Skip to content

This repository contains the Matlab code used to generate the results in the paper “Enhancement of a state-of-the-art RL-based detection algorithm for Massive MIMO radars” https://ieeexplore.ieee.org/abstract/document/9760145

License

Notifications You must be signed in to change notification settings

lisifra96/Improved_RL_algorithm_mMIMO_radar

Repository files navigation

Reinforcement Learning based algorithm for massive MIMO radar with adaptive parameters

The present repository contains the Matlab code used to simulate a RL-based algorithm for a massive MIMO radar. The RL-based algorithm shapes the transmitted beampattern by selecting the weighting matrix of the transmitted waveforms based on the position of the detected targets. The ε and α parameters can be adaptively selected (default option). The quasi ε-greedy policy with target recovery is the default one. The user can compare the performance of the algorithm over one parameter, such as ε, α or the policy, while keeping all the others constant.

Offline mode

To speed up the algorithm the user can compute the W matrices that are solutions of the optimization problem associated with all the possible Omega sets and store them in a cube (Wcube). Then the algorithm doesn't have to solve the optimization problem at each iteration but can select the correct W matrix from the cube. This mode of operation is called "offline" mode. Before running the code in "offline" mode the user has to run the W_opt_generator script to generate the cube.

UPDATE

The optimization problem in eq. (19) of the paper [1] and in eq. (6) of the paper [2] can be solved in closed form by using the procedure discussed in Sec. III.B of [3]. An extensive analysis of this point will be proposed in a paper in preparation.

The related Matlab function is Closed_Form_W.m. The files W_opt_generator.m and getWfromTargetIndexes_online have been updated accordingly. The updated MonteCarlo_online.m script exploits the parfor loop and the closed-form solution of the optimization problem to reduce the computational time. As a consequence, we recommend using the "online" mode. The updated version of the code does not require the user to install cvx and Mosek anymore

The function Alg2v1.m that solves the optimization problem using the cvx package is not used in the current version of the algorithm. If a user wants to change the optimization function and use cvx, he can update the objective function in Alg2v1.m and update the functions W_opt_generator.m and getWfromTargetIndexes_online accordingly.

[1] A. M. Ahmed, A. A. Ahmad, S. Fortunati, A. Sezgin, M. S. Greco and F. Gini, "A Reinforcement Learning Based Approach for Multitarget Detection in Massive MIMO Radar," in IEEE Transactions on Aerospace and Electronic Systems, vol. 57, no. 5, pp. 2622-2636, Oct. 2021.

[2] F. Lisi, S. Fortunati, M. S. Greco and F. Gini, "Enhancement of a State-of-the-Art RL-Based Detection Algorithm for Massive MIMO Radars," in IEEE Transactions on Aerospace and Electronic Systems, vol. 58, no. 6, pp. 5925-5931, Dec. 2022.

[3] P. Stoica, J. Li and Y. Xie, "On Probing Signal Design For MIMO Radar," in IEEE Transactions on Signal Processing, vol. 55, no. 8, pp. 4151-4161, Aug. 2007.

About

This repository contains the Matlab code used to generate the results in the paper “Enhancement of a state-of-the-art RL-based detection algorithm for Massive MIMO radars” https://ieeexplore.ieee.org/abstract/document/9760145

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages