TIME-DOMAINCHARACTERIZATION OF

SAW DEVICES

Robert C. BrayTim L. BagwellRoger L. JungermanScott S. Elliott

Hewlett-PackardMicrowave Technology Division1412 Fountaingrove Parkway

Santa Rosa, CA 95401

RF & MicrowaveMeasurementSymposiumand Exhibition

rli~ HEWLETT~~ PACKARD

Time Domain Characterizationof SAW Devices

Surface Acoustic Wave (SAW) devices have assumed increasing importanceas signal processing components in the frequency range 20 MHz to 2 GHz.SAW resonators, filters, and delay lines offer performance unachievableby other means, and are attractive for their small size, repeatablecharacteristics, and low cost. Characteristic SAW device specificationsinclude insertion loss, bandwidths, out-of-band rejection, group delayripple, input and output impedances, and phase linearity. These arestraightforward measurements with a network analyzer with sufficientfrequency accuracy and dynamic range. SAW devices are allfundamentally based on the delay of acoustic signals travelling 5orders of magnitude slower than electromagnetic signals. Ideal deviceresponse is shown to be degraded by spurious acoustic reflections inthe device, which are a key limitation on device performance. Thebuilt-in time domain capability of the HP 8753A network analyzer isshown to be an indispensable tool for complete analysis of deviceresponse including time spurious.

Bob Bray, Project Manager, Physical Wave Device R&D, HP MicrowaveTechnology Division, Santa Rosa, CA. B.S., M.S., Physics, Universityof Michigan, 1966., 1967. Taught high school physics, 1969-1977. MSEE,Ph.D., Stanford University, 1979, 1981. Thesis work in acoustic andphotoacoustic microscopy. Joined HP in 1981, where he has been involvedwith R&D in acoustic and optical devices.

Tim Bagwell, R&D Engineer, HP Microwave Technology Division, SantaRosa, CA. BSEE, University of Illinois, 1980. Joined HP in 1980.Worked on HP 853A spectrum analyzer, and on millimeter-wave devicecharacterization. In 1983-84, worked on the antiproton source atFermilab, Batavia, Illinois. Returned to HP in 1984, where he has sincebeen involved in acoustic wave device R&D.

Roger Jungerman, R&D Engineer, HP Microwave Technology Division, SantaRosa, CA. B.A. Physics, University of California, Santa Cruz, 1978.With Watkins-Johnson Co., 1978-1981, working on thin film coatings.M.S., Ph.D, Applied Physics, Stanford University, 1983, 1985, wherethesis work concerned scanning optical microscopy, fiber optics, andacoustics. Joined HP in 1985, where he has been involved with acousticand optical device R&D.

Scott Elliott, section Manager, Wave Technology R&D, HP MicrowaveTechnology Division, Santa Rosa, CA. BSEE, MSEE, University ofCalifornia, Berkeley, 1969, 1971. Worked with several firms inmicrowave devices, 1971-1975. Ph.D., University of California, SantaBarbara, 1978, with thesis work in acoustic imaging. with HP since1978, working on GaAs FETs, acoustic devices, millimeter wave ~,integrated circuits, and optics.

-0.2

1. INTRODUCTION TO SAW TECHNOLOGY2. SAW DEVICE MEASUREMENTS IN

FREQUENCY AND TIME DOMAINSA. FILTERSB. DELAY LINESC. RESONATORS

3. CONCLUSION

- Phase velocity vR

SAW SURFACECONFINEMENT

1.0

ill 0.8Cl

~:J 0.6[L::;-0:ill 004>~_ 0.2illa::

oI-\~"-----T;;--=~~~--:;~1.5 2.0 2.5

DEPTH (WAVELENGTHS)

1

Surface acoustic waves can exist on the freesurfaces of solids. The natural waveguiding of thefree surface allows a wave whose particle motion atthe surface is a retrograde ellipse [4,5J Thesewaves exist as one of the components of earthquakes(the L wave or slowest to arrive) where they havewavelengths measured in kilometers and frequenciesof fractions of 1 Hz. They were predicted to existby Lord Rayleigh in 1885, and experimentallyobserved 20 years later [1J. The same type of waveon a crystalline substrate is the basis of SAWdevice technology, which has grown rapidly sinceits origin in 1965 [2J.

The free surface of a crystalline solid proVides anatural waveguide for SAWs. The amplitude decaysexponentially into the interior of the substratefor both the vertical and longitudinal componentsof the motion. Note that most of the energy istrapped within 1 wavelength of the surface [4,5J.This has the great advantage of access to signalprocessing elements which can be fabricated easilyon the exposed surface.

ACOUSTIC ATTENUATIONIN GASES, LIQUIDS & SOLIDS

~.(\ (\ Q I\~ = .z- VV V\!)"J ==

I ! ,A (t,z) = Ao cos (kz-wt)e- OCz

For most conditions, 1 medium0<

12 = Constant

ATTENUATION EXAMPLES(f= 100 MHz)

MATERIAL ATTENUATION WAVE TYPE

Air,STP 2000dB/mm LongidudinalWater 3dB/mm Longitudinal

ST-Quartz .OO8dB/mm SAWYZ-LiNb03 .OO2dB/mm SAW

2

A sound wave of any type is attenuated as ittravels through any medium, by scattering orviscous damping, for example. In a given medium theattenuation constant is typically proportional tothe square of the frequency.

