[Paper Review] Nested Polar Codes Achieve the Shannon Rate-Distortion Function and the Shannon Capacity

Aria G. Sahebi, S. Sandeep Pradhan|arXiv (Cornell University)|Jan 25, 2014

Error Correcting Code Techniques3 references3 citations

TL;DR

This paper demonstrates that nested polar codes achieve the Shannon capacity for arbitrary discrete memoryless channels and the Shannon rate-distortion function for arbitrary discrete memoryless sources by leveraging group structure and channel polarization over finite Abelian groups. The approach generalizes Arikan's original polar codes using a nested structure that enables optimal performance across binary and non-binary alphabets.

ABSTRACT

It is shown that nested polar codes achieve the Shannon rate-distortion function for arbitrary (binary or non-binary) discrete memoryless sources and the Shannon capacity of arbitrary discrete memoryless channels.

Motivation & Objective

• To extend the optimality of polar codes beyond symmetric capacity to the full Shannon capacity of arbitrary discrete memoryless channels.
• To establish that nested polar codes achieve the Shannon rate-distortion function for arbitrary discrete memoryless sources, not just symmetric rate-distortion functions.
• To generalize polar coding techniques from binary to non-binary alphabets using finite Abelian groups and group operations.
• To demonstrate that nested polar codes achieve optimal performance with vanishing error probability and negligible rate loss in both channel and source coding settings.

Proposed method

• Utilizes polar codes with the original $(u, u+v)$ kernel over finite Abelian groups to generalize polarization to non-binary alphabets.
• Employs a nested coding structure where subgroups $K \leq H \leq \mathbf{G}$ are used to partition code indices based on polarization behavior.
• Defines sets $A_{H,K}$ and $B_{H,K}$ based on the Bhattacharyya parameter $Z^H(W_{c,N}^{(i)})$ and $Z^H(W_{s,N}^{(i)})$ to identify reliable and unreliable channels.
• Uses transversals $T_H$ and $T_{K \leq H}$ to decompose group elements into components for encoding and decoding, enabling structured message injection.
• Applies a two-stage encoding: for indices in $A_{H,K}$ with $K \leq H$, $[v_i]_K$ is known and $[v_i]_{T_{K\leq H}}$ carries the message; for $K \nleq H$, $v_i$ is transmitted via a separate polar code.
• Decoding uses maximum likelihood over cosets: $\hat{v}_i = \arg\max_{g \in [v_i]_K + [v_i]_{T_{K\leq H}} + T_H} W_{c,N}^{(i)}(z_1^N, \hat{v}_1^{i-1} | g)$, ensuring reliable reconstruction.

Experimental results

Research questions

• RQ1 Can nested polar codes achieve the Shannon capacity of arbitrary discrete memoryless channels, including non-binary ones?
• RQ2 Do nested polar codes achieve the Shannon rate-distortion function for arbitrary discrete memoryless sources with general reconstruction alphabets?
• RQ3 How can the polarization framework be extended from binary to non-binary alphabets using finite Abelian groups?
• RQ4 What is the rate loss due to transmitting unreliable indices in the nested structure, and can it be made negligible?
• RQ5 Can the nested structure be designed such that the fraction of indices requiring explicit transmission vanishes as block length increases?

Key findings

• Nested polar codes achieve the Shannon capacity of arbitrary discrete memoryless channels, including non-binary channels, by using group-based polarization and nested coding over finite Abelian groups.
• The fraction of indices $i$ for which $K \nleq H$ and thus require explicit transmission via a separate polar code vanishes as $N \to \infty$, ensuring negligible rate loss.
• The probability of error in both channel and source coding settings converges to zero as block length $N$ increases, due to the polarization of Bhattacharyya parameters.
• The symmetric rate-distortion function is achieved for binary sources in prior work, but this paper extends the result to arbitrary discrete memoryless sources with general reconstruction alphabets.
• The use of transversals and subgroup decomposition allows for structured message injection and reliable decoding even in non-binary settings.
• The analysis shows that $Z^H(W_{c,N}^{(i)}) \to 0$ and $Z^H(W_{s,N}^{(i)}) \to 1$ for $i \in A_H$, and $Z^H(W_{c,N}^{(i)}) \to 1$ and $Z^H(W_{s,N}^{(i)}) \to 0$ for $i \in B_H$, confirming polarization.