I wrote a matlab code for obtaining BER of ldpc codes using sum product algorithm. But its BER performance is relatively close to uncoded BPSK. What can I do to change this? What should be the dimension of parity check matrix, number of iterations???

Questions: Range for noise power N0 when bit energy=1 using BPSK modulation? No'of iterations required for sum product algorithm for different parity check matrix? No'of frames required or total length of codeword required to have a coding gain compared with uncoded BPSK?

[OUDOMSACK PIERRE PASQUERO]

I also implemented a LDPC decoder based on Min-Sum algorithm, for 802.11ac in C language. The obtained BER is good, better than that of BCC decoder.

Then, I implemented the Sum-Product algorithm (a couple of lines to modify in my code). I have the same problem as you. The obtained performance is close to the case there is no channel coding.

After a deep analysis, I found where the problem comes from. The check-nodes messages values are very low compared to LLR. Without explaining to much in detail, it is due to the function f(x) = ln((e^x+1)/(e^x-1))... I don't manage to resolve the problem.

Have you found a solution ?

[Aitor López Hernández]

Hello,

I am currently having the same issue. Were you finally able to solve it? If so, how?

Answer 1

[Tuong Nguyen Minh]

You might consider the log sum exp trick

https://stats.stackexchange.com/questions/105602/example-of-how-the-log-sum-exp-trick-works-in-naive-bayes

when seeing this kind of function pop up f(x) = ln((e^x+1)/(e^x-1))