Asymmetric code design for remote multiterminal source coding

We perform asymmetric code design for remote multiterminal source coding (the CEO problem [1,2]) in the quadratic Gaussian case. For an i.i.d. Gaussian source we have two noisy � � � � � � observations � � � � � � and, where � ¡ £ ¥ § © � � � � � and � ¡ £ ¥ § © � � � � are i.i.d. random noises that are independent of each other and � .For remote multiterminal source coding [1–3] of , to achieve the minimum sum rate of the two independent encoders subject to a fidelity criterion � , theoretical bounds were derived independently by Oohama [1] and Chen et al. [2] as where � � � � � � � � � � � � � � © § � and the sum-rate bound (2) is tight. Our main idea is to quantize the � � first observation and apply Wyner-Ziv � � coding on by using the of � � quantized version as side information efficient � � � � � � � � � � in an coding scheme. Specifically, we rely on TCQ � � for quantizing and employ Slepian-Wolf coded TCVQ described in our recent work [4] for Wyner-Ziv � � coding of. Uniform-threshold quantizers are used, and a uniformly distributed dither is added to make the quantization noise independent of the input signal in Wyner-Ziv coding. Assuming optimal quantization, we theoretically show that by varying the rate allocation between the for � � two � � encoders and, our scheme can achieve the two � corner points and on the sum-rate bound (as depicted by the dashed lines in Fig � 1). R 1