Adaptive channel modeling method and system of enhanced conditional generative adversarial network

Abstract

The invention discloses an adaptive channel modeling method and system for enhancing a conditional generative adversarial network, and the method comprises the following steps: S1, initializing a generative adversarial network model, and obtaining a channel sample sequence; s2, a potential variable method is set through priori knowledge, and a potential variable sequence is obtained; s3, obtaining a reconstructed sample sequence meeting neural network processing requirements through a complex sequence reconstruction method; s4, obtaining a penalty sample sequence through a penalty sample construction method; s5, obtaining an input sequence and a generative adversarial network objective function through an enhancement condition generative adversarial network error function optimization method; and S6, performing network adversarial training, and outputting a generator model capable of simulating real channel data distribution. According to the method, the improved generative adversarial network is selected to accurately capture random channel behaviors, and self-adaptive channel modeling without manually assuming a physical model is realized.

Description

technical field [0001] The invention belongs to the technical field of digital communication, and in particular relates to an adaptive channel modeling method and system for enhancing a conditional generation confrontation network. Background technique [0002] In the field of wireless communication in recent years, channel modeling for massive multiple input multiple output (MIMO) systems has received extensive attention and research in the industry. As an important branch of artificial intelligence, Deep Learning (DL) is considered to be a powerful tool for analyzing measurement data, understanding the propagation process, and constructing nonlinear models. Channel modeling based on deep learning has the characteristics of self-adaptation, and the network structure is relatively fixed. If a wireless channel needs to be remodeled, it can be done by training it with different data. Popoola S I et al. used measured data to train a neural network (Neural Network, NN) model to...

AI Technical Summary

Problems solved by technology

[0003] Aiming at the problems of difficult parameter analysis and lack of versatility of the modeling method in the case of limited measurement data in the prior art, the present invention provides an adaptive channel modeling method and system using an enhanced conditional generative adversarial network. The present invention utilizes training The sequence and the corresponding received sequence are used as conditional information, and the Wasserstein distance is used as the measure between distributions, which can effectively improve the training stability and learning ability of GANs; use prior knowledge to explore the distribution of latent variables in the training process, and significantly enhance the proposed Simulation performance of the Enhanced Conditional Generative Adversarial Networks framework of

Images