In the testbench precision is taken from the SNR (sigma_noise), but really, it is required to take it from MER.
Thus, the real performance starts from 0.5 SNR on 50 iterations.
Yes, there are many problems in there and I haven't touched LDPC codes since playing with Polar codes. Would you like to implement the fixes and create a pull-request?
In the testbench precision is taken from the SNR (sigma_noise), but really, it is required to take it from MER. Thus, the real performance starts from 0.5 SNR on 50 iterations.