«Challenges and Solutions for Removing Fixture Effects in Multi-port Measurements Robert Schaefer, Agilent Technologies schaefer-public ...»
Challenges and Solutions for
Removing Fixture Effects in
Robert Schaefer, Agilent Technologies
As data rates continue to rise in today’s digital networks, the need to accurately remove
the fixture effects continues to become more important. This paper will discuss several
calibration and measurement techniques needed for removing fixture effects. One
common technique used is to design a TRL calibration. A question that always arises is:
How can I determine the quality of this cal kit? Techniques for designing good cal kits, validating the quality of existing cal kits, and alternative methods to remove fixture effects, including considerations for simplifying multiport fixture removal calibrations will be discussed.
Author(s) Biography Robert Schaefer is the Technical Leader and R&D Project Manager for the Signal Integrity Group of Agilent Technologies in Santa Rosa CA. His responsibilities include product management, planning, and strategy.
For more than 20 years of his career he worked in Research and Development as a designer and project manager. He has worked in several divisions of Hewlett Packard and now Agilent Technologies.
His design and management experience covers the breadth of GaAs IC and microcircuit design, RF and analog circuit design, instrument firmware and computer aided test software, device modeling and design of modeling systems, microwave and RF CAD products, and signal integrity.
Prior to joining Hewlett Packard in 1976, he completed his BSEE and MSEE degrees from the University of Missouri at Rolla.
Introduction A common problem when measuring physical layer devices is the measurement of just the device under test (DUT). There are two general solutions to the problem. One is to probe the DUT (after calibrating to the probe tips) and the other is to build a fixture to convert from the coaxial connectors that are common on measurement equipment, to the actual DUT. The problem in the second case is how to remove the effects of these fixtures. This paper will describe two common methods to accomplish this.
Description of DUT The DUT for this paper will be a typical Signal Integrity device, a backplane (Figure 1).
As with nearly all Signal Integrity devices, such as connectors, cables, packages, PCB interconnects, there are not coaxial connectors that are part of the device. In this case, to connect to the backplane connector, a fixture card was designed. It was constructed from Rodgers 4350 material. The fixture card has a backplane connector on one end and a number of SMA connectors on the other end. The connectors are a surface mounted connector that is held in place with machine screws from the backside of the fixture card.
Single-ended transmission lines connect the SMA connectors to the backplane connector.
Each line is the exact same length and impedance. When performing a typical coaxial calibration with a Network Analyzer, these fixture cards are included in the measurement.
The fixture can have a significant effect on the overall measurement.
Figure 1 – Backplane with fixture cards.
Fixture Removal To obtain the measurement of just the backplane the effect of these cards must be removed from the measurement by either calibration techniques or by de-embedding the model of the fixtures from the measurement. There are several calibration techniques that can be used. Two common ones are SOLT (Short, Open, Load, Thru) and TRL (Thru, Reflect, Line). Before describing these, a description of the calibration reference planes is in order.
To de-embed the fixture, a model of the fixture card is needed. This model is usually in the form of an S-parameter file representing the fixture. The actual mathematics to deembed the fixtures is quite simple. The difficult part of the problem is to determine the Sparameters for the fixture. One technique is to model the fixture using a 2D or 3D electromagnetic field simulator. Reasonable results are achievable is the material properties and dimensions are known accurately. Another approach will be described in this paper. That technique is a two-tier calibration. This will result in the measurement of the fixture’s S-parameters.
Calibration Reference Planes Figure 2 shows the typical DUT and fixture cards connected to the test-port cables of a network analyzer. The reference planes for an E-cal or coaxial SOLT calibration are at the end of the test-port cables. In this case the measurement includes both fixtures as well as the DUT. For a TRL calibration or an in-fixture SOLT calibration, the reference planes are at the plane of the DUT. For this measurement, just the DUT is measured and the fixtures are removed as part of the calibration.
