by David Ballo, Senior RF Application Development Engineer, Keysight Technologies
This two-part article describes how a modern vector network analyzer (VNA), in this case Keysight’s PNA-X Series, can be used for comprehensive testing of mixers and frequency converters. The first installment in the series was in the September issue of MPD and continues here with further details on specific applications that are well served by modern VNAs. Part one of this article appeared in the September 2023 issue of Microwave Product Digest and is available here.
Power Sweeps for Gain and Phase Compression
Power sweeps at a fixed frequency are useful for characterizing non-linear behavior such as gain and phase compression (also called phase versus drive). Power sweeps are also used for the related measurements of AM-to-AM and AM-to-PM conversion, where the slope of the gain and phase plots versus input power are calculated.
Gain compression is one of the most common figures of merit for active devices. It is generally considered the approximate point where a device transitions from linear to nonlinear behavior and is defined as the input or output power where the device’s gain drops by x dB compared to the linear or small-signal gain, where the user defines x. Typically, x is 1 dB and the compression point is P1dB.
Gain compression is easily measured on a frequency converter by performing a power sweep at the device’s input while measuring conversion gain (SC21). This can be repeated with different mixing plans, for example, with different input and LO frequencies or a fixed input frequency and different LO frequencies.
In Figure 1, Tr 1 shows a measurement of gain compression, where the input was swept from -40 dBm to -20 dBm. A compression marker (a marker-search function) shows the gain and input and output power associated with the 1 dB gain compression point. Using the AM Distortion feature (AM-AM), Tr 2 shows the compression response normalized to zero at the start of the power sweep, which is assumed to be in the DUT’s linear region of operation. By enabling the checkbox for “Y-axis = Slope Calculated Over Aperture” in the AM distortion feature, the derivative of the gain compression trace is directly displayed (Tr 3) and shows the change in decibels of gain per decibel change in input power. At the 1 dB compression point, the slope of the compression curve is -0.48 dB/dB.
A similar plot but with the traces formatted as phase is shown in Figure 2, where Tr 1 shows phase versus drive, and Tr 2 is the same response but with the AM Distortion feature (AM-PM) enabled, which normalizes the phase at the start of the power sweep to zero. Tzvhe phase compression marker on this trace shows that 1 deg. of phase compression occurs with considerably less power than the 1 dB gain compression point. With the Y-axis aperture checkbox enabled (Tr 3), the phase change per decibel change in input power at the 1-deg. compression point is shown as -0.33 degree/dB.
A related measurement to phase compression or phase versus drive is a measurement often called phase transfer. In this case, a large out-of-band signal is swept in power while the phase response of a smaller in-band signal is measured. This measurement must be done outside of SMC+Phase using a standard channel’s frequency-offset mode (FOM). The setup takes advantage of the PNA-X’s built-in signal combiner, which routes source 2 through the combiner and out through port 1, as shown in the path configuration dialog in Figure 3. The primary source supplies the swept-power out-of-band signal at 6.41 GHz, while the second source supplies a constant-power in-band signal at 5.91 GHz. The receivers are tuned to the in-band output frequency (1.51 GHz), and source 3 (available with Option XSB) provides the 4.4 GHz DUT LO from the rear panel to port 3. An external signal generator could also be used for the LO signal.
The Power and Attenuators dialog is used to configure the power levels. Here, Port 1 (the out-of-band signal) is swept from -50 dBm to -25 dBm, while Port 1 Src2 (the in-band signal) is kept at a constant -35 dBm. The LO power (Source3) is set to 18 dBm, which overcomes some system loss. The measurement is performed with a single receiver measuring output power and relative phase.
Because calibrating a standard channel cannot be done while using FOM as it is automatically turned off during the calibration, creating a second channel for the calibration that includes all the frequencies used in the phase-transfer measurement is required. The easiest way to accomplish this is to set up the second channel in segment-sweep mode, where three single-point segments correspond to the input in-band frequency, the input out-of-band frequency, and the in-band output frequency.