For representative media, the 100 MHz soundattenuation values are shown here. Notice that incrystalline solids the attenuation for SAWs isorders of magnitude lower than for longitudinalsound waves in liquids and gases. This is thefundamental material property that makes SAWtechnology possible [4,6].

A SAW filter chip exhibits features common to allSAW devices. It consists of a piezoelectricsubstrate upon which input and output interdigitaltransducers (IDTs) are deposited. The IDTs, whichare metal electrodes connected to two bus bars asshown, convert electrical signals to surfaceacoustic waves and vice versa at the input andoutput of the device. Bidirectional transducerslaunch SAW energy in both directions. Eachelectrode in a bidirectional transducer is split tosuppress internal reflections within thetransducer. The split elctrodes are one eighthwavelength wide, separated by one eighthwavelength. The impulse responses of the twotransducers are shown in the center ofthe figure. Notice the correspondence betweenenvelope of the impulse response and the overlappattern of· the fingers in the transducers. Theconvolution of these impulse responses gives theoverall filter impulse response. The frequencyresponse of the transducer is obtained bymUltiplying the Fourier Transforms of the impulseresponses, apart from a phase factor due to theacoustic delay. Acoustic propagation loss must beconsidered at high frequencies [4,7].

NASTY REALITIES OF SAW

END REFLECTION......- n

MAIN SIGNAL

---~"-- -TRIPLE TRANSIT

~ BULKI IWAVES

SAW TRANSDUCER EQUIVALENT CIRCUIT

Some of the nasty realities (time spurious)encountered with SAW devices are indicated in thisslide. Waves launched towards the end of thesubstrate out the back of the transducer will bereflected towards the output transducer if anabsorbing layer is not deposited to suppress them.The triple transit signal is a major time spuriousand a major source of passband ripple. Its level issensitive to electrical matching conditions and isthus adjustable by the circuit designer employingthe SAW device. Direct electromagnetic coupling isa source of time spurious and passband ripple aswell. It can be reduced by increasing the delaypath and by careful packaging design. Interdigitaltransducers also launch longitudinal and shearwaves into the bulk of the crystal substrate, wherethey may reflect back to the output transducer,causing a spurious response, typically atfrequencies higher than the SAW passband.

A SAW transducer at its simplest may be repre­sented by the equivalent circuit shown here. ctrepresents the interdigital static capacitance,which is independent of frequency. Ga is theacoustic radiation conductance, representing SAWenergy launched away from the transducer. Itsfrequency dependence is corresponds to thatoutlined before. Ba represents non-radiatingenergy storage asociated with the acousticexcitation of the transducer [5].

SIGNAL GENERATOR

YIN(!)

SURFACE WAVE TRANSDUCER

Where: CT static lOT capacitanceGa(f) = acoustic radiation conductanceBa(f) = acoustic susceptance

SAW Technology - Why?

• 20 years old• Unique signal processing

20 MHz - 2 GHz+

• Low loss for acoustics incrystalline solids

• Piezoelectric coupling to circuits• An IC-Iike te(:hnology

*Repeatable process*Small, easy to mass produce*Inexpensive

3

SAW device technology is barely 20 years old. Ithas developed rapidly because it makes possibleunique signal processing capabilities in thefrequency range 20 MHz - 2 GHz. The lower limit isset by impractically large device size andavailibility of alternatives. At frequencies muchabove 1 GHz optical lithography is no longerfeasible, as the transducer fingers drop below 1micron in width. The technology is based on thefact that SAWs in the RF, UHF, and low microwavefrequency ranges experience low propagation loss incrystalline solids such as quartz and lithiumniobate that are piezoelectric and thus can coupleelectrical energy to acoustic energy. SAW devicefabrication is an integrated circuit - liketechnology, where many small, easy to mass producechips may be produced on a single wafer.

SAW DEVICE TYPES

• Bandpass Filters*Undirectional: low loss*Bidirectional: precise shape

• Delay lines• Resonators• Multipole resonators• Dispersive delay lines• Matched filters• Storage correlators

HIGH-VOLUMESAW APPLICATIONS

* TV IF Filters* Garage door openers* CATV set-top converters* SAW wireless labels?

SAW FILTER RESPONSE

RIPPLE AMPLITUDE

4

Many types of signal processing devices may befabricated with SAW technology. In this paper weexamine measurements necessary to characterize twotypes of SAW bandpass filters, as well as delaylines and resonators.

SAW filters for different applications have avariety of very demanding performance requirements.This results in the need to measure several aspectsof the filters' frequency and time domainresponses. These measurement needs are not onlypresent in the design process but also in themanufacturing of such devices. Typical SAW filterspecifications are insertion loss, ripple I

bandwidth, stopband bandwidth, stopband rejection,passband ripple amplitude, phase ripple amplitude,phase linearity, and group delay. Additionally,The more demanding applications require minimizingtime spurious signals resulting in specificationsin the time domain as well as those in thefrequency domain listed above.

BRICKWALL FILTER LAYOUT

A brickwall filter is one with a shapefactor approaching unity. These filter shapes areusually designed to yield the maximum possiblepassband bandwidth while providing adequaterejection of spurious signals such as thefeedthrough of a local oscillator. A typicalapplication is to provide adjacent channel andspurious rejection at an IF frequency. Almost alltelevision tuners now manufactured in the worldcontain a SAW IF filter of this type foradjacent-channel rejection. Brickwall filters aretypically designed with two bidirectional apodizedIDTs, of the split finger type, coupled via amUlti-strip coupler. The use of an MSC allowshigher stopband rejection per device since twoapodized lOTs can be used. Another advantage ofthe MSC design is that the lOTs can be offsettransversely from one another, thus reducingcoupling to bulk modes.

