The selection of a network extraction method is heavily dependent on the standards that are available at the inner plane and the results can vary wildly depending on the quality of and knowledge about those standards. While we cannot be complete in this guide in a discussion on uncertainties and sensitivities, we can give some general thoughts. Many publications exist on the topic that may provide more information, such as J. Martens, “Common adapter/fixture extraction techniques: sensitivities to calibration anomalies, 74th ARFTG Conf. Dig., Dec. 2009, and references therein.

The Type C extraction is, in some sense, the most complete since it uses all of the error terms in two full port calibrations to extract the S-parameters of the individual fixture arms. For that reason, Type C is also the most sensitive to standards quality at the inner plane. Uncertainties will follow those of the underlying calibrations quite closely, subject to a repeatability penalty associated with the two calibrations.

Type A is a bit different in that there is only one fixture arm, but it does still use two full-term calibrations. The computation difference is that it does not use transmission or load match terms to calculate the adapter/fixture S-parameters and hence is more immune to problems with those calibration steps than is Type C. Even if there is a problem with a reflect standard, Type A may be less sensitive since the reflect errors propagate to load match in the standard calibrations (particularly true for reciprocal methods as is discussed in Reciprocal Measurements). As an example, a mismatched pad was evaluated using both Type A extraction and an SOLR calibration, but a perturbation was introduced in one of the reflect calibration standards. As shown in Figure: Comparison of Reflected Standard Problem for Type A Extraction and SOLR Calibration., the sensitivity with Type A (labeled AR for adapter removal) is lower. The comment ‘one-sided-distortion’ is included in the plot since symmetric errors in a Type A setup tend to cancel (another advantage in some cases), so the introduced error was made on only one side in this measurement to create a worst-case result. Each trace in the plots represents a different assumed reflect magnitude value for the standard (over a 10 % interval).

Although less sensitive to some effects, it may not always be possible to do reasonable full port calibrations to support Type A. The Type B method requires only one port standards, which can be far more convenient. The internal computations are similar between Type A and Type B (in the full-standards case) except in how output match is determined. Type B is a more source match-intensive computation, so there is greater sensitivity to the high reflect standards’ behavior with Type B than with Type A (and less dependence on low reflect behavior). A comparison is shown in Figure: Comparison of Output Match Sensitivity for Type A (AR) and Type B (BP) Methods for Type A (again labeled AR for adapter removal) and Type B (labeled BP for Bauer-Penfield, two of the original authors to work on this method class). The sensitivity of output match to a reflect standard problem (a short in this case) is indeed greater. Again, each trace represents a different reflect magnitude value. Type B with flexible standards falls into the category of ‘partial information methods’ like Type D. With the partial information techniques, it is useful to distinguish between sensitivity to standards problems from absolute error (even if the standards were perfect). With network extraction methods A, B (in its full-standards form) and C, the absolute error is zero. Type B when using one or two standards allows one to accept some non-zero absolute error in exchange for reduced sensitivities to standards errors and repeatability. These sensitivities for Type B were discussed in the section describing Type B Network Extraction.

Correspondingly, the load sensitivity is somewhat reduced, but it will have an impact on high RL adapters and high IL adapters, as suggested in Figure: Effect of Load Standard Problem on Extracted Insertion Loss (1 of 2).

Type D is another approach entirely and is appropriate when it is difficult to create any reasonable standards at the inner interface. Also, when repeatability is problematic (for example, spring contacts, poorly positioned probes, etc.) trying to achieve knowledge of fewer parameters can be useful. The concept is to reduce weight on the inner plane match and/or make additional structural assumptions and focus on improving the accuracy of the insertion loss extraction. The absolute accuracy will be degraded from what one could get from either of the other methods (if good standards were available), but repeatability sensitivity can be greatly improved and the net practical uncertainty can actually be better. This follows from the 1/(1-x) kind of behavior of match terms in the standard methods. If the measurements are not terribly repeatable and the ‘x’ term is moving near unity, the uncertainty on the final parameter in practical terms can balloon. One can see this in a wafer probing example where probe placement was not that accurate. A series of calibrations were done using SOLR and using Type D extraction with a single thru (labeled partial information in Figure: Comparison of Repeatability Effects). The removal of match dependence lowered the scatter in the final insertion loss values. While repeatability sensitivity is better, there is sensitivity to inner plane match and to problems with the underlying calibration at the outer planes (although reduced).

While repeatability sensitivity is better, there is sensitivity to inner plane match and to problems with the underlying calibration at the outer planes (although reduced).

The effect of an inner plane distortion is shown in Figure: Sensitivities of a Type D Extraction to Underlying Calibration Issues. The effects can be substantial, but the errors in this particular example would have exceeded 4 dB with a Type C methodology because the media was pin-based and not repeatable.

Some of the Type D sensitivities were covered in the previous section. As with the one- and two-standards variants of Type B, Type D allows one to accept some non-zero absolute error in exchange for reduced sensitivities to standards errors and repeatability. Within Type D, there are a number of choices depending on which standards are available. If a thru is available and the fixture halves are relatively symmetric, phase localized D with a thru standard can do very well unless the fixture is very mismatched at the inner plane. Multi-standard variations of D (using two lines or a line and a reflect) can take over in the latter case. If only a reflect standard is available, reflect-based phase-localized D will usually outperform Type B with one standard.

The various extraction methods discussed here present a variety of choices dependent on standards quality, the media involved, and the possible measurement repeatability. In some sense it is a continuum (C->A->B->D) of choices more appropriate as the environment becomes decreasingly metrology-friendly. This is an oversimplification, but some of the sensitivities presented may help in making a good choice for a given measurement setup.