Stochastic BP Polar Decoding and Architecture with Efficient Re-Randomization and Directive Register

Abstract

Recently, polar codes have gained their popularity because of its capacity achieving property. On the other hand, representing continuous values by unary bit streams, stochastic decoding shows its advantages in both hardware efficiency and error resilience. Therefore, this paper devotes itself in combining both together to offer a stochastic belief propagation (BP) polar decoder, which is suitable for applications requiring low complexity, high fault tolerance, but not sensitive to processing speed. In this paper, several approaches such as bit length increasing and efficient re-randomization are proposed and carefully analyzed to improve the decoding performance. Numerical results have shown that for N = 64 and N = 256, the proposed stochastic BP decoder can achieve comparable decoding performance as the deterministic one. Hardware architecture employing directive register is also proposed in detail for efficient implementation of the proposed stochastic decoder.