Adaptive Demodulation Techniques for Next Generation Software ...

Modulation ClassifierFrom: http://www.ottawa.drdc-rddc.gc.ca

1 2 34 5 67 8 90What is Automatic ModulationClassification ?signalFilterA/DIFDemodulationDemodulationDemodulationDemodulatedsignalLOA non-cooperativecommunicationtechnique which usesstatistical methods toestimate the signalmodulation typesAutomatic ClassificationFilterChannelEqualizerBWCenter Estimation ChannelFrequencyEstimation Symbol RateEstimationEstimationStatistical EstimationModulationRecognition……………………………………… SNREstimation PhaseEstimation

Modulation ClassificationSNR EstimationTemplates BuildingAnalogAnalogPreprocessingAnalogFeatureExtractionPSK/QAMPreprocessingPSK/QAMFeatureExtractionPreprocessed IFCoarse ModulationEstimationDigitalFSK/MSKPreprocessingFSK/MSKFeatureExtractionA non-cooperativecommunication techniquewhich uses statisticalmethods to estimate themodulation type of aunknown signalAnalogModulationEstimationPSK/QAMModulationEstimationFSK/MSKModulationEstimationClassificationDecisionEstimationConfidenceRatingUnknownTypeModulation SchemeClassification ConfidenceModulation ParametersFailure

SDR Applications (1)‣ Overcome channel fading‣ Monitor communication spectrum‣ Remove co-channel interferences1 2 34 5 67 8 901 2 34 5 67 8 90145 2 3678 90145 2 3678 90145 2 3678 90

1 2 34 5 67 8 901 2 34 5 67 8 90SDR Applications (2)Deep Space CommunicationReduce thescheduling andconfigurationburdens ofcommunications52341678 90

Modulation ClassificationOverview

Feature Extraction: Amplitude,Differential Phase, and Frequency•Input: IF•Feather: Amplitude, phase, diff phase, frequency•Statistics: histogram, STD•Classifier: max correlation, decision tree•Reference: Liedtke 1984baud ratedetectorBPFtiming circuittemplatesIFQI/tan -1Δφ2ΔtΔφΔTFreq.STDDeltaphasehistogramcorrelationrecognition treeCWPSK2PSK4PSK8ASK2FSK2( ) 2 ( ) 2 +sqrtAmplitude

Higher-order Transform of Constellationsc20c21c40c41c42c60c61c62c63c80c81c82c83c84QPSKPSK8PSK16V29-8 and V29-16 constellationsQAM16QAM64QAM32QAM4-12QAM16-16QAM44-20The 4th order constellations

Mr k2 nd order1 st orderHigher-order StatisticalFeaturesPower-law:r 4 ( k )4 th orderdominantpoints4 th order transformation of QPSKG(Hi| rK) =K M⎧i1 ( i)⎨ ∑ pj( rk)= ⎩M= ⎭ ⎬⎫1 i j 1∏k2 nd orderMoment:mK440= ∑ r ( k)k = 1Cumulant:C240= m40− 3m20QI4 th order transformation of QAM16

Cumulants vs. SNRs

Cyclic Spectral AnalysisBasebandRaxx*Time varying autocorrelation1 T / 2( τ ) = lim→∞ ∫ R * ( t,tT − T / 2 xxTSaxx*( f )*{*x(t)x ( )}* ( t,t + τ ) = E t + τCyclic autocorrelationSpectrum correlation density+ τ ) e− j 2πat∫ ∞ a − j 2 τ= τπ fR ( ) e dτ−∞xxdtR xxDecision Templates•Input: IF•Features: cycle frequencies•Reference: Menguc, 2004Cycle FreqTheoretical spectrum correlation magnitudeGardner and Spooner 1992

BasebandFeature Classification: MaximumLikelihood (ALRT)TemplatesPDF BPSKPDF QPSKPDF 8PSKK∏k = 1K∏k = 1K∏k = 1(.)(.)(.)l( H | r(k))BPSKl( H | r(k))QPSKl( H8 | r(k))PSKMAXBPSKModulationScheme• Input: baseband• Feature: Complex envelop• Classifier: maximum likelihood• References: Polydoros andKim 1995unknownQPSK8PSK

ALRT1M i∑p( i)jMi j=1( i)p j( rk)Feature Classification: HistogramCorrelation( r )k1M i∑p( i)jMi j=1( r )kG ( Hp( i)ji( r| rk)KKM⎧i1 ( i )) = ∏ ⎨ ∑ pj( rk)= ⎩ = ⎭ ⎬⎫k 1 Mi j 1⎧( i)2⎫1 ⎪ rk− b ( j)⎪exp ⎨−2⎬2πσ⎪ 2σ⎩⎪⎭=2QuantizeHIST(i )p q(i )p qL(H | r ) =iKK∑k = 1log p( i)( r ) =kQQ( i)∑∑logp ( rk) ←∑q= 1 k∈Ωq k=1log p( i)qD( i)q(i )D q• Input: baseband/IF• Feature: frequency/diff phase• Classifier: max correlation• References: Liedtke 1984

