by David Ballo, Senior RF Application Development Engineer, Keysight Technologies
Part 1 of 2. See our October 2023 issue for part 2.
Frequency-translating devices translate a band of input frequencies to a different band, one that is either higher (upconverted) or lower (downconverted) in frequency. This crucial function makes them integral to most RF applications, including wireless communications systems and in the transmit and receive portions of radar and electronic warfare systems. This article, the first of a series, describes how a vector network analyzer (VNA) can be used to comprehensively test mixers and frequency converters.
Mixers are the core elements for providing frequency translation. An input signal is mixed or multiplied with a local oscillator (LO) signal that is used to switch diodes or transistors on and off, producing sum and difference products. In most cases, only one of these products is desired, and the other is suppressed by filtering. Mixers can be passive or active devices (Figure 1), and depending on the application, the input signal is provided to the IF port for upconversion or the RF port for downconversion.
While the frequency-translating or mixing process is inherently nonlinear, mixers are otherwise expected to behave linearly. For example, a 1 dB input signal change should result in a 1 dB output-signal change, and a mixer’s magnitude and phase responses versus frequency should be independent of input power. However, just like amplifiers, mixers have linear and nonlinear regions of operation, depending on the input drive level.
Frequency converters are more complex assemblies containing one or more mixers, amplifiers, filters, and possibly other signal-conditioning components like attenuators, isolators, limiters, and phase shifters. Each of the mixers within the frequency converter requires an LO signal, which for test purposes is either supplied from the VNA, an external signal generator, or from an internal or embedded oscillator within the device under test (DUT).
Many of the measurements performed on mixers and frequency converters are RF tests commonly performed on amplifiers, such as gain, gain flatness, group delay, gain and phase compression, intermodulation distortion (IMD), and noise figure. So, this series of articles will examine them in detail.
Many frequency converters are still tested on legacy systems consisting of a stack or rack full of RF instruments. Compared to the modern methods covered in this article, these test systems are significantly slower, less accurate, and more difficult and expensive to configure and maintain. The most common approach for testing frequency converters has been to use stand-alone signal generators for RF and LO signals and a spectrum analyzer as a measurement receiver. This approach is conceptually simple, but in practice has many drawbacks:
The test configuration is difficult to set up, as many external signal-conditioning components are needed, such as combiners, filters, and attenuators. It is often difficult to switch between different tests, as components need to be added or removed.
Typically, only transmission-magnitude measurements like conversion gain are performed, requiring VNAs to characterize port matches, deviation from linear phase, and group delay.
Error correction is limited to magnitude-response corrections, so attenuators are often required to reduce mismatch errors at the expense of signal-to-noise ratio (SNR).
Automated, swept-frequency or swept-power measurements require a computer and software to synchronize the various instruments, resulting in sweeps that are many times slower than those of a VNA.
A modern VNA like Keysight’s PNA-X is well suited for testing frequency converters because of its flexible hardware coupled with many software measurement applications. This combination means that many different tests can be performed while maintaining a single set of connections to the DUT. These tests include forward and reverse linear characterization with magnitude and phase and nonlinear tests of compression, distortion, and noise figure.
Since most of the needed hardware is contained entirely within the instrument (Figure 2), speed can be optimized, resulting in throughput improvements of up to 100 times compared to typical legacy systems. Advanced error correction methods for highly accurate results have been developed for all measurement applications. For example, mismatch errors are removed for swept gain, delay, and noise figure measurements, and power meters are used to calibrate the test system to get high accuracy for both setting and measuring absolute power, which is essential for compression and IMD measurements. Much effort has been put into designing simple, intuitive user interfaces and guided calibrations to make a complex instrument relatively simple.
Challenges of Converter Testing
When characterizing transmission parameters of frequency-translating devices, standard S-parameter methodology cannot be applied. S-parameters are defined as complex (magnitude/phase) ratios of incident (an) and test signals (bn). When the stimulus and response frequencies are the same, which occurs for non-frequency-translating devices, complex ratios can be directly measured as receiver ratios like b2/a1, which is the core measurement of S21, or b1/a1, which is the core measurement of S11.
Data acquisition for all the receivers is simultaneous, so a single sweep in each measurement direction (forward and reverse) is sufficient to gather the necessary data for a two-port set of error-corrected S parameters. Testing port matches of mixers and frequency converters is similarly easy since the incident and test signals are at the same frequency for reflection measurements, allowing the use of normal S-parameters.
However, the input and output frequencies are not the same for frequency-translating transmission measurements, so direct receiver ratios cannot be used. Two challenges must be overcome when testing transmission parameters of frequency-translating devices: offsetting the stimulus and response frequencies and calculating receiver ratios when the frequencies differ.
With modern VNAs like the PNA-X, the frequency-offset challenge is easily overcome since all the internal RF sources (including the one that provides the LO signal for the measurement receivers) are fully synthesized and can be independentlyset to any frequency within the range of the instrument.
