Advanced Spectrum Sensing in Cognitive Radio Networks: Riemannian Geometry Techniques

Khaja Kamaluddin Research Associate, Research & Development.

In Brief about the paper. Cognitive Radio Networks (CRNs) enable SUs to access underutilized licensed bands. Spectrum sensing ensures PU protection. Traditional methods fail in low SNR. Riemannian geometry offers robust detection via SPD matrix modeling. Proposal: Riemannian Distance Detector (RDD). Experiments show RDD > conventional detection..

Introduction. Spectrum scarcity problem ⚡ CRNs enable dynamic spectrum access. Spectrum sensing = core function. Conventional methods: Energy Detection Matched Filtering Cyclostationary Detection Problems: Noise, low SNR, uncertainty. Solution: Riemannian geometry-based techniques..

Traditional Spectrum Sensing Methods. 1. Energy Detection (ED) Simple, no prior info. SNR wall issue. 2. Matched Filtering Best when signal known. Needs full PU knowledge. High complexity..

Limitations of Traditional Methods. Sensitive to noise uncertainty. Poor performance in low SNR. High computational cost in matched & cyclostationary. Need for robust, adaptive approach..

Riemannian Geometry in Signal Processing. SPD matrices → covariance of signals. Lie on Riemannian manifold (curved space). Provides geometrically accurate distances. Applications: Brain-computer interfaces Radar Cognitive radio spectrum sensing.

Mathematical Foundations. 7. [image] Riemannian Distance: Riemannian Mean: log c-1äC2c-1ä — arg min d2R(C, Ci).

Proposed Riemannian Distance Detector (RDD). Steps: Estimate covariance Cx​. Reference noise covariance C0. Compute Riemannian distance dR(Cx,C0). Threshold γ using false alarm probability. Decision: PU present / absent..

Simulation and Analysis. Simulation 1: PU Detection Noise-only vs Signal-present. Threshold based on Pfa. False alarms observed but expected. Histogram visualization..

Simulation 2: Performance Comparison. SNR range: -20 dB to 10 dB Monte Carlo runs: 100 Metrics: Pd vs Pf RDD outperformed ED, MF, Cyclostationary..

Comparison Table. 11. [image] Method Energy Detection Cyclostationary Matched Filter RDD (Proposed) Pd @ Pf=O.1 -0.15 -0.65 -0.90 Complexity Low High High Moderate Robust to Noise.

Discussion. RDD captures geometric structure of signals. Strong performance at low SNR. Resilient to noise uncertainty. Moderate computational load. Suitable for 5G/6G & SDRs..

Future Research Directions. Cooperative sensing with Riemannian metrics. ML on manifolds (deep learning + Riemannian). Hardware acceleration (FPGA/GPU). Evaluation under non-Gaussian noise & wideband..

Conclusion. Riemannian geometry enhances CRN spectrum sensing. RDD > traditional methods (especially at low SNR). Balance between accuracy & complexity. Future: combine with ML & cooperative sensing..

15. Questions and Suggestions ?. Thanks for you time.