15 BPF 01 - home.deib.polimi.ithome.deib.polimi.it/rech/Download/15_BPF_01.pdf · SignalRecovery,...
Transcript of 15 BPF 01 - home.deib.polimi.ithome.deib.polimi.it/rech/Download/15_BPF_01.pdf · SignalRecovery,...
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Sensors, Signals and Noise 1
COURSEOUTLINE
• Introduction
• SignalsandNoise
• Filtering:Band-PassFilters1- BPF1
• Sensorsandassociatedelectronics
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Band-Pass Filters 1 2
• Narrow-BandSignals
• RecoveringNarrow-BandSignalsfromNoise
• MovingSignals inFrequency(SignalModulation)
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Narrow-Band Signals 4
Powersignalswithanarrowpowerspectrum,thatis,apeakwith• center-frequencyfs• bandwidthΔfswhichissmall inabsolutevalue,typicallyΔfs <10Hz,
and/orwithrespecttothecenterfrequencyΔfs <<fsTheyapproximatewellasinusoidoverawidetimeintervalTs ≈1/Δfs
QUESTION:howcanwemeasuresuchnarrow-bandsignals inpresenceofintensewhitenoise?Andwhatifalso1/fnoise ispresent?
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Recovering Narrow-Band Signals from noise 6
Let’sseesometypicalexamplesofsignalswith• narrowlinewidthΔfs=1Hz• smallamplitudeVs ≤100nV
forbringingthemtohigherlevel(suitableforprocessingcircuits:filters,meters,etc.)theyareamplifiedbyaDC-coupledwide-bandpreamplifierwith
§ upperband-limitfh =1MHz§ noisespectraldensity(referredtoinput) with
«white»componentand1/f componentwithcornerfrequencyfc=2kHz
Letusconsiderthreecaseswithdifferentcenter-frequencyfs :Ø Case1:high frequencyfs = 100kHzØ Case2:moderatelylow frequencyfs = 1kHzØ Case3:low frequencyfs = 10Hz
5 /bS nV Hz=
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Recovering Narrow-Band Signals 7
CASE1:signalVs ≤100nV athighfrequencyfs = 100kHz
a) observingthevoltagewaveformintheTIMEDOMAIN,i.e.onoscilloscopedisplay
Thesignaltoberecoveredisatfrequencyfs = 100kHzmuchhigherthanthenoisecornerfrequencyfc=2kHz,sothatwecanuseasimplehigh-passfilterwithband-limitfi =10kHztocutoffthe1/fnoiseandobtainarmsnoise(referredtothepreampinput)
andtherefore
EventhehighestsignalVs =100nV ispracticallyinvisibleontheoscilloscopedisplay!Thenoisecoversaband ≈5xrmsvalue≈20μVandthesinusoidalsignal isburiedinit!
20,02 1S
n
VSN v= ≤ <<
( )2 5n b h i b hv S f f S f Vµ= ⋅ − ≈ ⋅ =
Verticalscale50μV/div
Horizontalscale5μs/div
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Recovering Narrow-Band Signals 8
CASE1:signalVs ≤100nV athighfrequencyfs = 100kHz
b)observingthepowerspectruminFREQUENCYDOMAIN,i.e.onspectrumanalyzerdisplay
SIGNAL:thepower iswithinabandwidthΔfS =1Hz
sothattheeffectivepowerdensityofthesignalis
NOISE:theeffectivepowerdensityatfs = 100kHzis
Onthespectrumanalyzerdisplaythesignalpeakisvery wellvisibleabovethenoise!
Conclusion:goodS/Ncanbeobtainedwithabandpassfilterhavingbandwidth∆𝑓#matchedtothesignalband∆𝑓# ≈ ∆𝑓%
216 250 10
2S
SVP V−= = ⋅
70SS
S
PS nV Hzf
≈ =Δ
5 /bS nV Hz=
14 1SS S S
b b b b b
SP S fSN S f S f S
Δ= = ≈ =
Δ Δ?
14 1S
b
SS
= ?
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Recovering Narrow-Band Signals 9
CASE2:signalVs ≤100nV atmoderatelylowfrequencyfs = 1kHza)observingthevoltagewaveformintheTIMEDOMAIN,i.e.onoscilloscopedisplayThesignal isnowatfs = 1kHzjustbelowthecornerfrequencyfc=2kHz.Forreducingthe1/fnoisewecanstilluseahigh-passfilter,butinordertopassthesignaltheband-limit fi mustbereduced:fi <<fs = 1kHz,typicallyfi =100Hz.Thermsnoisereferredtotheinput is
andtherefore
Thesituation ispracticallyequaltothatofCase1:thesignal ispracticallyinvisibleontheoscilloscopedisplay,it’sburiedinthenoise!
