DEV Community

Yang Fei
Yang Fei

Posted on

Switches in an EMC Test Setup: What Actually Breaks the Measurement

An EMC test setup is a measurement instrument built by hand. Antennas, cables, amplifiers, a receiver and — usually — one or more switches that let a single receiver look at several antennas or several paths. The switch is the part of that instrument nobody calibrates properly, which is odd because it is the only part whose loss changes every time a relay closes.

This note is about the specific ways switches degrade an EMC measurement, and what to specify so they do not.

The path loss you calibrated is not the path loss you have

Radiated emissions measurements are made by subtracting a substitution loss from a receiver reading. That loss is the sum of antenna factor, cable loss, amplifier gain and switch loss. The first three are stable. The fourth depends on which contact is closed, how many operations it has seen, and the ambient temperature.

A typical mechanical switch specified at 0.3 dB insertion loss, maximum 0.5 dB, drifts by a few hundredths of a dB per 10 °C. That sounds negligible until the measurement is a 1 dB margin against a limit at 1 GHz and the whole band shifts. Two habits remove the problem:

  • Calibrate the switch in place, in the state used. Not the switch on a bench, but the assembled path, per state. If the switch has four positions, the calibration file has four entries.
  • Record insertion loss versus frequency as measured, not from the datasheet. A 0.6 dB datasheet loss that is 0.45 dB at 200 MHz and 0.8 dB at 1 GHz tilts the whole measurement.

If the switch sits before a low-noise amplifier, its drift is amplified by nothing — but the noise figure of the chain moves with it. For emissions work that is irrelevant; for immunity work at low field strengths it is not.

Isolation and the field you are standing in

An EMC chamber at 10 V/m is a hostile place for a switch. The fields are the point of the test, and any leakage path through a switch housing becomes a direct injection path into the measurement chain that bypasses the antenna entirely.

Shielding effectiveness matters more than the datasheet isolation figure, and the two are not the same thing. Isolation is measured port to port with the housing intact and the fields excluded. Shielding effectiveness is what happens when a 10 V/m field arrives at the seam of the die-cast box.

Practical requirements:

  • Metal housing, gasketed at the control connector. A switch with a plastic-bodied control header is a hole.
  • Filtered or fibre-optic control interface. Every copper control lead entering the chamber is an antenna.
  • Isolation specified with the unused ports terminated, not open.

I have seen a 70 dB isolation switch injected 40 dB down because the control cable ran alongside the antenna feed. The switch was fine.

Reflective switches create the standing wave you then measure

In a radiated setup the switch often sits between the antenna and the preamplifier. A reflective switch in the off state leaves the unselected antenna port open. At the frequencies where the cable run is a significant fraction of a wavelength, that open end produces a standing wave on the feed, and the standing wave changes the effective antenna factor in a way that is neither stable nor documented.

Consequences show up as:

  • Emissions readings that shift when a second antenna is connected but not selected.
  • Resonance-like ripple across a swept measurement, at intervals matching the cable length.
  • Chamber-to-chamber disagreement on the same EUT.

Absorptive switches and 50 ohm terminations on unused ports remove this. The cost is insertion loss and the need to rate the internal loads for the power they will absorb — in an immunity test that can be tens of watts during a sweep.

Contact resistance, and why it matters even at RF

At radio frequencies, a few milliohms of contact resistance is nothing. At the low end of the radiated emissions band — 30 MHz, or in conducted emissions work at 150 kHz — the measurement chain is looking at microvolts and the switch's contact quality starts to appear as noise and as intermittent level jumps.

Reed relays are the usual answer below a few hundred megahertz: sealed contacts, low thermal EMF, tens of millions of operations. Their weakness is that they are not tolerant of hot switching and they are not 50 ohm structures, so return loss at higher frequencies is poor. Mechanical coaxial relays are the opposite trade: excellent RF behaviour, limited life, and contact resistance that rises with wear.

The specification that predicts field behaviour is contact resistance stability over life, not initial contact resistance. Ask for the change after the rated number of operations.

Control noise is the failure mode that looks like an EUT problem

A relay coil is an inductor with a diode across it, driven by a logic line that runs from outside the chamber. When it de-energises, the flyback and the ground bounce couple into the measurement chain, and the result is a broadband burst recorded at exactly the moment the switch changed state. If your test software averages over a sweep, that burst lands in one bin and looks like an emission from the EUT.

Three fixes, in order of preference:

  1. Sequence the measurement so the switch is not operated during a measurement dwell — obvious, and frequently not done.
  2. Opto-isolate the control lines and give each coil a local reservoir capacitor with a short return path.
  3. Use latching relays so the coil is energised only during the change, not for the whole test.

What to write in the requirement

Parameter Why it matters in EMC work
Insertion loss, max over band and temperature Directly subtracted from every reading
Loss repeatability per operation Sets the drift between calibrations
Isolation, min over band, ports terminated Determines leakage floor in a field
Shielding effectiveness of the housing The figure that predicts chamber behaviour
Absorptive or reflective Standing waves and antenna factor
Termination power rating Immunity sweeps can be tens of watts
Contact resistance stability over life Low-band noise and level jumps
Control interface Copper in a chamber is an antenna
Life at rated power, not cold cycles Replacement interval

Two of these — loss repeatability and shielding effectiveness — are almost never on a datasheet and are the two that decide whether the setup is stable enough to defend a 1 dB margin.

A short calibration procedure that catches most of it

  1. Terminate all ports. Measure through-loss of each state against a through connection. Record, do not just note pass/fail.
  2. Repeat step 1 after 1,000 operations. The difference is your drift budget.
  3. Connect a signal generator to one port at the highest level the test will use, and measure leakage at the other ports with a spectrum analyser. Do this inside the chamber with the field generator running, if the setup allows.
  4. Sweep the same path twice, ten minutes apart, with the chamber at working temperature. Any ripple that moves is thermal or mechanical, not the EUT.

Step 4 is the one most laboratories skip and the one that explains the disagreements.

The switches and terminations used in these setups are described under coaxial switches; attenuators are usually needed in the same chain to keep levels inside the receiver's linear range.

Top comments (0)