Between DC and 18 GHz a coaxial setup is forgiving. The connectors are the same SMA parts everyone has used for thirty years, the cables can be ordinary flexible assemblies, a two-port calibration can be done in ninety seconds, and the residual errors sit comfortably below whatever the DUT is doing. Move the same bench to 40 GHz and every one of those assumptions expires at once — not because the equipment is worse, but because the electrical size of every mechanical feature has grown by a factor of two.
This note is about what actually changes above 18 GHz and how to keep a 40 GHz measurement honest. It is written for engineers who already know how to calibrate a VNA and are now finding that the results move when nothing was touched.
18 GHz Is the Real Dividing Line, Not 26.5 GHz
An SMA connector is specified to 26.5 GHz, and that number is honest — the interface will not launch a higher-order mode below it. What changes at 18 GHz is not the mode cutoff, it is the tolerance budget. An SMA has a 1/4-36 thread, a PTFE bead and a centre pin held to a few thousandths of an inch. At 18 GHz a 25 µm error in pin depth is a small fraction of a wavelength and costs a little return loss. At 40 GHz the free-space wavelength is 7.5 mm, and inside an air-dielectric line it is the same 7.5 mm — so 25 µm is now 1/300 of a wavelength at the mating plane, and the discontinuity it creates is a measurable reflection rather than a rounding error.
Above roughly 26.5 GHz the problem becomes explicit: SMA is out of its mode-free range, and the interface that owns the 40 GHz band is 2.92 mm (also sold as K). It shares the SMA thread, which is exactly why it is dangerous — a 2.92 mm female will accept an SMA male, and an SMA male has a slightly larger centre pin that opens the 2.92 mm socket on contact. Once opened, that socket is out of tolerance permanently.

| Interface | Mode-free to | Mates with | Watch out for |
|---|---|---|---|
| SMA | 26.5 GHz | SMA, 3.5 mm, 2.92 mm (one way) | PTFE bead, pin depth, wear after a few hundred cycles |
| 3.5 mm | 33 GHz | SMA, 3.5 mm, 2.92 mm | Air line, but the same thread as SMA — easy to mix up |
| 2.92 mm (K) | 40 GHz | SMA, 3.5 mm, 2.92 mm | An SMA male will open the socket — mark the two apart |
| 2.4 mm | 50 GHz | 2.4 mm, 1.85 mm | Smaller thread; nothing else fits |
| 1.85 mm (V) | 67 GHz | 2.4 mm, 1.85 mm | Pin diameter tolerance dominates everything |
The rule that follows is unglamorous but absolute: at 40 GHz, keep 2.92 mm and SMA hardware physically apart, and label which instrument port is which. Our 2.92 mm adapter range exists partly to make that possible — a 2.92 mm to SMA adapter is a designated, dimensioned transition, while a direct mate between the two families is a slow way to destroy a test port.
Calibration: What the Kit Is Actually Correcting
At 40 GHz the calibration is not a formality that removes a few tenths of a decibel. It is the thing that defines the measurement plane, and its residual errors become the floor of everything you can claim. Three practical points:
- SOLT needs a kit definition, not just standards. The open, short, load and thru only mean something through the model in the kit file — the open’s fringe capacitance, the load’s inductance, the thru’s delay. Using a generic definition with somebody else’s standards puts a systematic error into every subsequent measurement that no amount of averaging will remove.
- TRL removes the standards problem, not the fixture problem. A thru-reflect-line calibration builds the error terms from the fixture itself, so it is more accurate when the connectors are exotic — but the line length has to span a usable band, and below the first valid frequency the calibration is simply undefined.
- ECal is a time saver, not an accuracy upgrade. An electronic calibration module is repeatable and fast; whether it is more accurate than a well-kept mechanical kit depends on the module’s age and its own calibration history. At 40 GHz, an ECal that has been dropped is worse than a mechanical kit that has been looked after.
Two habits pay for themselves immediately. First, verify the calibration with an independent artifact — a known attenuator or a beatty line — before trusting a result, because a bad calibration looks exactly like a good one until something disagrees. Second, write down the reference plane in the test record. If the plane moves and nobody notices, the numbers are comparable only to themselves.
The Cables Are Part of the Calibration
Every calibration defines its reference plane at the end of the test cables. That makes the test leads part of the instrument: when their electrical length changes, the calibration is wrong, and nothing on the VNA says so. At 40 GHz a 1 m assembly holds tens of thousands of degrees of phase, so a temperature coefficient of a few hundred ppm per °C is enough to move the plane by tens of degrees per °C.
What that means in practice is short test leads, clamped rather than draped, and re-verified when the bench changes temperature. The cable assembly selection guide goes through the numbers family by family; the short version is that a phase-stable or armored assembly is not a luxury at 40 GHz, it is the difference between a calibration that holds and one that quietly expires.
Reading Ripple: Finding the Fault by Its Period
The most useful diagnostic at 40 GHz is also the cheapest. A reflection that appears at a fixed electrical distance from the reference plane produces a ripple in the sweep whose period is set by that distance:
Δf = c / (2 · L · √εr)
Measure the spacing between two adjacent ripple peaks in GHz, and you have the distance to the reflection in millimetres. In an air-dielectric line a 0.5 GHz spacing means the reflection is about 30 cm away; 0.25 GHz means 60 cm; 1 GHz means 15 cm.

