Repository containing the codes developed while producing my master's thesis, with the title Using Deep Learning Techniques to Optimize Access Demand for MMTC Traffic. The .mat files contain stored data. Specifically, the ones in this repository contain data for the trained Deep Q-Network (DQN) agents number 7501 and 7504, which correspond to the ones with the highest reward and average reward (on a 5-episode window), respectively, according to our 1e4-episode training. The .mat files corresponding to the trained neural networks (NNs) are stored in a Google Drive folder (https://drive.google.com/drive/folders/1m4HQ-DcVsMBgddrx7haMwFdN3dtZfg9c?usp=sharing) because of Github's file size limitations.

All files are written in MATLAB R2021b.

• comparative_analysis: this script is the one generating the results and graphics presented in Section 3.5. This script uses other subroutines, functions and stored data which are listed below:
  • epi_sim_estimated_action: function that generates measurements collected throughout an extended 4000-RAO episode
  • single_RAO_loop: function generating a single execution of the ACB and CBRA protocols
  • Agent7501.mat: trained DQN agent number 7501
  • Agent7504.mat: trained DQN agent number 7504
  • trained_net_21_1_1_1 [in the Google Drive folder]: first NN listed in Table 5
  • trained_net_21_16_1_2 [in the Google Drive folder]: second NN listed in Table 5
  • trained_net_21_1_1_1_Max_10 [in the Google Drive folder]: third NN listed in Table 5
  • trained_net_21_16_1_2_Max_10 [in the Google Drive folder]: fourth NN listed in Table 5