Statistical detection theory for stochastic-resonance-aided spectrum sensing in the low-SNR regime

Zheng, Li, Djouadi and Wang (2009) proposed detecting weak primary-user signals in cognitive radio by inserting a bistable stochastic-resonance (SR) filter before an energy detector, but the scheme was left statistically unprincipled: the distribution of the SR output statistic is unknown, forcing a heuristic CUSUM-like accumulation rule; the criterion for engaging SR instead of ordinary baseline detectors (the energy detector and the matched filter) is empirical; SR parameter tuning was illustrated on one toy setting; and the mixer downconversion front-end was only sketched. Later work has pursued heuristic parameter adaptation and alternative SR realizations without deriving the output law or exact optimal tests. The open problem is to construct the statistical detection theory of SR-aided spectrum sensing: characterize the probability law of the SR-filter output statistic under the signal-present and signal-absent hypotheses, build from it exact likelihood-ratio block tests and CUSUM-type sequential tests with provable false-alarm and delay guarantees, derive a principled criterion for when SR preprocessing outperforms those baselines together with a detection-optimal parameter tuning rule, and analyze the RF downconversion front-end that feeds the SR filter.

Background

Cognitive radio is a wireless-networking paradigm in which secondary users (SUs), devices holding no license, opportunistically transmit in frequency bands licensed to primary users (PUs); to avoid harmful interference, an SU must sense its channel and detect PU transmissions reliably even when the PU signal arrives extremely weak. The IEEE 802.22 standard, for example, requires detecting the PU within about 2 seconds at false-alarm probability below 10% and detection probability above 90%, at received powers near -116 dBm - conditions that force operation at very low signal-to-noise ratio (SNR). The standard sensing detectors are the energy detector (a radiometer comparing measured received power to a threshold), feature detectors (exploiting known PU structure, such as the sinusoidal pilot tone of an ATSC (Advanced Television Systems Committee, the North American digital-television broadcast standard) signal), and the matched filter (the optimal linear detector when the waveform is fully known). Energy detection degrades badly at low SNR and is fundamentally limited by the SNR wall - a threshold, induced by uncertainty in the noise power, below which detection is impossible regardless of how long one senses - while feature detection needs many samples and the matched filter needs full knowledge of the PU waveform.

Zheng, Li, Djouadi and Wang proposed to insert a stochastic-resonance (SR) filter before the energy detector. Stochastic resonance, introduced by Benzi, Sutera and Vulpiani to explain periodic ice-age recurrence, is the phenomenon in which the response of a nonlinear system to a weak periodic input is enhanced by an appropriately chosen amount of noise. Their SR filter is the classical bistable system: a state $x(t)$ obeying the Langevin stochastic differential equation $\dot{x}(t)=ax(t)-bx(t)^{3}+A\sin(2\pi ft)+\eta(t)$, which is over-damped motion in the symmetric double-well potential $V(x)=-ax^{2}/2+bx^{4}/4$ with stable points $\pm\sqrt{a/b}$; $A\sin(2\pi ft)$ is the (down-converted) PU pilot tone and $\eta$ is additive white Gaussian noise of power $D$. Noise-driven hopping between the wells occurs at the Kramers rate $R=(a/\sqrt{2\pi})\exp(-\Delta V/D)$ with barrier height $\Delta V=a^{2}/4b$; when the hopping rate is matched to the forcing, the switching synchronizes with the weak sinusoid and the output spectrum at frequency $f$ is strongly amplified, so the output SNR at $f$ can exceed the input SNR. Detection-theoretic studies of SR - detection statistics of stochastic resonators (Inchiosa and Bulsara 1996), SR in sine detection (Zozor and Amblard 2002), and SR effects in locally optimal detectors built from static memoryless nonlinearities (Zozor and Amblard 2003) - established that noise-enhanced detection is real, but only through approximate performance analyses. The exact finite-time probability law of the output of the dynamical bistable filter is not known in closed form: the time-dependent Fokker-Planck equation (the evolution equation for the probability density of a diffusion process) has no closed-form solution under periodic forcing, and the standard theory rests on the two-state adiabatic approximation of McNamara and Wiesenfeld (1989).