It has not always been this easy, however, as earlier generations of VNAs required the insertion of an external mixer into a phase-locked loop to offset the source and receiver LO frequencies. Due to the abundance of spurious signals with this arrangement, it was difficult to achieve phase lock over a broad frequency range, and spurious signals often showed up in the measurement results.
The second challenge involving receiver ratios is also not difficult to overcome for transmission magnitude measurements, as direct receiver ratios are unnecessary. At the expense of measurement speed, input power can be measured in one sweep, output power in another, and the individual receiver measurements can be ratioed. In receiver terms, the magnitude-conversion response |SC21| = |b2|/|a1|, where |a1|and |b2|are measured with two sequential sweeps. This measurement is very accurate because the PNA-X performs match-corrected power measurements that are calibrated using a power meter and an S-parameter calibration kit or ECal electronic calibration module.
Transmission-phase measurements are particularly challenging for frequency-translating devices because phase is generally measured between signals at the same frequency. With earlier VNAs, it was necessary to use a reference mixer whenever phase or group delay measurements were required. The reference mixer was placed in series with the main signal path (often called the down/up- or up/down-converting method) or in a parallel path to the DUT (Figure 3). Each approach had pros and cons, but either way, the reference mixer had to cover the frequency plan of the DUT, and it required its own LO signal. A separate calibration mixer was often used to calibrate the test system.
Modern VNAs eliminate the need for a reference mixer for most frequency-converting measurements of phase and group delay by using modern synthesizer hardware that supports phase-coherent frequency sweeps, which earlier generation VNAs could not do. With phase coherence, the phase response versus frequency of a single receiver is repeatable, allowing the phase responses of different receivers measured over different frequency ranges to be ratioed, similar to the previously discussed ratio of magnitude responses. This capability eliminates the need for a reference mixer, greatly simplifying test setups.
Core Measurements Using SMC, SMC+Phase
The Scalar Mixer/Converter (SMC, S93082B) and Scalar Mixer/Converter Plus Phase (SMC+Phase, S93083B – a superset of S93082B) are the core measurement classes for mixer and frequency-converter characterization using a PNA-X. A reference mixer is not needed and users can perform frequency and power sweeps, sweeps to measure output power, gain, gain flatness, gain compression, phase deviation, phase-versus-drive (AM-to-PM conversion), group delay, and port matches. Examples of these measurements discussed below use the down-converting DUT shown in Figure 4.
A Swept-IF Example
In a typical frequency converter, the operating frequency ranges of the ports are quite different. The RF port (the input port of a down-converter or the output port of an up-converter) generally covers a much broader frequency range than the IF port, which is typically band-limited by a filter. Swept-IF measurements primarily show the IF-bandwidth response versus frequency and are commonly achieved in single-stage converters by sweeping the input port while the LO signal is fixed at a single frequency (as it often is during operation) and measuring the swept response at the output.
Figure 5 shows the user-specified mixing plan and test results of such a swept-IF measurement of the example down-converter. From this measurement, which is a combination of the converter’s input and output frequency responses, narrowband gain and gain flatness can be observed, as well as absolute output power. Swept LO measurements can be configured to isolate just the input or output response, as shown in Figure 6.
Figure 7 shows the mixing plan and results of measuring the same down-converting DUT as above but with simultaneous sweeps of the RF input and LO signal provided to the DUT. With this setup, the output frequency is fixed and, in this example, positioned in the center of the IF filter response. The results show the combined frequency response of the DUT’s input amplifier and mixer unaffected by the frequency response of the IF filter. Broadband gain and gain flatness are observed with this measurement.
SMC+Phase can easily configure single- and dual-stage converter setups. Dual-stage converters have two mixing stages and typically two IF filters—one at the input (for up-converters) or output (for downconverters) and a second filter in between the two mixing stages. Figure 8 shows the user interface for configuring a dual-stage setup where both LOs are supplied externally, and Figure 9 shows which converter sections are measured as the result of various combinations of fixed and swept input and LO signals.
Measuring Converters with More than Two Mixing Stages
SMC+Phase can be used for converters with more than two conversion stages, with the limitation that only one LO can be controlled. While the user interface only shows one-and two-stage mixing plans, as long as the input, controlled LO, and output frequencies along with the direction of sweeps are consistent with the actual DUT’s mixing plan, the internal frequencies in the user interface of each conversion stage are not important.
Creating the proper mixing plan in the user interface is achieved by calculating the frequency of the uncontrolled LO, which is a mathematical combination of the actual DUT LOs that are not defined or controlled. Care must be taken to select the proper high-side or low-side mixing condition so that when the PNA sweeps up, the receivers sweep in a direction that matches the output of the DUT.
Figure 10 shows the actual DUT mixing plan and the calculated two-stage mixing plan for an example three-stageconverter with one controlled LO at 9 GHz (LO1DUT). LO2 in the PNA’s mixing plan equals LO2DUT – LO3DUT (5 GHz – 1.1 GHz = 3.9 GHz). As can be seen, the input and output frequencies of the three-stage DUT and the PNA-X’s two-stage mixing plan are the same.
