Finite Alphabet Fast List Decoders for Polar Codes

TL;DR

This work incorporates the fast decoding schedules for polar codes into their quantized finite alphabet decoding, and shows that the finite alphabet decoders can also leverage the special decoder nodes found in the literature.

Abstract

The so-called fast polar decoding schedules are meant to improve the decoding speed of the sequential-natured successive cancellation list decoders. The decoding speedup is achieved by replacing various parts of the serial decoding process with efficient special-purpose decoder nodes. This work incorporates the fast decoding schedules for polar codes into their quantized finite alphabet decoding. In a finite alphabet successive cancellation list decoder, the log-likelihood ratio computations are replaced with lookup operations on low-resolution integer messages. The lookup tables are designed using the information bottleneck method. It is shown that the finite alphabet decoders can also leverage the special decoder nodes found in the literature. Besides their inherent decoding speed improvement, the use of these special decoder nodes drastically reduces the number of lookup tables required to perform the finite alphabet decoding. In order to perform quantized decoding using lookup operations, the proposed decoders require up to 93% less unique lookup tables as compared to the ones that use the conventional successive cancellation schedule. Moreover, the proposed decoders exhibit negligible loss in error correction performance without necessitating alterations to the lookup table design process.

Figures (7)

• Figure 1: Graph representation of a polar code with $N=8$.
• Figure 2: Decoder-tree representations of a polar code with $N=8$ and $R=1/2$ with $\mathcal{A}{=} \{ 3,5,6,7\}$. (a) , (b) Simplified (c) Fast simplified . Node types: $SC$, $R0$, $R1$, $Rep$, $SPC$.
• Figure 3: Factor graph of the building block and the transmission channel.
• Figure 4: Quantized decoding under and schedules. $N{=}1024, R{=}0.5, N_{\text{CRC}}{=}16, N_L{=}32$, and BPSK over AWGN channel.
• Figure 5: Quantized decoding under and schedules. $N{=}1024$, $R{=}0.25$ and $0.75$, $N_{\text{CRC}}{=}16, N_L{=}32$, and BPSK over AWGN channel.