Eliminating Effects of PCB Fixtures Using De-embedding with CMT VNAs

Accurate Vector Network Analyzer (VNA) measurement of a Device Under Test (DUT) mounted on a printed circuit board fixture requires some form of de-embedding to eliminate the effects of the fixturing. Port extension, onboard calibration, and TRL might be exploited for this purpose, but each of these methods has limitations and drawbacks. The Automatic Fixture Removal (AFR) program from Copper Mountain Technologies (CMT) provides a superior method of fixture removal for measurements made with CMT VNAs. This brief article describes the exceptional utility of this software.

De-embedding and the Need for Good Fixturing

A Device Under Test (DUT) may be mounted on a printed circuit board if it isn’t otherwise connectorized. The effects of the circuit board and connectors must be eliminated to measure the DUT alone.

Figure 1 - PCB Mounted DUT

The input and output connectors will invariably generate reflections, and the transmission lines will add delay and loss. If a circuit board is to be used to carry a DUT, you can fabricate a board with connectors and a thru line, which may be used to evaluate the performance of the connectors and traces alone. This fabricated board may also be evaluated with the 2x thru the method in the AFR program. The thru line should be as long as the input and output traces of the PCB fixture are put together.

Usually, the coplanar waveguide or microstrip transmission lines on the circuit board will have very good return loss, if properly designed. However, each of the connectors will generate some reflection. Figure 2 shows a typical PCB with a microstrip thru-line and end-launch SMA connectors. The insertion loss is reasonable. The return loss of the transmission line alone might exhibit a slope from about -50 to -25 dB; however, the reflections from the connectors dominate. At some frequency –1.13 GHz in this case – the reflection from the second connector interferes destructively with the reflection from the first connector, resulting in a return loss dip. This occurs because the distance between the two connectors is a quarter wavelength at 1.13 GHz. Subsequent dips in the return loss occur at twice and three times this frequency.

Figure 2 - IL and RL of a 3.4" PCB Thru

The peaks of the return loss bumps (S11) occur where the reflections from the two connectors are in phase and add to each other. The dips in the return loss occur where the two reflections are out of phase and cancel. At these points, the reflection due to the transmission line alone remains.

Not that the choice of connectors and the interface between them and the PCB are critical to achieving good overall return loss. Reasonably low insertion loss and good return loss are critical to achieving good results with any method of de-embedding. It is impossible to properly de-embed a fixture with resonances in the measured DUT frequency range. These show up as discontinuities in the insertion loss and return loss.

De-embedding Methods

Using port extension, you can remove the delay (Figure 1) on each side of the DUT. It is also possible to remove either a constant or linearly changing insertion loss. Copper Mountain Technologies (CMT) VNAs have the capability to automatically set the delay and loss based on the measurement of an open or short on the end of the transmission line. If the DUT can be removed, you would measure an open, otherwise, you would short the input and output of the DUT to ground and perform automatic port extension to the shorts on each side.

Port extension corrects for the delay and loss of the transmission lines, and therefore phase measurements of the DUT are correct. However, the port extension does not correct for reflections from the connectors. This can result in substantial measurement errors. In practice, the assumption of a single delay value over frequency is only valid over a narrow bandwidth due to dispersion in the transmission lines. For example, the port extension may be used for a 2.45 GHz Wi-Fi measurement but may be limited to only 2400 to 2500 MHz.

SOLT calibration may be built into a PCB similar to the one holding the DUT. The short and open traces are easily fabricated, as shown in Figure 3.

Figure 3 - PCB with SOLT

A custom calibration kit definition would need to be created for the VNA to perform SOLT. The open would normally have an end-effect fringing capacity specified, but this would not be known unless a full EM analysis was performed on the circuit board design. The short would have a small, unspecified parasitic inductance as well. These two errors will introduce minor inaccuracies for measurements at high frequencies. The worst contributor to error would be the load design. It is nearly impossible to fabricate a load with better than 20 dB of return loss to 20 GHz. The calibration load is the most important piece in the calibration kit. Its return loss sets the floor for all subsequent reflection measurements. A 20 dB return loss calibration load results in reflection accuracy of ±1 dB at 0 dB, ±3.3 at 10 dB, and no accuracy at all at 20 dB return loss.

There should be no issues with the thru as long as it is identical to the lead-in and lead-out transmission lines for the DUT.

TRL may also be used to perform calibration. This requires a thru line as long as the lead-in and lead-out transmission lines, a match line that is 90 degrees longer than the thru at the center frequency of measurement, and a pair of reflect items – usually shorts – as in Figure 4.

Figure 4 - TRL Calibration

If the thru length is set to 0 length in the calibration kit definition, the VNA will place the measurement plane in the middle of the thru at length L, a desirable result that puts the reference planes at the input and output of the DUT. This method has two drawbacks:

It will work between 0.222*Fc and 1.78*Fc, where Fc is the frequency where the match line is 90 degrees longer than the thru. The match line operates properly when it is between 20 degrees and 160 degrees longer than the thru (with 90 degrees in the middle). The method is therefore bandlimited.

The residual directivity is no longer determined by a calibration load but is determined by the characteristic impedance of the match line. For best results, the return loss of the line itself should be better than 30 dB. Achieving this over a broad frequency range can be challenging.