127. 00 0 o MHz

1

( \\

...... ,w·."

CHI 821

Co,..

log MAG 10 d8/ REF 0 dB 1.1-22.039 dB

The frequency response of a typical brickwallfilter is shown in this figure. In order to improvethe dynamic range of the instrument, the IFbandwidth was reduced to 30 Hz. Marker 1,positioned at the peak of the response, shows aminimum insertion loss of 22.04 dB. The stopbandrejection can be seen to be about 50 dBc.

CHI CENTER 1:34. 000 000 MHz SPAN 50.000 000 MHz

o. 00 oeo MHzBW. 15. 12 1 B MHz

CQntl 133 8314 4 MHz

1 Ch B. 6275

~ ~

/ \I \I \

'I

CHI 821

Cor

log MAG 10 dB/ REF 0 dB l.J .0074 dB

This figure shows the filter passband. Here marker1 has been zeroed so that the marker width searchfeature can be used to determine the 3 dBbandwidth. Markers 3 and 4 are positioned at points3 dB lower than marker 1. The user can immediatelyread the 3 dB bandwidth which is 15.51 MHz. Thecenter frequency of the filter has also beencomputed as the average of the 3 dB frequencies andcan be seen to be 133.83 MHz.

CH 1 CENTER 134. 000 000 MHz SPJ\N 20. 000 000 MHz

5

O. 00 0 o MHzSW. 19. 71 6 6 MHz

clillnt. 133 8936 4 MHz

,!, Q. 7. 0206

"""'"/ \

I \I \~ ~

CHl 521

Co~

log MAG 10 dB/ REF 0 dB 1.1 0095 dB

The 40 dB bandwidth, of 19.07 MHz, is just aseasily obtained using the marker width searchfeature. The shape factor can now be computed asthe ratio of the 40 dB bandwidth to the 3 dBbandwidth which gives a shape factor of 1.23 forthis filter. Such a shape factor is routinelyachievable with SAW filters due to theirtransversal character [4,5].

CHl CENTER 134.000 000 MHz 5PAN 20. 000 000 MHz

O. 00 DC o MHzsw. 13. 59 31 4 MHz

clillnt. 133 8858. 9 MHz

lJ. 9. 8018

fJ1\ II. ,J..M MA~ ~. AJv..~ 'V' IV ., f'{V 1\

4

CHl 521

Co~

log MAG 2 dB/ REF -21 99 d8 1.1 0066 dB

Reducing the scale to 0.2 dB/div allows thepassband ripple to be measured. Here marker 1 hasbeen used to shift the peak of the response to thereference line. The ripple amplitude is seen to beabout 0.4 dB peak to peak. It is a common practicewith brickwall filters to specify the ripplebandwidth. This bandwidth is defined to be thewidth of the response where the amplitude isreduced by an amount equal to the peak to peakripple amplitude. Again using the marker widthsearch feature to find the 0.4 dB bandwidth, theripple bandwidth is found to be 13.66 MHz.

CHI CENTER 184. 000 000 MHz SPAN 20. 000 DOD MHz

BRICKWALL FILTER REFLECTIONS

CHIPEDGE

REFLECTION

C

lOTBACKEDGEREFLECTION

--------------.~ SINGLE TRANSIT

)

'---------------••~ TRIPLE TRANSIT

6

Three different ripple periods dominate thepassband response. The ripple with the shortestperiod is due to beating between the single transitacoustic signal and the triple transit acousticsignal. An approximate formula for computing thisripple period is given by p = v/(2L) where p is theperiod in HZ, v is the acoustic velocity inmeters/second, and L is the center to centerdistance between IDTs in meters. One can also seea ripple component which is about twice that of thetriple transit ripple. This ripple is due tobeating between the acoustic signal and theelectromagnetic feedthrough signal which is coupledbetween the IDTs through stray parasiticcapacitance. The approximate period of this ripplecan be computed from p = v/L. The longest periodripple component is Fresnel ripple due totruncation of the IDT samples. It is related to thelength of the lDT and the number of fingers in theIDT.

CH2 START-20 ne

CH2

Cor

CH2

Cor

Got

521 log MAG 10 dB/ RE:F -25 dB L -26. 766 dB

I.' 92 us

I~\

~ I 1,.\ A ~

~tJl ~~ ~III

~' IIISTOP 4.98 us

521 leg MAG 10 dB/ REF -25 dB l.J -26. 769 dB

I. 92 we

I~.

I~ ~II~~jr ,A It, II. ILJ LI .1.

The time domain transformation capability is anindispensable tool for analyzing SAW filters. Herean 801 point transform has been performed. It isimportant to adjust the frequency sweep to give theproper time domain resolution. The simple formulafor determining the time domain resolution is t1jspan where t is the time resolution inmicroseconds, and span is the frequency span inMhz. In this case,the span is 50 Mhz which gives atime resolution of 20 nsec. The time resolutionshould be adjusted to give adequate sampling of thefinest detail of interest in the time response.The time response shows the computed impulseresponse for S21. Here again an IF bandwidth of 30Hz has been selected to improve the dynamic rangeof the instrument. A great deal of information canbe gathered about the filters performance from thetime response. Most notable is the impUlse responseof the single transit signal at 1.592 usec. This isan approximate picture of the convolution of theapodization pattern of the launching and receivingtransducers. The electromagnetic feedthrough at 0usec can also be observed to be approximately 47dBc. At 4.78 uS, the triple transit signal isobserved to be about 50dBc. These two time spurioussignals are responsible for most of the fastpassband ripple. In addition to these signals,many other spurious signals can be observed in theregion between the single transit and the tripletransit signals. These signals are due to acousticreflections from various interfaces such as IDTfront and back edges, MSC front and back edges,chip edges and others. By determining the times ofthese spurious reflections and knowledge of thedevice layout, one can determine the source of thevarious reflections.