“Cal All” should be used to calibrate the second channel, and Enable Extra Power Cals should be configured to calibrate the output power of Port 1 Src2, and, if desired, the LO power. After completing the calibration, choose Save As User Calset and then apply the saved calset to the phase-transfer measurement channel using the Cal Set Selection dialog. When selecting the calset, use the “Do not change the active channel’s stimulus settings” to preserve the measurement settings and use interpolation as needed. Because of how error coefficients are calculated, averaging must be used for the phase-transfer measurement when calibration is applied.
Characterizing group delay versus LO frequency of broadband frequency converters that have narrowband IF filters is difficult. This is because the IF filter prevents broadband frequency sweeps, and swept LO/fixed-IF phase measurements of delay are impossible with SMC+Phase. One way around this is to use segment sweeps, each with a frequency span consistent with the IF response but different RF and LO frequencies that can cover the full range of the DUT.
An example of using segment sweeps is shown in Figure 4, where the segment spans are 315 MHz, and the LO is stepped in frequency from 4 GHz to 9 GHz in 500 MHz steps using 11 segments. In the top window, delay and magnitude traces are shown, where the segments are all plotted over the common output-frequency span of the downconverter. In the bottom plot, a mode called “X-Axis Point Spacing” is used, where the data in each segment is concatenated instead of overlaid, and the X-axis units are points instead of frequency.
Measuring Devices with Embedded LOs
Special considerations are required when measuring phase and group delay on frequency converters with embedded (internal) LOs that are inaccessible to provide or receive a frequency reference to ensure frequency synchronization between the DUT and the PNA-X. This situation is common with satellite transponders, where size, weight, and power limitations eliminate easy access to the LOs in the satellite. Historically, this challenge meant that VNAs were not used for transponder phase and delay measurements, dramatically hindering efforts to improve the speed of transponder characterization. With a modern instrument like the PNA-X with advanced measurement applications, the VNA is now the primary tool for transponder characterization.
SMC+Phase requires tuning the PNA-X receivers to frequencies that precisely match the output frequencies of the DUT. However, without establishing frequency locking between the DUT and the PNA-X, the measurement receivers will not be tuned to the actual output frequencies of the DUT unless extra measurements are made to characterize and compensate for LO-frequency offsets. The frequency offset for most embedded-LO converters ranges from a few kHz to several tens of kHz.
Still, any offset is enough to cause rapid changes in the measured output phase, obscuring the actual response of the DUT and making it impossible to correctly measure group delay. The solution for this class of devices is the S93084B Embedded-LO Application. This software application performs background sweeps to measure the DUT’s output-frequency offset compared to the nominal mixing plan (within a defined frequency tolerance).
The measured offset is then added to the nominal LO value in the mixing plan. This modifies the output frequency of the mixing plan, which then shifts the tuning of the PNA-X receivers to the correct DUT output frequency. From then on, the SMC+Phase measurements are performed as described in part one of this article (September 2023 issue of Microwave Product Digest). The embedded-LO feature works on DUTs with multiple mixing stages and also works in other converter measurement classes such as gain compression, intermodulation distortion, and noise figure.
LO Offset Determination
The algorithm used to determine the frequency offset of the DUT’s internal LO(s) is split into two parts. For the first part, a CW signal within the measurement span (default is center) is applied to the input of the DUT, and the DUT output is measured with a broadband receiver sweep over a user-selectable frequency span (default is 3 MHz), centered at the nominal output frequency based on the mixing plan. A peak search is done to find the output signal frequency, and the difference between the peak and nominal output frequency is used as the coarse value of the LO offset. However, this process does not provide the necessary frequency resolution for a successful SMC+Phase phase or delay measurement.
The second part of the LO-offset determination is called the precise sweep. The same CW signal is applied to the DUT input, but instead of performing a broadband frequency sweep, the receiver measuring the DUT output is tuned to the output frequency determined from the broadband sweep and a phase-versus time sweep is taken. The slope of the trace is calculated to determine the fine LO frequency offset, which is then combined with the coarse value to create a new estimate of the overall LO frequency offset and a new estimate of the output frequency.
