HF direction finding networks: long-range geolocation and skywave

The HF band – 3 to 30 MHz – is the only part of the radio spectrum where a single transmitter can be heard thousands of kilometers away without satellite relay or any infrastructure beyond the ionosphere itself. That reach makes HF essential for long-range military communications, over-the-horizon radar, and maritime messaging. It also makes it the subject of the most technically demanding branch of passive radio geolocation: HF direction finding. Unlike VHF/UHF DF, where signals travel line-of-sight and bearings map directly to emitter azimuths, HF DF must contend with ionospheric propagation that bends, scatters, and splits signals before they reach the antenna. This article examines how multi-station HF DF networks are architected to overcome those challenges and produce reliable geolocation fixes on emitters operating beyond the horizon.

Skywave propagation: what makes HF DF hard

A VHF signal travels in a straight line from transmitter to receiver. An HF signal at the right frequency departs the transmitter at an elevation angle, enters the ionosphere, undergoes total internal reflection at the layer boundary, and returns to earth at a skip distance determined by the reflection height, the transmission angle, and the ionospheric electron density at the reflection point. The receiver sees the signal as if it arrived from the direction of the ionospheric reflection point – not from the transmitter itself.

This geometry has four consequences for DF systems. First, the observed azimuth at any station is the azimuth to the reflection point, not to the emitter – and the reflection point moves with the ionosphere. Second, the signal arrives at a non-zero elevation angle (typically 5–25 degrees for single-hop F2 propagation), which means a DF array calibrated for horizontal arrival will measure a systematically biased azimuth unless elevation angle is measured and corrected. Third, a single emitter commonly produces multiple signal arrivals at the same receiver: one via a single F2 hop, one via a two-hop path at slightly different azimuth, and sometimes a groundwave component at close ranges – each appearing as a separate bearing. Fourth, the ionosphere is time-varying: solar flux, geomagnetic activity, and local time drive large changes in layer height and electron density that shift skip distances and reflection points over minutes to hours.

Groundwave versus skywave in the HF band

At ranges below approximately 200–500 km (depending on frequency and ground conductivity), HF signals propagate primarily via groundwave – hugging the earth's surface without ionospheric involvement. Groundwave DF is geometrically equivalent to VHF DF: the signal arrives at low elevation, the bearing maps directly to the emitter azimuth, and accuracy of 1–5 degrees RMS is achievable with a well-calibrated array. Groundwave range decreases rapidly with frequency – at 30 MHz it barely reaches 100 km over average ground, while at 3 MHz it can extend past 500 km over seawater.

Skywave dominates beyond the groundwave range and enables the long-distance geolocation that makes HF DF strategically valuable. The transition zone – where both modes coexist – is the most difficult regime for DF, because arriving groundwave and skywave signals from the same emitter may differ in azimuth by several degrees due to the ionospheric geometry, and the DF software must classify each arrival mode before it can apply the correct geometric correction.

Network architecture: stations, timing, and data paths

A practical HF geolocation network requires a minimum of three DF stations with good angular geometry relative to the intended coverage area, a reliable communications backbone for exchanging bearing data and ionospheric parameters, and a central network management server that fuses station reports into position fixes. Each component places demands on the others.

Station spacing and geometry. Station separation of 200–800 km is typical for a network designed to geolocate emitters at ranges of 500–3000 km. Closer spacing reduces the baseline for triangulation and degrades fix accuracy; wider spacing risks losing simultaneous intercept of short-duration transmissions because propagation conditions may allow the signal to reach one station but not another. The stations should form a triangle with interior angles no smaller than 30 degrees when viewed from the center of the primary coverage area – elongated or collinear station geometries produce high dilution of precision (DOP) for emitters on or near the axis of the network.

