Field notes from European Microwave Week 2026, London.
The banner above one of the Anritsu benches at EuMW 2026 read "Demystifying Scalar & Vector Mixer Measurements". The two words in that title describe very different levels of difficulty. Scalar mixer measurements, meaning conversion loss, match and isolation, are magnitude-only and well understood. Vector mixer measurements, meaning conversion phase and group delay, run into a basic problem. The phase of a frequency-converted signal includes the phase of the LO, so it means nothing unless the LO phase is defined. Two snapshots of the demo screen show why that matters.
The setup
The instrument was an Anritsu MS46644A, the four-port model in the Tensor family, with all four test ports cabled to the bench. A small connectorised three-port device sat on the bench, presumably the mixer under test. An Anritsu MN25243A USB calibration module was also part of the setup.
According to Anritsu's published information, the platform supports scalar and vector mixer characterisation: conversion loss, isolation, phase synchronisation, group delay and gain compression. It has four internal sources and eight receivers, an optional second LO, multiple-source control and measurement wizards, and it handles multi-stage mixers and multiple frequency plans.
The booth card sorted mixer parameters into four groups:
- Scalar parameters: conversion gain, match, and isolation between the LO, RF and IF ports.
- Conversion parameters: conversion phase and group delay.
- Power parameters: IMD, compression, IP3 and noise floor.
- Phase and synchronisation: phase sync, a second LO, and phase noise with an external reference.
Why "vector" is the hard half
A VNA measures phase relative to its own reference receiver. In a frequency-converting device, the output phase is the input phase plus or minus the LO phase. If the LO phase is unknown, the absolute conversion phase is unknown too.
Group delay helps, because it is the derivative of phase with respect to frequency, so a constant LO phase offset drops out. An LO phase that changes between sweep points does not drop out. It shows up as a local spike in group delay that looks exactly like DUT behaviour.
So in a vector mixer measurement, keeping the phase reference continuous across the whole sweep matters as much as the calibration itself. The two usual approaches are a characterised reference or calibration mixer, and phase-synchronised sources sharing a common reference.
Snapshot 1: four traces before correction
The first snapshot showed four traces from 11 to 12 GHz at IFBW 1 kHz. The channel status read "MEAS State", with no correction indicator.
- S11: a regular ripple, with peaks around −24 to −27 dB, nulls between about −37 and −45 dB, and roughly 0.2 GHz between nulls.
- Receiver ratios [B4/1](P1) and [A1/1](P1): about −51 to −56 dB, and a flat −36 to −37 dB respectively.
- S41: magnitude around −14 to −18 dB, and phase as a descending sawtooth.
The S11 ripple is a useful diagnostic in its own right. Periodic ripple in an uncorrected reflection measurement means two reflections are interfering. The ripple period gives their separation, because the one-way delay ≈ 1/(2Δf). A period of about 0.2 GHz works out to roughly 2.4–2.5 ns, which is about what you would expect from a reflection at the far end of a typical test cable. After calibration this ripple should disappear. If it does not, the second reflection sits beyond the calibration plane.
One small detail is worth noting. Near 11.92 GHz, S11 shows a tiny spike, and the S41 phase shows a small step at almost the same frequency. A feature that appears at the same frequency in both reflection and transmission is more likely an instrument-internal transition, such as a band crossing in the sweep, than something the DUT is doing. That is a hypothesis, and it is easy to test: shift the sweep or the LO and see whether the feature moves with the setting or stays fixed in RF frequency.
Snapshot 2: data, memory and the phase steps
Later in the show, the screen showed a single S41 trace in "Data & Memory" mode at 3 dB/div. The operator was set to Data / Mem, and the channel status showed a correction indicator. Two magnitude traces sat roughly 6 dB apart: one ran from about −8 to −15 dB, the other from about −15 to −21 dB. Without knowing what condition was stored to memory, that offset cannot be attributed to anything specific.
The phase trace is more informative. Two separate descending segments give the same slope: about 320° over 0.245 GHz, and about 215° over 0.165 GHz. Both correspond to roughly 3.6 ns of group delay, including whatever electrical length remains between the calibration planes. Two independent segments agreeing like this is a good sign of a stable phase slope.
The jumps between the segments are less tidy. A linear phase that wraps produces only full ~360° jumps. Some jumps here are close to that, at about 300–345°. Others are only about 100–140°. A step that is not 360° is a real phase discontinuity, either in the measured path or in the reference. If you switch this trace to group delay, every such step will appear as a spike.
The plausible causes are the ones that matter in any vector mixer setup:
- An LO or source band transition, or a re-lock, during the sweep.
- A phase reference that is not continuous across the sweep.
- A genuine feature of the DUT.
Photos cannot tell these apart. A second sweep with a shifted LO can.
A short checklist for vector mixer measurements
- Write down the frequency plan first: RF, LO, IF and sideband, and which products fall inside the IF filter.
- Use power calibration at both the RF and IF planes for scalar conversion loss. S-parameter correction alone does not cover a frequency-converting path.
- Establish a continuous phase reference before you trust conversion phase or group delay.
- Inspect the wrapped phase: only ~360° jumps are wraps. Investigate any smaller step before you report group delay.
- Choose the group-delay aperture deliberately. A wide aperture smooths noise, and it also hides discontinuities.
- Verify the LO drive at the DUT's LO port. Conversion loss and phase both depend on it.
- Use data/memory to change one thing at a time: LO level, DUT sample, cable position, or correction on and off.
Takeaway
Scalar mixer measurements are now routine. The difficult part of vector mixer measurement is keeping the phase reference continuous, more than the calibration math. The demo screen showed both sides of that. A clean phase slope gave a consistent ~3.6 ns, and a few phase steps that are not 360° each need an explanation before the group delay can be trusted.
Notes: Observations and interpretations are my own, based on what was running at the booth during the show. Screen values are read from my photos and estimated from the plot grid where marked "about". Instrument capabilities are paraphrased from Anritsu's booth material and published information. The DUT, frequency plan, calibration type and the stored memory condition were not confirmed with the exhibitor.