Field notes from European Microwave Week 2026, London.
Characterising an amplifier has traditionally meant several setups: a VNA for S-parameters, a power sweep for compression, a second source and a spectrum analyser for two-tone intermodulation, and some re-cabling and re-calibration between them. At EuMW 2026, Anritsu's booth demo under the banner "Simplifying Complex Amplifier Characterization" made the opposite case. A single four-port instrument, one connection to the DUT, and four measurement channels were all running on the same screen.
The instrument was the new Tensor VNA, an MS46644A. Below I go through the four channels one at a time. In each one, the screen shows something real, and it also shows something that is easy to over-read.
The instrument
According to Anritsu's published information, the MS46644A is a four-port VNA covering 10 MHz to 43.5 GHz. It uses a source-per-port architecture and a specified dynamic range of 147 dB. For amplifier work, Anritsu positions it as a single-setup solution for gain, compression, AM/AM, AM/PM, THD and two-tone IMD. Anritsu markets the Tensor platform as the first VNA with a built-in AI assistant, which configures measurements from natural-language requests. On the demo screen, the toolbar showed an "AAI" button and a Wizard panel listing an "Amplifier Measurement Wizard" and a "Mixer Wizard".
On the bench, all four test ports were cabled. Each port on the front panel had its own row of direct-access connectors. An Anritsu MN25243A USB calibration module sat on top of the instrument.
Channel 1: 0.4 dB of compression at 0.05 dB/div
The top-left plot is the classic compression view: S31 transmission against input power at a fixed 5 GHz. The trace header reads "S31 Trans LogMag RefLvl 12.03 dB Res 0.05 dB/Div". The channel line reads "Eff Pwr −10 dBm to 5 dBm, 5 GHz, IFBW 1 kHz".
The curve looks dramatic. It starts flat and then bends sharply downward towards the right edge. Now read the axis. The gain starts at approximately 12.19 dB at −10 dBm and ends at approximately 11.79 dB at +5 dBm. That is a total compression of about 0.4 dB across the whole 15 dB drive sweep. At 0.05 dB/div, a 0.4 dB change fills most of the plot.
Three engineering points follow from this:
- The 1 dB compression point was not reached in this sweep. Reading the plot, the 0.1 dB point falls at about −1 dBm input and the 0.2 dB point at about +2 dBm. If you need P1dB, the drive range has to extend further, or the result is an extrapolation.
- The compression criterion is a choice, not a constant. For an amplifier with soft compression, P0.1dB and P1dB can be several dB apart. A data sheet number only means something if you know which criterion was used.
- At this resolution you also see the instrument. Near −4.5 dBm there is a small step of about 0.01 dB in an otherwise smooth curve. It could be the DUT, but at this level a source level-range change or a receiver setting switching over is just as likely. Before you attribute a 0.01 dB feature to the amplifier, repeat the sweep with different source attenuation and see whether the feature moves with the setting.
The x-axis is labelled "Eff Pwr", meaning effective power. That is a reminder that a compression measurement on a VNA depends on a source power calibration at the DUT reference plane, not just an S-parameter calibration. If the power axis is off by 0.5 dB, every compression point you report is off by 0.5 dB too.
Channel 2: gain at compression across frequency
The top-right plot shows "S31@CP", the transmission at the compression point, swept from 5 to 8 GHz at 10 dB/div. At 5 GHz it reads approximately +12 dB, which is consistent with Channel 1. It then falls steadily to roughly −2 dB at 8 GHz. Whatever this DUT is, it is clearly band-limited towards the top of this range.
What the screen does not show is the compression criterion behind "CP". For a frequency-swept compression measurement, that criterion, together with how the instrument searches for it at each frequency point, defines what the trace means. When you read a plot like this, the first question to ask is "compression at what?"
Channel 3: IP3 across frequency, and a deliberate non-reading
The bottom-left channel, labelled "Ch3 (IMD)", sweeps from 5 to 10 GHz at 10 dB/div. Its header lists four parameters: GAIN1, OIP3, IIP3 and PWR3. Three traces are visible, roughly parallel, and each one falls by about 8–10 dB across the band.
I am deliberately not quoting values for these traces. From the photo I cannot reliably match each colour to its label, and at least one parameter is either hidden behind another trace or off-scale. That matters because the readings are tied together: input-referred and output-referred intercept points differ by the gain, so a value assigned to the wrong trace produces a confident but wrong conclusion. If you photograph a screen like this at a show, photograph the marker table as well.
Channel 4: the two-tone spectrum
The bottom-right plot is a spectrum view inside the same IMD channel. It runs from 4.997660156 GHz to 5.002339844 GHz, a span of about 4.68 MHz centred on 5 GHz, at 20 dB/div with IFBW 1 kHz.
It shows ten discrete lines, evenly spaced at roughly 0.5 MHz and symmetric about the centre, with levels falling away from the middle. The natural interpretation is a two-tone stimulus plus its odd-order intermodulation products, IM3 through IM9 on each side. The noise floor rises towards the centre of the span. That is consistent with source noise skirts around closely spaced tones, which in a close-spaced two-tone test often sets the practical floor for the higher-order products.
I am not reading the line levels off this plot either. The axis is in dB rather than dBm, the tone power for the IMD channel is not visible, and at 20 dB/div a photo is not a measurement. There is also a general point about any swept-receiver spectrum view: a CW line only shows its true peak if a measurement point lands within the IF filter. With a 1 kHz IFBW over a 4.68 MHz span, you need on the order of 4,700 points before every line is guaranteed to be caught at its peak. With fewer points, the lines that fall between points read low, and the IM ladder becomes partly a sampling artefact.
What "simplifying" actually simplifies
The demo makes a real point. Compression, gain across frequency, intercept points and the IMD spectrum all came from one connection on one instrument, without re-cabling between them, so the measurements share the same reference planes. A built-in wizard and an AI assistant can take much of the setup work out of that.
The decisions this article kept running into are a different kind of problem, though:
- Which compression criterion to use, and over what drive range.
- Whether the power axis is calibrated at the DUT plane.
- Whether a 0.01 dB feature belongs to the DUT or to the instrument.
- Which trace belongs to which parameter.
- Whether the spectrum view has enough points for its IF bandwidth.
None of these is a setup step. They are judgement calls. Tools can make the measurement faster, but the engineer still has to own the interpretation.
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 marked as approximate where read from the plot grid. Instrument specifications and product claims are taken from the manufacturer's published information. The DUT and the compression criterion used in the demo were not confirmed with the exhibitor.