As a high-speed interconnect supplier, we need to present measurement results that are accurate, reliable, and of good quality. Many of our customers use these results to decide if a certain connector can be used in their next-generation system. Impedance corrected de-embedding is a method that helps to achieve this requirement.
Before I explain impedance corrected de-embedding, first I need to explain test-board de-embedding. When characterizing a new high-speed connector, we mount it on a test fixture, typically a PCB terminated with a coaxial launch connector. As a result, we are measuring not only the performance of the connector but also the performance of the test-boards. These test board measurements are unwanted since the objective is to measure the performance of the connector only. To remove the impact of the test boards, a model for the test-fixture is required. Once this model is available it can be de-embedded, or you might say it’s “subtracted” from the original measurements. To obtain the test-fixture model, typically a 2x calibration trace is bifurcated. This is illustrated in Figure 1:
Figure 1: To obtain the test-fixture model, typically a 2x calibration trace is bifurcated. The top image shows he connector plus test fixture impedance combined, the middle shows them separating, and the bottom shows the impedance of the connector.
The bottom image in Figure 1 shows the impedance of the connector only after test-board de-embedding. This impedance is clearly not causal, which is the main problem of standard de-embedding methods.
A 2x calibration trace is used to derive a model for the test-fixture. Meaning the PCB used to generate the test-fixture model is physically different from the test-fixture PCB. And due to PCB manufacturing process and material variations (e.g., fiber weave effect), the impedance of the test-fixture model differs from the actual test-fixture impedance.
This is the main difference with impedance corrected de-embedding methods. They also make use of the actual test-fixture PCB to generate the test-fixture model and as such, as is shown in Figure 2a, the impedance of the test-fixture model perfectly matches the impedance of the actual test-fixture. After de-embedding of this test-fixture model, a perfect causal connector only model is obtained (see Figure 2b).
Figure 2: (a) top image shows a comparison of test fixture models used for de-embedding the connector, and (b) bottom image shows the subsequent connector model
The achieved improvement depends on the impedance variation between the impedance of the 2x calibration trace and the impedance of the test-fixture. The greater the mismatch, the greater the improvement will be.
Figure 3 shows the results of a simple experiment. A perfect thru is measured multiple times with various test-fixtures. If no test-fixture is used, a return loss of about -60 to -100 dB would be measured, depending on the network analyzer settings. Using a standard de-embedding method, the measured return loss is about -20 dB, while using an impedance matched de-embedding method, the measured return loss is closer to -40 dB. A 20 dB improvement is obtained.
Figure 3: A comparison of two methods for de-embedding. Using an impedance matched de-embedding method results in a 20 dB improvement.
Editor’s Note: Here’s a link to Stefaan’s webinar – Impedance Corrected De-Embedding.
An earlier version of this post appeared on the Samtec blog.