The sign on the bench said it clearly:
300 GHz Dual Polarization OTA Link.
285 GHz center frequency. ~12.5 GHz bandwidth. 63 Gbps.
At EuMW 2025 in the Netherlands, the Virginia Diodes booth had one of the most technically dense demos on the floor — and one of the most visually striking.
A linear rail running the length of the table. TX module on one end, RX module on the other. A precise gap of free space between them. Two VDI Dual Compact Converters (CCU-D and CCD-D) handling the upconversion and downconversion. Micro Harmonics OMT and Flann horn antennas providing the dual-polarization interface. Blue phase-stable coaxial cables looping back to a Keysight UXR-Series real-time oscilloscope.
On the screen: live OFDM Error Summary. Two streams. Real numbers.
Most people stopped walking when they saw it.
What the hardware was doing
The signal chain had four distinct sections — and understanding each one explains why this demo is harder than it looks.
Generation: A Keysight M8199B arbitrary waveform generator produced the custom OFDM baseband waveform. The control software on the adjacent PC showed the carrier setup — a wideband OFDM signal with the characteristic flat-topped yellow spectral shape, bandwidth approximately 12.5 GHz. This is the signal that encodes the 63 Gbps data.
Upconversion: A Keysight PSG Analog Signal Generator provided the LO reference — display reading 22.416 656 666 667 MHz at 16.00 dB — feeding into the VDI CCU-D (Compact Converter Upconverter, Dual polarization). The CCU-D multiplies the LO frequency up to the 285 GHz carrier, while mixing in the baseband signal. Two of these modules — one per polarization — handled the H and V channels simultaneously.
OTA propagation: The Micro Harmonics OMT (Orthomode Transducer) separated the H and V polarizations into the waveguide structure feeding the Flann horn antenna. The signal propagated across the free-space gap on the optical rail. At the receive end, a second OMT + horn antenna separated the arriving wavefronts back into H and V, feeding the CCD-D (Compact Converter Downconverter, Dual polarization) modules for downconversion.
Analysis: The Keysight UXR1104B Infiniium real-time oscilloscope — 110 GHz bandwidth, 256 GSa/s — captured the downconverted signal, running Keysight VSA software for OFDM demodulation in real time.
The OFDM Error Summary screen showed: → Center frequency: 285 000 000.000 9 Hz → Stream 1 EVM: −25.570 dB / Stream 2 EVM similar → Peak EVM: −24.659 dB → Data EVM: −20.842 dB → Throughput: 18,270 μBps — approximately consistent with the 63 Gbps headline at the demo's operating parameters → Power: −24.979 dBm received
The numbers were not perfect. The EVM values show measurable impairment — expected at this frequency and bandwidth. But the link was closing. The data was flowing. The demodulation was working in real time.
The 2×2 MIMO architecture — what dual polarization actually provides
The "dual polarization" in the demo name is not just a technical flourish. It is the mechanism that achieves the headline data rate.
A standard single-polarization link at 285 GHz with 12.5 GHz bandwidth, using a moderate modulation order, might achieve 20–30 Gbps. To reach 63 Gbps, you need either higher modulation order (which requires much better SNR) or more spatial channels.
Dual polarization provides two orthogonal channels — horizontal and vertical — that can carry independent data streams simultaneously without mutual interference, as long as the cross-polarization isolation is sufficient.
At 300 GHz, maintaining that isolation across a real free-space link is non-trivial: → The OMT must cleanly separate H and V at the waveguide level → The horn antennas must maintain polarization purity across the full 12.5 GHz bandwidth → Any mechanical misalignment between TX and RX antennas introduces cross-polarization coupling — H energy leaking into the V channel, and vice versa → Even a slight rotation of one module relative to the other on the linear rail will shift the polarization basis and degrade isolation
The Micro Harmonics OMT visible in the hardware chain is specifically designed for this — it is a precision passive component that defines the polarization separation quality. The Flann horn antennas, designed for the H-band frequency range, provide the aperture efficiency and polarization purity needed to launch and receive the two streams cleanly.
In the controlled demo environment, this architecture achieves 2× the data rate of a single-polarization link at the same spectral efficiency.
In a deployed system, maintaining that 2× gain requires that the polarization alignment is stable over time, temperature, and mechanical stress.
What the UXR was measuring — and what it was not
The Keysight UXR1104B is the instrument that makes this measurement possible. At 110 GHz analog bandwidth and 256 GSa/s sample rate, it has the raw capture capability to digitize the downconverted sub-THz signal and pass it to the VSA software for real-time OFDM demodulation.
