Map source-controlled channel decoding for image transmission system using CPFSK and ring convolutional codes.

摘要

Recently, many novel information technologies involve the transmission of imagery over noisy channels such as satellite and wireless mobile channels. In general, a low-bit-rate image transmission system requires an outstanding image encoder that provides both an excellent quality for the reconstructed image and a high compression ratio. However, the resulting compressed bit stream becomes highly sensitive to channel noise. There have been several approaches to add error resiliency to an image coder. In this work we concentrate on the use of joint source-channel (JSC) methods. In particular, source-controlled channel decoding, based on the residual redundancy in MPEG-4 compressed imagery, is considered. Here an embedded zerotree wavelet (EZW) algorithm is used to generate a compressed bit stream, which is then passed through a ring convolutional encoder (CE) and a CPFSK modulation system. The overall polynomial encoder is the combination of the CE and the continuous phase encoder (CPE). The source-controlled channel decoder exploits the source transition matrix (STM) of the zerotree symbols in computing the combined trellis branch metrics, giving MAP decoding. Simulation results for both the AWGN and flat Rayleigh fading channels show the performance improvement compared to conventional ML decoding.; Moreover, we investigate the design of trellis codes using ring convolutional codes and CPFSK for MAP decoding. The goal is to further improve the performance of the image transmission system when MAP decoding is used. Conventionally a ring convolutional encoder was designed for maximum likelihood (ML) decoding over the AWGN channel. The criteria is to find a code that has the maximum of the minimum squared Euclidean distance. Without considering the source information, this criteria may not be suitable for the case of using MAP decoding. In this work the STM is used in the design of trellis codes for a particular source and value of noise power N0. The "Lena" and "Barbara" images for both single quantization and multi-quantization mode are used.