ELINTIR ML/RF MAPPER

Angel round

elintir.com/investors

The site is the short version of everything: three directions, the company, the hardware, contacts. The data room behind it is the long version — with measurements, the protocol specification, and the results that did not hold up.

Data room access

Granted person by person

The data room is not public and does not open from a direct link. Write to us and we will send the materials personally, agreeing NDA terms by mail.

Please include your name, organisation and the context of your interest — round, partnership, procurement. It speeds up the reply.

Request access

An automated single-use code and self-service entry to the data room are in development. Today the process is manual, and we say so plainly rather than pretending the automation exists.

Where the edge is

Value compounds in the data, not only in the hardware

The station does not have to carry the whole analysis. It writes multichannel coherent raw IQ in a standard format with provenance metadata, and the heavy processing happens on the ground. This is a proven capability, not a plan.

A cheap node feeds an expensive analysis
The carrier only has to receive and record. Signature processing, classification and training stay where there is compute and power.
A recording outlives the algorithm
When the processing improves, the same archive is run again. Getting a new result needs no second flight — and the flight is the expensive part.
A recording is evidence
Provenance metadata travels with the samples, so a measurement can be checked rather than merely reported. That is what separates a result from a claim.

What was done before asking for money

We already know what the expensive board will compute

Buying an RFSoC and working it out afterwards is the most expensive approach available. So the arithmetic was done first, and it starts from an inconvenient fact: the raw stream off such a board fits nowhere.

Raw converter stream
393 Gbit/s across eight channels. The fastest interface on the board gives 262 — the stream overruns the budget by 3.9×. There is no "just stream it to a computer", whatever the board costs.
So the fabric computes rather than forwards
The programmable logic accumulates the covariance matrix, and that is what leaves the board instead of the stream. This is not a compromise: for the direction-finding algorithm the matrix is a sufficient statistic — it looks at nothing else. Thousands of times less data, with the task itself intact.
And it fits — checked by arithmetic
An 8×8 covariance at full rate needs 590 GMAC/s against a capacity of 1709 — 34.5% loading. Accumulators for 1024 subbands take 1.16% of block memory. What remains is enough for the rest of the chain.
One bank covers the whole band
A polyphase channeliser splits the band into subbands but does not change the total rate — so a single covariance bank serves every subband at once rather than one each. That property is what makes the arithmetic work out.
What it costs, stated
The matrix discards time, phase, and the ability to listen to a capture again. So raw IQ stays a separate path: the matrix for the live bearing, the recording for signatures and re-analysis. Two paths, not one instead of the other.

That is the difference between "give us money for boards" and "give us money for boards for which it is already worked out what goes in them and how much stays free".

What the round is for

Not only boards — drones and airborne trials

The next step in range runs into geometry rather than the receiver: altitude and position move the radio horizon from single kilometres into tens. So the round has to cover more than components.

100 m 40 km single km station height exaggerated — a diagram, not to scale
The same geometry as on the Hardware page: from the ground the horizon is single kilometres, from 100 metres it is tens. This is line of sight, not a promised DF range; only flight tests turn one into the other — which is what the round is for.
Boards and instruments
The RFSoC chain, analysers and a generator — without them a measurement stays a claim.
Airborne carriers
Platforms built for the array: clearance from motors and power runs, fibreglass near the phase elements, attitude synchronised from the flight controller.
A flight test programme
A fixed beacon at 5, 10 and 20 km, heights of 50/100/150 m, band by band, plus passes over the beacon to check the nadir cone. This is exactly the measurement missing today before range can be stated as a number.

Allocation, valuation and deal structure are in the deck and the data room.

Where this grows

Radar — a direction that already has code in it

The station is passive today, and its whole position — from staying unseen to its export category — is built on exactly that. Radar changes that position, which makes it a deliberate decision rather than a default. But it is the direction where the market value may turn out to be larger than everything above — and we would not be starting it from nothing.

Passive radar: the code is written, the physics checked
A mode that uses someone else's illumination — a DVB-T2 tower instead of a transmitter of our own. Wiener clutter cancellation and the cross-ambiguity function are written and covered by tests; coherent IQ capture off a live station works. What blocks it is measured too: an omnidirectional array is no good as the reference — the direct path drowns among the elements, and no amount of processing substitutes for it. A directional antenna on the reference channel is required. The strongest illuminator at our site is 698 MHz, direct path +21 dB.
Active radar: the same three skills
An active mode is waveform generation, coherent multichannel reception and correlation. We already do all three: we write our own physical layer for s.Link, we hold five coherent channels for direction finding, and the correlation maths exists for passive radar. What is missing is not the skill but the decision to radiate, and hardware for the power.
What it changes, and why it is the owner's call
Radiating makes the station visible and moves the product into a different regulatory category. That runs directly against what makes it valuable today, so radar is not a replacement for the passive station but a second product off the same shelf of skills. We name it as an R&D direction, and not as something that already works.

What is inside

Before that

The public half of the same material — the station specifications and the measured numbers — is already on the hardware page.