ALMITA — 21 cm Hydrogen Line Radio Telescope

Antenna Listening Mostly to Interference, Tentatively Astronomy

ALMITA is an open-source amateur radio astronomy project developed in Chile by Felipe Fridman (ffridman@gmail.com), built around a 21 cm neutral hydrogen (HI) radio telescope. It listens to the 1420 MHz hydrogen line with RTL-SDR receivers, a modified grid dish and an INDI/OnStep equatorial mount. A Raspberry Pi in a shielded box sits at the instrument, and a notebook runs the web console. ALMITA turns raw radio data into traceable spectra, velocity cubes and maps. It is an experimental amateur radio astronomy project that works fully offline in the field.

Source code on GitHub Project README Project overview

From antenna to maps: OBSERVE → REDUCE → SCIENCE

OBSERVE  →  REDUCE v1.1  →  SCIENCE
   ↑ ALIGN · CALIBRATE · RFI_REF supporting systems

REDUCE and SCIENCE V1 were frozen in October 2026 as a fixed, verifiable reference. They were reopened twice on purpose: REDUCE v1.1 folds frequency-switched captures, and SCIENCE maps use an exact spherical frame so they can cross the celestial pole. Other improvements, such as ESTIMATOR v2 (a better spectrum estimator inside REDUCE), are opt-in and leave the engine unchanged.

Automated observing
Plans sky mosaics, steps the equatorial mount and records raw I/Q with pointing and environment metadata in HDF5.
Deterministic reduction
REDUCE produces reproducible Level 1 spectra with masks, LSRK velocity, uncertainty and quality information.
Beam-aware science products
SCIENCE grids the spectra into an LSRK velocity cube and relative integrated maps, with uncertainty and coverage maps.
Relative calibration
Receiver gain is characterised first, then kept fixed during a session, so observations can be compared.
Live RFI masks
Every capture is checked as it is recorded: comb teeth, narrow lines and time pulses are flagged, never filled in, and REDUCE can leave them out.
Web console
Twelve pages behind a login, from STATUS and PREFLIGHT to OBSERVE, MONITOR, REDUCE, SCIENCE, SESSIONS and a remote desktop of the capture Pi.
Solar transits
LOBE fits the beam from a transit with the mount; TRANSIT logs the band power every 10 s while the Sun drifts through a still antenna.
Offline operation
Everything runs locally on a notebook and a Raspberry Pi; no internet connection is needed to observe.

First results

Hydrogen with frequency switching, repeated and matched · 12 November 2026

A week later the same 5 × 5 mosaic around l ≈ +20° was observed again, now with frequency switching (40 s per pointing) and a new feed with a Pawsey balun and a reflector plate. The line shows up in all 25 pointings. For the six pointings inside the reference cube, the switched spectra match HI4PI's prediction for the same pointing and 20° beam with a shape correlation of +0.94 to +0.99 (5 November: +0.43 to +0.77), on a flat baseline. The receiver's Y-factor was 4.71 dB, a system temperature of about 120 K.

Six frequency-switched 21 cm spectra from ALMITA in blue against the HI4PI prediction for the same pointing and 20 degree beam in black
Blue: ALMITA, frequency-switched and folded by REDUCE, LSRK velocity. Black: HI4PI predicted for the same pointing and beam. Still relative, not Kelvin. Peaks sit 3–10 km/s above HI4PI's, not yet explained. The Magellanic Clouds, at ~1–2 % of the system noise, were not detected the same night: the switching leaves a ±1–2 % baseline that depends on where the dish points.

First 21 cm hydrogen detection · 5 November 2026

On its first night on a rooftop, in a 5 × 5 mosaic around galactic longitude l ≈ +20°, ALMITA saw the hydrogen line where the HI4PI all-sky survey says it should be. For the six pointings inside the reference cube, the measured line has the shape HI4PI predicts for the same pointing, the same 20° beam and the same instant (shape correlation +0.43 to +0.77). With the velocity axis flipped, the match fails.