The sweep is repeated as necessary until the estimated output frequency is within a defined tolerance value (default is 1 Hz) or the maximum number of iterations is reached. The combination of coarse and precise sweeps to determine the actual LO value is fast while providing the necessary frequency resolution for successful phase and delay measurements. The mixing-plan graphical-user interface always shows the nominal frequency values, but the measured LO-offset frequency is displayed in the embedded-LO setup dialog box and on the lower left of the display, which is updated by default for each sweep.
For accurate measurements, the phase noise of both the PNA-X and the LOs in the DUT must be sufficiently low so that errors caused by non-ratioed receiver-phase measurements are not excessive. The phase-noise performance of the DDS sources in the PNA-X is quite good, allowing wide-IF bandwidths for speed, with minimal or no sweep averaging. For older PNA-Xs with fractional-N-based sources, an IF bandwidth between 10 and 30 kHz is recommended, with at least 10 sweep averages.
Figure 5 shows the effect of LO phase noise on group delay measurements. The top trace is a reference measurement using port 3 of the PNA-X for the LO signal. The other traces used an external signal generator with a phase-noise impairment feature, which allows the level of the phase-noise pedestal to be set (in dBc/Hz) over a defined offset range, 1 kHz to 30 kHz in this example. The added noise on the delay trace starts to become noticeable at about -80 dBc/Hz.
Two-tone intermodulation distortion (IMD) is the most widely used measurement of in-band distortion for microwave devices. Two closely spaced signals are applied at power levels that cause the DUT to behave nonlinearly, which creates higher-order mixing products on either side of the two primary signals. As the PNA-X has two internal filtered RF sources and a signal combiner, it can create two-tone IMD stimuli.
The swept-IMD class is the primary measurement tool, providing swept measurements of center frequency, tone spacing, tone power, or LO power. The application measures the primary signals and IMD products of order 2, 3, 5, 7, or 9. IMD parameters can be displayed as absolute power, relative to the main-signal power (dBc) or as input- or output-referred intercept points.
LO Power Sweeps
LO power sweeps are an easy way for converter designers to find the optimum LO power that balances distortion performance with power consumption. Too little LO power starves the mixer and increases IMD. An LO with more power than needed for proper mixing adds needless components and wastes power, which is undesirable for devices with a limited energy supply.
The test is difficult and time-consuming with the traditional approach using external signal generators and a spectrum analyzer, but relatively easy with a PNA-X. Figure 6 shows an example of sweeping the power of an external LO signal while measuring the converter’s gain and IMD performance. Marker 1 has been placed in a reasonable spot where the mixer operates away from starvation (i.e., in the region where the tone gain is constant). Doubling the LO power from 10 to 13 dBm (a 3 dB increase) only results in about a 0.36 dB increase in the input-referred intercept point IIP3 (from -14.87 dBm to -14.51 dBm), which is probably not a worthwhile trade-off.
IM Spectrum Converters
The IMD application S93087B includes a measurement class that displays the frequency spectrum of the IMD signals. This tool provides a quick way to check for IMD imbalance between lower and upper sideband pairs and to look for higher-order products. The marker feature “Marker -> IM Spectrum” opens an IM Spectrum channel where the stimulus is set to the same frequency and power values corresponding to the marker’s position in the Swept-IMD channel. Figure 7 shows an example of the spectrum display after using Marker -> Spectrum and adding markers. Markers 2, 3, 4, and 5 show IMD products of order 3, 5, 7, and 9, respectively. For example, marker 3 shows the amplitude of the fifth-order product generated as 3 * 509.5 MHz – 2 * 510.5 MHz = 507.5 MHz.
Single- and Double-Sideband Converters
Double-sideband (DSB) downconverters typically provide the most broadband frequency coverage, but the lack of a filter in front of the mixer means they have more downconverted noise than the equivalent SSB converter. Without the filter, input noise (typically from a front-end LNA) enters the mixer, and conversion occurs at sidebands above and below the LO frequency, resulting in both sidebands getting mixed to the IF output of the converter. The noise contributions of the two sidebands may not be equal, as it depends on the frequency response of the converter’s front-end.
