The Year the Robots Listened to the Ionosphere
The Year the Robots Listened to the Ionosphere
One receiver listened to 20 metres for a year and logged 30,819,847 FT8 messages. That is a remarkable view of amateur-radio path availability—provided we keep activity, propagation and measurement separate.
When I read Gergely Vakulya and Helga Anna Albert-Huszár’s 2025 Signals paper, one image stuck immediately: a ham-radio robot sat quietly in Hungary and listened for a year. No operator had to remain at the dial. A receive-only station decoded 20-metre FT8, hour after hour, and left behind a dataset large enough to expose patterns that a weekend log never could.
The robot heard amateur activity through the ionosphere; it did not measure the ionosphere directly. A successful decode proves that one complete transmitter–path–receiver combination crossed the decoder threshold at that moment. Turning millions of such events into propagation evidence requires explicit control of human activity, equipment, missing data, geolocation, filtering and uncertainty.
A Dipole, One Receiver and One Year
Vakulya and Albert-Huszár describe a deliberately modest, remotely accessible station. It used a simple wire dipole, an mcHF 0.6 transceiver with transmission disabled, and an Intel NUC running Debian Linux and WSJT-X. The receiver’s attenuator/preamplifier block was used in pass-through mode. Remote shell, desktop and VPN access kept the system manageable without a local operator.
The station monitored the 20-metre FT8 channel; the public ALL.TXT records show a 14.074 MHz dial frequency. Collection ran from 25 June 2024 through 25 June 2025. The paper reports 30,819,847 decoded messages.
That number is message records, not 30.8 million unique stations, contacts or independent propagation paths. One QSO can produce several decoded messages, one station can appear thousands of times, and multiple messages can be decoded in one 15-second interval. Power failures created gaps, and the authors explicitly discuss false decodes.
The public supplementary repository contains the split raw ALL.TXT archive plus derived histogram and grid-map files. That is valuable: readers can inspect receiver-generated records rather than relying only on a plotted map. Reproducing a figure still requires documenting the parser and filters that turn those records into events.
What an FT8 Record Actually Knows
WSJT-X writes a local decode record with time, dial frequency, mode, estimated signal-to-noise ratio, time offset, audio-frequency offset and decoded message text. The message itself changes during a contact. A CQ may carry a callsign and four-character Maidenhead locator; a later exchange may carry two callsigns and a signal report instead. Not every decoded message contains a grid square, and the report inside a transmitted message is not necessarily the local receiver’s SNR estimate.
A four-character Maidenhead locator covers 2° longitude by 1° latitude. There are 32,400 possible four-character squares. That is the size of the address space, not a count of squares observed by this station. A locator is also message content supplied by an operator; it is not an independent GNSS position measurement.
The current WSJT-X user guide defines a signal report as SNR in dB referred to a 2500 Hz noise bandwidth. That is useful for comparing decodes under controlled receiver conditions. It is not calibrated electric field strength at the antenna. Receiver gain and AGC state, filter response, antenna pattern and polarization, feed-line loss, local noise, overload, clock accuracy and decoder settings all lie between the arriving field and the number in the log.
| Recorded item | What it supports | What it does not establish alone |
|---|---|---|
| UTC time and 14.074 MHz dial frequency | When and where the receiver listened | Continuous availability if outages and configuration changes are not logged |
| Decoded callsign and message | A structured payload passed the decoder | A unique station, completed QSO or independently authenticated identity |
| Four-character locator in a message | A coarse, operator-supplied source location | Exact transmitter coordinates or proof that a mobile station was inside that square |
| WSJT-X SNR estimate | Received SNR on the software’s stated 2500 Hz reference under that setup | Calibrated field strength, transmitter power, path loss or antenna gain |
| Decode count | Observed traffic above the complete system’s decode threshold | Ionospheric state without activity and equipment controls |
The Map Shows Activity Through One Receiving Window
The paper’s CQ-message map is dominated by Europe and the eastern United States, with visible activity from Central America, Japan, parts of Australia and New Zealand, Indonesia and several islands. That is an honest description of the plotted decodes. It is not a map of ionospheric quality alone.
Every cell combines at least four layers:
- transmitter population and behaviour: where licensed operators live, when they transmit, which power and antennas they use, and which awards or rare locations attract activity;
- the path: ionospheric refraction and absorption, ground interaction, multipath, solar illumination and space weather;
- the receiving station: its dipole pattern, polarization, attic environment, feed system, local noise, receiver state and downtime; and
- the data pipeline: message type, decoder threshold, duplicate definition, false-decode filter and callsign–locator validation.
The authors found message locators in open-ocean squares and attributed them to maritime amateur stations. They also plotted activity in Antarctica, notably around the Neumayer Station III area. Those are interesting observations, not independent position fixes. A defensible follow-up would retain the callsign, repeated locator, timestamps and message types, then corroborate station identity and operating location from an independent record before calling the source a ship or Antarctic installation.
Rare islands can appear disproportionately bright because amateurs actively seek them and DXpeditions generate intense short-lived traffic. Conversely, a quiet cell may contain an excellent ionospheric path but no transmitter, no active operator, the wrong antenna direction or a signal below this receiver’s threshold. Decode density is therefore activity filtered by observability.
Distance–Time Patterns Are Evidence, Not a Mechanism Label
For preliminary analysis, the researchers grouped decoded messages into 500 km distance bins and compared three 10-day periods: 12–22 December 2024, 1–11 May 2025 and 1–11 June 2025. Their heat maps show hour-dependent and season-dependent changes in which distance ranges were observable from the receiving site.