( )2 ln ln 5h hn b h i b c b h b c b h
i i
f fv S f f S f S f S f S f Vf f
µ⎛ ⎞ ⎛ ⎞
≈ − + ≈ + ≈ ≈⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠
20,02 1S
n
VSN v= ≤ <<
ln hb c b h
i
fS f S ff
⎛ ⎞⎜ ⎟⎝ ⎠
=1/fnoiseisnegligible
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Recovering Narrow-Band Signals 10
CASE2:signalVs ≤100nV atmoderatelylowfrequencyfs = 1kHzb)observingthepowerspectruminFREQUENCYDOMAIN,i.e.onspectrumanalyzerdisplay
SIGNAL:thepower iswithinabandwidthΔfS =1Hz
sothattheeffectivepowerdensityofthesignalis
NOISE: duetothe1/fnoise,theeffectivepowerdensityatfs = 1kHzissomewhathigher
Anyway,onthespectrumanalyzerdisplaythesignalpeakisstillwellvisibleabovethenoise
Conclusion: abandpassfilterwithbandwidth∆𝑓# matchedtothesignal∆𝑓# ≈ ∆𝑓%stillgivesafairlygoodS/N
( ) 1 3 8,7c cn s b b b b
s s
f fS f S S S S nV Hzf f
= + = + = ⋅ ≈
216 250 10
2S
SVP V−= = ⋅
70SS
S
PS nV Hzf
≈ =Δ
( )8 1S
n S
SS f
= >
( ) ( ) ( )8 1SS S S
n S b n S b n S
SP S fSN S f f S f f S f
Δ= = ≈ = >
Δ Δ
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Recovering Narrow-Band Signals 11
CASE3:signalVs ≤100nV atlowfrequencyfs = 10Hza)observingthevoltagewaveformintheTIMEDOMAIN,i.e.onoscilloscopedisplayThesignal isnowatfs = 10Hzmuchbelowthecornerfrequencyfc=2kHz.Forreducingthethe1/fnoisewecanstilluseahigh-passfilter,butwithstronglyreducedband-limitfi <<fs = 10Hz,typicallyfi =1Hz.Thermsnoisereferredtoinputis
andtherefore
( )2 ln ln 5h hn b h i b c b h b c b h
i i
f fv S f f S f S f S f S f Vf f
µ⎛ ⎞ ⎛ ⎞
≈ − + ≈ + ≈ ≈⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠
20,02 1S
n
VSN v= ≤ <<
Thesituation ispracticallyequaltothatofCase1:thesignal ispracticallyinvisibleontheoscilloscopedisplay,it’sburiedinthenoise!
ln hb c b h
i
fS f S ff
⎛ ⎞⎜ ⎟⎝ ⎠
=1/fnoiseisnegligible
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Recovering Narrow-Band Signals 12
CASE3:signalVs ≤100nV atlowfrequencyfs = 10Hzb)observingthepowerspectruminFREQUENCYDOMAIN,i.e.onspectrumanalyzerdisplay
SIGNAL:thepower iswithinabandwidthΔfS =1Hz
sothattheeffectivepowerdensityofthesignalis
NOISE:duetothe1/fnoise,theeffectivepowerdensityatfs = 10Hzisnowmuchhigher
Onthespectrumanalyzerdisplaythesignalpeakisbarely visible,it’sequaltothenoise
Conclusion:theS/Nisinsufficientevenwithabandpassfilterwithnarrowbandwidth∆𝑓#matchedtothesignal∆𝑓# ≈ ∆𝑓%
( ) 1 14,2 71c cn s b b b b
s s
f fS f S S S S nV Hzf f
= + = + = ⋅ ≈
216 250 10
2S
SVP V−= = ⋅
70SS
S
PS nV Hzf
≈ =Δ
( )1S
n S
SS f
≈
( ) ( )1SS S S
b b n S b n S
SP S fSN S f S f f S f
Δ= = ≈ ≤
Δ Δ
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Recovering Narrow-Band Signals 13
SUMMARY• Foranarrow-bandsignalplungedinwhitenoise(i.e.withfrequencyfS higher
thanthe1/fnoisecornerfrequencyfc)abandpassfiltermatchedtothesignal
bandisveryefficientandmakespossibletorecoversignals evensosmallthat
theyareburiedinthewide-bandnoise.
