Reinforcement Learning for Bit-Flipping Decoding of Polar Codes

. 2021 Jan 30;23(2):171. doi: 10.3390/e23020171

Algorithm 1 Optimized design of action space

Input:

The number of trials:

T

Information bits length:

K

Discrete SNRs:

s n r

Output: The action space:

A

Initialization:

A \leftarrow 0

;

f r a m e \leftarrow 1

;

For

s

s n r

S p a c e_{1}^{K} \leftarrow 0

;
While

f r a m e < T

:
Updating the received LLR:

y_{1}^{N}

;

F E b i t \leftarrow N u l l

;
//The

F E b i t

is the first error bit index in SC decoding.

[{\hat{u}}_{1}^{N}, F E b i t] \leftarrow S C d e c o d e r (y_{1}^{N}, K, s, F E b i t)

;
If

F E b i t

N u l l

: Continue; else:

S p a c e [F E b i t] + = 1

;

f r a m e + = 1

;
End if
End while

S p a c e = S p a c e / T

;
For

i

K

:
If

S p a c e [i] > θ

:
Move

i

into

A

;
End if
End for
End for