If you find this algorithm useful, please cite the following paper. Thanks.
M. Rowshan, A. Burg and E. Viterbo, "Polarization-Adjusted Convolutional (PAC) Codes: Sequential Decoding vs List Decoding," in IEEE Transactions on Vehicular Technology, vol. 70, no. 2, pp. 1434-1447, Feb. 2021, doi: 10.1109/TVT.2021.3052550.
https://ieeexplore.ieee.org/abstract/document/9328621
Description: This is an implementation of the successive cancellation list (SCL) decoding algorithm for polar codes, CRC-polar codes, and PAC codes with the choice of various code constructions/rate-profiles in Python. The list decoding algorithm is an adaptive two stage successive cancellation list (SCL) algorithm. That means first it tries L=1 and then L=L_max. The performance is the same as list decoding with L_max. This trick has been implemented in the simulator.py file. The rest of the files are the same as the standard list decoding algorithm.
The main file is simulator.py in which you can set the parameters of the code and the channel.
To switch between decoding polar codes and PAC codes, you need to change the generator polynomial conv_gen to conv_gen=[1] for polar codes or any other polynomial such as conv_gen=[1,0,1,1,0,1,1].
- the encoding process where we have one more stage that we call convolutional precoding or pre-transformation. If conv_gen = [1], that means no precoding is performed, hence the result will be the simulation for polar codes, not PAC codes. If you look at pac_encode() method in polar_code.py file, you would find the convolutional precoding method, conv_encode(), as U = pcfun.conv_encode(V, conv_gen, mem).
- the decoding process where we consider both 0 and 1 values for every v and obtain the corresponding u by conv_1bit() method based on the current state. Then, we calculate the path metric based on the LLR and these u values (the values 0 and 1 of each extended branch on the tree). Obviously, we need to update the current state as well by getNextState() method.
Note that the "copy on write" or "lazy copy" technique has been used in this algorithm.
Please report any bugs to mrowshan at ieee dot org