GW2S

Code for exploring generalization in deep learning for mmWave beam selection using sub-6 GHz channels.

This repository contains MATLAB experiment scripts for training and evaluating a neural network that predicts mmWave beam indices from sub-6 GHz channel measurements. The workflow is designed around wireless communication experiments where sub-6 GHz channel state information is transformed into an angle-delay profile and used as the input to a beam-selection classifier.

Related Publication

This repository is associated with the PARAMOUNT paper:

Katarina Vuckovic, Mahdi Boloursaz Mashhadi, Farzam Hejazi, Nazanin Rahnavard, and Ahmed Alkhateeb. "PARAMOUNT: Toward Generalizable Deep Learning for mmWave Beam Selection Using Sub-6 GHz Channel Measurements." IEEE Transactions on Wireless Communications, vol. 23, no. 5, pp. 5187-5202, 2024.

• Paper DOI: https://doi.org/10.1109/TWC.2023.3324916
• Author version PDF: https://lcwnlab.eecs.ucf.edu/wp-content/uploads/2023/12/PARAMOUNT_IEEE.pdf

Research Focus

The experiment studies whether a deep learning model trained on sub-6 GHz channel features can generalize to mmWave beam-selection tasks under different SNR and transmit-power settings.

The pipeline measures:

• Top-1, top-3, and top-5 beam prediction accuracy
• Average achievable rate for predicted beams
• Upper-bound achievable rate from the best beam
• Performance across multiple transmit-power and SNR values

Repository Structure

File	Purpose
main.m	Main script for training, testing, and saving experiment results
dataPrep_train.m	Loads, normalizes, shuffles, and labels training data
dataPrep_test.m	Prepares test data for evaluation
buildNetconv4.m	Builds the neural network used for beam classification
CSI2ADP_theta_N.m	Converts channel state information to angle-delay profile features
UPA_codebook_generator.m	Generates the beamforming codebook
LICENSE	Repository license

Expected Data Layout

The main script expects .mat channel datasets in this structure:

Data/
  train/
    sub6Train_org_4_64.mat
    mmTrain_org_64_64.mat
  test/
    sub6test_LOSB_4_64.mat
    mmtest_LOSB_4_64.mat

Each dataset should include a channel variable. For NLOS cases, labels may also be required.

Requirements

• MATLAB
• Deep Learning Toolbox
• Communications Toolbox, for awgn
• Wireless channel datasets matching the expected dimensions in main.m

Running

From the MATLAB console:

main

Before running, update the paths and experiment settings in main.m if your dataset names or dimensions differ from the defaults.

Key Settings

Important configuration values live near the top of main.m:

• tx_power: transmit-power sweep
• snr_db: SNR sweep
• num_ant: number of sub-6 GHz antennas
• num_ant_mm: number of mmWave antennas
• options.case: channel case, such as LOS
• options.top_n: number of top beams used for top-N evaluation

Notes

This is a research codebase, so the scripts assume a prepared dataset and MATLAB environment. For a new experiment, first verify data dimensions, codebook size, and the options.inputSize settings.