论文标题:软判决译码的研究
Research on Soft Decision Decoding
论文作者 陈军
论文导师 王新梅,论文学位 博士,论文专业 通信与信息系统
论文单位 西安电子科技大学,点击次数 81,论文页数 89页File Size3365k
1999-12-01论文网 http://www.lw23.com/lunwen_858452122/ 信道编码,软判决译码,广义门限,格图,搜索技术
Channel Coding,Soft Decision Decoding,Generalized Threshold, Trellis,Search Technic
信道编码是提高通信可靠性的重要手段。达到Shannon限（即信道容量）是信道编码理论研究的根本目标。实践证明，软判决译码是最有希望达到这一目标的重要方法之一。目前，研究最佳（或准最佳）性能、低复杂度的软判决译码已经成为信道编码研究的重要课题。本文围绕这个课题研究了快速有效的软译码算法，主要研究成果包括：1.针对基于分组码树图的A~*译码算法，推导了一个简单的广义门限，增加了一 条新的搜索规则，提出了带门限检测的A~*译码算法，减少了不必要的搜索， 使得译码速度更快、同时误码性能接近最大似然译码性能。2.针对基于分组码格图的序列译码Stack算法，推导了更一般的广义门限，获 得了两种门限—Fano广义门限和A~*广义门限，提出了带门限检测的序列译 码Stack算法，利用最佳门限对侯选码字进行最优性检测，满足广义门限的 侯选码字作为译码输出，减少了Stack算法不必要的搜索。分析了采用不同 的度量函数、偏差项和搜索方向对译码性能、计算复杂度的影响。3.针对基于分组码有向树（图）的最大似然软判决译码，提出采用新的度量函 数，使计算更简单；采用更有效的搜索算法—DA算法，搜索分组码有向树 （图）上的最佳图样；建立错误图样的广义门限，进一步加快搜索速度。同 时指出：采用非最佳信号形式会导致性能损失近3dB。4.利用遗传算法固有的并行特性和启发式搜索能力，提出了基于遗传算法的、 有效的软判决译码。将分组码的软判决译码问题转化为相应的组合优化问题， 采用遗传算法进行快速优化计算，完成快速软判决译码。指出卷积码M译码 算法存在路径选择策略过于单一、容易丢失最佳路径的缺陷，提出将卷积码 格图上的单向和双向搜索译码转化为遗传空间的群体进化过程，利用遗传算 法的群体多样性好、全局搜索能力强的特点，提高搜索质量，改善译码性能。 对于纠错能力较强的分组码和卷积码，这类算法具有较高的实用价值。5.指出Chase算法在试探译码时会产生重复的侯选码字，提出采用人工智能搜 索技术—A~*算法，快速生成试探序列集合，并利用已经试探译码的信息，对 试探序列集合进行分类，生成试探序列的等价类及其代表，仅对代表进行试 探译码和最佳测试，从而获得了一种迭代的分组码快速软判决译码，其译码 速度更快而性能与Chase算法完全相同。
AbstractChannel coding is a fundarnental method to improve the conununication reliabilityTO reach the channel caPacity is the ulthate objechve of the research on channel codingtheory. Extensive prachce has shOwn tha soft decision decoding is one of the mostprospective aPproaches to this goal. Recentiy, the problem of finding computationallyefficient and practically imPlementable softdecision opted (or sub-optimal) decodingis still an open and challenging problem of channel coding theory. This thesis put itsemphasis on this problem to devise efficient soft decision decoding algorithms. Themain resultS are as following.l. Derive a simple form of generalised threshold, and proPOse a new search nde fortree-based A* decoding of block codes, and then obtain a threshold A* decodingalgorithIn. The proPOsed algorithm avoid redUndant search, thus speed up the A*decoding while mai-ng maximum lthelihood decwhng performance.2. Derive a generai form of generalized threshold, and obtain two kinds of thethreshold, i.e., Fano and A* generalized thresholds, and then propose a trellis-basedthreshold sequential decoding Stack algorithIn, which can avoid redundant searchand improve the decoding speed of the Stack algorithIn. The effects of path metricfunction bias item and search direction are corresPOndingly analyzed on bothdecoding Performance and complexity3. Use a computational efficient metric function and a more et1icient searchalgorithm --Dijkstra"s algorithIn to search raPidly through the directed tree (graPh)of biM block codes for the maximurn likelihood error pattern. and performoptimality test on ermr pattems to speed up the decoding fiJwher. It is emphasizedthat usage of non-optimal bosmission signal wou1d result in nearly 3dBPerformance loss.4. ProPOse a fast soft decision decoding, which makes use of parallel structure andheuristic searching ability of the genetic algorithm (GA). \\!e fOrmalize thedecoding of block codes as an aPpropriate combinaorial optimization problem, andthen use GA to complete oPtimiZaion computaion raPidl}r, thus obtain a fast soft-decision decoding algoritIun. We indicate the some drawacks of M decodingalgorithIn of convolutional codes and convert the decoding into search process ingenetic space. It utilizes GA"s good characteristic in POpulation diversity, widesearch region and global search ability to imProve the decoding performance.5. Point out that Chase algorithIn would generae reduplicate candidate codeworddining test decoding, which can lead to slow decoding speed. So we propose a novel iterative threshold Chase algorithm (ITC). It partitions the set of test patterns into the equivalence classes and generates the equivalence class representatives. Both test decoding and optimality test are performed only on the representatives. Therefore, much more redundant test decoding can be avoided and much more hard decision decoding time also can be saved while its decoding performance keeps same as Chase algorithm.
|
