Real-time coding of Gauss-Markov sources over burst erasure channels

We study real-time coding of Gauss-Markov sources over burst-erasure channels. A time-invariant encoder sequentially compresses a sequence of vector Gaussian sources, which are spatially i.i.d. and temporally correlated according to a Gauss-Markov model. The channel is a burst erasure channel that erases up to B packets in a single burst. The decoder is interested in instantaneously reconstructing all the source sequences within a quadratic (mean square error) distortion D, except for packets lost in the erasure burst window. We study the minimum achievable rate for the encoder under these constraints and define it as the lossy rate-recovery function R(B, D). We develop lower and upper bounds for the function and observe that the bounds coincide in the high resolution limit. Numerical comparisons indicate that the proposed joint source-channel coding scheme provides significant gains over a separation based scheme.