Unbalanced Quantized Multiple Description Video Transmission using Path Diversity Sila Ekmekci* , Thomas Sikora Technical University Berlin, Einsteinufer 17, Berlin, 10587 Germany ABSTRACT Multiple Description Coding is a forward error correction scheme where two or more descriptions of the source are sent to the receiver over different channels. If only one channel is received the signal can be reconstructed with distortion D 1 or D 2 . On the other hand, if both channels are received the combined information is used to achieve a lower distortion D 0 . Our approach is based on the Multiple State Video Coding with the novelty that we achieve a flexible unbalance rate of the two streams by varying the quantization step size while keeping the original frame rate constant. The total bitrate R T is fixed which is to be allocated between the two streams. If the assigned bitrates are not balanced there will be PSNR (peak signal to noise ratio) variations between neighbouring frames after reconstruction. Our goal is to find the optimal rate allocation while maximizing the average reconstructed frame PSNR and minimizing the PSNR variations given the total bitrate R T and the packet loss probabilities p 1 and p 2 over the two paths. The reconstruction algorithm is also taken into account in the optimization process. The paper will report results presenting optimal system designs for balanced (equal packet loss probabilities) but also for unbalanced path conditions (different packet loss probabilities). Keywords: multiple description coding, path diversity, rate-distortion optimization, rate allocation, unbalanced quantization, unbalanced loss probabilities, state recovery, flicker reduction. 1. INTRODUCTION The explosion in multimedia communication over the Internet has increased the conflicting requirements on high compression and high error resilience. Among the several methods investigated for error resilience, Multiple Description Coding is a forward error correction scheme where two or more descriptions of the source are sent to the receiver over different channels. If only one channel is received the signal can be reconstructed with distortion D 1 or D 2 . On the other hand, if both channels are received the combined information is used to achieve a lower distortion D 0 . In recent years multiple description approach was applied to some major coding techniques such as scalar quantization, vector quantization, correlating transforms or quantized frame expansions. Reference [6] gives a good summary of the proposed system designs. As compared to other error resilience approaches requiring retransmissions or long channel codes MD codes have low delays. Low delay on the other hand is especially critical for streaming video which makes MD Video Coding interesting. Streaming video is still a challenging problem due to high data rates and low latency requirements. Several MD coding techniques have been proposed for video, e.g. MD protection of the most significant DCT coefficients, MD coding of motion vectors or altering of prediction loops, listed in [7], [8] und [9,10] respectively. Video Redundancy Coding presented in [4] and [5] adapts the idea of framewise channel splitting to video. The system we will discuss in this paper builds upon the Multiple State Video Coding Approach presented in [1] and which is composed of two subsystems: multiple state encoding/decoding and a path diversity transmission system. In multiple state coding the video is coded into multiple independently decodable each with its own prediction process and state. The advantages are: 1) if one stream is lost the other stream is still decodable, 2) the correctly received stream can enable state recovery for the corrupted stream using bidirectional information from past and future frames. The block diagram of the Multiple State Video Coding Approach is given in Figure 1, Figure 2 shows the state recovery method. The path diversity system on the other hand enables the submission of different subsets of packets over different paths. The assumption is made that the probability of simultaneous losses on two paths is smaller than the individual loss probabilities of the paths. In this work we will additionally assume that the loss probabilities of the chosen paths are independent. On the other hand the distance between frames on the same thread increases with increasing number of descriptions causing a decrease in prediction gain as compared to single description coding of the same sequence. Error resilience however increases due to the state recovery ability of the system. To sum up we tradeoff the conflicting requirements of high coding gain and error resilience. In a two-description system we split the video sequence into the sequences of even and odd * ekmekci@nue.tu-berlin.de; phone: ++49 30 314-23835; fax: ++49 30 314-22514
