Improving Network AnalyzerMeasurements of Frequency-translating DevicesApplication Note 1287-7

LO-RFRF

LO

LO+RF

IFIF

RFLO

IF

LO-RFRF

LO

LO+RF

IFIF

RFLO

IF

LO-RFRF

LO

LO+RF

IFIF

RFLO

IF

LO-RFRF

LO

LO+RF

IFIF

RFLO

IF

2

Table of ContentsPage

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

Network Analyzer Mixer Measurement Configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3Scalar Network Analyzer Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4Vector Network Analyzer in Frequency Offset Mode Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5The Upconversion/Downconversion Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

Conversion Loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6Definition and Importance of Conversion Loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6Measurement Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

Mismatch Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7Considerations Unique to the Scalar Network Analyzer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

Importance of Proper Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9Frequency Response Error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

Considerations Unique to the Vector Network Analyzer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10Importance of Proper Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10Sampling Architecture and Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10R-Channel Phase-Locking Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12LO Accuracy and Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12Power-Meter Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

Accuracy Comparison of the 8757D and a Vector Network Analyzer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16Fixed IF Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

Relative Phase Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18Relative Phase and Magnitude Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18Group Delay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

Important Parameters when Specifying Group Delay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19Absolute Group Delay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20Upconversion/downconversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20Modulation delay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20Time domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21Measuring delay linearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

Reflection Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24Isolation Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

Feedthrough Measurement of Converters and Tuners . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

Absolute Group Delay — A More Accurate, Lower Ripple Technique . . . . . . . . . . . . . . . . . . . . 27Measurement Configuration Using the Mixer Pair . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

Labeling Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28Proper Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29LO Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29System Calibration and Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

Calibrating the Test System with the Calibration Mixer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29Calibration Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29Calibration Error Terms and Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30Test Procedure for Calibrating the Test System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

First-Order Error Correction: Frequency Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31Second-Order Error Correction: Frequency Response and Input Match . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32Third-Order Error Correction: Frequency Response, Input and Output Match . . . . . . . . . . . . . . . . . . . . . . . . 33

Appendix A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 Calibration mixer attitudes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

Appendix B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 Program for Fixed IF Measurements with One External LO Source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

Appendix C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 Uncertainty in Mixer Group Delay Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

Appendix D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 Related Application and Product Notes/Other Suggested Reading/Third Party Companies . . . . . . . . . . . . . . . . . 39

3

Introduction

Frequency-translation devices (FTDs)such as mixers, converters, and tunersare critical components in most RFand microwave communication systems. As communication systemsadopt more advanced types of modulation, FTD designs are increas-ingly complex, tests are more stringent with tighter specifications,and the need to reduce costs is moreimportant than ever.

The measurement trade-offs for frequency-translating devices varywidely among different industries.Measurement accuracy, speed, costand ease of setup are among the considerations for determining thebest test equipment. This applicationnote explores current test equipmentsolutions and techniques that can beused to accurately characterize andtest frequency-translating devices.Frequency-translating devices presentunique measurement challenges sincetheir input and output frequencies differ. These require different measurement techniques than thoseused for a linear device such as a filter. This note covers linear frequecy-translation measurements,such as magnitude, relative phase,reflection and isolation. Correspondingaccuracy issues are also discussed.

To get the most from this note, youshould have a basic under-standing offrequency translation terminology,such as “RF port,” “IF port” and “LOport.” Understanding of fundamentalRF and network analyzer terms suchas S-parameters, VSWR, group delay,match, port, full two-port calibration,and test set is also expected. For abetter understanding of such terms, alist of reference material appear in theAppendix section.

Network analyzer mixer measurement configurations

Network analyzers used for testingfrequency-translation devices includescalar network analyzers, vector network analyzers with frequency offset capability, and vector networkanalyzers using an upconversion/downconversion configuration. Each solution has its own advantages anddisadvantages. This section provides asynopsis of the three configurations soyou can quickly evaluate which is thebest fit for your measurement needs.Detailed information about each solution is discussed in later sections.

4

Scalar network analyzer configuration

The most economical instrument forFTD tests is a scalar network analyzer.A scalar network analyzer uses diodedetectors that can detect a very wideband of frequencies. This capabilityenables a scalar network analyzerto detect signals when the receiver frequency is different from the sourcefrequency. Magnitude-only measure-ments such as conversion loss,absolute output power, return loss andisolation can be made, as well as nonlinear magnitude measurementssuch as gain compression. Group delayinformation is available in some scalarnetwork analyzers using an AM-delaytechnique, which employs amplitudemodulation.

AM-delay measurements are less accurate than group delay measure-ments obtained with a vector networkanalyzer. AM delay typically has anuncertainty of around 10 to 20 ns,whereas group delay with a vectornetwork analyzer has an uncertaintyas good as 150 ps. Advantages of thescalar solution include low cost andgood magnitude accuracy. As shown inFigure 1, fully integrated scalar network analyzers such as the 8711Cor 8713C provide economical RF measurements up to 3 GHz, andinclude AM delay capability. The8757D scalar network analyzer, shownin Figure 2, measures up to 110 GHz,and provides very good absolutepower measurements, particularlywhen installed with an internal powercalibrator and used with precisiondetectors. In certain cases, such asmeasuring FTDs with an internal filter,the 8757D with internal power calibrator and precision detector cantypically make more accurate magni-tude measurements than a vector network analyzer.

8711C RF network analyzer

External LO source Lowpass filter

10 dB

10 dB

External LO source

Directional bridge

Precision detector

Sweeper8757D

scalar network analyzer

Lowpass filter

Figure 2. 8757D scalar network analyzer configuration

Figure 1. 8711C scalar network analyzer configuration

5

Vector network analyzer in frequencyoffset mode configuration

A more versatile solution for FTD testis a vector network analyzer. A vectornetwork analyzer uses a tuned-receiv-er narrowband detector, which allowsmeasurements of both magnitude andrelative phase. The vector networkanalyzer’s frequency offset mode off-sets the analyzer’s receiver from it’ssource by a given LO frequency, andmakes frequency-translation measure-ments possible.

There are two common vector net-work analyzer configurations for FTDmeasurements. The simplest configu-ration is shown in Figure 3, and ispractical for testing upconverters anddownconverters. This configurationallows magnitude-only measurementswith a limited dynamic range. Forexample, if you are interested in themagnitude response of the FTD’s passband, the 8753E vector networkanalyzer has 35 dB of dynamic rangein the R channel and provides a quickand easy solution

To also measure the FTD’s relativephase and out-of-band response,Figure 4 illustrates a high-dynamicrange configuration. An alter-nativehigh-dynamic range configuration canbe achieved by splitting the analyzer’sRF output power between the deviceunder test (DUT) and the referencemixer (This configuration is similar tothe one shown in Figure 24). In bothconfigurations the vector network analyzer has around 100 dB of dynam-ic range. A signal into the reference Rchannel is always necessary for properphase-locking of the vector networkanalyzer. In addition, the R channelprovides a reference for ratioed mea-surements such as relative phase ormagnitude and phase tracking. Vectornetwork analyzers such as the 8720Dseries and the 8753E have frequencyoffset capability to 40 GHz and 6 GHz,respectively.

The upconversion/downconversion technique

A vector network analyzer in normaloperating mode can also be configuredfor frequency-translation measure-ments. This configuration has twomain advantages. First, the instrumentcan be used to measure a FTD’s magnitude and relative phase responsewithout the need for frequency offset.

As shown in Figure 5, two mixers areused to upconvert and downconvertthe signals, ensuring the same frequencies at the network analyzer’ssource and receiver ports. Second, thisconfiguration provides a potentiallymore accurate method for measuringabsolute group delay. You can simplymeasure two mixers and halve theresponse, accepting the resultinguncertainty.

Vector Network Analyzer

Lowpass filter

10 dB10 dB

RF in

Start: 900 MHzStop: 650 MHz

Start: 100 MHzStop: 350 MHz

Fixed LO: 1 GHzLO power: 13 dBm

FREQ OFFSON off

LOMENU

DOWNCONVERTER

UP CONVERTER

RF > LO

RF < LO

VIEW MEASURE

RETURN

Figure 3. Vector network analyzer in frequency offset mode

Vector Network Analyzer

Lowpass filter Reference mixer

CH1 CONV MEAS log MAG 10 dB/ REF 10 dB

START 640.000 000 MHz STOP 660.000 000 MHz

RF out

RF in

10 dB10 dB

Power splitter

External LO source

Figure 4. Vector network analyzer, high dynamic range configuration

External LO source

Vector network analyzer

Bandpass filter

Bandpass filterDUT Mixer

6 dB 6 dB6 dB6 dB6 dB

Powersplitter

RF

LO

IFIF RF

LO

Figure 5. Upconversion/downconversion configuration

6

Conversion loss

Definition and importance of conversion loss

Conversion loss, as shown in Figure 6,measures how efficiently a mixer converts energy from one frequencyto another. It is defined as the ratio ofthe output power to the input powerat a given LO (local oscillator) power.A specified LO power is necessarybecause while the conversion loss of amixer is usually very flat within thefrequency span of its intended operation, the average loss will varywith the level of the LO, as the diodeimpedance changes. As shown inFigure 7, conversion loss is usuallymeasured versus frequency, either theIF frequency (with a fixed LO) or theRF frequency (with a fixed IF). Theconfiguration for a fixed IF measure-ment is different from those describedup to this point. (See the Fixed IF

Measurement section.) Figure 8 illustrates the importance of a flatconversion-loss response. The DUT isa standard television-channel convert-er. The input signal consists of a visualcarrier, audio carrier and a color subcarrier. Since the frequencyresponse of the converter has a notchin the passband, the color subcarrieris suppressed and the resulting outputsignal no longer carries a valid color-information signal.

Frequency

Pow

er le

vel

Conversion loss

Conversion loss =

20*log [ ]mag(f )

mag(f )

RF

LO

IFIF

RF

Figure 6. Conversion loss

RF IF

LO

RF IF

LO

Loss

IF freq

Loss

RF freq

Conv loss vs IF freq(fixed LO freq)

Conv loss vs RF freq(fixed IF freq)

0 0

Converter response

Input Signal DUT Output signal

Audio carrier

Visual carrier Color

sub-carrierColor sub-carrier

attenuated

LO

Figure 7. Two types of conversion loss measurements

Figure 8. TV tuner conversion loss example

7

Measurement considerations

Conversion-loss measurements can bemade with either a scalar networkanalyzer or a vector network analyzer,using the configurations shown inFigures 1 through 5. The measure-ment uncertainties are different foreach type of analyzer. For both typesof analyzers, the two main systematicerrors are port mismatch and frequen-cy response. The scalar network ana-lyzer approach requires additionalcare to minimize errors due to theanalyzer’s broadband detector. Forsome vector network analyzers, aninternal process, called sampling, andphase-lock requirements can also cre-ate errors. Next we will examine eachof these error terms and explore tech-niques to minimize their effects.

Mismatch errorsMismatch errors result when there is a connection between two ports thathave different impedances. Commonly,a device’s behavior is characterizedwithin a Z0 environment, typicallyhaving an impedance of 50 or 75 ohms.Although the test ports of a networkanalyzer are designed to be perfect Z0impedances, they are not. The imper-fect source and receiver ports of thenetwork analyzer create errors in thecalibration stage. Therefore, evenbefore a device under test (DUT) isconnected, some errors have alreadybeen created in the calibration stage(see Figure 9). Once the DUT is connected, the total measurementuncertainty is equal to the sum of thecalibration error plus the measurementerror.

Once the DUT is connected, interac-tion between the DUT’s ports and thenetwork analyzer’s ports cause mis-match errors. As shown in Figure 9,mismatch effects generate three first-order error signals. The first is interaction between the network analyzer’s source port and the DUT’sinput port. The second is between thenetwork analyzer’s receiver port andthe DUT’s output port.

The third is between the network analyzer’s source port and receiverport. For an FTD measurement, thisthird interaction is usually negligiblebecause the conversion loss and isolation of the FTD will attenuate thereflected signals. As frequency transla-tion precludes conventional two-porterror correction, attenuators can beused to improve port match.

Source

Receiver

Measurement:

Calibration:

Source

Receiver

Total Uncertainty = Calibration Error + Measurement Error

ρ ρ

(ρ ρ )

(ρ ρ )

(ρ ρ )

RF IF

LO

source

receiver

DUT input

DUT output

source receiver

Calibration plane

source receiverDUT

Figure 9. Mismatch effects

8

By adding a high-quality attenuator toa port, the effective port match isimproved by up to approximatelytwice the value of the attenuation. A high-quality attenuator has around32 dB of port match. The effectivematch is a function of the quality ofthe attenuator as well as its attenua-tion, as shown in Figure 10.

As shown in Figure 11 and Figure 12, awell-matched attenuator can significantly improve the effective portmatch. For example, a 10-dB attenua-tor, with a port match of 32 dB, cantransform an original port match of 10dB into an effective match of 25 dB.However, as the match of the attenua-tor approaches the match of the original source, the improvementdiminishes. As shown in Figure 12, thelarger the attenuation, the more nearlythe resulting match approaches that ofthe attenuator. However, excessiveattenuation is not desired since thiswill decrease the dynamic range of themeasurement system. The port matchof an FTD can be poor, typicallyaround 14 dB. Therefore, it is recommended that attenuators beplaced at the FTD’s input and outputports. Scalar network analyzers usedifferent detection methods than vector network analyzers that shouldbe considered when testing FTDs.

0 5 10 15 20 25 30 3510

15

20

25

30

35

Original match (dB)

Effe

ctiv

e m

atch

(dB

) 32 dB attenuator match

26 dB attenuator match

21 dB attenuator match

18 dB attenuator match

Region when attenuator no longer results in improved match

Figure 11. Effective match as a function of attenuator’s match (fixed 10 dB attenuator)

0 5 10 15 20 25 30 350

5

10

15

20

25

30

35

Original match (dB)

Effe

ctiv

e m

atch

(dB

) 20 dB attenuation

10 dB attenuation

6 dB attenuation

3 dB attenuation

Region when attenuator no longer results in improved match

Figure 12. Effective match as a function of attenuation (attenuator match = 32 dB)

ρ ρ

Attenuator

(ρ )

(ρ )( ) 2

Source

(ρ ) + = (ρ 2

source E source match

attenuator

source attenuation

attenuator source attenuation)( )

ff

ρE source matchff

Figure 10. Effective match as a function of attenuator’s match.

9

Considerations unique to the scalar network analyzer

Scalar network analyzers use broad-band diode detectors. Although capable of both narrowband andbroadband detection, the 8711 series,which includes the 8712C and 8714Cvector network analyzers, uses broadband detection for FTD measure-ments. Therefore, if you use an 8712or 8714, use the same FTD test considerations as you would for ascalar network analyzer.