If the front end is flat between the upper and lower sideband, the DSB converter has 3 dB more noise than the filtered SSB equivalent. If the response is not flat, the difference can be larger if the desired response is around the sideband with lower noise conversion. When using the Y-factor method, the measured noise figure would be the same for both DUTs since a ratio of noise-power measurements is made, and the excess noise of the DSB converter (relative to the SSB converter) is ratioed out.
For most DSB converters, Y-factor-based noise figure measurements typically read between 0 and 4 dB better (lower) than the actual value. The PNA-X noise figure application S93029B, typically used in conjunction with an internal low-noise receiver (Option 029), uses the cold-source noise figure measurement method. As noise power is measured only once, there is no ratio effect, and the DSB and SSB converters will measure differently, and each will have the correct value for noise figure.
LO Noise Contribution
A special consideration when measuring the noise figure of downconverters concerns broadband noise on the LO signal. Noise on either side of the LO offset by the IF can mix within the first mixing stage of the DUT to add noise to the converted output. For example, consider a satellite communications downconverter with an IF of 140 MHz. LO noise offset from the LO frequency by +/- 140 MHz converts to the IF and adds to the noise coming from the converter’s front end. Depending on the mixer’s conversion sensitivity to LO variations, the relative level of the LO noise, and the noise contributed by the converter’s input path, the effect of LO-noise contribution varies from negligible to significant.
The more gain present in the input path (and therefore more noise going into the first mixer), the less that LO noise affects the overall NF. If the LO is embedded within the DUT, the test system does not influence its noise contribution. However, for test systems where the LO is provided externally, LO-noise contribution can be a significant source of measurement error.
This effect can be seen in Figure 8, which shows measurement results from a downconverter with a 1.5 GHz LO provided by the PNA-X and an output of 321.4 MHz. The DUT consisted of a mixer and an output filter with results shown in the lower-left window. The orange trace (Tr 8-mem) shows the noise figure without LO filtering. The red trace is the noise figure with a 1.5 GHZ bandpass filter added to the LO signal. The delta-marker shows the difference in noise figure between the two conditions as 13.1 dB, a considerable difference. In the upper-left window, an input amplifier was added with 15.3 dB of gain, reducing the difference between a filtered and unfiltered LO to 1 dB. In the upper-right window, the gain of the input amplifier was increased by 20 dB to 35.4 dB, reducing the difference between a filtered and unfiltered LO to just 0.1 dB.
Characterizing harmonic and non-harmonic spurious signals from the ports of a frequency converter is a critical piece of the test suite, as these signals can cause in-band and out-of-band interference during operation. Despite the liberal use of filters within a typical converter, spurious signals generated by internal mixers, amplifiers, and frequency synthesizers can leak out of any RF ports. This is especially problematic for devices with a direct antenna port, as these spurs would then be broadcast to the operating environment. Spurious testing using legacy methods with standalone sources and spectrum analyzers typically takes up the most test time. Using the S93090xB Spectrum Analysis application, any of the test receivers within the PNA-X can be used as a spectrum analyzer, allowing spurious tests at all the converter’s RF ports (Figure 9).
Within the spectrum analysis (SA) measurement class, any of the internal RF sources can be set to the desired frequency and power to provide the necessary RF, IF, and LO stimuli during spurious tests. Using VNA error-correction methods, calibrated SA measurements can be made at any desired reference plane, whether coaxial, waveguide, or on-wafer.
Measurements are very fast (often ten to hundreds of times faster than conventional swept-LO benchtop spectrum analyzers) because the PNA-X does not have preselector filters in front of its receivers and because the spectrum analysis application uses stepped Fast Fourier-Transform (FFT) sweeps with optimized data processing.
