【作者】 杨帆；

【导师】 田宝玉；

【作者基本信息】 北京邮电大学 ， 信号与信息处理， 2009， 博士

【摘要】 Turbo码的产生和低密度奇偶校验(Low Density Parity Check,LDPC)码的重新发现是信道编码研究领域的巨大进展。虽然这两种码的纠错性能已逼近香农信道容量的极限,但其较高的编译码复杂度和较大的处理时延仍然限制了它们的应用。因此提出高性能,低复杂度和低时延的信道编译码算法是信道编码技术实用化的重要环节,也是当前学术界关注的研究课题。本文在对卷积码、Turbo码和LDPC码研究的基础上,提出了几种低复杂度和低时延的改进算法,分析了这些算法的性能和译码复杂度,并进行了仿真验证。主要创新点包括如下几方面。针对传统LDPC串行译码算法计算存在冗余的问题,提出了一种改进的LDPC串行译码算法。与传统算法不同的是,该算法只在初始化时完整地计算一次变量节点对校验节点软信息的求和,而在后续的迭代译码过程中通过对它局部更新得到求和结果,从而显著减少了计算量。与传统算法相比,所提出的算法在不降低纠错性能的前提下,具有更低的计算复杂度和更快的处理速度。提出了一类级联的卷积码混合译码算法。该算法包括两级译码过程,第1级采用置信传播(Belief-Propagation,BP)算法,第2级则采用修改的维特比(Modified Viterbi Decoding,MVD)译码算法。在中高信噪比条件下,BP预译码能够纠正接收序列中的大部分错误,余下的少量的错误可以通过MVD进一步纠正。仿真表明,与传统的维特比译码算法相比,混合译码算法的误码性能只有很小的损失,且平均译码复杂度在中高信噪比条件下有明显降低。基于扩展咬尾递归系统卷积(Extended Tail Biting RecursiveSystematic Convolutional,ETB-RSC)码的思想,提出了一种用于长帧数据传输的高度并行的咬尾Turbo码(High Paralleled Tail Biting TurboCode,HP-TBTC)结构,其特点是在发/收端,信息序列/接收序列被拆分为若干等长的子序列同时作编/译码处理。虽然HP-TBTC结构是以牺牲一定的硬件资源为代价换取编译码的并行处理,但它可以成倍地提高编译码处理速度。仿真表明,在适中信噪比条件下,HP-TBTC的性能也略好于传统的咬尾Turbo码。提出了一种适用于短帧数据传输的Turbo-like码(TLC)。该码的分量编码器采用的是消除环4的前向系统卷积(Feed-forwardSystematic Convolutional,FSC)编码器,与之对应的分量译码器采用的是改进的LDPC串行译码算法。仿真表明,当码字长度低于某个值时,TLC的性能比Turbo码更好,并且在中高信噪比时,TLC具有更低的功率损耗和极低的译码复杂度。更多还原

【Abstract】 The birth of Turbo codes and the rediscovering of low density parity check (LDPC) codes are the significant process in channel coding. The performance of the two error-correcting codes has been close to the Shannon limit, however, the high decoding complexity and large processing delay restricts the application of the two codes. Therefore proposed for channel coding techniques with high performance, low decoding complexity and latency is an important aspect for practical applications, it is also the academic community concerned about.This dissertation researches on the encoding and decoding techniques of convolutional codes, Turbo codes and LDPC codes, proposes some modified algorithms, and analyzes the performance and decoding complexity of the proposed algorithms via computer simulations. The main innovations include the following aspects.By the computation redundancy in conventional serial decoding algorithm, a modified serial decoding algorithm for LDPC codes is proposed. Unlike conventional approaches, this algorithm calculates the sum of variable-to-check messages only once at the initialization stage, and then updates it by simple recursions during the decoding stage. Compared with the conventional serial LDPC decoding algorithm, the proposed algorithm has lower computational complexity and faster processing speed without any performance deterioration.A type of hybrid decoding (HD) algorithm for convolutional codes is proposed. HD algorithm can be implemented by using two-stage decoding, where the first stage uses the belief propagation (BP) algorithm, while the second stage uses the modified Viterbi decoding (MVD) algorithm. In moderate-to-high signal-to-noise ratio (SNR), BP pre-decoder corrects most of the errors in received sequence, the remaining few errors can be further corrected by MVD. Simulations show that compared with the conventional Viterbi decoding algorithm, the proposed algorithm has a little performance deterioration with much lower average complexity in moderate-to-high SNR.According to the concept of extended tail biting recursive systematic convolutional (ETB-RSC) codes, a high paralleled tail biting Turbo code (HP-TBTC) structure is proposed for large frame transmission. The best idea of HP-TBTC is that the source/received bit sequence can be divided into several sub-streams at the transmitter/receiver, which can be can be encoded/decoded simultaneously. HP-TBTC can be implemented in parallel to obtain much higher encoding and decoding speed at the cost of more hardware resource. Simulations show the performance of HP-TBTC is better than that of the tail biting Turbo codes at the appropriate SNR.A type of Turbo-like code (TLC) for short frame transmission is proposed. The component encoder of TLC is feed-forward systematic convolutional (FSC) code designed without 4-cycles, while the associated component decoder uses serial belief propagation rule. Simulations show that TLC outperforms Turbo coding when the frame sizes are less than a threshold. Moreover, compared with Turbo coding, TLC has far less power consumption and decoding complexity in the medium-to-high SNR region.更多还原