Importance of proper filteringA scalar network analyzer’s broadbanddiode detector will detect any signalthat falls within its passband. Althougha broadband diode detector is an economical way to measure FTDs, italso can allow certain detection errors.The diode detector will detect thedesired IF signal, as well as other mixing products or spurious signals. To minimize the detection of undesiredsignals, a filter should be placed at thedetector port to pass the desired IFsignal but reject all other signals.Figure 13 shows an example of theincorrect measurements that mightresult when improper IF filtering isused in a scalar network analyzer configuration.

In Figure 13, the conversion loss measurement without the IF filterappears to be better than it really is.The lack of an IF filter generates erroneous results. The broadbanddiode detector cannot discriminate thefrequency of the received signal(s) —it measures the composite response. If the source is set at 1 GHz, it is“assumed” that this is the frequency of the detected signal. Any signal thatfalls within the passband of the diodedetector will be detected. If the output of a DUT is composed of thedesired IF signal plus the image frequency, LO and RF feedthroughand other spurious signals, the diodedetector will detect the composite ofall the signals within its passband.This composite signal will be incor-rectly displayed as a response thatoccurs at 1 GHz.

Frequency response errorWithout performing any sort of calibration on a scalar or vector network analyzer, the frequencyresponse of the test system cannot beseparated from the FTD’s response.One way to correct these errors is toperform a frequency-response normal-ization or calibration, using a throughconnection in place of the DUT.

For scalar network analyzers such asthe 8757D, which very accuratelymeasures absolute power, the normal-ization calibration can be performed in two steps. See Figure 21. First, theabsolute RF power is measured andstored in memory. Second, the DUT is inserted and the absolute IF poweris measured. Conversion loss is displayed using the Data/Memory format. The conversion loss value isvery accurate since the measurementsof the two absolute power levels, RFand IF, are very accurate. Ratioing twovery accurate absolute power levelsremoves the frequency response error.In some cases, a scalar 8757D with aninternal power calibrator and precisiondetector can make more accurate conversion loss measurements than avector network analyzer. In theAccuracy Comparison of the 8757D

and a Vector Network Analyzer

section, error terms are used to illustrate how a scalar analyzer withinternal power calibrator can be moreaccurate than a vector network analyzer.

For analyzers that do not preciselymeasure absolute power, correctionsfor the frequency response error areless accurate. The input and output ofthe DUT are at different frequencies,but the normalization can only be performed over one frequency range.The result is that part of the test system is characterized over a differ-ent frequency range than that which isused during the actual measurement.

There are two choices for the frequen-cy range used for the normalization:either the DUT’s input (RF source)range, or the DUT’s output (receiver)range. The normalization should bedone to correct the portion of the testsystem that contributes the largestuncertainty; for example, this wouldbe the portion with the most loss orfrequency roll-off. Systems and components tend to have poorer performance at the higher frequen-cies, therefore the calibration shouldnormally be performed at the higherfrequencies. In general, high-quality,low-loss cables and connectors shouldbe used to minimize frequency-response errors.

For higher accuracy, combine a nor-malization calibration with externalerror-term correction. During the normalization, only one section of thetest configuration should be connect-ed, either the DUT’s input range orthe DUT’s output range. For highestaccuracy, the removed section can becharacterized separately. An externalcomputer is used to extract theremoved section’s S-parameters fromthe network analyzer. This data isthen used to modify the network ana-lyzer’s error terms to account for theeffects of the removed section.

Start 900.000 MHz Stop 1 000.000 MHz

2:Conv Loss /M Log Mag 1.0 dB/ Ref 0.00 dB

Swept Conversion Loss

1

2

1:Conv Loss /M Log Mag 1.0 dB/ Ref 0.00 dB

-9

-8

-7

-6

-5

-4

-3

-2

-1

Abs

dB

IF filter

No IF filter

2

1

Ch1:Mkr1 1000.000 MHz -6.38 dB

Ch2:Mkr1 1000.000 MHz -4.84 dB

Figure 13. Conversion loss response with and without an IF filter

10

Considerations unique to the vector network analyzer

Now that we have covered the important measurement considera-tions of the scalar network analyzer,let’s continue with a discussion of thevector network analyzer. The impor-tant considerations include: the needfor proper filtering, an accurate andstable LO, and power meter calibra-tion for the most accurate measure-ments.

Importance of proper filteringA vector network analyzer has a narrowband tuned receiver. Since thereceived signal is heavily filtered by an internal narrowband IF filter,broadband detection issues encoun-tered by the scalar network analyzerare not present. However, proper filtering is still very important for vector network analyzers with sampler-based receivers, such as the8753E and the 8720D.

Sampling architecture and issuesA sampler-based receiver consists of a voltage-tunable oscillator (VTO), apulse generator, and a sampler(switch). The VTO drives the pulsegenerator, which in turn drives thesampler. As a result, with proper tuning of the VTO, this combinationreplicates a down-converted input signal at the correct intermediate frequency (IF) for further processing.This combination is similar to a har-monic mixer in which the harmonicsof the LO are generated in the mixer,and the input signal can mix with anyharmonic. With proper tuning of theLO, one of the LO harmonics is offsetfrom the input signal to produce thecorrect IF signal.

Since there are many LO harmonics,any signal (desired or not) that is one IF away from any of the LO harmonics will be downconverted tothe network analyzer’s IF and detect-ed. To illustrate this sampler effect,let’s use the 8753E as an example.

The IF of the 8753E vector networkanalyzer is 1 MHz. Errors might resultbecause the incoming signal is not filtered until after it is downconvertedto the IF. If there is only one signal atthe receiver, this signal will mix withone LO harmonic and is properlydownconvert to 1 MHz. However, ifthere are multiple signals that are 1 MHz away from any of the LO harmonics, these signals will be down-converted to 1 MHz, which createserroneous responses.

Figure 14 illustrates an example ofthis sampler effect where the desiredIF output signal of the mixer is 110 MHz. In order to correctly detectthis signal, the 8753E will use a VTOof 54.5 MHz, where its second harmonic (109 MHz) will properlydownconvert 110 MHz to the desired 1 MHz IF signal. In the illustration, weshow two mixer products (6 LO-2RFand 9 LO-RF) that would also produceIFs at 1 MHz. Notice that these twospurs occur on either side of the LOharmonics (18 VTO and 42 VTO,respectively), but as long as they are 1 MHz away, they will be downcon-verted to 1 MHz. Aside from the signals which downconvert to 1 MHz,signals that will directly pass throughthe finite passband of the 1 MHz band-pass filter can cause problems. In the8753E, IF BWs from 10 Hz up to 6 kHzare available.

(RF – LO) – 1

Spur: 18 VTO – (6LO – 2RF) 981 – 980 = 1 MHz

Spur: 42 VTO – (9LO – RF) 2289 – 2290 = 1 MHz

MixerIF output

LOharmonics

RF – LO

6LO – 2RF 9LO – RF

110 980 2290

2VTO 18VTO 42VTO

109Frequency

Frequency

1 MHz 1 MHz 1 MHz

IF 1 MHz

LO harmonics

2

Desired signal: 2 VTO – (RF – LO) = 109 – (410 – 300) = 1 MHz

Given RF = 410 MHz IF = RF – LO = 110 MHz LO = 300 MHz

VTO = = 54.5 MHz

981 2289

Figure 14. Diagram of spurious measurement responses

11

As shown in Figure 15, wide IF BWallows any signal that falls within thepassband to be detected. For this rea-son narrow IF BWs are recommended.However, narrower IF BWs result inslower measurement speed.

Several techniques can significantlyminimize spurious responses. Onetechnique is to use adequate filteringat the network analyzer’s receiver port.Another technique, as illustrated inFigure 15, is to reduce the instru-ment’s IF bandwidth. Reducing the IFbandwidth will more selectively filtersignals in the instrument’s IF signalpath. A third technique is to avoid frequency spacings equal to the IF. Inthe case of the 8753E, the IF is 1 MHz.Therefore 1 MHz and multiples thereofshould be avoided when choosing frequencies and frequency spacings. A fourth technique is to avoid or eliminate certain frequencies thatmight cause spurious responses. Aspur prediction program, available inthe 8753-2 product note, predicts thefrequencies that might cause spuriousresponses. Modifying the frequencyselection or use the analyzer’s frequency list mode, in which theinternal source will step from one continuous wave (CW) signal to thenext, to avoid these frequencies. Thisprogram, written in BASIC 5.0, is customized for the 8753 series andswept IF mixer measurements,although it can be modified for othermeasurements such as fixed IF. Thisprogram only predicts the possibleoccurrence of a spur — it does notpredict its power levels, and it doesnot consider RF and LO subharmonics.

RF

LORF – LO RF + LO

1 MHz

Mixer RF

Mixer LO

Mixer IF

8753E VTO

harmonics

8753E IF

responses

IF detectors

BPF

LO harmonics

Mixer IF signals1 MHz

Selectible passband

Figure 15. Effects of IF BW on spurious measurement responses

12

R-channel phase-locking considerations For FTD tests, using frequency offsetmode with a vector network analyzer,many times you are required toremove the R-channel jumper andinput a signal into the R Ref-In port.Once an external signal is introducedinto the R channel, proper filtering atthe Ref-In port is essential. Certainrequirements must be met in order forthe R channel to phaselock properly.First, the signal into the R channelmust be within a specified power levelthroughout the entire IF range. Forexample, in the 8753E vector networkanalyzer, the R-channel signal must bebetween –35 dBm and 0 dBm through-out the selected IF range, otherwiseproper phase lock will not occur. If the R-channel filter does not coverthe entire IF range, modifying theselection of the IF range accordinglywill eliminate the phase-lock error. For example, if the IF range is set from10 MHz to 300 MHz, but the R-channelfilter is already rolling off at 300 MHz,then change the IF stop frequency toaround 250 MHz, or use a different filter. The second requirement is thatthe R-channel sampler needs to phase-lock with the correct signal. If thereare spurious signals entering the R-channel sampler, it might try to estab-lish phase lock with an improper frequency. This situation can be avoid-ed with proper R-channel filtering. (If you are using the 8712C or 8714Cvector network analyzers, R-channelconsiderations are irrelevant becausebroadband detetors are used.)

LO accuracy and stabilityFor the vector network analyzer, theuse of an accurate, stable LO isrequired for accurate magnitude andrelative phase measurements. Asshown in Figure 3, a network analyzerin the frequency offset mode will automatically set its internal source tosweep over the corresponding RFrange once you have entered the necessary information (i.e., IF range,LO information, upconversion ordownconversion, RF <> LO).Therefore, if the LO source is notaccurate, a network analyzer will notreceive the anticipated IF signal. As aresult, the IF signal can fall on theskirts of the IF filter or worse, it canfall entirely outside of the filter pass-band. In the first case, this inaccuracyis transferred to the measurementresults, and in both cases, phase lockcan possibly not even occur.

This LO accuracy and stability require-ment can be explained with the help ofFigure 16, which illustrates the blockdiagram of the 8753E in frequency offset mode. At the start of the sweep,the RF source is pretuned to the IFfrequency plus the LO offset, then themain phase-lock loop (PLL) is lockedup. The receiver will sweep over the IF range, and the source will track itwith the fixed offset. As shown, themixer's IF signal is downconverted inthe R-channel sampler to provide the1MHz phase-lock signal. Therefore, theLO source must be stable and accuratein order to provide an LO signal thatwill properly convert the mixer’s RFinput to its desired IF output. Thismixer’s IF signal, once downconvertedby the sampler, is compared to aninternal 1-MHz reference signal wherea resulting voltage, proportional totheir phase difference, is used to fine-tune the RF source. If the mixer’sIF signal is too far from the expectedreceiver’s frequency, then it might lieoutside the acquisition range of thePLL. The phase-lock algorithm will notwork and a phase-lock error is displayed. For example, the 8753Erequires a LO frequency accuracy ofwithin a few kHz of the nominal frequency.

1 MHzBPF

1 MHzBPF

IF detectors

Pulsegenerator

Pretune DAC

RF source0.03 – 3000 MHz

RF out

DUT

ExternalLO

source A / B

R

FLO

15 – 60 MHz

Reference1 MHz

Main PLL

RF IF

LO

Figure 16. 8753E block diagram of frequency offset mode

13

Power meter calibrationSome vector network analyzers, suchas the 8753E and the 8720D, are capable of using power meters toenhance their source and receiveraccuracy. This power-meter calibrationcan significantly minimize frequencyresponse error, as well as correct forsome mismatch effects. The followingdiscussion relates to a downconvertermeasurement, although the same concepts also apply for upconverters.It is a three-step process:

1. Perform power-meter calibration over the IF range.

2. Perform response calibration over the IF range.

3. Perform power-meter calibration over the RF range.

The power-meter calibration proce-dure needs to be done correctly or itcan lead to unexpected errors. Let’slook at each step for performing apower-meter calibration.

Step 1. Perform power-meter

calibration over the IF range.

The purpose of this step is to calibratethe network analyzer’s source for avery accurate power level over the IF range. After this, a response calibration completely corrects thereceiver’s frequency response over theIF range. With the power-meter calibration, the accuracy of the powermeter is transferred to the networkanalyzer.

First, the power meter is preset andzeroed. The network analyzer is set tosweep over the IF range.

Before connecting the equipment asshown in Figure 17, you need toensure that the R channel will phaselock properly. For network analyzers,such as the 8753E, that do not havean internal/external switch for the R channel, you must turn on the frequency offset before disconnectingthe R-channel jumper. This step isimportant since it prevents the network analyzer from attempting todo a “pretune calibration” when thereis no valid R-channel signal. If the network analyzer attempts a pretunecalibration without a valid R-channelsignal, the pretune constants would beinvalid.

Once the pretune constants are erroneous, proper phase lock will notoccur even with a valid R-channel sig-nal. If this should happen, reconnectthe R-channel jumper and allow theautomatic pretune calibration to fixitself. This step is not necessary for an8720D with Option 089 which providesan internal/external R-channel switch.

At this point, the LO frequency shouldbe at 0 Hz, so that the source andreceiver are at the same frequency.Connect the equipment as shown inFigure 17. The power at point A and Bare assumed to be the same, but actually are not. Therefore, it is important to make points A and B assymmetrical as possible. For example,if point B is connected to one arm ofthe power splitter, then ideally thepower sensor should also be connect-ed directly to the other arm of thepower splitter (that is, without anyadditional accessories such as anadapter or cable). Figure 17 is config-ured so that the mixer’s IF port willeventually be connected to point B.

A 0 dBm power level is optimal. Power meters are most accurate at 0 dBm because they use a precision50-MHz 0-dBm source as their refer-ence. If the frequency response tableof the power meter is used and themeasured power is close to 0 dBm,very high measurement accuracy canbe achieved (typically better than 0.05 dB). However, point B differsfrom point A by the tracking error ofthe splitter and this adds approximate-ly another 0.2 dB of uncertainty. Selectthe network analyzer’s source powerso that point A and B will be 0 dBm.

