using multiple power spectrum measurements to sense...
TRANSCRIPT
UsingMultiplePowerSpectrumMeasurementstoSenseSignalswith
PartialSpectralOverlapMihirLaghateandProf.Danijela Cabric
7th March2017IEEEDySPAN 2017,Baltimore
D. Markovic / Slide 2
Outline
w Goals,Motivation,andExistingWorkw SystemModel
– Assumptions– Time-FrequencyMap
w Non-NegativeMatrixFactorization(NNMF)– Challengeswithexistingalgorithms
w ProposedAlgorithm:GreedyEnergyMinimizingNNMFw MATLABSimulationResultsw USRPMeasurementResultsw ConclusionsandFutureWork
2
D. Markovic / Slide 3
Goals
Thatis,w Countnumberofsignalsrcvdw Detectsets ofdiscreteFourier
transformbinsoccupiedbyeachsignal
3
[1] M. Laghate and D. Cabric, “Using the Time Dimension toSense Signals with Partial Spectral Overlap,” in IEEEGLOBECOM, Washington, USA, 2016.
Estimatenoisepowerspectrum
Challenges:– Colorednoise– Spursandalways-oninterferers
DistinguishSignalswithSpectralOverlap
D. Markovic / Slide 4
MotivatingApplications
– IEEE802.11nin5GHzbands– LTE-Advanced
4
Image Source:Wikipedia“ListofWLANchannels”
LTECarrierAggregation [2][2] H. J. Wu et al., “A wideband digital pre-distortion platform with 100 MHz instantaneous bandwidth for LTE-advanced applications,” in 2012 Workshop on Integrated Nonlinear Microwave and Millimetre-Wave Circuits, 2012, pp. 1–3.
– IEEE802.11b/gchannelsin2.4GHz
– ChannelbondinginIEEE802.11n
LackofGuardBands
Spectraloverlapbydesign Measurements@UCLA
D. Markovic / Slide 5
ExistingWorkforDistinguishingSignals
5
Based on Blind SingleAntenna
SpectralOverlap
DetectBands
BlindtoChannel
Transmission protocols[4],[5] û ü ü ü ü
Cyclicfrequency[7] û ü ü û ü
Channelmodel &location[6] û ü ü û û
Angle ofArrival[8] ü û ü û û
RandomMatrixTheory[9] ü û ü û ü
MultipleCRs[10],[11] ü û û ü ü
PowerSpectrumThreshold[12] ü ü û ü üMultiplePowerSpectrumMeasurements ü ü ü ü ü
Proposedmethodandourpriorwork[1]
D. Markovic / Slide 6
SystemModel
Widebandsensorw BasebandbandwidthW Hzw AdditivewidesensestationaryGaussiannoise𝜈 𝑡 ∈ ℝ%&
w WelchpowerspectrumestimatorusingFFToflengthF
IncumbentUsersw M distinctfrequencybandsw mth bandhasUm transmitterswithfreq.supportBm DFTbins
– Powerspectrumreceivedfromuth transmitter:𝑋(,* ∈ ℝ%&
– Activity𝑎(,* 𝑡 ∈ 0,1 isfractionoftth measurementthatuth
transmitterinmth bandisactive
w Receivedpowerspectrum:
6
D. Markovic / Slide 7
Time-FrequencyMap:1user/bandw Time-Freq.mapE ofreceivedenergy:E = [Y[1] Y[2] … Y[T]]T
w Definematrices:Atm= am[t],Smf = Xm( f ),andΔ/0 = 𝜈 𝑡 𝑓
7
=
+
+
E A= S +D
E
(1) (1)A ´S
Output computed by Non-Negative Matrix Factorization (NNMF) when given M = 3
Input:SimulatedPowerSpectrum
Output:Time-FreqofEachTx
(2) (2)A ´S
(3) (3)A ´S
Example:M = 3, F = 512, T = 30 Y [t] = am,u
u=1
Um
∑m=1
M
∑ [t]Xm,u[t]+ν[t]
D. Markovic / Slide 8
Non-NegativeMatrixFactorization(NNMF)
w Let𝑀4 =Estimatednumberofreceivedsignalsw NNMFfinds,tominimize
Challenges:w Estimating𝑀4 ishardwhen
w Non-convexcostfunctionÞ convergencetoglobalminimanotguaranteed
w CostfunctionisnotprobabilisticÞ Notrobusttonoise
w Non-uniquesolutionandisnotbinaryÞ ,i.e.,thresholdingwillnotdetectalloccupiedDFTbins
8
2ˆ ˆ FE A- S
T F<
AS » S/ S
A∈! +T ×M Σ ∈! M×F
D. Markovic / Slide 9
PriorWork:NNMF-basedAlgorithm
9
Initialization
NNMFofwith signals
Increment
No
DetectOccupiedBandsYes
ˆ 1M =
EnergyDetectionE
'E'E
Mˆ ˆ,A S
M
{ }ˆ1 2ˆ ˆ ˆ, ,..., 0,..., 1MB B B FÌ -
Noisebanddetected?
