Research On Inteingent Signal Receiving Algorithm For Large-scale Antenna System

Large-scale Multiple Input Multiple Output(MIMO)systems and Orthogonal Frequency Division Multiplexing(OFDM),among other key technologies,are widely applied in various complex communication scenarios to meet the performance demands of current wireless communication technology,including high data rates,high energy efficiency,low latency,and high throughput.MIMO can significantly enhance transmission performance without increasing system bandwidth or transmit power,while OFDM can effectively combat selective fading of signals in channel propagation.However,the use of MIMO and OFDM technologies also introduces challenges that traditional communication techniques struggle to address.Firstly,multi-antenna systems heavily rely on the effective acquisition of Channel State Information(CSI).For largescale MIMO-OFDM systems,obtaining downlink CSI at the User Equipment(UE)and transmitting it back to the Base Station(BS)through feedback links incurs significant overhead.Traditional methods quantize and compress CSI at the UE and rely on compressive sensing techniques for recovery,but the achieved performance is not satisfactory.Additionally,factors such as multipath effects,bandwidth limitations,multi-user interference,and noise in wireless communication make it more challenging to obtain accurate channel models.This,in turn,raises the hardware requirements and design complexity for traditional receivers,posing challenges for signal recovery in wireless communication receivers.Firstly,to address the problem of CSI compression feedback,this paper proposes a network called IACCsiNet based on an asymmetric convolutional structure.This algorithm not only surpasses traditional algorithms but also exceeds the current state-of-the-art deep learning-based implementations in indoor and outdoor work scenarios.To further reduce the parameter quantity of the IACCsiNet network,this paper introduces the depth-wise separable convolutional neural network algorithm to further optimize the design of IACCsiNet and proposes IACCsiNet-light.The implementation of this algorithm maintains comparable performance to IACCsiNet while reducing the parameter quantity by 22.9%,27.8%,and 30.0%when the compression ratios are 16,32,and 64,respectively,compared to the state-of-the-art low-parameter algorithm ShuffleCsiNet.Secondly,to solve the problem of channel model estimation in wireless communication scenarios,this paper proposes TransGAN based on the Transformer algorithm constructed with a generative adversarial network.This algorithm differs from traditional deep learning algorithms trained based on supervised learning.Through adversarial learning between the constructed generator and discriminator,the performance of the generator algorithm is continuously improved for channel estimation.Compared to the state-of-the-art cGAN algorithm trained based on self-supervised learning,this paper performs dimensionality reduction on multidimensional data and constructs the algorithm structure based on attention mechanisms for extracting sequence features.Experimental results demonstrate that this algorithm design significantly outperforms performance based on convolutional neural networks,fully connected neural networks,and cGAN.Finally,based on the multi-head attention mechanism algorithm,this paper considers channel estimation and signal detection in wireless receivers and proposes a purely data-driven deep learning algorithm called TDNet as the signal receiver algorithm.The experimental design comprehensively evaluates the algorithm performance of TDNet,considering peak clipping,cyclic prefix,and noise effects,demonstrating its superior performance compared to state-of-the-art deep learning algorithms and the feasibility of receiver design based on deep learning algorithms.