The CEO Problem With Inter-Block Memory

An $n$ -dimensional source with memory is observed by notation="LaTeX">$K$ isolated encoders via parallel channels, who compress their observations to transmit the decoder noiseless rate-constrained links while leveraging of past. At each time instant, receives new codewords from observers, combines them past received codewords, and produces a minimum-distortion estimate latest block symbols. This scenario extends classical one-shot CEO problem multiple rounds communication communicators maintaining We extend Berger-Tung inner outer bounds inter-block memory, showing that minimum asymptotically (as notation="LaTeX">$n \to \infty $ ) achievable sum rate required achieve target distortion bounded minimal directed mutual information problems. For Gauss-Markov AWGN we show bound tight solve corresponding problem, thereby establishing rate. Finally, explicitly loss due lack among observers; attained equality in case identical observation channels. The general coding theorem proved nonasymptotic uses stochastic likelihood coders whose asymptotic analysis yields an extension causal setting. Gaussian facilitated reversing channels observers.