Port 1, Port 2, or both VNA ports as well. Ca-pable of very precise and accurate results, aVNA’s performance is predicated upon the sta-bility and repeatability of the measurementsystem, that is, the DUT-to-VNA interface,which includes the test port extension, if used.

A test configuration using a flexible testport extension on Port 2 is illustrated in Fig-ure 1. The resulting calibrated measurementsof a given DUT will contain notable inconsis-tencies. When observing the analyzer display,S22 return loss will differ from S11 return loss,appearing somewhat “noisy” in nature. If thetest port extension is flexed, changes in bothS22 and S11 return loss data will be observed.If measuring a low loss DUT, the aforemen-tioned behavior is exaggerated. The intent ofthis article is to offer a qualitative explanationfor this behavior and its underlying mecha-nisms.

UNDERSTANDING HOW A VNA WORKS:VNA MEASUREMENT ERROR TYPESDirectivity Error

A vector network analyzer utilizes direc-tional couplers for measuring forward and re-verse propagating signals. Directivity is a mea-

PAUL PINOW.L. Gore & Associates Inc.Elkton, MD

Today’s modern vector network analyz-ers (VNA) are the product of evolution-ary advances in technology that literally

span decades — a marrying of the latest de-velopments in microwave, integrated circuitand software technologies. To a great extent,the precise nature of vector network analysisresults from its highly developed calibrationalgorithms and methodologies. In many VNAapplications, a flexible test port cable assemblyor “test port extension” may be required. Thetest port extension is used to establish athrough connection for calibration purposesand for making measurements of the deviceunder test (DUT). Generally, a single port ex-tension is used, but this is by no means therule. Port extensions may be employed on

DISTORTION INHERENTTO VNA TEST PORTCABLE ASSEMBLIES

Port 2

Test Port 2extended to

here

“Test Port extension”

Port 1

Fig. 1 VNA shown with aflexible test port cableassembly on Port 2. ▼

Reprinted with permission of MICROWAVE JOURNAL® from the September 2006 issue.©2006 Horizon House Publications, Inc.

sure of a coupler’s ability to discrimi-nate against a signal that is propagat-ing against the “coupled direction.”

Reflection Response/TransmissionResponse Tracking Error

These errors represent the differ-ences between reference signal pathsand signal measurement paths.

Source Match ErrorError generated as a result of the

mismatch between the source testport and the measurement system(this includes any device connectedto the source test port).

Isolation ErrorError generated as a result of sig-

nal leakage between reference andmeasurement channels within thevector network analyzer.

Load Match ErrorError generated as a result of the

mismatch between the load test portand the measurement system (this in-cludes any device connected to theload test port).

Port 2 Match ErrorError generated as a result of the

mismatch between the VNA test portand the test port extension.

The flow chart shown in Figure 2is a popular format for describingboth the desired signal flow and errorsignal flow associated with VNA mea-surements. Specifically, the diagramdescribes a “forward” S21 transmis-

sion measurement where the intend-ed signal flow is left to right, fromPort 1, through the DUT, and intoPort 2. A representation of the flexi-ble test port extension used in thisstudy is highlighted in red. This errormodel represents the test configura-tion depicted in Figure 1.

The diagram can be applied to S11return loss measurements as well.The various errors listed previouslyare corrected for during the calibra-tion process. A mirror image of thediagram would apply to “reverse” S12transmission data and S22 return lossdata, port reference planes are high-lighted in blue.

SUMMARY OF FINDINGS Through the calibration process,

the VNA corrects for the phase andamplitude characteristics of the mea-surement system using a processknown as “error correction.” A testport cable extension, if used, is consid-ered part of the VNA measurementsystem. The error correction operationis static in nature. The correction algo-rithm “assumes” the test port exten-sion’s phase and amplitude responseremain constant. In short, the errorcorrection algorithm is unable to com-pensate for future system changes.

The “noisy” quality noted primarilyin, but not limited to, the DUT’s S22 re-turn loss data stems from minute physi-cal changes within the test port cableassembly — physical changes that oc-cur during flexure. As a result, the mea-surement system has been altered fromits original calibration state; the VNA’serror correction algorithms cannotcompensate for this change.

