Decoding Algorithms And Decoder Architecture Design For Double Binary Convolutional Turbo Code

The non-binary convolutional turbo codes that introduced in1999based on binaryturbo codes by C. Berrou et al., innovators of binary turbo code, have beendemonstrated with higher coding efficiency and less equivalent decoding delay.Presently, double binary convolutional turbo code (DB-CTC) has been adopted as thechannel coding scheme by several nowadays international communication standards,such as IEEE802.16m in2010. Although there are on-shelf achievements for binaryturbo codes but there still are spaces for DB-CTC’s decoding algorithms andengineering requirements. Therefore, it is a realistic importance for developingsimplified decoding algorithm and memory reduced decoder for DB-CTC.DB-CTC is symmetry in both time and spatiality. The time symmetry appears initerative calculations for forward and backward state metrics and it is possible todecrease the iteration time only by an extra data exchanging at the end of the iteration.In spatial the trellis diagram of DB-CTC can be decomposed into two sub-trellis ofclassic turbo code. Such parallel characteristics in trellis diagram hereafter results in apossibility of simplifying the decoder architecture of DB-CTC.In this dissertation, the encoding structure of DB-CTC is analyzed, an improvedMAP algorithm is proposed and the decoding complexity is decreased by optimizing themultivariablemax*operation. Secondly, to accelerate the decoding speed, abidirectional decoding architecture for DB-CTC is suggested. Thirdly, to decreasememory size of the decoder, a novel strategy called the “Compare-Select-RecalculationProcessing (CSRP)” is recommended, and the corresponding three decodingarchitectures are proposed. Moreover, this strategy is applied to the logarithm domain,and related decoding structure is also proposed.Therefore, researches of this dissertation are outlined as the following four aspects.1. Proposition of an improved MAP algorithm for DB-CTC.A new definition for branch metrics is proposed based on classic MAP algorithmand the exponential and logarithm computations are transferred into a pre-processingunit ahead of decoding. Subsequently, to reduce the decoding complexity, the exponential calculation is linearly approximated in the pre-processing unit. Then theimproved decoding algorithm is made up while the performance is only0.01dBdeteriorated comparing to that of the classic MAP algorithm.2. Simplification of the multivariablemax*operation for DB-CTC.By investigating two kinds of approximation for the Jacobian logarithm,complexity of the recursive multivariablemax*operation is analyzed. Subsequently, anew approach is proposed to decomposed the recursive multivariablemax*operationinto compare operations and one Jacobian logarithm, by which complexity of therecursion is reduced, while the BER performance is improved by0.03~0.06dBcompared to the recursive approach.3. Proposition of a bidirectional decoding scheme for DB-CTCRedundancy in the computation of the branch metrics is analyzed. By introducingthe branch metrics matrix, we rewrite computation of the forward (backward) statemetrics with the departed factors, and find the forward factor matrix can bere-constructed by the backward factor matrix, and vice versa. Therefore, a decodingarchitecture is proposed by using this symmetry, at the cost of low hardware overhead,the decoding speed is one time accelerated comparing to that of the conventionaldecoding architecture.4. Architecture design of memory-reduced decoder for DB-CTC(1) It has been found that there is no need to store all of the state metrics in thestate metrics cache (SMC). And a new strategy to decrease the SMC size is proposed,i.e. the CSRP methodology. Only parts of the state metrics are stored, while theun-stored state metrics are recalculated by a CSRP unit.(2) By adopting the butterfly decoding scheme and the CSRP unit, a bidirectionaldecoding architecture with low SMC size is proposed, by which memory size in theDB-CTC decoder is considerably reduced.(3) In this dissertation, the CSRP strategy is extended into the logarithm domain,and themin*operation is defined. By approximating the correction term with linearfunctions, error of the approximation is limited, and has been shown that the SMC sizeis reduced by45%, while BER is only0.05dB inferior to that of the Log-MAPalgorithm.Seen from the conclusions of this dissertation, although DB-CTC has been recommended as the coding scheme for the fourth generation wireless communicationstandards, there still are spaces to optimize the decoding algorithms and architectures.Therefore, results of this dissertation lay a foundation for the engineering applicationand further efficiency improvement of DB-CTC.