Time synchronization. All stations must timestamp their bearing measurements to a common time reference with sub-millisecond precision. GPS-disciplined oscillators (GPSDO) provide the reference; each station's bearing processor applies the GPS second pulse to synchronize its sample clock and timestamps each bearing report with UTC time to better than 100 microseconds. The network management server uses these timestamps to associate simultaneous bearing reports into intercept batches – bearings that are not simultaneous do not necessarily correspond to the same transmission from the same emitter and cannot be meaningfully fused into a fix.

Communications backbone. Bearing data is compact – a single bearing report is under 100 bytes – but latency matters for real-time operations. A latency budget of under 2 seconds from signal intercept to fix publication is achievable over any IP-capable link (satellite, cellular, leased line), but links with variable latency (satellite VSAT, cellular in congested areas) require the fusion engine to handle late-arriving bearing reports from slow stations by holding the association window open for a configurable duration before computing the fix.

DF array technology for HF: wullenweber, adcock, and compact arrays

The antenna array is the most operationally constraining element of an HF DF station. HF wavelengths range from 10 m at 30 MHz to 100 m at 3 MHz, which means a physically large array is required for good bearing accuracy at the low end of the band.

Wullenweber arrays. The Wullenweber (also known as CDAA – Circularly Disposed Antenna Array) is the classic large-aperture HF DF array. A full-size Wullenweber has an outer element ring diameter of 300–900 m and provides bearing accuracy of 0.5–1.0 degree RMS across the full HF band. These systems were the backbone of Cold War SIGINT DF networks. They require large land areas and are fixed installations. Their primary advantage – besides accuracy – is that the very large aperture provides inherent discrimination between simultaneous signals arriving from different azimuths, reducing the effect of co-channel interference on bearing quality.

Adcock arrays. The Adcock DF array uses four or more vertical elements arranged in a cross or circular pattern at spacings of 5–30 m. Adcock arrays are direction-sensitive to vertically polarized signals only, which is an advantage for HF DF: horizontally polarized signals (including the unwanted sky noise contribution from horizontal polarization) are rejected. A compact Adcock (10–20 m diameter) provides useful coverage across the upper HF band (10–30 MHz); extending coverage below 10 MHz requires either larger element spacing or interpolation from an ionospheric model. Adcock arrays are used in mobile and tactical HF DF applications where a Wullenweber is not feasible.

MUSIC and superresolution processing. Modern compact HF DF arrays apply superresolution bearing estimation algorithms – MUSIC (Multiple Signal Classification), ESPRIT, or Capon's minimum variance – to extract bearing accuracy beyond the classical Rayleigh limit imposed by the array aperture. MUSIC, in particular, applies an eigendecomposition of the array covariance matrix to separate signal and noise subspaces, enabling bearing accuracy of 1–3 degrees RMS from an array whose aperture would classically limit accuracy to 5–10 degrees. The tradeoff is computational cost and sensitivity to array calibration errors – MUSIC requires an accurate array manifold measurement to perform near its theoretical limit.

Ionospheric correction: from observed azimuth to emitter bearing

Once each station has computed an observed azimuth for an intercept, the geolocation engine must correct that azimuth for the ionospheric geometry to recover the true great-circle bearing to the emitter. The correction process has three steps.

Propagation mode identification. The engine first determines the dominant propagation mode – single-hop F2, two-hop F2, or groundwave – by comparing the observed elevation angle (measured by the DF array if it has elevation capability, or inferred from the ionospheric model) against the expected elevation angle for each mode at the observed frequency. For skywave modes, the expected elevation angle for a single-hop path is approximately arcsin(2h/d), where h is the virtual height of the F2 layer and d is the range. If the measured elevation angle is consistent with single-hop geometry, the single-hop mode is selected.

Skip distance and reflection point computation. Given the propagation mode and the ionospheric parameters (virtual height h'F, critical frequency foF2), the engine computes the skip distance using the standard flat-earth approximation for ranges below 2000 km or the spherical-earth formula for longer paths. The ionospheric reflection point is placed at the midpoint of the transmitter-to-receiver path for single-hop propagation. The engine then computes the great-circle bearing from the station to the reflection point and the bearing from the reflection point to the transmitter.