As there are no filters ahead of the mixers in the PNA-X receivers, the display would show image signals if image-rejection methods were not employed. Images occur in swept measurements when an input signal is mixed into the receiver’s IF when the LO is below and above the signal, creating both a desired and a false response.
To overcome this problem, a software preselection algorithm borrows from a method employed with unpreselected microwave spectrum analyzers for many decades, often called “signal identification” or sig-ID. In its simplest form, two sweeps are taken, with one sweep using low-side mixing and the other high-side mixing. Real signals will remain in the same position on the display, while false signals will move to a different position.
The image signal is not displayed by using the smallest value of the two sweeps for any trace point. However, in dense signal environments, two LO frequencies may not be enough to eliminate image responses. The spectrum analysis application employed in the PNA-X defaults to four LO acquisitions for each displayed data point. The user can decrease it to one LO acquisition when measurement speed is most important or increase it to six or eight acquisitions when better image rejection is desired. For even better rejection, the LO frequencies are randomized by default to make it even less likely that an erroneous signal will be displayed.
The digital filters used for S-parameter measurements are optimized for speed and don’t require high selectivity, as stimulus/response testing always places the measured signal in the center of the IF bandwidth. However, high selectivity is very important for spectrum analysis to distinguish between closely spaced signals. This requires Gaussian-shaped digital filters, which typically require more filter taps than an equivalent-bandwidth filter used for network analysis.
One essential requirement of frequency-converter design is knowing the level of spurious signals coming from the internal mixers. This information is necessary to design filters with sufficient stopband rejection to meet the overall spurious-signal specifications of the converter. All mixers generate signals at m*FRF ± n*FLO where m and n are integers. Usually, only one set of m and n are used for the desired mixing product (e.g., m = n = 1, or m =1 and n = 3), while the other sets represent spurious mixing products of various orders. While the spectrum analysis application can be used to characterize unfiltered mixers, a better choice (especially for swept measurements) is the S93089B Differential and I/Q Devices application, which is suitable for single-ended and differential mixers.
For a mixer used as a downconverter, the sum spurious products (m*FRF + nFLO) are often far out of the band and easily filtered, but some of some of the difference products (m*FRF – nFLO) often fall close to the desired band of operation. Figure 10 shows the frequency range definitions used to test a single-ended downconverting mixer where the RF and LO signals are swept, and only the difference products are measured. Frequency ranges were first defined for the RF and LO signals, then for the LO’s second and third harmonics using multipliers.
With this core set of frequency ranges, it is then easy to use the multiplier and offset fields to define additional frequency ranges for the desired signal and for the spurious mixing products, where, in this example, m and n are 1, 2, and 3. An example of the frequency-setup dialog box is shown for F13. This approach makes it easy to add ranges for higher-order products beyond those shown in the example or for including the sum products (checking the “Up” box will add the coupled and offset frequencies).
Looking at the frequency range F12 (which represents 2*FRF – 3*FLO), it is clear from the start and stop frequencies that a spur will cross the desired IF of 1.5 GHz during the sweep (this will occur when the RF is at 6 GHz and the LO is at 4.5 GHz: 2*6 GHz – 3*4.5 GHz = -1.5 GHz). Although the frequency range shows mathematically correct negative values, the actual signals will show up as positive frequencies.
The results of the measurement are shown in Figure 11. All the mixing products are relative to the input signal power. The marker on Tr 3 shows the mixer’s conversion loss for the desired RF – LO product as -6.0 dB. The marker on Tr 10 shows the crossing spur at -1.5 GHz as -62 dB, which is 56 dB below the desired IF product. Since spurious products are often a function of input power, this set of measurements can be repeated in the same channel or duplicated in other channels with different input power levels. The user can set the frequency axis for each parameter so all products can share a common frequency axis if desired.
This article has shown that the flexible hardware of the PNA-X combined with many software measurement applications enable a broad range of frequency-converter measurements used to characterize linear and nonlinear behavior, all with a single set of connections to the DUT. The article is a condensed version of Keysight’s PNA-X Application Note 1408-23, which can be downloaded from the Keysight website.