Those changes are compatible with established HF behaviour, but a distance band is not a measured hop count. The same great-circle distance can be supported by different launch angles, layer structures, chordal or multi-hop paths, and ground reflections. A gap can also come from transmitter schedules or a local-noise change. Calling every bright band a particular numbered hop requires propagation modelling or independent ionospheric evidence.
Nighttime deserves particular care. NOAA’s ionospheric-region definition identifies the D region as a principal source of HF absorption and the F region as central to radio communication. As darkness arrives, D-region absorption generally reduces, while F-region electron density and usable frequencies also evolve. Those effects can push a 14 MHz path in different directions. A nighttime hole in 20-metre decode counts is therefore not proof that “the D layer disappeared,” nor does correlation with local time identify one causal layer.
To test a mechanism, align the FT8 observations with independent quantities: receiver uptime, local solar zenith, ionosonde foF2 and virtual-height data, geomagnetic indices, solar flux and flare records, plus a propagation model for the specific paths. NOAA’s ionogram guidance explains that ionosondes provide virtual-height and critical-frequency measurements. Those are different observables from an oblique amateur decode and make a causal comparison much stronger.
What FT8 Contributes
FT8 gives propagation research something genuinely useful: a dense stream of time-stamped, frequency-specific detections from ordinary amateur activity. The primary FT4/FT8 protocol paper documents the structured 77-bit payload, 15-second cadence, 50 Hz occupied bandwidth and error-control coding that make automated collection practical.
At one receiver, a decode is a binary piece of path-availability evidence plus an SNR estimate under the station’s current conditions. Repeated consistently, it can support:
- diurnal and seasonal statistics for that site, band, antenna and decoder configuration;
- distributions of observed path distance and bearing;
- before/after comparisons when receiver state and amateur activity are controlled;
- hypothesis generation for unusual openings or long-duration changes; and
- model validation when matched with independent ionospheric and space-weather data.
FT8 also carries more human selection than a dedicated beacon experiment. Operators choose when, where and why to transmit, and power and antenna data are normally absent. WSPR and purpose-built beacons answer different questions because their operating assumptions and metadata differ. More FT8 records improve statistical resolution; they do not automatically remove selection bias.
A Reproducible Analysis Boundary
If I were extending this year-long experiment, I would freeze the following before looking for a seasonal explanation:
- Pin the evidence: record the paper version, dataset commit and hashes of every raw archive.
-
Publish the schema: define every
ALL.TXTcolumn, time zone, WSJT-X version, receiver setting and configuration change. - Define an event: state whether repeated messages in one slot, one QSO or one callsign–locator pair count once or many times.
-
Declare filters: identify message types eligible for geolocation, publish the false-decode threshold
k, validate callsign–locator pairs and show sensitivity to alternative thresholds. - Measure exposure: build an uptime mask for outages and discard or flag hours without a known listening state.
- Bound location: use square centroids only with the uncertainty implied by a 2° × 1° locator, and distinguish fixed, portable, maritime and expedition evidence.
- Characterize the receiver: log gain/AGC, bandwidth, antenna orientation and pattern, feed loss, local noise, overload indicators and calibration checks.
- Separate counts from causes: normalize for observed transmitter activity where possible, then compare predeclared hypotheses with independent ionosonde and space-weather data.
- Publish code and uncertainty: make the parser, plots, distance calculation, exclusions and confidence or sensitivity analysis executable by another researcher.
That is the real beauty of this work. The station was simple enough for an amateur to understand, patient enough to collect a year, and open enough to invite another analysis. The robot did not explain the ionosphere for us. It gave us 30,819,847 carefully time-stamped reasons to ask better questions.
Primary sources and public evidence
- Vakulya and Albert-Huszár, Signals 6(4), 58: station architecture, collection period, message total, filtering and preliminary maps.
- Signals FT8 data repository: public raw split archive and derived histogram/grid-map artifacts.
- WSJT-X user guide: decoder records and SNR’s 2500 Hz reference bandwidth.
- Franke, Somerville and Taylor, “The FT4 and FT8 Communication Protocols”: FT8 messages, timing, modulation and coding.
- NOAA/NCEI ionospheric regions and vertical-incidence soundings: D/F-region roles and independent ionosonde observables.
Mini-FAQ
- Did this receiver measure the ionosphere directly? No. It decoded amateur transmissions that arrived through a complete transmitter–path–receiver system. Independent ionosonde or space-weather data are needed to test a specific ionospheric mechanism.
- Were there 30,819,847 unique stations or paths? No. That is the reported number of decoded message records. A station, QSO or path can contribute many messages, and several signals can be decoded in one FT8 interval.
- Is WSJT-X SNR the same as calibrated field strength? No. It is an SNR estimate referred to a 2500 Hz noise bandwidth. Field strength or path loss requires calibration and control of the antenna, feed system, receiver, gain, bandwidth and local noise.
- Do ocean and Antarctic grid squares prove where the transmitter was? No. They are coarse, operator-supplied locators. Repeated callsign–locator evidence can be consistent with a vessel or Antarctic station, but independent station records are needed for confirmation.
- Can a distance–time heat map identify D-layer, F-layer or hop count? Not by itself. Decode counts also depend on human activity and the receiving system, while one distance can arise from several propagation modes. Test the explanation against independent ionospheric data and a path model.
- What makes a repeat analysis reproducible? Pin the dataset, publish the parser and filters, define one counted event, mask downtime, document receiver state, bound locator uncertainty and report sensitivity to filtering and activity bias.