Differential or balanced measurements performed with a vector network analyzer (VNA) are typically done with a "virtual" approach. The device under test (DUT) is stimulated with unbalanced signals as in the manner of testing single-ended devices, and the VNA mathematically transforms these unbalanced wave quantities into balanced S-parameter results.¹ The approach is adequate for evaluating active or passive DUTs under small-signal (linear) conditions. But the technique falls short when testing active DUTs under large-signal or nonlinear conditions, yielding less-than-accurate results. Although numerous attempts have been made to resolve this problem, involving an "ideal" broadband hybrid coupler^2,3 or power divider,^4,5 these methods do not allow for full system error correction. Fortunately, an off-the-shelf solution exists in the form of an option for the four-port ZVA and ZVT VNAs from Rohde & Schwarz (Columbia, MD) that is inherently broadband, precise, and simple to apply.

The R&S ZVx-K6 option is the result of a fundamentally new approach and has resulted in several patents along the way. The company has evaluated the technique on a variety of active devices from various manufacturers, and the measurement have shown that gain compression occurs at either higher or lower output levels than measurements produced by the "virtual" technique. A typical example is shown in Fig. 1, in which the two techniques are shown together on the display of an R&S ZVA40 analyzer when measuring a 2-GHz monolithic-microwave-integrated-circuit (MMIC) amplifier. The results are identical in the linear small-signal region. However, the results from the two techniques show serious deviations for a device under test (DUT) under conditions of when gain compression. Using the virtual differential technique, gain compression starts about 4 dB sooner than under true differential stimulus, with peak gain that is about 0.5 dB less.

With this in mind, the potential ramifications (and benefits) of this new technique are threefold:

1. Most RF amplifiers used in mobile telephones, smart-phones, data access cards, and many other portable applications are balanced devices, and the use of balanced, differential amplifiers is increasing in portable wireless and mobile electronic products. Balanced RF power devices and amplifiers have been almost universally measured using the virtual differential mode, since there has been no viable alternative. Consequently, a large number of these products have been characterized incorrectly.
2. If gain compression occurs at drive levels lower than specified by their manufacturers (who thus far had only the virtual technique with which to evaluate them), amplifiers will produce unacceptable levels of intermodulation products under conditions that were previously thought not to produce them.
3. Handset manufacturers that have built systems and tested them using the virtual differential method may need (or have already had to resort to) "backing off" drive levels to achieve linear performance over the expected operating range based on their performance in service.

However, backoff conditions can result in a greater number of active devices required to achieve a specific output level, which can require a redesign of the entire transmit section.

Nevertheless, going forward, the ability to more accurately measure active device performance will allow manufacturers of components, subsystems, and systems to create products that achieve excellent results under their anticipated operating conditions before they leave the factory—not once they have failed in the field.

When two-port (balanced) components are characterized with a VNA, the instrument mathematically calculates balanced (single-ended) S-parameter results from the single-ended stimuli signals to obtain differential S-parameters. The DUT is not excited with differential signals but is measured as an unbalanced device. The resulting mixed-mode S-parameters are then calculated using unbalanced or single-ended S-parameters. This virtual differential approach is subject to inaccuracies since the DUT is not tested with the type of signals that are present in the actual application, such as a cellular handset. The approach is valid under linear, small-signal conditions, but does not accurately represent an application's operating environment under nonlinear, large-signal conditions.

When active devices are driven harder and their behavior becomes nonlinear (usually specified at 1-dB and 3-dB gain compression points), the technique has long been suspected of producing optimistic results. That is, an amplifier (for example) might well reach its 1-dB compression point at a drive level lower than previously measured and thus be rated at too high a power level. However, the "virtual" technique has been the only one offered by network-analyzer manufacturers for making these measurements because (among other challenges) it is extremely difficult to control the magnitude and phase of the two signal sources.

The technique developed by Rohde & Schwarz allows nonlinear RF and microwave balanced components such as amplifiers to be stimulated with actual differential signals for the first time to frequencies as high as 40 GHz (depending on the analyzer). The technique is based on vector-corrected wave quantities and uses a patented technique to control magnitude and phase of the two internal sources. The sources in the R&S ZVA and R&S ZVT generate two signals identical in magnitude and phase-shifted 0 or 180 deg., with phase uncertainty below 1 deg. This differential-mode signal then stimulates the device, and mixed-mode S-parameters are then directly calculated from the ratios of the error-corrected differential or common-mode wave quantities.

With the virtual differential measurement method, at every frequency point, a signal is applied at port 1 (now a single-ended port), all transmitted signals (wave quantities) to port 2, 3, and 4 are measured, and the reflected signal at port 1 is measured. The signal is applied to port 2 and the above measurements are repeated. The process is repeated for ports 3 and 4. The 16 single-ended Sparameters (S11 to S44) are calculated and then used to calculate the mixed-mode Sxxyy-parameters. However, for nonlinear devices, the instrument's single-ended port 1 and single-ended port 2 are never simultaneously stimulated, so the device is never driven the way it would be in actual operation.