• Foranarrow-bandsignalplungedindominant1/fnoise(i.e.withfS lowerthan
the1/fnoisecornerfrequencyfc)abandpassfiltermatchedtothesignal isstill
quiteefficientandinmanycasesmakespossible torecoverthesignal.However,
ifweconsidersignalsatprogressivelylowerfrequencyfS,the1/fnoisedensity
at fSprogressivelyrises,sothattheavailableS/Nisprogressivelyreduced.
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Recovering Narrow-Band Signals 14
OPEN QUESTIONS• Weneedefficientband-passfilterswithverynarrowband-width.
Weneedtounderstandhowtodesignandimplement suchnarrow-bandfilters,
butweshalldealwiththisissueafterdealingwiththefollowingquestion.
• Iftheinformationiscarriedbytheamplitude ofalow-frequencysignal,ithasto
facealso1/fnoise.Itwouldbeadvantageoustoescapethisnoisebypreliminarly
transferringtheinformationtoasignalathigherfrequency.However:
a)howcanwetransferthesignaltohigherfrequency?
b)ifwetransfertothehigherfrequencyalsothe1/fnoisethatfacesthe
signal,thismakesthetransferuseless: howcanweavoidit?
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Recovering Narrow-Band Signals 15
OPEN QUESTIONS• Forescaping1/fnoise,alow-frequencysignalshouldbetransferredtohigherfrequency
beforeitmixeswith1/fnoiseofcomparablepowerdensity:thatis,frequencytransfer
shouldbedonebeforethestagewherethesignalmeets the1/fnoisesource.
• Thefrequency-transferstageshavetheirownnoise,withdifferentintensity indifferent
types.Unluckily,thetypeswithlowestnoisebearotherdrawbacks,typicallyalimited
capabilityoftransfer,restrictedtomoderatelyhighfrequency.
• Forachievingourgoal,thesignalmustbehigherthanthenoisereferredtotheinputof
thefrequency-transferstage.Ifwithagivenstagethesignalisnothighenough,
preamplifyingisnotadvisablebecauseapreampbringsits1/fnoise.Inmostcases itis
bettertotransferthesignal«asitis»bymeansofafrequencytransferstagewithlower
noiseandacceptthelimitationsofthisstage,typicallyamoderateoperatingfrequency.
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Amplitude Modulation with DC signal (ideal) 17
• Informationisbroughtbythe(VARIABLE)amplitudeA ofaDCsignalx(t)=A .(NB:arealDCsignal isasignalatverylowfrequencywithverynarrowbandwidth)
• Ananalogmultiplier circuitcombinesthesignalwithasinusoidalwaveformm(t)(calledreferenceorcarrier)withfrequencyfm andCONSTANT amplitudeB
• Theinformationistransferredtotheamplitudeofasinusoidalsignaly(t) atfrequencyfm
y(t)=A·B cos(2π fmt+φm)
ANALOGMULTIPLIER
x(t) y(t)=x(t)·m(t)A
t
t t
B
A·B
m(t)=Bcos(2π fmt+φm)m(t)
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Amplitude Modulation with DC signal (ideal) 18
y(t)=x(t)·m(t)=
=A·Bcos(2π fmt+φm)
A
t
t
B
t
A·B
TIMEDOMAIN FREQUENCYDOMAIN
Aδ(f)
f
f
f
|X|x
m
y
fm- fm
- fm fm
|M|
|Y|
( )2 mB f fδ −( )
2 mB f fδ +
Y(f)=X(f)*M(f)
( )2 mBA f fδ⋅ + ( )
2 mBA f fδ⋅ −
Thesignal isshiftedinfrequencyby+fm and–fmandinphaseby+φm and– φm respectively
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Convolution in the Frequency Domain 19
Inthetimedomain(TD) the amplitudemodulation isthemultiplicationofthesignalx(t)(withvariableamplitudeA)
bythereferencewaveformm(t) (withstandardamplitudeB)
