Privacy Protection and Efficient Incumbent Detection in Spectrum Sharing Based on Federated Learning

Spectrum sharing techniques can greatly promote spectrum efficiency and mitigate the congested wireless spectrum. During spectrum sharing, the lower access users, e.g., commercial users, must protect the privilege of the incumbent users, e.g., navy ship radar operators. To achieve this task, environment sensing capability (ESC)-based incumbent detection is a commonly used method, where multiple ESC nodes with a fixed geo-location sense the spectrum and upload the spectrum events to the spectrum access system (SAS). However, existing ESC-based methods may incur a high communication overhead and lead to the leakage of sensitive information, e.g., navy ship route information. To tackle these issues, in this paper, we propose a compressed sensing (CS)-based federated learning framework to achieve incumbent user detection for improving communication efficiency while protecting the privacy of training samples. In particular, the local learning models transmit the updating parameters instead of the raw spectrum data to the central server, and these parameters are aggregated based on a multiple measurement vector (MMV) CS model. The central server can gain a global learning model based on the aggregation of the parameters and get the updating of global parameters back to the local learning models to achieve federated learning. The security analysis and simulation results are provided to validate the effectiveness of the proposed schemes. In the proposed CS-based federated learning framework, the detection performance is as good as the scheme under the raw training samples, and the communication and training efficiency can be significantly improved.

I. Introduction

Recently, spectrum sharing has received great attention for mitigating the issue of the massive connectivity in the congested wireless spectrum and promoting dynamic access to spectrum resources. For example, according to a statement of the Federal Communications Commission (FCC), the citizen broadband radio service (CBRS) bands between 3550MHz to 3700MHz, was opened for cellular carriers and opportunistic spectrum sharing [1],[2]. These CBRS bands originally occupied by authorized federal and grandfathered fixed satellite services users (i.e., incumbent access users), will be shared with commercial users and other licensed occupants (i.e., lower access users). In the meantime, the spectrum access system (SAS) that manages spectrum access must protect the highest access privilege of the incumbent users in CBRS bands to prevent harmful interference from the lower access users. To achieve this task, environment sensing capability (ESC) is employed to achieve incumbent user detection for spectrum sharing, where multiple ESC nodes at a fixed geo-location sense the spectrum from its location and report the spectrum event to SAS [3],[4].