JPEG2000 TRANSMISSION OVER WIRELESS CHANNELS USING UNEQUAL POWER ALLOCATION by Mahin Torki B.Sc., Isfahan University of Technology, Iran, 2004 a Thesis submitted in partial fulfillment of the requirements for the degree of Master of Applied Science in the School of Engineering Science c Mahin Torki 2009 SIMON FRASER UNIVERSITY Fall 2009 All rights reserved. This work may not be reproduced in whole or in part, by photocopy or other means, without the permission of the author.
Declaration of Partial Copyright Licence The author, whose copyright is declared on the title page of this work, has granted to Simon Fraser University the right to lend this thesis, project or extended essay to users of the Simon Fraser University Library, and to make partial or single copies only for such users or in response to a request from the library of any other university, or other educational institution, on its own behalf or for one of its users. The author has further granted permission to Simon Fraser University to keep or make a digital copy for use in its circulating collection (currently available to the public at the Institutional Repository link of the SFU Library website <www.lib.sfu.ca> at: <http://ir.lib.sfu.ca/handle/1892/112>) and, without changing the content, to translate the thesis/project or extended essays, if technically possible, to any medium or format for the purpose of preservation of the digital work. The author has further agreed that permission for multiple copying of this work for scholarly purposes may be granted by either the author or the Dean of Graduate Studies. It is understood that copying or publication of this work for financial gain shall not be allowed without the author s written permission. Permission for public performance, or limited permission for private scholarly use, of any multimedia materials forming part of this work, may have been granted by the author. This information may be found on the separately catalogued multimedia material and in the signed Partial Copyright Licence. While licensing SFU to permit the above uses, the author retains copyright in the thesis, project or extended essays, including the right to change the work for subsequent purposes, including editing and publishing the work in whole or in part, and licensing other parties, as the author may desire. The original Partial Copyright Licence attesting to these terms, and signed by this author, may be found in the original bound copy of this work, retained in the Simon Fraser University Archive. Simon Fraser University Library Burnaby, BC, Canada Last revision: Spring 09
Abstract We address the issue of JPEG2000 transmission over wireless channels using Unequal Power Allocation (UPA). First, an algorithm for multicasting JPEG2000 images over MIMO systems is proposed. According to antenna selection orders requested by receivers, the channel assignment unit transmits the substreams with different powers from antennas. The decoder is modified such that when two copies of a substream are available, they are used jointly to decode the image. Second, we propose a UPA algorithm to transmit JPEG2000 images over flat-fading channels. The contribution of coding passes in the quality of the decoded image is extracted. According to these contributions, the optimization algorithm assigns different transmission powers to coding passes. When CSI is available, the powers are adjusted at the transmission time to compensate for the channel fading. Simulation results show that in both works presented, UPA acts as a simple yet fast method for image transmission over wireless channels. iii
To my parents, Abbasali and Fatemeh and my beloved husband, Ali iv
Opportunity is missed by most because it is dressed in overalls and looks like work -Thomas Edison v
Acknowledgments It is my pleasure to take this opportunity to thank those who helped me through my graduate studies and made writing this thesis possible. My supervisors Dr. Atousa Hajshirmohammadi and Dr. Ivan Bajić provided me guidance and support. I owe my deepest gratitude to Atousa, who has made available her support in a number of ways. Her encouragement, guidance and support from the initial to the final level enabled me to develop an understanding of the subject, and accomplish this thesis. I would also like to express my gratitude to Ivan, whose enthusiasm and expertise in providing valuable feedback helped me improve my work. I would also like to thank Dr. Sami Muhaidat for the patience to read through this thesis and providing constructive feedback. I would like to thank Dr. Parvaneh Saeedi for chairing the defence. Many thanks to my fellow graduate students in multimedia communication lab at Simon Fraser University. I would thank my parents for all the support they have provided me over the years, and my lovely husband who always took the time to listen, and encourage me to work harder and harder. Lastly, I offer my regards and blessings to all of those who supported me in any respect during the completion of this thesis. vi
Contents Approval ii Abstract iii Dedication iv Quotation v Acknowledgments vi Contents vii List of Tables x List of Figures xi 1 Introduction 1 1.1 Motivation..................................... 1 1.2 Literature Review................................. 3 1.3 Thesis Contribution................................ 6 1.4 Thesis Outline................................... 7 2 Background 9 vii
2.1 Mobile Radio Channel Model........................... 9 2.1.1 Multipath Fading Channels........................ 10 2.1.2 Parameters of the Fading Channel.................... 11 2.1.3 Types of Fading.............................. 15 2.2 The JPEG2000 Still Image Compression Standard............... 17 2.2.1 Overview.................................. 18 2.2.2 Features of the JPEG2000 Standard................... 21 3 Multicasting JPEG2000 Images over MIMO Systems 25 3.1 Introduction..................................... 25 3.2 System Description................................. 27 3.2.1 Channel Model............................... 28 3.2.2 Transmitter................................. 29 3.2.3 Receiver................................... 31 3.3 Proposed Method.................................. 32 3.3.1 Channel Assignment Algorithm...................... 32 3.3.2 Modified JPEG2000 decoder....................... 34 3.4 Simulation Results................................. 35 3.5 Conclusion..................................... 39 4 Wireless JPEG2000 Transmission Using UPA 44 4.1 Introduction..................................... 44 4.2 System Overview.................................. 45 4.3 Optimized UPA................................... 47 4.4 Implementation of the UPA............................ 50 4.5 Simulation Results................................. 53 4.6 Conclusion..................................... 61 viii
5 Conclusion 63 A User Guide 68 A.1 Files on the DVD.................................. 68 A.1.1 PartI-Chapter3............................... 68 A.1.2 PartII-Chapter4.............................. 69 Bibliography 72 ix
List of Tables 2.1 Typical measured values of RMS delay spread in different environments [16]. 13 2.2 Typical calculated values of coherence bandwidth in different environments based on the values given in Table 2.1...................... 14 x
List of Figures 2.1 Fading Types.................................... 16 2.2 The general block diagram of JPEG2000 codec structure............ 18 2.3 One-level decomposition of Lena image (512 512 pixels) by applying Discrete Wavelet Transform................................. 19 2.4 Two-level decomposition of Lena image (512 512 pixels) by applying Discrete Wavelet Transform. The image is decomposed into three resolution levels, R 0, R 1, and R 2..................................... 20 2.5 Structure of Precincts and Codeblocks...................... 21 3.1 System model for multicasting JPEG2000 images over MIMO channels.... 28 3.2 N T by N R MIMO channel matrix......................... 29 3.3 Division of the raw bitstream into blocks of length 2T. The channel assignment unit runs on each of these blocks independently.............. 30 3.4 Structure of the MMSE Receiver for separating the transmitted data bitstreams over a MIMO channel........................... 32 3.5 Modified JPEG2000 decoder............................ 35 3.6 PSNR curves for different schemes in transmission of Lena image (512 512) 37 3.7 PSNR curves for different schemes in transmission of Peppers image (512 512)......................................... 38 xi
