What Are Polar Codes?

Polar codes are a family of linear block error-correcting codes that provably achieve the symmetric channel capacity of a binary-input discrete memoryless channel when decoded with a successive cancellation (SC) algorithm, as the block length grows. Introduced by Erdal Arikan in his landmark 2009 IEEE Transactions on Information Theory paper, polar codes were the first class of codes demonstrated to be capacity-achieving with an explicit, low-complexity construction rather than a probabilistic ensemble argument. The codes are built on the phenomenon of channel polarization: by applying a recursive transform to N copies of a noisy channel, the resulting N synthetic channels polarize into two groups, those that are nearly noiseless and those that are nearly completely noisy. Information bits are assigned to the reliable synthetic channels and the remaining channels are fixed to known frozen values, giving the encoder its rate and structure. Polar codes drew on prior work in channel coding and information theory, and they share a structural relationship with Reed-Muller codes, a classical family predating them by more than fifty years.

Channel Polarization and Code Construction

The polarization transform is constructed by recursively applying a 2-by-2 kernel matrix, typically the matrix with rows [1, 0] and [1, 1], through Kronecker products to produce a transform matrix of size N by N where N is a power of two. When this transform operates on N independent uses of the same physical channel, it produces N synthetic channels whose individual capacities span from near zero to near one, while the sum of their capacities equals the total of the N original channels. The fraction of channels with capacity approaching one equals the underlying channel's symmetric capacity I(W), and the remaining fraction approaches zero; in the limit, all channels are fully polarized. The encoder places message bits on the K most reliable synthetic channels and zero-valued frozen bits on the remainder, achieving a code rate of K over N. Block lengths in practical deployments are powers of two, typically ranging from 32 to 1024 bits for control channel applications. As detailed in ArXiv preprint 2502.11053 on 5G polar and LDPC codes, polar codes in the 5G New Radio (NR) standard use a nested reliability sequence derived from offline density evolution analysis rather than computing polarization for each specific channel instance.

Decoding Algorithms

The original SC decoder introduced by Arikan processes the N-bit codeword in a binary tree structure, estimating each message bit sequentially, with each decision conditioned on all previously decoded bits. SC decoding has complexity of order N log N, making it computationally tractable for long codes. Its shortcoming is that at finite block lengths, the error-correction performance is mediocre compared to LDPC codes of the same rate and length. Successive Cancellation List (SCL) decoding improves performance by maintaining L parallel decoding paths through the tree simultaneously and selecting the most likely path at the end. When SCL is combined with a cyclic redundancy check (CRC) that identifies which list path contains the correctly decoded codeword, the resulting CRC-Aided SCL (CA-SCL) algorithm delivers error-correction performance that meets or exceeds LDPC codes at the block lengths relevant to 5G control channels. The arXiv analysis of list decoding for 5G NR polar codes describes three CA-SCL decoder variants distinguished by how they use distributed CRC bits placed at different positions throughout the message sequence, balancing early termination speed against detection reliability.

Relationship to Reed-Muller Codes

Reed-Muller codes, first described in 1954, are constructed by selecting row subsets from the same Kronecker-product transform matrix used in polar codes. Reed-Muller codes select rows by Hamming weight, taking all rows with weight at least a specified minimum. Polar codes instead select rows by channel reliability, producing a selection that coincides with Reed-Muller at high rates and diverges at intermediate rates. This overlap links polar codes to the structural theory of Reed-Muller codes and their error-correction properties, including established distance bounds and connections to Boolean function theory. The relationship has guided research into generalizing polar codes beyond binary kernels and power-of-two lengths.

Applications

Polar codes have applications in a range of fields, including:

• 5G NR control channels, including broadcast, paging, uplink control information, and downlink control information
• Satellite and deep-space communications, where capacity-approaching performance at moderate block lengths is valuable
• Data storage systems, as candidates for flash memory error correction
• Post-quantum cryptography research, where polar code structure appears in code-based cryptographic proposals
• Theoretical information theory, as a constructive proof of the channel coding theorem