Transcript of Blind-Spectrum Non-uniform Sampling and its Application in Wideband Spectrum Sensing
- 1. Blind-Spectrum Non-uniform Sampling and its Application in
Wideband Spectrum Sensing By. M. R. Avendi
- 2. Agenda Blind Spectrum Signal Model Parameters L, p, q, C
Spectral Recovery Subspace Method NLLS Method Simulation
Application for Spectrum Sensing Application for Spectrum Sensing
Cognitive Overview Spectrum sensing Current methods Proposed model
Simulation Summary and conclusion 2
- 3. Blind spectrum signal model Number of bands N Each band no
wider than B Maximum frequency fmax Locations unknown Landau lower
bound (F)=NB 3
- 4. Sampling Parameters Number of active slots : qmin < q
< qmax q=3 4 Minimum and Maximum number of active slots q=6
- 5. Sampling Parameters L p > qmax q = N(d+1) 5 qmax= N(d+1)
Sampling Rate
- 6. Sampling Parameters Sample pattern C - Exhaustive Search-
Exhaustive Search - Random Selection - Sequential Search 6
- 7. Spectral Recovery Ideal model y(f)=AC(k) z(f) Non-ideal
model y(f)= A (k) z(f)+ n(f)y(f)= AC(k) z(f)+ n(f) y is known, k
and z are unknown n(f) additive white noise spectral index set k ?
7
- 8. Spectral Recovery Subspace method 8
- 9. Number of active slots q 9 Ordered Eigenvalues
- 10. Number of active slots Information theoretic criteria
approaches AIC : Akaike Information criterion MDL: Minimum
Description length 10
- 11. Number of active slots Exponential fitting test (EFT) 11
Ordered Eigenvalues of signal and noise
- 12. Location of active slots MUSIC algorithm 12 The k-th column
of modulation matrix
- 13. MUSIC-Algorithm 13 Signal spectrum and MUSIC- Results
- 14. Spectral Recovery : NLLS method 14 Solution: -Exhaustive
search - Sequential search
- 15. NLLS method 15 Least square error criteria
- 16. Simulation N=3, fmax=20, B=0.9 16 Time and frequency domain
of multi-band signal
- 17. simulation L=floor(fmax/B)=22 qmax= N*2=6 => p=q+1=7 C=
{0 5 6 8 11 16 17} Sampling rate D= f *p/L= 6.3 !! In compare
Sampling rate D= fmax*p/L= 6.3 !! In compare 20!! Compute matrix
R7*7 17
- 18. simulation 18 simulation result
- 19. Simulation Sampling ratio=6.36 p=7 L=22 19 Reconstructed
signal in time and frequency domain RMSE=2.7%
- 20. spectrum sensing 20
- 21. Spectrum sensing Narrowbands Wideband Challenge: High
sample rate 21
- 22. Spectrum Sensing: Proposed Model 22
- 23. Model parameters Given channel B, , fmax L= fmax/B , p= L
Compression Ratio: CR=p/L > Sampling Parameters Sampling
Parameters Detection Probability: Depend on SNR and CR 23
- 24. Spectrum sensing: simulation =0.1 , fmax=2 GHz Resolution
Channel : B=10 MHz L=2GHz/10MHz= 200 p= L = 200 * 0.1= 20 p= L =
200 * 0.1= 20 fs= fmax * p/L= 200 MHz !! CR= 0.1 24
- 25. Simulation 25
- 26. simulation: 26 Detection probability vs. SNR for various
Compression Ratio
- 27. Summary & Conclusion Periodic non-uniform Sampling
& Reconstruction Sampling parameters: L, p ,sample pattern C
Spectral recovery : Subspace, NLLS Wideband spectrum sensing Future
works: Implementation aspects of non-uniform ADC General sample
pattern for blind spectrum signal Reduce the sample rate for
extreme case of blind signals Formulation of detection probability
vs. SNR and CR 27
- 28. Thank you for attention Question ? 28