3.8 PSNR curves for different schemes in transmission of Barbara image (512 512)......................................... 39 3.9 Barbara image results for different schemes at SNR=15 db (a) Received image by User 2 with antenna selection based on User 1, PSNR=18.41 db (b) Received image by User 1 with antenna selection based on User 1, PSNR=23.93 db (c) EPA-MD, PSNR=24.25 db for both users (d) UPA-JAAMD, PSNR=24.87 db for both users.................................. 40 3.10 Peppers image results for different schemes at SNR=15 db (a) Received image by User 2 with antenna selection based on User 1, PSNR=13.69 db (b) Received image by User 1 with antenna selection based on User 1, PSNR=20.35 db (c) EPA-MD, PSNR=25.81 db for both users (d) UPA-JAAMD, PSNR=28.82 db for both users.................................. 41 3.11 Lena image results for different schemes at SNR=15 db (a) Received image by User 2 with antenna selection based on User 1, PSNR=15.46 db (b) Received image by User 1 with antenna selection based on User 1, PSNR=25.96 db (c) EPA-MD, PSNR=26.69 db for both users (d) UPA-JAAMD, PSNR=27.78 db for both users.................................. 42 4.1 System model.................................... 45 4.2 Effect of different number of groups on the decoded PSNR of 512 512 Lena image at 0.25 bpp (n g is the number of groups)................. 52 4.3 Effect of the constant parameter α on decoded PSNR curves for the 512 512 Lena (0.25 bpp).................................. 55 4.4 PSNR curves for the transmission of 512 512 Lena image at 0.25 bpp with and without CSI, compared to EPA.................... 56 xii
4.5 Visual comparison in transmission quality of Lena image (at 0.25 bpp and f D = 10 5 ) for different schemes at SNR=10 db (a) EPA, PSNR=23.10 db (b) UPA without CSI, PSNR=24.13 db (c) UPA with CSI PSNR=29.85 db 57 4.6 Visual comparison in transmission quality of Peppers images (at 0.25 bpp and f D = 10 5 ) for different schemes at SNR=10 db (a) EPA, PSNR=23.52 db (b)upa without CSI, PSNR= 24.23 db, (c) UPA with CSI, PSNR= 27.43 db.......................................... 58 4.7 PSNR comparison for the transmission of the 512 512 Lena image (0.25 bpp), f D = 10 5.................................. 59 4.8 PSNR comparison for the transmission of the 512 512 Peppers image (0.25 bpp), f D = 10 5............................... 60 xiii
Chapter 1 Introduction 1.1 Motivation With the explosive growth in communication networks around the world in recent years, the geographical distance is no longer a barrier for communication and the world is now considered a global village. Information exchange has never been easier for users who cross the borders in the virtual world without any restrictions. Users are connected to the World Wide Web or the Internet through wired or wireless connections and enjoy a wide variety of services and applications such as video conferencing, video streaming [12], social networking, etc. Among various services available, of particular interest to the users are multimedia applications. Rapid developments in technology have made all kinds of multimedia services available in a single Personal Digital Assistant (PDA) or a cell phone handset. Having a mobile handset, the users can get pleasure from connecting to the web through wireless channels and watching high quality videos and images from online resources. Multimedia applications increasingly require efficient transmission of still and moving images over wireless channels and users are expecting higher quality image and video services. The ultimate goal of future-generation wireless communication systems is to provide 1
CHAPTER 1. INTRODUCTION 2 ubiquitous seamless connection between mobile terminals and computer servers, so that users can enjoy high-quality multimedia services at any time without wires. Increasing demand for high-speed and efficient multimedia transmission over wireless networks has driven tremendous research on enhancing the performance of multimedia communications over noisy channels. Fundamental physical challenges such as channel fading and interference, however, have put strains on the radio resources, which makes achieving robust multimedia communications difficult. With the advent of scalable image and video standards such as JPEG2000 [24] and H.264 [18], robust transmission of scalable image and video contents over wireless channels has received much attention in recent years. In this thesis, we focus on the transmission of scalable images over error prone wireless channels. Scalable bitstreams are more prone to channel errors caused by fading and interference than plain data bitstreams; Due to the type of data dependency in scalable bitstreams, even a single bit error in the encoded image can cause error propagation and affect large areas of the image, resulting in visible and often objectionable image quality degradations. The first step in the design of an efficient image transmission system is selecting an efficient source coder to reduce the amount of bits required for representing the image, as much as possible. JPEG2000 is the state-of-the-art image compression standard and is based on Taubman s Embedded Block Coding with Optimized Truncation (EBCOT) algorithm [23]. JPEG2000 provides high compression efficiency and has unique features such as scalability and error resilience, which will be described later in this thesis. These features make JPEG2000 a very appealing candidate for wireless multimedia applications. Moreover, due to its high scalability, JPEG2000 makes it possible for the service provider to provide its users with different levels of Quality of Service (QoS). In order to have acceptable image quality at the receiver, we need to optimize image transmission taking into account both the characteristics of the encoded image and also
CHAPTER 1. INTRODUCTION 3 the wireless channel. To describe the characteristics of the channel, we need to use an appropriate channel model which can represent the wireless channel behavior accurately enough to simulate the real world channels. In wireless mobile communications, one of the most important and widely used channel models is the flat-fading Rayleigh channel as explained in chapter 2. In this thesis, we focus on the transmission of JPEG2000 images over error-prone wireless channels. Several methods have been proposed in the literature to protect the scalable bitstreams from errors induced in the wireless channel. In the following section, these methods are reviewed. 1.2 Literature Review As stated above, JPEG2000 encoded bitstreams are scalable, meaning that there is a sequential dependency among the compressed bits. In other words, correct decoding of the bits that come first in particular sections of the bitstream called codeblocks is crucial in proper decoding of the bits which appear later in the codeblock (the structure of the JPEG2000 bitstream and the definition of the codeblock will be discussed in the next chapter). This property calls for Unequal Error Protection (UEP) of the bitstream, i.e., the beginning part of the bitstream needs the highest protection, the middle part needs less protection and the last part needs very little or even no protection. Unequal Error Protection (UEP) is one of the techniques appropriate for protecting scalable bitstreams against bit corruption. The idea behind UEP is driven by the fact that different portions of a scalable bitstream have different impact on the quality of the reconstructed image. The UEP scheme exploits the hierarchical structure of the scalable coded bitstream and assigns higher protection to the more important parts. Several UEP techniques have been proposed in literature for efficient transmission of
CHAPTER 1. INTRODUCTION 4 scalable bitstreams over wireless channels, which are based on assigning variable Forward Error Correction (FEC) coding (channel coding) to different portions of the bitstream according to their importance. In FEC, additional bits are added to the source bits to add redundancy to the bitstream. The redundant bits will help the receiver to recover the erroneous source bits. To protect the source bits unequally, different number of FEC bits are added to different portions of the source bitstream, providing higher protection for the more important bits and less protection for the bits with lower importance. In [19], a product channel code consisting of concatenated Rate Compatible Punctured Convolutional Codes/Cyclic Redundancy Check codes (RCPC/CRC) as the row code, and Reed-Solomon codes as the column code, is proposed for the Equal Error Protection (EEP) of progressive wavelet-based coded images transmitted over wireless and memoryless channels. In EEP, the same protection level is used for all parts of the source bitstream. However, the authors also present a UEP product code setting which gives the best simulation results in terms of the average reconstructed image quality. In [21], the method proposed in [19] is improved by reorganizing the source code into a set of independently decodable packets, which makes the system more robust under varying channel conditions. The RCPC code rate and the RS protection of the system presented in [19] are selected ad-hoc and without any optimization, while in [21], the authors propose a fast algorithm which finds the optimal EEP solution. In [25], a product coding scheme, including Reed-Solomon and Turbo codes, has been introduced for the transmission of JPEG2000 bitstreams over wireless channels. This scheme provides unequal protection against both errors and erasures. The method presented in [25] outperforms the methods of [19] and [21] in terms of the decoded image quality. All of the methods mentioned above have high computational complexity both at the transmitter and the receiver side. At the transmitter, the optimal selection of channel codes is very time consuming and also need to be performed for each channel condition separately.
