Polar code construction using the information bottleneck method

Abstract

The provably capacity-achieving error-correction capability of polar codes prospered huge interest in academia and industry. The explicit structure of polar codes leads to a different notion of code construction which refers to the assignment of information bits to certain positions in that structure instead of the design of the structure itself. State-of-the-art construction methods either rely on approximations or are computationally demanding to achieve an optimum assignment of the information bits. In this paper, we propose a novel construction technique which leverages the information bottleneck method to combine information-optimum construction and less complex, tractable processing of the required joint distributions. Furthermore, we consider the effect of a low-resolution channel quantizer explicitly in the code design which ensures applicability of the constructed code also in practically relevant settings.