Correlation-Based Sensing for Cognitive Radio Networks: Bounds and Experimental Assessment

Minimal missed detection rate of primary users is critical for adoption of cognitive radio networks, underlining the need for robust collaborative sensing combined with near-optimal single-node detection methods. Although correlation-based detection methods potentially provide needed per-node performance improvements for correlated signals, their performance for realistic blind sensing is unclear since the type and extent of correlation may be unknown in practice. Although standard Neymon-Pearson (NP) based detection can be applied when correlation is perfectly known, difficulty arises when the correlation is random, which is the focus of this paper. A tighter bound for the performance of correlation-based methods is developed herein based on a signal with random correlation and NP detection under the assumption of correlation distribution information (CDI). Simulations of existing ad-hoc correlation-based detectors are compared to the upperbound using a simple uniform random correlation model (RCM). Additionally, a measurement campaign is presented where radio-frequency (RF) spectra in many bands of interest are measured throughout a large sub-urban environment, generating realistic models for the random signal correlation. The measurement-based model indicates limits on performance gains possible with correlation-based detection and how well existing ad-hoc techniques can be expected to perform in practice.