Power efficient and low latency baseband processing for LTE femtocell

Minimize power while maintaining network QoS, about 2.5ms per frame

Network scheduling provides expected uplink network load for 5ms, and buffer provides current state. Use this information to set the baseband processor state.

There are 10^18 possible workloads, sample from this and perform reinforcement learning to find a good policy.

A policy is a lookup table from state to optimal action, use clustering to reduce the state space to 101 * 11. Action space is operating frequency {2,3,..,20} x number of active cores {1,2,3,4}