Scalable predictive coding as the Wyner-Ziv problem

An alternative to scalable predictive coding of first order Gauss-Markov processes is proposed in this paper. It is shown that conventional scalable predictive coding is inherently suboptimal. An alternative to scalable predictive coding, which achieves the rate-distortion performance of predictive coding for first-order Gauss-Markov processes is then proposed. The proposed approach is posed as a variant of the well-known Wyner-Ziv (1976) problem. By using coset codes with nested lattices, the present paper proves that the proposed approach achieves the predictive coding bound asymptotically at all scales while simultaneously providing the functionality of scalable coding.