This thesis describes a method to produce soft outputs from a, conventional stack algorithm with the intent of obtaining a relatively low complexity decoder, suitable for use in an iterative decoding framework. The soft stack algorithm proposed uses a simple averaging operation to derive soft ouputs from the stack algorithm. The increase in computation needed for the derivation of soft values is trivial. The increase in storage is also minimal, consisting of four integer arrays. Also, the Fano metric conventionally used in the stack algorithm is retained and modifications to the algorithm are minor. This retains all the reduction in computational complexity and latency obtained by using the sub-optimal stack.; In the thesis, the soft stack is applied to convolutional coding. A gain of about 1 dB is obtained when the soft values of the stack are used as opposed to the hard decisions of the stack. The soft values of the stack are compared with those produced by the BCJR (Bahl, Cocke, Jelenik and Raviv) algorithm and are found to correlate highly with those values especially at SNR (signal-to-noise ratio) above 3 dB. The BER of the soft stack, BCJR and the Viterbi are also compared. Finally, the soft stack is applied to turbo decoding scenario where we find that it does not gain enough over iterations to be comparable to the optimal BCJR. Thus, this thesis establishes a generic principle to obtain soft values from tree-based decoding algorithms. It also shows that these values may be utilized in iterative settings. Further, this is achieved with minimal increase in the computational complexity and storage requirements of these algorithms.
展开▼
