[Paper Review] From sequential decoding to channel polarization and back again

Erdal Arıkan|arXiv (Cornell University)|Aug 26, 2019

Cellular Automata and Applications120 citations

TL;DR

The paper traces the evolution from sequential decoding and cutoff-rate boosting to channel polarization, introduces polar codes and PAC codes, and analyzes their complexity and performance with practical decoding schemes.

ABSTRACT

This note is a written and extended version of the Shannon Lecture I gave at 2019 International Symposium on Information Theory. It gives an account of the original ideas that motivated the development of polar coding and discusses some new ideas for exploiting channel polarization more effectively in order to improve the performance of polar codes.

Motivation & Objective

• Motivate polar coding by linking sequential decoding, Massey/Pinsker schemes, and capacity concepts.
• Explain channel polarization via multi-level coding with MSD and the polar transform, and analyze complexity.
• Introduce 0-1 rate assignment leading to polar codes and discuss SC decoding performance and bounds.
• Propose PAC codes by fronting polar transforms with outer convolution, and evaluate decoding with Fano/sequential approaches.
• Provide practical design guidance and comparisons to dispersion performance for finite lengths.

Proposed method

• Review of convolutional codes and sequential decoding foundations.
• Analysis of Massey’s MEC labeling to illustrate capacity preservation and cutoff-rate boosting.
• Pinsker’s scheme with inner/outer block codes and multiple sequential decoders.
• Multi-level coding with MSD and the polarization transform P_n, proving polarization with complexity O(N log N).
• Construction of polar codes by selecting A to minimize the SC bound using Z(W_i) or R0(W_i).
• Introduction of PAC codes with outer convolution T before the polar transform, and decoding with sequential (Fano) methods.

Experimental results

Research questions

• RQ1 Can polarization be achieved with practical complexity to approach channel capacity for binary-input memoryless channels?
• RQ2 How should data indices A be chosen to minimize the SC error bound and maximize performance?
• RQ3 Do PAC codes outperform pure polar codes at finite lengths and under practical sequential decoding?
• RQ4 What decoding strategies (SC, list decoding, Fano) best leverage PAC code structure to approach dispersion performance?

Key findings

• Polarization causes the fraction of bit-channels with C(W_i) > 1−δ to approach C(W) and those with C(W_i) < δ to approach 1−C(W); the corresponding R0(W_i) polarizes similarly.
• With the P_n transforms, polarization can be achieved at complexity O(N log N) per mapper block and enables scalable construction.
• Polar codes with 0-1 rate assignments yield a stand-alone SC-decoded code with FER bounded by O(exp(−N^0.499)) for any fixed rate R < C(W).
• Polar codes are capacity-achieving for symmetric BMCs with low-complexity encoding/decoding and constructive data-index selection based on small Bhattacharyya parameters (or large cutoff rates).
• PAC codes place an outer convolution before the polar transform, creating an irregular tree code that can be decoded with sequential/Fano decoding, and simulations show FER approaching dispersion for moderate to high SNR and N=128; RM-based A selection performed best in reported results.
• The PAC approach reduces reliance on large outer codes and leverages polarization to improve finite-length performance, with design choices (A and c) impacting decoding speed and FER.