These now-uncorrected changesare manifested as properties of theDUT, and are visible within the asso-ciated S-parameter data — most no-ticeably, in the return loss/VSWRdata of the port equipped with the

test port cable assembly. As the fre-quency increases, adverse effects re-sulting from changes in phase andamplitude response become morepronounced. This is especially truewith phase, following the relationship

degrees = t (ns) • 360 • ƒ (GHz)

where a time period, t is expressed indegrees at a frequency, ƒ. This for-mula can be used to describe a par-ticular cable assembly feature, withthe feature in question having a timedelay measured in nanoseconds,whose electrical length in degrees in-creases with respect to wavelength.As the frequency increases, the wave-length decreases. Although constantin its time delay, the feature size cancover multiple wavelengths or a frac-tion of a wavelength, depending uponthe frequency of interest.

An unstable, flexible test port cableassembly will have a profoundly nega-tive impact on a VNA’s measurementaccuracy. By selecting a high qualitycable assembly having long-term sta-bility, coupled with proper measure-ment practices, one can significantlyaugment the capabilities of any VNA.

DESCRIPTION OF TESTPROCEDURE AND EQUIPMENT

The following equipment and pro-cedures were used in this study. Datawas captured using GORE propri-etary software.• Agilent Technologies PNA, modelE8364B, swept frequency range:0.010 to 67 GHz• Flexible test port cable assembly,2.4 mm socket to 2.4 mm ruggedized,NMD-style socket, GORE FE0BN0-BL038.0, length: 38 inches• Calibration standard: AgilentTechnologies E-CAL unit, 0.010 to67 GHz, equipped with 1.85 mm in-terfaces

The ambient test conditions were23° with relative humidity controlledat 20 percent. A full two-port calibra-tion, omitting isolation, was appliedover a frequency range of 0.0625 to50 GHz, 800 points with an IF band-width of 100 Hz. No smoothing or av-eraging was used.

The low loss “insertable” DUTconsists of an Anritsu 1.85 mm pin to1.85 mm socket adapter, commonlyknown as a “connector-saver” (seeFigure 3). This adapter is consideredto be a stable and sufficiently precise

TECHNICAL FEATURE

Incident Thru

DUT

D1

I

ReflectedVNA Port 1ref. plane

D = Directivity Tr = Reflection Frequency Response TrackingMs = Source Match I = IsolationMI = Load Match Tt = Transmission Frequency Response TrackingM2 = Match at Port 2

Flexible TestPort Extension

Port 2ref. plane(extended)

Port 2connectionat VNA

Tr Ms M2MI

Tt

S11 S22

S21

S12

S11 S22

S21

S12

▲ Fig. 2 Forward (S21) two-port error model showing port reference planes (blue) and flexibletest port extension model (red).

▲ Fig. 3 The 1.85 mm pin to 1.85 mmsocket adapter used as a low loss DUT.

device for the purposes of this study.After inserting the DUT and carefullytightening the interfaces to the prop-er torque specification, the test portcable assembly was examined to en-sure it was routed in a wide arc withno undue stress placed upon it. Sup-port fixtures were used to hold theDUT side of the test port cable as-sembly (see Figure 1). At this point, abase-line data set comprising all fourS-parameters was recorded.

After recording base-line data, thetest port cable assembly was bentsharply at a point just after the Port 2connection. A data set consisting ofall four S-parameters was collectedwith the cable assembly held in thisbent or flexed position. Phase andamplitude characteristics of the DUTin the “flexed” and “not flexed” stateswere compared.

S-PARAMETER ANALYSISAlthough very slight, Figure 4

demonstrates the differences be-tween S11 and S22 data when a testport assembly is used. In Figure 5,the test port assembly is flexed, as de-scribed in the test procedure. In thiscase, the differences are more promi-nent — S22 data differs noticeablyfrom S11, especially in terms of tracesmoothness. Observe that the “notflexed” S11 data differs from “flexed”S11 even though the distortion-induc-ing flexure occurred on the Port 2side of the measurement system.

Figure 6 illustrates S22 return lossdata for the low loss DUT using a testport extension in both the “flexed”and “not flexed” positions — note thebehavior at and around nulls. Figure7 is a detailed view concentrating onperformance between 30 and 40GHz. Differences are due to phase/amplitude changes in test port assem-bly affecting error correction.

A more revealing representationfor the change in return loss involvesapplying a subtraction operation tothe “flexed” and “not-flexed” data

sets, as seen in Figures 8 and 9. Inthis case, the change in S11, that is, aport not fitted with a port extension,is compared to the change in S22, aport fitted with a port extension. Theassumption being that S11 will changelittle, if not at all during flexure.

Figures 8 and 9 are comparisonsbetween “flexed” and “not flexed”states on Ports 1 and 2. To gain a bet-ter understanding of the magnitudeof change on Port 1 compared to Port2, refer to Figure 8. For an idea as tohow changes in magnitude compareto key VNA calibration error criteriasuch as source match, load match, re-flection tracking and directivity (allexpressed in dB), refer to Figure 9.