ˆ 4M =
Significantsignalenergy
NoiseBand
ReconstructedFactorsfor
“Leaked”energy
D. Markovic / Slide 10
MotivatingNNMFAlgorithm:RobustXRay
i.e.,conewithreceivedpowerspectraasextremerays
RecursiveAlgorithm:1. Choosepointwithmaximum residual
asextremeray2. Measureresidualtoallmeasurements3. Repeat1untilallmeasurementsliewithincone
10
[13] A. Kumar, V. Sindhwani, and P. Kambadur, “Fast conical hull algorithms for near-separable non-negative matrix factorization,” in International Conference on Machine Learning, 2013.
Image source: [13]
E Y [t] | a[t][ ] = am,u[t]E Xm,u[t]⎡⎣ ⎤⎦u=1
Um
∑m=1
M
∑ + E ν[t][ ]
D. Markovic / Slide 11
Separability Assumption
i.e.,conewithreceivedpowerspectraasextremerays
Separability Assumption:Requiresatleastonemeasurementofeachextremerayw Atleastonenoiseonlymeasurementw Foreachtransmitter,atleastonemeasurementwhereitisthe
onlyactivetransmitter
11
[14] D. Donoho and V. Stodden, “When Does Non-Negative Matrix Factorization Give a Correct Decomposition into Parts?,” in Advances in Neural Information Processing Systems, MIT Press, 2004, pp. 1141–1148.
Image source: [13]
E Y [t] | a[t][ ] = am,u[t]E Xm,u[t]⎡⎣ ⎤⎦u=1
Um
∑m=1
M
∑ + E ν[t][ ]
D. Markovic / Slide 12
ProposedGreedyEnergyMinimizingNNMF
12
Welch Power Spectrum Estimator
No
Energy Detection on Detected Signals
Yes
Received Signal
Append to Time Frequency Map E
Ø Center FrequencyØ Bandwidth
Detect Noise-Only Measurements ℒ
ℒ ≜ Measurements that are linear combinations
of Detected Signals
Noise power spectrum learned from data1. Initialize noise estimate as min. energy measurement2. Estimate noise as mean of measurements with
Mahalanobis distance < t to est. noise distribution3. Repeat Step 2 until convergence
Index in 1,… , 𝑇 \ℒ with Lowest Energy labelled
as Detected Signal
ℒ = 𝑇?
D. Markovic / Slide 13
ProposedGreedyEnergyMinimizingNNMF
13
Welch Power Spectrum Estimator
No
Energy Detection on Detected Signals
Yes
Received Signal
Append to Time Frequency Map E
Ø Center FrequencyØ Bandwidth
Detect Noise-Only Measurements ℒ
ℒ ≜ Measurements that are linear combinations
of Detected Signals
Noise power spectrum learned from data1. Initialize noise estimate as min. energy measurement2. Estimate noise as mean of measurements with
Mahalanobis distance < t to est. noise distribution3. Repeat Step 2 until convergence
Index in 1,… , 𝑇 \ℒ with Lowest Energy labelled
as Detected Signal
ℒ = 𝑇?
Constrained least squares minimization estimates activity1. Activity for noise ∈ (−∞, 1]2. Activity for detected signals ∈ [0,∞)
»0.3´ +0.2´ + 0.5´
Measurement Detected Signal 1 Detected Signal 2 Noise Estimate
D. Markovic / Slide 14
PerformanceMetrics
14
w Numberofdetectedbandsw Numberofextrabandsdetectedw RelativeErrorsinCenterFrequencyandBandwidth
D. Markovic / Slide 15
PerformanceMetrics
15
OurOutput
GroundTruthFullyConnectedBipartiteGraph
ComputedusingMaximumWeightMatching
EdgeWeights:
SymmetricDifference
1B 2B
2B 3B1B
9 564 8 10
w Numberofdetectedbandsw Numberofextrabandsdetectedw RelativeErrorsinCenterFrequencyandBandwidth
δ Bm1 , Bm2( ) = F − Bm1 ! Bm2
D. Markovic / Slide 16
MATLABSimulations:Performancevs.TReceiver:w Bandwidth100MHzw 512lengthFFT,averageof64
windowedoverlappingsegmentsw Upto50measurements,i.e.,8.32msw Parameters:Pfa = 0.01, 𝜏r = 0.1
16
802.11gTransmitters:w Eachnetwork=1AP+2STAsw Channels1,4,6,8,11w Saturateduplinkanddownlinkflowsw Shadowfadingchannelswith6dB
variance
NumberofDetectedSignals NumberofExtraSignals
D. Markovic / Slide 17
Simulation:Performancevs.SpectralOverlap
17
NumberofDetectedSignals
NumberofExtraSignals
RelativeErrorinBandwidth
FrequencyError
D. Markovic / Slide 18
USRPMeasurements:Example
w Device:USRPN210withCBXdaughtercard
w Measurementsat2.452GHzw 8-bitsamples@50MS/s
18
ColoredNoiseLearnt
802.11SignalsDetected
USRPMeasurements
Black arrows: detected supportsLabels: corresponding 802.11 channels
D. Markovic / Slide 19
MultipleUSRPMeasurements
w Contourplotof2Dhistogramofdetectedcenterfrequenciesandbandwidths
19
AndroidWiFiAnalyzerusedtoconfirm802.11channels6,7,8,and11inuse
D. Markovic / Slide 20
ConclusionsandFutureWork
w Noisepowerspectrumcanbeestimatedautomaticallywhensensingcommunicatingincumbentusers
w Multiplepowerspectrummeasurementscandistinguishrealworldspectrallyoverlappedsignalseveninunknownchannels
w Conventionalsignaldetectionandestimationtheorymaynotbesufficient
FutureWork:w Findstructuralpropertiesofoptimizationproblemtoreduce
computationalcomplexityw Estimatetimeofactivity,i.e.,,foruseintrafficestimation
20
A
Thankyou!
