On Predictive Coding for Erasure Channels Using a Kalman Framework

Abstract

We present a new design method for robust low-delay coding of auto-regressive (AR) sources for transmission across erasure channels. The method is based on Linear Predictive Coding (LPC) with Kalman estimation at the decoder. The method designs the encoder and decoder offline through an iterative algorithm based on minimization of the trace of the decoder state error covariance. The design method applies to stationary AR sources of any order. Simulation results show considerable performance gains, when the transmitted quantized prediction errors are subject to loss, in terms of Signal-to-Noise Ratio (SNR) compared to the same coding framework optimized for no loss. We furthermore investigate the impact on decoding performance when channel losses are correlated. We find that the method still provides substantial improvements in this case despite being designed for i.i.d. losses.

Fixed-Lag Smoothing for Low-Delay Predictive Coding with Noise Shaping for Lossy Networks

“…The source process is assumed stable. The encoder (shown in Figure 1) is defined by the following equations [1] …”

“…See [1] for a discussion of this linear model of the quantizer. The quantization noise feecback is de-correlated from d n by scaling d n by ρ in the path subtracting the quantizer input from the quantizer output.…”

“…This allows us to model the fed back noise as white Gaussian which simplifies the optimization of the encoder. The design of Lloyd-Max quantizers for use in the encoder is treated in [1].…”

“…CODING FRAMEWORK This section describes the coding framework. The framework and design method have been presented in [1]. In the following, we summarize and generalize important results of [1], which will be needed in order to establish our main results in Section III.…”

“…In [1] we have presented a method for design of linear predictive coding (LPC) with noise shaping optimized for transmission losses, which does not rely on ECCs in order to achieve the desired degree of robustness towards losses. The coding problem is formulated and solved as a state estimation problem and the coding performance is improved through optimization for the known transmission loss statistics over part of a state-space model representing the source and the encoder.…”