechanism requiring a large amount of

computational complexity. In order to reduce the

computation burden, the GPS position data can be

utilized in lane detection for predicting the exact

direction of the lane. This presentation introduces a

robust lane detection system by combining the

camera and GPS receiver on a smartphone.

B. Rear-side Panorama View Generation

This presentation also introduces a system of

generating rear-side panorama view using four

video cameras mounted on side mirrors and tail

lights of a car. The system requires camera

calibration and employs road segmentation using

entropy and saturation of the ground plane.

C. Real-time Pedestrian Detection

Many approaches have been studied for human

detection in various environments. They can be

applied to pedestrian detection for advanced driver

assistant systems [3]. In this presentation, three

promising techniques, Bidirectional PCA, PCA, and

HoG, are compared in the aspects of detection

rates and processing time.

Table 1: Performance comparison for pedestrian

detection.

Algorithm BDPCA PCA HOG

Training Fast Slow Fast

Feature Vertical gradient

Intensity Histogram of gradient

TP rate at FP rate of 0.01

95.93% 20.52% 91.32%

Proc. time 1x 18x 0.31x

III. CONCLUSIONS

Convergence of ICT in automotive industry

seems boundless and will be accelerated for the

next decades. Collaboration of experts from

multiple disciplines is the key to the success in this

realm.

REFERENCES

1. M. W. Park, K. H. Jang and S. K. Jung,

“Panoramic vision system for an intelligent

vehicle using a laser sensor and cameras,” The

17th Intelligent Transport Systems World

Congress, 2010.

2. S. Suh and Y. Kang, “A robust lane recognition

technique for vision-based navigation with a

multiple clue-based filtration algorithm", Int. J. of

Control, Automation, and Systems, vol. 9, no. 2,

pp. 348-357, 2011.

3. M. Enzweiler, and D. Gavrila, “Monocular

pedestrian detection: Survey and experiments,”

IEEE Trans. on PAMI, vol. 31, no. 12, pp. 2179 -

2195, 2009.

Acoustic and Image based Context Awareness Using Mobile Device

Hanseok Ko Woohyun Choi School of Electrical Engineering, Korea University, Seoul, Korea, hsko@korea.ac.kr

SUMMARY Multisensory fusion architecture on a mobile device can be designed that can sense and determine user contextual information in real time, such as where the user is, what the user does, or how healthy the user is by acquiring and processing relevant acoustic and visual signals as well as those personal health related signatures. By exploiting acoustic and visual signals using sensors equipped on mobile devices, we develop an effective context awareness system. An overall architecture of such system is outlined first and then the enabling algorithms performing detection and classification of various contextual scenarios are described. Representative experiments demonstrate the feasibility of the schemes explored wherein their actual implementation into mobile device such as smart-phone confirms its effectiveness toward practical realization.

I. INTRODUCTION

In this paper, we explore new capabilities for enabling the intelligent mobile devices. In particular, it is desirable for smart-phones with ability to sense and recognize user’s contextual information such as user activities, surrounding environment, user’s personal health condition, and provide context relevant information for user’s current needs. In short, one of the key anticipated future capabilities of smart devices is “Context Awareness (CA)” and CA is expected to provide new insights to improving the life quality by life-logging and timely warning of health hazardous situations.

II. CONTEXT AWARENESS SYSTEM

The envisioned CA can be thought of as the “blackbox onboard a vehicle”, which continuously monitor and provide the time essential information to users. Though many earlier studies use acoustic signal alone reported good performance, there are obviously limitations in accuracy and extent of the complex context the acoustic only CA can perceive and recognize. It is also desirable to further classify the surrounding environment into either indoor or outdoor and whether it’s dark or bright. In other words, given just an acoustic sensing capability, a person may have to estimate the visual context or

environment, which limits the scope of overall CA capability. It becomes very difficult to sense or recognize the visually meaningful context by using acoustic signal alone and has limitations for recognizing the complex contexts.

Many resources such as microphone, camera, gyroscope, accelerometer, luminance, Global Positioning System (GPS), and etc., are available for sensing and capturing the contextual information. Among them, Audio/Acoustic and visual sources sensing from a microphone and a camera respectively provide much more information for CA capable system. Furthermore, microphone and camera are readily available as they are embedded in many mobile devices. Thus, in this paper, we address a more general CA capable system employing both the acoustic and visual signals from microphone and camera respectively without additional sensors, and develop a feasible implementation in a smart-phone system.