Set the power-meter calibration powerlevel. The number of calibration pointsshould be set at this time. Using nomore than 26 or 51 points usuallyyield very good results. The number ofpoints can then be increased beforecalibrating the network analyzer’sreceiver. The power-meter calibrationconstants will be interpolated over thenew points. Another time-saving tech-nique is to use one receiver calibrationfor different frequency ranges. This isaccomplished by calibrating over thelargest frequency span, then turningcalibration interpolation on and changing the span and number ofpoints. Interpolation errors are quitesmall, usually adding less than 0.05 dBto the total uncertainty.

During the power-meter calibration, itis important to connect the equipmentas shown in Figure 17. If point B isunterminated, errors will result due toreflections at this port. Terminatepoint B with the components withinthe IF path, such as the attenuator,filter and cabling. This closely reproduces the same condition duringthe calibration as in the transmission measurement. With the componentsconnected, the power-meter calibra-tion will account for mismatches dueto these components (such mismatch-es contribute to unleveled power levels at both points A and B). Analternative technique would be to terminate point B with a good Z0termination.

During the power meter calibration,via the GPIB interface, the networkanalyzer will automatically read theoutput from the power meter at eachfrequency point and adjust its sourcepower exactly by successively reset-ting and measuring the power. In mostinstances, two readings are sufficientto settle to within 0.02 dB of therequired power at point A.

Vector network analyzer

Lowpass filter

Power meter

GPIB

Power splitter 10 dBPower sensor

Point A Point B

Figure 17. Configuration for power meter calibration over the IF range

14

Step 2. Perform response

calibration over IF range.

With a very accurate power level atpoint B, now calibrate the networkanalyzer’s receiver for accurateabsolute power measurements overthe IF range.

The equipment is still connected asshown in Figure 17. As mentioned earlier, the power into the R channelmust be within the correct powerrange (–35 to 0 dBm, in the case of an 8753E), throughout the selected IF range, or proper phase lock will notoccur.

Examine the R-channel response. Thevalues displayed for the R channel canappear surprising. Notice that the values displayed are positively offsetby approximately 16 dB. For example,with a signal of 0 dBm, the R channelwill display approximately +16 dBm.Normal operation usually displaysratioed measurements (A/R or B/R). A mathematical offset is used toaccount for the differences betweenthe R-channel path and the A-or B-channel path. The network analyzercannot discern that the signal isdirectly applied to the R sampler, sothe signal level is displayed approxi-mately 16 dB too high. This offset andthe frequency-response error areaccounted for after a responsecalibra-tion is performed on the R channel.

Go to the calibration menu and perform a response calibration. Thethrough standard in Figure 17 is composed of the attenuator, filter, andcabling. After the response calibration,the R channel will indicate 0 dBm. At this point, the accuracy of thepower meter has been transferred tothe network analyzer's receiver atpoint B. In addition, the analyzer hasaccounted for the frequency responseof the through standard. Point B isnow calibrated for a very accurate 0 dBm absolute power reading.

Step 3. Perform power-meter

calibration for RF range.

This last step, another power-metercalibration over the RF range, provides a very accurate power levelat the RF port of the mixer.

An alternative is to do one power-meter calibration for both the RF andIF ranges. As mentioned earlier,power-meter calibration interpolationis a useful, time-saving tool. To savetime, the first power-meter calibrationcan be performed over a frequencyrange that encompasses both the RFand IF range. Then turn on interpola-tion to preserve the calibration in theseparate RF and IF ranges. This doescompromise accuracy to some extent.

Connect the equipment as shown inFigure 18. Once the appropriate information such as the LO informa-tion, upconversion/downconversion,and RF <> LO have been set, the network analyzer will automaticallyset the RF range. After selecting thedesired calibration power at port B,perform the power-meter calibration.

Upon completion of the power-metercalibration, a very accurate reading ofthe DUT’s absolute output power isdisplayed on the R channel, which hasbeen calibrated to 0 dBm. Notice thatthe reading is not necessarily the conversion loss of the DUT; it is theabsolute power reading of the DUT’soutput power. For example, if theDUT has a 6 dB conversion loss whenstimulated with –10 dBm of RF inputpower, the R channel will display –16 dBm. If you want the R channel todisplay the conversion loss directly,then in Step 1 select the power atpoint B to be the RF drive power forthe DUT’s RF port. The disadvantagehere is some loss in accuracy; powermeter calibration is most accurate at 0 dBm power level.

Vector network analyzer

Lowpassfilter

External LO source

Power meter

GPIB

Power splitter

10 dB

6 dB

Power sensor

Point APoint B

DUT

Figure 18. Configuration for power-meter calibration over the RF range.

15

Power-meter calibration for a high-dynamic-range measurementFor a high-dynamic-range configura-tion, Figure 19 is analogous to Figure17 above. Notice that the R-channeljumper is connected, since a valid signal is required into the R channelfor phase locking. In this configura-tion, the R channel is used only for thepurpose of phase locking. It is notincluded as a direct path in the actualmeasurement. This is the high-dynam-ic range configuration you should usefor performing Steps 1 and 2.

Figure 20 is analogous to Figure 18.This high-dynamic-range configurationis useful for devices that includes a filter, as shown. A filter is especiallyimportant in the R channel to suppress spurious signals that cancause false phase locking. A filter intothe B channel might not be necessary.This is the configuration you woulduse for performing Step 3.

Vector network analyzerPower meter

GPIB

Power splitter 10 dB

Power sensor

Point A Point B

Figure 19. Power meter calibration over the IF range, high-dynamic-range configuration.

Vector network analyzer

External LO Source

Power meter

GPIB

Power splitter 10 dB

Power sensor

RF in

RF out

Reference mixer

Point APoint B

DUT

Lowpass filter

Figure 20. Power meter calibration over the RF range, high-dynamic-range configuration

16

Accuracy comparison of the 8757D and a vector network analyzer

In some cases, an 8757D with an internal power calibrator and precisiondetectors will make more accurateconversion loss measurements than avector network analyzer with powermeter calibration. To illustrate the difference in accuracy, let’s look at thethe error terms involved in a down-converter measurement in either case.

In Figure 21a, the source is set to theRF range. The precision detector veryaccurately detects the absolute powerat its port over the RF range. Theresulting errors are m1, the mismatcherror, and rdetector RF,the uncertaintyin the detector’s calibration over theRF range. The rdetector RF error is verysmall, approximately ± 0.18 dB.

Next, the RF response is stored intomemory. The purpose of this step is to correct for the frequency responseof the attenuator and the source,rsource RF.

In Figure 21b, the DUT is connected.The errors consist of mismatch errors,m2 and m3 , and the uncertainty of thedetector’s calibration over the IFrange, rdetector IF. The latter error isvery small. Notice that an attenuatoris not used at the output of the DUTsince the precision detector measuresmost accurately at its port. If addition-al devices, such as an attenuator or filter, are added, their effects will addadditional calibration steps and there-fore added uncertainty. This configu-ration is especially advantageous forFTDs with internal filters.

In this 8757D calibration and measurement, the total uncertainty is equal to the sum of three uncorrect-ed mismatch error terms, m1, m2, and m3. The detectors errors are verysmall and are negligible.

With the vector network analyzer withpower-meter calibration, five mis-match errors result. Let’s take a closerlook at the error terms involved in avector network analyzer with powermeter calibration measurement.

In Figure 22a, the source is set to theIF range. A power meter calibration isperformed. The errors consist of themismatch error, m1, and the sourcefrequency response error over the IFrange, rsource IF. The latter error issmall.

In Figure 22b, a normalization calibra-tion is performed. Errors = m2 +rreceiver IF, where m2= m1 + rdetector IFand rreceiver IF is the receiver frequen-cy response error over the IF range.

In Figure 22c, the source is set to theRF range and a power meter calibra-tion is performed. Errors = m3 +rsource RF , where rsource RF is thesource frequency response error overthe RF range.

In Figure 22d, the DUT is connected.Mismatch errors, m4 and m5, are generated. As illustrated, the total uncertainty is due to five uncorrectedmismatch error terms, m1, m2, m3, m4,m5. The other errors are relativelysmall.

In summary, the 8757D with its internal power calibrator and precisiondetectors can be more accurate than avector network analyzer with powermeter calibration since there are lessconnections, resulting in less mismatch effects and therefore lessmeasurement uncertainty. However,the vector network analyzer is capableof further error correction using error-term manipulations with anexternal computer. With additionalerror correction, the vector networkanalyzer might be the more accurateinstrument. In the section, Absolute

Group Delay — A More Accurate,

Lower Ripple Technique, proceduresfor external error-term manipulationsare discussed.

Sweeper

8757D with internal power calibratorSweeper

8757D with internal power calibrator

Precision detector

Precision detector

m1

m3m2

LO

(a)

(b )

B

B

System test ports

Figure 21. (a) Absolute power measurementover the RF range(b) Absolute power measurement over the IF range.

Vector network analyzer

Vector network analyzerVector network analyzer

Power meter

Vector network analyzer

m1

m3

m2

m5m

4

Power meter

Detector

Detector

System test ports

LO

RF IF

(a)

(d)

(b)

(c)

System test ports

System test ports

Figure 22. (a) Power meter calibration over the IF range. (b) Normalization calibration over the IF range. (c) Power meter calibration over the RF range. (d) DUT is measured.

(a)

(a) (b)

(c) (d)

(b)

17

Fixed IF measurements

When a communication systemreceives a signal using a superhetero-dyne receiver, the information isdownconverted to an appropriate frequency called the intermediate frequency. The components in the IFsection of the receiver usually providemost of the gain and frequency selectivity of the communication system. Using a fixed IF allows adesigner to optimize the IF filteringand amplification which yield the highest performance. Typical IF frequencies for FM radios are 10.7 and 21.4 MHz, radar receivers are 30and 70 MHz and satellite receivers are70 and 140 MHz.

The other way in which conversionloss or gain flatness can be measuredis by keeping the IF at a constant frequency. This is known as a fixed-IFmeasurement. In order to accomplishthis, the LO must sweep in conjunc-tion with the RF input signal, keepinga constant frequency offset equal tothat of the IF. In many cases, this measurement more closely matchesthe operation of the DUT in the actualapplication.

The most common way to perform this measurement is to use the vectornetwork analyzer in a tuned-receivermode (without frequency offset).Figure 23 illustrates this configuration.As shown, two external sources areused. One external source providesthe RF signal and the other providesthe LO signal. Both sources sweep in a stepped-frequency mode under thedirection of a measurement controllervia the GPIB bus. This measurement is an ideal candidate for automationusing the built-in test-sequencing orIBASIC capability of many modernnetwork analyzers. Also, notice that anexternal reference must be connectedbetween the VNA and the sources (i.e. a 10 MHz reference). When testing a tuner with a fixed-IF sweep,the controller must be able to tune the internal LO. If an analog voltage is used, then a programmable powersupply or digital-to-analog converteris needed. If the LO is tuned digitally,the proper interface must be used.

Testing a mixer under fixed IF conditions can also be accomplishedwith one external source and the network analyzer under the control ofan external computer. The configura-tion is very similar to Figure 23,except there is only one external LOsource and a computer interface forboth the network analyzer and theexternal LO source. The configurationis shown in Appendix C.

Two cases should be considered for a fixed IF measurement. In both, theLO and RF frequencies are steppedover their respective ranges but ineach case the network analyzer is configured slightly differently.

The first considers the FTD as adownconverter, where the RF outputfrequency of the analyzer is steppedover the RF range and the analyzermeasures a fixed IF frequency at theR-channel input. The second uses theFTD as an upconverter and the outputfrequency of the analyzer (test port 1)is fixed at the IF frequency. The analyzer must measure the steppedRF frequency at the R-channel input.Both measurements are accomplishedusing the frequency offset mode capability available on many modernnetwork analyzers. This tuned receiver mode allows measurement tobe made at frequencies other thanthat of the analyzer’s internal source.

The Appendix provides a program fora fixed IF measurement on a mixerthat operates as a downconverter. This Basic for Windows program usesthe 8753E as the RF source andreceiver and the LO signal is providedby a ESG-D3000A signal generator.The LO source commands are writtenin SCPI, so any compliant signal generator will work. The program setsan absolute power level to the mixer’sRF input port and measures theabsolute power level at the IF port.The analyzer’s receiver is first calibrated at the fixed IF frequency,then the analyzer’s RF source level isset. The difference between the twosignals is the conversion loss of themixer under test. The programassumes the RF drive level from thetest port remains constant over thestepped frequency range and alsodoes not include insertion loss fromthe test port cable. The accuracy ofthe measurement can be enhanced byincluding a power-meter calibration atthe IF and RF ranges as detailed inthe 8753E Operating Manual. Caremust be taken to ensure adequate signal level to the R-channel input orthe system will not phase lock to theIF signal. The R-channel inputrequires a signal level between 0 dBmand –35 dBm. Adequate filtering isalso required to prevent spuriousresponse from entering the receiver.

Vector network analyzer

External RF Source External LO Source

GPIB

10 MHzbandpass

filter

Externalreference

Externalreference

10 dB

3 dB6 dBRF LO

IF

DUT

GPIB

Figure 23. Fixed IF configuration

18

Relative phase measurements

Measuring mixer relative phase parameters such as tracking, relativephase linearity and group delayrequires a vector network analyzer.Relative phase between two mixerscan be measured, but the absolutephase response of a single mixer cannot. In order to make a phase measurement of a FTD, a secondmixer is needed to provide a referencephase signal. This reference mixer isneeded because when the analyzer isin frequency-offset mode, the sourceand receiver function at different frequencies and phase is not definedbetween two different frequencies.This reference mixer needs to be driven by the same RF and LO signalsthat are used to drive the mixer undertest. The IF output from the referencemixer is applied to the referencephase-lock R channel of the vectornetwork analyzer.

Relative phase and magnitudetracking

The capability to measure the amplitude and relative phase matchbetween frequency-translation devicessuch as mixers is increasing in impor-tance as the number of multichannelsignal processing systems increases.These multichannel systems, such as direction-finding radars, requirethat the signal transmission througheach channel be amplitude- andphase-matched. To achieve therequired match between channels,each channel usually is manufacturedwith matched sets of components.

The match between FTDs is definedas the difference in amplitude and relative phase response over a specified frequency range. Also, thetracking between FTDs is measuredby how well the devices are matchedover a specified interval after theremoval of any fixed offset. For exam-ple, this interval can be a frequencyinterval or a temperature interval, ora combination of both.

