Soft Dequantization aided by Error Control Coding in Various Communication Scenarios

• Main Authors: Feng-Chun Lu, 盧逢春
• Other Authors: kuen-Tsair Lay
• Format: Others
• Language: zh-TW
• Published: 2004
• Online Access: http://ndltd.ncl.edu.tw/handle/13848558169787416645

碩士 === 國立臺灣科技大學 === 電子工程系 === 92 === In digital communications of multimedia signals, it is desired that the receiver reliably receives as much message as possible. To achieve these goals, the techniques of error correction coding (also called channel coding) and data compression (also called source coding) are applied. Traditionally, they are performed in two separate stages. Recently, however, joint schemes for source and channel coding have been proposed in the literature. This thesis is also about a joint source channel coding scheme. Our goal is to reduce the distortion suffered by a signal/waveform when it is quantized at the transmitter and then sent through a noisy or fading channel to the receiver. In the part of source coding, scalar quantization and trellis-coded quantization (TCQ) are considered. In the part of channel coding, convolutional codes and turbo codes are considered. The major contribution of this thesis is that we propose to take advantage of some soft-decision information available in the channel decoding. More specifically speaking, we propose to decode the received bit sequence with maximum a-posteriori (MAP) estimation, for which the BCJR algorithm is one famous method to accomplish. In the MAP estimation, the a-posteriori probabilities (APP''s) of received bits/symbols can be obtained. Then, we take into account those probabilities when we try to reconstruction (i.e. dequantize) the sampled signal intensities. Treating the APP''s as soft information, we call our proposed scheme soft dequantization. As compared to hard dequantization, we learned from experiments that soft dequantization performes much better when the signal-to-noise ratio (SNR) in the communication channel is low. Moreover, the improvement is even more obvious when the technique of residual dequantization is adopted. In residual dequantization, quantization of sampled intensity is performed progressively. Approximation to the sampled is first quantized into one bit. Then, the residual between the original value and the reconstruction value is further quantized into another bit; And so on. As to the types of communication channels, two cases are considered : additive white Gaussian noise (AWGN) channel and Rayleigh wireless channel. The various scenarios mentioned above, including different coding and quantization/dequantization schemes, and different channel models, are simulated. Their results are shown and compared.