Zheng et al. therefore built their detector heuristically. The received samples are passed through the SR filter, a fast Fourier transform (FFT) is applied blockwise, and the per-block energy is $E_{n}=\max_{t}|Y(n,t)|^{2}$, where $Y(n,t)$ is the $t$-th frequency-bin output of the FFT of block $n$, so $E_{n}$ is the largest FFT-bin power at the output of block $n$; block detection compares the average $M=(1/K)\sum_{n}E_{n}$ to a threshold, and sequential (quickest) detection runs the CUSUM-like recursion $m(n)=\max\{m(n-1)+E_{n}-E_{0},0\}$ until it exceeds a threshold. CUSUM (cumulative sum) here refers to the classical quickest-change-detection procedure that accumulates log-likelihood ratios and provably minimizes worst-case detection delay at a given false-alarm rate. Because the authors could not obtain the distribution of the SR output - their footnote states 'We are unable to obtain the distribution of the output signal of SR system... we cannot apply exact CUSUM test' - the likelihood increment is replaced by the energy difference $E_{n}-E_{0}$, where $E_{0}$ is a fixed reference level set for the SR output energy at the pilot frequency, so the false-alarm and delay behavior of the sequential test is only empirical. Simulations in the paper further show that SR preprocessing helps only in the very-low-SNR regime (at -10 dB it is worse than plain energy detection); a parallel architecture of one SR-based and one ordinary energy detector is sketched but the combination rule is left unspecified; the parameters $a=b=1$, $f=10$ Hz, $A=0.3$ carry a single empirically optimal noise power $D=0.43$, found for that one toy setting with the optimization details omitted; and downconversion of the ATSC pilot (about 309.44 kHz or 328.84 kHz, with frequency error within about 10 Hz) via a mixer and low-pass filter into the SR filter's low-frequency range is proposed without any analysis of its tradeoffs or error sensitivity.

Problem Statement

Establish the statistical detection theory for the stochastic-resonance (SR)-aided spectrum sensing scheme of Zheng et al. (2009), replacing its heuristic elements with principled, analyzable designs, under the source paper's system model: the binary hypothesis test $H_{0}$: $Y(n)=W(n)$ versus $H_{1}$: $Y(n)=S(n)+W(n)$ with $W$ additive white Gaussian noise and $S$ a sinusoidal primary-user (PU) pilot, filtered by the bistable SR system $\dot{x}(t)=ax(t)-bx(t)^{3}+A\sin(2\pi ft)+\eta(t)$ and observed through the fast-Fourier-transform (FFT)-based statistics $E_{n}=\max_{t}|Y(n,t)|^{2}$ (where $Y(n,t)$ is the $t$-th frequency-bin output of the FFT applied blockwise to the SR-filter output of sensing block $n$, not the raw received sequence $Y(n)$ of the hypothesis test) and $M=(1/K)\sum_{n=1}^{K}E_{n}$. Under $H_{1}$ the forcing amplitude $A$ is that of the down-converted pilot and vanishes under $H_{0}$; the SR noise power $D$ is a separately tunable quantity, as in the source, where $D=0.43$ is optimized while the received SNR is varied independently. The source does not fix the sampling and windowing conventions linking the continuous-time SDE to the discrete sensing blocks.

All requested results serve one objective - a principled statistical theory of SR-aided sensing - and are scientifically linked:

1. Output law: characterize the probability law of the SR-filter output statistic - $E_{n}$ under $H_{0}$ and $H_{1}$, the block statistic $M$, and the dependence structure across blocks needed for sequential testing - either exactly in closed form or through a computational procedure with certified error bounds, for the bistable SR filter in the parameter regimes relevant to low-SNR sensing.

2. Optimal tests: from this law, construct the likelihood-ratio block test and an exact CUSUM-type (cumulative-sum) sequential test for the emergence of a PU, with thresholds set so that the false-alarm probability and the average detection delay provably meet specified targets, replacing the heuristic accumulation rule $m(n)=\max\{m(n-1)+E_{n}-E_{0},0\}$, in which $E_{0}$ is a fixed reference energy level.

3. Engagement criterion and tuning: derive from the same theory a criterion, as a function of input SNR, pilot frequency, and SR parameters $(a,b,D)$, for when SR preprocessing improves detection relative to the energy detector and the matched filter; from it, give the principled combination/switching rule for the parallel SR-based and ordinary detectors and a parameter-selection rule that optimizes detection performance - rather than output SNR alone - for the pinned sinusoidal pilot, with an optional extension of the rule across PU signal families and time-varying SNR beyond that pilot (motivated by the source's remark that PU signals are generally periodic).

4. Front-end feasibility: analyze the mixer-plus-low-pass downconversion stage that maps the RF pilot (about 309.44 or 328.84 kHz for ATSC, with frequency error of order 10 Hz) into the SR filter's low-frequency operating band, including its noise/aliasing tradeoffs and its sensitivity to frequency-offset error, and quantify the resulting end-to-end detection degradation.

Results 1-3 form the tightly coupled core: the law enables the tests, and the tests define the engagement criterion and the tuning objective. Result 4 addresses the downconversion front-end that the source only sketches, leaving its tradeoffs and error sensitivity unanalyzed. Partial heuristic progress exists on parameter adaptation and alternative SR realizations.