The configuration shown in Figure 24can be used to make magnitude andrelative phase matching measure-ments. First, one DUT is measuredand its response is stored in memory.

Then the second DUT is measuredand the analyzer’s data/memory feature is used to compare them.Theanalyzer accounts for theresponse ofthe test system since the two DUTsare measured with exactly the sameconditions.Figure 25 shows the magnitude and relative phase matchbetween the two DUTs.

If both DUTs must be measured simultaneously (for tuning), then youcan place one DUT in the B channeland one in the R channel. However, inthis configuration you are limited bythe dynamic range of the R channel.

Therefore, the DUT in the R-channelpath should not have a filter. If it does,you can only view the DUT’s passbandmatch. For two DUTs with internal filters, you cannot simultaneously viewthe response using the configurationin Figure 24.

Figure 26 illustrates a configurationthat takes advantage of the highdynamic range of the A and B channels. The analyzer compareschannel A and B and can be used toactively display A/B responses, or formaking high-dynamic-range measure-ments such as tracking of out-of-bandresponses. In this case, the R channelis used only for phase locking.

Vector network analyzer

Power splitter 10 dB

DUT 1

Power splitter

DUT 2External LO Source

RF in

LO

RF

IF

LO

RF

IF10 dB

Low pass filter10 dB

Step 1: Measure DUT 1, store data in memory Step 2: Measure DUT 2, display data/memory

10 dB

Figure 24. Mixer matching — magnitude and group delay

CH1 S21/M log MAG .02 dB/ REF 0 dB

CH1 START 250.000 000 MHz STOP 400.000 000 MHz

PC

Cor

Ofs

CH2 S21/M phase REF 0 200 m /

CH2 START 250.000 000 MHz STOP 400.000 000 MHz

PCCor

HldOfs

Vector network analyzer with direct receiver access

External LO source

LPFDUT 1

DUT 2

RF

LO

IF

RF

R A B

Powersplitter

Powersplitter

Figure 26. Simultaneous tuning and high-dynamic range configuration

Figure 25. Tracking responses

19

Group delay

Group delay is a measure of a signal’stransition time through a device. It is classically defined as the negativederivative of phase vs. frequency.Relative group delay measurementscan be made with the same setup as that shown for relative phase measurements.

Important parameters when specifyinggroup delay

The result of a group-delay measurement depends upon manymeasurement factors, including aperture, averaging and IF bandwidth.The true negative derivative of phasevs. frequency cannot be determinedbecause any measurement made witha network analyzer is discrete, withone phase point per frequency point,so there is no continuous functionfrom which to take a derivative.Instead we calculate the slope of thephase over a frequency range. We usethe values of two frequency points oneither side of the frequency of interestand calculate the corresponding phaseslope. As illustrated in Figure 27, thismethod can yield a result that is dif-ferent than the true derivative of thephase.

The number of frequency points overwhich the slope is calculated is theaperture. The default aperture is thesmallest difference between any twofrequency points. The aperture can beadjusted via the “smoothing aperture”control to use a wider frequency rangeto calculate the phase slope. The maximum aperture is limited to 20%of the frequency span of the measure-ment.

Aperture = frequency span /(number of points – 1)

The aperture determines the informa-tion that is transferred from oneresponse to the other. Any effects onthe phase response carry over directlyto the group-delay response. Forexample, noise in the phase responsedirectly translates into noise in agroup delay response.

For example, if you set a narrow aperture, the noise in the phaseresponse becomes more significant,making the group-delay response noisier. At the same time, a narrowaperture allows you to see variationsin the group-delay response, whichyou would miss otherwise. But, if youincrease the smoothing aperture toomuch, you begin to lose resolutionsince the peaks start to disappear.Figure 28 illustrates how dramatic the trade-off between resolution andnoise as a function of aperture can be.

If you do not want to trade off resolu-tion and noise, you can achieve a balance by trading off some measure-ment speed. Using averaging or narrowing the IF bandwidth resultsin slower measurement speeds

without degrading resolution or noise.

The smoothing aperture should always be specified for a group delaymeasurement. A group delay measure-ment is only truly valid if its apertureis specified. Figure 28 illustrates howdramatically different the group delayresponses can be, depending on thesmoothing aperture settings. Alwaysspecify the smoothing aperture whenmaking a group delay measurement. Ifyou do not want to trade off resolutionand noise, you can achieve a balanceby trading off some measurementspeed. Using averaging or narrowingthe IF bandwidth results in slowermeasurement speeds, withoutdegrading resolution or noise.

Phase

φ∆φ

∆ω

Frequency (ω)

Tangent

∆ω∆φ

Average delay

Frequency

Group delay ripple

t o

Group delay (t )g =− d φ

d ω= –1

360°d φd f*

fo

Gro

up d

elay

Figure 27. Group delay

20% Aperture

0

0.2

0.4

0.6

0.8

1

1.2

Frequency

Del

ay (

ns)

Minimum Aperture

0

0.2

0.4

0.6

0.8

1

1.2

Frequency

Del

ay (

ns)

2% Aperture

0

0.2

0.4

0.6

0.8

1

1.2

Frequency

Del

ay (

ns)

10% Aperture

0

0.2

0.4

0.6

0.8

1

1.2

Frequency

Del

ay (

ns)

Figure 28. Group delay as a function of aperture

20

Absolute group delay

If you are interested in absolutegroup-delay measurements a calibra-tion mixer is required. When measur-ing the group delay of a linear device,standard vector error correction callsfor a through connection (delay=0) to be used as a calibration standard. The solution to the problem of measuring the delay of a nonlineardevice like a mixer is to use a calibra-tion mixer with very small group delayas the calibration standard. There areseveral techniques that can be used tocreate a calibration mixer with knowngroup delay. Two mixers have beencharacterized by Agilent Technologiesfor this purpose:

Mixer ANZAC MDC-123MCL ZFM-4

Frequency range 30 MHz to 3,000 MHzdc to 1,250 MHz

Group delay 0.5 ns0.6 ns

Upconversion/downconversion

The group-delay values above wereobtained by measuring two mixers,using the upconversion/downconver-sion configuration shown in Figure 5.The measured group delay was veryconsistent, with flat frequencyresponse. For example, a pair of MDC-123 mixers was measured,resulting in 1 ns of group delayexcluding the effects of the IF filterand components placed between themixer pair. A single mixer, then, musthave a group delay between 0 and 1ns. Therefore, one can state that thegroup delay of the MDC-123 is 0.5 ns,with an accuracy of ± 0.5 ns. This is avery conservative uncertainty window.Therefore, the uncertainties of bothmixers are equal to their respectivegroup delay, although these are worse-case estimates. After the through calibration you are ready to test thefrequency-translating DUT, enter avalue of electrical delay to compensatefor the group delay of the calibrationmixer (–0.5 ns for the MDC-123 or–0.6 ns for the ZFM-4). The resultingaccuracy derived from this techniqueis typically the group-delay uncertain-ty of the calibration mixer.

Choosing a calibration mixer witheven smaller absolute group delay willfurther improve the measurementuncertainty because the worst caseerror is half the stated value.

Another important consideration toremember when selecting a calibrationmixer is that the delay of the devicemust have a relatively flat group delayresponse over frequency. The conven-tion is to select a mixer with very widebandwidth compared to the mixer tobe tested — the wider the bandwidth,the better the mixers’ delay linearitywill be. For further informationregarding the desirable characteristicsof a calibration mixer, see theAppendix section, Calibration Mixer

Attributes. When using the calibrationmixer in the test system, typically onlya reponse calibration is used to calthrough the mixer. The mismatchinteraction between the test systemand the cal mixer and subsequentmeasurements with the DUT willintroduce uncertainty in the form ofripple into the measurement. Addingattenuation before and after eachmixer can greatly improve this ripple.

Another technique for improving themeasurement uncertainty in absolutegroup delay and delay linearity usesan upconversion/downconversion withadditional error correction at themixer's test ports. The technique isdiscussed in the section, Absolute

Group Delay — A More Accurate,

Lower Ripple Technique.

Modulation delay

Measuring the group delay of convert-ers and tuners with inaccessible internal LOs is very difficult using themethod described above because welack the necessary phase-coherent LO for phase locking. One alternatemethod uses a definition of groupdelay as the time delay through themixer of a modulated signal. Thistechnique can measure group delay ondevices without needing a referencemixer. Using a frequency-responsenormalization, absolute and relativegroup delay can be measured.

As shown in Figure 29, the modula-tion-delay method for measuringgroup delay is accomplished by modulating an RF carrier. AM, FM or pulse modulation can be used. The carrier is swept through the frequency-translating device, the IFoutput is demodulated, and finally the phase of the demodulated signal iscompared with the original basebandmodulation tone (which can bederived by demodulating the carrier atthe input to the DUT). If square wavemodulation is used, zero-crossings canprovide the delay detection mecha-nism; if sine wave modulation is used,the phases of the sine waves must becompared. One main drawback forthe AM-delay technique is that itrelies on the device not having ampli-tude limiting, which would strip offthe amplitude modulation.

AM ModulatorRF

Sweep fmod

DUT

Phase detector

Gd = e (360 * f )– ÷ mod

LO

θ

Measure phasebetween twodemodulated signals

°

Figure 29. Modulation delay method for measuring group delay

21

As the carrier is swept through theDUT, the phase of the modulationvaries proportionately to the phaseresponse of the device itself. Groupdelay is calculated as follows:

Gd = – θe° ÷ (360*fmod)

where θe° is the phase differencebetween the demodulated input andoutput carriers in degrees, and fmodis the modulation frequency. The aperture of this measurement is equalto twice the modulation frequency.The effects of aperture are the sameas the phase-derivative method ofmeasuring group delay previously discussed. As the aperture gets larger,the measurement becomes less noisy,but resolution degrades.

AM-delay measurements are easilyperformed with the 8711C series ofnetwork analyzers (see Figure 30). A response normalization calibration is performed without the DUT connected. Then the DUT is connect-ed for relative or absolute group-delaymeasurement. The aperture of thismeasurement is fixed at 27.8 kHz,with typical accuracy around ± 10 to20 ns. The accuracy of this techniqueis limited by the diode detectors,which have less sensitivity than thenarrowband detectors used in the8653E or 8720D. Group-delay accura-cy with the 8753E or 8720D is better,typically around the picosecond range,depending on the smoothing aperturesettings. As shown in Figure 30, thephase-derived method (top response)is less noisy than the AM delaymethod (bottom response). Vectornetwork analyzers use a phase-derivedmethod of group-delay measurement,making it difficult to measure groupdelay of FTDs with internal inaccessi-ble LOs.

The FM-delay technique is very similar to the AM technique, in that ituses a modulated signal to measurethe group delay. The key difference is that FM modulation is used, whicheliminates errors due to amplitudenoise, and is not sensitive to ampli-tude limiting in the device under test.However, since the frequency at theoutput is dependent on the LocalOscillator (LO) frequency, the FMtechnique has difficulties when the LO is not synthesized.

Time domain

The measurement of group delaythrough a mixer can be easily accom-plished using the time-domain optionavailable on some vector network analyzers. This does not require theextensive hardware and filtering thatother traditional methods demand. Allthat is required to measure absolutegroup delay and delay linearity is asingle mixer, a 50-ohm airline, a short,and a VNA capable of time-domaintransformation. (This capability isavailable on the 8510C and 8720ESanalyzers with Option 010.)

The technique uses the measuredreflection coefficient from the mixerthat is terminated in the 50-ohm airline and the short. The measuredfrequency response is transformedinto the impulse response using thetime-domain option on the VNA.Knowing the absolute delay of the airline, the absolute delay through the mixer can be calculated by examining the two-way reflection from the short in time. In addition, the delay linearity of the mixer as a function of frequency can be mea-sured using the gating function on the VNA. The gating function filtersthe effects of reflections internal to the mixer and isolates only thetransmitted signal through the mixer.This gated signal contains the delaydistortion introduced by the frequencytranslation process. In this way, delaylinearity through the mixer can now be directly measured in the frequency domain.

The impulse response on a VNA simulates a traditional Time DomainReflectometry (TDR) measurement.TDR is a technique to generate animpulse or step waveform in time that is propagated down a transmis-sion line. The reflections from discontinuities can then be detected intime on the TDR display. The measured time delay represents thetwo-way electrical distance to the discontinuity in the transmission line.Individual discontinuities can beexamined in time if there is adequateelectrical separation between them.This same TDR measurement can besimulated using a high-performanceVNA capable of time-domain transfor-mation. In this case, the measured frequency response from the VNA ismathematically transformed into theimpulse response using the InverseFourier Transform algorithm presentwithin the VNA.The measurementaccuracy is improved by applying the standard one-port vector errorcorrection to the initial reflection mea-surement.

Ext detX – input

Ext detY – input

DUT

Detectors

871XC scalar orvector network analyzer

Normalize without DUT first,then measure with DUT

Power splitter

Figure 30. AM delay measurements using the 8711 series

Center 175.000 MHz Span 100.000 MHz

5 10 15 20 25 30 35

Abs

2

1:Memory Delay 5 ns/ Ref 0 sns

35

Center 175.000 MHz Span 100.000 MHz

5 10 15 20 25 30

Abs

ns 2:AM Delay /M 55.6 kHz 5 ns/ Ref 0 s

1

Ch2: Mkr1 25.370 MHz -16.45 ns

M1

1

Ch1: Mkr1 25.480 MHz -16.31 ns

22

As a measurement example, thereturn loss as a function of frequencyof the IF port of a broadband mixer ismeasured on the VNA. The RF port ofthe mixer is terminated with an airlineand a short. The proper LO drive is also applied to the mixer. The hardware configuration is shown inFigure 31. The VNA is used to mea-sure and transform the frequencyresponse of the mixer under test. The resolution of the time-domaintransform on the VNA is directly related to the measured frequencyspan. In order to examine two closelyspaced discontinuities in the timedomain, a large frequency span mustbe measured. However, limitations inthe operating frequency range of the device under test also limit thetime-domain resolution. It may benecessary to electrically separate the discontinuities whenever possible.Because we wish to isolate the reflected signal from the short in the time domain, placing the shortdirectly on the output of the mixermay not provide enough separationbetween the mixer’s own internalreflections and the signal reflectedfrom the short. Placing a 50-ohm airline between the mixer and theshort can improve the measurementresolution by electrically separatingthe reflections from the mixer and theshort, as shown in Figure 31.

For example, the VNA is calibrated foran S11 one-port cal over the IF fre-quency range of 50 MHz to 10.05 GHz.The LO was set to a fixed value of 20GHz. This IF range is chosen for tworeasons; it is within the operatingrange of the mixer’s IF port, and itallows the VNA to be used in the low-pass mode of operation. The low-passmode yields the highest resolution inthe time domain.

