Single carrier MIMO channel equalization algorithm based on InceptionNet

This article proposes a channel equalization neural network based on InceptionNet for the channel equalization of a single-carrier MIMO system. This scheme does not require statistical channel information and directly uses the received pilot signals for offline training. It uses improved convolution neural network extracts features so that it can learn the channel information features contained in the pilot signal and then equalize the channel through the neural network, making the model output the equalized signal. The simulation results show that compared with the traditional channel equalization method, the network model proposed in this paper has better performance in reducing the signal error rate.

I. Introduction

MIMO technology can make full use of space resources, expand channel capacity, improve data transmission rate and communication quality, and it has been widely used in various fields of communication. Channel estimation and equalization technology, as an important part of the communication field, also has a lot of Research. Traditional channel equalization techniques are divided into linear equalization and non-linear equalization. Linear equalization has a simple structure, but its performance is not ideal. Paper [1] further improved the linear equalization technology and achieved better performance. Non-linear equalization mainly includes Decision Feedback Equalization (DFE) [2] [3], Successive Interference Cancellation (SIC) [4] etc. Non-linear equalization technology can improve the equalization performance by further eliminating inter-symbol interference. However, the process of feedback and iteration also increases the complexity of the receiving algorithm. With the continuous deepening of machine learning research, machine learning has also begun to be widely used in the field of channel estimation and equalization. Compared with traditional algorithms, machine learning can better find the inherent characteristics in the data. In some specific scenarios, channel estimation and equalization methods based on machine learning have better performance. Paper [5] proposed a channel equalizer structure based on radial basis function (RBF) neural network for Bayesian estimation, which can better adapt to channel changes and accelerate system performance. Paper [6] trained a back propagation (BP) neural network to compensate the nonlinear error caused by the high-power amplifier in the MIMOM-OFDM system. Paper [7] constructed a recurrent neural network structure based on long short memory (LSTM) and feedforward neural network (FNN) to solve the channel tracking problem under the wireless energy transfer (WET) system. The CNN network is one of the classic deep learning networks, and it has also attracted a lot of attention in the field of channel estimation and equalization. The paper [8] proposed a joint estimation method for large-scale millimeter wave (mmWave) MIMO systems based on deep convolutional neural networks. This method uses the spatial correlation and frequency correlation to perform more accurate channel estimation by simultaneously inputting the preprocessed pilots of multiple adjacent subcarriers into the CNN.