Inthefrequencydomain(FD) itistheconvolutionofthetransformedsignalX(f) bythetransformedreferenceM(f)
( ) ( ) ( )y t x t m t= ⋅
( ) ( ) ( ) ( ) ( )Y f X f M f X M f dα α α∞
−∞= ∗ = −∫
Convolutionismorecomplicated inFDthaninTDbecause:1. thefunctionstobeconvolvedaretwofold,thatis,theyruninthepositive
andnegativesenseofthefrequencyaxis2. Complex valuesmustbesummedateveryfrequencyforobtainingY(f).IngeneraltheresultofFDconvolutionisnotasintuitiveasthatofTDconvolutionandthemodule|Y(f)|isNOTgivenbytheconvolutionof|X(f)|and|M(f)|
wemustfirstcomputetherealandimaginarypartsofY(f)andthenobtain|Y(f)|
( ) ( ) ( )Y f X f M f≠ ∗
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Amplitude Modulation with Narrow-Band Signal 20
Inthecaseshereconsidered, however,the issue isremarkablysimplified becausea) X(f) isconfinedinanarrow bandwidthΔfsb) M(f)hasalinespectrumwith(fundamental)frequencyfm thatismuchgreater
thanthesignalbandwidthfm >>ΔfsIntheconvolutionX(f)*M(f) eachlineofM(f)actsonX(f) asfollows§ ShiftsinfrequencyeverycomponentofX(f) by+ fm and -fm
(i.e.addstoeachfrequency+ fm and -fm )§ ShiftsinphaseeverycomponentofX(f) by+φm and - φm
(i.e.addstoeveryphase+φm and - φm )IncaseswithΔfs << fm,thereisnosumofcomplexnumbers tobecomputedbecauseatanyfrequencyfthereisatmostonetermtobeconsidered,allothertermsarenegligible.Theresultoftheconvolutioniseasilyvisualized:everylineofM(f) shifts X(f) infrequencyandaddstoX(f) itsphase.Therefore,|Y(f)|iswellapproximatedbytheconvolutionof|X(f)|and|M(f)|and|Y(f)|2 bytheconvolutionof|X(f)|2 and|M(f)|2
( ) ( ) ( )Y f X f M f≅ ∗ ( ) ( ) ( )2 2 2
Y f X f M f≅ ∗
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Amplitude Modulation with Narrow-Band Signal 21
y(t)=x(t)·m(t)
f
f
f
|X|
fm- fm
- fm fm
|M|
|Y|
( )2 mB f fδ −( )
2 mB f fδ +
Y(f)=X(f)*M(f)
( )2 mB X f f+ ( )
2 mB X f f−
m(t)=Bcos(2π fmt+φm)
( ) ( )1 S
tT
S
Ax t t eT
−
= ⋅
( ) 11 2 S
X f Aj f Tπ
=+1
S
fTπ
Δ ≈
FREQUENCYDOMAINExampleofquasi-DCNBsignal(withverylongTS)
with fm >>1/TS
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Amplitude Modulation with Sinusoidal Signal 22
f
f
f
|X|
fm- fm
- fm fm
|M|
|Y|
( )2 mB f fδ −( )
2 mB f fδ +
( ) cos (2 )Sx t A f tπ= ( )2 SA f fδ + ( )
2 SA f fδ −
( )4 m SAB f f fδ − −( )
4 m SAB f f fδ − +( )
4 m SAB f f fδ + + ( )
4 m SAB f f fδ + −
fS-fS
( ) cos (2 )m mm t B f tπ ϕ= +
with fm >>fS
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Amplitude Modulation with sinusoidal signal 23
Byexploitingtheawellknowntrigonometric equation
incaseswithsinusoidalsignalandsinusoidal reference
theresult isdirectlyobtained
( ) cos(2 )Sx t A f tπ=
( ) cos(2 )m mm t B f tπ ϕ= +
( ) ( )( ) ( ) ( ) cos 2 cos 22 2S m m S m mAB ABy t x t m t f f t f f tπ ϕ π ϕ= ⋅ = − − + + +⎡ ⎤ ⎡ ⎤⎣ ⎦ ⎣ ⎦
( ) ( )1 1cos cos cos cos2 2
α β α β α β= − + +
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Squarewave Amplitude Modulation 24
Modulationwithasquarewavereferencem(t) canbeimplementedwithcircuitsbasedsimplyonswitchesandamplifiers,withoutanalogmultipliers
A
t
A·Bx(t)
t
t
y(t)=x(t)·m(t)
B
B
• Insuchcases,thecircuitnoisereferredtotheinputisduemostlytotheswitch-devices andismuchlowerthanthatofanalogmultiplier circuits.
