Get our free email newsletter

Do Measurements Validate Simulations?

Or Do Simulations Validate Measurements?

I expect the title of this article might raise a few eyebrows! It is very common for people doing simulations to make a measurement of a similar set up to validate the simulation. This is a reasonable precaution since modern simulation tools will give a very accurate answer to whatever question it is asked. The real issue is did the tool user understand the problem well enough to capture the important features, and did the user understand the tool well enough to use it correctly.

However, we usually do not expect a measurement to be validated. After all, measurements are a great emotional comfort!  I have seen many test laboratories claim measurement uncertainty in the 1.5 to 2 dB range. However, whenever I ask an experienced EMC test person how well they might expect to correlate between two different laboratories, I often get a response that anything better than 8 dB is ok, and certainly, I have never been told that better than 6 dB is expected!  This tells me what the ”real” laboratory accuracy is. And this is when all the equipment is functioning to specification.

When we look at a typical EMC chamber, we all know and accept that the chamber effects can be +/- 4 dB from the theoretical. This alone could account for 6 dB or more difference between laboratories!  When we add the difference between different antenna’s response to the nearness of the metal chamber floor as it travels up/down the antenna mast (which can be as much as 4 dB), the potential for site-to-site variation continues to climb. Then we have an antenna factor that was probably measured in a different test environment than where we use it, cable loss, receiver accuracy, etc. So maybe when a simulation is not validated in the test laboratory, we should simply try another test laboratory?  (I am NOT recommending this practice! But I think you see my point.)

- Partner Content -

A Dash of Maxwell’s: A Maxwell’s Equations Primer – Part Two

Maxwell’s Equations are eloquently simple yet excruciatingly complex. Their first statement by James Clerk Maxwell in 1864 heralded the beginning of the age of radio and, one could argue, the age of modern electronics.

Obviously it would be cost prohibitive to improve all the things in the previous paragraph so the site-to-site repeatability is reduced to 1-2 dB. However, I do think we should be careful to understand exactly how accurate the measurements are and not place too much credibility in the numbers resulting from such measurements.

Again, all the above assumes the equipment is operating correctly. I recently heard a story where a salesman was demonstrating a comb generator source to a potential customer in their chamber. The receiver measured fine over a portion of the total frequency range. However, there was one band where there were no comb harmonics!  It turned out that the receiver had a broken band, and the operators were not aware of it and had been using the receiver with the broken band on product measurements for a while. I have also heard many stories of how a cable from the antenna had a broken connector without operators realizing it. This points to the importance of having (and using) a test artifact on a regular basis.

Usually, benchtop measurements are better controlled with fewer chances for error (although the examples above could also happen in a benchtop setting). However, these measurements often introduce other, more subtle issues. Many years ago, I wanted to make measurements of the impedance between power and ground-reference planes on a printed circuit board (PCB) in order to validate some simulations of the same PCB. The measurements were very different than the simulation results, and this was because the measurement VNA had 50-ohm ports. I had not loaded my simulation ports with 50 ohms. (Why would anyone ever put 50 ohms between power and ground reference?). Once I modified the simulation to include the loading, the simulation and measurements agreed very well. This was a clear example where the measurement changed the thing I was trying to measure!

Of course, the story is not completely one-sided. Simulations can have subtle issues that can cause errors. Years ago, I was involved in a project at IBM where we wanted to know the impedance of vias transitioning through 250 mil thick PCB up to 50-60 GHz. Test equipment, probing techniques, and de-embedding probe effects were not as advanced as they are now. So a group of five engineers teamed up to do simulations on the via structure using five different simulation techniques since it is commonly accepted that if very different simulation techniques give the same result, it is likely the correct result.

The simulation techniques we used included the Method of Moments, the Partial Element Equivalent Circuit technique, two different Finite-Difference Time-Domain tools, and the Finite Element Method. Figure 1 shows the initial results. Since the goal was to have data up to 50‑60 GHz, the agreement above 10 GHz is not good.

