Open multi-spectral sensing · v0.1.0
Most of the sky is invisible to you.
Your eyes cover one narrow band out of fourteen. NBAND is a sensor node you can actually build that watches the rest of them at the same time, stamps every reading against a satellite-disciplined clock, and publishes what it records so that anyone can check the work.
The instrument, by wavelength
The idea
One sensor can be fooled. Fourteen disagreeing is information.
A bright dot on a camera is almost nothing on its own. It could be a satellite, an aircraft, a bug lit by a porch light, or a dead pixel. The picture alone cannot tell you, and no amount of arguing about the picture will settle it.
What settles it is asking the other thirteen bands what they saw at the same instant. Something genuinely hot shows up in the thermal band. Something with an engine makes noise the microphones hear a few seconds later. Something transmitting lights up the radio receiver. Something close and metallic bends the magnetometer. A dead pixel does none of that, and neither does a satellite.
Everything is stamped to the same clock
Every node disciplines its clock against GNSS satellites with a hardware pulse-per-second signal, holding a few hundred nanoseconds. That is what makes "at the same instant" a measurement instead of a figure of speech, and it is why two nodes 60 km apart can triangulate a real altitude rather than guess at one.
Agreement across bands is the trigger
A single channel crossing a threshold is noise until something else agrees with it. When two bands trigger inside 250 milliseconds of each other, the node promotes the whole buffered window to permanent storage, including the 15 seconds that happened before the trigger fired.
Known things are subtracted first
Before anything is called interesting, the discriminator checks it against aircraft transponders, satellite orbits, lightning networks, the site’s own radio interference fingerprint, weather, and space weather. Most events are explained and closed. That is the system working.
What it sees
Each band answers a different question
Ionizing photons above 100 keV.
Radioisotope decay, cosmic-ray secondaries, and any source emitting hard photons. The only band in the stack that responds to nuclear rather than thermal or electronic processes.
Ultraviolet100 to 400 nanometres, above the violet edge of vision.
Corona discharge, electrical arcing, plasma, combustion, and lightning leaders. Solar UV sets the daytime floor, so the band is far more informative after dark.
Visible380 to 750 nanometres, the band your eye already covers.
Anything that reflects sunlight or emits its own light: aircraft, satellites, meteors, balloons, birds, and the occasional thing that fits none of those. Provides the astrometry that turns a detection into a bearing.
Near infrared750 to 1400 nanometres, just past the red edge.
Hot exhaust, incandescent surfaces, IR illuminators and rangefinders, and haze-penetrating reflected light. Silicon sensors are natively sensitive here, so removing the IR-cut filter from a normal camera buys the band for free.
Short-wave infrared1.4 to 3 micrometres. Reflective, not thermal.
Sees through haze, thin smoke, and some fog far better than visible light. Discriminates materials by reflectance in a way no other band in this stack can. Night-sky airglow illuminates targets passively at 1.5 to 1.7 micrometres.
Long-wave infrared8 to 14 micrometres. Pure thermal emission.
Everything warmer than absolute zero, by its own emitted heat rather than reflected light. Works in total darkness and through smoke. A radiometric sensor reports actual temperature per pixel, which turns a track into an energy-budget measurement.
Millimetre wave24 to 81 gigahertz active radar.
The only band in the stack that measures range and radial velocity directly, by illuminating the target and timing the return. Gives the discriminator a physical distance, which is what converts an angular track into a real trajectory.
Radio frequency500 kilohertz to 6 gigahertz, received passively.
Emissions rather than reflections. Aircraft transponders, satellite downlinks, control links, broadband impulsive noise from discharge events, and anything transmitting where nothing should be. Also feeds passive radar: an aircraft crossing a broadcast transmitter's illumination produces a Doppler-shifted echo.
Magnetic and ELF/VLFDC to 30 kilohertz field measurement, not photon detection.
Static and slowly varying magnetic fields, sferics from distant lightning, power-line harmonics, and any moving ferromagnetic or current-carrying mass close enough to perturb the local field. This is the band that has historically carried the most repeatable anomalous reports.
The discriminator
The top of the ladder is “unresolved”, not “alien”
The signal originated in the instrument: sensor glitch, hot pixel, shutter event, self-interference from the node's own emitters, condensation on optics, or a cable fault.
Matched to a specific catalogued object. An ADS-B airframe by hex code, a satellite by NORAD ID, a lightning stroke by network fix, a licensed transmitter by frequency and bearing.
Consistent with a known class but not matched to a specific object. A bird, an insect near the lens, a balloon, a meteor, an aircraft not transmitting ADS-B. Common and uninteresting, and by far the largest bucket after known sources.
Insufficient data to classify. Too few bands, too short a track, clock quality too poor, or the only witness channel was one the discriminator does not score alone. Not a mystery, just a bad measurement.
Survived every catalogue subtraction available, was witnessed coherently in two or more bands, and has kinematics or energetics the discriminator could not reconcile with any conventional class it knows. This is a statement about the limits of the catalogues, not a claim about the object.
An event cannot reach the top rung on one channel, or with a degraded clock, or when a catalogue the discriminator wanted to check was unavailable. Every lookup it performed is recorded, including the ones that found nothing and the ones it could not run. “We checked ADS-B and found no aircraft” and “we could not reach ADS-B” are different claims, and the archive keeps them different.
Build tiers
Start at the bottom. The grid treats every tier the same.
Tier 1 - Baseline
$456Visible, near-infrared, radio, environmental, and disciplined time. The minimum configuration that can contribute usefully to the grid.
Bill of materials →Tier 2 - Core
$1445Adds long-wave infrared, ultraviolet, millimetre-wave radar, acoustic, and magnetometry. The configuration the build guide is written against.
Bill of materials →Tier 3 - Extended
$4433Adds short-wave infrared, wideband SDR, gamma spectroscopy, and a seismometer. Research-grade coverage without research-grade cost.
Bill of materials →Costs are the sum of sourced part prices as of July 2026 and exclude tools, shipping, and tax. Silicon pricing is unstable this year: Raspberry Pi boards have risen three times since December 2025 as memory supply moved to AI datacentre demand. The entry tier is held near its target by specifying a 2 GB board and writing firmware disciplined enough to run on it, not by pretending prices did not move.
Before you build one
What this will not do for you
It will not prove anything on its own
A single node produces angular tracks with no range. Without range you have no size and no speed, only a direction and a brightness. The honest output of one node is a well-characterised question. Range comes from radar or from a second node, and the second node is usually the better buy.
Most of what it records is boring
Aircraft, satellites, birds, insects near the lens, and the neighbour’s motion light. Expect the overwhelming majority of events to close as explained. A system that frequently finds mysteries is a system with a calibration problem.
It demands a real site, not a windowsill
Glass blocks ultraviolet and is opaque in the thermal band. A node indoors is a node with four dead channels. It needs sky, power, a horizon survey, and somewhere the magnetometer is not sitting next to a refrigerator.
A null result is the likely outcome, and it counts
Five hundred hours of coverage that turns up nothing unexplained is a real measurement: it puts a bound on how often anything unusual crosses that patch of sky. The archive is built to make that bound computable rather than to make headlines.
The build guide starts with one camera and a clock.
Every step ends in something you can verify before you move on: a command that prints an expected value, a capture you can look at, a timing offset you can read. If a step does not verify, the guide tells you what usually causes that.