Field notes from European Microwave Week 2026, London.
An average power reading gives you one number for an entire pulse train. For a pulsed amplifier, though, the important behaviour happens at the edges: how fast the output rises, how flat the top stays, and how cleanly it switches off. At EuMW 2026, Anritsu had a compact bench at its booth set up to look at exactly that. It used three synchronised peak power sensors around a bi-directional coupler, with a live time-domain display above the bench. Below I walk through what was on the screen, and why the most interesting part of the picture is also the easiest one to misread.
The setup
Reading the bench from left to right:
- Source: an Anritsu MG36221A Rubidium analog signal generator. Its status bar showed the internal rubidium reference active ("REF INT RB"). Pulse modulation was set to Internal Pulse with a four-pulse train: each pulse 100 µs wide, with 200 µs delays between pulses.
- DUT: an amplifier from Microwave Amps, a Bristol-based UK manufacturer, mounted together with a large finned heatsink assembly.
- Sampling: a Narda Model 3022 bi-directional coaxial coupler, rated 1.0–4.0 GHz. Its printed coupling chart is centred around 20 dB. The coupler's band also limits the usable test frequency for this setup.
- Sensors: two Anritsu MA24418A and one MA24406A microwave peak power sensors. Two of them sat on the coupler's coupled ports, and one was on the main line.
- Host: an Anritsu ML2439A power meter, with the sensors connected to its front-panel "Channel Sync" ports.
One detail is easy to overlook: the three sensors are not identical. According to Anritsu's published data, the two models differ in ways that matter for pulse work:
| Parameter | MA24418A | MA24406A |
|---|---|---|
| Frequency range | 50 MHz – 18 GHz | 50 MHz – 6 GHz |
| Dynamic range, average | −34 to +20 dBm | −60 to +20 dBm |
| Dynamic range, pulse | −24 to +20 dBm | −50 to +20 dBm |
| Video bandwidth | 70 MHz | 195 MHz |
| Rise time (fast mode) | 5 ns | 3 ns |
| Sampling | 100 MSa/s continuous, 10 GSa/s effective (both) | |
In the 18 GHz sensor, you trade about 26 dB of pulse-mode dynamic range, and some video bandwidth, for frequency coverage. If you put both types on the same time axis, every comparison between channels inherits that difference.
What the screen showed
The meter was in Pulse Mode, triggered, sampling at 100.00 MS/s and 1.0000 k sweeps/s. The display spanned −240 µs to +260 µs at 50 µs/div. CH1 and CH2 were set to 5 dB/div, and CH3 to 10 dB/div. Two markers bracketed the pulse: marker 1 sat in the off-state before the rising edge, and marker 2 sat just after the falling edge. I photographed the display twice, a few minutes apart. Here are both readouts:
| Readout | CH1, snapshot A | CH1, snapshot B | CH3, snapshot A | CH3, snapshot B |
|---|---|---|---|---|
| Mk1 level (pre-pulse) | −30.440 dBm | −29.576 dBm | −Low− | −Low− |
| Mk2 level (post-pulse) | −23.882 dBm | −23.738 dBm | −15.628 dBm | −15.618 dBm |
| Marker-gated average | −1.005 dBm | −1.007 dBm | 18.736 dBm | 18.734 dBm |
CH2 showed no marker readout in either snapshot, so two traces were active on the display: CH1 in yellow and CH3 in violet.
Reading 1: the 19.74 dB that did not move
Take the difference between the two gated averages: 18.736 − (−1.005) = 19.741 dB in snapshot A, and 18.734 − (−1.007) = 19.741 dB in snapshot B. The absolute values drifted by 2 mdB, but the ratio between the channels did not change in the third decimal place.
That is what synchronised acquisition should give you. Both channels see the same pulses within the same gate, so slow source or DUT drift cancels in the ratio. It is also very close to the coupler's nominal ~20 dB coupling. The simplest reading is that CH1 samples a coupled port and CH3 reads the main line. I could not confirm the channel mapping from the photos, so treat that as an interpretation, not a fact.
