Having More VNA Ports Improves Throughput

Multiport VNAs reduce wear on connectors and save time.

Integrated circuits found in today�s consumer products often contain multiple RF ports because there may be several different radios on the same chip. When multiple frequency bands are combined with multiple-input-multiple-output (MIMO) technology as used in IEEE 802.11n, it�s possible to have up to 10 RF ports on a device.

Differential-to-differential, single-ended-to-differential, and differential-to-single-ended modes are available as well as the usual single-ended-to-single-ended configuration. These capabilities are valuable, but they do require a greater number of VNA ports.

In addition, the constant drive to reduce the cost of test means that there�s a need to test more chips in parallel. One way to do that is to have more ports in the test equipment.

In a related development, very high-speed copper backplanes require simultaneous testing across multiple channels. Even though the channels are intended to be independent, there is crosstalk among them that can be measured efficiently by multiport RF instruments.

For all these reasons, several manufacturers provide VNAs with multiport capabilities. Of course, a number of multiport implementations are possible, each with its own advantages and disadvantages.

Multiport Implementations
The lowest-cost approach is to expand the number of ports by adding switches to a two-port VNA. This method typically provides only one source and does not measure more than two ports simultaneously. It may be a very good choice for production testing of RF devices, for example.

Agilent Technologies introduced the E5091-016 option for use up to 8.5 GHz with four-port ENA Series VNAs. This 16-port testset has the switching done after the directional couplers associated with the basic VNA ports, as shown in Figure 1. In addition, nine connections to three independent SPDT switches increase the flexibility of the option when used in custom test setups.

Straightforward switching provides a very cost-effective, high-volume test platform for components used in multiband cell phones and wireless systems. But, for stable microwave-frequency multiport measurements, a directional coupler must be supplied at each port. Switching after the directional coupler causes some instability, which becomes unacceptable in these higher-frequency applications.

Agilent�s Joel Dunsmore, a senior R&D engineer/scientist, described how PNA Series instruments and multiport test systems interface. �All of the testsets used with the PNA Series have their capabilities described in a special testset file. This file, loaded by the firmware, contains specific information such as number of ports, port control, and switching configurations. Interfacing in this way means that additional testsets can be developed and controlled by providing new testset files.�

The modular Rohde & Schwarz Model ZVT8 VNA is available with as few as two ports or as many as eight. Each test port has separate reference and measurement receivers for measurements to 8 GHz. Each pair of ports includes a separate source. The sources and the receivers are externally accessible in ZVT8s with the direct generator/receiver access option shown in Figure 2.

Improved throughput is one advantage claimed for the ZVT8 because multiple sources support truly simultaneous multi-DUT testing. For example, two four-port DUTs or four two-port DUTs can be tested simultaneously. In addition, multiple sources enable intermodulation (IMD) tests as well as tests on mixers. A five-port ZVT8 with three sources can perform IMD tests on a mixer and simultaneously provide the LO signal.

According to Rich Pieciak, a Rohde & Schwarz product manager, dedicated per-port couplers make a big difference in performance. �Higher accuracy and greater speed result from this architecture. In approaches that provide additional switched ports after the base unit�s directional couplers, the switch attenuation reduces the dynamic range that leads to lower throughput. Attenuation also decreases measurement accuracy because directivity is affected.�

A further option, the ZVT-B16, makes inputs to the reference and measurement receivers from the eighth port available. This feature enables the performance of DUTs with one input and eight outputs to be measured in a single test.

Anritsu also provides multiport testsets. As explained by Dr. Jon Martens, staff scientist, �For up to 18-port, 9-GHz applications, we design a switch matrix in front of a four-port engine. This approach is cost-effective yet provides >90-dB isolation and relatively low insertion loss. At very high frequencies such as 40 and 65 GHz,� he continued, �we use a hybrid approach with receiving couplers at every port (Figure 3). This keeps insertion loss very low, which is necessary to maintain dynamic range, and raw port performance very high while allowing almost full calibration flexibility.�

Calibration
A wide range of options available for VNAs supports testing different types of DUTs. For example, multiple sources are required to develop IMD test signals and test frequency-translating devices such as mixers. Some instruments have higher power outputs as well as pulsed source capability. Pulsed drive capability is particularly important if a device can only be used in a pulsed mode, for example, because of power dissipation considerations.

