Dataset associated with the conference paper "A Novel MADDPG Algorithm for Efficient Power Allocation in Cell-Free Massive MIMO"

Description This repository contains a dataset that represents the simulation of 4 distinct cell-free massive MIMO scenarios. Each scenario includes 21,000 setups, derived from a customized version of the MATLAB code package available here. The simulations take into account three power allocation optimization schemes: max-min Spectral Efficiency (SE) fairness, sum SE maximization, and Fractional Power Allocation (FPA). In addition, two different types of user equipments (UEs) are considered: static UEs (fixed_UEs) and dynamic UEs (moving_UEs). The results for the simulation for each scenario are split across 200 MATLAB storage files ('train_part#.mat') for the training part, and accros 10 files ('test_part#.mat') for the testing part, with each file encompassing 100 setups. Denoting the number of access points (APs) as L, the numer of user equipments (UEs ) as K, and the number of setups as M. The variables within these files are defined as follows: 'gainOverNoise': This is a [L x K x M] matrix that contains the large-scale fading coefficients. 'rhokl_DL_LPMMSE_X': This is a [L x K x M] matrix that includes the power values with; (1) X = 'equal' when the maximum power is distributed equally among all the UEs, (2) X = 'maxmin' for the max-min SE fairness, (3) X = 'sumSE' for the sum SE maximization, and (4) X = 'FPA' for the FPA power scheme. 'SE_DL_LPMMSE_X': This is a [K x M] matrix that includes the SE per UE with; (1) X = 'equal' when the maximum power is distributed equally among all the UEs, (2) X = 'maxmin' for the max-min SE fairness, (3) X = 'sumSE' for the sum SE maximization, and (4) X = 'FPA' for the FPA power scheme. 'bk_dist': This is a [L x K x M] matrix that represents the average channel gain of the desired signal. 'ck_dist': This is a [L x L x K x K x M] matrix that represents the average channel gains for the respective interfering signals. 'D_dist': This is a [L x K x M] matrix where (l,k,m) is one if AP l serves UE k in setup n and zero otherwise. Additonally, the file 'setup.mat' contains information about the simulated environment. Please refer to the conference paper for more detailed information about the simulations. The simulation code is available here on GitHub for training and testing the MADDPG model. Acknowledgements This work is supported by the Spanish ministry of economic affairs and digital transformation and the European Union - NextGenerationEU [UNICO I+D 6G/INSIGNIA] (TSI-064200-2022-006).