Six measured 21 cm spectra in blue against the HI4PI prediction for the same pointing, beam and instant in black
Blue: ALMITA, relative to an off-plane reference. Black: HI4PI predicted for the same pointing, beam and instant. One night, not yet repeated and not calibrated. The ±2–6 % waves around the line are the receiver, not the sky. The peaks sit 3–6 km/s above HI4PI's, which is not yet explained, and an earlier strip at l = +10…+30° showed nothing.

Frequency switching · November 2026

Those waves are the receiver's own band shape, which is ±15 % and drifts. With frequency switching the SDR alternates its tuning by ±0.6 MHz every 2 s: the hydrogen line moves with the tuning, the receiver's shape does not, and (A − B) / B cancels the shape.

Bench test: receiver band shape of ±14.6 % versus 0.29 % noise in one switched 10 s pointing and 0.06 % in the mean of 400
Bench test with the LNA and the SDR but no antenna, so no sky signal is expected. One 10 s pointing: 0.29 % noise. Mean of 400 pointings: 0.06 % rms left of the receiver. A steady local carrier at 1420.040 MHz is masked. The first open-sky test across the Milky Way is next. Tap a figure for full size.

What is new (November 2026)

The web console

Twelve pages behind one login, shown as each one looks when you land on it. Tap a page for the full screenshot. Network addresses are hidden and the mount camera is blurred; the data is a frequency-switching bench test with the mount on its simulator.

MONITOR page: live view of a run: sky map, 3D spectral stack, RFI masks, Unknown PleasuresMONITOR · live view of a run: sky map, 3D spectral stack, RFI masks, Unknown Pleasures STATUS page: instrument health, LNA power, RFI check, hydrogen-line testSTATUS · instrument health, LNA power, RFI check, hydrogen-line test PIPELINE page: the whole chain end to endPIPELINE · the whole chain end to end OBSERVE page: plan a sky mosaic over the HI4PI mapOBSERVE · plan a sky mosaic over the HI4PI map CALIBRATE page: receiver profile and Y-factorCALIBRATE · receiver profile and Y-factor ALIGN page: pointing alignment on the Sun or on HIALIGN · pointing alignment on the Sun or on HI LOBE page: beam from a solar transitLOBE · beam from a solar transit TRANSIT page: the Sun through a still antennaTRANSIT · the Sun through a still antenna REDUCE page: raw captures to spectraREDUCE · raw captures to spectra SCIENCE page: cubes and mapsSCIENCE · cubes and maps SESSIONS page: catalog of every sessionSESSIONS · catalog of every session DESKTOP page: the capture Pi's own desktopDESKTOP · the capture Pi's own desktop

Hardware

An RTL-SDR Blog V3 as the science receiver with a Nooelec SAWbird+ H1 LNA, a DIY 1420 MHz dipole feed on a modified grid reflector of about 90 × 60 cm, an equatorial mount with an OnStep controller and INDI, a Raspberry Pi 4 capture box and a notebook. The reflector plate of the feed is fitted; a Pawsey balun comes next. A second, independent RTL-SDR chain for RFI reference is supported but currently disconnected. The hardware is accessible, experimental and repairable. Until November 2026 everything ran on a single Raspberry Pi 5.

Scientific honesty

Current HI measurements are relative and instrumental: ALMITA does not claim an absolute calibration, and the first hydrogen detection (above) is a single night that still has to be repeated. Next come an open-sky night with frequency switching across the Milky Way, solar alignment, a Y-factor with the current feed and a Pawsey balun. The antenna beam used for mapping is an operator-configured, provisional model until a proper beam characterisation is done, and no astrophysical interpretation is automated. Results are those of an experimental amateur instrument, not of a professional observatory.

Why ALMITA exists

The project explores how far accessible hardware can be pushed toward careful, auditable amateur radio astronomy: raw data are preserved, uncertainties are explicit and claims are conservative.

Some parts are carefully engineered. Some parts are carefully attached with plastic cable ties. Both approaches have survived surprisingly well so far.