However, this is still the best choice as compared to port extension and onboard SOLT calibration, if no other choice is available.

S-Parameter De-embedding

The most accurate method of de-embedding is to find the full 2-port S-parameters of the lead-in and lead-out sections of the fixture. If we label the lead-in section A and the lead-out section B with S-parameters SA and SB, then the measurement diagram is as shown in Figure 5.

Figure 5 – S-Parameter Measurement Diagram

We need only find the full 2-port S-parameters for the left and right (A and B) sides of the fixture and perform some mathematics. This calculation is automatically performed by the VNA when the de-embedding files are applied in the S-Parameter Files function. This function can be found in the VNA user interface under Analysis > Fixture Simulator > De-Embedding.

If the A and B sides of the fixture were connectorized on each side, de-embedding would be simple. We would simply measure each side and use the resulting 2-port S-parameters as the de-embedding files for the VNA. However, this is rarely the case, and here the A and B sides only have a connector on one side, as seen in Figure 6.

Figure 6 - Fixture with A and B sides

AFR provides the means to characterize the two sides and allows for two possible configurations. In the first configuration, 2x Thru, the A and B sides must be connected directly together and measured end to end as in Figure 7 (top). In the second, each side is measured separately, with either an open (DUT removed) or a short at the DUT interface as in Figure 7 (bottom).

Figure 7 - 2x Thru and 1x Reflect Configurations

There are three possible methods of characterization in the 2x Thru configuration: time gating, a proprietary filtering method, and bisection.

Figure 8 - AFR 2x Thru Methods

For the time gating method to be accurate, the A and B sides of the fixture must be long enough such that the delay is greater than two impulse responses of the virtual time domain signal. For example, if each side is measured to 20 GHz, then the impulse response is approximately 1/20e9 or 50 pS, and the A and the B side must be at least 100 pS in length. For Rogers 4350B material with εr = 3.48, 100 pS is 633 mils minimum length from the connector to the interface of the A or B sides. Characterization of a fixture smaller than this with time domain gating will produce undesirable results.

In the time gating method, the four S-parameters of the fixture with sides A and B are connected as in Figure 8 and measured to obtain S11, S12, S21, and S22. S11A, the S11 measurement of the A side alone, is found by gating S11 from the A connector to the A/B interface, and S22B is found by gating S22 from the B connector to the A/B interface.

The rest of the S-parameters are derived via the following calculations:

Figure 9 - 2x Thru A and B S-Parameters

From Reference [1] -

Thus, all four S-parameters of the A and B sides are derived.

The filtering method also uses gating to derive the A and B S-parameters but using a proprietary least-squares fit method. The filtering method is more accurate for smaller fixtures, which are right at the four rise-time threshold in length.

For small fixtures – less than four rise times in length – you must use the bisection method. This method amounts to converting the S-parameters to cascade parameters, taking the square root of the matrix, and then converting them back to S-parameters. The algorithm is more complicated than this since the square root has two solutions and only one is valid.

After the 2x Thru measurement is completed, the measured S-parameters of the A and B sides of the fixture may be automatically applied to the VNA for de-embedding, and/or saved to a pair of 2-port touchstone files for later use.

Figure 10 - Apply or Save De-embedding Files

If it is impossible to connect the A and B sides of the fixture together, then use the 1x Reflect method as seen in Figure 11.

Figure 11 - 1x Reflect Method

The 1x Reflect method derives all four S-parameters of the A and B sides while only measuring from the connector with the other side of A or B either open or shorted. This method requires time gating, and both the A and B sides must be longer than two virtual rise times.

To determine the S-parameters, the measured S11 is bandpass gated up to, but not including, the open or short. The resulting response is simply S11A. In the next step, S11 is bandpass gated at the short or open. This response is equal to Γso = S21A*Γ*S12A, where reflection Γ is 1 for an open and -1 for a short.

Figure 12 - Two Gate Positions

We now have all four S-parameters for A, and B is found similarly.

After the 1x Reflect measurement is completed, the two S-parameter files may be applied for de-embedding or saved as a pair of 2-port touchstone files for later use.

The mathematics makes the time domain gating process look very straightforward, but there are a number of proprietary operations – performed by the AFR software – which make the process work well.

For best results, the fixture characterization using the AFR program should be performed to as high a frequency as possible, even if the DUT measurements occur only at lower frequencies. The higher frequencies provide more time domain resolution, and the de-embedding process is therefore more accurate. On the other hand, there can be no discontinuities or resonances in the fixture response over the frequencies of characterization or the process will fail. The insertion loss and return loss of the fixture must be reasonably smooth with no glitches. If resonances do occur, the highest frequency used during the AFR process must be well below them.

Summary

Four methods of de-embedding have been presented: port extension, onboard SOLT calibration, onboard TRL calibration, and finally full 2-port de-embedding using 2-port touchstone files derived by 1x Reflect or 2x Thru methodology. The benefits and limitations of each were discussed. Full 2-port de-embedding using the AFR program from Copper Mountain Technologies is by far the best approach to this problem. The program is easy to use with built-in help screens for each section. Note that the AFR program will only operate with VNAs from CMT.

Click here to learn more about Copper Mountain Technologies on everything RF.