Bearing correction and quality weighting. The difference between the observed azimuth and the computed bearing to the reflection point is the ionospheric correction. After applying it, each station reports a corrected bearing to the emitter along with a quality metric derived from the SNR, the elevation angle measurement uncertainty, and the consistency of the ionospheric model at the current conditions. The fusion engine weights each corrected bearing by its quality metric before computing the fix.

Multi-station bearing fusion and fix computation

With corrected bearings from three or more stations, the fusion engine computes a position fix. The standard algorithm for HF DF bearing fusion is the Stansfield estimator or its weighted generalization, which finds the geographic point that minimizes the sum of weighted squared angular residuals between the computed bearings from each station to the candidate point and the observed corrected bearings.

The fix computation outputs a position estimate and a covariance matrix that describes the fix uncertainty. The covariance matrix is projected to produce the 50% and 90% confidence error ellipses published to the analyst display. A fix with a circular 50% error radius under 50 km is considered high confidence for strategic HF geolocation; fixes with error radii exceeding 200 km are flagged as indicative of poor geometry, strong ionospheric disturbance, or multipath contamination.

Handling multipath and co-channel interference

Multipath – multiple propagation paths from the same emitter arriving at slightly different azimuths – is the primary cause of bearing quality degradation in HF DF. A station receiving a two-path arrival may report a bearing that is a weighted average of the two path azimuths, or it may oscillate between them as the phase relationship between the two arrivals changes over seconds. The fusion engine handles multipath by running a consistency check: if a station's reported bearing is incompatible with the best-fit fix position given the ionospheric model, the station is flagged as multipath-contaminated and excluded from the fix computation.

Co-channel interference – a different emitter transmitting on the same frequency simultaneously – produces bearing errors that the multipath filter cannot reliably distinguish from genuine multipath. The primary mitigation is temporal: short-duration transmissions (frequency-hopping, burst communications) are less likely to coincide in time with an interferer on the same hop frequency. The collection software should log the signal duration and duty cycle of each intercept; very long, continuous transmissions on active HF frequencies are most susceptible to co-channel contamination, and their fixes should carry wider confidence intervals.

Operational siting and network management

Beyond the technical architecture, the operational performance of an HF DF network depends critically on how stations are sited, maintained, and tasked.

Electrical noise environment. HF DF station performance degrades in proportion to the local man-made noise floor. Industrial zones, power transmission corridors, and urban areas introduce broadband noise that raises the minimum detectable signal level and reduces the effective intercept range. A rural site with a noise floor at the ITU Recommendation P.372 quiet-rural reference level provides 20–30 dB more sensitivity than a peri-urban site – equivalent to extending the intercept range by a factor of 3–5. Site surveys should characterize the noise floor across the full HF band at multiple times of day, because some noise sources (VDSL broadband, industrial equipment) are active only during business hours.

Array maintenance and recalibration. The calibrated array manifold is the DF system's most operationally sensitive asset. Mechanical changes to the array – element bending from wind loading, ground settling, vegetation growth near elements, and moisture ingress in cable runs – shift the measured phase and amplitude response away from the calibration table, introducing systematic bearing errors that may not be immediately apparent to operators. Scheduled recalibration every 90 days, supplemented by continuous monitoring using a reference signal from a known azimuth (a co-located calibration transmitter), prevents silent accuracy degradation.

Collection tasking and frequency coordination. An HF DF network must coordinate its collection tasking carefully, because the HF band is shared with civilian services and the network's own receivers are susceptible to intermodulation from strong local transmitters. The collection manager assigns frequency monitoring tasks to stations based on coverage geometry: a task that benefits from high SNR at a southern station (emitter is in the south) may yield poor bearing quality at northern stations that receive the signal via a weaker multi-hop path. Adaptive tasking – routing collection tasks to the subset of stations most likely to achieve high-quality intercept – improves fix quality without adding hardware.