Figure 2 - Calibration Reference Planes
In Fixture SOLT calibration To create an in fixture SOLT calibration kit, a short, open, and load calibration standards must be fabricated on the same panel of the printed circuit board (PCB) as the fixture cards. The short, open, and load each need to be located at the end of the same structure that is used on the fixture card (1x). That is the exact same connectors, launches, and transmission lines used to connect to the DUT, in this case the backplane connector. The thru is two fixtures back to back (2x). Broadband loads are difficult to fabricate on PCBs.
The best results are typically two 100 Ohm small surface mount resistors in parallel with several vias to ground to minimize the inductance. A short is fabricated using several vias and the open is typically just an open-ended transmission line. For higher frequencies ( 1 GHz) the short and open should be modeled. The inductance of the short and fringing capacitance of the open can be modeled as shown in Figure 3.
Figure 3 - In fixture SOLT calibration kit.
In Fixture TRL calibration Like the SOLT calibration kit, the TRL calibration standards must be fabricated on the same PCB panel as the fixtures themselves. The variations from board to board are usually significant enough to limit the accuracy of the TRL cal. The TRL math assumes that the connectors and launches are identical. The thru and lines are identical except for their length. The reference plane is usually set to be the center of the thru (the same as the reflect). For this case the delay for the thru and reflect are set to zero. Typically it will take three to four lines to cover the frequency range of interest. The frequency range over which each line can be used is limited by the phase of the line relative to the thru. The range is 20-160 degrees.
The reflect standard, either a short or open, usually does not need to be modeled. A load standard is typically used to get to very low frequencies ( 100 MHz). It is easy to fabricate since it is only used for very low frequencies, unlike to one in the SOLT calibration kit.
Figure 4 - TRL calibration kit.
Figure 5 is the TRL calibration board for the fixtures. It is made with Rogers’ material and has vertical surface mount SMA connectors. The thru is in the upper left, with the short and open reflects below. The three lines go from left to right across the bottom, with the longest one on the right. Finally, there are two loads on the upper right. The transmission lines are stripline transmission lines on a buried layer.
Figure 5 - The TRL calibration kit
Checking the TRL Lines The first step after constructing the TRL cal kit is to check the quality and length of the lines. To do this, start with the thru, doing a 2-port transmission measurement. Display the phase and save the trace into memory. Display the “Data/Memory” trace. It should be a flat line. Connect each line and measure where the phase is 20 degrees. This is the lowest frequency that can be specified for the line. Measure the first 160 degree point.
This is the highest frequency that can be specified for the line. Also look to see that the phase is clean and linear. With this information you can define your cal kit standards.
Figure 6 is magnitude (top trace) and phase (bottom trace) of line 3 divided by the thru measurement.
Figure 6 - Magnitude and Phase of Line 3 relative to Thru With the calibration kit defined, the next step is to do a 2 port calibration on the network analyzer, using either an E-cal or coaxial SOLT kit. This sets the reference plane at the end of the test-port cable. Measure each of the standards and look at the thru and lines to make sure there are no resonances or strange behavior. Next convert the reflection parameters (S11 and S22) of each of the standards into the time domain using an IFFT. For this paper, the Physical Layer Test Software (PLTS) was used for calibration, measurements, transforming and displaying the data. Overlaying all the S11 and S22 measurements in the time domain, makes it easy to observe how identical the connectors, launches and line impedances are to each other.
Looking at the lines and thru in the frequency domain (Figure 7), it is easy to observe there are no resonances, just a ripple due to the mismatch of the connectors and the length of the line.
Looking at the time domain (Figure 8),it is easy to see that the connectors on the thru are nearly identical. There are two features. The first is the capacitive dip, followed by the inductive peak. Line 1’s input connectors are also nearly identical. It’s output connector T44 is has about 1 ohm less for the inductive peak. Line 2’s inductive peak is almost 1 ohm higher for the output connector (T66). Line 3 is nearly identical to the other connectors. Overall the repeatability looks very good and should provide a good TRL calibration. However, the variation in impedance will ultimately limit the calibration accuracy. Note the impedance of the line is about 51.5 ohms. This will be the impedance reference for the measurement.