CHAPTER 1. INTRODUCTION 5 These schemes try to find a sub-optimal but fast solution instead. At the receiver side, they need to use iterative decoding algorithms such as list-viterbi algorithm which has a high computational complexity. Furthermore, the correct decoding of each row code can be delayed in the cases of row decoding failures. In these cases, the decoder needs to receive the corresponding RS column code to be able to correct the error. The dependency of the decoder to the reception of the column RS code will introduce some delay and therefore rapid improvement in progressive image quality can not be achieved. Alternatively, UEP can also be achieved by means of Unequal Power Allocation (UPA), i.e., using different transmission powers over the bitstream ([11][17]). In [11], a Hybrid FEC-UPA algorithm is proposed for transmission of JPEG2000 digital cinema streams over wireless channels. The bitstream is partitioned into several packets, which are protected by light FEC and transmitted through the channel at different powers. In this scheme, the author assumes that each packet can be decoded only if all the previous packets have been correctly received, and the distortion is calculated based on this assumption. However, because of the independent decoding of codeblocks in JPEG2000 decoder, it is in fact possible to refine this assumption by calculating the distortion at the coding pass level, as will be explained later in this thesis. All of the above methods employ either error correction coding only, or a combination of coding and UPA, to enhance the quality of the received image. They use code rates in the order of 0.3, i.e., 70% added redundancy. In [17], however, a UPA algorithm for transmission of JPEG images over a MIMO system is proposed without employing any FEC. The JPEG compressed image is divided into different quality layers and different layers are transmitted with unequal transmit power. The results are compared to different equal power allocation schemes and show improvement as high as 15 db in the received image quality at low signal to noise ratios. To the best of our knowledge, no research results for the transmission of JPEG2000 images using UPA only has been published yet.
CHAPTER 1. INTRODUCTION 6 1.3 Thesis Contribution In this thesis, we focus on the transmission of JPEG2000 bitstreams over wireless channels using UPA. To achieve higher bandwidth efficiency and lower system complexity, we do not use any error correction coding. Instead, we take advantage of the inherent scalable properties of the JPEG2000 bitstream along with Channel State Information (CSI) to enhance the quality of transmitted images by assigning different transmission powers over the bitstream. We validate the use of UPA using two different scenarios. In the first part of this thesis, we consider multicasting of JPEG2000 images over Multi- Input Multi-Output (MIMO) systems [4]. We present an antenna selection algorithm which efficiently transmits JPEG2000 images over a multiuser MIMO network, while making use of both spatial multiplexing and diversity benefits of MIMO systems. Due to the quality scalability of the JPEG2000 bitstreams, various parts of the encoded bitstream differ in importance, and it is thus crucial to assign the best channels in terms of Signal to Interference and Noise Ratio (SINR) to the most important parts in order to minimize distortion at the receiver. The challenge of antenna selection in a multicast MIMO system is that different users impose different transmitting antenna orders, and the best antenna selection strategy for one user may be the worst for another user. Our proposed algorithm assigns substreams to transmitting antennas with different powers to reduce the distortion of the received images for different users. To ensure that all users receive the important parts of the data through their best channels, the algorithm uses diversity and sends the important parts of the bitstream through multiple channels. Hence, the chance that users accurately receive the important parts of the bitstream increases. Also, we may use this extra information at the JPEG2000 decoder by jointly decoding multiple copies of the same codestream. It is shown that applying antenna selection simultaneously for all the users, decreases the maximum distortion in the received images. The results of this research are published in [26].
CHAPTER 1. INTRODUCTION 7 In the second part of this thesis, we consider the transmission of JPEG2000 bitstreams over point-to-point wireless channels using UPA. Here, the power allocation over the bitstream is performed in a finer scale, i.e., the coding pass level. The image is encoded in the error-resilient mode of the standard when the RESTART/ERTERM option is enabled. In the first stage, the structural information of the JPEG2000 encoded image is retrieved. This information includes the number of codeblocks in the bitstream, the number and lengths of coding passes within each codeblock and the impact of each coding pass in the decoded image quality. A distortion model is developed to evaluate the distortion of the reconstructed image in terms of the power assigned to each coding pass within the bitstream. Based on this model, the UPA algorithm assigns the appropriate amount of power at coding pass level using the simulated annealing optimization. In the next step, the algorithm takes advantage of the CSI to improve the quality of the received image further by removing the effect of fading. The low complexity of our system allows us to find the optimum UPA levels for different channel conditions. Simulation results show up to 7 db improvement in reconstructed image quality compared to transmission with Equal Power Allocation (EPA), and up to 4 db improvement compared to the method presented in [25]. The results of this research are published in [27]. 1.4 Thesis Outline The thesis is organized as follows. In chapter 2, we provide the background knowledge which is required in understanding the ideas presented in this thesis. In section 2.1, the characteristics and the model of the mobile radio channel is discussed. Section 2.2, provides an overview of the structure of the JPEG2000 standard and discusses its major features. In chapter 3, we investigate an introductory unequal power allocation strategy for multicasting JPEG2000 bitstreams over MIMO systems. Section 3.1 includes an introduction to
CHAPTER 1. INTRODUCTION 8 the research. In section 3.2, descriptions of different parts of the system are provided. Section 3.3, describes the proposed algorithms. Simulation results are presented in section 3.4, and section 3.5 concludes this chapter. In chapter 4, an unequal power allocation scheme is proposed for the transmission of JPEG2000 images over point-to-point wireless channels. The scheme assigns unequal power over the JPEG2000 bitstream in a finer scale compared to the method of chapter 3. Also, the the transmission powers are selected from continues power levels. The proposed system is described in section 4.2, followed by the proposed optimized UPA algorithm in section 4.3. In section 4.4, we present the implementation of our algorithm and section 4.5 provides the simulation results. Finally, section 4.6, concludes the chapter. Chapter 5 gives a summary of the thesis contributions along with directions for future work.
