The characterization of high-speed test fixtures can either be accomplished via a probe station or through more standard RF test equipment such as a VNA through a microstrip to coax transition. The latter can be particularly challenging as a poor transition relates to an impedance mismatch, reflections, and ultimately signal degradation that becomes more pronounced at high frequencies. While this is not entirely avoidable in a real system, there are methods and techniques to mitigate its effects. This article covers the various testing techniques for DUTs mounted onto planar test fixtures.
VNA Error Correction
Vector network analyzers are among the most prolific tools for measuring RF parameters (e.g., S-parameters, gain, time delay, group delay) with swept frequency and power measurements as well as time domain measurements. Generally speaking, a VNA involves either a shared stimulus source or two for each port as well as receivers to detect incident, reflected, and transmitted signals across the device under test (DUT) as shown in Figure 1. From collected information, the processor is able to extrapolate the S-parameters, with measurements such as insertion loss, gain, VSWR, and return loss. Even with highly precise equipment, there are intrinsic errors that can occur with VNA measurements; these are known as systematic errors (e.g., directivity, matching, tracking) and they are typically time invariant in nature. These can be removed by performing a calibration—the most common of which is the short, open, load, and thru (SOLT)—bringing the measurement plane from the VNA ports through the coaxial test cables or various passive interconnects, all the way up to the DUT. Random errors also occur but are not predictable as they can be from random manufacturing defects such as the durability of the internal switches and connectors as well as from measurement noise.
Drift error is attributed to temperature changes in the testing environment; the linear coefficient of thermal expansion (CTE) will cause the internal dimensional changes to devices within the VNA as well as the interconnect between internal and external equipment components. This may be negligible at low frequencies but can change the behavior of high frequency devices. For any transmission line (e.g., microstrip, coplanar waveguide, coaxial), changes in temperature will cause changes in electrical length—a parameter related to physical length and the dielectric constant—ultimately, this changes the transmission phase, causing any measurements taken thereafter to be less accurate. To mitigate the effects of this, testing can be performed in a temperature-stable environment. Moreover, phase-stable coaxial test cables can be leveraged to extend the time a calibration is viable.
Planar Transmission Lines in Test Fixtures
High frequency signals require transmission lines to enable a signal to propagate with minimal distortion and attenuation. Among these transmission interfaces are coaxial and strip-type transmission lines. Coaxial cables are best used in long distance transmissions (>6 inches) while the shorter distances in microwave modules require a smaller, lower profile interface. Among the most common planar configurations are stripline, microstrip, coplanar waveguide (CPW), and grounded CPW (GCPW). Fixtures will often leverage the microstrip transmission or GCPW as they are relatively simple to design and fabricate while also relatively straightforward to integrate with active and passive devices (Figure 2). The microstrip transmission line offers adequate performance in the microwave frequencies, however, it tends to experience excessive radiative losses at the millimeter-wave frequencies. At higher frequencies, the GCPW structure with two additional ground planes and ground straps/vias allows for more stable impedance over frequencies and is less prone to frequency dispersion.
Understanding the Transition
The coaxial feed from the DUT to the VNA involves an edge-mounted coaxial connector—typically SMA or precision —fixed onto a substrate with the center pin soldered on the conductor plane while the outer conductor is electrically connected to the ground planes (Figure 3). Removable/replaceable end-launch variants can form a sufficient electrical coupling to the circuit trace through a compressible conductive button such as a fuzz button, or densely packed wire bundles, as well as pogo pins. The relatively bulky coaxial interface requires more bulkiness/thickness on the strip-type transmission line to effectively accommodate the assembly.
Additionally, the signal line widths are typically much larger than the coaxial center pin; the abrupt transition from cylinder to plane creates a discontinuity in the electric fields and transmission line impedance. Ultimately, a signal propagating along this space will have reflections. Adjusting the transition geometry, or tapering the ends of the signal path, can mitigate the unwanted signal distortion. Tying ground together with ground vias is especially important at high frequencies to prevent the ground planes from acting as additional transmission line segments; without this, the ground planes can resonate at ¼ of the wavelength of the signal propagating through it, generating undesired radiation and an unintentional antenna. The spacing of vias is an additional consideration when leveraging signals operating at higher frequencies. Some rules of thumb for via spacing involve spacing no further apart than λ/10 while some recommend spacing at or below λ/8. Regardless, additional vias can prevent unwanted bulk modes and resonances from the PCB that ultimately partly radiate and reflect an incoming signal. Moreover, it has been found for the GCPW that placing vias near the excitation region of the ground patch—closest to the coplanar line—is sufficient to suppress unwanted resonant modes and a random arrangement of vias on the PCB is not necessarily more effective .
