The last piece was about infrasound in air — about organ stops with 18-meter pipes, difference tones that do not arise in air, and the low notes of a grand piano whose fundamental physically does not exist. One thread remained untangled at the end: ocean whales, whose infrasound travels thousands of kilometers without loss. The excuse was “the physics is different there,” and that was an honest excuse right up to the moment I decided to get into that physics.
It turned out to be about one specific layer of water, around a kilometer deep, that behaves like a pipe. It was discovered twice, almost simultaneously, by two hostile countries; conceived not for the navy but to rescue downed pilots; used to track submarines; and still serving as a detector of nuclear explosions. It also lets you hear icebergs and one whale singing in the wrong key.
The speed of sound in seawater depends on two things, and they pull in opposite directions. Temperature: sound travels faster in warm water, while temperature falls with depth — so sound should slow down as it goes lower. Pressure: it increases with depth, and sound speeds up again. Between these two mechanisms is a point where the speed is at its minimum. The depth of that point is the axis of the sound channel.
In warm waters, the axis lies at 750–1000 meters; in the Gulf of Alaska, just 100–200 meters; north of Kauai, nominally at 800. Then geometry takes over. A ray leaving the axis upward enters warmer water, speeds up, and bends back downward; a ray heading down speeds up under pressure and bends upward. Nothing reflects off the surface or the bottom — sound simply cannot escape: the gradients turn it back. The energy is trapped in a horizontal layer and spreads not as a sphere but closer to a cylindrical wave, losing orders of magnitude less along the way.
NOAA explains this layering through salad dressing: oil and vinegar in an unstirred sauce, except the oil is warm, less salty water. The metaphor works, but it hides the scale. Over a distance of 5000 kilometers, sound travels for nearly an hour. And arrives.
The first serious proposed use of the deep channel was rescuing pilots who had gone down at sea. The idea was elegant: the pilot drops a small charge into the water; three or more hydrophones on the channel axis, spaced horizontally apart, triangulate the burst. The system was named SOFAR — SOund Fixing And Ranging. Stations were built. They were never used for their intended purpose even once — instead, they became platforms for studying sound and noise.
The channel was discovered twice, nearly at the same time. Maurice Ewing and J. Lamar Worzel at Columbia University; Leonid Brekhovskikh in the USSR, working with explosions in the Sea of Japan. In 1944, Ewing and Worzel lowered a hydrophone from a sailing vessel while a second ship set charges at distances of up to 900 nautical miles. Earlier, in the Bahamas, a pound of TNT had been heard 2000 miles away, off West Africa. A quote from the first report: the end of transmission in the channel was “so sharp that it could not be missed even by the most unskilled observer.”
Then the war, already drawing to a close, decided everything. By its end, 4733 Allied ships had been sunk in the Atlantic; the Germans had lost 966 boats; and the Germans’ last argument was the snorkel — a diesel submarine beneath periscope depth, with a claimed range of 20,000 miles. Low-frequency sound was ideal against it: a boat emits narrow lines around 25–200 Hz, and narrowband analysis pulls them out of the noise.
In October 1950, Admiral Sherman ordered Bell Labs to begin work on low-frequency sonar. The contract with Western Electric: one million dollars. The project was called Jezebel, the analyzer LOFAR, its analysis bandwidth 1.5 hertz, operating in real time; the first working prototype was delivered in May 1951. In July 1951, six hydrophones were installed in the Bahamas. In April 1952, a group of admirals came to watch the “lofargrams”: a submarine was asked to change course, speed, and depth, but the admirals did not wait for the results — they had already seen the recordings and left for Washington to launch the next stage.
In June 1952, six stations were ordered; in September, already nine; in 1954, another ten were added, six of them on the U.S. West Coast. That same year, the system first received the name by which it is still known: SOSUS, SOund SUrveillance System. From committee report to first operational station: less than four years. For a project on the scale of a transatlantic cable network, that is obscenely fast.
The most revealing story about SOSUS is the one where it failed.
On March 11, 1968, the Soviet GOLF II-class submarine K-129 went down in the northwestern Pacific, with 98 people aboard. The sounds of its destruction were recorded by bottom hydrophones of the AFTAC system — an Air Force service created to register nuclear explosions, not to monitor navies. The world’s largest hydroacoustic network, built precisely for that task, said nothing. This is stated plainly in the U.S. Naval Institute’s Proceedings.
