Under a cold-switching specification a coaxial switch is rated for one to five million operations. Under a hot-switching specification the same switch may be rated for fifty thousand. The hardware is identical. The difference is whether the contact opens while RF power is still flowing through it.
Almost every argument about switch life in a test rack comes down to this one clause, and it is often the least visible line on the datasheet.
What actually happens at the contact
An electromechanical coaxial switch makes contact through a spring-loaded blade pressed against a centre conductor. When the circuit is dead, the blade closes onto a clean surface and the only wear is mechanical. When the circuit is live, the blade approaches the conductor while a voltage difference still exists across the gap. Below roughly ten volts the arc does not form and the contact is effectively cold. Above that, and especially above a few tens of volts, a micro-arc erodes a small crater at the contact point on every operation.
The crater matters for two reasons. It raises contact resistance, which shows up as insertion loss creeping up over months. And it roughens the surface, which degrades the repeatability of the return loss. A switch with a million cold operations behind it can still measure like new. A switch with thirty thousand hot operations at 50 W may already be out of specification on isolation, because the eroded contact no longer seats flat.
Manufacturers know this, which is why the two ratings differ by a factor of twenty or more. What they cannot tell you is which one applies to your rack, because that depends on how the software drives the switch.
The sequence is the specification
The sequence that keeps a switch in its cold-switching rating is simple:
- Turn the RF source off, or set it to zero output.
- Wait for the source to actually settle. On many synthesizers this is longer than the command latency, and on amplifiers with a large output capacitor it can be tens of milliseconds.
- Switch.
- Wait for contact bounce to finish. A few milliseconds for a relay, microseconds for a PIN diode.
- Turn the source back on.
In practice most racks skip steps one and four. The switch driver fires as soon as the instrument accepts the command, and the source is still ramping down. That is a hot switch, and no amount of careful hardware selection fixes it.
The measurable symptom is a switch that meets its insertion loss specification when new and drifts by a few tenths of a dB after a year of production testing, with the drift concentrated on the channels that carry the highest power.
Where hot switching is unavoidable
Some architectures cannot avoid hot switching. A switch matrix that reconfigures while a carrier is running, a redundancy switch that has to move on failure, a calibration path that has to switch during a sweep: in these cases the specification has to be chosen for hot switching from the start, and the derating applied to power and to life.
For a switch rated 100 W cold, a hot-switching rating of 10 to 20 W at the same frequency is typical for a relay with a good contact design. For a PIN diode switch there is no arc at all, so the limit becomes the diode's power dissipation and the bias network, not contact erosion. That is one of the few real advantages of solid-state switching that survives contact with a test plan.
What to write into the test software
Three changes cover most of the exposure, and none of them costs money:
- Put the source-off command before the switch command in the driver, not after it, and let the driver wait a fixed settling time that you have measured rather than guessed.
- Add a per-channel operation counter to the log. A switch that is 60 percent through its hot-switching life should be scheduled for replacement, not discovered by a failed measurement.
- Record power level with the counter. A channel that ran at 2 W for a year and then moved to 50 W has a completely different remaining life, and without the power column nobody can tell.
Choosing the part for the sequence you actually run
If a rack genuinely cold-switches, a standard electromechanical coaxial switch from SPDT to SP12T or DPDT is the right choice, and the failure mode you will meet is mechanical wear after millions of operations. If the rack hot-switches, either accept the reduced life and buy a PIN diode switch, or derate an electromechanical part and treat it as a consumable.
The same logic applies at higher frequencies where the switch is a waveguide switch: the actuator and the contact geometry change, the arc problem does not. And whatever the switch type, the unused throws still need coaxial terminations whose power rating is chosen for the same hot or cold assumption, because a termination that absorbs a live carrier during reconfiguration is doing exactly the job the switch was derated to avoid.
One last line worth writing into the RFQ: ask for the hot-switching life at the frequency and power you will actually use, not the headline cold-switching figure. The answer tells you more about the contact design than any other number on the sheet.