A Geometry-based Stochastic Wireless Channel Model using Channel Images

In this work, we obtain channel parameters from the ray-tracing simulation in a specific area and process them in the form of images to train any generative model.

For the detailed descriptions of this implementation, we kindly suggest to read the paper, A Geometry-based Stochastic Wireless Channel Model using Channel Images

Extended decription -> Geometry-based Stochastic Wireless Channel Model using Generative Neural Networks

Channel Image Generation and Model Training

Firstly, we obtain channel parameters from ray-tracing simulate and create the matrices $\boldsymbol{D}$ corresponding to each Tx-Rx link.

After then, we normalize the matrices with Min-Max scaling and create images by enlarging the matrices horizontally and vertically.

Next, we train WGAN-GP with the channel images. The followings show the output images at each epoch of training.

After training the WGAN-GP, we sample data from the outputs of the model.

Performance Evaluation

CDFs of pathloss and delay obtained from the trained model are compared with those from the original data.

In addition, we compare the link state probabilities (LOS and outages).