- From Our Sponsors -
Initial Via Model Results
Figure 1: Initial Via Model Results

Some careful analysis discovered that the various techniques/engineers all made assumptions about the geometry that made sense to them individually but were slightly different from each of the others. First, it is important to understand that no simulation technique uses round objects, even if the software tool displays a round object. The round object must be converted to rectangular or triangular objects in order for the solution software to properly grid the object. Users should always check the gridding to see that this conversion has been done correctly.

In the above instance, one of the engineers converted the round via, via pad, and via keep out to a square that would fit inside the round object. Another converted the round to a square where the round fit inside the square. Another engineer made the area of the round the same as the area of the square. Once these differences were understood, and the models modified so that all were using the same dimensions, then the simulations all agreed very well. The main point is that each engineer, being very experienced in modeling/simulation, made assumptions that seemed reasonable. The tools gave very accurate answers to the models, but the models were not 100 percent correct!

Another common issue with simulation tools are the source and load ports. A ‘lumped’ port is basically a port that connects at a point. Figure 2 shows an example of a simple microstrip trace using a lumped port. While this is fine for some applications, the current that is spread across the width of the trace must neck down to all flow through the point connection. This will increase the inductance of the connection by increasing the current density in that region. Figure 3 shows an example of this effect.

Lumped Port on Simple Microstrip Model
Figure 2: Lumped Port on Simple Microstrip Model
Current across trace
Figure 3: Current across the trace width must narrow to the single point of contact
Figure 4: Current does not narrow when a number of ports are used in parallel

This increase in inductance can be avoided when a number of ports are used in parallel, often called a ‘face’ port. The impedance must be adjusted to provide the correct desired impedance, with many in parallel.

The other commonly used port is a ‘wave’ port. This type of port is often used to drive or load a transmission line in a printed circuit board model because, when used properly, it will automatically ensure the EM modes are correct. One important point is that the circumference of the wave port is a perfect electrical conductor (PEC). This will connect any metal object that touches the edges of the port. For example, if a stripline is intended to be modeled, and the upper reference is considered a power plane, and the lower reference a ground-reference plane, then the two planes would not physically be connected, but the wave port would force them to be connected.

Another potential issue with wave ports is their size. Users must ensure that they are large enough to allow the correct modes to be created. Figure 5 shows two examples of a PCB with a wave port. Figure 5a shows a small wave port, while Figure 5b shows a larger wave port. Figure 6 shows the electric field with the small wave port. Note that the fields are strongest between the microstrip and the wall of the wave port (PEC), and this is not correct. Figure 7 shows the correct fields are created when the wave port is larger.

Figure 5a: Small wave port
Figure 5b: Large Wave port
Figure 6: Incorrect electric fields due to too close PEC walls of wave port
Figure 7: Correct electric field modes for microstrip trace

Conclusion

There are many ways to make errors, both with measurements and with simulations. Engineers should constantly double-check themselves and not assume that either gives the correct answer. Measurements are often considered the “gospel” but while they can be considered “real world,” they will often change the thing they are trying to measure. Understanding how the measurement devices work and where issues can be created is extremely important to make sure that the results are correct.

On the other hand, simulations will usually not change the thing they are try to measure, but subtle issues can creep in the model, yielding incorrect results. These simulation tools are extremely powerful, and should be a tool in the engineer’s tool box, but they cannot be used as simply as a screwdriver! Training on the simulation technique to understand its strength and its weaknesses is vital.

The bottom line? Question everything.

Related Articles

Digital Sponsors

Become a Sponsor

Discover new products, review technical whitepapers, read the latest compliance news, trending engineering news, and weekly recall alerts.

Get our email updates

What's New

- From Our Sponsors -

Sign up for the In Compliance Email Newsletter

Discover new products, review technical whitepapers, read the latest compliance news, trending engineering news, and weekly recall alerts.