Field notes from European Microwave Week 2026, London.
On the Smiths Interconnect stand at EuMW 2026, a compact Anritsu USB VNA was connected to a green test board with a smaller board stacked on top of it. The product was EZiCoax, a single-piece, solderless, spring-loaded coaxial contact for board-to-board RF connections. The product card quoted about −1 dB insertion loss at 40 GHz. The screen showed a transmission trace that reached about −18 dB at 40 GHz.
Both numbers can be correct. They describe different things, and the difference between them is one of the most common sources of confusion in interconnect measurements.
The setup
The instrument was an Anritsu MS46122B-043, a two-port ShockLine compact USB VNA. According to the booth card, the series is available with upper frequency limits of 8, 20 or 43.5 GHz.
According to the exhibitor's product card, the connector itself is a double-ended, spring-loaded coaxial contact rated for board-to-board signalling up to 40 GHz. It has a 3.0 mm pitch, a minimum mated height of 3.30 mm, and up to 0.2 mm of compliance in the Z axis. It can be combined with digital, power and DC contacts in an interposer.
Both VNA cables were connected to the same pair of launch connectors at one corner of the base board. The measured path therefore presumably runs from a launch connector, along a base-board trace, up through a board-to-board transition, across the daughter board, back down through a second transition, and along another trace to the second launch connector. I did not confirm the exact routing with the exhibitor.
What the screen showed
Channel 1, transmission, with correction on. The trace starts at about 0 dB and falls steadily: about −5 dB at 8 GHz, about −10.5 dB at 20 GHz, and about −18 dB at 40 GHz. That is roughly 0.45 dB/GHz on average, almost linear in frequency. A few small local features appear near 2.5, 10 and 18 GHz.
Channel 2, reflection, labelled UNCORR. The sweep runs from 1 MHz to 43.5 GHz. Both markers read −19.6096 dB at 40 GHz. Two traces vary between about −15 and −60 dB, and both run flat above roughly 39–40 GHz.
Reading 1: the board, not the connector, sets the loss
The path on this board includes two launch connectors, two lengths of PCB trace, and two board-to-board transitions. If each transition contributes about 1 dB at 40 GHz, as the card claims, the two transitions together account for only a small part of the 18 dB total. Most of the loss comes from the PCB traces and the launches.
The shape of the curve supports this. Dielectric loss in dB rises roughly in proportion to frequency, while conductor loss rises roughly with √f, plus some extra loss from surface roughness. A trace that falls almost linearly in dB from 1 to 40 GHz points to dielectric loss as the dominant term. That is what you would expect from a demonstration board built on a standard laminate rather than a dedicated low-loss RF material. I did not confirm the laminate.
So the −1 dB connector figure and the −18 dB path figure do not contradict each other. One describes a component and the other describes a fixture. The mistake would be to read one as the other.
Reading 2: the "UNCORR" label matters
The reflection channel was not corrected, so its traces include the VNA's raw source and load match plus the reflections from the test cables. They are not the board's return loss, and certainly not the connector's. The flat segments at the top of the sweep also suggest data outside the valid sweep or calibration range. Before reading any reflection value off a screen, check the correction indicator. It is a small label, and it changes everything about what the trace means.
Reading 3: how you would isolate the connector
To extract connector-only performance from a board like this, you need to remove the fixture:
- 2x-thru de-embedding: build a reference path with identical launches and traces but no board-to-board transition, and use it to de-embed the fixture, for example with IEEE P370-style methods.
- TRL calibration: put thru, reflect and line standards on the board, so the calibration reference planes move to the transitions.
- Time-domain gating or TDR: separate the transitions from the traces in time, and locate whichever discontinuity causes the small features near 2.5, 10 and 18 GHz.
A test board with several connector pairs, like this one, often carries exactly such reference structures. That is worth asking about when you evaluate any interconnect demo.
Reading 4: compliance is a test variable
A spring-loaded contact with 0.2 mm of Z-axis compliance will not have the same RF behaviour at minimum, nominal and maximum compression. Contact force, contact area and the effective transmission-line geometry all change with compression. A meaningful qualification therefore measures insertion and return loss across the compression range, for example with shims, and repeats the measurement over mating cycles. For a solderless contact, that variation over compression and mating cycles is just as much a specification as the number at 40 GHz.
A short checklist for board-to-board RF interconnects
- State what every S21 includes: connector only, transition plus traces, or the full path.
- Design a 2x-thru or TRL structure into every connector test board.
- Use TDR or time gating to look at each transition individually.
- Confirm that correction is on and valid across the sweep before reading return loss.
- Measure across the full compression range and over mating cycles.
- Check crosstalk at minimum pitch when RF contacts share an interposer with digital and power.
Takeaway
At 40 GHz, the interconnect is rarely the biggest loss in the path. On a demo board, the board usually is. The screen at the Smiths Interconnect stand illustrated that cleanly: an almost linear 18 dB of path loss from a fixture whose connectors are specified at about 1 dB each. Reading interconnect data well means knowing which of those two numbers you are looking at, and whether the correction was even switched on.
Notes: Observations and interpretations are my own, based on what was running at the stand during the show. Screen values are estimated from the plot grid in my photo. Connector specifications are paraphrased from the exhibitor's product card and are the manufacturer's figures. Board routing, laminate and calibration details were not confirmed with the exhibitor.