During the search the Itasca deployed an experimental high-frequency direction finder on Howland Island manned³² by Coast Guard Radioman Second Class Frank Cipriani. This equipment was used on July 5 to obtain an approximate bearing on a signal near 3105 kHz.

Sandretto’s description³¹ of the PAA Adcock DF system illustrates the importance of signal strength and duration. A signal bearing was indicated by an aural null. But instead of measuring the null bearing directly, the operator observed bearings on each side of the null where the signal level was high enough for accurate measurement, over a period of 2 to 3 minutes, and averaged those bearings to obtain the null bearing. the accuracy of the bearings on each side of a null, and thus the accuracy of the average bearing, would be adversely affected if a signal was weak or of short duration.

The Adcock DF system produces a figure-eight antenna beam pattern, which is rotated via a goniometer to obtain a null, or signal minimum. the antenna pattern is symmetrical about the null axis, which allows bearing ambiguity because it is possible to get two null bearings 180 degrees apart. Adcock DF system designs of the 1930s usually incorporated a vertical sense antenna to eliminate bearing ambiguity. But some remarks in the PAA DF site logs indicate the possibility of bearing ambiguity, suggesting that PAA might have simplified their system design by omitting the sense antenna, since each DF site would know the general bearing of a Clipper flight. this is consistent with the recollection of Captain Almon A. Gray, USNR (ret), who was the Assistant Communications Superintendent for the PAA Pacific division during the Clipper era. Gray said³² of the PAA DF system:

The system was bi-directional hence one had to be ever mindful of the possibility of reciprocal bearings. That was not much of a problem however as one usually knew the general location of the aircraft.

This analysis assumes the PAA DF system was bi-directional and examines the possibility of a reciprocal bearing source in each case.

The distance from Wake Island to Midway Island is approximately 1,080 nautical miles (nmi), and the distance from Midway to Honolulu is approximately 1,150 nmi. thus a flight on the route between Wake and Honolulu would never be more than about 600 nmi from the nearest DF site. In contrast, the distances from Gardner Island to Wake, Midway, and Honolulu are 1,825nmi, 1,990 nmi, and 1,850 nmi respectively.

High frequency direction finding sites take bearings on skywaves, which travel long distances via refraction from the ionosphere. Short term variations in the ionosphere can cause signal strength fading, reducing the SNR and increasing the difficulty of getting an accurate bearing. If fading occurs on an already weak signal, the SNR can drop below the reception threshold, making it impossible to obtain a bearing. On the other hand, increasing atmospheric noise can cause the SNR of an already weak signal to drop below the threshold even without fading.

The PAA DF system³³ site locations, radio equipment, DF equipment, and operating procedures clearly were chosen to ensure accurate bearings would always be available for support of PAA flights.

The 70-watt output power of transmitter³⁴ on the Pan Am Clippers was roughly comparable to the 50-watt output of the transmitter on NR16020. However, as noted earlier, the distances from Gardner Island to the PAA DF sites were about 3 times the maximum distance from a Clipper to the nearest DF site on the route between Wake Island and Honolulu. Hence, the propagation loss for signals from Gardner Island was at least 10 times greater than for signals from a Clipper to the nearest DF site, making the SNR for any signals from Gardner Island correspondingly lower. And signals from the other potential sources, which were even further away, were likely to be much weaker than those expected from Clipper aircraft. the result, not surprisingly, was that bearings on potential post-loss signals were tenuous and approximate at best. that the DF site personnel were able to get any bearings at all in such conditions is a tribute to their skill and dedication.

Each PAA DF site³⁵ used two concentrically arranged sets of Adcock antennas, covering the frequency band 200 kHz to 6,000 kHz. Adcock antennas were used³⁶ because, although they are electrically equivalent to loop antennas, they are immune to the effects of polarization shifts³⁷ in radio waves refracted by the ionosphere.

A downcoming skywave, will have both vertically and horizontally polarized components which fade³⁸ independently of each other. the presence of a horizontally polarized component in a radio wave causes “night effect”³⁹ in a loop antenna by inducing voltages that do not cancel out when the plane of the loop is perpendicular to the bearing of the radio wave. this results in an inaccurate bearing, or an indistinct minimum, or both. the Adcock antenna, being insensitive to horizontally polarized waves, avoids this problem.⁴⁰

Notwithstanding its ability to obtain usable skywave bearings, the performance of the Adcock antenna is limited⁴¹ by the fact that its ability to extract energy from a passing radio wave is the same as that of a loop antenna with a single turn of wire, and hence is quite small. Consequently, even with the Adcock antennas at the DF sites, it was difficult to obtain usable bearings on weak signals.

