Repository files navigation

Source code of the paper

B. Fesl, M. Baur, F. Strasser, M. Joham, and W. Utschick,"Diffusion-Based Generative Prior for Low-Complexity MIMO Channel Estimation," in IEEE Wireless Communications Letters, 2024.

This letter proposes a novel channel estimator based on diffusion models (DMs), one of the currently top-rated generative models, with provable convergence to the mean square error (MSE)-optimal estimator. A lightweight convolutional neural network (CNN) with positional embedding of the signal-to-noise ratio (SNR) information is designed to learn the channel distribution in the sparse angular domain. Combinedwith an estimation strategy that avoids stochastic resampling and truncates reverse diffusion steps that account for lower SNR than the given pilot observation, the resulting DM estimator unifies low complexity and memory overhead. Numerical results exhibitbetter performance than state-of-the-art estimators.

The code is tested withPython 3.10 andPytorch 2.1.1. For further details, seeenvironment.yml.

1. Load channel data fromhttps://syncandshare.lrz.de/getlink/fi93y1AnwmsvHrAGNqq5zX/ (password: Diffusion2024) and move it into folderbin.

2. To evaluate the pre-trained models used for the plots in the paper, run

python load_and_eval_dm.py -d cuda:0

3. To train a DM from scatch and evaluate the performance afterward, run

python diff_cnn.py -d cuda:0

4. To evaluate the baseline estimators, run

python baselines.py

The code is based on the implementation ofhttps://github.com/benediktfesl/Diffusion_MSE.