lynshao / gbfac_lite

GBFAC_lite

The source code of GBAFC with enhaced feature extractor.

Design C of the enhaced feature extractor is adopted, the architecture of which is:

Linear + ReLu + Linear + ReLu + Linear

There is no belief update and successive decoding.

Well-trained models

This repo provides two well-trained models for noiseless feedback channels and three well-trained models for a noisy feedback channel of 20 dB. The results we obatined are as follows:

Noiseless feedback:

Feedforwad SNR	Feedback SNR	BLER
-1.5 dB	100 dB	1.4e-4
-1 dB	100 dB	1.3e-9
0 dB	100 dB	4.1e-11

Noisy feedback:

Feedforwad SNR	Feedback SNR	BLER
-1 dB	20 dB	2.33e-2
0 dB	20 dB	4.05e-5
1 dB	20 dB	4e-7

To reproduce the results, please run

python main.py --snr1 [input] --snr2 [input] --train 0 --batchSize 100000