Correlating measurements to simulations
Do you use an oscilloscope to verify the operation of your design or do you simply trust your simulation? Being a conscientious engineer, you probably probe your DUT and view the waveform data on an oscilloscope. You also believe that you have an accurate device or component level schematic model, perhaps a SPICE model. You simulate the behavior at your test point (TP) in your model, and the results don't exactly match what is shown on the oscilloscope. Is your model correct or is the measurement system, including probes and oscilloscope, at least partly to blame?
You may also notice that your circuit behavior changes when you probe the TP. In some cases, it gets worse or in others it gets better. Regardless, the simulation and measurements aren't aligning and you're stuck. How can you explain—or better yet simulate—why this is happening?
One option would be to obtain a model of the oscilloscope's probe and insert this model into your simulation. You can either obtain this model from the manufacturer or go ahead and measure it yourself. While this could help you understand loading-related DUT behavior changes, it doesn't fill in all the missing pieces. That's because you're missing the complete oscilloscope system response from probe tip to display.
In this article, we'll walk through a more inclusive model along with examples that will give you some useful tools for closing this gap between simulations and measurements. All probe and system modeling will be done using Touchstone S-parameter files. The modeling requires a simulation environment with the ability to incorporate S-parameter files, including 2-port and 3-port models, of which several are available.
The whole picture
A problem with using a probe model in simulation is that no matter how good the model, it won't tell you the whole picture. What you see at your simulated probe output omits the oscilloscope response and, most importantly, its DSP correction filter. The DSP correction in the oscilloscope is designed to flatten (sometimes called "normalize" or "equalize") the total measurement system response for a specific DUT impedance. Usually, but not always, this is a constant, real impedance over frequency, typically 50 O single-ended, or 100 O differential. This impedance may not describe your circuit.
What's needed is a complete model that combines the probe's behavior, the oscilloscope's analog response, and the DSP correction filters—all of which sounds like a lot of work. Fortunately, there is an easier way to get an accurate model of the entire probe and oscilloscope system, leveraging the fact that the DSP filter generation process produces a known overall system frequency response when driven from an ideal 50 O or 100 O source.
In the remainder of this article, I'll use what's called the probe/scope/DSP (PSD) model along with an example of the model in action. If you want to generate a similar model for your oscilloscope and probe (assuming you have a circuit you wish to test and compare to simulation), the sidebar, Generating a PSD model (page 3), shows how the PSD model was generated using the appropriate data files on an Tektronix oscilloscope, the nominal system bandwidth response, and a MATLAB script. Note: Other oscilloscope manufacturers also provide S-parameter (DSP) correction for their higher performance oscilloscopes and probes.
The DUT being tested in our examples is a test fixture. Figure 1 shows the unloaded (left) and loaded (right) test fixture. It's a symmetrical fixture with 2.92 mm input and output connectors connected to a VNA (vector network analyzer). The fixture accepts a probe's browser tip at its center test point. The fixture is single-ended, with a grounded coplanar transmission line. Measurements are taken simply by placing one browser input pin on the center test point, and the other input pin on adjacent fixture ground. The fixture could also be used with a solder tip and would yield similar results.
Figure 1. Placing a probe on the test figure (right) loads it. The left photo shows the fixture unloaded.
Figure 2 is the schematic with the PSD system model's differential response. The corresponding probe-only model, obtained directly from the de-embedding files contained on the oscilloscope's hard drive, is shown for comparison purposes.
Figure 2. You can use this circuit to evaluate the differential response of the PSD system mode, and compare it to the probe-only model.
Figure 3 shows simulation results from these schematics. The pink line on the chart represents the analog response of the probe and its browser tip. The PSD system model uses a MATLAB script to convert the analog through-response of the probe into the smooth, 20 GHz system response shown in black. This recreates exactly what the oscilloscope/probe DSP software does: correct the overall system response so that it matches the "ideal" 20 GHz target response when driven from a 100 O differential source.
Figure 3. Differential response of PSD system model compared to browser probe-only model.
In contrast to the different through-responses of the two models, the input return loss of the two models are identical. In this case, extending the model from probe-only to the more inclusive PSD system model doesn’t change the input impedance of the system. This condition is a result of using a probing system that contains active buffers that isolate the probe tip circuitry from the effect of any down-stream connections, such as connecting the probe output cable to the oscilloscope. Hence the oscilloscope (the "S" in the PSD system) has negligible effect on the total loading introduced by probing the DUT. The DSP, ("D" in PSD) has no effect on the input loading, of course. The S and D parameters will affect the THRU response.
Figure 4 gets to the heart of what we are trying to accomplish: using the PSD model to predict probe loading effects on the DUT, and to simulate the expected waveform on the oscilloscope screen or in oscilloscope memory. Figure 4 shows a model of the DUT/fixture used for this article. The fixture is represented as two half-fixture models in series, so that the probing point is at the junction of the two models. These models were derived from fixture S-parameter measurements using fixture-splitting methods, which are beyond the scope of this article.
Figure 4. Use this circuit to evaluate the response of the PSD model at the fixture midpoint and compare it to simulated loaded and unloaded actual fixture behavior.
In this and in the simulations that follow, a voltage controlled voltage source (VCVS) was used as a "virtual ideal probe" to examine the simulated waveform at a point, without adding any loading. Of course, these kind of probes exist only in simulations because all real probes have some finite loading effects. This is shown in the other two sub-circuits of Figure 4 where the DUT test point is loaded with the PSD and probe-only models shown earlier in Figure 2. This time, however, the models are used in a single-ended configuration, where one input of the probe is attached to the DUT probing point, and the other input connected to ground. We can use a single-ended probe because the DUT fixture used for this article is a single-ended trace. Using a differential probe in this fashion to measure single-ended DUTs is also a common practice. To observe the loaded waveform at the probing point, ideal VCVS probes are also connected in shunt with the probe models.
Continue to page two on Embedded's sister site, EDN: "System models help correlate measurements to simulations."