The center frequency on the screen read: 140.171752297 GHz.
Not 140 GHz. Not "approximately 140 GHz."
140.171752297 GHz — seven significant figures, displayed in real time on a Keysight UXA Signal Analyzer N9042B at the microwave show in the Netherlands.
That number alone tells you something about where RF measurement is heading.
What was on the bench
Left side: Keysight VXG Vector Signal Generator — a two-channel instrument capable of generating complex modulated signals up to millimeter-wave frequencies. The display showed the signal chain configured with I/Q modulation, RF output parameters, and fading settings. Status line: "REF INT Locked" and "Channel 1: System Alignment Recommended" — the instrument mid-calibration cycle during the demo.
Center: VDI WR 6.5 frequency extender, Model CCD, S/N: CCD 390. The WR 6.5 designation refers to the waveguide aperture size — 6.5 thousandths of an inch — corresponding to the D-band, 110–170 GHz. Two blue coaxial cables feeding in from the VXG. Gold WR 6.5 waveguide port on the output side. A small integrated fan on the bottom face for thermal management.
Right side: Keysight UXA Signal Analyzer N9042B, 2 Hz to 50 GHz native — extended to 170 GHz via a second VDI module on its input port. The UXA is Keysight's highest-performance signal analyzer in the X-Series family.
The signal path: VXG generates a modulated signal → upconverted by VDI extender to D-band → transmitted → received and downconverted by second VDI module → analyzed by UXA.
What the UXA screen was showing
The UXA display was divided into four panels, all running simultaneously:
Top left — IQ constellation diagram A 16-QAM constellation, showing the 16 symbol states as clusters of yellow-green points. The cloud density around each point reflects phase noise, amplitude noise, and residual EVM. Visually, the clusters were tight — indicating good signal quality at 140 GHz — but not perfect. The spread visible around each symbol is real degradation, not display scaling.
Top right — EVM vs. symbol A bar chart showing error vector magnitude over the captured symbol sequence. The bars showed consistent but non-zero EVM — the kind of flat EVM profile that indicates a systematic noise floor rather than burst errors or intermittent problems.
Bottom left — spectrum view The signal spectrum centered at 140.171752297 GHz with instantaneous bandwidth of approximately 1 GHz. The spectral shape showed the characteristic flat top of a well-filtered 16-QAM signal, with rolloff shaped by the Root Raised Cosine filter (alpha 0.25, visible in the Meas Setup panel).
Bottom right — measurement summary table Key values readable from the screen: → Modulation Format: 16-QAM → Symbol Rate: 500.000000 MHz → Points/Symbol: 10 → Meas Filter: Root Raised Cosine Y → Ref Filter: Raised Cosine → Filter Alpha: 0.25 → TX Power: −27.31 dBm → EVM (RMS): 1.36 Nrms / Phase Error: 0.84 Nrms / Freq Error: 3.22 %rms → IQ Offset: −75.46 dB → Quad Error (MER): 34.56 dB → Gain Imb.: 0.08°
What 16-QAM at 500 MHz symbol rate means at 140 GHz
16-QAM encodes 4 bits per symbol — meaning this link was carrying 2 Gbps of raw data rate (500 MHz × 4 bits/symbol).
That is not a demonstration signal. That is the data rate profile of a real D-band backhaul link.
5G mmWave network densification requires high-capacity point-to-point links between base stations. D-band (110–170 GHz) is one of the primary candidate bands for this application — large available spectrum, short-range high-capacity links, minimal licensing friction in many jurisdictions.
The WR 6.5 waveguide dimension defines the propagation mode. At 140 GHz, the free-space wavelength is approximately 2.14 mm — shorter than a grain of rice. At this scale:
→ A connector torqued 1 Nm too tight changes the waveguide cross-section measurably → Surface roughness of the waveguide walls begins to add significant resistive loss → Thermal expansion of the waveguide assembly shifts the passband by tens of MHz between room temperature and operating temperature
None of these effects were visible in the constellation diagram on the UXA screen. Which is the point.
The measurement setup removes exactly what a real system must survive
The VXG and UXA in this demo were sitting 40 cm apart on a flat white table. The VDI extender was connected with short, pre-characterized blue coaxial cables of known length and insertion loss. The reference clock was shared between source and analyzer — the REF INT Locked status on the VXG means both instruments were phase-locked to the same 10 MHz reference.