As the change between the “flexed”vs. “not flexed” condition approaches–45 dB, anomalies appear in the S22 re-turn loss data. This behavior can beverified by examining Figures 4 and 5.In general, desirable corrected errorlevels for source match, load match, re-flection tracking and directivity are inthe realm of –45 dB. Measurement sys-tem distortion approaching this level

TECHNICAL FEATURES 1

1, S

22 (

dB)

FREQUENCY (GHz)

−15

−20

−25

−30

−35

−40

−4530 34 38 42 46 50

VNA

Port 2

Port 1

DUT

▲ Fig. 4 S11 and S22 data for the low lossDUT with flexible test port extension in “notflexed” position.

S 11,

S22

(dB

)

FREQUENCY (GHz)

−15

−20

−25

−30

−35

−40

−4530 34 38 42 46 50

VNA

Port 2

Port 1

DUT

▲ Fig. 5 S11 and S22 data for the low lossDUT with flexible test port extension in“flexed” position.

RET

UR

N L

OSS

S22

(dB

)FREQUENCY (GHz)

0

−10

−20

−30

−40

−50

−60

−700 10 20 30 40 50

“not flexed”“flexed”

▲ Fig. 6 S22 return loss of the low loss DUTusing the flexible test port extension; S22“flexed” and “not flexed” data are overlaid.

RET

UR

N L

OSS

S22

(dB

)

FREQUENCY (GHz)

−15

−20

−25

−30

−35

−40

−45

−5030 32 34 36 38 40

“not flexed”“flexed”

▲ Fig. 7 Enlarged view showing S22 returnloss of the “flexed” vs. “not flexed” states.

DEL

TA M

AG

NIT

UD

E

FREQUENCY (GHz)

0.012

0.010

0.008

0.006

0.004

0.002

030 34 38 42 46 50

S11 abs (mag A - mag B)S22 abs (mag A - mag B)

▲ Fig. 8 S11 and S22 difference magnitudeof low loss DUT with the flexible portextension, as measured before “B” and after“A” the test port cable assembly flexure.

DEL

TA M

AG

NIT

UD

E (d

B)

FREQUENCY (GHz)

−30

−40

−50

−60

−70

−80

−90

−10030 34 38 42 46 50

S11 20 log (mag A - mag B)S22 20 log (mag a - mag b)

▲ Fig. 9 S11 and S22 difference magnitudein dB, as measured before “B” and after “A”the test port cable assembly flexure.

will have a detectable influence uponthe S-parameter data of the DUT. Dis-tortion levels less than –50 dB are gen-erally beyond the system’s ability to dis-tinguish such detail.

In Figures 6 and 7, a change in bothS11 and S22 return loss is recorded.During calibration, the test port exten-sion connected to Port 2 acts as theload match measurement path for Port1. When the cable assembly is in the“flexed” state, any consequent distor-tion will affect the load match correc-tion for Port 1. This explains the smallchange in S11 return loss even thoughthe test port cable assembly was flexedvery close to Port 2. If the DUT hadpossessed a significant amount of inser-tion loss, the change in S11 return lossbetween “flexed” vs. “not flexed” stateswould have been less pronounced. Ahigh insertion loss DUT would isolatethe two ports from one another, effec-tively reducing the influence of Port 2-related distortion on Port 1. AppendixA provides detailed explanations andexamples of the basics of vector errorcorrection.

INSERTION LOSS AND PHASEANALYSIS OF LOW LOSS DUT

Figures 10 and 11 display the re-sulting apparent changes in insertionloss and insertion phase of the lowloss DUT used in this experiment.The data clearly indicates a differ-ence in phase and amplitude perfor-mance between the “not flexed” vs.“flexed” states of the test port cableassembly. To further emphasize animportant point, these changes in

phase and amplitude should not beconfused with DUT performance;they are solely the result of changeswithin the measurement system thatcannot be accounted for by the VNAerror correction algorithm. As statedearlier, changes in the measurementsystem can easily be misinterpretedas changes in DUT performance.