Another powerful feature of the HP 8753A is thetime domain gating capability. This capabilityallows the user to gate out the spurious timesignals in order to determine the filters frequencyresponse without the interfering spurious signals.By placing the gate center at the peak of the timeresponse and adjusting the gate span to include allof the filter sidelobes, we can perform an inversetransform of the gated data to obtain the idealizedfrequency response.

CH2 START 20 n9 STOP 4.98 us

7

127. 00 0 o MHz

,I,

( \I \

~,...JtM Vl.M I~ ..(\/1 ~ ~

~~ r 'I II~~I i I~AI\hW

CHI' S21

Cor

Got

log MAG 10 dB/ REF 0 dB 1.J -22 061 dB

The gated frequency response shows improvedstopband rejection due mostly to reducedelectromagnetic feedthough as a result of thegating. Now the acoustic sidelobes are readilyobserved in the response.

CHI CENTER 134. 000 000 MHz SPAN 50.000 000 MHz

CHI S21

127. 00 DC o MHz

A1M \ 1\V V

Cor

Got

log MAG 2 d8/ REF -22 08 dB 1.J -22. 062 dB

The passband of the gated response shows ripplewhich is due to acoustic processes within the lOTregions only. Ripples caused by electromagneticfeedthrough, triple transit, and other spuriousacoustic reflections, have been suppressed by thegating. The gated response allows the user tostudy the acoustics in the lOT region itselfwithout the effects of spurious time reflections.It is also useful to see what the filter responsewould look like if the filter was actually embeddedin circuitry designed to minimize regenerativereflections and electromagnetic feedthrough. Theuser can, however, make these measurements in asimple 50 ohm system.

CHI CENTER 134.000 000 MHz SPAN 50.000 000 MHz

127. 00 0 o ...

MAR ER 11~7. MH ..

.0. ~VV' "VT

CHI 921

Cor

100 ne/ REF 1- 589 ue L 1 5892 ue

This figure shows the group delay over the passbandregion. The group delay is a measurement of thetime delay of the filter as a function offrequency. Deviation of the group delay from aconstant value over the passband region can lead tosignal distortion since different frequencycomponents are delayed by different amounts. Thisis especially important for broadband filters.

CHI CENTER 134.000 000 MHz SPAN 20. 000 000 MHz

8

A response at three times the fundamental frequencyis also observed in this particular filter. Thisresponse is produced by the inherent spatialasymmetry of the electric field produced by thesplit finger lOT's commonly used in brickwallfilters.

.1.; -28 18 dB

STOP 500 . 000 000 MHz

10 dBI REF 0 dB

.300 000 MHz

log MAG

407. 85 4 o MHZ

1

/

~

~rSTAAT

CHi 521

Coe

1 :34. DOD 000 MHz

Measurement of 511 allows quick determination ofthe lOT radiation conductance and capacitance. Themarker can be set to display impedance oradmittance. In this case, admittance is dispayed.The value of capacitance or inductance is alsocomputed and displayed. This allows equivalentcircuit parameters to be measured at a glance.

CT

YIN(fj

SURFACE WAVE TRANSDUCER

.l.o 07.3:38 lOIS 22.97 illS 27.282 pI'

1 :34. 000 DOD 14Hz

SIGNAL GENERATOR

SPAN 50. 000 000 MH:Il

1 U FS

CH2 S.-:-"c--,-1..:0-,,9'---,MrA_G_-.--_'_D-,d_8_/_,RE_F_D~d_8_-,-_-=L;--.:.5.:.'.:...9::r1:..;3:.....:d::8:,...

:3. 88

Transforming 511 frequency data to the time domaincan give yet more insight into the operation of thefilter. Here the main response is due toelectrical reflections at the launching lOT. Marker1 is placed on the main peak of the signal whichwas reflected from the receiving lOT at 3.188 uS.Again, other spurious reflections can be observedin the region between the single and double transitsignals.

Cor

9

BULK MODE PROPAGATION PATHS

LaunchinglOT

--j r-rReceiving

lOT

SURFACE

W

CHIP BOTTOM

Various wave modes, other than the intended SAW,can be launched by the IDT's. Each mode propagateswith a certain angle to the surface for a givencrystal cut. From a knowledge of the slownesssurface for a particular cut, one can compute boththe group velocity and the phase velocity and theirassociated angles for each mode. Once these areknown, the resonant frequency of the mode and thetime delay can be computed. Wagers [9] has studiedthe various bulk modes in 128 Y LiNb03 and hascalculated the velocities and angles. Some ofthese modes propagate along the surface similar tothe SAW. Others propagate into the bulk andreflect from the bottom of the crystal. Uponreflection, some of the energy can be coupled to adifferent mode. Mode conversion at the bottom ofthe crystal accounts for some of the moretroublesome spurious signals in SAW filters builton 128 rotated Y LiNb03.