Figure 1. Overall structure of context awareness system.

The indoor/outdoor classifier uses two features such as a color distribution histogram as a color feature [13] and Local Binary Pattern (LBP) as a texture feature of all sub-images [14], the joint classifier, k-means clustering algorithm and k Nearest Neighbor, for classification

A few representative scenarios were identified with advantages and issues addressed. The experimental results show that the mobile device capable of CA sensitive to user’s situation using acoustic and visual signal is feasible.

Acknowledgment This research was supported by Seoul R&BD.

Power Optimization in a Parallel Multiplier using Voltage Islands

Seok Won Heo Computer Science Department

University of California at Los Angeles Los Angeles

comace@cs.ucla.edu

SUMMARY

We present an approach to reducing power

dissipation in the design of parallel multipliers by

utilizing voltage islands corresponding to non-

uniform arrival of inputs to the final adder. Our

approach reduces dynamic power dissipation up to

16.2% in the 32 x 32 bit multiplier with a 3.6%

increase in delay.

I. INTRODUCTION

Dynamic power dissipation is the dominant

factor in the total power consumption of a CMOS

circuit and typically contributes over 60% of the total

system power dissipation [1]. It can be described by

Pdynamic = 0.5 ∙ CL ∙ VDD2 ∙ fp ∙ N

where CL is the load capacitance, VDD is the power

supply voltage, fp is the clock frequency, and N is

the switching activity. The equation indicates the

power supply voltage has the largest impact on the

power dissipation due to its squared term factor.

Unfortunately, the lowering power supply voltage

causes speed penalties. A great deal of effort has

been expended in recent years on the development

of the techniques to utilize the low power supply

voltage while minimizing the performance

degradation. Using voltage islands is one way to

mitigate such performance degradation by

architectural changes of the circuit [2]. This paper

proposes a scheme to achieve power savings in a

parallel multiplier of a tree type by utilizing voltage

islands.

II. PROBLEM & SOLUTION

We propose to exploit the non-uniform arrival

time profiles of the tree multiplier to achieve power

savings with minimal performance degradation.

Specifically, we apply voltage islands technique to

the regions of non-uniform input generated by

partial product reduction tree. That is, adder is

partitioned into blocks that operate with different

power supply voltages. A voltage island occupies a

contiguous physical space and operates at one

supply voltage. Such voltage island techniques are

applied to the tree multiplier so that the units of the

multiplier get different levels of voltage support, as

profiled by their performance requirements. The

slowest region of the tree multiplier is the middle

region at which the arrival time is large and

constant. It requires higher supply voltage level in

order to maximize element’s performance. On the

other hand, the other regions may run at lower level

of supply voltage because they are not on the

critical path. These regions are (1) the least

significant part at which arrival time increases from

the least significant bit (LSB) towards middle region,

and (2) the most significant part where arrival time

decreases from the middle region towards the most

significant bit (MSB). An example of a partition into

Low-High-Low islands is shown in Fig. 1.

Figure 1: Structure of the partial product reduction

tree by using voltage islands

REFERENCES

1. J. Rabaey, Low Power Design Essentials,

Springer, 2009

2. D. E. Lackey, P. S. Znchowski, T. R. Eednar, D.

W. Stout, S. W. Gould, and J. M. Cobn,

“Managing power and performance for system-

on-chip designs using voltage islands,” in Proc.

ICCAD, Nov. 2002, pp. 195–202

Improvements of Visual Inspection System for Semiconductor Package Using Moire Effect

Young Hwan Jang* Samsung Institute of Technology

San #24 Nongseo-Dong, Giheung_Gu, Yongin-City,

Gyonggi-Do 446-711 Korea yw00.jang@samsung.com

Kum Jin Yun Samsung Electronics co., Ltd.

San #74 Buksu-Ri, Baebang-Eup, Asan-City, Chungcheongnam-Do

336-851 Korea yun117@samsung.com

SangKeun Kwak Samsung Institute of Technology

San #24 Nongseo-Dong, Giheung_Gu, Yongin-City,

Gyonggi-Do 446-711 Korea skkwak@samsung.com

SUMMARY

A semiconductor package visual inspection system using Moire effect is presented. Currently, stereo and laser method is being mainly used for semiconductor visual inspection process. The experiment of this paper compared with the projection Moire method using 4-bucket phase-shifting algorithm to the conventional method of the stereo and laser method. The method for measurement used in an experiment used gage R&R method. The experiment results of %R&R and %accuracy data showed the result that Moire method was improved than conventional methods. Therefore, we verified that a package visual inspection method using Moire effect is the improved method for measurement rather than conventional methods.

