Polar Codes: Theory Meets Practice to Reduce Latency and Increase Throughput

This talk focuses on polar codes, the latest breakthrough in coding that achieves channel capacity with low-complexity encoding and decoding. There are two main challenges to achieve the promise of polar codes in wireless systems. First, state-of-the-art decoding algorithms for polar codes are sequential in nature, decoding bits one by one. Thus, they suffer from high latency and low throughput. Second, the construction of polar codes that achieve capacity is a difficult task that depends on the underlying channel characteristics and decoder design. To address the first issue, we identify specific sub-codes in the construction of polar codes that can be decoded in one shot with low complexity, while preserving the error-correction performance of the code. We further develop a new theoretical framework to characterize the latency of the proposed decoding algorithm, in which hardware area constraints and the speed at which the code approaches channel capacity are explicitly encapsulated. We present hardware implementation results for the proposed decoder based on this theoretical framework, and show analytical bounds on its latency that are confirmed via simulations and hardware measurements. To address the second issue, we exploit the sequential decoding process in polar codes and map the construction of polar codes to a game in which an agent is trained to traverse a maze. Thus, we minimize the error rate of the code by maximizing the game’s expected return. We then use emerging reinforcement learning algorithms to solve the game. We show that when the state-of-the-art decoders are used, the proposed game-based construction results in a significant performance gain in the error-correction performance of polar codes with respect to the conventional construction methods. The talk concludes with a discussion of open problems in the design of energy-efficient hardware for wireless communications, coding for distributed systems, and coding for new spectrum frontiers.