Over-the-horizon radar waveform detection and classification from LEO

HF over-the-horizon radars broadcast distinctive chirp and FMCW signatures across thousands of kilometres of skywave path. LEO wideband receivers can detect, classify and schedule-track those emissions, within the real limits imposed by ionospheric geometry and networked emitter ambiguity.

Sensors

HawkEye 360 cluster satellites (HF extension bands): Three-satellite formation flying at roughly 575 km altitude; TDOA/FDOA geolocation across HF and VHF bands. Geolocation accuracy published at 1–10 km depending on geometry and signal duration. Revisit at any given point varies by latitude but is typically several passes per day for a cluster of this size.
Spire Global LEMUR-2 constellation (HF/VHF SDR payloads): Software-defined radio payloads covering portions of HF through UHF. Large constellation (100+ satellites) gives sub-hourly revisit in aggregate, though individual satellite dwell over a target is short. Useful for detecting scheduled OTHR transmission windows rather than continuous waveform characterisation.
ITU-R M.1462 coordination database (ground truth reference): Not a sensor, but the published ITU coordination record for OTHR systems lists notified frequency ranges and coordination zones. Cross-referencing detected emissions against M.1462 notified bands distinguishes known coordinated OTHR from uncoordinated or clandestine operation.
Dedicated government SIGINT payloads (classified, context only): National SIGINT satellites carry purpose-built HF receivers with larger apertures and higher dynamic range than commercial platforms. Their data is not available commercially, but their published existence confirms the physical feasibility of LEO HF intercept at the waveform level.

What makes OTHR waveforms recognisable from 500 km up

Over-the-horizon radars operate in the HF band, broadly 5–30 MHz, and illuminate targets at ranges of 1,000–4,000 km by refracting energy off the ionosphere. That long path is also the source of their detectability from orbit. To achieve useful range resolution, OTHR systems transmit wideband frequency-modulated continuous-wave or linear frequency-modulated chirp waveforms, sweeping bandwidths typically between 10 kHz and several hundred kHz within a single transmission. That sweep is not subtle. A LEO receiver passing over captures the characteristic ramp in instantaneous frequency, a signature that survives the additional propagation through the ionosphere to the satellite.

The Australian Jindalee Operational Radar Network (JORN) is the most extensively documented open-source example. Published technical literature describes its operation across multiple HF sub-bands, with frequency agility used to track the optimal ionospheric reflection layer. Chinese and Russian OTHR systems, including the Chinese 'Sky Wave' OTHR and the Russian Konteyner system (declared operational in 2019), follow the same physical logic and produce broadly similar waveform families, though their specific parametric choices differ. From a classification standpoint, the chirp rate, sweep bandwidth, pulse repetition interval and dwell time on a given frequency are the discriminating features.

Geolocation from orbit: what the physics allows and what it does not

A single LEO satellite cannot geolocate an HF emitter usefully. The signal arrives via skywave, meaning the apparent angle of arrival at the satellite reflects the ionospheric reflection point, not the transmitter. Formation-flying clusters such as HawkEye 360 use time-difference and frequency-difference of arrival (TDOA/FDOA) across their baselines to estimate emitter location, but the ionospheric refraction introduces an additional unknown. The reflection height varies with solar activity, time of day and season, typically between 150 and 350 km for F-layer propagation. Without an independent ionospheric model constrained to the actual conditions at the moment of intercept, the geolocation error budget for an HF skywave emitter is tens to hundreds of kilometres, not the single-digit kilometres achievable for line-of-sight VHF/UHF emitters.

This is an honest and important limit. For OTHR systems, it matters less than it might appear. The transmit sites of the major known OTHR networks are already documented in open sources, ITU filings, and commercial satellite imagery of the antenna fields. What space-based intercept adds is not the discovery of a site's coordinates but rather the detection of activation, the characterisation of waveform parameters in use, and the inference of operational tempo from scheduling patterns. That is a different and genuinely useful intelligence product.

Reading the schedule: operational tempo as the primary intelligence product

OTHR systems are not always on. They are expensive to operate, ionospherically constrained, and often tasked against specific areas of interest. A system monitoring the southern ocean approaches behaves differently from one cued to track a surface group in the South China Sea. Repeated LEO passes over weeks and months build a transmission schedule: which frequency sub-bands are active, at what times of day, and whether the dwell pattern shifts. Shifts in dwell direction are harder to infer from a single pass but become visible in aggregate when the emitter changes its sweep rate or centre frequency in ways consistent with re-tasking.

Scheduling analysis requires patience and archive depth. A single intercept event is ambiguous. Thirty intercepts across sixty days, cross-referenced against known solar flux indices (which govern ionospheric conditions and therefore optimal OTHR operating frequencies), begin to reveal whether a system is running a routine surveillance schedule or responding to specific triggers. This is the kind of longitudinal analysis that commercial RF monitoring from LEO is well positioned to support, provided the revisit cadence is sufficient and the spectral coverage spans the system's known frequency range.

The networked emitter problem

JORN is not one radar. It is a network of geographically separated transmit and receive sites whose coverage sectors overlap and complement each other. The Russian Konteyner system similarly uses multiple transmit nodes. From a LEO intercept perspective, this creates ambiguity: a waveform detected over the Indian Ocean could originate from any of several transmit sites in the network, and the ionospheric path uncertainty compounds the problem. Resolving individual sites within a networked OTHR architecture from commercial LEO assets alone is not currently achievable with published accuracy figures.