CONCLUSIONThe intent of this article was to ex-

plain the basic principles of VNA er-ror correction and how VNA mea-surement system effects can be con-fused with DUT performance. Thistopic was deliberately approached ina qualitative manner so as not to bur-den the reader with the complexity oferror correction algorithms. In clos-ing, the following practices can beused to minimize the changes inphase/amplitude response associatedwith test port cable assemblies:• For measurements requiring ahigh degree of precision and accura-cy, limit test port assembly movementduring the calibration and measure-ment processes• Before using any cable assembly asa test port extension, measure itsphase/amplitude response with flex-ure in two orthogonal planes of mo-tion, noting which plane of motionyields the smallest change inphase/amplitude response. When at-taching the assembly to a VNA, orientthe cable so that its movement is inthe plane of motion yielding the leastamount of phase/amplitude distortion

• To track performance degradation,periodically check the phase/ampli-tude stability of the flexible test portcable assembly by using the manufac-turer’s advertised test method• Ensure that all VNA and DUTconnections are clean and the speci-fied connector tightening torque isobserved. Inspect all DUT and testsystem connectors for damage; in-spect all DUT and test system con-nectors for proper pin/socket heightspecifications before connecting tomating connectors• Perform VNA measurements in atemperature-controlled environment,ensuring instrumentation is posi-tioned well away for heating/coolingducts. Avoid placing instrumentationnear exit/entrance ways, as these ar-eas can experience intermittent fluc-tuations in temperature

When considering the purchase ofa test port cable assembly, carefullyreview and compare manufacturers’claims of stability and repeatabilitywith flexure. To realize optimal VNAperformance, a high quality cable as-sembly with consistent long-termperformance is required. ■

ACKNOWLEDGMENTThe author would like to thank

Harmon Banning — mentor, teacherand friend — for his guidance in thepreparation of this paper.

Paul Pino received hisBS degree in electricalengineering from theUniversity of Delawarein 2000 after leaving along career in theautomotive industry.He joined W.L. Gore &Associates Inc. in 1999and has worked withvarious groups,including Gore’s SignalIntegrity Lab, the

Planar Cable Team and the Fiber OpticTransceiver Team. For the past four years hehas worked within the ATE Microwave Group.

TECHNICAL FEATURED

ELTA

IN

SER

TIO

N L

OSS

S21

(dB

)

FREQUENCY (GHz)

0.010

0.006

0.002

−0.002

−0.006

−0.0100 10 20 30 40 50

▲ Fig. 10 Apparent change in S21 insertionloss of the low loss DUT with the flexible testport extension in “flexed” vs. “not flexed” states.

DEL

TA P

HA

SE S

21 (

°)

FREQUENCY (GHz)

0

−0.5

−1.0

−1.5

−2.0

−2.5

−3.00 10 20 30 40 50

▲ Fig. 11 Apparent change in S21 phase ofthe low loss DUT with the flexible test portextension in “flexed” vs. “not flexed” states.

TECHNICAL FEATURE

The S22m vector represents a measuredor uncorrected S22 data point. The S22a vec-tor is an actual or corrected data point. Whenmaking a calibrated measurement, correcteddata is displayed on the VNA screen. The er-rtotal vector is the sum of all error correctionassociated with a particular calibrated VNAmeasurement at a specific data point.

In reality, the errtotal vector is comprisedof several error correction vectors, includingcorrection for the error associated with thecalibration standards, as well as error associ-ated with Port 2 of the VNA — hereafter re-ferred to as errgeneral. In this example, a testport cable assembly connected to Port 2 isused as part of the calibrated measurementsystem. A portion of the error associated withPort 2 of the VNA is related to the test portcable assembly.

S22a

S22m

errgeneral

errtest port cable

S22a

S22m

errtotal

Once the test port cable assembly isflexed (after a calibration is completed), thefollowing occurs:

Error correction vectors are fixed (inmagnitude and phase) for a given calibration,therefore all measurements subsequent tocalibration have correction applied that as-sumes the measurement system is static.

With the test port cable assembly flexed,phase/amplitude distortion occurs. This dis-tortion is manifested as a property of theDUT, not the test port cable assembly. Theeffect is S22m-new.

The result: S22m has changed in magni-tude and phase, thus S22a must change sinceit is calculated from S22m. The outcome —S22a-new — results from a calibration correc-tion based upon having the test port cable inits original “calibration position” (shown ingrey).

S22a-new

errgeneral

S22m-new

errtest port cable

APPENDIX A

SIMPLIFIED EXPLANATION FOR THE ILLUSTRATED BEHAVIOR IN FIGURES 4 THROUGH 7

Condition is for stable calibration – idealS22m = S22a + errtotal

To arrive at corrected reading:S22a = S22m – errtotal

For this example, total error is:errtotal = errgeneral + errtest port cable

Grey vectors represent initial S22a and S22mresults under ideal conditions

Colored vectors represent S22a and S22mresults with test port cable flexed