DEV Community

Yang Fei
Yang Fei

Posted on

Connector torque and repeatability: where 0.2 dB actually comes from

You tighten an SMA connector to what feels right, take a measurement, swap the cable, tighten it again, and the reading moves by 0.2 dB. Then you spend an afternoon hunting a "bad" component that was never bad. The variation came from the connector, not the part.

This is the most common source of unexplained scatter in RF work above about 8 GHz, and it is almost always a torque problem rather than a component problem.

Why torque matters more than people expect

An SMA interface is a precision lapped pair held together by thread friction. The centre pin contact depth is only a few tenths of a millimetre. Under-torque, and the outer conductor contact is not fully seated: you get a small series inductance and a reflection that grows with frequency. Over-torque, and the dielectric is compressed, the pin may bottom out, and the connector is permanently damaged — it will still mate, it just will not repeat.

For SMA the specified value is usually 0.8 to 1.1 N·m (7 to 10 in-lb). For 3.5 mm and 2.92 mm connectors, the range narrows and the consequences of over-torque get more expensive. For N-type, 1.4 to 1.7 N·m. For 2.4 mm and 1.85 mm, the parts are small enough that a hand-tight guestimate is not a measurement.

The practical rule: use a torque wrench, and use the same one. Repeatability comes from consistency, not from hitting the nominal value precisely.

The three mistakes that produce the 0.2 dB

1. Torquing the wrong side. Rotating the connector body instead of the coupling nut drags the cable, twists the braid and changes the launch geometry. Always hold the body still and turn the nut.

2. Mixing connector sexes and grades without checking. A 2.92 mm male will mate with an SMA female and "work". It will also wear the SMA female's outer conductor, because the 2.92 mm design has a smaller air gap by specification. After a few dozen matings the SMA part is out of tolerance and every subsequent measurement drifts. This is how a calibration kit silently degrades.

3. Reusing adapters as if they were transparent. Every adapter adds two interfaces. A "known good" adapter that has been in a drawer for two years has likely been over-torqued at least once.

What to do about it in practice

  • Torque every interface the same way, every time. Write the number on the bench.
  • Use a torque wrench with a break-over action so the value does not depend on wrist feel.
  • Keep a log of mating cycles for anything above 18 GHz. Most precision connectors are rated for 500 to 1000 matings, not infinite.
  • When a measurement disagrees with expectation, re-mate the connector before blaming the DUT. Re-mating twice and seeing the value move is diagnostic information, not a nuisance.

Where this shows up in a switch matrix

A switch matrix multiplies interfaces. Eight ports, plus the DUT, plus the instrument, plus the calibration reference — a single measurement path can cross six connectors. If each contributes 0.05 dB of random variation from inconsistent torque, the path-to-path repeatability of the matrix is 0.3 dB before any switch repeatability is counted.

That matters because people usually attribute matrix repeatability problems to the switches. In a well-built matrix, the switch is often the most repeatable element in the chain: a good SPDT holds insertion loss within 0.05 dB over a million cycles. The connectors are the loose variable.

If you are building the matrix from scratch, budget the connector interfaces explicitly, and specify the switch repeatability separately from the cable and adapter repeatability. On our coaxial switch assemblies the repeatability figure is quoted at the connector reference plane, which is only meaningful if the mating torque is controlled.

A two-minute check worth doing

Take your worst-behaving measurement path. Measure it five times, re-mating one connector between each measurement with the same torque. If the spread is under 0.05 dB, your connectors are fine and the problem is elsewhere. If it is over 0.1 dB, you have found the culprit — and it is probably a worn adapter, not the instrument.

The same logic applies to terminations. An unterminated port is an open circuit and reflects everything back; a termination that is not fully seated behaves like a partly open port. When a coaxial load measures worse than its datasheet, check the mating before you check the resistor.


Yang Fei works on RF and microwave switching hardware at DB Design / MeiXun (Wuxi) Communication Technology, where coaxial and waveguide switches, loads and attenuators are built for test, radar and satellite systems.

Top comments (0)