Abstract
As the limited communication spectrum cannot meet the demand of the exponential growth of intelligent connected devices in the Internet of Things (IoT) and typical smart city applications, this paper proposes a tractable spectrum sensing method based on Rao detection. The method addresses non-Gaussian noise—such as generalized Gaussian noise (GGN), Gaussian mixture noise (GMN), and symmetric alpha-stable ($S\alpha S$) noise—within multi-path fading channel environments. By utilizing adaptive kernel density estimation (AKDE) for noise types without closed-form probability density functions (PDFs), the proposed framework ensures reliable spectrum identification for secondary users. Simulation results validate the accuracy of the estimated PDFs and the sensing performance across indoor, outdoor, and vehicle fading channels.
Introduction: The Challenge of Spectrum Scarcity in Smart Cities
The rapid expansion of the Internet of Things (IoT) has led to a critical shortage of available radio spectrum. Cognitive Radio (CR) technology provides a promising solution by allowing "Secondary Users" (SUs), such as IoT devices, to opportunistically access frequency bands when the "Primary User" (PU) is inactive.
However, real-world smart city environments present significant detection challenges. Traditional spectrum sensing methods often assume additive white Gaussian noise (AWGN). In reality, urban electromagnetic environments are plagued by non-Gaussian, impulsive noise—sharp spikes caused by power lines, electronic interference, or atmospheric events. Furthermore, signals are subject to multi-path fading as they bounce off buildings and moving vehicles, making it difficult to distinguish a weak PU signal from background noise.
System Model and Methodology
The proposed framework focuses on a cognitive radio scenario where an SU must detect the presence of a PU signal amidst complex noise and fading.
Rao Detection for Computational Efficiency
To accommodate the low-power nature of IoT devices, the authors employ Rao detection. Unlike the Generalized Log-Likelihood Ratio Test (GLRT), which requires parameter estimation for both the "signal present" ($H_1$) and "signal absent" ($H_0$) hypotheses, the Rao test only requires estimation under $H_0$. This significantly reduces the computational overhead while maintaining robust performance in fading channels.
Adaptive Kernel Density Estimation (AKDE)
A major hurdle in non-Gaussian environments is that some noise models, particularly Symmetric Alpha-Stable ($S\alpha S$) noise, lack a closed-form PDF. To overcome this, the authors utilize Adaptive Kernel Density Estimation (AKDE). By "learning" the noise distribution from data samples and adjusting the bandwidth to capture "heavy tails" (impulsive spikes), the system can accurately model the interference environment. To further refine accuracy, the authors enforce symmetry properties on the estimated PDF, reflecting the physical nature of the noise.
Performance Evaluation
The framework was evaluated using ITU-R M.1225 standard channel models for indoor, outdoor, and vehicle environments.
Key Findings
- Estimation Accuracy: The AKDE method achieved high precision, with absolute errors for $S\alpha S$ noise PDF estimation typically falling below 0.02.
- Detection Robustness: The system maintained a high probability of detection ($P_d$) even at low false alarm rates ($P_f$), demonstrating resilience against impulsive noise spikes.
- Environmental Adaptability: While effective across all tested scenarios, the method showed peak performance in outdoor multi-path fading environments.
- Theoretical Consistency: Monte Carlo simulations (10,000 experiments) closely aligned with theoretical derivations, confirming the mathematical validity of the Rao-based approach.
Conclusion and Future Directions
This research provides a tractable and robust solution for spectrum sensing in the complex electromagnetic landscapes of smart cities. By combining Rao detection with AKDE, the proposed method allows IoT devices to find "white spaces" in the spectrum efficiently.
Future work may explore cooperative sensing, where multiple IoT devices share data to further enhance detection reliability. Additionally, validating the algorithm with real-world RF data from dense urban environments would help account for hardware-specific impairments like IQ imbalance.
Summary updated