Questions?
ThismaterialisbaseduponworksupportedbytheNationalScienceFoundationunderGrantNo.1527026:DynamicSpectrumAccessbyLearningPrimaryNetworkTopology
D. Markovic / Slide 22
SelectedReferences[1] M. Laghate and D. Cabric, “Using the Time Dimension to Sense Signals with Partial Spectral Overlap,” in IEEE GLOBECOM, Washington, USA, 2016.[2] H. J. Wu et al., “A wideband digital pre-distortion platform with 100 MHz instantaneous bandwidth for LTE-advanced applications,” in 2012 Workshop on Integrated Nonlinear Microwave and Millimetre-Wave Circuits, 2012, pp. 1–3.[3] Z. Quan, S. Cui, A. H. Sayed, and H. V. Poor, “Optimal Multiband Joint Detection for Spectrum Sensing in Cognitive Radio Networks,” IEEE Transactions on Signal Processing, 2009.[4] I. Bisio, M. Cerruti, F. Lavagetto, M. Marchese, M. Pastorino, A. Randazzo, and A. Sciarrone, “A Trainingless WiFi Fingerprint Positioning Approach Over Mobile Devices,” IEEE Antennas Wirel. Propag. Lett., vol. 13, pp. 832–835, 2014.[5] M. Ibrahim and M. Youssef, “CellSense: An Accurate Energy-Efficient GSM Positioning System,” Veh. Technol. IEEE Trans. On, vol. 61, no. 1, pp. 286–296, Jan. 2012.[6] H. Yilmaz, T. Tugcu, F. Alago¨z, and S. Bayhan, “Radio environment map as enabler for practical cognitive radio networks,” IEEE Commun. Mag., vol. 51, no. 12, pp. 162–169, Dec. 2013.[7] S. Chaudhari and D. Cabric, “Cyclic weighted centroid localization for spectrally overlapped sources in cognitive radio networks,” in 2014 IEEE Global Communications Conference (GLOBECOM), Dec. 2014, pp. 935–940.[8] J. Wang and D. Cabric, “A cooperative DoA-based algorithm for localization of multiple primary-users in cognitive radio networks,” in IEEE GLOBECOM, Dec. 2012, pp. 1266–1270.[9] L. Wei, P. Dharmawansa, and O. Tirkkonen, “Multiple Primary User Spectrum Sensing in the Low SNR Regime,” IEEE Transactions on Communications, vol. 61, no. 5, pp. 1720–1731, May 2013.[10] M. Laghate and D. Cabric, “Identifying the presence and footprints of multiple incumbent transmitters,” in 2015 49th Asilomar Conference on Signals, Systems and Computers, 2015, pp. 146–150.[11] M. Laghate and D. Cabric, “Cooperatively Learning Footprints of Multiple Incumbent Transmitters by Using Cognitive RadioNetworks,” submitted to IEEE Transactions on Cognitive Communications and Networking, Sept. 2015.[12] T.-H. Yu, O. Sekkat, S. Rodriguez-Parera, D. Markovic, and D. Cabric, “A Wideband Spectrum-Sensing Processor With Adaptive Detection Threshold and Sensing Time,” IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 58, no. 11,pp. 2765–2775, Nov. 2011.[13] A. Kumar, V. Sindhwani, and P. Kambadur, “Fast conical hull algorithms for near-separable non-negative matrix factorization,” in International Conference on Machine Learning, 2013.[14] D. Donoho and V. Stodden, “When Does Non-Negative Matrix Factorization Give a Correct Decomposition into Parts?,” in Advances in Neural Information Processing Systems, MIT Press, 2004, pp. 1141–1148.
22