EC-22.6 Measuring at Radio Frequency: Calibration, and What Goes Wrong

What this is and why it exists

At these frequencies the cable, the connector and the fixture are part of the circuit.

An uncalibrated measurement describes the whole arrangement rather than the component you meant to test. That is not a small error. A well-matched part can appear badly matched purely because of thirty centimetres of cable.

Calibration is the procedure that removes everything except the thing of interest. It is why a network analyser measurement is trustworthy in a way most laboratory readings are not. The second half of this topic is the list of ways it still goes wrong.

The vocabulary

• Fixture — the board, connectors and cables between instrument and component.
• Calibration standard — a component of known behaviour used to characterise errors.
• Error term — one of the instrument imperfections calibration solves for.
• De-embedding — mathematically removing a known section of the fixture.
• Resolution bandwidth — the width of the filter a spectrum analyser sweeps.
• Sweep time — how long the instrument takes to cover the frequency range.
• Input attenuation — deliberate loss added to protect the analyser's front end.
• Torque wrench — the tool that tightens a connector to a repeatable force.

The mental model

Every connector, cable and length of board between the instrument and the component has loss, delay and reflections of its own. Without correction the instrument reports the sum of all of them. That is the whole reason calibration exists.

Calibration works by measuring known things. Connect a short, an open and a matched load in turn. The instrument then has enough equations to solve for its own error terms and subtract them. Each standard removes specific terms, which is why a partial calibration can be worse than none. It corrects some errors and leaves others, and you no longer know which.

Once the instrument is calibrated to the ends of its cables, the fixture between there and the component remains. De-embedding models that section and removes it too. Getting this wrong shifts phase, and a phase error makes a good component look badly matched, which is a very common false alarm.

A spectrum analyser has three interacting settings. A narrower resolution bandwidth lowers the noise floor and reveals smaller signals. It also slows the sweep, sometimes to the point where a transient signal is missed entirely. Most misleading spectrum measurements come from one of those three settings being wrong for the question being asked.

The instrument also produces its own artefacts. An overloaded front end generates intermodulation products indistinguishable from the ones you are looking for. The standard test is to step the input attenuation and see whether the suspect peak moves. A real signal stays where it is; an artefact changes with the attenuation.

Last, the unglamorous part. A connector tightened by hand gives a different answer each time, and a damaged connector damages every connector it subsequently meets. This costs teams real weeks, and a torque wrench and a connector gauge cost very little by comparison.

What you should now be able to explain or do

• Explain why an uncalibrated high-frequency measurement is nearly meaningless.
• Describe what each calibration standard lets the instrument solve for.
• Say what de-embedding removes and what a mistake in it looks like.
• Choose resolution bandwidth and sweep time for a given measurement.
• Test whether a peak on a spectrum analyser is real or an artefact.
• Say why connector care is a technical matter and not a formality.

Check yourself