Latency reduced method for modified successive cancellation decoding of polar codes

Abstract

A latency reduced method is proposed based on the modified successive cancellation (MSC) decoder for decoding polar codes. In the MSC decoder, it was shown that latencies for both the rate-zero and the rate-one nodes can be reduced. By redistributing the information bits, the proposed method can obtain a good rate-zero and rate-one nodes distribution, in which decoding latency can be further reduced under the MSC decoder. Simulation results show that the new polar code (obtained by the proposed method) achieves 8.5% latency reduction with neglected error performance loss compared with the original polar code. Furthermore, it is easy to adjust the trade-off between the error performance and decoding latency by the proposed method.Introduction: Recently, polar codes [1] have attracted a great deal of interest since they are the first codes to provably achieve channel capacity with low-complexity encoding and decoding algorithms in a diverse set of scenarios. Although theoretically very interesting, polar codes have some drawbacks practically. Because of the nature of sequential decoding, the traditional successive cancellation (SC) decoders suffer from a very high latency. In [2], it has been shown that the latency of a rate-one node (introduced later on) can also be reduced like a rate-zero node, in which significant latency reduction is achieved with no sacrifice in error performance. For a polar code with fixed block length, the rate-one and rate-zero nodes distribution is determined by the information set. In this Letter, we propose a method to give a minor adjustment to the original information set and to obtain a new information set with good rate-zero and rate-one nodes distribution, in which decoding latency can be further reduced with small performance loss compared with the original information set.

High-Speed Decoders for Polar Codes

“…Arıkan's original polar code construction only focuses on maximizing the reliability of the information bits. Several altered polar-like code constructions have been proposed in the literature [44]- [46] and their objective is to trade off error-correction performance for decoding complexity reduction by slightly changing the set of information bits, while keeping the code rate fixed. The main idea behind all the altered code constructions is to exchange the locations of a few frozen bits and information bits in order to get more bit patterns that are favorable in terms of decoding latency.…”

“…In all cases, care must be taken in order to avoid using bit locations that are highly unreliable to transmit information bits. The method in [44] first defines a small set of bit locations which contains the n s − h least reliable information bit locations along with the h most reliable frozen bit locations. Then, in order to keep the rate fixed, it performs an exhaustive search over all n s h possible combinations of the n s elements containing exactly h frozen bit locations and selects the combination that leads to the smallest decoding latency.…”

“…As such, the results can not be directly applied to Fast-SSC decoding, where several additional types of special nodes exist. For this reason, in this work we follow the more general exhaustive search method of [44], augmented with a human-guided approach. More specifically, bit-state alterations that would lead to smaller latency are identified by visual inspection of the decoder tree for the unaltered polar code.…”

“…More specifically, bit-state alterations that would lead to smaller latency are identified by visual inspection of the decoder tree for the unaltered polar code. This list of bit locations is then passed to a program to be added to the bit locations considered by the technique described in [44]. Hence, two lists are composed: one that contains frozen bit locations proposed by the user as well as locations that were almost reliable enough to be used to carry information bits, and one that contains the information bit locations proposed by the user and the locations that barely made it into information bit locations.…”

“…That list containing the best candidates is further trimmed by removing all candidates that feature both a greater latency and worse error-correction performance compared to those of their predecessor. Similarly to the technique of [44], our proposed technique does not alter the code rate as the total number of information and frozen bits remains the same.…”