Power-Aware Deep Unfolding Learned Optimization for Modular Receive Beamforming

This repository contains the source code for simulations and experiments presented in the paper Rapid and Power-Aware Learned Optimization for Modular Receive Beamforming.

Paper Link

Access the full paper on ArXiv: arxiv.org/2408.00439

Overview

This repository includes the code and results for the experimental study described in Section V of the paper.

Some simulations require external data, specifically generated using the QuDRiGA channel generator. The relevant data file, H_1100_32_12_5, is provided as a MATLAB file, representing the matrix H with:

• 1100 samples of 32 frequency bins
• MIMO channel realizations with 12 antennas and 5 users

Note: In our simulations, only 4 of the 32 frequency bins were used.

Contents

The simulations are structured into multiple Jupyter Notebooks, each corresponding to different sections of the study:

1. Notebook 1 - Unconstrained Simulations

• Contains simulations for Scenario 1 and Scenario 2.

2. Notebook 2 - Model Training & Transfer

• Example of training the model on Scenario 4 and evaluating on a different scenario with more users (Scenario 3).
• Includes a parameter comparison between the CNN network and the unfolded network.

3. Notebook 3 - Model Training on Different Antenna Configurations

• Example of training on Scenario 5 and testing on a different configuration with more antennas and panels (Scenario 6).

4. Notebook 4 - Power-Aware Sparse Beamforming

• Demonstrates the power-aware implementation of sparse beamforming.

5. Notebook 5 - Power-Aware Quantized Beamforming

• Shows results of quantized low-resolution beamforming using our algorithms.

6. Notebook 6 - Runtime Analysis

• Provides an analysis of the running time of different benchmarks, specifically on Scenario 2.

External Data Requirements

For some simulations, you will need the QuDRiGA-generated channel data. The dataset is provided as H_1100_32_12_5.mat. Ensure that the file is placed in the appropriate directory before running the simulations.

Citation

If you use this code or data in your research, please cite our paper:

@misc{levy2024rapidpowerawarelearnedoptimization,
      title={Rapid and Power-Aware Learned Optimization for Modular Receive Beamforming}, 
      author={Ohad Levy and Nir Shlezinger},
      year={2024},
      eprint={2408.00439},
      archivePrefix={arXiv},
      primaryClass={eess.SP},
      url={https://arxiv.org/abs/2408.00439}, 
}