Logo der Uni Stuttgart
Channel Polarization

Channel polarization is the concept upon which polar codes are based.

Channel polarization is the process through which $N$ distinct channels are generated ${\mathrm{W}_{N}^{(i)}:1\leq i\leq N}$, originally from $N$ independent copies of a BI-DMC.

The $N$ generated channels are polarized and have mutual information either close to 0 (i.e. noisy channels) or close to 1 (i.e. noiseless channels).

The synthesized channels get perfectly noisy/noiseless as $N$ approaches infinity.

The process of channel polarization consists mainly of two phases:

$\bf{\text{1. Channel Combining:}}$

In this phase, $N$ distinct channels are created in $n=\log_2\left(N\right)$ steps, through recursively combining $N$ copies of a B-DMC to form a vector channel $\mathrm{W}_{N}:X^{N}\rightarrow Y^{N}$, where $N$ can be any power of two, $N=2^{n}\,,\,n\geq0$.

$\bf{\text{2. Channel Splitting:}}$

In this phase, the channel $\mathrm{W}_{N}$ is splitted into $N$ binary-input channels $\mathrm{W}_{N}^{(i)}:X\rightarrow Y^{N}\times X^{i-1},1\leq i\leq N$.

Uncoded information bits are transmitted over the reliable (i.e. noiseless) channels with rate 1 and frozen (known) bits are transmitted over the unreliable (i.e. noisy) channels.