Regardless of test-setup details, an accurate calibration must be performed. In general, all elements of the test setup should be accounted for, the DUT being the only thing altered in the calibration process. Typically, forward and reverse measurements are taken with known impedances instead of the DUT.

Many different algorithms have been developed, such as short-open-load-through (SOLT), through-reflect-line (TRL), and through-reflect-match (TRM). There are others as well, but all calibration schemes accumulate sufficient measurements that the effects of internal delays, mismatches, and various connection and switching losses can be compensated for.

The reasons to choose one scheme instead of another involve the availability of practical standards as well as the level of accuracy required. For example, a perfect open is difficult to manufacture for use at very high frequencies. And, some types of standards are more practical as coaxial elements than as striplines.¹

According to one reference, �Compared to the SOLT cal, the requirements on these [through, reflect, line, match] standards are much easier to meet so they are more easily constructed. During the calibration process, there are more measurements made of the standards than unknown error terms. The calibration math ends up with 10 equations and eight unknowns, so there is redundant information, which can be used to calculate the propagation constant of the line standard, and the reflection coefficient of the reflect standard.�²

Nevertheless, whichever scheme is used, as many as 12 error terms are associated with each pair of ports. Calibration is the process of removing or de-embedding the error terms from the raw measurement values. In addition, there are N² S-parameters to be measured, where N is the number of ports.

It�s easy to see that for N>2, calibration quickly becomes unwieldy if you must manually connect the various standards between all possible combinations of ports. To alleviate this problem, VNAs have largely automated and simplified calibration.

Automation typically is accomplished by providing a convenient calibration device. For example, Agilent offers several two- and four-port ECal modules with a number of different connector styles covering the 300-kHz to 67-GHz range. These modules do not provide perfect shorts, opens, or loads, but they do have highly repeatable impedance states controlled by a VNA�s USB port. The modules use fully traceable and verifiable electronic impedance standards and, in this sense, actually are transfer standards.

Calibration beyond two ports is problematic given the definition of S-parameters. Although each term in an N�N S-parameter matrix only involves two ports, for example S_ij refers to the ratio of the output on port i when port j is driven, all ports other than i and j must be terminated in the reference impedance.

If a two- or four-port calibration standard is successively inserted between only N sets of ports, as in some schemes, many of the S-parameter matrix terms must be derived: They are not being directly measured. In addition, depending on the quality of the terminations on all ports other than i and j, the quality of the derived calibration may be questionable for certain types of devices.

For example, in couplers and dividers, the performance at one port is sensitive to the load on others. This effect is not necessarily as pronounced with diplexers and triplexers because the output ports on these devices are better isolated from each other. However, reflections at outputs can and do affect measurements at the common node.³

Anritsu�s Dr. Martens commented that hybrid combinations of the various algorithms could be useful in a multiport calibration to handle a complex measurement. For example, this situation might arise when only a few good through lines are available. He also made the important point that, in addition to calibration kits, manufacturers offer verification kits.

According to Anritsu, �The�kits contain precision components with characteristics traceable to NIST. Used primarily by the metrology laboratory, these components provide the most dependable means of determining the system accuracy of your VNA. A disk containing factory-measured test data for all components is supplied for comparison with customer-measured data.�

Agilent�s Mr. Dunsmore described some of the calibration techniques recently applied to the company�s PNA Series VNAs. �Calibration types such as unknown through, also known as SOLR where the SOL refers to short-open-load on each port and the R implies a reciprocal through, QSOLT in which the SOL is applied to only one port and a known through is used at every other port, and multiport ECal have greatly reduced the difficulty of calibration. These techniques can be combined to support different calibrations on different ports.

