Six Test-Bench Mistakes That Make Good RF Parts Look Bad
Every few weeks someone sends us a switch or an attenuator with a note saying it measures worse than the datasheet. Most of the time the part is fine. The bench is not.
This is not a defence of our datasheets. It is a list of the six things that account for almost every "out of specification" RF component I have seen, in the order they usually turn out to be the cause.
1. The calibration plane is at the analyser, not at the part
You calibrate at the front panel because that is where the cal kit connects. Then you add two adapters, a cable, and a fixture, and measure the DUT. The analyser reports the loss of everything between the plane and the DUT, and calls it the DUT's insertion loss.
At 3 GHz the two adapters might be 0.1 dB. At 18 GHz the same pair can be 0.4 dB, and the mismatch term adds another 0.1 dB that changes sign with frequency. That is more than the specification you are arguing about.
Fix it with adapter removal, or calibrate at the DUT plane. If neither is practical, at least write the reference plane on the test record, so the next person knows what the number includes.
2. Nobody torqued anything
A hand-tight SMA is not a connection, it is a hope. The specification for a 3.5 mm connector is around 0.9 N·m; for SMA about 0.45 N·m. Under-torque gives you a gap in the dielectric and a reflection. Over-torque deforms the outer conductor and permanently degrades both connectors.
The practical symptom is measurement repeatability of 0.05 to 0.1 dB per re-mate, which people then attribute to the component drifting. Buy a torque wrench. It costs less than the argument.
3. The cables moved between calibration and measurement
A flexible cable is a mechanical device that happens to carry RF. Bend it and the impedance changes; the change is small but it is real, and at high frequency it appears as a phase shift and a small loss change.
Tape or clip the cables down before calibrating, and never use a cable as a handle again. If a measurement changes after someone reached across the bench, suspect the cable before the DUT.
4. The DUT is better than the test setup
Measuring a 0.3 dB component on a bench with 0.15 dB of uncertainty produces a number with a 50 % error bar. That is not a measurement, it is an opinion with decimals.
The four contributors, roughly: calibration standards, test port match, trace noise, connector repeatability. Averaging only addresses one of them, and it is usually the smallest. Fix the match and the connectors first — a 6 dB pad in front of the analyser often improves accuracy more than 1000 averages.
5. The unused ports were left open
A multi-throw switch measured with its unused ports open is not the switch you will install. An open port reflects, and the reflection couples back into the selected path. On a SP4T at X-band this can move the measured insertion loss by 0.1 dB and do strange things to the return loss.
Terminate every unused port with a load rated for the power present, and say in the test record which ports were terminated. This matters even more for a coaxial switch with more than four throws, where the geometry makes the coupling larger.
6. The part is at a different temperature than the calibration
A resistive film has a temperature coefficient. A connector has a different one. Twenty degrees of ambient difference between the factory and the receiving inspection is enough to move a 0.3 dB attenuator by 0.02 dB and a return loss by a couple of dB at the band edge.
This one is genuinely hard to eliminate, which is why the specification should be read as "typical at 25 °C" and not as a guarantee at whatever temperature the rack runs at.
What to do instead
A short, boring routine removes most of these:
- Torque every connection, every time, with the same wrench.
- Calibrate at the plane you will quote, and record it.
- Keep a verification artefact — a short, a load, a stable fixed attenuator — and measure it at the start and end of a session.
- Terminate unused ports.
- Let the analyser and the DUT reach the same temperature before calibrating.
- When comparing with a datasheet, check whether the figure is typical or maximum, and whether connectors are included.
None of that is exotic. It is just the difference between a measurement and a number.
The uncomfortable part
The uncomfortable part is that a vendor's specification and a customer's measurement are often both correct, and they are simply describing different things. A coaxial termination measured on a bench with 0.1 dB of adapter uncertainty, at 40 °C, with an open reference port, is a different object from the one on the datasheet.
Which means the useful question is rarely "is this part in specification" but "what exactly did each of us measure". Answering that takes ten minutes and saves a month.
Notes from the engineering team at DB Design / MeiXun (Wuxi) Communication Technology, who make coaxial switches, waveguide switches, loads and attenuators, and who have had this conversation more times than they can count.
Top comments (0)