Converters with Internal Multipliers or Dividers
Some converters have built-in multipliers or dividers in the main or LO paths, which become part of the overall mixing plan. The example in Figure 11 shows how an LO multiplier can be considered.
The ability to enter multipliers in the user interface can also be used to test frequency multipliers. With a single-stage mixing plan, setting the numerator at Port 1 to the desired harmonic number N and setting LO1 to zero results in an output frequency that is N times the input frequency.
Phase and Delay Measurements
To enable measurements of phase and group delay, the checkbox for Enable Phase on the Sweep tab of the SMC Setup dialog box must be checked. For instruments with fractional-N-based sources (synthesizer version 6), the starting phase of the sources cannot be controlled, so phase traces are normalized to zero at one point in the phase response (user-selectable). This eliminates the effect of a random starting phase for each sweep, giving stable, normalized phase traces. For this normalization, the user should select a measurement point that has a good SNR, typically in the middle of the band of interest.
Phase normalization does not impact deviation-from-linear-phase measurements or group delay measurements since absolute phase is not a factor when calculating the slope of the phase-versus-frequency response. However, for these instruments, phase normalization means it is not possible to use SMC+Phase to compare phase offsets between multiple paths, multiple DUTs, or phase changes within a DUT. Figure 12 shows fixed-LO measurements of conversion gain and group delay through the example one-stage converter. The trace-analysis feature has been used to show the delay statistics over a 200 MHz span centered in the middle of the passband.
For instruments with direct-digital-synthesis (DDS)-based sources (synthesizer version 7), an additional choice is available called “Use Absolute Phase.” With this choice, the internal sources and internal LO start up with repeatable phase at at every point, giving stable and repeatable phase measurements sweep-to-sweep without the need for phase normalization.
This feature allows SMC+Phase to make measurements that previously required the Vector Mixer/Converter (VMC) measurement class and the associated reference and calibration mixers. One example is measuring phase changes that occur within a device because of an internal phase shifter in a transmit/receive module. Another example is measuring the absolute phase variation among various frequency converters or mixers due to internal path-length differences.
As the phase of the LO contributes to the converter’s absolute output phase, any difference in LO path lengths among multiple test stations will influence the test results. This issue can be overcome using the proper phase calibration method, which will be described in the calibration section of part 2 of this series.
The “Use Absolute Phase” choice can only be selected when LO1 (and LO2 if used) is set to an internal source that isn’t being used for the input stimulus. The LO can come from a test port or the Option XSB rear panel source (Source 3). The “Use Absolute Phase” choice is grayed out if either LO is set to “Not controlled” or to an external signal generator. Absolute phase does not work for converters with embedded LOs, since the phase of the DUT’s LO would not be synchronous with the sources within the PNA-X, even if they share a common frequency reference.
Deviation from Linear Phase
When measuring a device’s phase response, it is generally not useful to display wrapped phase without electrical-delay compensation, as shown in Tr 1 of Figure 13. Measurements of deviation-from-linear phase are accomplished by removing the linear, negative slope from the phase-versus-frequency response, leaving the phase deviations.
This process can be accomplished in one of two ways. The traditional approach is to place a marker in the center of the response and use the Marker → Delay feature, the results of which are shown in Tr 2. The firmware calculates an electrical delay value corresponding to the linear portion of the phase response and then mathematically removes that amount of electrical delay from the measured results. This effectively flattens the phase trace around the active marker.
With this approach, the average response is not centered around zero, so the user must either calculate the phase-deviation values with delta markers using the trace-statistics feature (with a user-defined span if desired) or with external software. The second approach is shown for Tr 3, where the Trace Deviation feature is used with the Linear choice. With this approach, the response is centered around zero degrees, so the phase deviation values can be directly read from the trace.
Figure 14 shows the results of measuring the deviation-from-linear phase of a converter with a 4.4 GHz LO and two different LO path lengths, created in this example by adding an adapter with 94 ps of delay, shown in the bottom plot (Ch3). Both Ch1 and Ch2 have a fixed reference marker set for the case of the shorter LO path. In Ch1, the phase is normalized to the middle point, and after the adapter is added to create a longer LO path, the delta marker shows that the difference in phase between the two LO path conditions cannot be measured.
In Ch2, the “Use Absolute Phase” method was chosen, and in this case, the phase shift due to the additional LO-path length is directly measured with the delta marker as 151.8 degrees. The additional LO phase due to the adapter can be calculated as 94 ps x 4.4 GHz x 360 degrees = 149 degrees. This is very close to the measured value, with the small difference likely due to small match changes between the two LO path conditions.
The next article in this series will examine power sweeps for gain and gain compression, phase transfer, segment sweeps, calibration, and other key elements of the mixer and frequency-converter test process.