The measured return loss of the IFport as a function of frequency isshown in Figure 32. The return lossshows the typical peaks and valleysthat occur when multiple reflectionsare added and subtracted over themeasured frequency range. This fre-quency response is then transformedinto the impulse response using thetime-domain option on the VNA.Figure 33 shows the impulse responseof the mixer terminated in theairline/short.

This figure represents the TDRresponse as a function of time. Itshows the individual reflections from several discontinuities as theimpulse waveform propagates alongthe transmission line. The position of each impulse shows the electricaldistance in time from the calibrationplane.

The amplitude of each impulse showsthe average amount of reflected signalfrom each discontinuity. The threereflections to the left of the figure arecaused by discontinuities presentwithin the mixer: for example, theinput and output connectors andtransformers.

H

PORT 2PORT 2PORT 1PORT 1

to LO Source

Airline

ShortMixer

Figure 31. Hardware configuration for measuring the absolute delay and delay linearity of frequency translation devices

Figure 32. Measured return loss from the IF port of the mixer as a function offrequency, measurement made using a shorted airline placed on the RF port

CH1 S11 LIN 50mU / REF -100mU1

STOP 1.9 nsSTART -100 ps

1: 204.42 mU 888 ps

Lowpass Mode

Airline / Short

Mixer Response

Second Reflection fromAirline / Short

Figure 33. Low-pass impulse response of the mixer using a shorted airline placed on the RF port

23

The largest impulse on the figure iscaused by the reflection from theshort placed at the end of the airline.This is the discontinuity of interest, as it represents the signal that istransmitted through the mixer undertest. A marker is used to measure the position of the short at the end of the airline in time. Because the dis-continuities are electrically separatedin time, each reflection can be individ-ually examined. The measured markervalue of 888 ps represents the two-way reflection from the short as thesignal passes through the mixer andthe airline. By knowing the electricaldelay of the airline, we can subtract it from the measured delay, and thenby dividing the result in half, we can obtain the mixer’s absolute groupdelay of 168 ps. Here we are assumingthat the delay characteristics of the mixer are the same as the signalpasses through the mixer from the IFto RF and RF to IF ports.

Measuring delay linearity

Delay linearity is also an importantcharacteristic of components used inbroadband communication systems. In order to measure the delay linearityof the mixer using this technique, a time filter needs to be applied to the frequency domain data. The timefilter is used to isolate the reflectionfrom the shorted airline. The reflec-tion from the short contains the delaydistortion introduced by the frequencytranslation process as this signal ispassed through the mixer. The timefilter or gating function is positionedon the time-domain response aroundthe reflection from the short, asshown in Figure 34. This gated mea-surement now represents only theresponse of the signal that passedthrough the mixer. Eliminated are allother reflections from the measure-ment: the mixer’s own internal reflec-tions and any secondary reflectionsbetween the mixer and the short. Onlythe reflection from the short isallowed to pass through the time filter.

Once the gate is properly centered,the time-domain function can then be turned off while leaving this gateactivated. This will yield the frequencyresponse of just the first reflectionfrom the short, the signal that passedthrough the mixer.

The delay of this gated frequencyresponse is shown in Figure 35. Thisfigure shows the mixer’s delay lineari-ty as a function of frequency. In thiscase, the measured delay linearity isless than 100 ps peak-to-peak over a10 GHz span, and only 50 ps over themiddle 8 GHz.

Note that the ends of the frequencyrange show a downturn in the measured delay response. This may be the result of the windowingfunction and data extrapolation that is applied to the measured frequencyresponse data. Generally, this window-ing process distorts the measurementsat the very ends of the gated frequen-cy response. The initial frequencyspan should be chosen larger thanrequired to avoid possible errors inthe frequency range of interest.

The measurement of group delaythrough the mixer can also be charac-terized using the reflection from theRF port of the mixer. In this case, theIF port of the mixer is terminated with the airline and short. Here, thetime-domain transformation is typical-ly operated in the band-pass mode,because the frequency range of themixer’s RF port does not typicallyoperate down to DC. The band-passmode allows arbitrary selection of thestart and stop frequencies, which isvery useful for band-limited devices.Keep in mind that the band-pass modewill reduce the time-domain resolutionwhen compared to the low-pass modecovering the same frequency range. In this case, a larger frequency rangeor longer airline may be required inorder to improve the time-domain resolution.

STOP 1.9 nsSTART -100 ps

CH1 S11 LIN 50mU / REF -100mU 1: 204.42 mU 888 ps

Lowpass Mode

CH1 S11 DEL 50 ps / REF 880 ps

1

STOP 10.050 000 000 GHzSTART .050 000 000 GHz

1: 881 psec 5.000 000 GHz

Gating ON

Figure 34. Low-pass impulse response using a time gate to isolate the reflection from the short

Figure 35. Measured delay linearity of the mixer as a function of frequency

24

Reflection measurements

Now that transmission measurementshave been covered, let’s move on toreflection measurements. Reflectionmeasurements are linear, even whentesting frequency-translating devicessince the reflected signal does notundergo a frequency shift. Therefore,these measurements are essentiallythe same as for filters and amplifiers,with a few minor variations. The systematic errors in the test systemand the calibration techniques used to remove them are the same. Narrowband detection should be usedif available.

Mixers are three-port devices, and thereflection from any one port dependson the conditions of the other twoports. When measuring reflection on athree-port device, it is important toterminate the ports that are not beingtested with an impedance typical ofactual operation. While this is oftenthe characteristic impedance Z0(usually 50 or 75 ohms), it could alsobe more complex. For example, if theIF port of the mixer is directly connected to a filter, then this filtershould be used when testing RF- orLO-port reflection. In this case, itmight be necessary to disconnect theIF port of the mixer from the test setand connect the appropriate loadimpedance.

When testing the RF or IF ports, theLO signal should be applied at thepower level that will be used in actualoperation. Since the bias point andimpedance of the mixer diodes is afunction of LO level, the reflection atthe other ports will also be a functionof LO level.

Figure 36 shows the different RF-portVSWR results obtained with a broad-band load versus the IF terminationthat would be used in actual opera-tion. The lower trace was measuredwith the IF port terminated with a 50-ohm load. The upper trace was terminated with a filter and then a 50-ohm load. The effect of mismatchbetween the mixer and filter is readilyapparent. Usually, the mismatch iseven worse in the stopband where thematch is typically quite poor. This is avery common operating environmentfor a mixer.

This measurement was done usingnarrowband detection.The measure-ments of VSWR on the LO and IFports are very similar. For IF-portSWR, the RF port should be terminat-ed with a matched impedance and theLO signal should be applied at its normal operating level. For the LOport VSWR, the RF and IF portsshould both be terminated in conditions similar to what will be present during normal operation.

Testing converters and tuners is easiersince there are only two ports and theLO will already be at the correctpower level. The same requirement forproper termination of either the RF orIF port also applies.

Start 900.000 MHz Stop 1 000.000 MHz

RF Port SWR

1:Reflection &MSWR 0.1 / Ref 1.00 C

1.1

1.2

1.3

1.4

1.5

1.6

1.7

1.8

1.9

AbsCh1

1

M1

IF filter

50-ohmload

Figure 36. Reflection measurement

25

Isolation measurements

Isolation is a measure of the leakageor feedthrough from one port toanother. The more isolation a mixerprovides, the lower the feedthroughwill be. Isolation is a transmissionmeasurement since the stimulus isapplied at one port and the responseis measured at another. However, youare measuring a signal at the same frequency as the stimulus, and not thefrequency-shifted signal. As is the casefor reflection measurements, isolation/feedthrough is dependent also on thetermination and LO power on theports not being tested.

Three main isolation terms are ofinterest: RF-to-IF feedthrough, LO-to-IF feedthrough and LO-to-RF feed-through. The latter term is oftenimportant if the mixer is near thefront end of the receiver or tuner thatis connected to a cable or antenna. Inthis case, significant LO leakage couldcause interference in other frequencybands. Both the LO isolation terms aresmall for single- and double-balancedmixers. RF to IF feedthrough is alsolow in double-balanced mixers. RF feedthrough is usually less of aproblem than LO to IF feedthroughbecause in most cases, the LO powerlevel is significantly higher than theRF power level.

RF-to-IF feedthrough is measuredwith the same instruments and setupused to measure conversion loss(except frequency-offset mode is notneeded when using a vector networkanalyzer such as the 8753E). Becausethe source and receiver frequenciesare the same, set up the network analyzer to use narrowband detectionto make the measurement. The onlydifference in the hardware configura-tion is that the IF filter needs to beremoved so the RF feedthrough willnot be filtered out. Depending onwhich network analyzer is used, a frequency-response calibration or fulltwo-port calibration should be performed to improve measurementaccuracy.

The RF feedthrough level is verydependent on the level of applied LOsignal. For this reason, the measure-ment should be made with the LO signal present at its normal operatinglevel.

LO feedthrough measurements aremade the same way as describedabove. Remember to terminate the unused port with the properimpedance. The LO power level usedfor the measurement should be thesame level that is used during actualoperation of the mixer.

LO to IFIsolation

RF Feedthrough

V = 0RF(1)

VLO(2)

LO(3)V

LO to RFIsolation V = 0

IF(3)V

LO(2)

LO(1)V

V = 0LO(2)

VRF(1)

RF(3)V

VRF

VIF

VLO

1

VLO

VLO

VRF

2

3

Figure 37. Isolation

Start 900.000 MHz Stop 1 000.000 MHz

RF Feed through

1:Transmission /M Log Mag 2.0 dB/ Ref 0.00 dB

–18

–16

–14

–12

–10

–8

–6

–4

–2

Abs

dB

Ch1

1

Remove IF filterApply proper LO power

Figure 38. RF feedthrough

26

Feedthrough measurement of converters and tuners

The procedure for measuring RF to IF feedthrough of a converter or tuner is identical to that of a mixer.Since an IF filter is often included inthe DUT, the RF leakage is often verysmall. An instrument with highdynamic range should be used tomake the measurement, and averagingand a reduced IF bandwidth can alsobe required to lower the noise floor ofthe test instrument.

Measuring LO leakage requires a different technique and test instru-ment. Since the LO port is typicallyinaccessible, a normal network-analyzer transmission measurementcannot be used. The best instrumentto use for measuring LO leakage is aspectrum analyzer. The spectrum analyzer is connected to either the RFor IF port and tuned to the frequencyof the LO signal. Again, the unusedport should be terminated. If the LOfrequency of the converter or tuner isknown to within a few kHz, then thetuned-receiver mode of a networkanalyzer such as the 8753E could alsobe used to make this measurement.

The plot in Figure 39 shows a fairlyhigh level of LO leakage present at the output of a converter. If a scalarnetwork analyzer were used to measure conversion gain of thisdevice, this signal could cause somemeasurement inaccuracy because theanalyzer’s broadband detectors wouldinclude the LO as well as the desiredIF signal. In cases such as this, additional filtering should be added to reduce the level of the LO feed-through when measuring conversiongain.

Up to this point, the possible testequipment configurations have beendiscussed, along with linear measure-ments such as conversion loss, relativephase, reflection, and isolation. Oneconfiguration that has not been thoroughly discussed is the upconver-sion/ downconversion configuration.This configuration is useful for testingFTDs on vector network analyzersthat do not have a frequency offsetfeature. Furthermore, this configura-tion along with external error-termmanipulation via an external computercan yield very accurate, lower ripplemagnitude and phase measurements.

DUT

ATTEN 10dB CNT –26 . 00dBmRL 0dBm 10dB/ 5. 14983 GHz

MKR5. 149850 GHz

CENTER 5. 15 0075GHZ SPAN 5 .000MHzRBW 30 kHz VBW 30 kHZ SWP50. 0ms

–26 . 00 dBm

Spectrum analyzer

LO

Figure 39. Feedthrough of tuners and converters

27

Absolute group delay — a more accurate, lower ripple technique

Although this section describes a more accurate technique for measur-ing absolute group delay, the sameprinciples can be applied for makingmore accurate, lower ripple phase-related measurements such as relativephase and relative group delay, as well as very accurate conversion lossmeasurements.

This section provides more advancedtechniques and the operator shouldunderstand network-analyzer datatransfer and error-term manipulation.An external computer is required.Unlike the previous technique forabsolute group delay in which aresponse-only calibration is performedthrough a characterized mixer, theaccuracy of this technique dependsupon using additional vector error correction to calibrate the test system in order to reduce measurementuncertainity created by mismatchinteraction between the mixer and the system.

This technique uses an upconversion/downconversion configuration. It consists of a two-step process.

First, a calibration standard needs tobe characterized for both magnitudeand delay or phase. The standard'sdelay response can be characterizedusing any one of the techniquesdescribed in the earlier section ongroup delay. As an example, a mixerpair is measured in which one mixer isused as an upconverter and the otheras a downconverter. Use the simpleapproximation that a single mixer'sgroup delay response is just half themeasured response of the pair. Themagnitude response of the cal standard can be determined usingeither the frequency offset mode on a vector network analyzer such as the 8753E or a scalar network analyzer such as the 8757D. Once the magnitude and delay response is determined, a common data file can be created using the measuredresponses as a function of frequency.This data file will be used to modifythe error terms within a vector network analyzer during system calibration.

Second, the calibration mixer is usedto calibrate the test system. Duringthe calibration, the calibration mixer is the through standard. Since the calibration mixer has been character-ized, its effects can be removed.

The effects of the through standardare removed by modifying error-termsextracted from the network analyzer.The calibration mixer data file alongwith additional measured data areused to modify the extracted error-terms. These modified error terms are imported back into the networkanalyzer where the calibrationcalculations are performed. Therefore,the network analyzer perform all theerror-correction calculations, but theexternal computer is used to modifythe error terms.

Three different levels of error correction can be achieved dependingon how much accuracy is needed. The first level removes frequency-response error, the second levelremoves frequency response andsource mismatch, and the third levelremoves frequency response, sourcemismatch, and receiver mismatch. The complexity and the steps requiredincrease for each level of accuracy.Each will be discussed in detail.

Let’s begin with a discussion of thefirst step.

28

Measurement configuration usingthe mixer pair

When making FTD measurements the measurement accuracy of a network analyzer depends upon knowing and correcting for the reference (or calibration mixer standard) used during the calibrationprocess. The calibration mixer, alongwith additional standards, is used tocorrect for errors within the testsystem. The calibration mixer alsoprovides the necessary frequencytranslation required during the systemcalibration process. In order to use thecalibration mixer as a standard, itmust have certain attributes. SeeAppendix A for a list of attributesrequired in the calibration mixer.

The typical measurement configura-tion consists of two mixers, placed asan upconverter and a downconverterin order to obtain the same RF inputand output frequency at the networkanalyzer’s test ports. The vector network analyzer supplies and measures these RF signals. A commonLO source is shared between the two mixers. A typical configuration of atwo-mixer pair including accessoryhardware is shown in Figure 40.