Chapter 2 Background In order to have reasonable image quality at the receiver, we have to optimize image transmission over wireless channels taking into account the characteristics of both the coded image and the wireless channel. This chapter provides the background knowledge which is required in understanding the characteristics of the wireless channel and the JPEG2000 coded image data. 2.1 Mobile Radio Channel Model A precise knowledge of mobile radio channels is indispensable for the development, evaluation and testing of multimedia transmission systems. In order to have an efficient transmission of multimedia content, nearly all relevant components of the multimedia transmission system need to take the channel characteristics into account. Wireless channels suffer from undesirable phenomena such as interference and fading, which can distort the data transmission. This section deals with a brief description of mobile fading channels. It provides a fundamental understanding of the major issues in mobile fading channels and helps to understand the assumption of slowly flat fading channel that is being considered in this 9
CHAPTER 2. BACKGROUND 10 thesis. 2.1.1 Multipath Fading Channels The term signal fading refers to the rapid fluctuations in the amplitude of a received radio signal over a short period of time or travel distance. A fading channel is a communication channel in which the transmitted signal experiences fading. Fading is caused when two or more versions of the transmitted signal, called multipath signals, arrive at the receiver at slightly different times and with different amplitude attenuations. These multiple replicas of the signal can therefore, interfere constructively or destructively. In wireless communications, multipath propagation occurs due to reflections of the transmitted signal from the ground and surrounding structures existing between the transmitter and the receiver. If the pulse, s(t), is transmitted over a multipath channel, the received signal might appear as the sum of pulses with different delays and amplitudes. The transmitted signal can generally be represented as: [ ] s(t) = Re s l (t)e j2πf ct, (2.1) where s l (t) is the equivalent low-pass signal and f c is the carrier frequency. Therefore, the received signal in a multipath channel is expressed as: x(t) = n α n (t)s[t τ n (t)], (2.2) where α n (t) and τ n (t) are the attenuation factor and the propagation delay associated with the nth path of the multipath channel, respectively [14]. The propagation delays and the attenuation factors associated with each path in a multipath fading channel of a mobile system are often time-variant as a result of changes in the physical characteristics of the channel. On the other hand, due to the time variations in the structure of the channel, different transmissions of the same pulse can result in different
CHAPTER 2. BACKGROUND 11 received pulse trains. Therefore, the channel is said to have a time-varying impulse response: h(τ; t) = α(τ; t)e j2πfcτ, (2.3) where h(τ; t) represents the response of the channel at time t to an impulse applied at time t τ [14]. Because these time variations in the impulse response of the channel appear to be unpredictable for the user of the channel, they can only be characterized statistically. When there is a large number of paths, the central limit theorem can be applied to model the time-variant impulse response of the channel, h(τ; t), as a complex-valued Gaussian random process. When there is no line of sight between the transmitter and the receiver, the impulse response is modeled as a zero-mean Gaussian random process. Thus, h(τ; t) will be Rayleighdistributed and the channel is called a Rayleigh fading channel [14]. If a line of sight exists between the transmitter and the receiver, the Gaussian process is no longer zero-mean. Therefore, the envelope h(τ; t) will have a Rician distribution and the channel is said to be a Ricean fading channel [14]. 2.1.2 Parameters of the Fading Channel This section provides a brief overview of the parameters which specify the characteristics of fading channels. These parameters are used to compare different multipath channels and to develop some general design guidelines for wireless systems. Understanding these parameters will help in specifying the type of the fading channel, whether it is fast or slow, frequency-selective or frequency-nonselective (flat). Delay Spread and Coherence Bandwidth The transmitted signal in a multipath channel will arrive at the receiver through different propagation paths. Each received signal arrives at the receiver with a different delay. The
CHAPTER 2. BACKGROUND 12 mean excess delay and rms delay spread are multipath channel parameters that can be determined in terms of delays and attenuations associated with each path and are used in specifying the characteristics of the multipath channel. The mean excess delay, τ, is defind to be [16]: αkτ 2 k k τ = αk 2, (2.4) k where α k and τ k are the attenuation factor and the propagation delay associated with the kth path of the multipath channel, respectively. The rms delay spread, σ τ, is defined to be: σ τ = τ 2 (τ) 2, (2.5) where αkτ 2 k 2 τ 2 k =. (2.6) Analogous to the delay spread parameters in the time domain, coherence bandwidth, B C, is used to characterize the channel in the frequency domain. Coherence bandwidth indicates the range of frequencies over which the channel passes all spectral components of the transmitted signal with approximately equal gain and linear phase. Therefore, two sinusoids with frequency separation greater than the coherence bandwidth, are affected with different gains and different phase changes. The rms delay spread and coherence bandwidth of the channel are inversely proportional to one another, although their exact relationship is a function of the exact multipath structure. B C 1. (2.7) σ τ Measurements show that the typical value of the rms delay spread is in the order of microseconds for outdoor mobile radio channels and in the order of nanoseconds for indoor k α 2 k
CHAPTER 2. BACKGROUND 13 Table 2.1: Typical measured values of RMS delay spread in different environments [16] Environment Frequency (MHz) RMS Delay Spread (σ τ ) Notes Urban 1 910 1300 ns avg. New York City Urban 2 892 10-25 µs Worst case San Francisco Suburban 1 910 200-310 ns Averaged typical case Suburban 2 910 1960-2110 ns Averaged extreme case Indoor 1 1500 10-50 ns Office building Indoor 2 850 270 ns max Office building Indoor 3 1900 70-94 ns avg. Three San Francisco buildings radio channels. Typical values for measured rms delay spread in different environments are shown in Table 2.1 [16]. If the coherence bandwidth is defined as the bandwidth over which the frequency correlation function is above 0.9, then the coherence bandwidth is approximately [16]: B C 1 5σ τ. (2.8) If the definition is modified for the frequency correlation function to be above 0.5, then the coherence bandwidth is approximately: B C 1 50σ τ. (2.9) In order to give an idea of the range of values of the coherence bandwidth for different environments, the coherence bandwidth is calculated for the RMS delay spread values of Table 2.1 and the results are reported in Table 2.2. Doppler Spread and Coherence Time Delay spread and coherence bandwidth do not offer any information about the time-varying nature of the channel caused by either changes in the physical structure of the channel or
CHAPTER 2. BACKGROUND 14 Table 2.2: Typical calculated values of coherence bandwidth in different environments based on the values given in Table 2.1. σ τ (ns) B C = 1 5σ τ (KHz) B C = 1 50σ τ Urban 1 1300 154 15.4 Urban 2 15000 13.3 1.33 Suburban 1 250 800 80 Suburban 2 2000 100 10 Indoor 1 25 8000 800 Indoor 2 270 741 74.1 Indoor 3 80 2500 250 (KHz) the relative movement of the transmitter and the receiver (e.g. movement of the mobile device). Doppler spread and coherence time are the parameters which describe the time varying nature of the channel. If the receiver in a communication system moves with respect to the transmitter, the frequency of the received signal may not be equal to the frequency of the transmitted signal. The change in the frequency of the received signal due to the movement of the receiver is called the Doppler effect. Because of the Doppler effect, if a pure sinusoidal tone of frequency f c is transmitted, the received signal spectrum will have components in the range, f c f d to f c + f d, where f d is called the Doppler shift. Doppler shift is a function of the relative velocity of the receiver with respect to the transmitter, v, the angle between the direction of motion and the direction of the arrival of the scattered waves, θ, and the wavelength of the wave,λ [16]: f d = v. cos θ. (2.10) λ The Doppler spread, B D, is defined as the maximum Doppler shift given by: B D = v/λ. (2.11) Coherence time, T C, is the time domain dual of the Doppler spread. The Doppler spread
CHAPTER 2. BACKGROUND 15 and coherence time are inversely proportional to one another [14]: T C 1 B D. (2.12) A slowly changing channel has a large coherence time, or equivalently a small Doppler spread. Coherence time quantifies the similarity of the channel response at different times. In other words, coherence time is the time duration over which the amplitudes of the two received signals are highly correlated. The definition of coherence time implies that two signals arriving with a time separation greater than T C are affected differently by the channel. If the time duration of the baseband signal is greater than the coherence time of the channel, the channel will change during the transmission of the baseband message, thus causing distortion at the receiver. 2.1.3 Types of Fading There are different types of fading experienced by a signal propagating through a mobile radio channel. The type of fading depends on the relation between the parameters of the transmitted signal (such as bandwidth, symbol duration, etc), and the channel parameters (such as delay spread and Doppler spread). Figure 2.1 summarizes the four different types of fading as will be explained shortly. In this figure, B S indicates the signal bandwidth and T S is the symbol duration. Frequency-nonselective vs. Frequency-selective Fading If the coherence bandwidth of the channel, B C, is small in comparison to the bandwidth of the transmitted signal, the channel is said to be frequency-selective. On the other hand, if the coherence bandwidth of the channel B C is large in comparison to the bandwidth of the transmitted signal, the channel is said to be frequency-nonselective or frequency-flat. In a frequency-flat channel, all frequency components of the signal are affected in almost the same manner. According to the values of coherence bandwidth reported in Table 2.2,
CHAPTER 2. BACKGROUND 16 Figure 2.1: Fading Types it can be seen that for indoor environments, the coherence bandwidth is larger than the suburban or urban environments. Therefore, since the transmitted signal bandwidth should be less than the coherence bandwidth for a flat-fading channel, for indoor environments the channel is more likely to be considered flat. In suburban and urban environments, for the same signal bandwidth, because of the smaller coherence bandwidth of the channel, the channel is more likely to be modeled as frequency-selective fading. However, if we consider the signal bandwidth of today s wireless standards, such as GSM, IEEE 802.11b or IEEE 802.11n, which is 200KHz for GSM and in the order of MHz for IEEE 802.11 standards, it is inferred that in almost all the cases the channel should be modeled as a frequency selective channel. Slow vs. Fast Fading If the coherence time of the channel, T C, is small in comparison with the signal duration, then the channel is likely to change during the transmission of the baseband signal and is said to be fast fading. This will cause the received signal at the receiver to be distorted.