Reading 2: the tail, and why it looks different on each channel
The most eye-catching thing on the display is the falling edge. On CH3 (violet), the trace drops sharply and then decays slowly over roughly the next 100 µs, with small ripples, before it settles. On CH1 (yellow), the trace seems to collapse almost straight into the noise.
The obvious conclusion would be that the two ports see different physics. The numbers say otherwise.
- At marker 2, CH3 reads about −15.6 dBm. If CH1 sits about 20 dB below CH3, it should read around −35 dBm at the same instant.
- −35 dBm is below the MA24418A's published pulse-mode floor of −24 dBm. CH1's actual marker readings, −23.7 to −30.4 dBm in both the off-state and the post-pulse window, sit right at that floor.
- CH3 is displayed at 10 dB/div and CH1 at 5 dB/div. The same decay looks twice as long on CH3's scale.
So the difference in how the tails look is fully explained by a 20 dB offset, a sensor floor and a display scale. You do not need different physics to get it. CH1 probably has a tail too; it is simply below what that channel can see.
That leaves the question of whether the tail on CH3 is real. With this setup, there are three candidates to separate:
- DUT behaviour: drain or bias-network discharge, the timing of gate or supply switching, or thermal recovery in the output stage. For a pulsed PA, all of these are legitimate tail sources.
- Sensor response: MA244xxA sensors offer a fast mode (ns-class rise time) and a standard mode with a rise time around 10 µs. A slow mode visibly stretches a falling edge. Check the mode before you characterise any edge.
- Source or modulation leakage: the RF that the generator leaks in its "off" state during the pulse train, then amplified by the DUT.
The three-sensor arrangement is what lets you separate these. Put the generator output alone on a sensor and you get a baseline for candidate 3. A sensor in fast mode rules candidate 2 in or out. Comparing the forward-coupled and main-line channels after you account for the floor and the scale tells you whether the energy is travelling forward or being reflected.
Why the tail matters
In pulsed systems, especially radar and anything else with a shared transmit and receive path, the receive window opens shortly after the transmit pulse ends. Whatever power the transmitter is still delivering at that moment becomes self-interference in the receiver. The relevant question is not just the peak power or average power. It is how far below the pulse top the output has fallen, and how many microseconds after the edge that happens. You can only answer that with a time-resolved measurement that has enough dynamic range on the channel that sees the tail.
This is where the sensor table above becomes a design decision. If the tail you care about is 40–50 dB below the pulse top, you need a channel with that much pulse-mode dynamic range at the right point in the chain. Otherwise you need to rearrange the chain, for example by using less attenuation ahead of the sensor that watches the off-state.
A short checklist for pulsed amplifier power measurements
- Match sensor floors to the question. Before you compare edges between channels, write down each channel's pulse-mode floor and the expected signal level on that channel.
- Put channels on a common vertical scale before you compare decay shapes on the display.
- Check the sensor's rise-time mode before you interpret any falling edge.
- Use ratios between synchronised channels for stability work. The 19.741 dB ratio held while the absolute values moved.
- Know your coupler. In a coupler-based setup, the coupling factor's variation across frequency and the coupler's directivity often dominate the forward and reflected uncertainty. The sensors usually do not.
- Baseline the source alone before you attribute any off-state energy to the DUT.
Takeaway
The demo was simple: one generator, one amplifier, one coupler and three sensors. It still carries a useful lesson. A multi-channel, time-aligned power measurement is powerful because it lets you compare channels, and that is also where it can mislead you. The stable 19.74 dB ratio shows what synchronisation buys you. The two different-looking tails show that the floor of the channel you are reading can shape what you think you see.
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 photos. Sensor specifications are taken from the manufacturer's published data. Channel-to-port mapping and the DUT's operating parameters were not confirmed with the exhibitor.