Two key components are requiredwhen making upconverter and down-converter measurements on the mixerpair:

IF filter: The function of the IF filter is to separate the desired mixing product from the undesired product.

Attenuators: These are used to insurewell-matched ports on the individualmixers and the test system. Using sufficient attenuation and vector errorcorrection during the characterizationprocess will reduce these effects. Youshould avoid excessive attenuationsince this may introduce unwantednoise in the measurement system, andpossibly making the errors larger thana system with less attenuation. Thevalue of attenuation to use will dependon your selected configuration. Toreview the effects of attenuation,please refer back to the Magnitude

Measurements section.

Another approach would be to replacethe attenuation with an isolatorbetween the first mixer and the filter.The isolator may be a good way toreduce mismatch effects while main-taining a low insertion-loss measure-ment path in the system.

Labeling conventionsThe RF port of the mixer A (mixer Awill later be used as the calibrationstandard for the test system) will bereferred to as ARF, the IF port as AIF,and the LO port as ALO. The S-parameter for a reflected signal fromthe RF port will be referred to asS11ARF. All measured S-parameterswill consist of a complex number with real and imaginary values. Themeasurement group delay of mixer Awill be represented by GDA. Thegroup delay is specified by GDA with avalue in the units of time. To simplifythe equations throughout this paper,refer to Figure 41 as a guide.

The test configuration is as shownin Figure 42. A good measurementpractice is to maintain shortdistances between the componentsand the VNA. Shorter cable lengthsresult in less ripple effects. If possible,connect test components directly toport 2 of the VNA and use a cablefrom port 1 to complete the measure-ment path. This configuration willminimize ripple in S21 measurementscaused by the receiver mismatch of the analyzer. Mixer B, the two filters and any additional cables andattenuators connected to port 2 of thenetwork analyzer will be referred toas the hardware interface chassis (see Figure 42). The function of thischassis is to provide the necessary frequency conversion and filtering forthe vector network analyzer.

LO input

IF filter

LO input

RF input RF output

IF output

IF input

Figure 40. Typical configuration for a mixer pair

S22AIFS11A RFS21A

GDA

IF output RF input =AMixer A

LO input

AIFRFA

A LO

Figure 41. Mixer labeling example

Vector network analyzer

A BF

CablePort 1 Port 2

RF

LO

IFRF

LO

IF Hardware interfacechassis

Figure 42. Diagram of the suggested component connections

29

Proper filteringDepending on the level of filtering andisolation provided in the measurementsystem, some spurious signals may bemeasured by the network analyzer. Itis beneficial to make an initial test onthe VNA to verify the effects of "spurs"in the measured response. Performingmeasurements on a pair of mixersover the required RF and LO frequen-cy ranges is required to determine if additional filtering or isolation is needed in the test system. This measurement will also be used to optimize the VNA settings for IF BWand IF averaging.

LO effectsIf measurements at different LO frequencies are required, it isnecessary to verify the frequencyresponse of the system componentsacross the LO range. It is assumedthat the LO drive level to the mixer isconstant as the LO frequency isstepped through its specified range. If the LO drive level changessufficiently to alter the mixer perfor-mance then adjustments may berequired to minimize these effects.Mismatch effects may also change theLO drive to the mixers, therefore theaddition of padding and isolationmight be required to maintain theaccuracy of the measurements. TheLO source might also require amplifi-cation to sufficiently drive the twomixers.

System calibration and testDuring the calibration and measure-ment, it is best to use the step sweepmode. This sweep mode verifies phaselock at each measurement point andyields the highest accuracy. If themixer pair is a noninsertable device,you should calibrate the VNA with the “swap-equal-adapter” technique.Please refer to the VNA user guide for the details of this calibration procedure.

Calibrating the test system with thecalibration mixer

Calibration configuration

When making accurate measurementsusing a VNA, the analyzer must be calibrated to remove the effects ofsystematic errors from within the test system. In a traditional VNA calibration, a series of knownstandards is connected at the test

ports and measured. Next, a set oferror terms is created, based on themeasurements and knowledge aboutthe standards. These error terms areused to correct subsequent measure-ments from errors contained withinthe test system.

To provide absolute measurements offrequency-translation devices, such asmixers, upconverters and downcon-verters, it becomes necessary tocalibrate the system with a knowndevice that can also translate thefrequency during the test systemcalibration. A calibration mixer is used to calibrate the system after itsabsolute S-parameters are known. Thesystematic error terms can then be calculated based on measurements andknowledge of the standard(s)used.

A typical configuration of the test system, which includes the hardwareinterface chassis, is shown in Figure 43.The calibration mixer standard is connected between test ports 1 and 2during the calibration stage. Once thesystem has been calibrated, and the effects of the calibration mixeraccounted for, a DUT can be insertedbetween the test ports and measured.As a calibration verification check, thecalibration mixer can be measured andcompared to its stored data files. If thetwo sets of data files are not equal,then the test system has not beenproperly calibrated.

All connections to the test system willbe performed at the system test port 1and 2 as labeled in Figure 43. A coaxi-al calibration kit will be required whenperforming reflection calibrations ofthe test system at the system test port 1. If the calibration mixer or DUTis “noninsertable” into the test systemat the system test ports 1 and 2, thenuse the swap-equal-adapter techniqueto compensate for the test-port connections.

As previously described in Figure 43,the hardware interface chassis contains the necessary filtering,padding and frequency translationrequired by the test system and the DUT. The chassis is to remain connected to the VNA for the durationof the calibration and test unless otherwise stated. If the chassis is disconnected from the test systemafter calibration is performed, a newcalibration would then be required.Notice that since the chassis containsa mixer for frequency translation, this mixer requires a phase-locked LO, similar to the one provided to thecalibration mixer or DUT. A signalgenerator set to the appropriate LOfrequency can provide theses two signals. If the DUT contains an inter-nal LO, then a portion of this signalcan be supplied to the calibrationmixer and hardware interface chassisduring the calibration and test.

Figure 43. Typical configuration of test system

Vector network analyzer

Hardwareinterfacechassis

LO

System test port 1

System test port 2

Port 1 Port 2

Cal mixer or DUT

30

Calibration error terms and equations

This section introduces you to thepertinent calibration error terms andequations. It serves as an overview. Amore detailed, step-by-step procedurefor the calibration is presented in thesucceeding Test Procedure for

Calibrating the Test System section.

The first step to measuring a DUTrequires calibrating the test systemusing the calibration mixer standard.During the calibration, the S21 of thecalibration mixer and test system ismeasured as a two-port device thatyields (Equation A).

As shown, the measured transmissionresponse, S21 A

MXR, is a function of theactual S21 of the mixer, S21 A

MXR, and several error terms. Only theerrors terms associated with the for-ward direction are considered becausethe hardware interface chassis prevents measurements to be made inthe reverse direction, at the systemtest port 2.

S21 AMXR is the actual S21of the mixer.

This information should be containedin the stored data files of the calibra-tion mixer or it can be measured.

ESF, is the source match at the systemtest port 1. It can be determined using a standard one-port calibration techniques over the RF frequencyrange.

ELF is the input match of the hard-ware interface chassis, S11A

CH, which

is the system test port 2. The S11ACH

can be measured directly by measur-ing the return loss of this port afterthe VNA is calibrated over the IFrange using a standard one-port coaxial calibration.

S11 AMXR and S22 A

MXR are the actualinput and output match of thecalibration mixer. These are measuredat the RF and IF frequencies,respec-tively. This information should be contained in the stored data files ofthe calibration mixer or it can be measured.

Using the test system’s error termsand the characterization data file ofthe calibration mixer, the forwardtracking error term, ETF, of the system can be obtained (Equation B).

This adjusted error term, ETF, is latersubstituted into the VNA when makingmeasurements on the DUT. As shownabove, ETF can be corrected for theforward mismatch effects in thesystem. If the source and receiver mismatch is low then it may not benecessary to correct for all the errorterms in the equation. This will becovered in the test-system calibrationsection to follow.

Now that all the forward error termsof the test system have been obtained,a measurement on a DUT can be per-formed. The DUT measurement canbe represented with a similar equationreplacing the calibration mixer termswith the DUT (Equation C).

The actual match of the DUT, namelyS11A

DUTand S22A

DUT, can be

measured using a standard one-portcalibration over the RF and IFfrequency ranges, respectively. Usingthe error terms and the measuredvalue of the DUT, the actual S21of theDUT can be calculated. As mentionedpreviously, not all error terms may berequired to obtain an accurate estimate of the actual DUT response,S21A

DUT. If the mismatch of the test

system or the DUT is low, then correcting for the source and loadmatch effects may not be necessary.Equation D shows the calculation forS21A

DUTincluding source and load

match correction.

It is now time to apply these equationsto a system calibration and test. Thefollowing procedures outline the stepsused to calibrate the system using various levels of error correction. Eachlevel of error correction increases theaccuracy in the system calibration byincluding additional error terms asdescribed above. Once the system iscalibrated, the same techniques canbe applied to measurements of theDUT.

E x E SF LF 21 M

MXR= xS

21 AMXRS

ETF 1 – S11 AMXRx x 1 – S22 A

MXR

1–S11 ADUT

= x1

x E SF

21ADUTS 21M

DUTS x 1–S22 A

DUT

1

x E LF

x E TF

1–S11 AMXR

= x1

x E SF

21AMXRS 21M

MXR S x 1–S22 A

MXR

1

x E LF

x E TF

21 1 – 11 x E1 – S22x E ADUT SF LF

21MDUT= A

DUTADUT

ETF

xS SS

Equation A

Equation B

Equation C

Equation D

31

Test procedures for calibrating the test system

The basic concept for calibration ofthe test system involves inserting thecalibration mixer into the measure-ment path between the system testports 1 and 2. A measurement sweepis performed and the calibration errorterms are calculated based on themeasurement. Because the calibrationmixer is removed from the systemwhen performing a measurement on a DUT, the system's calibration errorterms are adjusted to remove the contribution from the calibrationmixer. The calibration mixer essentially provides the frequencytranslation required during the calibration of the test system.

There are several levels of error correction that can be performedwhen calibrating the test system. Thesimplest error correction techniquecharacterizes the frequency responseof the test system and does notinclude any source and receiver matcheffects. Here, the calibration mixer isplaced in the test system and a S21measurement is performed on the system and mixer. The measurementis adjusted for the calibration mixerresponse and the remaining responseterm is used to normalize subsequentmeasurements of the DUT. This technique does not remove the mis-match effects between the source andinput to the calibration mixer. It alsodoes not remove the effects of thereceiver match interaction with thecalibration mixer. Depending on thelevel of mismatch within the systemand calibration mixer, an error orripple in the calibration term might beintroduced. This ripple will presentitself when performing measurementson the DUT. If the ripple is too large,then additional error correction mightbe required.

Other levels of error correctioninclude characterizing the source and receiver match effects during the calibration procedure. Takingthese mismatches into accountduring calibration reduces the levelof ripple in the DUT measurements.Additional calibration standards and measurements are required inorder to characterize and removethese mismatch effects. To accurately characterize the effects from both the source and receiver mismatch,measurements are required at the RFand IF frequency ranges, respectively.

Below are the calibration proceduresfor the three types of error correctionthat can be implemented. Each procedure corrects an additional errorterm to increase the accuracy of thecalibration.

First-order error correction

during system calibration:

Frequency-response term only

For the first level of error correction,the forward tracking term, ETF, for the system components (network analyzer, hardware interface chassis,cables, etc.) is measured directly. This error term is adjusted for the calibration mixer response and used to normalize DUT measurements. Forthis level of error correction, there isno adjustment for source and receivermismatch effects that might introduceripple in the error term, ETF.

•Place the calibration mixer into thetest system at the system test ports 1and 2 and supply the appropriate LOto the mixer and the hardware chassis. Perform a response throughcalibration and extract the error term,ETF, from the VNA. Using the datasupplied with the calibration mixer,recall the group delay and themagnitude of S21. The S21 phase ofthe calibration mixer will be recon-structed from the group delay usingthe expression below.

•Adjust ETF by removing the contribution of the calibration mixer from the system’s through response.Note: a different ETF term should be created for each setting in LO frequency using the appropriate calibration mixer data from the file.Based on the calibration mixer’sresponse, a new forward trackingerror term, ENEW

TF , will be created

using the following calculation.

•Insert the new tracking term in the VNA’s memory. The response calibration has now been adjusted forthe calibration mixer. In other words,the actual S21 of the calibration mixerhas been removed from the forwardtracking error term. The VNA will nowapply the appropriate error correctionto subsequent measurements.

The system is now ready to measurethe frequency translation device undertest. The absolute group delay andconversion loss will be displayed onthe VNA. As mentioned, this level oferror correction does not include mismatch effects from the source and receiver. As a system verification, the calibration mixer should now be measured. Compare the measuredgroup delay and the magnitude of S21to the data stored on the mixer’s datafile. The two data sets should beequivalent. Note any ripple in the displayed response. At this point, you should determine if any additionalerror correction is required to reducethe ripple in the frequency response.

NEWTF

TF=EE

S21 AMXR

θ = mod 360 (–Gd x 360° x Frequency)

32

Second-order error correction

during system calibration:

response and input match terms

In addition to the forward trackingerror term used in the calibration, correction of errors resulting frommismatch between the source and calibration mixer RF input can bemade. These mismatch effects canintroduce error and ripple in the forward-transmission tracking termETF, which is used to normalize measurements of the DUT. By charac-terizing the source match of the VNAand measuring the input match of thecalibration mixer, the forward-trackingterm can be adjusted for input mis-match errors. A one-port calibration is used to characterize the inputmatch terms and a response throughcalibration is used for the forward-tracking term. The one-port calibra-tion yields three more error terms: the forward directivity, EDF, thesource match, ESF, and the reflectiontracking, ERF. A substitute full two-port calibration will be created butonly the above errors terms will beused by the analyzer, the other errorterms will be eliminated from theerror correction. This allows the VNA to provide the required error correction calculations. The VNA willalso display the corrected S11 and S21of the device tested. Following theprocedure below will create all thenecessary calibration informationrequired for a second level of errorcorrection.

•Create a full two-port calibrationplace holder, which will be used tostore the modified error terms. This can be accomplished by using a pre-existing full two-port calibration,which covers the desired RF frequen-cy range. If one does not exist, youcan emulate a full two-port calibrationover the desired RF frequency range. To emulate a calibration, press all analyzer’s keys as if you were perform-ing a calibration, but do not connectany calibration standards. Save thisinstrument state to memory or diskfile.

•The calibration configuration isshown in Figure 38. Perform a S11one-port calibration over the specifiedRF range at the system test-port 1.Extract the three error terms EDF,

ESF, ERF associated with the systemtest-port 1 using a computer and HP VEE or similar software.