I. INTRODUCTION

3D visual inspection methods of package generally are divided into laser and stereo method. The necessity of the precise and rapid method in the visual inspection system can be increased so as to shrink the size of the semiconductor package. Because the defect of the package outline can cause damage of the whole Module which a package assembled. Therefore, the value of package inspection result must need to have high reliability. Nowadays, as for the 3D method for measurement used in package visual inspection process, laser and stereo method make the mainstream of the inspection methods. In this paper, we proposed that inspection method using Moire effect and verified the proposed model using the comparison with conventional model and proposed model for accuracy, reliability through the experiment. The paper is organized as follows: we first introduce the conventional model in Section II. Then we describe the proposed model in Section III. Measurement results are provided in Section IV. Finally, conclusions are given in Section V.

II. CONVENTIONAL MODEL

Machine vision system which has been used for the package visual inspection uses two kinds of 3D algorithms. First, there are two ways which point laser and line laser (slit beam). Because the optical displacement sensor of the point laser method measures only a piece of height at a time, there are problem that it takes a lot of measuring time. To overcome this issue, the slit beam projection methods are used as shown in Fig. 1(a). In order to concentrate the laser beam from the laser diode using the condenser lens and we create the slit beam using index lens or cylindrical lens. The slit beam projection method get deformed the slit beam depending on shape by reflected slit beam and acquire deformed the slit beam using charge coupled device (CCD) camera. We calculate the 3D coordinates from geometric relationships. This method is measured a cross sectional three-dimensional coordinates with a non contact type. Therefore, it must transfer a measurement cross section with machine drive system to measure a whole shape. The slit beam reflects in a mirror and scans the object surface through the angle variation of the mirror using galvanometer. The measurement resolution and accuracy of the slit beam is highly dependent on the thickness of the slit beam and camera calibration.

(a) (b)

Fig. 1. (a) Line laser method, (b) stereo method.

Second, the stereo method using CCD cameras are shown in Fig. 1(b). In case of the stereo method is passive method that does not project to the object. If we knew the distance between the camera and the camera, the included angle, and the coordinates of the two cameras when we measured the same points can be calculate the 3D position information by principle of trigonometry. The stereo method has the limiting factor that is the correspondence problem. This problem cannot measure the distance of the object does not have a characteristic. For example, in case of the inspection results of the white walls from left and right camera, the image of the results is all white. So, the distance to the surface of the walls cannot be calculated.

III. PROPOSED MODEL

Moire effect is a kind of interference fringe created by overlap among the two or more periodic pattern. Currently, it is used to the various fields. By changing the Moire fringe, it can be obtained the 3D coordinates and information of much amplified displacement of actual object.

Fig. 2. Displacement amplification effect.

(1)

(2)

In Fig. 2, ① is linear grating which have a period

of A, ② have a same period as ①, but the rotation state is counterclockwise direction as θ. If these two gratings were overlapped, it can be obtained the

Moire fringe such as ③. The period B have a relationship such as formula (1) in this moire fringe and it is very large number than original grating period A. If 2/A moves a grating to the horizontal direction in this status, the Moire fringe was

obtained such as ④. Eventually, the movement of grating was amplified as the formula (2) times. When we move the grating on an object in three dimensional space, the Moire fringe is changed by depending on the shape of an object. By calculating this effect, the behavior of an object or a non-contact shape measurement is possible to measure. Fig. 3 shows the method that it used this principle in package visual inspection. This method adopted the phase shifting method using the projection Moire, and the movement of grating is the half period of grating. Among them, we adopted 4-bucket algorithm that it was 4 times to move the phase. The reason why using this method, we can have less error than the measurement method using the straightness of the fringe or the order of a fringe, and also it have an advantage that it was not affected by the shape of the object.

Fig. 3. 4-Buckets Algorithm.

Fig. 4. Inspection image of BGA package ball using 4-bucket algorithm.

2sin2

AB

2sin2

1

B

Fig. 5. Image results of the 3-dimension using 4-bucket algorithm.

Fig. 6. The problem of 2π ambiguity during the case of the phase unwrapping.

Fig. 7. The principle of the vernier caliper.

