Field notes from European Microwave Week 2026, London.
At the "RF to Sub-THz Characterization" wall on Keysight's EuMW 2026 booth, a four-port PNA-X was characterising an I/Q mixer, the kind of device found at the heart of every direct-conversion transmitter. The screen showed four channels. One of them, a sweep of the Q-channel drive phase, holds a lesson that applies to anyone who works with quadrature hardware. The best image rejection did not occur at 90°.
The four channels
Every channel displayed the same set of traces. "LO+I/Q" is the wanted sideband, "LO−I/Q" the unwanted image, "Image rejection" the ratio between the two, and "GainHi" the conversion gain.
- Ch1, LO frequency sweep from 6 to 15 GHz.
- Ch2, I/Q frequency sweep from 10 to 500 MHz. Gain and wanted-sideband power are flat up to roughly 400 MHz and then fall steeply. That is the baseband bandwidth of the device or the test path.
- Ch3, Q phase sweep from −180° to +180°. The phase of the Q-channel drive is swept while the I channel is held fixed.
- Ch4, port matches: S11, S22, S33 and S44 from 10 MHz to 15 GHz.
This is the PNA-X's main selling point in this application. Its multiple internal sources can drive the I and Q ports with a controlled relative phase, while the receivers measure both sidebands at the same time. The booth panel listed what makes that possible: four pulse-modulated sources, two-tone stimulus on two ports, two low-noise receivers, receiver amplifiers and preselection, and accessible source loops.
The Q phase sweep: 46.93 dB at −88.2°
The markers on Ch3 read:
- Q phase −90.000°: LO+I/Q = 6.59 dBm
- Q phase +90.000°: LO−I/Q = 6.86 dBm
- Image rejection maximum at −88.200°: 46.93 dB
The physics behind it is straightforward. Driving Q at −90° relative to I selects one sideband, and +90° selects the other. Hence the near-identical powers of about 6.6–6.9 dBm at the two settings. If the mixer's internal LO quadrature split were perfect, image rejection would peak at exactly −90°. It peaks at −88.2°, so the device's own quadrature network is off by about 1.8°. The sweep found the drive phase that compensates for it.
Two quick calculations show why this matters:
- Phase error alone. A 1.8° quadrature error with perfect amplitude balance limits image rejection to about 36 dB. If you drive this mixer with an ideal 90° I/Q signal, that is roughly what you would measure.
- What the 46.93 dB peak means. At the optimum phase, the remaining image is set by amplitude imbalance. 46.93 dB corresponds to an I/Q gain imbalance of only about 0.08 dB.
So the same device delivers about 36 dB or about 47 dB of image rejection, depending on whether the drive compensates for its 1.8° error. In a real transmitter, that compensation is done by I/Q calibration in the baseband or DAC. The measurement tells you how much correction is needed, and how good the result can be once it is applied.
Why sweep the phase instead of measuring at 90°?
A single measurement at nominal quadrature gives one number, and it mixes two error sources together. A phase sweep separates them:
- The position of the peak gives the phase error. Here, 1.8°.
- The height of the peak gives the amplitude imbalance. Here, about 0.08 dB.
Repeat that across LO frequency, as Ch1 does, and you get the frequency dependence of both errors. That is exactly the data a calibration algorithm needs. The image rejection trace on Ch1 varies by tens of dB across 6–15 GHz, a reminder that a correction valid at one LO frequency will not hold across the band.
One practical detail from the status bar
The status line at the bottom of the screen read "WARNING: Receiver b2 overload". With wanted-sideband outputs near +7 dBm, this is not surprising. It is still worth a note for anyone reproducing the measurement: a compressed receiver reduces the apparent level of the stronger sideband, which corrupts the image rejection ratio. Add attenuation ahead of the receiver, or reduce drive, until the warning clears, before trusting a 47 dB result.
Takeaway
The most useful number on this screen was not the 46.93 dB. It was the −88.2° where it occurred. That 1.8° offset is the device's quadrature error, and it alone costs about 10 dB of image rejection if left uncorrected. Sweeping the drive phase turns image rejection from a pass/fail number into a diagnosis: how much phase error, how much amplitude error, and how much correction the system will need.
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. The DUT identity and the exact test configuration were not confirmed with the exhibitor. Image rejection figures derived from the markers are my own calculations using the standard quadrature-error model.
