Polar Coding for Arbitrary Channels Based on Gaussian Decomposition

Kangjian Qin,Zhaoyang Zhang
DOI: https://doi.org/10.1109/vtcfall.2018.8690944
2018-01-01
Abstract:How to construct polar codes for arbitrary channels has been one of the most appealing and challenging problems in polar coding theory. In this paper, recalling the fact that any continuous distribution function can be approximated by a linear combination of several Gaussian functions with given accuracy, we firstly decompose the underlying channel into a set of independent parameterized Gaussian channels based on its arbitrary channel coefficients. Secondly, we prove that, the constituent channels can still be polarized with probabilistic channel switching using the general polarization matrix, and the average capacity can still be achieved. In this sense, constructing polar codes for an arbitrary channel is equivalent to handling its constituent channels. Thirdly, the extrinsic information transfer (EXIT) analysis is exploited to achieve the mutual information of the constituent channels. Finally, numerical results validate the superior performance of the proposed scheme compared to the existing methods.
What problem does this paper attempt to address?