Vpifi = = bulk mode resonant frequency

r cos8pi

T__d1 _d2+ = time delay of bulk mode- Vg1 Vg2

di = W csc 8gi = pathlength , i-1,2

xi= W (cot 891 + cot 892)= horizontal distance traveledper bounce

whereVpi = phase velocityVgi = group velocityW = thickness of crystal

8pi = angle of Vpi vector8gi = angle of Vgi vector

CHi 521 log MA~

Gat

10 dBI REF 000 dB

~

\ ,"""! I j IlYV~,\y,

~WI Ij~, ~'!I

r~~""'.I' ~"'I I' ,F'I III

This plot shows a wide sweep of an in-lineBrickwall filter. This device has a centerfrequency of about 300 MHz and a time delay ofabout 320 nS. The bulk mode spurious response canbe seen in the region around 600 MHz. Most of thebulk modes are launched at frequencies higher thanthe SAW frequency. In fact, most of the energy islaunched at about twice the frequency of the SAW.

START 200. 000 000 MHz STOP 1 000. 000 000 MHz

10

CH1 S21

Gat

log MAG 10 dB/ REF . 000 dB

To measure these bulk modes, it is useful to coatthe device surface with an acoustic absorbingmaterial such as black wax. This suppresses the SAWbut has a negligible effect on the bulk waves whichpropagate within the crystal. This figure shows theeffect of using black wax on the surface of thedevice. The SAW response is completely suppressedand the bulk mode response is clearly visible. Nowthe user can transform to the time domain and seethe time responses caused by the various bulkmodes.

START 200.000 000 MHz STOP 1 000. 000 -000 MHz

30 .5 ns

N ul

r f'l ~~ I A ~

\~II ~~L'I \ I~~ !IAII I

r~,1"1~ ~ ~ l Ul~~ ~M 1Iit,I I r '''l

CH1 S21 log MAG 10 dS/ REF -30 dB .1.; -75 699 dB

Each response corresponds to a different bulk waveor a multiple bounce of one of the waves. Note thatone of the stronger responses has a time delaywhich is close to the SAW delay of 320 nS. Thiscould be a potential source of frequency responsedegradation, especially if this energy is launchedclose to the SAW frequency. To see the frequencycontent, the user can place the gate around thetime response of interest and transform back to thefrequency domain.

CHi START-20 ns STOP 780 ns

hI'h.~f \ r'1 IA

AI 11 V '\ I,\, ~N

'W~~~~ nn~W rv~WI ~\.~I

CH1 521

Gat

log MAG 10 dB/ REF a dB

The frequency response of the two most dominanttime responses is shown here. These responsesaccount for most of the bulk mode energy that islaunched in this device. Fortunately, these wavesdo not contain appreciable energy at the SAWfrequency, but they do affect the far out stop bandrejection. without using the black wax absorber,these waves could not be observed in the presenceof the stronger SAW response.

START 200. 000 000 MHz STOP 1 000. 000 000 MHz

11

LOW LOSS GUDT(Group-type Unidirectional Transducer)

SAW FILTERS

RF o--l=::;=====:;------;::====;\'Input

'--------"---+++-0 RFOutput

PiezoelectricSubstrate

Bidirectional transducers offer precise control offilter shapes, but not without drawbacks. There isa minimum 6 dB loss due to the fact that half theenergy launched is away from the output. Practicalinsertion losses are often 15-35 dB in order tomeet demanding passband ripple specifications,since the only way to reduce triple transit rippleis to mismatch the electrical ports 6f thetransducers. A low-loss approach to SAW filtersuses unidirectional transducers. One configuration,employing group-type unidirectional transducers, isshown here. Each transducer in this case is reallytwo phases offset on the substrate from each otherby lambdaj4 or 90 degrees. The two phases share thesame ground electrode. This ground follows ameander pattern through the device and isalternately lambda/2 and lambda wide. This tricksupplies the necessary lambda/4 offset betweenphases. An external electrical phase shift of 90degrees is necessary between the two phases of eachtransducer, supplied by two inductors pertransducer. The filter using group typetransducers can be fabricated in a single,self-aligned layer with no crossovers. [8J

Shown here is the response of a group-type SAWfilter with passband at 400 MHz. The passbandminimum insertion loss is seen to be about 7.6 dB.Feedthrough is clearly visible at the 40 - 60 dBlevel below the input. Much of this feedthrough istest-fixture related, and will be eliminated in aproperly shielded circuit board application.

,--- --t----

j ,- ~ _.._~.--+--~--I

j

__+- __ L __+ : _i ' .

- -----j" !----i--! I

. -1 ··-~'-----+---'\>---+I--I i I

CH 1 I'"S2;;.1~~1.;;.o",9_M"l'A_G l_0--rd_8_/_R7E_F_O_dr-8'---.,.-::-:7"''-'~-:::7:-.-::5-;,5:::33""7:7d':"18':I 401. 00 oqo MHz\

.• \

~,,)_:1 ;:I~", +" ,_, __ I

Cor

START 250.000000 MHz STOP 550. 000 000 MHz

An 80l-point transform to the impUlse response inthe time domain is shown here. The main filterresponse indicated by marker 1 has delay 660 nsec.The triple transit signal at marker 2 is 50 dBdown, demonstrating that this type of filter designcombines low loss with good triple transitsuppression. At t=O the electrical feedthroughsignal is a pronounced spike. The group offeatures visible from -1100-1250 nsec in theimpUlse reSponse correspond to an acoustic mUltiplereflection within metal shield bars located betweenthe input and output transducers, which could beeliminated at the expense of new masks. ThisconClusion was reached only after examination ofthe present time domain response.

