List Decoding of Polar Codes

Abstract: Channel coding is an important instrument used in communication to correct errors that occur on channels. It is interesting to find the best-suited channel code for different communication systems. Polar codes have been in the spotlight lately for their simple structure and performance when in combination with list decoding and cyclic redundancy check code. Polar codes have a recursive structure that makes them interesting to implement in hardware, and they have lately been chosen as a standard for short code communication in 5G to correct bit errors. However, polar codes by themselves are shown to work poorly for practical block lengths, and it is therefore of interest to research them further. This thesis investigates polar codes with a suggested combination of list decoding and CRC. The combination is shown to improve short polar codes enough to compete with the best-known channel codes today for short block lengths. This thesis investigates why this combination works so well with polar codes. The focus lies on the selection of frozen bits in polar codes, in comparison with the similar Reed-Muller codes, and on the size and bit-placement of the CRCs. All investigations focus on codes with length 128 bits and code rate 0.5. We find that a slightly modified frozen bit selection can result in huge performance changes of polar codes. We also find how the use of a list decoder with a large list size improves Reed-Muller codes such that they challenge polar codes both with and without added CRCs. We study if a long CRC is preferred, or if the code performance can be improved by dividing it into several shorter CRCs spread out over the polar code. Results from different modifications to polar codes are presented and discussed.