•Using the one-port calibration, measure the input match, S11A

MXR, to

the calibration mixer’s RF port. This measurement might have beenpreviously stored during the mixer characterization.

•Perform a response through calibration with the calibration mixer connected between test port 1 and 2.Extract the transmission trackingterm, ETF, from the VNA.

•Using the data file associated withthe calibration mixer, adjust ETF bythe calibration mixer’s S21 and inputmatch using the following equation.

•Recall the previously saved two-portcalibration. Insert the following termsinto the error terms of this two-portcalibration.

The VNA can now be used to measurea DUT. The absolute group delay andconversion loss will be displayed onthe VNA. Using this error correction,it is also possible to display the corrected input match, S11A

DUT, of the

DUT. The error correction implement-ed adjusts for systematic errors in thefrequency response and source matchof the system. If different LO settingsare required, it might be necessary to create a new set of error terms foreach LO frequency used. The calibra-tion mixer should be measured to verify system performance. Comparethe group delay and the magnitude ofS21 to the data stored on the mixer’sdata file. The two data sets should beequivalent. Note any ripple in the displayed response. You will need todetermine if additional error correc-tion is required to further reduce theripple in the frequency response.

NEWTF

TF=EE

S21 AMXR

x E SF 1 – S11 AMXR

NEWTFDFE , RFSF E ,E , E

LFE = 0

Third-order error correction

during system calibration:

response, input and output

match terms

This level of error correction includesall the above terms for transmissionfrequency response and source match,but also adds an additional correctionfor the receiver match of the test system. Only error terms in the for-ward direction can be implemented inthis system. The resulting error termsconsist of the frequency responseterm, ETF, the system test-port 1terms, EDF, ESF, ERF and the systemtest-port 2 receiver match term, S11A

CH. The input and output match

and the transmission characteristics ofthe calibration mixer are also requiredto implement this error correctionscheme during system calibration.

•Create a full two-port calibrationplace holder, which will be used tostore the modified error terms. Thiscan be accomplished by using a pre-existing full two-port calibrationthat covers the desired RF frequencyrange. If one does not exist, you canemulate a full two-port calibrationover the desired RF frequency range.To emulate a calibration, press all theanalyzer’s keys as if you were perform-ing a calibration, but do not connectany calibration standards. Save thisinstrument state to memory or diskfile.

•The calibration configuration isshown in Figure 38. Perform a S11one-port calibration over the specifiedIF range at test port 1. Measure boththe match of the hardware interfacechassis, S11A

CH, and the IF port of the

calibration mixer, S22 AMXR. The

calibration mixer data file may containthe IF port match of the mixer. Savethis data to the computer.

•Perform a S11 one-port calibrationover the specified RF range at the system test-port 1. Extract the threeerror terms EDF, ESF and ERFassociated with the system test-port 1using a computer and HP VEE or similar software.

•Using this one-port calibration, measure the input match, S11 A

MXR, to the calibration mixer’s RF port. This measurement may have been previously stored during the mixercharacterization.

•Perform a response through calibration over the RF range with the calibration mixer connected to the test ports. Extract the forward-tracking term, ETF , from the VNA.Remove the contribution of the cali-bration mixer from the calibration byadjusting the forward-tracking term by the mixer’s transmission response.Also include the effect of the inputand output match terms at the RF andIF frequencies, respectively, using Equation A.

•Recall the full two-port calibrationand insert the following error termsinto the VNA’s two-port calibrationmemory. Verify that the correct isused for each setting in LO frequency(see Equation B).

The VNA can now be used to measurethe DUT. The absolute group delayand conversion loss will be displayedon the VNA. It is also possible to display the corrected input match,S11A

DUT, of the DUT. The error correction implemented adjusts forsystematic errors in S21 caused by thefrequency response and source matchof the system during both calibrationand test. It only corrects for receivermatch effects during system calibra-tion. To include the effects of receivermatch during measurement of theDUT, the output match of the DUTwould be required. Using a one-portcalibration over the IF range, the output match of the DUT, S22A

DUT ,

can be measured. Additional calcula-tions are required to remove thereceiver match effects from the actualS21 of the DUT. The match of thehardware interface chassis, S11A

CH , is also required over the specified IFrange.

The corrected S21 for the DUT includ-ing receiver match effects as shown inEquation C.

This calculation can easily be performed on the computer. It is possible to change the error terms ofthe full two-port calibration to includethis effect thereby using the VNA toperform all error correction in theinstrument.

As previously discussed, the systemcalibration can be verified by leavingthe calibration mixer connected to thetest system and performing S11 andS21 measurements. After calibration isperformed, the calibration mixer char-acteristics are removed from the cali-bration terms. At this point the calibration mixer can be measured asa DUT. These measured S-parameterscan then be compared to those valuessaved in the characterization data fileprovided with the calibration mixer. If the system was calibrated correctlythen the two sets of data should beequivalent. Depending on the level of error correction used during systemcalibration and test, there can beslight differences between the mea-sured results and the stored charac-teristics.

At any time, if the system calibrationneeds to be re-verified, then placingthe calibration mixer into the test system and measuring its S-parame-ters will provide the system check. As long as the calibration mixer’s measured response agrees with thestored characteristics, then the systemcalibration is still valid.

33

x E SF x 1 – S11 AMXRx x 1 – S22 A

MXRNEWTF

TF=EE

S21 AMXR

S11 ACH

21 1 – 22 x NEWDUT = A

DUTxS S 21A DUTS S11 ACH

NEWTFDFE , RFSF E ,E , E

LFE = 0

Equation A

Equation B

Equation C

34

Appendix A

Calibration mixer attributes

The quality of the response measure-ment of the calibration mixer dependssomewhat on the characteristics ofthat mixer. These attributes are listedbelow:

Frequency response The calibration mixer should have abroad frequency response for all theports, RF, IF and LO ports. A low conversion loss for the mixer is notimportant in this mixer. It is especiallyimportant that there are no sharp discontinuities or variations in the frequency response of the calibrationmixer, as these can indicate poormatch conditions which can causeerrors during calibrations. Finally, themixer response should not be sensitiveto the LO level, provided a sufficientLO level is present. A low level mixermay be desirable for calibration so that the LO may be padded with awellmatched attenuator.

Return lossThe calibration mixer must have goodmatch (low return loss) on the RF andIF ports. The mismatch between theseports and the test instrument is amajor source of error in the measure-ment of the mixer response. In mostcases, the mixer should have wellmatched attenuator pads added to the RF and IF ports as a compositecomponent with good match. Thereturn loss of the LO port is not soimportant, but if the LO drive is notwell matched, the mismatch can causevariations in the LO drive level,which may affect the mixer response.

IsolationThe calibration mixer should havegood isolation, so that the RF to IF and LO to IF leakage does not cause“spurs” in the response of the systemafter calibration. Using proper filteringin the test system will reduce theeffect of spurs in the final test system,but the leakage signals may reflectback into the mixer from the LO and output ports of the mixer, re-mix,and cause ripple in the calibrationresponse.

Spurious generationThe spurious generation of the calibration mixer should be low.Driving the mixer at relatively low levels, and padding the input of themixer, will help reduce spurious generation. Spurs created in the mixer may cause spurs in the network analyzer measurement response of the calibration mixer.

SizeSurprisingly, the calibration mixershould typically be of a small physicalsize. A small mixer will have less phaseresponse (and thus lower group delay)so any ripples caused by phase changethrough the mixer will be less and“smoother”, with peaksfarther apart infrequency. This helps in measurementswhere small gain or phase peak-to-peak variation over a small frequencyaperture must be measured.

In practice, the calibration mixershould have a much broader band-width than the mixer under test, thus ensuring that any mismatch orconversion loss rolloff effects will“common mode” out of the calibrationmeasurement. Broad bandwidth usually means smooth frequencyresponse. The mixer should be doublebalanced, or even triple balanced toreduce leakage and image responses.The mixer should be mated with highquality attenuators that have goodreturn loss over the entire range of LO,IF, RF and the sum products. The final calibration mixer may havethese pads “permanently” attached(e.g., using shrink tubing over theconnectors) so that the pad becomespart of the calibration mixer. The loss of the calibration mixer is not too important. The input and outputpads should be at least 6 dB, providedthe measuring instruments have sufficient dynamic range at these low signal levels.

The network analyzer should also havepadding on its input and output ports,placed at the ends of the test cables.Keeping the lengths between the test-port pads and the calibration mixer short will reduce the rate atwhich mismatch ripples appear, thusmaking narrow band differential gainand phase measurements less prone toerrors from mismatch. Finally an RFband pass or high pass filter should beadded between the analyzer ports andthe mixer under test, or better yet,between the analyzer coupled portat the sampler. This will reduce thenumber and size of any "spikes" in themixer response caused by out of bandsignals which might be sampled intothe measurement IF of the networkanalyzer.

Appendix B

Example program listing for fixed IF measurements using a single external source as an LO

1 ! *** Fixed IF Downconverter Measurements using the 8753E Network Analyzer and ESG-D3000A Signal Generator ****10 ! RE-SAVE "FIXED_IF" 20 PRINTER IS CRT 30 DIM Marker(51) !Create an array to hold the conversion loss values 40 ASSIGN @Ana TO 708 ! 8753E Vector Network Analyzer Address is 70850 ASSIGN @Lo TO 719 ! ESG-D3000A Signal Generator Address is 719 60 OUTPUT @Ana;"Pres" ! Instrument Preset 70 OUTPUT @Lo;"*RST" ! Instrument Preset using the SCPI command 80 OUTPUT @Ana;"FORM4" ! 8753E Data Transfer Format, ASCII with no header 90 ! 100 ! Enter Frequencies and Power Levels for Mixer under Test 110 ! 120 INPUT "Enter the Fixed IF Frequency (MHz)",If_freq 130 INPUT "Enter the LO Start Frequency (MHZ)",Lo_start_freq 140 INPUT "Enter the LO Stop Frequency (MHZ)",Lo_stop_freq 150 INPUT "Enter the Number of Frequency Point to Measure(3,11,21,26,51)",Pts ! Can be increased160 IF Pts<>3 AND Pts<>11 AND Pts<>21 AND Pts<>26 AND Pts<>51 THEN GOTO 150170 INPUT "Is the RF>LO (1) or RF<LO (2) ? (1/2)",Rflo 180 IF Rflo<1 AND Rflo>2 THEN GOTO 170190 Lo_increment=(Lo_stop_freq-Lo_start_freq)/(Pts-1) 200 INPUT "Enter the RF Power to the Mixer Under Test (dBm) (RF>-30 and RF<0)",Rf_power 210 IF Rf_power<-30 OR Rf_power>0 THEN GOTO 200 ! maintain adequate level for R-Channel220 INPUT "Enter the LO Power to the Mixer Under Test (dBm)",Lo_power 230 ! 240 PRINT "FIXED IF DOWNCONVERTER MEASUREMENTS" 250 PRINT "IF Frequency = ";If_freq;" MHz" 260 PRINT "LO Frequency = ";Lo_start_freq;" MHz to ";Lo_stop_freq;" MHz" 270 PRINT "LO Frequency Spacing = ";Lo_increment;" MHz" 280 PRINT "RF Port Power = ";Rf_power;" dBm" 290 PRINT "LO Port Power = ";Lo_power;" dBm" 300 PRINT 310 OUTPUT @Ana;"POIN";Pts ! Set the Number of Points on the 8753E 320 ! *** Set the 8753E RF Attenuator for the proper range ***330 IF Rf_power<=0 AND Rf_power>=-15 THEN Range$="00" 340 IF Rf_power<-15 AND Rf_power>=-25 THEN Range$="01" 350 IF Rf_power<-25 THEN Range$="02" 360 OUTPUT @Ana;"PWRR PMAN" ! Set Test Port Power Range Setting to Manual370 Power_range$="POWR"&Range$ 380 OUTPUT @Ana;Power_range$ ! Set Test Port Power Range 390 OUTPUT @Ana;"POWE ";Rf_power ! Set Test Port Power Value400 OUTPUT @Lo;"POW ";Lo_power ! Set LO Power on the ESG-D3000 410 ! *** Set the 8753E to measure absolute power at the R channel for the Fixed IF 420 OUTPUT @Ana;"MEASR" ! Measure and Display the R channel for absolute power measurements 430 OUTPUT @Ana;"CWFREQ ";If_freq;"MHZ" ! Set the 8753E to CW at the fixed IF Freq.440 OUTPUT @Ana;"LOFREQ 0HZ" ! Temporarily set the 8753E for an LO of 0 HZ 450 OUTPUT @Ana;"FREQOFFS ON" ! Set the 8753E to Freq Offset Mode 460 OUTPUT @Ana;"VIEM ON" ! Show the measured data (CW Freq) on the 8753E

Vector network analyzer

External LO source

Bandpass filter

HP-IB

3 dB

10 dB

RF in Bandpassfilter

Personal computer

10 dB

Figure 44. Connection diagram for fixed IF measurements using an external computer

35

36

470 ! *** Perform a Receiver Calibration on the 8753E at the Fixed IF Frequency *** 480 INPUT "Disconnect the R-CHANNEL jumper from the front panel, Press ENTER",A$ 490 INPUT "Connect any IF Components (filter, pad) placed after the mixer to the R-CHANNEL IN, Press ENTER",A$ 500 INPUT "Connect a cable from the TEST PORT 1 to the IF components on the R-CHANNEL IN, Press ENTER",A$510 OUTPUT @Ana;"REIC";Rf_power ! Set the 8753E Power Level Reference for Rcvr Cal 520 OUTPUT @Ana;"TAKRS" ! Calibrate the 8753E's Receiver for absolute power 530 ! *** Connect the Mixer to the 8753E and set initial Freq Offset conditions 540 INPUT "Insert the Mixer into the setup and Connect the LO, Press ENTER",A$ 550 OUTPUT @Ana;"LOFREQ ";Lo_start_freq;"MHZ" ! Set LO Freq setting on the 8753E 560 OUTPUT @Lo;"FREQ:CW ";Lo_start_freq;"MHZ" ! Set the LO Freq on the ESG-D3000A 570 OUTPUT @Ana;"LOPOWER ";Lo_power ! Set LO Power setting on the 8753E(reference only)