STOP 2.4 us

MAG

CHI

10 dB/ REF -10 dB L -26. 658 dB

'I 659.:38

I---i---f--+--+-I--~.'--i--- ..~--~

Cor I il,: I

.. I 1---+---+-_~__.L...1 I 1 I; i

I--+---+---+H---+--L---·'----r---+--t----t---++-+--f-~-:--L-J---l- _

II I..L.--++--1-+----+-r-.1..--+----'r-

I i·-i---~--_·_-t-

12

Here we use the time domain gating features of the8753A to remove the effects of feedthrough tosimulate performance in a properly shielded circuitboard application.660

l~ -26. 6~3 dBREF G d810 dB.'log MAG

,-----r------ --r- -----~----------

i--t----

Gat

Cor

CHI 521

300. DO 0 o MHz/

1\\ •,""

.- 1.1l!' 1I111l'\'1'\ ,,~-

.It'{ \IUIII' '1"- WI PI 'Ii '1WI '11

300. 00 0 a MHz

/ \

Iv,1 .r"f ¥\. ) i.J."l1\

!W .N.)I, Ii IJI\!, _JI"l/ \I.... 111 .~ -"'\lII '....In I

1

CH1 521 log MAG

COr

Gat

CH2 521 log MAG

10 dB/ REF 0 dB

10 dB/ REF a dB

.1.; -55 918 dB

1. -51 484 dB

Shown here in dual channel display are the gated,retransformed frequency domain data from theprevious plot in Channell. This simulates thecircuit performance. For comparison we display inchannel 2 the measured data. A criticalspecification for this filter is out-of-bandrejection at 300 MHz. Marker 1 in both plots is setto this point. The presence of feedthrough in theresponse on channel 2 actually may make the SAWdevice's rejection at 300 MHz look better or worse.This is due to interference between the electricaland acoustic signals. The rejection due to theacoustic signal alone, which simulates circuitperformance, is seen at marker 1 on Channell.This rejection is 55.9 dB, whereas the measuredrejection is degraded to 51.5 dB by feedthrough inChannel 2.

START 250. 000 000 MHz STOP 550.000 000 MHZ

DELAY LINESLow propagation loss for SAWs in crystals makespossible compact, convenient delay lines of 10 usecor more delay. These are highly useful inoscillator control, in FM discriminators, and otherapplications. The HP 8753 network analyser can beused to measure important characteristics of SAWdelay lines in both the frequency domain and thetime domain.

• OSCILLATORS

• FM DISCRIMINATORS

• 10 psee DELAY IN 5 em

13

310. 50 0 o MHz

MAR ER 1 1/ '\310. 5 M-Iz I \

f\J \r~ tv V\ /'"

N \/ 1/ \ tr..

II W J

CHI 521

Cor

109 MAG 10 dB/ REF 000 dB l..J -8 9264 dB

We show here the frequency response of a devicewith transducers similar to those employed in thelow loss filter previously shown. The centerfrequency of this device is 310 MHz. The overallsubstrate length between input and output is about10 times as great, however, giving a delay of 6.062usec. The fixture and cable loss in themeasurement are calibrated out by first doing aresponse calibration with a through line. Due tothe long acoustic delay present in the device it isnecessary to increase the resolution bandwidth andsweep time of the analyser. We have found that abandwidth of 3 kHz and a sweep time of 2 secondsare sufficient for the 6 us device delay.Increasing the resolution bandwidth and sweep timeare required to prevent anomalously high insertionloss measurements.

START 280.000000 MHz STOP 340. 000 000 MHz

Here we use 1601 points to transform the frequencydomain data into a time plot. By setting thevelocity factor in the calibrate menu to .0000132times the velocity of light (for surface waves onlithium niobate), positions of the marked timespurs are calibrated in distance on the device.The start time can be set to a small negative valueto move any feedthrough signal off of the t=o axis.If the stop time is set to some large value it isautomatically set to the maximum value allowed toprevent aliasing. The figure shows no feedthroughin the long device, although both single and tripletransit signals are clearly evident. The tripletransit signal can be reduced by selecting optimumvalues for the tuning inductors of the group typeunidirectional transducers. The use of 1601 pointsis required to prevent aliasing of the tripletransit signal in the transform. The user shouldnote that to use 1601 points, Channel 1 and Channel2 must be decoupled and the number of points in theunused channel reduced to prevent memory overflow.

2J -56 703 dB10 dB/ REF -10 dBlog MAG

18. 19 us

MAR ER1 B. 1 f us

71. 9 3 mm

2

,~ I, ,IIV

,A ••. ~~~. ~

WJ~~IIV, l'I.lAJM!.

CHI 521

CHI START-20 ne STOP 28.683 UO!;l

4.86 us

MAR ER 24 86 us

19.2 2 mm

2

In. A ,111

~I~lW

'"~~~

~I'I~ NI~ '~ '~I

CHI 521 log MAG 10 dB/ REF -10 dB 2J -56 123 dB

Here we show the transform using 801 points. Thetriple transit signal occurs, incorrectly, beforethe single transit signal due to aliasing. Insteadof employing more points, aliasing can beeliminated by reducing the frequency span of theoriginal measurement.

CHI START-20 ns STOP 13. a5 U&

14

CHI S22 109 MAG

Co~

5 dB/ REF 000 dB

11'\

Reflected signals returning to the sendingtransducer are also easily measured on the HP8753A. This figure shows 822 for a signal injectedat the apodized transducer and reflecting from theunapodized transducer. The ripples indicateinterference between an electrically reflectedsignal and an acoustically delayed signal.

