| dc.description.abstract | Iterative turbo decoders have become popular in error correction because of their excellent Bit Error Rate (BER) characteristics, approaching Shannon limits. They have shown great potential in a variety of areas like wireless, DSL, and are included in standards like 3GPP. However, they are computationally intensive, and due to iterative processing, decoding delays are large, making them unsuitable for high-speed data transfers. This thesis aims to reduce the decoding delay in iterative convolutional turbo decoders, leading to improved turbo decoder performance.
In this thesis, a performance evaluation of the two major turbo decoding algorithms-Maximum A Posteriori (MAP) algorithm, popularly known as the BCJR algorithm (named after Bahl, Cocke, Jelinek, and Raviv), and the Soft Output Viterbi Algorithm (SOVA)-is carried out. It is known that the MAP decoder outperforms the SOVA decoder. We focus on the implementation of the MAP decoder. The MAP decoder requires complex units like logarithmic, exponential, division, and multiplication operators for its implementation. The Log-MAP algorithm, an implementation-friendly version of the MAP algorithm (where all state metrics required for MAP are computed in the log-domain), is investigated in detail. The variation of BER with frame size, interleaver, and encoding polynomial for different Signal-to-Noise Ratios (SNRs) is simulated.
Total decoding delay in an iterative decoder is proportional to the sum of delays in the decoder and interleaver, multiplied by the number of iterations required before stopping the iterative decoding process. The number of iterations becomes a multiplication factor and has a significant effect on decoding delay. We propose the concept of adaptive scaling for iterative convolutional Log-MAP decoders to reduce the average number of iterations required to achieve a given BER for a range of SNRs. An adaptive scaling scheme for low SNRs (0.2-1.5 dB), where turbo decoding is most useful, is proposed and shown to reduce decoding delay considerably. By reducing the number of iterations, adaptive scaling not only decreases decoding delay but also reduces the power consumption of the decoder, making it suitable for low-power applications.
Next, we evaluate the performance of two parallel MAP schemes known in the literature, aimed at reducing decoding delay in Log-MAP decoders by increasing hardware. The Parallel MAP algorithm based on the belief-propagation algorithm is shown to perform better than the Parallel MAP based on the sliding window algorithm, and has no requirement for long learning lengths to avoid degradation due to interleaving of window edges among themselves. Finally, the proposed adaptive scaling scheme is shown to decrease the number of iterations required in the parallel MAP scheme as well. | |