• Metal-contactswitcheshavethelowestnoise,buttheycanoperateatlimitedswitchingfrequency,typicallyuptoafew100Hz
• Electronicswitches(MOSFET,diodes,etc.)operateuptoveryhighfrequenciesbuthavefairlyhighernoise(anywayMOSFETsoperatingasswitch-devicehavelowernoisethanMOSFETsoperatingasamplifierdevices).
+
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Squarewave Amplitude Modulation 25
y(t)=x(t)·m(t)
+Bx(t)
Switchingexample:differentialamplifierwithalternated inputpolarity
t
A·Bt
A
y(t)=x(t)·m(t)+Bx(t)
Switchingexample:chopper(ON-OFFmodulation)
t
A·Bt
A
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Squarewaves and F-transforms 26
Bmsq(t) =symmetricalsquarewave(from+1to-1)atfrequencyfm
ffm- fm 3fm 5fm
B·|Msq|
-3fm-5fm
2 Bπ
23B
π⋅ 2
5B
π⋅
2 Bπ
23B
π25B
π
[ ]2 12 1 2 1
0( ) ( ) ( )
2k
sq k kk
bB M f B f f f fδ δ∞
++ +
=
⋅ = ⋅ − + +∑ with ( )( )
2 1
2 1
(2 1)
1 42 1
k mk
k
f k f
bk π
+
+
= +⎧⎪
−⎨= ⋅⎪ +⎩
t
B·msq
B0
Intheamplitudemodulation:• eachlineofthereferenceMacts likeasimplesinusoidalreference,i.e.shiftsbyits
frequencyanditsphasethesignalXandmultiplies itbytheamplitudeB·b2k+1• ifthesquarewaveisnotperfectlysymmetrical(e.g.ithasasymmetricalamplitudeand/or
durationofpositiveandnegativeparts)thereisalsoafiniteDCcomponentwithamplitudeB0 (possiblyverysmall)
• theDCcomponentdoesNOTtransferthesignalXinfrequency,just«amplifies» itbyB0
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Squarewaves and F-transforms 27
t
B mch(t) =choppersquarewave(from+1tozero)atfrequencyfm
B·mch
ffm- fm 3fm 5fm
B·|Mch|
-3fm-5fm
Bπ 1
3Bπ⋅ 1
5Bπ⋅
2B
( )( (2
))2ch sqB M f B fB M f δ⋅ = + ⋅⋅
Bπ
13
Bπ⋅1
5Bπ⋅
2B
2B
ChoppersquarewavewithamplitudeB==Symmetricalsquarewavewithamplitude B/2+DCcomponentwithamplitudeB/2
Intheamplitudemodulationbyachopper:• a replicaofthesignalX«amplified»byB/2istransferredinfrequencybythesquarewave• anotherreplicaofX«amplified»byB/2isNOT transferred,itstayswhereitis
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Squarewave Amplitude Modulation 28
ffm- fm 5fm
B·|Msq|
-3fm-5fm
B0
|X|
ffm- fm 5fm
|Y|=B·|Msq|*|X|
-3fm-5fm
2 B Aπ
⋅
25B A
π⋅B0·A
A
3fm
3fm
2 Bπ 2
3B
π25B
π
23B A
π⋅
Narrow-bandDCinputsignal
Squarewavereference(slightlyasymmetricalcase)
Squarewavemodulatedoutputsignal
Signal Recovery, 2017/2018 – BPF 1 Ivan Rech
Summary and Prospect 29
• Asintuitive,narrow-bandfilteringisveryeffectiveforrecoveringnarrow-bandsignals immersed inwide-bandnoise
• Besideswide-bandnoise,however,othercomponentswithpowerdensityincreasingasthe inversefrequency(1/fnoise)areubiquitous inelectroniccircuitry(amplifiersetc.).Inthelow-frequencyrangetheyareindeeddominant.
• Atlowfrequenciesthe1/fnoiseaddedbythecircuitryisoverwhelming,sothatthesolutionofnarrow-bandfilteringbecomesprogressivelylesseffectiveandfinallyinsufficientforrecoveringsignalswithprogressivelylowerfrequency
• Aneffectiveapproachtorecoveralow-frequencysignal istomoveittohigherfrequencybeforetheadditionof1/fnoise.Thatis,tomodulatethesignalbeforethecircuitrythatcontainsthe1/fnoise sources
• Narrow-bandfilteringcanthenbeemployedtorecoverthemodulatedsignal;wewillnowproceedtoanalyzemethodsandcircuitsfornarrowbandfiltering