The
constellation optimization method for terahertz non-ideal channel based on
deep learning belongs to the field of terahertz communication.The method of the present application is as follows: a
transmitter based on DFT-s-OFDM modulation is designed to generate a transmission
signal; a communication
system model for terahertz non-ideal channel is established to obtain a received symbol sequence after the transmission
signal is transmitted through the channel and processed by a
receiver; a
constellation optimization model based on a multi-layer neural network is constructed, a joint
loss function containing three types of constraint terms is designed,
adaptive optimization training is performed on
constellation symbols, and an optimized constellation
point sequence suitable for terahertz non-ideal channel is obtained; the optimized constellation points output by the multi-layer neural network are screened and constrained, projected and mapped to a standard
QAM constellation to obtain a final
constellation diagram satisfying the standard modulation format.The present application can reduce the
symbol error rate under the condition of terahertz non-ideal channel, improve the robustness of the
system to non-linear
distortion and
phase noise, and simultaneously consider power efficiency and implementation complexity.