CHAPTER 2. BACKGROUND 17 On the other hand, if the coherence time of the channel T C is large in comparison with the signal period, the channel is said to be slow fading. In slow fading channels, the transmitter is able to track the channel variations by the feedback sent by the receiver. The feedback about the channel condition can be taken into account at the receiver to enhance the unequal power allocation algorithm. As discussed in the above paragraph, in today s wireless standards the bandwidth of the transmitted signal is very high, and therefore the systems are most likely to be considered as slow fading. For example, if a mobile user is moving at a speed of 5 km/h, and the transmission is at the frequency of 900 MHz, the Doppler spread of the channel will be approximately 4.17 Hz, which is much lower than the bandwidth of the transmitted signals. In this thesis, we assume the wireless channel to be a slow flat fading channel. This assumption has been adopted in many publications, such as [8], [22], and [25]. 2.2 The JPEG2000 Still Image Compression Standard With the continual expansion of multimedia and Internet applications, the JPEG2000 still image compression standard was developed to fulfill the advanced requirements of today s diverse applications. The JPEG2000 standard has superior performance compared to previous standards in terms of reconstructed image quality. It also incorporates many interesting features which satisfy the needs of today s digital imagery applications and addresses areas and applications where previous standards such as JPEG fail to produce the best quality or performance [20]. The applications in which JPEG2000 performs better include Internet transmission, printing, scanning, digital photography, remote sensing, mobile communications, medical imagery, digital libraries/archives, and E-commerce [20]. This section provides a brief overview of the JPEG2000 structure and discusses major features of the standard.
CHAPTER 2. BACKGROUND 18 2.2.1 Overview The block diagram in Figure 2.2 illustrates the JPEG2000 codec structure. To encode the raw image, the encoder first decomposes the image into its color components (i.e., the components included in the color model, e.g., RGB). An optional color transformation may be applied to each color component. JPEG2000 supports both Irreversible Component Transformation (ICT) for lossy coding, and Reversible Component Transformation for lossless or lossy coding. Component transformations improve compression efficiency by decorrelating the color components and converting the color space to a reasonable color space for quantization with respect to the human visual system. The image components are then (optionally) divided into rectangular and non-overlapping tiles. The tile-component is the basic unit of the original (or reconstructed) image, and is compressed (or reconstructed) independently, as an entirely distinct image. Figure 2.2: The general block diagram of JPEG2000 codec structure Each tile-component goes through the forward Discrete Wavelet Transform (DWT). DWT decomposes the tile into its frequency components by passing it horizontally and vertically through lowpass and highpass filters. The four possible combinations of lowpass (L), and highpass (H) filtering decompose the tile into four subbands, namely LL 1, LH 1, HL 1, and HH 1 as depicted in Figure 2.3.
CHAPTER 2. BACKGROUND 19 1-Level DWT Figure 2.3: One-level decomposition of Lena image (512 512 pixels) by applying Discrete Wavelet Transform LL 1 subband is a down-sampled low-resolution representation of the original tile component. Other subbands contain residual information for the tile; hence, to reconstruct the tile with finer details, we need to have the subbands LH 1, HL 1, and HH 1 in addition to LL 1. The LL 1 subband is the lowest resolution of the tile and is called resolution level R 0. Subbands LH 1, HL 1, and HH 1 constitute the next resolution level, R 1. DWT can be applied further to resolution level R 0. For instance, if DWT is applied twice, the subband LL 1 will be decomposed further into four subbands, LL 2, LH 2, HL 2, and HH 2 as depicted in Figure 2.4. In this case, the lowest resolution of the tile is LL 2 and there are three resolution levels as illustrated in this figure. Each resolution of each tile component is then divided into rectangular-shaped precincts as shown in Figure 2.5. Within each subband, precincts are further sub-divided into squareshaped codeblocks (Figure 2.5). It should be noted that the precinct partition does not affect the transformation or encoding of the data. Instead, the precinct s role is in organizing the compressed data into the final code stream [24]. Each codeblock is encoded independently, and hence decoded independently as well. This
CHAPTER 2. BACKGROUND 20 Resolution Level LL 2 HL 2 R 0 Resolution Level R 1 Resolution Level HL 1 LH 2 HH 2 R 2 LH 1 HH 1 Figure 2.4: Two-level decomposition of Lena image (512 512 pixels) by applying Discrete Wavelet Transform. The image is decomposed into three resolution levels, R 0, R 1, and R 2 feature of JPEG2000 standard confines the error propagation to the code block boundaries. In other words, the unsuccessful decoding of a codeblock will not affect the decoding of the following codeblocks. Bits of equal significance across all the coefficients in a code block are referred to as a bit plane. Each bitplane is then entropy coded by an arithmetic encoder within three coding passes, namely the significance propagation, the magnitude refinement, and the cleanup pass (The order in which the bit planes are coded is from the most significant to the least significant). Each of these coding passes collects contextual information about the bitplane data. The arithmetic coder uses this contextual information along with its internal state and generates a compressed bitstream. In this way, a separate bitstream is generated for each code block. The compressed bit streams of code blocks within a precinct comprise the body of a packet. A collection of packets, one from each precinct of each resolution level, comprise the layer. A packet could
CHAPTER 2. BACKGROUND 21 Figure 2.5: Structure of Precincts and Codeblocks be interpreted as one quality increment for one resolution level at one spatial location, since precincts correspond roughly to spatial locations. Similarly, a layer could be interpreted as one quality increment for the entire full resolution image. Each layer successively and monotonically improves the image quality, so that the decoder is able to decode the code block contributions contained in each layer in sequence. Layers are ordered successively to generate the final bitstream. The final generated code stream has a main header at the beginning that describes the original image and the various settings that are used in the encoding process. This information enables the decoder to reconstruct the image with the desired characteristics. 2.2.2 Features of the JPEG2000 Standard The JPEG2000 standard supports many interesting features that are unique to this standard and therefore distinguish this standard from the previous image compression standards. A brief overview of some of the major features relevant to the material of this thesis is presented in this section.
CHAPTER 2. BACKGROUND 22 Scalability One of the most attractive features of the JPEG2000 standard is that it supports scalability. In scalable coding of still images, the image is encoded once, achieving coding of more than one quality and/or resolution simultaneously. In the decoding stage, simple truncation of the bitstream at specific points results in obtaining lower qualities or resolutions of the original image. In other words, by decoding an appropriate portion of the scalable bitstream, a complete picture can be generated. The quality and/or resolution of the generated picture is proportionate with the length of the bitstream decoded. A key advantage of scalable compression is that the target bit rate or reconstruction resolution need not be known at the time of compression. A related advantage of practical importance is that the image need not be compressed multiple times to achieve a target bit rate, as is common with the JPEG compression standard. While low performance decoders may decode only small portions of the bitstream producing basic quality, high performance decoders may decode much more and produce significantly higher quality. An additional advantage of scalability which is employed in this thesis is its ability to provide resilience to transmission errors by providing different layers with different importance. In the transmission time, the most important data of the lower layer can be sent over the channel with more protection, while the less critical parts of the data can be sent with less protection. Scalability feature provided by JPEG2000 standard has many applications in today s digital imagery industry. Efficient progressive transmission of still images in web browsing applications, flexibility in image storage based on the available system storage and the desired quality, adaptation to bandwidth or power requirements of the image transmission system are examples of scalable image coding applications.