CH 1 START 280. 000 000 MHz STOP 340. 000 000 MHz

CHI S22 log MAG 10 d8/ REF -10 d8 1.J-3413 dB

12. :3:3 us

The transform of the 822 data reveals both a doubleand quadruple transit signal. Due to the long timespan, 1601 points are required for this transform.

MAR ER 1112. 1 3 U"

48.012 mm

r '"\~ .1

~n 1\ N''~ I'J \ I/V~ (\ r IrJ

V \h {II I~ NlJ,~ V II

CH1 522

Gat

log MAG 10 dB/ REF 0 d8

Transforming the double transit signal gives afrequency response response dominated by the filterapodization.

CHI START 280.000 000 MHz STOP :340. 000 000 MHz

15

I--~ ~ / "" ~/" .--

~"-...../

CH1 S22 109 MAG

Gat

10 dB/ RC;F 0 dB

The frequency response of the signal reflected attime t=o is more independent of frequency but showsvariation due to the tuning inductors and theacoustic impedance of the transducer.

SURFACE ACOUSTIC WAVERESONATOR (2-PORT)

CH1 START 2BO.000 000 MHz

auartzSUbstrate~

InputlOT

STOP 840. 000 000 MHz

Grooved ArrayReflector

Another type of SAW device is the high-Q resonator[4,5,10]. The principal application is as theoscillator frequency control element at fundamentalfrequencies from -100 MHz to -2 GHz. Unloaded Qfactors vary from 50,000 at the low end to 3,000 atthe higher frequencies. These devices are usuallyconstructed on quartz rather than on lithiumniobate because of the extremely low temperaturesensitivity of the former. The processing andpackaging for resonators is much more difficultthan for filters because of the sensitivity of thecenter frequency to surface contamination andprocess variations. The test system requirementsare also different. High-Q resonator measurementsdemand much higher frequency accuracy, but do notneed the dynamic range required in filtermeasurements. The HP 8753A network analyzer isagain well suited for manual or automatic SAWresonator measurement.

16

The layout of a typical SAW resonator is depictedhere. The device is composed of 2 symmetricinterdigital transducers (IDTs) surrounded byreflective arrays of synchronously-placed shallowgrooves. The IDTs, connected to the outside worldvia bonding pads, serve two purposes. They act toconvert electrical energy into acoustic energy andback again, and they form part of the reflectivearrays. Each of the grooves (or IDT fingers)reflect a small portion of an incident acousticwave back toward the center of the cavity. At afrequency such that the spacing between grooves (orfingers) is 1/2 wavelength, the total reflection iscoherent and virtually 100% from each array. Atother frequencies the combined reflectioncoefficient is very small.

~~C~ JL R1 C~ L 1 JLc:r -co

within a few hundred kilohertz of the centerfrequency, this device can be modeled as a seriesresonant circuit. Note that both I-port and 2-portresonator configurations are possible. The 2-portconfiguration is usually preferable because of thehigh isolation out-of-band between input and output·ports. The I-port model is the same as that for abulk crystal resonator.

r1

CIlII

w/-,

l,j

l,j

0-.J

o

-10

-20

-30

-40

~

J\f\ !\ Nr ,\

j j \ Ai \f\rv1\

Although the resonant frequency is roughlydetermined by the grating and lDT periodicity, manydesign and processing factors can affect the exactcenter frequency. optimization is required toachieve the highest Q and lowest resonantresistance at a desired frequency. Accuratemathematical models have been developed to predictthe response of a SAW resonator [10]. Theshoulders and ri~ples around this responserepresent the si&elobes of the reflective gratingand lDT response; These ripples are important tomeasure and model accurately, because they giveimportant information about the quality of thesurface, the metal fingers and the grooves.

739.000 743.000Frequency [ MHz

748.000

I--r!

CHI 521 log MAG

Cor I.- t-

III-ri

000

5 dB/ REF 0 dB L -40. 839 dB

i 747. 00 0 0 MHz

17

A simple, coaxial two-port fixture is used here tomeasure the transmission response of a SAWresonator centered at 742.5 MHz. A singlethru-line calibration was done to remove the lossesof the fixture and cables using the "RESPONSE"option in the calibration menu. Measuring the logof the magnitude of S21 shows good agreement withthe modeled response above.

Sma

For an ideal, single-pole resonator, the loss andunloaded Q are sufficient data for an oscillatordesigner to predict his noise characteristics. Fora real resonator, however, the resonance may benonideal, causing unwanted noise and pUllingproblems. The group delay is a sensitive measureof the resonant characteristics. In this Figure,we use the GROUP DELAY display option on Channel 2and the DUAL DISPLAY option to view the group delayresponse concurrently with the transmissionresponse over a 300 KHz span. Note that the groupdelay response is smoothed and averaged with afactor of 100, requiring 20 seconds or so ofmeasurement time. The response reveals a slightdip in the group delay of about 7% of the totaldelay, which can be read as 1.57 microseconds. Theripple in the group delay was measured by settingMarker #1 at the maximum (using the SEARCH MAXfunction), pressing the :MARKER ZERO soft key, thenmoving the marker to the local minima of the dip toread the relative difference of 115.7 nSeconds.Note that the markers are independent on the twochannels. This group delay dip can give rise to alower than expected Q value on resonance, and isvery difficult to detect from the amplitude dataabove.

