A CGAN-LSTM-Based Framework for Time-Varying Non-Stationary Channel Modeling

TL;DR

This work tackles the challenge of modeling long-term, time-varying non-stationary wireless channels by introducing a hybrid CGAN-LSTM framework that generates PDP-based channel realizations with preserved temporal evolution. A stationarity constraint and TPCC-based supervision guide the generator to produce temporally coherent sequences that closely match ground-truth statistics, beyond what short-horizon models can achieve. The method is validated on a RT-based V2V dataset, showing close alignment with ground-truth distributions for path loss, shadow fading, WSS intervals, RMSDS, and multipath count, and yielding the lowest Fréchet Inception Distance among competitive baselines. The approach extends available channel measurement data, providing high-fidelity long-term channel realizations for system design, evaluation, and training of robust communication strategies in highly dynamic environments. $L_{gen}=\lambda_1 L_G + \lambda_2 L_{linear} + \lambda_3 L_{TPCC}$, $L_{D} = -\dfrac{1}{2} ( E_{P \sim p_{real}}[\log D(P|y)] + E_{z}[\log(1 - D(G(z|y), y))] )$, and $L_{TPCC}$-based constraints underpin the learning objective and physical realism of the generated channels.

Abstract

Time-varying non-stationary channels, with complex dynamic variations and temporal evolution characteristics, have significant challenges in channel modeling and communication system performance evaluation. Most existing methods of time-varying channel modeling focus on predicting channel state at a given moment or simulating short-term channel fluctuations, which are unable to capture the long-term evolution of the channel. This paper emphasizes the generation of long-term dynamic channel to fully capture evolution of non-stationary channel properties. The generated channel not only reflects temporal dynamics but also ensures consistent stationarity. We propose a hybrid deep learning framework that combines conditional generative adversarial networks (CGAN) with long short-term memory (LSTM) networks. A stationarity-constrained approach is designed to ensure temporal correlation of the generated time-series channel. This method can generate channel with required temporal non-stationarity. The model is validated by comparing channel statistical features, and the results show that the generated channel is in good agreement with raw channel and provides good performance in terms of non-stationarity.

Figures (7)

• Figure 1: Network structure of the CGAN-LSTM model with temporal stationarity constraints.
• Figure 2: Examples of RT simulation scenario layouts: (a) one of the sparse scenario layouts, (b) one of the dense scenario layouts.
• Figure 3: Examples of RT simulated PDPs and the generated PDPs based on the proposed CGAN-LSTM in sparse and dense scenarios: (a) RT simulated the PDP in sparse scenarios, (b) CGAN-LSTM generated PDP in sparse scenarios, (c) RT simulated PDP in dense scenarios, and (d) CGAN-LSTM generated PDP in dense scenarios.
• Figure 4: Comparison of the distribution of fundamental statistical characteristics in sparse and dense scenarios:(a) shadow fading, and (b) path loss.
• Figure 5: Examples of RT simulated WSS intervals and the generated WSS intervals with the proposed CGAN-LSTM-SC in sparse and dense scenarios: (a) RT simulated WSS intervals in sparse scenarios, (b) CGAN-LSTM-SC generated WSS intervals in sparse scenarios, (c) RT simulated WSS intervals in dense scenarios, and (d) CGAN-LSTM-SC generated WSS intervals in dense scenarios.