Source-linked AI summary
Cooperative Spectrum Sensing based on the Limiting Eigenvalue Ratio Distribution in Wishart Matrices
Federico Penna, Roberto Garello, Maurizio A. Spirito
TL;DR
Eigenvalue-ratio detectors for cooperative spectrum sensing have relied on asymptotic assumptions because the exact ratio distribution is difficult to characterize. This paper derives a limiting ratio distribution from Wishart-matrix results, uses it to set detection thresholds, and reports more accurate distribution matching and improved detection performance, particularly for realistic sensing sizes.
Problem
Existing eigenvalue-ratio detection rules rely on asymptotic approximations because the exact eigenvalue-ratio distribution is difficult to characterize.
Method
The paper derives a limiting distribution for the ratio of the largest and smallest eigenvalues of a complex Wishart covariance matrix and applies it to threshold selection.
Results
The novel analytical CDF matches empirical data, and the resulting detection rule outperforms previously proposed eigenvalue-based rules, especially with realistic numbers of sensing samples and cooperative receivers.
Takeaways & Limitations
The limiting eigenvalue-ratio distribution provides a basis for eigenvalue-based spectrum detection beyond the prior asymptotic approximations.
Takeaways & Limitations
The asymptotic approach is valid for very large N and K, while the evaluated setting with N = 1000 and K = 50 remains far from that regime.
Abstract
from arXiv · showhide
Recent advances in random matrix theory have spurred the adoption of eigenvalue-based detection techniques for cooperative spectrum sensing in cognitive radio. Most of such techniques use the ratio between the largest and the smallest eigenvalues of the received signal covariance matrix to infer the presence or absence of the primary signal. The results derived so far in this field are based on asymptotical assumptions, due to the difficulties in characterizing the exact distribution of the eigenvalues ratio. By exploiting a recent result on the limiting distribution of the smallest eigenvalue in complex Wishart matrices, in this paper we derive an expression for the limiting eigenvalue ratio distribution, which turns out to be much more accurate than the previous approximations also in the non-asymptotical region. This result is then straightforwardly applied to calculate the decision threshold as a function of a target probability of false alarm. Numerical simulations show that the proposed detection rule provides a substantial performance improvement compared to the other eigenvalue-based algorithms.
I. INTRODUCTION
Eigenvalue-based cooperative spectrum sensing uses covariance-matrix eigenvalues to detect primary signals without prior signal or noise-power information. Existing decision rules rely on asymptotic approximations, motivating a limiting eigenvalue-ratio distribution and a new detection rule.
- Motivation: Eigenvalue-based methods use random-matrix properties of the received signal covariance matrix for cooperative spectrum sensing.Their advantage over energy and cyclostationary detection is that they require no prior information about the primary signal or noise power.
- Problem: Existing eigenvalue-based decision rules rely on asymptotic approximations that can be inaccurate in practical scenarios.The stated limitation motivates a distributional treatment beyond the asymptotic regime.
- Contribution: The paper derives an analytical limiting distribution for the ratio of the largest and smallest covariance-matrix eigenvalues.The derivation uses recent random-matrix-theory results.
- Contribution: The derived distribution is used to obtain a novel eigenvalue-based decision rule that outperforms previously proposed rules.The paper applies the result to the detection problem and reports numerical evaluation.
A. System Model
The system collects K receiver streams over N sensing samples and forms their sample covariance matrix. Its eigenvalues are used to infer whether a primary signal is present.
- System setup: K collaborating receivers collect N samples during the sensing interval.The received samples are organized into a K × N matrix Y.
- Hypotheses: The model distinguishes H0, containing only noise, from H1, containing a primary signal plus noise.Under H0 the samples are noise; under H1 the received signal includes the primary waveform and channel contribution.
- Covariance matrix: The normalized sample covariance matrix converges to the population covariance matrix as N approaches infinity.The sample covariance is constructed from Y and its conjugate transpose.
- Detection input: The eigenvalues of the sample covariance matrix provide information for inferring primary-signal presence or absence.The system model identifies covariance-matrix eigenvalues as the detection input.
B. Previous Results
Under H0, the normalized covariance matrix follows a complex white Wishart model with bounded eigenvalue support. Under H1, a signal spike raises the largest eigenvalue, motivating an eigenvalue-ratio test with a threshold γ.
- Test statistic: The test statistic is T = λmax/λmin, the ratio of the largest to smallest eigenvalues.The ratio is used because the largest eigenvalue responds to the signal while the smallest eigenvalue remains part of the covariance-spectrum characterization.
- H0 behavior: Under H0, the normalized covariance matrix is a complex white Wishart matrix whose eigenvalue support is finite.This behavior follows from the Marchenko-Pastur law.
- H1 behavior: Under H1, the covariance matrix is a spiked population model and its largest eigenvalue moves outside the Marchenko-Pastur support.The signal-induced spike increases the largest eigenvalue.
- Decision rule: The detector decides H0 when T < γ and H1 otherwise.The threshold γ is the decision parameter whose selection is addressed by approaches in the literature.
1) Asymptotic Approach [1]:
The asymptotic approach sets the detection threshold from limiting eigenvalue behavior under H0 and H1. It uses almost-sure limits for the smallest and largest eigenvalues and leads to an asymptotic rule.