580 OUTPUT @Lo;"OUTP:STAT ON" ! Turn ON the LO Power on the ESG-D3000A 590 OUTPUT @Ana;"DCONV" ! Set the 8753E to measure a downconverter 600 IF Rflo=1 THEN 610 OUTPUT @Ana;"RFGTLO" ! Set the 8753E RF>LO 620 ELSE 630 OUTPUT @Ana;"RFLTLO" ! Set the 8753E RF>LO640 END IF 650 OUTPUT @Ana;"MARK1 0" ! Turn Marker ON for reading data, can be placed anywhere for CW meas 660 OUTPUT @Ana;"VIEM OFF" ! Turn on Freq Offset Block Diagram for Connection Verification670 INPUT "Examine the 8753E/ESG Screens and Verify the Connections, Frequencies & Power for theMeasurement",A$ 680 OUTPUT @Ana;"VIEM ON" ! Turn on the measurement display690 PRINT 700 PRINT "LO Freq (MHz) LOSS (dB)" 710 FOR I=1 TO Pts 720 Lo_freq=Lo_start_freq+(I-1)*Lo_increment 730 OUTPUT @Ana;"LOFREQ ";Lo_freq;"MHZ" ! Set LO Freq setting on the 8753E 740 OUTPUT @Lo;"FREQ:CW ";Lo_freq;"MHZ" ! Set LO Freq on the ESG-D3000A 750 WAIT 2 ! Wait to allow 8753E to phase lock, this setting should be optimized 760 OUTPUT @Ana;"OUTPMARK" ! Command to output marker value 770 ENTER @Ana;Mark1,Mark2,Stim1 ! Enter the marker value 780 Marker(Pts)=Mark1-Rf_power ! Calculation for conversion loss (dB): Meas-Input levels(dBm) 790 PRINT USING 800;Lo_freq,Marker(Pts) 800 IMAGE 1X,4D.DD,10X,4D.D 810 NEXT I 820 PRINT830 PRINT "Measurement Complete" 840 END

Example output: mixer is the HP Avantek TFX-18075L

FIXED IF DOWNCONVERTER MEASUREMENTSIF Frequency = 170 MHz LO Frequency = 2400 MHz to 2600 MHz LO Frequency Spacing = 20 MHz RF Port Power = -20 dBmLO Port Power = 10 dBm

LO Freq (MHz) LOSS (dB)2400 -8.52420 -6.92440 -7.62460 -8.52480 -7.32500 -7.22520 -8.52540 -7.12560 -7.52580 -8.62600 -6.9

37

Appendix C

Uncertainty in mixer group delaymeasurements

Limitations of equipmentVector network analyzers (VNA) are commonly used to measure gain,phase and group delay of lineardevices. The accuracy of such mea-surements depends upon the stabilityof the test equipment, and the qualityof the user calibration. For these calibrations to work, the VNA typicallydepends upon the fact that the inputand output frequencies are the same.The user calibration consists of measuring several carefully controlledcalibration standards, and using theresults of these measurements toremove the effects of the test equipment response from the devicemeasurement. Unfortunately, no suchcalibration devices exist for mixers.For mixer amplitude response, thisdoes not create a great barrier. Othertechniques can be used to measurethe power applied to a mixer and thepower delivered to a load from amixer. The use of precision powermeters and scalar network analyzersin this application is common.Further, a precision power meter canbe used to calibrate a VNA which canmake these measurements faster, andwith more accuracy. But for phaseresponse and group delay, there exists no reference-standards thatmay be employed to perform such acalibration.

For the case of group delay, thereexists a further level of uncertainty.Since any measurement containsnoise, which is roughly constant withfrequency, as the frequency spacinggets smaller and smaller the group-delay calculation gets increasinglynoisy. Wider spacing leads to frequen-cy uncertainty as well as errors if thegroup delay response is not flat withfrequency, as in a filter. We’ll seeshortly that the aperture definitionalso limits delay measurement meth-ods that don’t depend of phase mea-surements.

Uncertainties in methodologiesEach methodology for measuringgroup delay, up/downconversion, modulation delay, and time domain,contains its own uncertainties.

Up/down conversionThe major uncertainty in the up/downconversion method is in determiningthe delay and the amplitude responseof the calibration mixer. One earlyapproach to remove the effects of the second mixer in the up/downconversion was to use two small calibration mixers in an up/down conversion configuration, with thetotal delay being measured. A filter is required between the mixers toremove the effect of the unwantedside band. The total delay consists ofthe delay of the first mixer and thedelay of the second mixer and thedelay of the filter, which may be measured independently. After subtracting off the delay of the filter,the remaining delay is divided byassigned to the each mixer. In thiscase, the uncertainty is on the order ofthe delay of the mixer. One can arguethat since neither mixer has negativedelay (this argument might breakdown if unwanted side bands areallowed to mix in) then even if one mixer has zero delay, the mostdelay in the other mixer is the delaymeasured. Thus, if half the measureddelay is assigned to each mixer andthe maximum error is also half themeasured delay. By using very smallmixers, which are available withdelays less than 500 ps., this calibra-tion mixer may be used to measuredevices with greater delays.

This argument does not assume thatthe mixers are the same, only thatthey do not have negative delay. If themixers are of the same type, onemight argue that the delay in eachshould be approximately the same andthus the error much less thanassumed above. One approach wouldbe to measure the mixers in bothdirections, first forward then reverse.However, even if the result gave thesame reading, it does not guaranteethat the mixers’ delays are the same.At most, it shows that the mixers aresimilar when both are used as up converters, or both used as down converters.

In fact, a single mixer may behaveentirely differently in as an up-con-verter than as a down converter, butthe up/down combination would stilllook the same. For example, supposemixer A has 900 ps of delay as an up-converter, and 100 ps as a downconverter, and mixer B is similar.Connect A as an up converter and Bas a down converter gives 1000 ps ofdelay. Also, connecting B as an upconverter and A as a down convertergives 1000 ps of delay. But clearly, theup conversion delay is not at all closeto the down conversion delay.

In controlling the unwanted sidebands, the filter used in between the mixers will have mismatch at theunwanted side band (as well as somemismatch of the desired side band)than can cause signals to re-mix andcreate further errors. In this approach,one can estimate the errors by know-ing the match of the converters andthe match of the filter. Since only onemeasurement is being made, only oneaccounting for these errors is needed.

Modulation delayUncertainty and errors in the AMdelay occur in several main areas. Thefirst is the frequency of modulationand the modulation waveform. If thedevice being measured does not haveflat delay, higher frequencies cancause errors due to modulation sidebands being delayed differently. If asquare wave modulation is used, themodulation waveform consists ofmany, discrete modulation frequencieswith the higher frequencies beingdelayed differently than the lower frequencies. A second area of concernis with the detector response to themodulation frequency. If a broad banddetector is used, both side bands ofthe mixer will show up in the detector.If a filter is used, the filter responseand mismatch must be accounted for.In most detectors, noise in the detection system limits the lowest frequency to the range of kilohertz tomegahertz, thus setting the minimumaperture.

38

An additional error may occur in theamplitude modulator. If there is delayin the modulating circuitry, this mustbe accounted for in determininguncertainty. Some systems allow cali-brating for this delay by connectingthe detector to the source and measuring the delay at the input frequency. However, if the detector isfiltered, this is not practical, and ifnot, it depends upon the demodula-tion response of the detector to be flatacross the input and output range ofthe mixer.

For AM delay, an additional uncertain-ty is limiting and AM noise in the measurement. Some devices have builtin amplitude limiting, which strips offthe amplitude modulation, so cannotbe measured using the AM delay technique. Others have noisy LO’swhich limit the delay sensitivity.

In practice, the AM delay techniquecan give delay measurements in thelow nanosecond range. With theadvent of small, low delay mixers, the AM technique may not have theresolution of the up/down conversiontechnique.

Many of the difficulties in the AMtechnique can be overcome by usingFM modulation. It is not sensitive toamplitude limiting in the device undertest, and it eliminates errors due toamplitude noise. However, AM to PMconversion can cause some errors.Also, since the frequency at the output is dependent on the LocalOscillator (LO) frequency, if the LOused is not synthesized (as in somelow cost converters) the FM techniquecannot be used. A further difficulty isthat the FM detector is quite complex,and may not have flat broad bandresponse. Also the FM modulationmay not have the same delay at all frequencies, particularly if the sourceused has different filtering for differ-ent bands, and the modulation occursbefore the filtering, which is typical.The FM technique is primarily used to measure differential delay across a mixer-under-test that might havenarrow filtering following. It is usuallynot practical to measure absolutedelay with the FM technique.

Time domain The first uncertainty to consider forthe time-domain technique it due tothe reflection from the short repre-senting both the upper and lower sideband products from the mixer. If themixer has significantly different delayfor the upper (sum) products and the lower (difference) products, then the reflection from the short can be“smeared” and it will not represent thedesired delay. This is particularly truewhen looking at the gated response. In the extreme, there may be twoseparate responses from the shortrepresenting the upper and lower side band individually.

A second source of error occurs if themixer is not broad band. In this case,the delay through the mixer may besignificantly different at different frequencies. The delay measured bythe short’s reflection is the compositedelay of the entire response. However,the gated response should show thedifferential delay from which the delayat any frequency may be calculated.These measurements are of coursesubject to the uncertainty mentionedabove with respect to the unwantedside band response.

As noted in the chapter on relativephase measurements, the gated time-domain response can produceerrors at the edges of the frequencyresponse, mostly due to windowing,gating and re-normalization. In generalthe first and last 10 percent of thetime-gated frequency response is suspect.

Finally, this method depends upon the mixer having identical responseas an up- or down-converter, since the signal reflected from the shortmust be converted twice. Any non-symmetry in these response will pro-duce uncertainty in the measurement.

Comments on traceabilityIn specifying the performance of VNAsto measure linear devices, manufac-turers depend upon traceability tofundamental measurements. Thisoften leads back to such organizationsas the National Institute for Standardsand Technology (NIST). But there are no such traceable standards for converter delay. Traceability may notbe a requirement for an R&D engineerworking at his bench, where he under-stands the block diagram and interac-tions between components. But in the manufacturing arena, as well as in product acceptance, where therecannot be assumptions made aboutthe internal design of a component,strict limits on the error bounds ofmeasurements are important. Thedetermination of these limits must bebased on sound engineering practicesand not on wishful thinking about thequalities of devices.

With the introduction of new measure-ment instruments, such as vectorspectrum analyzers and vector signalgenerators, more techniques arebecoming available to lessen theuncertainties in these measurements.Further, new investigations into mea-surement methodologies may providethe path to generate calibration tech-niques appropriate to these devices.

Finally, it is clear that new transferstandards are needed in this area.Similar problems have been addressedin other areas such as electro-opticconverters, but we are still lackingfundamental standards and techniquesfor frequency converter measurementtraceability.

39

Appendix D

Related application and product notes

Understanding the Fundamental Principles of Vector Network Analysis, Application Note 1287-1, 5965-7707E

Exploring the Architectures of Network Analyzers,Application Note 1287-2, 5965-7708E

Applying Error Correction to Network Analyzer Measurements,Application Note 1287-3, 5965-7709E

Network Analyzer Measurements: Filter and Amplifier Examples,Application Note 1287-4, 5965-7710E

In-fixture Microstrip Device Measurements Using TRL*Calibration,Product Note 8720-2, 5091-1943E

Specifying Calibration Standards for the 8510 Network Analyzer,Product Note 8510-5A, 5956-4352E

Applying the 8510 TRL Calibration for Non-Coaxial Measurements,Product Note 8510-8A, 5091-3645E

Measuring Noninsertable Devices, Product Note 8510-13, 5956-4373E

Improving throughput in Network Analyzer Applications, Application Note 1287-5, 5966-3317E

Using a Network Analyzer to Characterize High-Power Components,

Application Note 1287-6, 5966-3319E

Other suggested reading

“Design of an Enhanced Vector Network Analyzer,” Frank David et al., Hewlett-Packard Journal, April 1997.

“Calibration for PC Board Fixtures and Probes,” Joel Dunsmore, 45th ARFTG Conference Digest, Spring 1995.

“Techniques Optimize Calibration of PCB Fixtures and Probes,” Joel Dunsmore, Microwaves & RF, October 1995, pp. 96-108,November 1995, pp. 93-98.

“Improving TRL* Calibrations of Vector Network Analyzers,” Don Metzger, Microwave Journal, May 1995, pp. 56-68.

“The Effect of Adapters on Vector Network Analyzer Calibrations,” Doug Olney, Microwave Journal, November 1994.

“A Novel Technique for Characterizing the Absolute Group Delay and Delay Linearity of Frequency Translation Devices,”Michael Knox, 53rd ARFTG Conference Digest, Spring 1999, pp 50-56.

40

Third-party companies

Test fixturesInter-Continental Microwave1515 Wyatt DriveSanta Clara, California 95054-1524USAE-mail: icmfixture@aol.comPhone: (408) 727-1596Fax: (408) 727-0105

Multiport test setsATN Microwave, Inc.85 Rangeway RoadNo. Billerica, Massachusetts 01862-2105USAE-mail: info@atn-microwave.comPhone: (978) 667-4200Fax: (978) 667-8548World Wide Web Home Page: http://www.atn-microwave.com

Wafer probes and stationsCascade Microtech14255 SW Brigadoon CourtBeaverton, Oregon 97005USAE-mail: sales@cmicro.comPhone: (503) 626-8245Fax: (503) 626-6023

For more information about AgilentTechnologies test and measurement products,applications, services, and for a current salesoffice listing, visit our web site:

http://www.agilent.com/find/tmdir

You can also contact one of the following centers and ask for a test and measurementsales representative.

United States: Agilent Technologies Test and Measurement Call CenterP.O. Box 4026Englewood, CO 80155-4026 (tel) 1 800 452 4844

Canada:Agilent Technologies Canada Inc.5150 Spectrum WayMississauga, Ontario, L4W 5G1(tel) 1 877 894 4414

Europe:Agilent TechnologiesEuropean Marketing OrganisationP.O. Box 9991180 AZ AmstelveenThe Netherlands(tel) (31 20) 547 2000

Japan:Agilent Technologies Japan Ltd.Measurement Assistance Center9-1, Takakura-Cho, Hachioji-Shi,Tokyo 192-8510, Japan(tel) (81) 426 56 7832(fax) (81) 426 56 7840

Latin America:Agilent TechnologiesLatin American Region Headquarters5200 Blue Lagoon Drive, Suite #950Miami, Florida 33126, U.S.A.(tel) (305) 267 4245(fax) (305) 267 4286

Australia/New Zealand:Agilent Technologies Australia Pty Ltd 347 Burwood HighwayForest Hill, Victoria 3131(tel) 1-800 629 485 (Australia)(fax) (61 3) 9272 0749(tel) 0 800 738 378 (New Zealand)(fax) (64 4) 802 6881

Asia Pacific:Agilent Technologies24/F, Cityplaza One, 1111 King’s Road,Taikoo Shing, Hong Kong(tel) (852) 3197 7777(fax) (852) 2506 9284

Technical data is subject to changeCopyright © 2000Agilent TechnologiesPrinted in U.S.A. 3/005966-3318E