The software-defined isolator without a magnet.

Every transmitter needs an isolator between its amplifier and its antenna, and for seventy years that has meant a ferrite disc and a magnet. Onuska Industries makes the same one-way behaviour from a printed transmission line and timing: a switching pattern that travels along the line lets signals through one way and blocks them the other, and because the pattern is firmware, the same board retunes. Given a different schedule, the line does other things a fixed circuit cannot; the physics section shows one of them live.

ONE-WAY PROPAGATION conceptual animation

traveling modulation forward wave — passes reverse wave — rejected

The moving stripes are the switching pattern sweeping along the line. A signal travelling with the stripes rides them and passes; a signal travelling against them is thrown out of the band and blocked. The animation sends one each way in turn. This is a schematic of the mechanism, not a measurement; measured performance for the real board is published in the product section when it exists, and not before.


01 · The product

Magnet-free non-reciprocity.

Isolators and circulators — the one-way valves of RF — have required magnetized ferrite for seventy years. A time-modulated line breaks reciprocity with a traveling switching pattern instead: no ferrite, no magnet, no fixed function — the same hardware retargets by schedule. Our first module targets VHF/UHF, where ferrite is at its worst.

Status · 4 September 2026

The first module is the 433 Alpha: a 96-cell printed line on a two-layer board, pumped by an on-board microcontroller — no ferrite, no magnet, no rings or resonators. Design review of the fabrication-ready board is complete (v0.9d); first articles are pending order. Predicted performance is below, labeled as a prediction until the first article is measured. The next design iteration targets under 6 dB of loss.

433 ALPHA — EVALUATION UNIT predicted · not yet measured

Rendered top view of the 433 Alpha board: four rows of 24 cells, on-board pump filters and controller, SMA ports
25–36 dBpredicted isolation after the firmware tunes pump frequency and varactor bias (independent SPICE model of the whole board; the tolerance Monte-Carlo is more optimistic)
10–16 dBpredicted insertion loss in the pass direction; the range spans the coil-Q uncertainty that the first measurement resolves
> 13 dBpredicted return loss, both ports, pump on
3 modelsa time-domain solver, an independent frequency-domain solver and a SPICE transient of the whole board; the first two agree to 0.3 dB, SPICE sets the isolation range; every number here is a prediction, replaced by VNA data when it exists
166 × 118 mmtwo-layer FR4, 96 varactor cells in four rows (selectable to 48 for scaling measurements), ~580 placed parts
USBpower and control through the on-board Raspberry Pi Pico; pump frequency, bias and on/off over serial, no external source
5 × SMAIN, OUT, a line tap, and ANT / RX OUT for circulator mode with a second board and the on-board hybrid
≈ −37 dBmpredicted pump residue at the RF ports near 55 MHz, behind on-board port filters — stated so an evaluator knows what the analyzer will show

Mechanism: a varactor-loaded transmission line with a traveling-wave pump; the direction the pump travels is the direction that passes. What gets measured first: pump-off transmission (which pins the coil quality), then pump-on transmission both ways, return loss, isolation against frequency, and spur levels — all published here with dates when they exist.

Evaluation partners. First units go to small-satellite and UHF radio teams who want a flat, magnet-free isolator on the bench and are willing to tell us what they measure. If that is you, write.

Ask about an evaluation unit →

Why the low band

The wavelength limit

A junction circulator's size is set by the wavelength itself — at VHF/UHF that means fist-sized and hundreds of grams, a floor no engineering removes. The lumped-ferrite workaround trades away loss, power handling, and bandwidth — and still keeps the magnet. We are subwavelength by construction: no gyromagnetic volume, no magnet.

The asymmetry

Time-modulation gets harder as frequency rises — the pump must run at a sizeable fraction of the carrier frequency — so our mechanism is at its easiest precisely where ferrite is at its worst. The funded efforts we have found targeted the GHz/integrated tier; the low-band module tier remains unaddressed.

