Compressed Wideband Spectrum Sensing for Whitespace Networking. Vishnu Vardhan ECE Dept

Transcription

1 Compreed Wideband Spectrum Sening for Whitepace Networking Vihnu Vardhan ECE Dept

2 roblem Decription Spectrum i heavil unutilized in man band 0-6 Gz Whitepace Networking: Cognitive Radio opportuniticall eploiting thee pectral hole. ow? UF Spectrum Map 7 th floor, CS dept, UW-Madion Each channel i 6 Mz wide, onl 5 out of 40 channel in UF band are occupied

3 Compreed Wideband Sening [Borrowed from Anant Sahai DSAN 05utorial] Currentl band b band ening i emploed which i time conuming Need high rate ADC to ene wideband which i prohibitive he Edge Spectrum i compreible hence Compreed Wideband Sening

4 A Compreed Sening Architecture ropoed b Yvanolo, Ying Wang et al, Compreive wide-band pectrum ening ICASS, 009 AIC can be viualized a Nquit rate ampling followed b Compreed ening r r where i related to M N N N

5 Edge Spectrum i given b z where W Wavelet Smoothing F Fourier ranform Γ ΓFWr L OLL - M ΓFW r i.e thet order difference r ΓFW z Gz Recover : z argmin z z.t. r Gz

6 Need M > logn/ Can we do better? Sample Required In a given location, the Support of V band i tatic, onl the Support of MIC i varing, that too lowl During initial ening we tr to accuratel gauge {upport of V band}, and ue thi knowledge in ubequent ening We eplore two approache in thi repect

7 A Modified l- recover ropoed b NamrataVawaniand Wei Lu, Modified-CS: Modifing Compreive Sening for roblem with artiall nown Support,to appear in IEEE ran. Signal roceing,00. Recover: arg min c.t. where i known upport and k Require m > c' u log ne / u k u log9 / δ ~ O u log n / u k ~ f u k

8 Eact Recontruction Normal Modified l-0 verion δ < δ k u < l- verion δ δ 3 < δ δ δ δ δ δ k u k u u 3u k <

9 A3 Selective Acquiition What if I don t care about the nown Support i.e. filter out the nown Support during ignal acquiition Chooe I ' Ψ Ψ Ψ Ψ ' ~ Require m > c' u log ne / u u log9 / δ ~ O u log n / u

10 erformance of the 3 cheme

11 Reult with Spectrum Data otal Channel 9, DV3,Analog, M0.4N, nown 9,0,3, Unknown6

12 Normalied MSE v m/n

13 B Compreed Detection In the previou approache we firt recover the Edge Spectrum and poe the detection problem on it. In thi ection we avoid the recover phae and poe the detection problem directl on the compreed ample. At R where : i n i i ' are orthogonal i Following approach for developed b M. A. Davenport,.. Boufouno, M. B. Wakin, and R. G. Baraniuk, "Signal proceing with compreive meaurement,"ieee Journal of Selected opic in Signal roceing, April 00.

14 Detection of nown Spare Signal σ σ σ 0 0, ~ :, 0, ~ : i fied, 0, ~, : : N N N N I N n R n n γ η : which implifie to log Log Likelihood Ratio tet i : t f f < > < > Λ

15 Detection of nown Spare Signal 0 : t : t where ~ ~ N0, σ N, σ, we et D α F α reulting in Q Q α / σ We know that if for an hu S, D α N/M δ M/N SNR α Q Q M N provide a δ table δ embedding M N of S, then Since M<N, Compreed Detector ha lower Detection robabilit, for e.g. if M0.5N then performance lo i 3 db in SNR

16 Detection of nown Spare Signal Upon further implification it can be hown that cm / N D α ce hu for fied SNR and N the probabilit of detection approache eponentiall fat with M Some Reult Channel 9, Mz, SNR~ 0-5 db

17 Different hi, fied m and hrehold

18 nown Signal: > n t t i i tatitic, tet he : i Rule Detection channel For 0 th < > γ n i n n i i i i i ~ ~ ~,,, > <

19 Q Q α α the interference power from other tranmiion i ~ where, ~ :, 0, ~ :, 0, ] E[i I I N t I N t σ σ D I Q Q, σ α α hu Interference further degrade the detector performance

20 Unknown Signal For th channel Detection Rule i t : ~ < > 0 γ It i an energ detector and it performance take another 3 db lo in both normal and compreed cae. Not enough data available for wideband cae iemulti rimar Uer

21 Compreed Spectrum Detection Step : UF ha about 40 V band with fied bandwidth 6 Mz and frequencie out of which onl few are occupied. Identif the unued band. Step : MIC can be preent at an frequencie in the unued band and have a bandwidth of 00 kz. Divide the unued pectrum into narrow band to detect them. Finall need to compare the Detection erformance and Running ime of RecoverDetection v Compreed Detection v Uncompreed Detection

22 hank You