Cache Update Algorithms for Multicasting Without File Splitting

Coded caching reduces peak transmission rates on shared wireless links by pre-storing content at user terminals during low-traffic periods and using multicasting during peak hours. However, most current methods require dividing each file into many segments and sending numerous messages, making these methods impractical for large systems due to complex book-keeping and sub-packetization. This work proposes and evaluates three decentralized algorithms for cache updating using intercepted messages: Max-Pairs, which maximizes expected pairings; a deep-Q-learning (DQL) algorithm for selfish users; and a reinforcement learning scheme that balances service times. These algorithms assume that files are cached as whole units and an uninterrupted stream of user requests. Simulations show that our approaches outperform Uniform Caching and Least Frequently Used (LFU) policies.