Resource Optimization for Signal Recognition in Satellite MEC with Federated Learning

Currently, limited resources and data privacy have influenced the development of signal modulation recognition in satellite communications. To solve the above issues, mobile edge computing (MEC) and federated learning (FL) are considered to signal modulation recognition on satellite communications in this paper. FL enables distributed learning with local datasets and allows model parameters instead of the whole training datasets being shared during learning process. MEC technology reduces transmission delay and energy consumption via setting up edge servers on Medium Earth Orbit (MEO) satellites close to Low Earth Orbit (LEO) satellites, High Attitude Platforms (HAPs) and Unmanned Aerial Vehicles (UAVs) replacing central server on Geostationary Earth Orbit (GEO) satellites. To accelerate the learning process and improve the resource allocation strategy simultaneously, we formulate a joint delay ratio and energy efficiency optimization problem. We introduce a Q-learning based algorithm to solve the problem. The experimental results indicate that the accuracy of the proposed scheme is almost the same as the scheme without resource optimization while the required resource is 10% less than the classical algorithms. Meanwhile, the computation complexity of the Q-learning based algorithm is
$O(n^{2})$
, much lower than the ergodic scheme
$(O(n^{3}))$
.