Measurement Methodologies for Test Fixtures
When test fixtures are employed, additional error correction methodologies such as port extensions must be leveraged to factor out the loss from the coax to microstrip transition. For this to occur, delay (and sometimes loss) is ascertained from a vendor-specific algorithm (e.g., best-fit straight-line model, coaxial/dielectric loss model) based upon open or short measurements . This involves calculating the loss of a fixture using open or short measurements. Even with port extension accomplished, the mismatch that occurs from the microstrip to the coax forms of transmission cannot be extrapolated and removed from the data as it is beyond the coaxial measurement plane and within the device plane. Moreover, the non-ideal transmission line characteristics such as attenuation and dispersion often go unaccounted for. For these reasons, the accuracy of port extension degrades at higher frequencies.
De-embedding can be a complex process depending upon the fixture. It is typically accomplished by mathematically modeling a fixture and removing the fixture effects from the measured S-parameter file that contains both the fixture and DUT measurement data. This modeling can be from a simple broadband equivalent circuit model or from full-wave 3D modeling of the fixture. The mathematical removal occurs first by converting the S-parameter matrix into cascadable parameters such as T-parameters or ABCD parameters and fixture elements are removed portion by portion . Naturally, the accuracy of the model will directly affect the accuracy of the resulting measurements. Moreover, the process of converting to cascadable matrices often varies on a case-by-case basis as issues can pop up requiring unique solutions. However, unlike port extension, the mismatch effects of the coaxial-to-microstrip transition are accounted for. Outside of this method, custom calibration kits can be fabricated using specific calibration standards to attempt to bring the measurement plane to the DUT.
Calibration Standards for Transmission Lines with Non-Ideal Conditions
As stated earlier, the SOLT calibration, as well as many other calibration standards, is typically meant for coaxial interfaces where systematic effects would be entirely characterized and removed. For this reason, calibration standards such as thru-reflect-line (TRL), line-reflect-line (LRL), and line-reflect-match (LRM) are used to de-embed S-parameters. These non-ideal calibration standards are often used for microstrip and CPW transmission lines to accurately de-embed the non-ideal effects of the fixture while minimizing the size of the calibration kit. A LRM calibration, for instance, leverages a symmetric reflect, a short line, and another line of longer length. Custom calibration kits are typically necessary as DUT-specific parameters such as line lengths must be calculated and fabricated. This method is meant to remove the parasitics at the end of the planar transmission line, including the center pin placement on the line as well as the transition to the coaxial interface as the calibration kit is intended to contain identical adapters, electrical length, propagation constant, and impedance.
Differential Signaling Calibration and De-embedding Methodologies
Signal integrity (SI) test-and-measurement applications often require robust frequency and time domain analysis over wide bandwidths to adequately characterize high-speed differential signals. While there are an array of tools that can be utilized for this process, a VNA can perform both frequency and time domain analysis through an inverse discrete Fourier transform (DFT). Differential signal analysis involves layers of complexity over single-ended signals—much of the noise generated equally effects both lines and can be rejected at the receiver (e.g., common-mode noise) however, cross-mode noise (e.g., common-to-differential mode, differential-to-common mode) that causes signal degradation must be characterized, often using the mixed-mode S-parameter matrix . High SI passive components with nearly identical electrical lengths are required for these test applications to adequately troubleshoot a DUT and discover various SI “pathologies.” The VNA calibration for these types of tests is particularly tricky as most of the devices tested (e.g., PCI Express, SERDES, multi-gigabit ethernet, high speed memory interface) are on planar substrates.
For this reason, a number of unique calibration methods have been innovated, including thru-line de-embedding, 1X-reflect smart fixture de-embedding (SFD), automatic fixture removal (AFR), and 2X-thru SFD with calibration sets that are customized to the test fixture with the same PCB panel, layer numbers/transitions, PCB orientation, and electrical length. The complexity of SI test applications would require the use of an EM simulator to mathematically de-embed the fixture effects. However, refining and simulating the fixture model is typically very involved. The common SI calibration methods use specialized cal kits combined with specific algorithms (e.g., time-domain channel characterization, wave peeling algorithm).
The VNA is an indispensable high frequency measurement tool with more complex testing considerations in order to accurately gather data on DUTs mounted onto test fixtures. There are a number of common test methods used to obtain better S-parameter data. From optimizing the design of the physical transition/test fixture to mathematically modeling the fixture to best de-embed the desired data, the characterization of a DUT on a test fixture is a complicated problem that often requires a custom solution with the right choice of design and components.
1. W. H. Haydl, “On the use of vias in conductor-backed coplanar circuits,” in IEEE Transactions on Microwave Theory and Techniques, vol. 50, no. 6, pp. 1571-1577, June 2002, doi: 10.1109/TMTT.2002.1006419.