The objective of calibration is to remove the largest contributor to measurement uncertainty: systematic errors. As you start working in mmWave frequencies, the objective is unchanged, but the actual process for achieving the calibration is quite different.
The mmWave frequency band from 30-300 GHz is enabling technologies such as 5G and radar. But as we move into these higher frequencies, wavelengths become smaller and margins for error become tighter. The opportunity at mmWave frequencies is substantial. But, you can’t forget to account for the unique measurement challenges that come with moving to this frequency band. Properly calibrating your measurement set up is critical if you want to get accurate and repeatable measurements.
Figure 1: Network analyzer with a test set controller and frequency extenders
Your Measurement is Only as Good as Your Calibration
If you regularly work with a VNA, you’re probably familiar with the necessity of calibration. VNA’s are designed to be incredibly precise measurement tools, but without proper calibration, you’re leaving that precision on the table. To maximize the precision of your VNA, you need to calibrate it using a mathematical technique called vector-error-correction. This type of calibration accounts for measurement errors in the VNA itself- plus all the test cables, adapters, fixtures, and probes you have hooked up between your analyzer and the DUT. But calibration at mmWave isn’t this simple.
New Calibration Challenges
The main calibration challenge that comes with working at mmWave frequencies is that you now need a broadband calibration over a very wide frequency range- often from 500 MHz up to 125 GHz or higher. But most calibration techniques aren’t designed to get a cal over such a wide range. What you’re really looking for is a load that offers this broadband frequency coverage. You can get reasonable accuracy by using a well-designed broadband load. But a sliding load isn't a good fit for mmWave. So, what other option do you have?
The Old Model: Polynomial Calibration
Well, you might first consider using a polynomial model. This is a common model used at low frequencies. With this model, you’d need three bands- low, high, and a type of broadband sliding load. This usually works fine at frequencies below 30 GHz, but as you get into the mmWave frequency range, you’ll notice some issues.
Figure 2 shows a short with three different polynomial models- low, high, and broadband. The x-axis is frequency in GHz and the y-axis is the error ratio. (So, low numbers are good in this case) The red trace is when we use a low band model, one that is optimized for low band performance. It has a good load, but potentially limited shorts. For this signal, around 40 GHz, we notice that it breaks down and the error starts to expand out.
The blue trace is when we use shorts without any low band load. In this case, with the multiple shorts, you limit the performance at 40 GHz and above.
However, if you can combine a broadband model that takes advantage of the lower band load of the red trace and the high band offset short corrections of the blue trace, your result would be something like the green trace.
Figure 2: Low vs high vs broadband load models across a frequency range of 0-70 GHz
This demonstrates the new challenge of working at mmWave frequencies. As we get into these broadband frequencies, we need to eliminate the load. To do this, you need to use multiple shorts to cover the broad frequency range that you are now working in. It’s no longer possible to find a single load that covers the full frequency range you are testing. Also, you can’t just combine multiple shorts to achieve this either. A new solution is required.
The New Model: Database Calibration
So, we know we need to use multiple shorts to cover this broad frequency range. But how? You need a calibration kit that eliminates the need for a broadband load. It should implement multiple shorts to cover the entire frequency range you’re working in- something like the Keysight calibration kit in Figure 3. This mechanical, coaxial calibration kit:
- Has a low band load, four shorts, and an open,
- Covers the low frequencies up to 50 GHz with the load, and
- Uses the offset shorts to provide states on the Smith Chart that represent different impedance conditions.
Figure 3: Mechanical calibration kit
This calibration kit uses a database model. This model is a good fit for mmWave testing. It characterizes each device using a specified dataset and uses a Smith Chart with known data of various components across a certain frequency range.
For example, for a source match type measurement, if we’re measuring a high reflect device, we can ask “what represents a good short at this frequency?” We plot that out, and we use this as our database calibration model. You can do that for any type of measurement you are working with: plot out the ideal conditions and use that as a model.
This dataset then allows us to calibrate our system.
The Keysight calibration kit in Figure 3 uses these techniques and allows us to effectively calibrate our system for mmWave testing. It’s important to realize that calibration kits and methods that work at lower frequencies simply do not work at these broadband frequencies. You need to consider selecting a new set of calibration tools that will optimize the accuracy of your mmWave test set up.
Tight margins at mmWave frequencies require new, more precise calibration techniques. You need to be able to make accurate, repeatable measurements or else risk design failures and missed deadlines.
Proper calibration across the broad frequency range is the first step to a reliable test set up. Consider re-evaluating your test set up, calibration tools and techniques. What changes do you need to make for working in the mmWave frequency range? How can you ensure you’re getting the most reliable measurements and avoiding costly test errors?