Eigenvalue based Spectrum Sensing Algorithms for Cognitive Radio

Spectrum sensing is a fundamental component is a cognitive radio. In this paper, we propose new sensing methods based on the eigenvalues of the covariance matrix of signals received at the secondary users. In particular, two sensing algorithms are suggested, one is based on the ratio of the maximum eigenvalue to minimum eigenvalue; the other is based on the ratio of the average eigenvalue to minimum eigenvalue. Using some latest random matrix theories (RMT), we quantify the distributions of these ratios and derive the probabilities of false alarm and probabilities of detection for the proposed algorithms. We also find the thresholds of the methods for a given probability of false alarm. The proposed methods overcome the noise uncertainty problem, and can even perform better than the ideal energy detection when the signals to be detected are highly correlated. The methods can be used for various signal detection applications without requiring the knowledge of signal, channel and noise power. Simulations based on randomly generated signals, wireless microphone signals and captured ATSC DTV signals are presented to verify the effectiveness of the proposed methods.

💡 Research Summary

This paper addresses the fundamental problem of spectrum sensing in cognitive radio (CR) systems, where secondary users must reliably detect the presence of primary users without prior knowledge of signal, channel, or noise characteristics. Traditional detectors—energy detection, matched filtering, and cyclostationary detection—either require accurate noise power estimates (energy detection) or detailed signal/channel information (matched filtering, cyclostationary). The authors propose two blind detection schemes that exploit the eigenstructure of the sample covariance matrix formed from the received multi‑antenna or oversampled signal vectors.

The first scheme, called Maximum‑Minimum Eigenvalue (MME) detection, computes the ratio λ_max/λ_min, where λ_max and λ_min are the largest and smallest eigenvalues of the sample covariance matrix R_x. Under the null hypothesis (no primary signal), R_x reduces to σ_η²I, yielding a ratio of exactly one. When a signal is present, the covariance acquires a low‑rank signal component H R_s H†, raising λ_max while λ_min remains close to the noise variance. Thus λ_max/λ_min > 1 signals the presence of a primary transmission. The second scheme, Energy‑with‑Minimum‑Eigenvalue (EME) detection, replaces the numerator with the average eigenvalue Δ = trace(R_x)/(ML), which is mathematically equivalent to the received signal energy (proved in the appendix). The decision rule becomes Δ/λ_min > γ₂.

A major contribution of the work is the rigorous statistical characterization of these ratios using modern random matrix theory (RMT). The authors invoke the Marčenko–Pastur law to describe the asymptotic distribution of eigenvalues for large sample sizes N_s, and the Tracy–Widom distribution to model the fluctuations of the extreme eigenvalues. This enables closed‑form expressions for the false‑alarm probability (P_FA) and detection probability (P_D) as functions of the threshold γ₁ (or γ₂), the dimensionality ratio c = ML/N_s, and the signal‑to‑noise ratio (SNR). By fixing a target P_FA (e.g., 10⁻³), the corresponding thresholds can be analytically derived, eliminating the need for empirical tuning.

The paper also discusses the notorious “noise uncertainty” problem. In energy detection, an uncertainty factor α (often expressed in dB as B) multiplies the true noise variance, leading to an SNR wall below which reliable detection is impossible. In the eigenvalue‑based methods, λ_min itself serves as an implicit estimator of σ_η², making the detectors virtually immune to α. Moreover, when the primary signals are highly correlated—such as in multiple‑microphone scenarios, oversampled digital modulation, or MIMO transmissions—the leading eigenvalue ρ₁ of H R_s H† becomes significantly larger than the remaining eigenvalues. Consequently, λ_max/λ_min (and Δ/λ_min) increase dramatically, allowing the proposed detectors to outperform ideal energy detection even when the latter has perfect noise knowledge.

Complexity analysis shows that both algorithms require an eigen‑decomposition of an ML × ML covariance matrix, nominally O((ML)³). However, the authors note that iterative methods (Power Iteration, Lanczos) can compute the extreme eigenvalues with far lower computational burden, making real‑time implementation feasible for moderate antenna counts.

Simulation results validate the theory across three signal classes: (i) randomly generated complex Gaussian symbols, (ii) wireless microphone recordings, and (iii) captured ATSC digital TV signals. Experiments span SNR values from –20 dB to 0 dB and noise‑uncertainty levels up to B = 2 dB. For a target P_FA = 10⁻³, both MME and EME achieve P_D ≈ 0.9–0.99, with a noticeable 10–15 % gain over energy detection when the primary signal exhibits strong correlation. The detectors also maintain stable performance across the entire noise‑uncertainty range, confirming their robustness.

In conclusion, the paper presents a compelling case for eigenvalue‑based spectrum sensing: (1) it eliminates the need for explicit noise power estimation, thereby solving the noise‑uncertainty issue; (2) it leverages signal correlation to achieve superior detection performance compared to classical energy detection; and (3) it operates blindly, requiring no prior knowledge of signal waveforms, channel responses, or noise statistics. The authors suggest future work on adaptive threshold selection, low‑complexity eigenvalue algorithms, and cooperative sensing extensions for multi‑user CR networks.