Deep Reinforcement Learning-based Distributed Dynamic Spectrum Access in Multi-user Multi-channel Cognitive Radio Internet of Things Networks
author={Zhang, Xiaohui; Chen, Ze; Zhang, Yinghui; Liu, Yang; Jin, Minglu; Qiu, Tianshuang} journal= {IEEE Internet of Things Journal}
This project includes: [DSA_env.py]: Define two classes, DSA_Markov and DSA_Period, which model different scenarios of DSA in a wireless communication environment. These classes are part of a simulation for evaluating the performance of Spectrum Sharing strategies. [PRDDQN.py]: The setup and training process of the PRD-DQN with prioritized experience replay using Echo State Networks. [pyESN_online.py]: The RNN structure for ESN.
[main.py]: Run this file, inclduing setting system parameters. Firstly, initialize the environment, including number of channel and su. Secondly, create a new folder in the result folder according to the environment configuration. For example, when the number of channels is 12 and the number of su is 7, you can create a folder named "channel_12_su_7_1" in the result folder to store the training data. Finally, adjust the save path through the code "file_folder = '.\result\channel_%d_su_%d_' % (n_channel, n_su)". Note that you only need to change the "" according to the folder created in the previous step. For example, according to the previous step, the code can be modified to "file_folder = '.\result\channel_%d_su_%d_1' % (n_channel, n_su)".
-Tensorflow 1.x. or Tensorflow 2.x. -numpy -matplotlib The versions of Python and libraries used in our work: Python = 3.7, Tensorflow = 2.7.0, numpy = 1.21.6, matplotlib = 3.5.3.
The experimental results of this article are based on Monte Carlo simulation results generated from random data, and the simulation data is presented in our paper "Deep Reinforcement Learning-based Distributed Dynamic Spectrum Access in Multi-user Multi-channel Cognitive Radio Internet of Things Networks".