CHAPTER 2. BACKGROUND 23 There are four types of scalability or progression in JPEG2000 bitstream, namely resolution, SNR (or quality), spatial location, and component (color). The term progression determines the ordering of elements within a scalable bitstream. Different types of progression are achieved by the appropriate ordering of the packets within the bitstream. The most important types of scalability are SNR scalability and resolution scalability. These types of scalability are very important for web and database access applications and bandwidth scaling for robust delivery [2]. Error Resilience Error resilience is one of the most desirable properties of JPEG2000 in mobile and Internet applications. JPEG2000 uses an arithmetic coder, which is a variable-length coder, to compress the quantized wavelet coefficients. Variable-length coding is known to be prone to channel or transmission errors. A bit error results in loss of synchronization at the entropy decoder and the reconstructed image can be severely damaged. To improve the performance of transmitting compressed images over error prone channels, error resilience tools are included in the standard. The error resilience approaches used are: data partitioning and synchronization, error detection and concealment, and quality of service (QoS) transmission based on priority. Here, data partitioning and synchronization approach is utilized to provide error resilience. As mentioned before, entropy coding of the quantized coefficients is performed within code block boundaries. Since encoding and decoding of the code blocks are independent processes, bit errors in the bitstream of a code block will be restricted within that code block. Although restriction of errors to codeblock boundaries reduces the error propagation to a large extent, a corrupted codeblock still may lead to objectionable artifacts in the decoded image. To enhance error resilience further, other mechanisms can also be utilized. Different termination mechanisms allow different levels of independent extraction of the coding pass data. Termination of the arithmetic coder is allowed after every coding pass
CHAPTER 2. BACKGROUND 24 and the contexts may be reset after each coding pass. This allows the arithmetic decoder to continue the decoding process even if an error has occurred. Also, synchronization markers can be added to the bit stream to produce error resilience features in the bitstream. In this thesis, we employ the ERTERM/RESTART option of the JPEG2000 standard. By enabling this option a separate predictably terminated codeword segment is created for each coding pass. Any error in the bitstream is likely to leave the decoder in a state which is inconsistent with the predictable termination policy. An error resilient decoder may detect this condition at the end of the coding pass in which the error occurred. In this way, the decoder discards only the corrupted coding passes [24].
Chapter 3 Multicasting JPEG2000 Images over MIMO Systems 3.1 Introduction Real-time transmission of image and video content over wireless channels is becoming very common in today s wireless networks. Much research is done aiming at improving the quality of the received image or video and providing a robust transmission system. Fundamental physical challenges such as channel fading and interference, however, have put strains on the radio resources, which makes achieving reliable wireless communication difficult. Development of Multi-Input Multi-Output (MIMO) systems has been a great achievement toward overcoming this problem. These systems provide high speed links while maintaining good quality of service. MIMO systems have the following desirable capabilities [4]: Increased bitrate if used in spatial multiplexing systems Decreased bit error rate if used in diversity mode Extended transmission range by utilizing beamforming methods 25
CHAPTER 3. MULTICASTING JPEG2000 IMAGES OVER MIMO SYSTEMS 26 In this chapter, we consider multicasting of JPEG2000 images over MIMO systems. JPEG2000 is the state-of-the-art image compression standard that outperforms other standards such as JPEG [20]. This standard generates a progressive bitstream with different scalable progressions, excellent error resilience features, and region of interest processing. The property of JPEG2000 focused upon in this chapter is its quality scalability, which means as more elements of the bitstream are received, the quality of the decoded image increases accordingly [20], [24]. The aim of this chapter is to investigate the effect of applying unequal power allocation for multicasting JPEG2000 images over MIMO channels. Here, we propose an antenna selection algorithm, which assigns the channels with higher Signal to Interference and Noise Ratio (SINR) to the more important parts of the bitstream and also assigns unequal transmitting powers to different antennas. The unequal power assignment strategy employed in this chapter is done in a coarse form, meaning that we simply divide the bitstream to parts with equal length and transmit each part with a power chosen from a set of determined powers and from a determined antenna. Details of the algorithm will be elaborated in section 3.3.1. The proposed antenna selection algorithm efficiently transmits JPEG2000 images over a multiuser MIMO network, while making use of both spatial multiplexing and diversity benefits of MIMO systems. Due to the quality scalability of the JPEG2000 bitstreams, various parts of the encoded bitstream differ in importance, and it is thus crucial to assign the best channels in terms of SINR to the most important parts in order to achieve the minimum distortion at the receiver. As stated above, we consider transmission of JPEG2000 images to multiple users over MIMO channels with multiple transmitting and receiving antennas. The challenge of antenna selection in a multicast MIMO system is that different users impose different transmitting antenna orders, and the best antenna selection strategy for one user may be the
CHAPTER 3. MULTICASTING JPEG2000 IMAGES OVER MIMO SYSTEMS 27 worst for another user. Our proposed algorithm assigns substreams to transmitting antennas to reduce the distortion of the received images for different users. To ensure that all users receive the important parts of the data through their best channels, the algorithm uses diversity and sends the important parts of the bitstream through multiple channels. Hence, the chance that users accurately receive the important parts of the bitstream increases. Also, we may use this extra information at the JPEG2000 decoder by jointly decoding the multiple copies of the same codestream. It is shown that applying antenna selection simultaneously for all the users, decreases the maximum distortion in the received images. Furthermore, wireless channels are subject to signal degradations such as noise, interference and fading and due to the nature of JPEG2000 coded bitstream, without adequate data protection, any transmission errors that occur in the coded image will be propagated to affect large image areas, causing visible and often objectionable, image quality deterioration. The JPEG2000 standard addresses the transmission error problem by including provisions for error resilience tools. Here we use the error resilience features of JPEG2000 to improve the quality of received images [24]. The organization of this chapter is as follows. In section 3.2, the system model is presented. In section 3.3, the proposed method is explained, and section 3.4 provides simulation results and discussion. Finally, we conclude this chapter in section 3.5. 3.2 System Description The system is modeled as a two-user MIMO system, in which both users have the same requirements for the quality of the received image and have the same priority for the base station. The transmitter has N T and each receiver has N R antennas. The general model of the system is shown in Figure 3.1 with two users and N T = N R = 4 antennas. It should be noted that although we consider two users in our proposed algorithm, the algorithm can be generalized for the case of more than two users by few modifications.
