Energy-Efficient Transmissions in Federated Learning-Assisted Cognitive Radio Networks
M. C. Hlophe, Bodhaswar T Maharaj, Malcolm M. Sande · 2021 IEEE 21st International Conference on Communication Technology (ICCT) · 2021
The estimation of critical parameters and catching the relationship between allocated resources, link reliability, and transmission latency to save communication resources for data transmission while avoiding data privacy issues will be realized in beyond 5G networks. Achieving this in cognitive radio networks (CRNs) requires the combination of autonomous realtime spectrum awareness and federated algorithms to improve spectral efficiency (SE) and data privacy, respectively. However, incorporating federated intelligence in CRNs requires sacrificing SE for data security. This paper analyzes the effects of incorporating federated machine learning (FML) in the operation of CRNs, which mandates the use of computation and communication resources to evaluate the implication of FML on the energy consumption economics. To achieve this, each cognitive user trains a local model using a multi-layer perceptron model and uploads an update to the serving base station, which in turn aggregates a global model and broadcasts the update to its associated cognitive users. A closed-form energy cost was derived based on time and bandwidth allocation, and the results show that when properly weighted, FML reduces the computation and communication time for cognitive users compared to noncognitive users.