Figure 7 - Insertion Loss of Thru and Lines Figure 8 - Thru and Line Connectors and Launches Thru Results after TRL calibration With the cal standards examined in both the time and frequency domains it is time to perform a TRL calibration using the kit that has been defined. The TRL calibration sets the reference plane in the middle of the thru. Therefore, measuring the thru should result in 0 db of loss and 0 degrees of phase shift. The return loss of the thru gives an indication of the repeatability of the connectors and calibration. It doesn’t indicate the accuracy of the calibration kit. As can be seen in Figure 9 the repeatability of the thru measurement is
0.01 db and 0.5 degrees. The return loss varies from 45 to 70 dB.
Figure 9 - Thru results after TRL calibration
Measurement of DUT (Backplane) With the TRL calibration complete, the next step is to measure the DUT (backplane).
Looking at the time domain of S11 (Figure 10), it is easy to see the effect of removing the adapter with the TRL calibration. The blue trace (upper) is the backplane with an E-cal calibration. It includes the backplane and both fixtures. The fixture is shown in the circle.
The red trace (lower) is the measurement after a TRL calibration. The fixture is removed and the trace is shifted to the left.
The difference in the impedance level is do to the fact that for E-cal (blue trace) the system impedance level is set to 50 as defined by the E-cal. For the TRL calibration the impedance is set by the line impedance. The impedance of the lines is slightly higher than 50 ohms (51.5) and it is dispersive i.e. it changes with frequency. That is why there is not a fixed offset between the red and blue traces. The difference increases with frequency as the line impedance increases with frequency.
Figure 10 - Backplane measurement with and without the fixture.
Two Tier Adapter Removal Calibration Now that the TRL calibration kit has been defined, we can use it, along with either an Ecal or a mechanical cal kit to perform a two tier adapter removal cal. The adapter removal calibration is normally used when measuring non-insertable devices. Normally this is done with two SOLT calibration kits or E-cal. In this case the calibration process is used to remove an adapter from the measurement.
For this case, port one is a SOLT calibration and port two is a TRL calibration. The reference plane for port one is the coaxial end of the test-port cable. The reference plane for port two is the TRL reference i.e. the center of the thru. If a thru is measured after completing the calibration, the resulting measurement will be half of the thru which is the left fixture. After saving the measurement as a touchstone file (.s2p), it can be used to deembed the fixture from the measurement done with an E-cal calibration. Reversing ports 1 and 2 of the.s2p file gives the file for the right fixture, which can also be de-embedded from the measurement. Figure 11 shows the reference planes and measurement in more detail.
Figure 11 - Reference Planes for two tier adapter removal calibration Figure 12 shows three different measurements of the thru. The bottom trace is the thru after and E-cal calibration. It shows the actual loss of the thru is about 2 dB at 20 GHz.
The top measurement is the thru after an E-cal. The thru is 0 dB of loss. The middle trace is the thru after the adapter removal calibration, showing the loss is ½ the loss of the bottom trace. The right plot shows the three phase measurements, corresponding to the different calibrations. The phase of the adapter removal cal is ½ the phase of the E-cal measurement. These indicate we were successful in creating the model of the fixture.
Figure 12 - Measurements of Thru after three different calibrations Using the.s2p files for the fixtures, and de-embedding the fixtures from the backplane measurement done with the E-cal calibration, gives the middle trace shown in Figure 13.
It also effectively removes the fixture like the TRL calibration, but the reference impedances is much closer to the E-cal (the 50 Ohm reference) measurement. The advantage of this method is a very quick multiport calibration with an E-cal, and then the de-embedded measurement. This is much faster than performing a TRL calibration on multiple ports.
Figure 13 - Backplane measurement with fixtures de-embedded