Eliminate Fixture Effects On Device Measurements
Increased accuracy has been an objective of the earliest vector network analyzer (VNA) measurements. By means of calibration and vectorerror- correction techniques, the accuracy of a VNA can be extended from the instrument ports to the ends of the test cables. When the device under test (DUT) connects directly to the test port cables, the calibration plane and the measurement plane are one and the same. In this case, calibration and error correction are straightforward processes involving mechanical or electronic coaxial calibration standards. For DUTs in pin-mounted or surface-mount packages, however, a test fixture must be used, and now the coaxial-calibration plane and the measurement plane are separated and additional errorcorrection techniques are needed to achieve high measurement accuracy. These methods often use the modeled response of the fixture to effectively move the calibration plane to the terminals of the DUT. Some engineers choose to assume minimal impact of the test fixture, and simply measure the total response of the DUT plus the fixture. This article discusses two model-based corrections that increase measurement accuracy and eliminate the need to ignore measurement errors induced by test fixtures.
Direct measurement involves measuring physical calibration standards and calculating error terms. This method provides accuracy that is primarily based on how precisely characteristics of the calibration standards are known. Much has been written over the years on various direct-measurement calibration techniques. Full details can be found in Agilent Application Notes 1287-3 and 1287-11.1-3 Calibration based on modeling uses mathematical corrections derived from a modeled response of a network. This modeled response could come from simulated results or theoretical behavior, but often is derived from actual measurements. Often, a combination of measurement and modeling helps achieve the highest quality results.
Port extension is the simplest of the modeling techniques shown in Fig. 1. It relies on a simple delay (and, in some cases, a loss) model of the test fixture. De-embedding uses a full S-parameter model of the fixture. Both techniques eliminate the need for building precise, in-fixture calibration standards, which are difficult to realize (especially the load standard), and take a lot of time and effort.
Test fixtures vary widely, depending on application and cost. While test fixtures used in manufacturing are rugged and tend to be expensive, printed-circuit-board (PCB) fixtures are especially common in research and development (R&D) laboratories. They are relatively inexpensive and easy to make, although signal losses cannot be ignored for frequencies beyond 3 GHz. Many of the devices used in today's wireless appliances must be tested at frequencies as high as 13 GHz. It is, therefore, necessary to reduce or eliminate both the loss and delay of the fixture, which allows the true characteristics of the DUT to be analyzed.
So, when measuring devices in fixtures, consider the traces on the PCB as extensions of the coaxial test cables that are between the network analyzer and the DUT. By performing port extensions on each portion of the fixture, the measurement plane is extended beyond the coaxial calibration plane right to the terminals of the DUT. When the loss and electrical length between the fixture connector and the DUT is known, it can be manually subtracted by most VNAs on the market.
Many test setups employ a PCB fixture with SMA connectors (Fig. 2). The fixture/VNA combination can be calibrated at the plane of the SMA connectors. But when the fixture is used to measure board-mounted devices, the electrical characteristics of the PCB fixture can change the DUT's measured amplitude and phase. Port extensions are used to add linear phase (constant delay) and loss-versus- frequency terms to the coaxial error-correction arrays to shift the reference plane to that of the DUT.
When the delay and loss of the fixture are not already known, they must be measured before port extensions can be applied. An automated method has been developed and integrated into the PNA series VNAs from Agilent Technologies. Agilent's Automatic Port Extensions (APE) provides a convenient way to calculate the loss and delay of the fixture using simple open or short measurements. Electrical delay is calculated using a best-fit straight-line model. The loss term is calculated in one of two ways, depending on the media used for the transmission line. The loss model is assumed to be either coaxial or dielectric. Both the coaxial and dielectric models give a variable loss versus frequency that is not a simple straight line. The dielectric model is used when the fixture is built on a PCB.
The APE algorithm measures an open or short and computes the insertion loss and electrical delay of the tested portion of the test fixture. This step is repeated for each portion of the test fixture. After this step, only the fixture mismatch remains as a source of error. The main source of mismatch error is the transition from coax to microstrip that occurs at the connectors of each of the fixture's ports. This mismatch is not removed by the coaxial calibration, since it occurs after the coaxial calibration plane.
Measurement accuracy can be increased by minimizing the reflection at this transition by using good quality edge-mounted connectors and by having good 50-Ohm transmission lines on the test fixture. The port-extension technique gives good results with medium- level accuracy. While not as accurate as using high-quality in-fixture calibration standards, it is by far an easier method for testing components in fixtures, and provides adequate accuracy for many applications.
The APE technique uses a curvefitting process to calculate low-order loss and phase responses. While the algorithm is tolerant of mismatch ripple, it does not remove the ripple itself. In most cases, only one high-reflection standard is needed to accurately calculate the loss and delay responses. Using only one high-reflection standard requires that the frequency span of the measurement be wide enough so that the ripple in the reflection measurement goes through at least one full cycle. In this case, the most convenient standard can be used, which is often an open. Using two standards makes little difference for broadband measurements, as ripples occur with either standard and the calculated loss is the same when using an open or short. Using two standards improves accuracy for narrowband measurements, where a full ripple cycle does not occur. The lower trace in Fig. 3 shows the response of one portion of the test fixture before APE is applied. The upper traces show responses after APE is applied. The loss compensation can either center the error on 0 dB (brown trace) or keep the ripple peaks below 0 dB (blue trace).
