Research On Detection And Estimation Schemes Based On Deep Learning

The traditional methods usually model the communication physical layer into a specific mathematical model so that the communication process can be explained to a certain extent.However,in some complex scenarios,the communication process is difficult to be accurately modeled or the established model is too complex.This increases the complexity of the algorithm and is difficult to support the constraints of low delay.In recent years,the application of data-driven deep learning technology in communication system has gradually become the current research hotspot.Deep learning technology has strong representational ability and can learn the complex relationship between data distribution from a large number of training data.Moreover,the distributed and parallel computing architecture guarantees its efficient computing and processing capacity.In this paper,deep learning technology is used to carry out in-depth research on spectrum sensing,channel estimation and signal detection technology of OFDM systems.The main research contents of this paper are as follows:1.In terms of dealing with the SNR-wall problem in traditional energy detector,this paper proposes a deep learning-based spectrum sensing algorithm without knowing the noise power and main user signal information.Firstly,this paper designs a Spectrum Sensing Network(S~2Net)to detect the received signal of a single node.When deployed online,it can directly predict the presence or absence of the signal of the main user.On this basis,this paper further proposes to use Automatic Decision Network(ADNet)to fuse the detection results of multiple distribution nodes,so as to make comprehensive decision.Relevant experiments show that the S~2Net based on deep learning proposed in this paper can obtain higher detection rate and lower false alarm rate than the traditional energy detector.Compared with the single node detection,the proposed ADNet which integrates the detection results of multiple distributed nodes can further improve the detection rate and reduce the false alarm rate.2.For the OFDM system,this paper proposes a channel estimation scheme based on deep learning.Based on the similarity between the time-frequency 2D signal in OFDM system and the 2D natural image,the channel matrix is regarded as a 2D natural image in this paper.This paper constructs a Channel Estimation Network(CENet)using an efficient convolutional neural network based on second-order attention,and estimates the channel response matrix from the pilot signals.On this basis,the influence of different loss functions for training the model is further explored.Relevant experiments show that the deep learning-based CENet can achieve better performance than the traditional channel estimation algorithm by combining the Charbonnier loss and the total variation loss as optimization function.3.For the OFDM system,this paper proposes a signal detection scheme based on deep learning.On the basis of the above research on channel estimation based on deep learning,this paper regards the signal detection task as a process of conditional signal recovery under the condition of known channel response.And a Channel Conditional Recovery Network(CCRNet)is designed to restore the transmitted signals.In the conditional fusion module of CCRNet,a bilinear residual layer based on low-rank decomposition is introduced to enhance the model characterization capability.Relevant experiments show that,by combining the channel conditioned recovery algorithm described in this paper with the aforementioned channel estimation algorithm,the joint channel and signal detection scheme based on deep learning can achieve lower bit error rate and better signal detection performance compared with other traditional schemes.