References

1. Kun Zheng, Husheng Li, Seddik M. Djouadi, Jun Wang, 'Spectrum Sensing in Low SNR Regime via Stochastic Resonance', arXiv:0906.0739 (2009); published in Proc. 44th Annual Conference on Information Sciences and Systems (CISS), pp. 1-5, 2010, DOI: 10.1109/ciss.2010.5464907, https://arxiv.org/abs/0906.0739
2. Wei Chen, Jun Wang, Husheng Li, Shaoqian Li, 'Stochastic Resonance Noise Enhanced Spectrum Sensing in Cognitive Radio Networks', Proc. IEEE Global Telecommunications Conference (GLOBECOM) 2010, pp. 1-6, DOI: 10.1109/GLOCOM.2010.5683957, https://doi.org/10.1109/GLOCOM.2010.5683957
3. Jun Wang, Xin Ren, Shaowen Zhang, Daiming Zhang, Husheng Li, Shaoqian Li, 'Adaptive Bistable Stochastic Resonance Aided Spectrum Sensing', IEEE Transactions on Wireless Communications 13(7):4014-4024 (2014), DOI: 10.1109/TWC.2014.2317779, https://doi.org/10.1109/TWC.2014.2317779
4. Yonghua Wang, Pin Wan, Qin Deng, Yuli Fu, 'Spectrum Sensing Based on Monostable Stochastic Resonance in Cognitive Radio Networks', TELKOMNIKA (Telecommunication Computing Electronics and Control) 13(2) (2015), DOI: 10.12928/telkomnika.v13i2.1429, https://doi.org/10.12928/telkomnika.v13i2.1429
5. Jin Lu, Ming Huang, Jing-Jing Yang, 'A Novel Spectrum Sensing Method Based on Tri-Stable Stochastic Resonance and Quantum Particle Swarm Optimization', Wireless Personal Communications 95:2635-2647 (2017), DOI: 10.1007/s11277-017-3945-5, https://doi.org/10.1007/s11277-017-3945-5
6. Haftu Tasew Reda, Abdun Mahmood, Abebe Diro, Naveen Chilamkurti, Suresh Kallam, 'Firefly-inspired stochastic resonance for spectrum sensing in CR-based IoT communications', Neural Computing and Applications 32(20):16011-16023 (2020; first online 2019), DOI: 10.1007/s00521-019-04584-0, https://doi.org/10.1007/s00521-019-04584-0
7. Bin Jiang, Yuefei Cao, Jianrong Bao, Chao Liu, Xianghong Tang, 'Novel Piecewise Normalized Bistable Stochastic Resonance Strengthened Cooperative Spectrum Sensing', IEEE Transactions on Cognitive Communications and Networking 9(5):1167-1182 (2023), DOI: 10.1109/TCCN.2023.3287236, https://doi.org/10.1109/TCCN.2023.3287236
8. 'High-frequency signal spectrum sensing method via stochastic resonance', in Information, Computer and Application Engineering, pp. 63-66 (CRC Press, 2018), DOI: 10.1201/b18658-14, https://doi.org/10.1201/b18658-14 (author list not stated in the publisher metadata)
9. Steeve Zozor, Pierre-Olivier Amblard, 'On the use of stochastic resonance in sine detection', Signal Processing 82(3):353-367 (2002), DOI: 10.1016/S0165-1684(01)00174-8, https://doi.org/10.1016/S0165-1684(01)00174-8
10. S. Zozor, P.-O. Amblard, 'Stochastic resonance in locally optimal detectors', IEEE Transactions on Signal Processing 51(12):3177-3181 (2003), DOI: 10.1109/TSP.2003.818905, https://doi.org/10.1109/TSP.2003.818905
11. Fabing Duan, Derek Abbott, Qisheng Gao, 'Evaluation of bistable systems versus matched filters in detecting bipolar pulse signals', Fluctuation and Noise Letters 5(2):L127-L142 (2005), DOI: 10.1142/S0219477505002483 (preprint arXiv:cond-mat/0410412), https://arxiv.org/abs/cond-mat/0410412
12. M. E. Inchiosa, A. R. Bulsara, 'Signal detection statistics of stochastic resonators', Physical Review E 53:R2021-R2024 (1996), DOI: 10.1103/PhysRevE.53.R2021, https://doi.org/10.1103/PhysRevE.53.R2021
13. B. McNamara, K. Wiesenfeld, 'Theory of stochastic resonance', Physical Review A 39:4854-4869 (1989), DOI: 10.1103/PhysRevA.39.4854, https://doi.org/10.1103/PhysRevA.39.4854
14. R. Benzi, A. Sutera, A. Vulpiani, 'The mechanism of stochastic resonance', Journal of Physics A: Mathematical and General 14:L453-L457 (1981), DOI: 10.1088/0305-4470/14/11/006, https://doi.org/10.1088/0305-4470/14/11/006
15. Rahul Tandra, Anant Sahai, 'SNR Walls for Signal Detection', IEEE Journal of Selected Topics in Signal Processing 2(1):4-17 (2008), DOI: 10.1109/JSTSP.2007.914879, https://doi.org/10.1109/JSTSP.2007.914879
16. Haitao Dong, Shilei Ma, Jian Suo, Zhigang Zhu, 'Matched Stochastic Resonance Enhanced Underwater Passive Sonar Detection under Non-Gaussian Impulsive Background Noise', Sensors 24(9):2943 (2024), DOI: 10.3390/s24092943, https://doi.org/10.3390/s24092943