Fig. 4 shows responses for a balanced- to-unbalanced 5.5-GHz wireless- local-area-network (WLAN) filter tested to 10 GHz. The delay and loss terms for port one of the test fixture measurement are shown in the Automatic Port Extension tool bar. The values were calculated automatically by the Agilent PNA network analyzer. The two traces below show DUT measurements with and without port extension. Without port extension, the measurement includes the DUT plus the fixture. The distorted response is due to not compensating for phase (especially important for a balanced port), and not compensating for the loss of the transmission lines on the PCB. With the port extensions, significant errors due to the test fixture are removed, providing a considerably more accurate insight into the actual performance of the WLAN filter.
Fixture de-embedding is a more rigorous modeling technique. The process begins by creating a model for the test fixture used with the DUT. The accuracy of the model directly affects the accuracy of the de-embedded measurement. De-embedding is used to remove the undesired effects of fixtures, adapters, and probes. Instead of simply subtracting electrical length and insertion loss, de-embedding uses a modeled response as a function of frequency and mathematically removes the fixture effects from the measurement. Unlike port extensions, deembedding removes the mismatch effects of the coaxial-to-microstrip transitions. The S-parameters of the test fixture circuit are stored in an .s2p file format.
Continued on page 2
Page Title
The easiest way to create the .s2p model of a fixture is to use a measurement probe that can interface with the DUT end of the transmission lines in the fixture (Fig. 5). In this case, the user performs an unknown thru calibration using coaxial standards on one side of the test fixture and a probe impedance-standard substrate (ISS) for the other side of the fixture. The fixture's transmission line is the unknown thru. After the calibration is performed, the fixture is simply measured, without moving the probe or coaxial cable. The measurement procedure is repeated for each section of the fixture, using the same calibration as was done for the first arm. In order for the probe to measure the end of the transmission line, ground pads must be placed on the fixture with the correct spacing to match the pitch of the test probe.
An alternative to probing the fixture is to use a technique where two one-port calibrations are performed. This technique assumes the fixture section is reciprocal (i.e., S21 = S12), which is always the case. The first one-port calibration is done at the end of the coaxial connector, using coaxial standards. The second one-port calibration is done at the spot where the DUT is placed, using in-fixture calibration standards. Both the Agilent PNA and ENA network analyzers provide a macro that extracts the .s2p data of the fixture section using the two sets of one-port-calibration data. While this method has the advantage that a probe is not needed, it does require that in-fixture calibration standards must be fabricated and characterized in order to do a oneport in-fixture calibration.
If direct measurement of the fixture is not practical, then simulations can be performed to determine the S-parameter behavior of the fixture sections. For accurate data based on this technique, a good loss model of the PCB material and accurate trace dimensions are required.
The mathematical flip side of fixture de-embedding is fixture embedding. If a network can be mathematically subtracted from a measurement, it is reasonable that one could just as easily be added to a measurement. VNAs are matched to make S-parameters in a 50-Ohm, singleended environment. When measuring devices that do not fit into this category, further processing of the data is required. Many of these softwarefixture tools are built into the PNA and ENA series vector network analyzers from Agilent Technologies. For non-50-Ohm devices, it is possible to restate the S-parameter data so that it looks like the DUT was measured using a VNA impedance other than 50 Ohms. It is also possible to embed virtual impedance matching circuits, which are often needed for devices such as surface-acoustic-wave (SAW) filters, without having to actually add inductors and capacitors to the test fixture. Mixed-mode (differential-, common-, and cross-mode) S-parameters can be calculated for devices with at least one balanced port. Figure 6 shows some common impedance matching networks built into a VNA for this purpose.
For devices with balanced ports, four-port de-embedding allows simulation of crosstalk between the test ports (Fig. 7). Although crosstalk is insignificant when using coaxial cables, it may be significant when fixtures or probes are used for a measurement. Using two two-port .s2p files will give different measurement results than using a single four-port file, because the crosstalk terms are not included with the two-port files.
Port extensions and de-embedding are important tools that should be added to every engineer's measurement tool kit in order to achieve the most accurate results. The PNA's automatic port extension functionality takes the guesswork out of the setup by guiding the user through the required measurements and automatically applies the results. Whenever practical, fixture de-embedding is recommended to achieve the most accurate measurement results. The PNA's calibration wizard makes the completed calibration steps for probe de-embedding easy with a step-bystep guided process.
REFERENCES
1. Application Note 1287-3, "Applying Error Correction to Network Analyzer Measurements," Literature Number 5965-7709E, Agilent Technologies, www.agilent.com.
2. Application Note 1287-11, "Specifying Calibration Standards and Kits for Agilent Vector Network Analyzers," Literature Number 5989-4840EN, Agilent Technologies, www.agilent.com.
3. Application Note 1364-1, "De-embedding and Embedding S-Parameter Networks Using A Vector Network Analyzer," Literature Number 5980-2784EN, Agilent Technologies, www.agilent.com.