ANN Based Adaptive Successive Cancellation List Decoder for Polar Codes

Abstract

Combined with cyclic redundancy check (CRC), SC list (SCL) decoder can achieve outstanding error correction performance, which is more obvious with increasing list size. However, the corresponding decoding complexity and latency increase with the list size. To this end, the selection of list size becomes essential for practical applications. A new artificial neural network (ANN) based framework is proposed in this paper to design a hardware-friendly adaptive SCL (DL-ASCL) decoder. First, the list size at each stage is predicted by an ANN predictor. The performance achieved based on the proposed DL-ASCL algorithm is close to the optimal SCL decoder with the same list size, especially in the high signal-to-noise ratio (SNR) region. Meanwhile, the computational complexity is significantly reduced compared with the conventional ones. Numerical results have demonstrated that the proposed deep learning based adaptive SCL decoder can achieve 56% computational complexity reduction compared with the conventional SCL decoder for the polar code with length 128 and rate 1/2. The hardware architecture of the adaptive SCL decoder based on the predicted list size is proposed and the folding technique is also adopted, which helps reduce the hardware cost by about 25%.