This repository is the implementation of "TinyQMIX", which is a cooperative MADRL policy for channel selection in mMTC networks.
Figure 1. mMTC devices uplink transmission model
We compare it with different static, tabular Q-learning, and deep Q-learning policies for distributed channel selection methods.
Figure 2. Moving average of the access delay. Traffic dynamic changes every 10 seconds
Over 5 minutes of testing trace, TinyQMIX has the lowest delay, approaching the empirical lower-bound WFLB.
This is the repository for the paper "TinyQMIX: Distributed Access Control for mMTC via Multi-agent Reinforcement Learning" - presented at VTC Fall 2022.
Contact: lethanh@nii.ac.jp