�For example, on an eight-port calibration, it is only necessary to connect the ECal unit between port 1 and each other port�seven connections in total�to complete a full 8�8 calibration,� he continued. �All the other 28 through paths are computed from these seven measurements.

�For an on-wafer application, a multiport calibration can be created by doing a TRL calibration between one pair of ports. Then, using a form of QSOLT, only known through connections are required between one of the TRL ports and every other port. As a result of this reduced connection calibration, the uncertainty on derived paths is greater than that of measured paths. For critical applications, particular or additional through paths can be specified [for additional calibration],� explained Mr. Dunsmore.

Not only is calibration complicated by multiple ports or a requirement for high measurement accuracy, but the actual DUT test fixture also comes into play. The vector part of the VNA name refers to the instrument�s combined magnitude and phase capabilities. Because phase is changed by line length, the physical connections to the DUT need to be accounted for in establishing the reference plane, the physical location at which measurements will be correctly calibrated.

For Agilent�s PNA Series, the automatic port-extensions feature measures the loss and delay of test fixtures so that these effects can be removed from subsequent measurements. Anritsu VNAs have automatic reference delay compensation that eliminates the linear phase difference resulting from a simple reference/measurement-path length difference. Using this capability allows only the residual phase characteristics of the DUT to be observed�the signal of interest.

Summary
As the complexity of communications systems increases, ICs have become evermore complicated as they address multiple standards as well as schemes such as MIMO to improve performance. The result is a need for multiport RF measurements both in design and test. VNAs combined with suitable testsets fill this need, and for many applications, newly developed forms of calibration make the instruments� use relatively straightforward.

Nevertheless, a VNA is a precision instrument, and its use demands attention to detail. This is particularly true if applications involving pulsed measurements or differential signals are undertaken.

For example, a repetitive fast RF pulse has a spectrum that may exceed the analyzer�s intermediate frequency (IF) bandwidth. This means that only part of the spectrum, such as the central line, can be measured. The implication is that S-parameters can be derived using this data only if the instrument calibration has been done in a similar way.⁴

For DUTs with differential inputs/outputs, two single-ended ports can be used to simulate a differential port if the signals remain within the DUT�s linear region of operation. However, many devices cannot be assumed to behave in the same way for both large and small signals, for example. In these cases, mode conversion may be taking place in which a portion of the differential signal appears as common mode and vice versa.

To correctly measure the performance of these kinds of DUTs, true differential signals must be used. This means that the two signals forming a differential input, for example, must be generated with 180� phase difference and equal amplitudes over the entire frequency band of interest. Conventional VNAs don�t operate in this way.

One solution to the problem has been described in a recent Agilent white paper Advanced Measurements and Modeling of Differential Devices. By controlling two specially modified signal generators with a third generator, a truly differential output was achieved. Traditionally, baluns have been used to overcome differential signal test problems, but baluns do not allow injection of common-mode signals, nor do they support evaluation of mode-conversion problems. The paper is particularly valuable because of the detailed discussion it presents relating DUT output anomalies to their likely causes.

Multiport VNAs are not unique in the types of calibration and measurement problems they present. They just have more of them because there are so many ports. In addition to calibration standards available from the manufacturers, there are proprietary software applications that guide you through the calibration process.

Because VNA measurements have been heavily relied upon for years, many articles and application notes are available on the manufacturers� websites. Reference 2 provides links to a wide range of VNA-related material. Discussion forums offering more direct help also can be found online.

References
1. Stripline TRL Calibration Fixtures for 10-Gigabit Interconnect Analysis, Agilent Technologies, 2006.
2. VNA Help for Microwave Network Analyzers, http://www.vnahelp.com/tip18.html
3. Three and Four Port S-Parameter Measurements, Anritsu, 2002.
4. Pulsed S-Parameter Measurements, Anritsu, 2003.

October 2006