Abstract:

Aiming at the problem that the bit error rate of low density parity check (LDPC) codes increases under the condition of correlated noise, a new decoder is designed combined with traditional decoding algorithm and convolutional neural network (CNN). The decoder introduces the weighted bit-flipping (WBF) algorithm into the belief propagation (BP) algorithm to generate a weighted BP (WBP) structure to solve the problem of high bit error rate at the critical point of codeword. Then, the noise is reduced by CNN, and the received signal is processed iteratively between WBP and CNN to make the estimated value of the signal approach the real value continuously, so as to reduce the influence of relevant noise. Simulation results show that compared with BP algorithm, the proposed algorithm can effectively reduce the bit error rate of LDPC decoding under correlated noise.

Key words: weighted bit-flipping (WBF), belief propagation (BP), low density parity check (LDPC) decode, convolutional neural network (CNN)