Abstract

After certain training epochs of the deep learning decoders, the binary cross entropy (BCE) loss function ignores the training of some unsuccessful decoded information bits, thereby degrading the training efficiency. In this letter, we propose the negative bit error rate (NBER) loss function to increase the training concentration degree on the unsuccessful decoded information bits by modifying the gradient on different information bits, so that the training efficiency and decoding performance can be improved. The simulation results show that NBER could achieve significant performance improvement over BCE both in the deep neural network decoders and the weighted belief propagation decoders.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call