Phase-locking the reference removes the single largest degradation factor in any real wireless link: oscillator phase noise across a physically separated transmitter and receiver.
In a real D-band backhaul deployment: → Transmitter and receiver are 100–500 meters apart → They have independent local oscillators with their own phase noise profiles → Wind, thermal gradients, and mechanical vibration cause antenna pointing variations → Atmospheric conditions at 140 GHz — particularly water vapor absorption — vary minute to minute
The EVM measured in this demo was already non-zero. That non-zero EVM exists even with all the real-world degradations removed. It represents the floor of what the instrument chain can achieve under the best possible conditions.
A real deployed system will have higher EVM than this. The question is: by how much, and does it still close the link budget?
This is precisely what this class of test equipment is designed to answer — but in a controlled, systematic way. You characterize the instrument floor first. Then you add impairments one at a time. Then you understand which ones dominate in your specific deployment scenario.
→ The clean constellation is the starting point, not the endpoint.
The VXG: what the left side of the bench was doing
The Keysight VXG is a two-channel vector signal generator introduced as the successor to the MXG and EXG families for complex signal generation requirements. Its relevance here:
→ Two phase-coherent RF channels allow simulation of multi-antenna transmit scenarios (MIMO, beamforming test) → Fading simulation built into the hardware — the "AWGN Off / Fading Off" settings visible on screen indicate this demo was running without intentional impairments → Baseband I/Q generation with symbol rates up to the instrument's bandwidth limit
The VXG was generating the 16-QAM signal at an intermediate frequency, with the VDI extender handling the upconversion to 140 GHz. The extender is not a simple frequency multiplier — it contains a local oscillator, a mixer, and filtering stages that determine the conversion gain and noise figure of the upconversion chain. Each of these contributes to the EVM seen at the UXA.
Why D-band, and why now
The choice of 110–170 GHz for this demo was not arbitrary.
The ITU and regional regulators in Europe, the US, and Asia have been allocating D-band spectrum for licensed point-to-point and point-to-multipoint fixed links. The European Commission's decision to harmonize 92–114.25 GHz spectrum in 2019, and ongoing work extending into the full D-band, has created a commercial pathway that didn't exist five years ago.
The technology drivers: → 5G backhaul density — each small cell needs a high-capacity return link; fiber is not always available → 6G research — sub-THz links (100–300 GHz) are a primary 6G physical layer candidate → Automotive sensing — next-generation radar operating above 100 GHz for higher resolution at shorter range → Satellite communication — high-throughput satellite terminals operating in adjacent bands
The measurement challenge in all of these: you need to characterize signal quality at frequencies where almost nothing was standardized a decade ago. The calibration practices, waveguide standards, connector interfaces, and modulation analysis methods are all being developed in parallel with the applications themselves.
The UXA + VDI extender combination is one of the few commercially available solutions that can perform traceable, calibrated modulation quality analysis at 140 GHz — which is why it was on the demo table at the microwave show.
What the IQ offset number tells you
One readout worth highlighting: IQ Offset: −75.46 dB.
IQ offset measures the DC component at the origin of the constellation — a non-zero IQ offset appears as a carrier feedthrough at the center frequency, visible as a fixed point at (0,0) in the constellation diagram.
−75.46 dB is an extremely good number. It means the carrier leakage is 75 dB below the signal power — essentially invisible in the constellation.
This reflects the quality of the VXG's internal I/Q modulator calibration. In a real transmitter chipset at 140 GHz — whether a silicon-germanium RFIC or a GaN MMIC — IQ offset values are typically 30–50 dB, not 75 dB. The difference is test equipment grade versus product grade hardware.
→ The demo is showing you what a perfect transmitter would look like. → Your actual device will not look like this. → But now you have a reference to measure against.
The value of the demo, correctly understood
Demonstrations like this are sometimes criticized as "unrealistic" — controlled environments, matched reference clocks, short cable runs, optimized settings.
That criticism misses the point.
The value is not that this represents a deployed system. The value is that it establishes a calibrated reference under known conditions. When you take this system and deliberately degrade it — add phase noise, remove the shared reference, introduce antenna misalignment, increase the cable length, vary the temperature — you can measure exactly how much each impairment costs you in EVM.
That systematic decomposition of impairments is how you design a real system that closes the link budget.
The constellation on the UXA screen at 140 GHz was not showing you what D-band looks like in the field.
It was showing you where D-band measurement starts.
All photos: Thomas · @SignalByThomas