Abstract
Increasing concerns on intelligent spectrum sensing call for efficient training and inference technologies. In this paper, we propose a novel federated learning (FL) framework, dubbed federated spectrum learning (FSL), which exploits the benefits of reconfigurable intelligent surfaces (RISs) and overcomes the unfavorable impact of deep fading channels. Distinguishingly, we endow conventional RISs with spectrum learning capabilities by leveraging a fully-trained convolutional neural network (CNN) model at each RIS controller, thereby helping the base station to cooperatively infer the users who request to participate in FL at the beginning of each training iteration. To fully exploit the potential of FL and RISs, we address three technical challenges: RISs phase shifts configuration, user-RIS association, and wireless bandwidth allocation. The resulting joint learning, wireless resource allocation, and user-RIS association design is formulated as an optimization problem whose objective is to maximize the system utility while considering the impact of FL prediction accuracy. In this context, the accuracy of FL prediction interplays with the performance of resource optimization. In particular, if the accuracy of the trained CNN model deteriorates, the performance of resource allocation worsens. The proposed FSL framework is tested by using real radio frequency (RF) traces and numerical results demonstrate its advantages in terms of spectrum prediction accuracy and system utility: a better CNN prediction accuracy and FL system utility can be achieved with a larger number of RISs and reflecting elements.
Introduction
As we transition toward 6G wireless networks, intelligent spectrum sensing—the ability to identify active frequencies and users—becomes critical for network efficiency. While Deep Learning (DL) provides powerful tools for this task, centralized training often faces significant hurdles regarding data privacy and the communication overhead of uploading massive raw datasets. Federated Learning (FL) offers a decentralized alternative by keeping data on local devices.
However, conventional FL in wireless edge networks is frequently hampered by deep fading and blocked links, which can prevent the base station (BS) from even identifying participating users. This creates a "chicken-and-egg" problem: effective resource allocation requires accurate sensing, but accurate sensing (via FL) requires reliable communication links. To address this, we propose the Federated Spectrum Learning (FSL) framework, which leverages Reconfigurable Intelligent Surfaces (RIS) to enhance both sensing and transmission.
The Federated Spectrum Learning (FSL) Framework
The FSL framework integrates RISs into the FL process in two distinct stages, creating a symbiotic relationship between learning accuracy and resource optimization.
Stage 1: RIS-Aided Resource Allocation (Sensing)
In the first stage, the system performs cooperative spectrum sensing. Each RIS controller is equipped with a pre-trained Convolutional Neural Network (CNN) and a small number of semi-active elements capable of analog-to-digital conversion. When users send requests, the RIS controllers use these elements to infer the set of requesting users, even when direct links to the BS are obstructed. This allows the BS to perform informed configurations and optimal resource allocation at the start of each training iteration.
Stage 2: RIS-Aided FL Training (Transmission)
Once users are selected and bandwidth is allocated, the FL training process begins. Users train local models and upload their updates. During this stage, the RIS acts as an intelligent mirror, adjusting its phase shifts to beamform signals toward the BS. This strengthens the uplink, ensuring that local model updates are received correctly and quickly, overcoming the unfavorable impacts of the wireless environment.
Optimization and System Design
To fully exploit the potential of FSL, we address a joint optimization problem aimed at maximizing System Utility—defined as the ratio of correctly received model parameters to the total time consumed. We decompose this complex problem into manageable parts:
- User-RIS Association: Utilizing matching theory (one-to-one and one-to-many algorithms) to pair users with the most suitable RIS.
- Bandwidth Allocation: Employing bisection search to optimally distribute limited bandwidth among users, mitigating the "straggler effect" where slow users delay global aggregation.
Experimental Evaluation and Results
The FSL framework was validated using real-world Radio Frequency (RF) traces collected via a USRP2 testbed in a coexistence environment featuring WiFi and LTE-U signals. Key findings include:
- High Accuracy: The converged CNN model at the RIS achieved an inference accuracy of over 95% for identifying spectrum users.
- Utility Gains: In a network with 256 users, the FSL scheme with four RISs achieved a system utility 60% higher than a single-RIS configuration.
- Scalability: Both prediction accuracy and system utility improved significantly as the number of RISs and the number of reflecting elements per RIS increased.
- The Interplay: Results confirmed that higher FL prediction accuracy directly improves resource allocation efficiency, which in turn facilitates more stable and accurate FL training.
Conclusion and Future Work
Federated Spectrum Learning demonstrates that endowing RIS with sensing capabilities can significantly boost the performance of decentralized learning in challenging wireless environments. While the current model assumes perfect Channel State Information (CSI) and quasi-static channels, future work will explore FSL's performance under imperfect CSI and in high-mobility scenarios, such as vehicular edge networks.
Summary updated