CHAPTER 3. MULTICASTING JPEG2000 IMAGES OVER MIMO SYSTEMS 28 Figure 3.1: System model for multicasting JPEG2000 images over MIMO channels However, the improvement in the received image quality may not be as good as the case of two users. Also, the two user scenario can be considered in cases where the users are clustered into two multicast groups, where each group includes several users with similar channel conditions but different from the other group. Users will subscribe to either of the multicast groups based on their channel conditions and each group can be considered as a single user [3]. Different components of the system are described below. 3.2.1 Channel Model We assume that the channels between the base station and different users have the same statistics. Each of these channels is a N T by N R MIMO channel and is assumed to be Rayleigh flat fading. Let H denote the N R N T channel matrix. The channel matrix entries, h ij (i = 1... N R, j = 1... N T ), are i.i.d. Gaussian complex random variables (Figure 3.2), with independent real and imaginary parts, each with zero mean and variance 1/2. We assume that the channel is unknown at the transmitter but is known at the receiver. A low bandwidth, error-free feedback channel provides limited channel information from the
CHAPTER 3. MULTICASTING JPEG2000 IMAGES OVER MIMO SYSTEMS 29 H = h h N R h h M 11 M L M 12 L 1 Lh 1 N N R T N T h 11 h NR N T Figure 3.2: N T by N R MIMO channel matrix receiver to the transmitter. The noise is assumed to be Additive White Gaussian Noise (AWGN) with unit variance. We use 4-QAM (Quadrature Amplitude Modulation) for modulating the bitstream, although extension to higher order modulation is readily possible. The channel is slowly time varying and is assumed to be constant over every T symbol intervals. 3.2.2 Transmitter JPEG2000 Encoder: As shown in Figure 3.1, the transmitter consists of two parts. The first part is the source encoder which converts the input images into compressed bitstreams. The input images to the system are encoded using JPEG2000, which is the state-of-the-art still image compression standard, in its quality progressive mode. As explained in Chapter 2, to encode the raw image, JPEG2000 first divides it into disjoint rectangular tiles. The subband/wavelet transform is applied to each tile-component to generate subbands, which are then divided into rectangular-shaped precincts, and further divided into square-shaped
CHAPTER 3. MULTICASTING JPEG2000 IMAGES OVER MIMO SYSTEMS 30 codeblocks. Each bitplane of a codeblock is then encoded by an arithmetic encoder in three coding passes. This provides a progressive bitstream for each of the codeblocks. Coding passes are then interleaved to create the scalable JPEG2000 bitstream. We enable the error resilient feature of the JPEG2000 encoder by using the RESTART/ERTERM option. We have used Kakadu software as our JPEG2000 codec with 64 64 codeblocks and 128 128 precincts [24]. In all simulations, the header information throughout the bitstream is separated and assumed to be transmitted error free. At the receiver, headers are re-inserted at their original location before JPEG2000 decoding. Channel assignment unit: The second part of the transmitter is the channel assignment unit. In this unit, after the removal of headers, the raw bitstream is divided into two equal length substreams, SS 1 and SS 2. SS 1 is the first part of the bitstream and SS 2 is the second part of the bitstream. Each substream is divided into non-overlapping blocks of lengths 2T bits, where T is the number of symbols for which we assume the channel to be constant, and the number of bits per symbol for 4-QAM modulation is two. Our proposed channel assignment algorithm then runs on each of these blocks independently. Figure 3.3 illustrates the decomposition of the bitstream into non-overlapping blocks of length 2T bits. Figure 3.3: Division of the raw bitstream into blocks of length 2T. The channel assignment unit runs on each of these blocks independently
CHAPTER 3. MULTICASTING JPEG2000 IMAGES OVER MIMO SYSTEMS 31 The total transmit power from all the antennas during each symbol period is kept constant. In effect, the channel assignment unit determines the substreams that are transmitted through each antenna and the required power for transmitting them. It also determines how the final bitstream should be composed from the substreams that each user receives, and sends this side information to the users. This part is explained in more detail in section 3.3.1. 3.2.3 Receiver We have used a Minimum Mean Square Error (MMSE) receiver to eliminate the problem faced by MIMO receivers which is the presence of multi-stream interference. Each received signal in a receiving antenna consists of a combination of all the transmitted signals. MMSE is a linear receiver, meaning that it first separates the transmitted substreams by multiplying the received signal vector, ȳ, by MMSE matrix filter, G MMSE, and then decodes each substream independently as illustrated in Figure 3.4 [13]. Let s denote the transmitted signal vector. If the transmitting antennas have different powers, p 1, p 2,..., p NT then the received signal will be: ȳ = H p1 0... 0 0 p2... 0......... 0 0 0 0 pnt s + n. (3.1) Let P denote the power matrix, therefore the received signal vector, ȳ, is: ȳ = HP s + n, (3.2) where n is the noise vector. The transmitted signal vector is then estimated by: ŝ = GMMSE.ȳ, (3.3)
CHAPTER 3. MULTICASTING JPEG2000 IMAGES OVER MIMO SYSTEMS 32 H Figure 3.4: Structure of the MMSE Receiver for separating the transmitted data bitstreams over a MIMO channel where: G MMSE = [P H H H HP + N 0 I NT ] 1 P H H H, (3.4) N 0 is the noise power, I NT the Hermitian operator. is the identity matrix of size N T, and the superscript H denotes The estimated substreams, ŝ, are then passed to the modified JPEG2000 decoder which will be discussed in section 3.3.2. 3.3 Proposed Method In this section, the proposed channel assignment algorithm and modified JPEG2000 decoding scheme are discussed. 3.3.1 Channel Assignment Algorithm The channel assignment algorithm first calculates the post processing SINR for all the receiving antennas of each user every T symbols [6]. Because we have used MMSE receiver, each transmitted substream will be decoded by its corresponding receiving antenna. The calculated SINRs determine the best transmitting antennas in terms of post processing SINR for each user. This constitutes the antenna selection order of each user. Based on the antenna selection order requested by each user, the algorithm assigns the
CHAPTER 3. MULTICASTING JPEG2000 IMAGES OVER MIMO SYSTEMS 33 antennas to substreams aiming at reducing the average distortion among all users. The algorithm is based upon the progressive nature of the JPEG2000 coded bitstream, i.e., the first substream, SS 1, needs more protection and should be transmitted through channels with higher SINR or equivalently lower Bit Error Rates (BERs). The challenge of antenna selection when transmitting images to multiple users will arise when different users request different transmitting antenna orders, and the best antenna selection order for one user may result in great quality degradation for another user. Assumes that the total transmit power at each symbol period is 4p. According to the antenna selection orders requested by each user, the algorithm decides on one of the following scenarios: Choose two best antennas and send one copy of each substream with power 2p from each of the selected antennas. The other two antennas are not used for transmission. Send two copies of each substream transmitting from all four antennas each with power p. Send one of the substreams with power 2p from one antenna and two copies of the other substream from two other antennas each with power p. The fourth antenna is not used for transmission. It should be noted that in cases where we send two copies of the same substream from two transmitting antennas, the MMSE receiver will decode each copy separately and each user will receive two copies of that substream, obviously with different BERs. We call the copy transmitted over the channels with higher and lower SINR, the main and the secondary copy, respectively. The transmitter also lets each user know which substream has been transmitted from each antenna. The channel assignment algorithm is summarized in Algorithm 1, where A ij refers to the j th best antenna of user i, i = 1, 2, j = 1,..., 4. The total transmit power at each
