Research On Low Resolution OFDM Receiver Based On Neural Network

Orthogonal Frequency Division Multiplexing(OFDM)has been widely used because of its high frequency utilization.However,for traditional OFDM receivers,due to the adoption of high resolution Analog-to-Digital Converter(ADC),the whole system has a large power consumption.Therefore,in order to reduce power consumption,it is a more direct scheme to use a low resolution ADC in OFDM system to quantify the received signal.But,the application of low resolution ADC will have a great impact on the Bit Error Rate(BER)performance of the whole system.So,how to design a suitable low resolution OFDM receiver,which can carry out accurate channel estimation and signal detection,so as to eliminate the impact of the quantization caused become the main research problem of this paper.Firstly,this paper proposed an one-bit OFDM receiver scheme(AE-OFDM)based on the deep neural network auto-encoder structure.The end-to-end OFDM communication system is regarded as an auto-encoder.Through two-stage training,network parameters are constantly adjusted until the receiver achieves the best detection performance.Then,a large number of simulation experiments are carried out to verify the accuracy of the theoretical analysis.The advantages of the deep neural network structure in channel estimation and signal detection are proved.Then,the limitations of the autoencoder structure are considered.In order to further improve performance,this paper proposed an one-bit OFDM receiver scheme based on model-driven deep learning,it is quite different from AE-OFDM scheme.Including channel estimation and signal detection module is constructed from the depth are based on neural network,at the same time use the LS/ZF traditional algorithm on the network's input data such as initialization processing.Finally,the advantages of the proposed scheme in BER performance and algorithm complexity are verified by simulation experiments.The above two receiver schemes based on deep learning have complex network structure and many parameters that need to be trained to adjust,resulting in high computational complexity.Therefore,this paper also proposed a design scheme of one-bit OFDM receiver(FCELM-OFDM)based on full complex extreme learning machine algorithm.This scheme is based on the design of one-bit OFDM receiver.At the same time,it aims at the problems in deep neural network,such as high algorithm complexity,long training time and mostly off-line training,poor generalization ability and so on.Finally,by comparing computer simulation with different schemes,this paper verifies the excellent performance of this scheme in terms of algorithm complexity,generalization ability and BER performance.