Three cases cover most of what you will see:
- Ripple with a long period (a few hundred MHz across a 40 GHz sweep) is a reflection far away — the other end of the test cable, or the DUT’s own input connector.
- Ripple with a short period (1 GHz or more) is something close: a gap at the mating plane, a damaged bead, a pin that is not seated.
- A single spike rather than a ripple is a resonance — a cavity, a stub, or a fixture that is radiating. Distance arithmetic will not help; change the geometry and watch whether it moves.
This is also the fastest way to tell a cable problem from a DUT problem. Disconnect the DUT and terminate the cable: if the ripple is still there, it was never the DUT.
Power, Level and the Parts in Between
Above 18 GHz, attenuators and terminations stop being accessories and become part of the error budget. Three things are worth stating plainly:
- Pads protect the instrument and cost dynamic range. A 10 dB pad in front of a receiver buys 10 dB of protection and spends 10 dB of sensitivity; at 40 GHz where cable loss is already 2–3 dB/m, that budget disappears quickly. Use the smallest pad that protects the port, and account for its loss in the calibration.
- Power ratings derate with frequency. A termination rated at 10 W at DC is not a 10 W part at 40 GHz; the resistive film is thin, the heat path is long, and the VSWR rises as it warms. Our 2.92 mm attenuators quote their ratings at the top of the band rather than at DC for that reason.
- Every extra connection is a reflection. At 40 GHz a setup with five adapters in series has five chances of a 0.3 dB ripple and a damaged pin. Count the connections and remove the ones that exist only because the hardware was within reach.
A 40 GHz Setup Checklist
- Interface audit. Confirm every port is 2.92 mm or better; tape over any SMA female that could be mated by mistake.
- Inspect before mating. 10–20× magnification on every mating plane, every time. A particle is a reflection; a burr is scrap.
- Torque to spec with the correct wrench: 0.9 N·m for SMA, 3.5 mm and 2.92 mm. The 1.0 mm care guide covers the smaller families.
- Short, clamped test leads. Route them once, tape them down, and leave them alone for the rest of the session.
- Calibrate with a defined kit, then verify against an independent artifact before measuring anything that matters.
- Record the reference plane and the date of the calibration in the test record, not in your head.
- Terminate the cable if a result looks wrong. Half of all “DUT problems” disappear at that step.
- Re-check the ripple period whenever the sweep changes shape, and convert it to a distance before theorising.
FAQ
Can I use SMA connectors at 40 GHz if the datasheet says 26.5 GHz?
The connection will be made and you will get a number, but above 26.5 GHz the interface is no longer mode-free: energy starts coupling into the TE11 mode, and both loss and return loss become unpredictable and position-dependent. For a 40 GHz measurement, use 2.92 mm hardware; it costs little more and it removes an entire class of unexplained variation.
Why does my thru measurement ripple after a good calibration?
Because a thru measurement shows every reflection in the path, including the ones the calibration does not remove: the cable’s own return loss, the adapter between cable and DUT, and any gap at a mating plane. The period of the ripple tells you the distance — long period means far away, short period means at the connector. Start by terminating the cable and re-measuring.
Is a 3.5 mm connector safe to mate with a 2.92 mm connector?
Yes. 3.5 mm shares the SMA thread and the air-line construction, and 2.92 mm was designed to be compatible with it. The direction that is not safe is SMA male into 2.92 mm female, because the SMA pin is oversize for the 2.92 mm socket and opens it permanently.
How often should a 40 GHz calibration be redone?
On evidence rather than on a schedule. Verify with an artifact after every setup change, after the bench has been moved, and whenever a result disagrees with an earlier measurement by more than the repeatability you have established. A calibration on a bench that has not moved or changed temperature can hold for hours; one done on a bench in a draughty room may not hold for twenty minutes.
Do I need a torque wrench at 40 GHz?
Yes. The window between “not seated” and “damaged” is small, and at 40 GHz an under-torqued connection shows up as a few tenths of a decibel of ripple rather than as an obvious failure — which means it gets blamed on the DUT. One wrench per interface family is the cheapest improvement most benches can make.
The Short Version
Above 18 GHz the tolerance budget, not the mode cutoff, is what changes; at 26.5 GHz SMA runs out of mode-free bandwidth and 2.92 mm takes over to 40 GHz. Keep the families apart, inspect and torque every mating, calibrate against a defined kit and verify with an artifact, keep test leads short and clamped, and use ripple period to find reflections instead of guessing. None of that is exotic — it is just the discipline that a lower frequency lets you skip.
If you are building a 40 GHz bench, the adapter range covers the transitions, the attenuators and terminations cover the level and the match, and the VSWR primer covers how to read the sweep once everything is connected.
About the author — this guide is maintained by the RFSPECT engineering team in Shenzhen, who design and test coaxial adapters, attenuators, terminations and cable assemblies to 110 GHz. If a number here disagrees with the drawing you received, tell us and we will correct it.
The terms in this note, from mode-free bandwidth to reference plane, are defined in the RF and microwave glossary.