Fig. 4 shows the image that the ball shape of the ball grid array (BGA) using the projection Moire with 4-bucket algorithm. The BGA is a kind of the packages. If we measure the distance of this image, we can obtain the information of three dimensional positions as shown in Fig. 5. A function showing height in Moire is trigonometric function as shown in Fig. 3. In other words, the phase unwrapping from the consecutive value was occurred because of periodic function. But we don't know the height that

suddenly jumping the height value in the 2 π period as shown in Fig. 6. This is known for the 2π ambiguity. In order to overcome the 2π ambiguity, we should be used both the stereo camera and Moiré effect. The small number was measured by using the Moire method, and the large number was measured by using the stereo method as shown in Fig. 7.

IV. EXPERIMENT A. Methods

For each test methods the results of gage R&R and accuracy dispersion were measured with the calibrated golden device as a reference. The test ability of inspection system only is considered. The experiment systems were used in the semiconductor package product line.

B. Procedures

First, auto-visual inspection equipment using laser, stereo, Moire method with 4-bucket algorithm was tested. Secondly, the measurement was repeated 3 times in the three people. Thirdly, 10 sample points of the golden devices was selected. Thus, 90 samples of the test case were measured.

C. Result

Fig. 8. Inspection image of Gage R&R using Moire method for the test package.

Fig. 9. Inspection image of Gage R&R using Slit Beam for the test package.

Fig. 10. Inspection image of gage R&R using stereo method for the test package.

We had gathered the results of inspection image and measurement values from each method as shown in Fig. 8 and Fig. 9 and Fig. 10. The measurement result of Moire method has been improved for the results with others method shown in Fig. 11. The gage R&R result of Moire method is 3.27% and the accuracy is 0.26%.

Fig. 11 Comparison results between Moire method and others method.

V. CONCLUSION

In this paper, the test method of visual inspection using Moire effect for semiconductor package was presented. The problems between laser and stereo method were slow down the test speed and had a limiting factor of correspondence problem. However, these problems can be overcome using Moire method. In order to apply Moire method to semiconductor package visual inspection process, Test samples must be considered for the test speed and durability because of moving the grating and complexity of mixing the stereo method to overcome the 2π ambiguity.

REFERENCES 1. S.W. Kim, “Reverse engineering: high speed

digitization of free-form surfaces by phase-shifting grating projection Moire Topography,” Machine Tools & Manufacture, pp. 389-401, 1999.

2. Yoon-Chang Park, “The Principle and Applications of Moire Phenomenon,” Journal of the Korean Society for Precision Engineering, Vol.17, No.6, pp25-31, 2000.

3.Frank Chen, “Overview of Three-dimensional Shape Measurement using Optical Methods.” Optical Engineering, Vol. 39 No.1, Jan. 2000.

4.Ju Hyun Park. “A Study on the N-Buckets Algorithm error of two Wavelength Projection Moire.” Master Thesis., the Chon Buk University, 2003.

Development of Automated Steering System for Unmanned Driving of Electrical Vehicle

Yong-jun Lee

Mokpo National University Department of Control & Robot

Engineering, Mokpo National University, Korea

yongjun@mokpo.ac.kr

Young-Jae Ryoo Mokpo National University

Department of Control & Robot Engineering,

Mokpo National University, Korea yjryoo@mokpo.ac.kr

Dae-yeong Lim Korea Institute of Industrial

Technology Honam Regional Division

dylim@kitech.re.kr

SUMMARY In this paper, a manned and an unmanned

steering mechanism using an electric clutch for an electric vehicle is proposed. When electric vehicles have unmanned driving function, to steer for a vehicle to follow along the path would be required. In this paper, a steering mechanism was developed of electric vehicles for unmanned driving. Steering motor is using BLDC motor, electric clutch is used to transmit motor power to steering axis steering mechanism. Also, electric vehicle driving was tested to examine usefulness of developed steering mechanism. I. INTRODUCTION

In unmanned driving system, precise steering angle control is an important task for stable driving.

Electric Power Steering (EPS) has a representative in the automatic steering system. But, the EPS has a problem that steering bar could be locked to protect itself from overheating of motors.

In this paper, how to overcome the problem of EPS as steering system using electric clutch is proposed. When Vehicle is unmanned driving, the automated steering system is connected to steering axis by electric clutch turn on. When vehicle isn't unmanned driving, the automated steering system is disconnected from steering axis by electric clutch turn off. Thus, steering bar locking problem can be overcome.

To implement the proposed method, a steering system has been developed. Usefulness and practicality of the proposed method was tested