"…a new category of subwavelength non-reciprocal circulators… operating at microwave frequencies without the need of magnetic materials… based on suitably tailored time-varying networks."
— DoD OSD STTR solicitation OSD21C-002 (2021). That is the component class we build.
status
433 Alpha v0.9d · design review complete, first articles pending
first target
VHF/UHF module
why it matters
SWaP + magnetically clean

02 · The physics

Boundaries in time, not just space.

Every conventional RF component is a structure in space: waves hit an interface, and the interface reflects, filters, or guides them. Frequency stays fixed; geometry does the work. That is the physics of every filter, antenna, and ferrite isolator made in the last century.

A time-modulated medium adds the other axis. Switch the line's properties at a chosen moment, and the roles invert: wavelength holds, frequency shifts, and a portion of the wave returns as a time-reversed echo. Sweep the switching pattern through space and time together, and the medium behaves as if it were moving — Doppler shifts, one-way propagation, and effects no static circuit can produce, without a magnet in sight.

What you are looking at: a pulse travelling along a line of coils and capacitors, computed live. Press FIRE BURST; while the pulse is mid-flight every capacitor changes value at once. Watch the pulse keep its shape but jump in frequency, and a faint echo run backward.

A WAVE CROSSING A BOUNDARY IN TIME computed in your browser

line voltage V(x) probe trace V(t) @ marker boundary instant fin 1.00 · predicted fout = β·fin 2.00 · measured

Real simulation shown: a discrete LC transmission line integrated live. Fire a burst; while it is mid-flight, every capacitor in the line switches at once — a boundary in time. The wave keeps its wavelength, jumps in frequency (f → β·f, β=√r), and splits into a transmitted wave and a backward echo. This effect was first measured in an electrical lattice in 2023 (Moussa et al., 2023); it is the physics our components are built on.

What you are looking at: position across, time upward. The shaded cone is how fast a wave can travel in the line; the violet line is how fast the switching pattern sweeps along it. Drag the slider. Slower than the waves gives the one-way behaviour our isolator uses; faster than the waves acts like the boundary in time above.

HOW FAST THE PATTERN MOVES DECIDES WHAT THE LINE DOES interactive

regime SUBLUMINAL βmod = vfront/vwave 0.50

Position across, time upward. The shaded cone is how fast waves travel in the line; the violet line is how fast the switching pattern sweeps along it. Slower than the waves, the pattern Doppler-shifts them differently by direction — the non-reciprocal regime our isolator uses. Faster than the waves — allowed, because a pattern imposed from outside carries no signal — it acts as a boundary in time. One hardware platform reaches every regime; only the schedule changes.


03 · The platform

Bubble One: temporal-boundary physics on an accessible bench.

Time-modulated media are usually studied on microwave metamaterial rigs. Bubble One is our validation platform: a 96-section switched transmission line, microsecond-scheduled by a microcontroller, instrumented with a bench oscilloscope — a programmable medium built from catalog parts for a few hundred dollars.

It exists to measure specific predictions of temporal-boundary physics on real hardware, openly and reproducibly. Results, hardware documentation, and the full build will be published here as they are validated.

MEASURED DATA Stage E campaign · 1 September 2026

Measured: a 25 kHz packet caught inside an opened segment exits at twice its frequency; measured spectrum peak against the 51.8 kHz prediction
×2.00frequency conversion at a temporal boundary, measured, against 2.07 ideal (sim 0–3 %)
6 / 6modulated segments certified by an automated phase-step ladder, 0.8–9 % from the time-domain model
87 mV vs 3 mVtemporal reflection at node 8 against the noise floor, confirmed by a no-packet control

Every shot scripted: scope, generator, and firmware driven from code, predictions written before measurement, model parameters measured independently and never fitted. Full bench log and hardware documentation publish with the open-source release.


04 · The company

Building empirically.

Onuska Industries LLC is an independent hardware company founded by Ian Onuska, based in Pittsburgh, Pennsylvania, with a lab in South Amboy, New Jersey. We work from first principles, publish the results we validate, and make no claims ahead of measurement. What you see on this page is either computed live, measured and dated, or labeled as a prediction.

ian@onuskabrown.com