Universal Polar Codes for More Capable and Less Noisy Channels and Sources (1312.5990v3)

Abstract: We prove two results on the universality of polar codes for source coding and channel communication. First, we show that for any polar code built for a source $P_{X,Z}$ there exists a slightly modified polar code - having the same rate, the same encoding and decoding complexity and the same error rate - that is universal for every source $P_{X,Y}$ when using successive cancellation decoding, at least when the channel $P_{Y|X}$ is more capable than $P_{Z|X}$ and $P_X$ is such that it maximizes $I(X;Y) - I(X;Z)$ for the given channels $P_{Y|X}$ and $P_{Z|X}$. This result extends to channel coding for discrete memoryless channels. Second, we prove that polar codes using successive cancellation decoding are universal for less noisy discrete memoryless channels.