LOADED Q:

QL =fol1f

UNLOADED Q:1 1-- =--

Qu QL

1

Qextelec

1

Qextair

The loss and loaded Q allow us to derive theunloaded Q of the SAW resonator. Electricalloading is due to the 50 ohm measurement system andacoustical loading may be present if the device isoperating at atmospheric pressure. The devicemeasured above is packaged in a vacuum, and thusthe unloaded Q is 7140.

Qextelec

= , = TRANSMISSIONCOEFFICIENT

Qextair = 66,700 FOR QUARTZ

- - f----- ---t-'11t\f-f---+-It---1f'fl--I'-t---+---f-----j

I--...L-+---,----+----+--+---+--+---+---1-+ ----1----,---- --II--JI!--f-e-~I---+--+----I~------+---+---,l-i!r--t--+---+--+--l

____ L +__+-1 .rfi --- ----+~-----+--+--j

---J---r--#' \INV\. rJ

CHI CENTER 742.471 500 MHz SPANCH2 521 109 MAG 3 dB/ REF -50 dB

The time domain analysis capability of the HP8753Acan also be very useful for analyzing SAWresonators. On the upper trace of this figure, weare viewing the amplitude response of the resonatorover a 50 MHz span-using 801 measurement points.On Channel 2, we have initiated the time domaincapabilities in BANDPASS stimulus mode to view theFourier transform of these data, which correspondsto the resonator impulse response. The start andstop times and scale maybe set independently ofthe top channel. Note the sharp pulse at timezero, corresponding to direct RF feedthrough, andthe basic exponential ring-down (linear on this logscale) of this impulse response. The peaks andvalleys appearing in the impulse response representacoustic reflections off of the IDTs, thereflective gratings, the ends of the quartz chip,and other physical elements of the resonator.

50.000000 MHz

5 dB/ REF -5 dBlog MAGCH1 521

CH2

18

CHi CENTER 742.500 000 MHz SPANCH2 521 I 09 ~lf\G 3 dB! RE~ -50 dB

I !I +-I , I

i +I c

, .AN -t---r---IV' 1\j' ,IV 'l\- i

; )'1 ! UI' I I"- I ~

CHI 521

Gat

log MAG 5 dB/ REF a dB

50. 000 000 MHz

The gating feature may also be used to powerfuladvantage. In this figure, gates were set toremove the zero-time impulse in the time domain dueto RF feedthrough. The GATE function was alsoturned-on in the upper, frequency domain trace toview the result. Note from comparison with theprevious figure that several apparentantiresonances along the .shoulders of the responsewere really due to destructive interference betweenthe acoustical response and the RF feedthrough.The rises of the frequency domain response at thetwo ends of the span range are artifacts of thegating window and sample period.

T

CHZ CENTER SPAN 4.02 us

19

The HP8753 is ideally suited to make the necessarymeasurements on SAW devices. The excellent dynamicrange allows the user to measure minute detail inthe response of the device. This is especiallyimportant for high insertion loss filters as istypical of brickwall filters. The internalsynthesized source permits extremely accuratefrequency measurements of SAW resonators. Thecalibration capabilities allow accurate measurementof insertion loss and impedance. The markerfeatures allow quick access to bandwidths,frequencies, time and loss. The time domain featureis indispensable in the measurement of thesedevices and the 1601 point transform allows highresolution time domain information to be computedrapidly. Time domain gating and retransformationto the frequency domain enables simulation ofcircuit performance of devices tested in testfixtures. Automated features ease large volumetesting applications.

REFERENCES

1. A. H. Cook, PHYSICS OF THE EARTH AND PLANETS. N.Y.:Wiley, 1973.

2. R.M. White and F.W. Voltmer, "Direct PiezoelectricCoupling to Surface Elastic Waves", APL 7 (Dec. 1965)

3. S. Elliott and R. Bray, "Surface Acoustic Wave Devices:Design and Measurement", Hewlett-Packard RF & MicrowaveSymposium, 1982-1984.

4. A.A. Oliner, ed.,Springer-Verlag, 1978.

ACOUSTIC SURFACE WAVES. N. Y. :

5. H. Matthews, ed., SURFACE WAVE FILTERS. N.Y.: Wiley, 1977.

6. L.F. Kinsler2nd . ed . N. Y. :

and A.R. Frey, FUNDAMENTALS OF ACOUSTICS,Wiley, 1962.

7. C.S. Hartmann, D.T. Bell, Jr., and R.C."Impulse Model Design of Acoustic Surface WaveIEEE Trans MTT-2l, 162 (1973)

Rosenfeld,Filters",

Elliott, and R.C. Bray,Using Parallel Coupled

Symp. Proc., IEEE Cat

8. K.Yamanouchi, F.M. Nyffeler, and K.Shibayama, "LowInsertion Loss Acoustic Surface Wave Filter UsingGroup-Type Unidirectional Interdigital Transducer", 1975IEEE Ultrasonics Symposium Proceedings, IEEE Cat. #75 CHO994-45U (1975).

9. R.S. Wagers and R.W. Cohn, "Residual Bulk Mode Levels in(YXl) 128 degree LiNb03", IEEE Trans. on Sonics andUltrasonics, SU-3l, 168-175 (May 1984).

10. P.S. Cross, W.R. Shreve, S.S."Very Low Loss SAW ResonatorsCavities", 1982 Ultrasonics#82CH1823-4, pp. 284-289.

F/iOW HEWLETT~~ PACKARD

March 1986

Printed in U.S.A.5954-1569