Field notes from European Microwave Week 2026, London.
Sometimes the most useful screenshot at a trade show is the one taken at the "wrong" moment. At Keysight's "Master Signal Analysis" station at EuMW 2026, I photographed the 89600 VSA software while it showed a yellow "Calibrating…" banner. The traces behind it were captured over an 8 GHz analysis span centred at 20 GHz, and they show clearly why ultra-wideband signal analysis depends on calibration far more than narrowband work does.
The setup
The bench combined an XA6-series signal analyser, with the VSA software running on it, a Keysight VXG vector signal generator below it, and a UXA signal analyser alongside. The VXG's screen listed several waveform files in "Playing" state. The VSA status bar showed "EXT REF" and "AUTOCAL: OK", and the hardware list referenced Keysight M8100-series and M819x AWGs. The booth panel summarised the station's purpose: analyse ultra-wideband signals, accelerate EVM measurements, and validate high-frequency systems.
What the screen showed
- Spectrum trace: centre 20 GHz, span 8 GHz, RBW 100 MHz, range 30 dBm.
- Time trace: main time from 0 to 38.184 ns, with a time length of 38.28125 ns.
- Measurement state: paused, with "Calibrating…" displayed and "DATA?" flags on both traces.
The spectrum, captured before calibration was applied, is the interesting part. Across the 8 GHz span it rises steadily, by roughly two to three divisions from left to right, and it is punctuated by a series of sharp notches at fairly regular intervals.
Why an 8 GHz span needs calibration
Over a narrow span, an analyser's frequency response is flat enough that a single amplitude correction is good enough. Over 8 GHz at 20 GHz, it is not. Several effects add up:
- Cable and connector loss rise with frequency, so a few metres of cable alone can tilt an 8 GHz span by several dB.
- Mismatch ripple between source, cables and analyser input produces periodic variations. The spacing of the notches relates to the electrical length between the mismatches.
- The analyser's IF and digitiser path has its own amplitude and phase response across the analysis bandwidth.
For a spectrum display, these errors distort levels. For modulation analysis they do much more damage. An equaliser can remove a smooth tilt, but amplitude and phase ripple across the signal bandwidth acts as linear distortion. Whatever the equaliser cannot remove shows up directly as an EVM floor. When the goal is "accelerate EVM measurements" on multi-GHz-wide signals, the quality of the channel correction effectively sets the best EVM the system can report.
That is why wideband analysis systems calibrate the entire path, including cables, using a known reference signal, and apply a complex (amplitude and phase) correction across the full analysis bandwidth. The "Calibrating…" banner on this screen was that process in action.
What to ask about a wideband analysis setup
- Where is the calibration plane? At the analyser input, or at the end of the test cable? Cable effects over 8 GHz are not negligible.
- Is the correction amplitude-only, or amplitude and phase? EVM needs both.
- What is the residual flatness after correction, across the actual analysis bandwidth?
- What is the system's own EVM floor, measured in loopback, before attributing EVM to the DUT?
- How stable is the correction with temperature and cable movement?
Takeaway
An 8 GHz-wide signal at 20 GHz exposes every imperfection in the measurement path: tilt from cable loss, ripple from mismatches, and the analyser's own response. The pre-calibration trace on this screen showed all of them at once. When EVM at multi-GHz bandwidths is the goal, calibration across the full path is not a preliminary step. It is part of the measurement.
Notes: Observations and interpretations are my own, based on what was running at the booth during the show. Screen values are as read from my photo, taken during a calibration cycle, so the traces do not represent calibrated performance. The exact analyser model and the signal being analysed were not confirmed with the exhibitor.