Terrain interference of skywave signals at a DF site can reduce signal strength and cause bearing errors. the elevation angle of arriving skywave signals is a key consideration in this regard. This angle depends on the height of the ionospheric reflection point and its distance from the receiving site, which in turn depend upon date, time, signal frequency, and conditions in the ionosphere.

Wake Island, Midway Island, and Howland Island are essentially flat, so the DF sites there had no terrain interference. But the situation at the Mokapu Point DF site was quite different.

The Mokapu Point site was at approximately 21° 27′ 21″ North latitude, 157° 45′ 44″ West longitude, near the seaward edge of a large flat area in what later became the U.S. Marine Corps Air Station at Kaneohe Bay. the DF site coordinates were derived by correlating topographic map⁴² features with two photographs of the antenna array and associated buildings. One photo⁴³ was taken from offshore looking westward toward the Koolau mountain range and shows the DF site near the shoreline at a height of about 40 feet above the beach. The other photo⁴⁴ was taken from onshore looking north toward the sea.

The crest of the Koolau range on the great circle bearing of Gardner Island (214 degrees) is 6.75 statute miles from the DF site coordinates. the mean orientation of the ridge line⁴⁵ in that area is 340/160 degrees true, with some segments deviating from the mean by up to 50 degrees.

The southwest slope⁴⁶ of the Koolau range is a forested watershed with numerous ridges, valleys, and running streams. the terrain in this area, along the path of a signal arriving from Gardner Island, has a compound slope rising from sea level at Honolulu to about 1,800 feet over a distance of 6.3 miles at a vertical angle of approximately 3.1 degrees, and tilting downward to the right at a slope angle of approximately 10.8 degrees across the path to the DF.

The terrain on the northeast slope of the range drops from the ridge line to near sea level in about one mile, and remains essentially flat from there to the DF site, and beyond to the sea.

The elevation angle of the ridge line from the DF site coordinates, on the bearing of Gardner Island, is approximately 2.9 degrees, hence the DF site would be in the radio “shadow” of the Koolau range for skywave signals arriving with lesser elevation angles.

But even signals with somewhat greater elevation angles could encounter terrain effects. the lower edge of the wavefront would contact the terrain on the southwest slope, potentially sustaining directional skewing. Moreover, diffractive scattering from the ragged ridge line could cause signals to arrive at the DF site from multiple directions, making it difficult to obtain a reliable signal bearing.

The elevation angle at the DF site for signals on 3105 kHz arriving from Gardner Island and the Itasca ranged from 1.4 degrees to approximately 12 degrees at the times of the Mokapu bearings. Angles near the low end of this range were small enough to put the DF site in the radio shadow of the Koolau ridge. Angles near the middle of the range were small enough that, although the DF site was not in the radio shadow, there could have been significant terrain effects. Angles near the upper end of the range were large enough that terrain effects can be assumed to be minimal.

The bearing of the Itasca from the Mokapu DF site varied between 225 and 232 degrees on July 3 through July 6. the height of the Koolau ridge in this sector varies between 2,600 feet and 2,800 feet, and the average terrain elevation angle from the DF site is approximately 4.8 degrees. this may have been a factor in the unsuccessful attempt to get a bearing on the Itasca on July 6, discussed later.

The bearing of Midway from the Mokapu site coordinates is 295 degrees, and the Koolau ridge elevation angle on that bearing is 2 degrees. PAA flights on the Midway – Honolulu route operated only during daylight hours,⁴⁷ and their signals arrived at the DF site with elevation angles of at least 30 degrees. Hence, terrain was not a significant factor in bearing accuracy for normal PAA operations.

And the Mokapu site had an unobstructed “view” of the radio path of signals from PAA flights on the route to and from Alameda.

Clearly, the Mokapu DF site was well-positioned for obtaining accurate bearings for its designed mission of supporting flights to and from Midway and Alameda, but was operating at a potential terrain disadvantage in the search for post-loss Earhart signals. ICEPAC does not take into account the terrain interference effects described above, so the SNR values given in this paper for low-angle signals from Gardner Island or the Itasca should be regarded as upper-bound values, i.e., the actual SNRs could have been less than the values computed by the model.