Using arrival times at spaced sensors, by May 1968 they had calculated coordinates near 40°06'N 179°57'E. A 2008 reconstruction based on the same recordings produced a chronology: three explosions inside the pressure hull, the first 62 seconds before the calculated missile launch, a signal duration of 1.5 seconds, energy equivalent to roughly 10 pounds of TNT. Then two R-21 missiles caught fire in their sealed silos, and the control system kept operating for another 8 minutes 36 seconds after the crew had, apparently, already died.
The Soviet search — submarines, ships, aircraft — found nothing. The Americans did. Five and a half years after the sinking, the CIA began Project Azorian: lifting part of the boat from a depth of 16,400 feet, roughly 5000 meters. The previous lifting record had been 245 feet. That gap is the scale by which the undertaking should be measured. The cover story was manganese-nodule mining under Howard Hughes’s name.
The ship reached the site on July 4, 1974, and a Soviet intelligence vessel immediately appeared nearby. It watched the “mining” proceed and saw nothing strange. In December of that year, a Soviet naval attaché told an American officer at a reception in Bethesda that they knew everything, adding: “if you return there, it will mean war.” By then, the American press had already been working the story for half a year — the leak came from the Los Angeles police and the FBI, not from spies.
The idea of measuring climate with sound was born from an arithmetic mismatch nobody wanted to notice for twelve years.
In 1960, a series of explosions off Perth, Australia, was recorded by hydrophones in Bermuda — roughly 19,820 kilometers, almost halfway around the planet. The result entered the literature as a triumph of the direct great-circle path. In 1988, Walter Munk and colleagues constructed a set of refracted geodesic rays from the blast site and got something unpleasant: the north-south temperature gradient associated with the Antarctic Circumpolar Current deflects the rays so strongly that they never reach Bermuda. The island lies in a geometric shadow. Munk wrote it himself — it was unclear what scattering process had pumped enough energy into the shadow zone for the explosions to be heard. In retrospect, the classic result turned out to be luck.
Munk drew from this a thought about climate, not acoustics. The ocean absorbs most of the atmosphere’s heat and therefore responds to warming more slowly, and more honestly, than air does. Warming of the ocean at a depth of 1000 meters is expected to run around 0.005 °C per year — almost impossible to measure with a thermometer from a ship, but possible through sound travel time: in warm water sound moves faster, and a Hawaii–California signal will arrive earlier if the water between them has warmed.
In 1991, an experiment was conducted near Heard Island: an acoustic source on a ship with a shaft through its hull transmitted coded signals at 57 hertz and 215 decibels, and they were picked up in every ocean basin on Earth, from California to Bermuda. Thirty-five tests, received on SOSUS arrays that had already been declassified. The eventual precision was better than expected: arrival time could be measured to 20–30 milliseconds over distances of 3000–5000 km, yielding the path- and depth-averaged temperature to ±0.006 °C. The fear that internal waves would smear the signal into mush did not materialize.
Next came ATOC — a two-year program meant to begin in spring 1994 with a budget of 35 million dollars. It began with letters to Congress.
In February 1994, a few weeks before anticipated approval from the National Marine Fisheries Service, two sperm-whale specialists from Dalhousie University posted online that the transmissions might deafen whales. Science News later noted that they had overestimated the potential for damage, but newspapers had already quoted them. The second source of scandal was linguistic: the permit application used the term “take” — a legal term covering a range from an animal’s death to a barely noticeable behavioral shift. Critics read it as “kill.”
The signal itself was more modest than its reputation: 75 hertz and 195 decibels, described by Science News as the volume of an ambulance siren at the pitch of the lowest cello notes, for 20 minutes every four hours. Ten percent of the budget went to biological monitoring. Sources on Pioneer Seamount and north of Kauai operated until 1998 and 1999; the same SOSUS arrays received the signals, along with a receiver near New Zealand and Russian scientists at a bottom receiver off Kamchatka. No obvious changes in marine mammal behavior or distribution were found; the statistics showed weak movement of humpbacks away from the source, as well as an unexpected result — during sound playback, the whales stayed at the surface longer and dove longer, contradicting the simple model of “stress → farther from the loud noise.”
