🤖 AI Summary
To address the high computational complexity of maximum-likelihood (ML) decoding for tail-biting convolutional codes (TBCCs), this paper proposes a two-stage A* graph-search decoding framework. In the first stage, the Viterbi algorithm is executed to obtain reliable survivor paths and state-level statistics. In the second stage, these statistics are leveraged to construct a high-accuracy heuristic function for A*, and a path-confidence-based early-termination mechanism is introduced. This work is the first to exploit Viterbi path statistics explicitly to enhance the precision of the A* heuristic, thereby substantially reducing the search space. Simulation results demonstrate that the proposed decoder achieves strict ML performance while significantly lowering computational complexity compared to standard A* and BCJR decoders—offering marked complexity advantages without compromising optimality.
📝 Abstract
The A* algorithm is a graph search algorithm which has shown good results in terms of computational complexity for Maximum Likelihood (ML) decoding of tailbiting convolutional codes. The decoding of tailbiting codes with this algorithm is performed in two phases. In the first phase, a typical Viterbi decoding is employed to collect information regarding the trellis. The A* algorithm is then applied in the second phase, using the information obtained in the first one to calculate the heuristic function. The improvements proposed in this work decrease the computational complexity of the A* algorithm using further information from the first phase of the algorithm. This information is used for obtaining a more accurate heuristic function and finding early terminating conditions for the A* algorithm. Simulation results show that the proposed modifications decrease the complexity of ML decoding with the A* algorithm in terms of the performed number of operations.