- Asymptotic Approach [1]: Under H0, the smallest and largest eigenvalues converge almost surely to limiting values.These limits are obtained from asymptotic Wishart-matrix properties.
- Asymptotic Approach [1]: Under H1, spiked-model theory makes the largest eigenvalue converge almost surely to a value b′ greater than b.The separation of the H1 largest-eigenvalue limit from the H0 value supports detection.
- Asymptotic Approach [1]: The approach uses these asymptotic eigenvalue results to define a detection threshold and decision rule.The cited method is presented as an asymptotic detection rule based on the limiting values.
- Asymptotic Approach [1]: A related approach uses the limiting distribution of the largest eigenvalue instead of its asymptotic value.This contrasts distribution-based thresholding with the asymptotic-value approach.
2) Semi-asymptotic Approach [2]:
The semi-asymptotic approach combines an asymptotic approximation for the smallest eigenvalue with the Tracy-Widom distribution for the largest eigenvalue to derive a tunable detection threshold.
- It retains the asymptotical limit for the smallest eigenvalue when forming the eigenvalue ratio.
- The approach uses the Tracy-Widom law of order 2 for the largest eigenvalue's limiting distribution.
- The resulting threshold is linked to a target probability of false alarm through the inverse Tracy-Widom CDF.
- The Tracy-Widom distribution is defined through a Painlevé II differential equation and an Airy-function boundary condition.
- Its CDF can be evaluated using tabulated values or a Matlab routine.
III. EIGENVALUE RATIO DISTRIBUTION AND NEW DETECTION THRESHOLD
The paper derives a limiting distribution for the largest-to-smallest eigenvalue ratio using Tracy–Widom limits for both eigenvalues, then uses it to set false-alarm-controlled detection thresholds. The resulting approach addresses inaccuracies of asymptotic and semi-asymptotic methods when observations are limited.
- Motivation: Asymptotic thresholds become unbalanced relative to the actual eigenvalue-ratio distribution and cannot be tuned to a target false-alarm probability.The semi-asymptotic approach permits false-alarm control but remains inaccurate as N decreases.
- Limiting ratio distribution: The smallest eigenvalue is modeled through its limiting Tracy–Widom distribution after appropriate rescaling, complementing the corresponding result for the largest eigenvalue.The ratio derivation uses limiting PDFs for the numerator and denominator, with an independence assumption for the limiting distributions.
- Limiting ratio distribution: The limiting PDFs are transformed and combined with the ratio-distribution formula to obtain the PDF and CDF of the eigenvalue-ratio test statistic.The integration is restricted to nonnegative eigenvalues and t > 1 preserves the ordering l1 > lK.
- Validation: The proposed ratio-based approach is compared with empirical and asymptotic CDFs, including a semi-asymptotic setting with N = 1000 and K = 50.The section frames the method for practical conditions involving a small number of observations due to time-varying operation.
- New detection threshold: The new CDF yields a detection threshold as a function of the target probability of false alarm, with numerical evaluation enabling receiver implementation through a look-up table.The table can be indexed by N, K, and the target Pfa; the inverse CDF required by the asymptotic implementation has no closed-form expression.
IV. NUMERICAL RESULTS
The novel ratio-distribution approach closely matches empirical eigenvalue-ratio data and improves detector performance over asymptotical, semi-asymptotic, and energy-based approaches. Its nearly exact distribution enables threshold selection for a target Pfa and yields lower Pmd in the reported finite-sample setting.
- Eigenvalue-ratio distribution: The novel analytical CDF matches empirical data, while asymptotic and semi-asymptotic results are unbalanced for N = 1000 and K = 50.These parameters remain far from the asymptotical region, so neither competing approach sets the decision threshold correctly for the target Pfa.
- Detector comparison: The complementary ROC compares the novel ratio-based detector with asymptotical, semi-asymptotic, and cooperative equal-gain energy detection.The graph reports achievable Pmd versus target Pfa, with N = 1000 and K = 50 in the simulation setting.
- Detector comparison: The novel ratio-distribution threshold provides lower Pmd than the other approaches for any given Pfa.The eigenvalue-based algorithms are insensitive to noise power uncertainty, whereas energy detection assumes 0.25 dB noise uncertainty.
- Detector comparison: 10^-1 target Pfa produces Pmd = 1.0 · 10^-2 for the novel approach versus 6.5 · 10^-2 for the semi-asymptotic approach.The nearly exact distribution permits selection of the lowest threshold for a given target Pfa and therefore the minimum Pmd.
- Detector comparison: The asymptotical approach fixes its threshold, achieving (Pfa, Pmd) = (4 · 10^-3, 1.15 · 10^-1), a Pmd lower bound that cannot be improved by changing target Pfa.This fixed-threshold behavior prevents control of Pmd versus Pfa.
V. CONCLUSION
The paper derives a limiting eigenvalue-ratio distribution for Wishart matrices and applies it to cognitive-radio signal detection. The resulting analytical distribution agrees with empirical data, and its detection rule outperforms previous methods, especially with realistic sensing samples and cooperative receivers.
- The paper derives an expression for the limiting eigenvalue-ratio distribution in Wishart matrices and applies it to signal detection in cognitive radio.
- The analytical distribution is consistent with empirical data.
- The novel detection rule clearly outperforms previously proposed methods, especially for realistic numbers of sensing samples and cooperative receivers.