Today, the channel works for an organization with nothing to do with sound: the CTBTO uses it to detect nuclear explosions. The International Monitoring System has 337 facilities, including 11 hydroacoustic stations — six hydrophone stations and five T-phase stations, seismometers on islands that hear sound when it reaches shore. A hydrophone station usually consists of two triplets suspended in the channel north and south of an island; the exception is the station at Cape Leeuwin in Australia, which has only one triplet, to the west.
The clearest demonstration of what these stations can do is the missing submarine San Juan. On November 15, 2017, it vanished roughly 600 km east of Comodoro Rivadavia, having transmitted its last message around 10:30 UTC. Two hydrophone stations — one on Ascension Island in the Atlantic, the other on the Crozet Islands in the Indian Ocean — recorded a short, broadband signal of unknown origin coming from the area of the submarine’s last known position.
Two weeks later, the Argentine Navy dropped a calibration charge into the ocean to test the stations’ capabilities at a point with known coordinates. Both heard the blast, and the source position was determined to within 39 kilometers of the announced drop point. On November 17, 2018, the submarine was found on the seabed at a depth of 900 meters — less than 20 kilometers from the point CTBTO had given for the November signal.
MH370 did not work out that way, and that too is worth keeping in mind as a measure of the method. In the official narrow time window, in the direction of the “seventh arc,” exactly one suitable signal was found — at Cape Leeuwin, and it was not corroborated at the Diego Garcia station, which calls its nature into question. The calculation is simple: a 200-ton aircraft traveling at 200 meters per second has the kinetic energy of a small earthquake, and hydrophones hear things like that from 5000 kilometers away. So the issue is not volume, but that something was missing from the signal.
Separate from all this serious work, the channel solved a sea monster. In 1997, autonomous NOAA hydrophones recorded a sound in the South Pacific that was heard by several sensors more than 5000 kilometers away, and whose frequency rose over the course of a minute. The source was roughly triangulated to 50°S 100°W and named Bloop. The answer: an iceberg. The sound matches an icequake, confirmed by tracking iceberg A53a as it broke apart near South Georgia in 2008. The legend of a giant creature grew from one researcher’s caveat at NOAA: he allowed that the sound profile resembled a living creature, but noted that the source was “much more powerful than the calls of any animal on Earth.”
Whales had been using this channel long before humans discovered it, and one of them entered history by accident.
In 1989, biologists at Woods Hole heard a call in the North Pacific at 52 hertz — two to three times higher than the usual 15–25 hertz for blue and fin whales. The sound returned in 1990 and 1991. And in 1992, the U.S. Navy partially declassified SOSUS and opened its recordings to oceanographic research, and then they began tracking this whale every year — for twelve consecutive seasons.
Each year, the signal was caught between August and December, then lost by January: the whale traveled from the California coast to the Aleutian Islands, covering 31–69 kilometers a day. Over a season, it covered anywhere from 708 kilometers to 11,062 kilometers in 2002/03. For comparison: ordinary tracking of an individual whale lasts hours. By 2004, the frequency of its call had fallen to 50 hertz — as it should in a maturing animal.
The press made it “the loneliest whale in the world,” and that is a bad metaphor because it contradicts the data. The uniqueness of the source rested on the absence of other recordings at that frequency, but since 2010, recordings of a second 52-hertz whale have been found sporadically elsewhere.
The ocean itself, in which all this can be heard, has grown noticeably louder over the past half-century. Ross predicted an increase in background noise of roughly 3 decibels per decade, and measurements only partly confirmed it: comparing 138 days from 2003–2004 with recordings from 1964–1966 near San Nicolas Island yielded 10–12 decibels in the 30–50 hertz band — those same 2.5–3 dB per decade. But above 50 hertz, the difference drops to 1–3 dB, while above 300 hertz the 1964–1966 level was higher than today’s — because of a daily component simply absent from the newer data. Nobody knows what that sound was.
In the ocean’s busiest areas, the total increase reaches 15–20 decibels, and then chemistry enters the game: acidification of the water from pH 8.21 to 8.10 changes sound absorption, and the ocean of the future may grow noisier on its own, without a single new ship.
There is one number for what this means to those who listen. For the North Atlantic right whale in Stellwagen Bank, the loss of communication space was calculated at 63–67 percent. A NOAA researcher compared it to a visually impaired person needing to move safely through a neighborhood beside a noisy airport.