Research on CommercialApplications

Research on Adaptive ModulationBased on SDR - Cooperative‣ Maintain a constent BER by varying modulation schemes‣ Modulation schemes: QPSK, 16QAM, and 64QAM‣ Data frame based modulation recognition‣ A pilot symbol is used in forward channel‣ Reference: Jain, P.; Buehrer, R.M, “Implementation of adaptive modulation on the Sunrisesoftware radio,” The proceedings of the 45th Midwest Symposium on Circuits and Systems,Volume: 3 , 4-7 Aug 2002. Pages:III-405 - III-408TransmitterPilotDATASlow FlatFadingChannelsPilotDATAReceiverPilotDataFeed BackChannel

Why Applying Non-cooperativeDemodulation‣ Environment limitation and restriction‣ Elimination of the signal overhead information‣ Attractive for packet data servicesRFAir-interfacePreprocessingIFDelayerChooseDemodulatorDemodDataoutputModulationRecognitionPreprocessingModulationRecognition

Deference Between Military andCommercial Applications. .SIGINT SDR .Real time classification demodulationSNR low highCandidates unlimited limitedQoS friend / foe packet lossPulse shape unknown knownBandwidth unknown knownBaud rate unknown knownBlindness more less

Assume:‣ Equally likely‣ Symmetrical QAM/PSKPurpose:‣ Reduce the processing timeIssues:Research of Nolan et al.Reduce Form Constellation‣ Low utilization of availableinformation‣ May need longer data length forrandomnessQConstellation for QPSKQReduced form constellationII

Adaptive Receiver – Ishii et al.‣ Automatically recognize BPSK, QPSK, 8PSK, pi/4QPSK,16QAM, FSK, MSK, GMSK, AM, FM, CW, and SSB usingdecision tree for spectrum, variance, and baud detectionanalysis.ThresholdsTransmitter•SpectrumDecisiontreeRFBPFIF•Envelope•Baud•AmplitudeModulationEstimationOSC•PhaseDemodulationDataoutput

Blind Modulation EstimationUmebayshi et al.‣ Automatically recognize BPSK, QPSK, 8PSK, and 16QAM usingamplitude and differential phase variances. Channel gainestimation is discussed (2000).‣ Automatically recognize BPSK, QPSK, and 8PSK usingTransmitter differential phase and maximum likelihood test.RFBPFIFDemodulationDataoutOSCNoise VarianceEstimationMaximum LikelihoodModulation Estimation

Research of Menguc and Jondral (1)Air Interface Identification for SDRIdentify (Verify?)‣ TDMA-GMSK‣ OFDM-PSK/QAM‣ CDMA-QPSKTransmitterRF•DetermineNumber ofInterfaces•EstimateCarrier andBWPreprocessingBase bandCalculateCyclicAutocorrelationDemodulationRecognizeAir InterfacesLikelihoodTestDataoutputThreshold

Research of Menguc and Jondral (2)Magnitude Plot of the Cyclic AutocorrelationEstimationsIssues‣ Processing speed‣ Need universal front endGMSK OFDM CDMA* O. Menguc, “Air interface identification for software radio systems,” Ph.D. Dissertation, University of Fridericiana Karlsruhe, Nov. 30, 2004.

Research of Simon and Divsalar (1)Data Format classification for SDR•Discriminate NRZ andManchester code•Reduce complexity•Extend to non-coherent caseUse two curves approximateln cosh(x) in order to simplifythe ML computationsmallK −1∑n=0−1⎛2 2Pln cosh⎜r⎝ N0k , NRZ⎞⎟

Research of Simon and Divsalar (2)Reduced Complexity ML ImplementationBasebandDataSNREstimation

Other Research Results on ModulationClassification Based on SDR‣ Gu et al. “Channelized receiver platform of SDR based onFPGAs, Proceedings of The 5 th IEEE International conferenceon ASIC, Vol.2, Oct. 2003, pp.840-843.‣ Yang, “An enhanced SOFM method for automatic recognitionand identification of digital modulations,” Proceedings of the 2 ndIEEE International Workshop on Electronic Design, Test andApplications (DELTA’04), Jan. 2004, pp.174-179.‣ Ko et al. “Modulation type classification Method using wavelettransform for adaptive demodulator,” Proceedings of 2004International Symposium on Intelligent Signal processing andCommunication System, Vol.46, Oct. 1995, pp.211-222.‣ Hooftand Darwish, “A reconfigurable software digital radioarchitecture for electronic signal interception, identification,communication and jamming,” COTS Joural, April 2002, pp.31-35.

RF signalNoise PowerEstimationQuantization andAddress MappingNormalizationLook UpTableDecisionProcessConfidenceCheckEstimated ModulationSchemeChoose DemodulatorBandPassFilterPulse Timing andMatched FilteringDelayerDemodulationProcessReal-timeDataOutputLocalOscillatorCarrier and CarrierPhase Tracking

Summary‣ Adaptive modulation is not only an importantinformation warfare practice but also an effective toolto maximize the data capacity and minimize thetransmission error in SDR applications.‣ Automated modulation classification is a solution inhandling the non-cooperative communicationproblem for SDR.‣ Blind estimation of modulation parameters such ascenter frequency offset, carrier phase, pulse shape,symbol rate, and bandwidth is critical to therobustness of modulation classification.‣ A good modulation classifier should be able toidentify modulation scheme fast and robust.

Future Work‣ Faster estimator‣ Shorter data length‣ Lower SNR‣ Better channel estimation‣ Better QoS