The EVM numbers visible on screen — approximately −25 dB for the OFDM streams — reflect the combined impairment of the entire signal chain: the AWG's signal quality, the LO phase noise through the multiplication chain, the upconverter and downconverter linearity, the OTA path, and the UXR's own noise floor.
What the EVM does not capture:
→ Long-term drift: The measurement was taken at a specific moment, in a stable lab environment at the show. EVM 10 minutes later, after the system has thermally stabilized to a new equilibrium, may be different.
→ Mechanical sensitivity: The linear rail on the bench allowed the TX-RX distance and alignment to be adjusted. The optimal alignment was set before the demo ran. What happens when the rail vibrates — from show floor foot traffic, from HVAC airflow, from the cable tension shifting slightly — is a dynamic quantity the static EVM snapshot doesn't show.
→ Phase noise under multiplication: The PSG LO at ~22.4 GHz is multiplied by a factor to reach 285 GHz. Phase noise multiplies as 20·log(N) — for a ×12 multiplication, that's 20·log(12) = +21.6 dB of phase noise degradation relative to the PSG's native performance. The EVM reflects the result, but the sensitivity to LO phase noise becomes a primary design constraint for any deployed 300 GHz system.
→ Cross-polarization under stress: The isolation between H and V streams was sufficient for the demo. Under real deployment conditions — outdoor temperature cycling, wind loading on antenna structures, long-term connector wear — the cross-polarization rejection will degrade.
The linear rail as a statement
One design choice in this demo that deserves specific attention: the linear rail on which the TX and RX modules are mounted.
It is adjustable. You can change the TX-RX distance. You can adjust the alignment precisely.
This is necessary for a research demo — it allows the team to find the optimal alignment and demonstrate the system at its best.
But it is also, implicitly, an acknowledgment that alignment matters enormously at 300 GHz.
The free-space wavelength at 285 GHz is approximately 1.05 mm. A pointing error of 1° at 300 GHz produces a displacement of approximately 17.5 mm per meter of path length. At 1 meter distance, 1° misalignment means the beam is pointing 17.5 mm off-center.
The Flann horn antennas have a narrow beamwidth — at 300 GHz, a typical horn gain of 20–25 dBi corresponds to a half-power beamwidth of roughly 10–15°. So 1° of misalignment at 1 meter is survivable, but only barely.
At 10 meters? At 100 meters? The alignment tolerance tightens dramatically.
The linear rail makes the demo work reliably.
A deployed system needs a different answer to the same alignment problem.
63 Gbps: what the number means, and where it comes from
The 63 Gbps headline is technically accurate for the demo configuration. Let's verify the arithmetic:
→ 12.5 GHz bandwidth × 2 polarizations = 25 GHz of total spectral resource → 63 Gbps ÷ 25 GHz = approximately 2.52 bits per Hz spectral efficiency → This corresponds roughly to 8-QAM or low-order 16-QAM with OFDM overhead factored in
This is not an extraordinary spectral efficiency — 5G NR achieves similar or higher. What is extraordinary is achieving this at 285 GHz over a real free-space link, with real hardware, in real time.
The data rate is a function of both the bandwidth available (which is vast at sub-THz frequencies) and the modulation order achievable (which is limited by the SNR and EVM floor of the system). At 300 GHz, the available bandwidth per channel is orders of magnitude larger than at mmWave frequencies — which is why even modest spectral efficiency yields headline-grabbing data rates.
What this demo represents in the 6G timeline
The VDI demo at EuMW 2025 sits at a specific point in the sub-THz technology development arc.
It is past the "proof of concept" phase — this is not the first 300 GHz link to be demonstrated. VDI and similar groups have been demonstrating sub-THz links for several years.
It is not yet at the "system prototype" phase — the hardware on the bench cannot be packaged and deployed without significant engineering work.
It occupies the space between: a fully functional, characterized, repeatable measurement platform that demonstrates the key capabilities the industry needs to understand before designing the next generation of systems.
The critical technical questions it helps answer: → What EVM is achievable at 285 GHz with current hardware? → Can dual-polarization MIMO close a link at sub-THz frequencies? → What are the dominant impairment mechanisms at 12.5 GHz bandwidth?
These are not marketing questions. They are engineering questions that have to be answered before 6G standards bodies can write specifications.
The demo is part of that answer.
All photos: Thomas · @SignalByThomas