CHAPTER 3. MULTICASTING JPEG2000 IMAGES OVER MIMO SYSTEMS 34 symbol period is assumed to be 4p. Algorithm 1 Channel Assignment Algorithm for Every T symbols do Calculate the post processing SINR for all the receiving antennas of each user and sort the antennas based on their corresponding SINRs. if The two best antennas of each user match then Transmit SS 1 from A 11 with power 2p Transmit SS 2 from A 12 with power 2p Do not transmit from remaining antennas end if if Only the best antennas of each user match then Transmit SS 1 from A 11 with power 2p Transmit SS 2 from A 12 with power p Transmit SS 2 from A 22 with power p Do not transmit from the remaining antenna end if if The best antennas of the two users are different then Transmit SS 1 from A 11 with power p Transmit SS 1 from A 21 with power p Send SS 2 from the remaining antennas with power p end if end for 3.3.2 Modified JPEG2000 decoder In the original JPEG2000 decoder (in the ERTERM/RESTART mode) if an error is detected in a coding pass, all the remaining coding passes within that codeblock are discarded [24]. The modified JPEG2000 decoder, however, takes advantage of the diversity in the received data during the time intervals in which more than one copy of the codestream is available. This is done by jointly decoding the two different received copies of the original data. The flowchart of the algorithm is shown in Figure 3.5. First, the decoder uses the main codestream to decode the image. Once the decoder detects an error in a coding pass, the
CHAPTER 3. MULTICASTING JPEG2000 IMAGES OVER MIMO SYSTEMS 35 Figure 3.5: Modified JPEG2000 decoder decoder tries to correct the erroneous byte(s) by using the information contained in the secondary codestream. Here, we assume that it is unlikely to have bit errors at the same bit locations in both the main and the secondary copy. If the decoder is successful in correcting the damaged coding pass, it restarts its operation from the beginning of the erroneous coding pass. Otherwise, the decoder will continue decoding of the codestream from the next codeblock of the main codestream. The modified JPEG2000 decoder is summarized in Algorithm 2. 3.4 Simulation Results In this section, we provide simulation results to investigate the performance of our proposed algorithms. Images are encoded by enabling the quality progression option of the JPEG2000 encoder. Also, to take advantage of the error resilient feature of JPEG2000, we enable the ERTERM/RESTART mode at the encoder. Consequently, the decoding should be done in the error resilient mode as well. We assume that the channel is constant for T = 250 symbols to ensure that the slowly time varying condition is satisfied [17]. The algorithm is
CHAPTER 3. MULTICASTING JPEG2000 IMAGES OVER MIMO SYSTEMS 36 Algorithm 2 Modified JPEG2000 Decoder for All the coding passes in the codestream do Decode the coding pass if Decoder declares occurrence of an error in the coding pass then Find all the mismatches between the main and the secondary coding passes for each mismatch do Swap the corresponding values of the two codestreams Reinitialize the decoding from the start of the coding pass if decoding is successful then Break; else swap the values of the two codestreams again and repeat the procedure with the next mismatch end if end for end if end for tested using the 512 512 Lena, peppers and Barbara test images with coding rates of 1.0 and 0.5 bits per pixel (bpp). We compare the performance of our proposed system with two other cases. The Peak Signal to Noise Ratio (PSNR) is used as a measure of the reconstruction quality [1]. The PSNR for an image is a function of the Mean Squared Error (MSE) between the decoded image and the original image: (2 L 1) 2 P SNR = 10 log 10 MSE, (3.5) where L represents the number of bits used to encode the pixel values in the original raw image, typically 8 bits. In the first case, we do not use any of our proposed algorithms. We simply encode the image at 1.0 bpp, and the generated bitstream is divided into four equal length substreams. We transmit the four substreams according to the antenna selection order requested by User 1. In effect, we are giving complete priority to the needs of User 1 and we are applying
CHAPTER 3. MULTICASTING JPEG2000 IMAGES OVER MIMO SYSTEMS 37 40 Average PSNR (db) 35 30 25 20 User1 based U1 User1 based U2 Ave U1&U2 EPA JAAMD UPA JAAMD 15 10 0 5 10 15 20 25 30 35 40 Channel SNR (db) Figure 3.6: PSNR curves for different schemes in transmission of Lena image (512 512) no antenna selection for User 2. The PSNR curves for User 1 and User 2 are labeled with User1 based-u1 and User1 based-u2, respectively as shown in Figure 3.6 for Lena, Figure 3.7 for peppers, and Figure 3.8 for Barbara images. We also calculated the average PSNR for the two users, labeled with Ave U1&U2 in the figure. In the second case, equal power antenna assignment along with the proposed modified JPEG2000 decoder algorithm is used. Here, the encoded image is divided into two equal length substreams. These substream together with a copy of each, are transmitted from all four antennas simultaneously and with equal power. We transmit the more important substream through the channel with better quality. We call this method Equal Power Allocation-Joint Antenna Assignment and Modified JPEG2000 Decoder, (EPA-JAAMD). In this case, the image is encoded at a rate of only 0.5 bpp, to ensure the same transmission
CHAPTER 3. MULTICASTING JPEG2000 IMAGES OVER MIMO SYSTEMS 38 40 35 User1 based U1 User1 based U2 Ave. U1&U2 EPA JAAMD UPA JAAMD Average PSNR (db) 30 25 20 15 10 0 5 10 15 20 25 30 35 40 Channel SNR (db) Figure 3.7: PSNR curves for different schemes in transmission of Peppers image (512 512) time and power consumption as the previous case, for fair comparison. In the third case, both of our proposed algorithms are implemented, i.e., we use the unequal power antenna assignment algorithm in conjunction with the modified JPEG2000 decoder algorithm. Again, the image is encoded at a rate of 0.5 bpp for fair comparison. Since this method may result in transmitting with different powers (0, p, or 2p), it is called Unequal Power Allocation-JAAMD (UPA-JAAMD). As can be seen from Figures 3.6, 3.7, and 3.8, our algorithm performs significantly better in terms of PSNR in the channel Signal to Noise Ratio (SNR) range of 0 30 db. For example, in Figure 3.6 at an SNR of 15 db, the EPA-JAAMD results in 5 db improvement in the average user PSNR, compared to the case labeled as Ave U1 & U2. Furthermore, by implementing the unequal power channel assignment algorithm we gain an additional 1
CHAPTER 3. MULTICASTING JPEG2000 IMAGES OVER MIMO SYSTEMS 39 40 Average PSNR (db) 35 30 25 20 User1 based U1 User1 based U2 Ave U1&U2 EPA JAAMD UPA JAAMD 15 10 0 5 10 15 20 25 30 35 40 Channel SNR (db) Figure 3.8: PSNR curves for different schemes in transmission of Barbara image (512 512) db. Simulation results show that it is more beneficial to transmit a lower rate encoded image with higher channel protection, than a higher rate encoded image with less protection. To show the visual quality of our proposed method, results for Barbara, peppers and Lena images transmitted with different algorithms through a channel with SNR of 15 db are shown in Figures 3.9, 3.10, and 3.11. Again, noticeable visual quality enhancement is achieved through our proposed algorithms. 3.5 Conclusion In this chapter, we presented an algorithm for multicasting of JPEG2000 images over MIMO systems. According to different antenna selection orders requested by different users, the
CHAPTER 3. MULTICASTING JPEG2000 IMAGES OVER MIMO SYSTEMS 40 (a) (b) (c) (d) Figure 3.9: Barbara image results for different schemes at SNR=15 db (a) Received image by User 2 with antenna selection based on User 1, PSNR=18.41 db (b) Received image by User 1 with antenna selection based on User 1, PSNR=23.93 db (c) EPA-MD, PSNR=24.25 db for both users (d) UPA-JAAMD, PSNR=24.87 db for both users
CHAPTER 3. MULTICASTING JPEG2000 IMAGES OVER MIMO SYSTEMS 41 (a) (b) (c) (d) Figure 3.10: Peppers image results for different schemes at SNR=15 db (a) Received image by User 2 with antenna selection based on User 1, PSNR=13.69 db (b) Received image by User 1 with antenna selection based on User 1, PSNR=20.35 db (c) EPA-MD, PSNR=25.81 db for both users (d) UPA-JAAMD, PSNR=28.82 db for both users
CHAPTER 3. MULTICASTING JPEG2000 IMAGES OVER MIMO SYSTEMS 42 (a) (b) (c) (d) Figure 3.11: Lena image results for different schemes at SNR=15 db (a) Received image by User 2 with antenna selection based on User 1, PSNR=15.46 db (b) Received image by User 1 with antenna selection based on User 1, PSNR=25.96 db (c) EPA-MD, PSNR=26.69 db for both users (d) UPA-JAAMD, PSNR=27.78 db for both users