Cooperative Cache in Cognitive Radio Networks: A Heterogeneous Multi-Agent Learning Approach

Deploying distributed cache in cognitive radio networks (CRNs), which spreads popular contents to the edge of network during off-peak time through spectrum sharing, can reduce deliver delay users nearby without causing severe interference primary network. However, due un-predicable requirement as well band occupation users, it is non-trivial optimize storage and fetching strategy dynamically. The letter proposes a heterogeneous multi-agent deep deterministic policy gradient (MADDPG) approach, takes servers two different types agents learn cooperation competition for mutual benefits. numeral simulation demonstrates that comparing with other single or homogeneous reinforcement learning (DRL) approaches, proposed MADDPG further delivery enhance efficiency SBSs.