What is achievable is network-level activation status. If the waveform family is detected at all, the network is operating. If multiple distinct frequency allocations are observed simultaneously, more than one node may be active. Combining intercept data with the known geometry of documented transmit sites allows probabilistic attribution, not definitive geolocation, but probabilistic attribution against a finite set of known candidates is still analytically useful.

Waveform taxonomy: classifying what you intercept

Published OTHR literature and ITU-R M.1462 coordination documents describe the main waveform classes in enough detail to build a classification framework. Linear FM chirp (used by JORN and most Western OTHR systems) is distinguishable from stepped-frequency CW and from the interrupted CW variants used by some Russian-heritage designs. Chirp rate, instantaneous bandwidth, and the presence or absence of a guard interval between sweeps are the primary discriminants. Secondary features include sidelobe structure visible in the power spectral density and the coherent integration time implied by the dwell duration.

Automated classification using matched-filter banks or machine-learning classifiers trained on synthetic waveform models is the practical approach at LEO intercept timescales. The challenge is that ionospheric dispersion distorts the received waveform relative to the transmitted one, spreading the chirp in time and introducing frequency-dependent phase errors. Dispersion correction using a concurrent ionospheric model (from, for example, the IRI standard model or real-time assimilation products) improves classification accuracy but adds processing complexity. This is an active area of published research, and honest practitioners should acknowledge that classification confidence degrades significantly during ionospheric disturbance events such as geomagnetic storms.

Satellize's analytical work sits within this published framework. Where a client requires scheduled monitoring of known OTHR bands, we design collection tasking against documented frequency ranges and deliver waveform-classification reports with explicit confidence intervals, not binary detections.

Cloud, darkness and the HF advantage

One genuine advantage of HF monitoring over optical or SAR surveillance of the same targets is weather indifference. Cloud cover, rain and darkness are irrelevant to an HF receiver. The ionosphere itself is the variable, and it follows predictable diurnal and seasonal patterns that can be modelled in advance. Night-time F-layer conditions generally support longer-range skywave propagation than daytime D-layer absorption, which means OTHR systems often shift frequency downward after sunset. That predictable shift is itself a scheduling signature that a monitoring programme can exploit.

Typical figures

Frequency coverage (commercial LEO RF)	HawkEye 360: HF through VHF/UHF; exact HF lower bound not publicly specified. ITU-R M.1462 OTHR coordination bands: 5–28 MHz principal range.
Geolocation accuracy (line-of-sight VHF/UHF)	1–10 km (HawkEye 360 published figures, formation TDOA/FDOA)
Geolocation accuracy (HF skywave path)	Tens to hundreds of km; ionospheric refraction uncertainty dominates. Probabilistic attribution to known site set is the practical product.
Revisit cadence	HawkEye 360 cluster: several passes per day per location. Spire LEMUR-2 (100+ satellites): sub-hourly aggregate, short individual dwell.
Minimum detectable signal	Not publicly specified for commercial HF receivers. Physical limit set by receiver noise floor and ionospheric noise floor (ITU-R P.372 galactic and atmospheric noise).
Waveform discriminants	Chirp rate, sweep bandwidth (typically 10 kHz to several hundred kHz for OTHR), pulse repetition interval, dwell time, guard interval structure.
Archive depth (commercial RF)	HawkEye 360 commercial archive: available from 2019 onward for contracted customers.
Ionospheric correction dependency	IRI standard model or real-time assimilation products required for dispersion correction; classification confidence degrades during geomagnetic storm conditions (Kp > 5 indicative threshold).
Delivery formats	Waveform classification reports (PDF/structured data), frequency-time occupancy logs (CSV/GeoJSON), scheduled activation timelines.

Analytics Satellize can run

OTHR activation detection	Power spectral density monitoring across ITU-R M.1462 notified frequency ranges; threshold detection against ionospheric noise floor baseline.	Alert feed with timestamp, detected frequency band, and signal confidence score.
Waveform classification report	Matched-filter bank or ML classifier trained on synthetic FMCW/chirp waveform families; secondary features include chirp rate, bandwidth and dwell structure.	Per-intercept classification report with waveform type, estimated parameters, and confidence interval.
Operational tempo and scheduling analysis	Longitudinal aggregation of intercept timestamps cross-referenced against solar flux index (F10.7) and time-of-day; anomaly detection against baseline schedule.	Monthly scheduling pattern report; alert on statistically significant deviation from baseline.
Network activation status assessment	Multi-frequency simultaneous detection analysis; probabilistic attribution to documented transmit sites using known network geometry.	Network status summary (active nodes, probable coverage sectors) delivered as structured report.
Frequency agility tracking	Centre-frequency time-series analysis correlated with IRI-modelled optimal OTHR frequency for the intercept geometry; identifies reactive vs. scheduled frequency changes.	Frequency usage log with ionospheric context annotation; GIS layer of inferred coverage sector shifts.
Dispersion-corrected waveform archive	Post-processing of raw intercept data using IRI or assimilated ionospheric model to remove propagation-induced phase distortion before classification.	Corrected waveform dataset for client-side analysis; quality flag per intercept indicating ionospheric disturbance level at time of collection.

Sources

Who does the work

We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.