Skip to content

Joint Deep Reinforcement Learning and Unfolding: Beam Selection and Precoding for mmWave Multiuser MIMO With Lens Arrays

Notifications You must be signed in to change notification settings

hqyyqh888/DDQN_BeamSelection

Repository files navigation

DDQN_BeamSelection

Joint Deep Reinforcement Learning and Unfolding: Beam Selection and Precoding for mmWave Multiuser MIMO With Lens Arrays

This repository contains the entire code for our work "Joint deep reinforcement learning and unfolding: Beam selection and precoding for mmWave multiuser MIMO with lens arrays", available at: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9448095 and has been accepted for publication in IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS (JSAC).

For any reproduce, further research or development, please kindly cite our JSAC Journal paper:

Q. Hu, Y. Liu, Y. Cai, G. Yu, and Z. Ding, “Joint deep reinforcement learning and unfolding: Beam selection and precoding for mmWave multiuser MIMO with lens arrays,” IEEE J. Sel. Areas Commun., vol. 39, no. 8, pp. 2289–2304, Aug. 2021.

Requirements

The following versions have been tested: Python 3.6 + Pytorch 1.9.0. But newer versions should also be fine.

Training and Testing

Run the main program "joint_trainer.py.py".

The introduction of each file

joint_trainer.py:main function,run this file to train jointly the DDQN and deep-unfolding neural network;

Config.py: System parameters;

DDQN:

Net_module.py:The architecture (dimension of trainable parameters) of DDQN;

Dueling_DDQN.py: The dueling architecture of DDQN;

Base_Agent.py: Define the class of agent that contains the basic functions of the DQN;

DQN.py: Define the class of DQN and it inherits from the Base_Agent;

DDQN.py: The class that inherits from the DQN and add the function of DDQN;

my_DQN.py: The class that inherits from the DQN and add some functions to deal with our problem, e.g., the computation of the reward function;

Replay_Buffer.py: The replay buffer of the DDQN;

The folder data_structures and exploration_strategies denote the data structure and exploration strategies (e.g., noise net), respectively;

Deep-unfolding:

WMMSE.py: Iterative WMMSE algorithm for digital precoding;

unfolder.py & model.py:Deep-unfolding neural network for digital precoding (unfold the iterative WMMSE algorithm);

complex_matrix.py: Some complex matrix operations, e.g., the matrix inversion and determinant of complex matrix;

Beamspace_channel.py:Beamspace channel model.

Releases

No releases published

Packages

No packages published

Languages