The camera pans in toward her face as an otherworldly sound begins pulsing, louder and louder. In the famous scene from the movie “Contact”, Ellie Arroway listens for aliens by plugging a set of headphones into the Very Large Array. In real life, the “Wow!” signal was found by a telescope called the "Big Ear,” and our most powerful current search is called "Breakthrough Listen.” But are SETI scientists really just listening for strange noises on the radio? 

Humans have pondered the existence of life beyond Earth since ancient times. In 60 BCE, Lucretius reasoned that “there must be other earths inhabited by different tribes of men and breeds of beasts.” A millennium and a half later, in 1440, Nicholas of Cusa theorized that the Sun, stars, and other parts of the heavens were likely inhabited. But another half millennium passed before realistic methods of communicating with them were proposed.

Marconi, Tesla, and other pioneers of radio technology soon recognized its potential for wireless communication between worlds. Indeed, Tesla was convinced he’d picked up broadcasts from Martians in 1899 (they may in fact have been natural radio signals coming from the interaction between Jupiter and its moon Io.) In 1924, when Mars was at its closest to Earth in over a century, radio operators across the world listened to their sets and picked up a variety of strange sounds.

“There is no proof that the signals come from another planet,” said Marconi. “No one can say definitively that abnormal sounds on the wireless originate on the earth or in other worlds.”

This problem of distinguishing between technosignatures (true signals from alien technology) and radio frequency interference, or RFI, continues to be the major challenge for modern SETI. The existence of radio transmitters engineered by humans motivates us to imagine that intelligence elsewhere might have followed the same path. But those same human-made transmitters cloud our searches with terrestrial noise. 

In 1960, a 29-year-old radio astronomer named Frank Drake, who had been working at the National Radio Astronomy Observatory in Green Bank, WV since obtaining his PhD from Harvard two years earlier, began Project Ozma, the first modern SETI search. Using the observatory’s 26-meter dish, Drake spent 150 hours scanning the stars Tau Ceti and Epsilon Eridani at frequencies around 1420 MHz. By comparing the signal from one feed horn pointed at the star being studied, and a second horn pointed off into space nearby, the Project Ozma radiometer was able to filter out local RFI, which would be likely to enter both horns with equal strength.

Next, the radiometer compared the signal strength from a filter designed to pick out a comparatively broad band of noise from the radio spectrum, with that from a filter designed to select a narrow (100 Hz wide) range of frequencies. Many astrophysical objects emit radio waves, including the vast amounts of hydrogen which permeate our galaxy, which radiates at 1420 MHz. Indeed, a year before Ozma, Giuseppe Cocconi and Philip Morrison published a paper, “Searching for Interstellar Communications,” which suggested searching for techno signatures at this “objective standard of frequency, which must be known to every observer in the universe.” Natural radio emissions such as those from neutral hydrogen cover a relatively broad swath of the radio spectrum, but only artificial transmitters are capable of producing signals narrower than a few hundred hertz.

The two filters employed by Project Ozma — a spatial filter to isolate signals coming from a particular direction in the sky, and a narrowband filter to select only engineered radio signals — still form the foundation of the majority of modern radio SETI experiments. However, despite reducing overall RFI volume, pernicious signals can still sneak through these filters. On its very first day of observations, Ozma detected an off-the-scale signal apparently coming from Epsilon Eridani. But after a few minutes, the signal disappeared. To try to confirm whether the signal was terrestrial or extraterrestrial, Drake pointed another small antenna (sensitive to a large area of the sky) out of the window of the control building. About ten days later, the Ozma radiometer and the smaller antenna both picked up the same signal. The team never determined its source, but it was consistent with a transmitter on board an aircraft that had passed close to the position of the star being studied. 

By the time the Breakthrough Listen initiative was launched in 2015, radio telescopes, and the instruments attached to them, had made huge leaps forward. Breakthrough Listen deployed its first digital instruments to two of the largest steerable dishes in the world — the 64-meter Parkes telescope, and the 100-meter Green Bank Telescope (located just one kilometer from the telescope used by Drake for Ozma.) But in contrast to Ozma’s single 100 Hz channel, Listen’s backend is capable of digitizing billions of 3-Hz-wide channels at a time. 

Unfortunately, artificial transmitters have also proliferated in the past five decades. At many of the frequencies observed by Listen, the instruments receive powerful radio signals wherever they point in the sky. A GPS satellite can be picked up by the antenna inside your cellphone, so when it gets anywhere near the direction in which a big radio telescope is pointing, the signal overwhelms the receiver. Bluetooth and wifi, satellites delivering phone and internet … noisy RFI is getting louder by the day.

But Drake’s filters can still help us find signals of interest. By pointing the telescope on and off its target star (either by using multiple feed horns or by physically moving the dish,) and by looking for narrow bandwidth signals, much of the interference can be rejected. Further, by throwing out signals that don’t change their frequency over time, we can even get rid of pernicious signals that survive the first two cuts. Any transmitter on a distant world would be expected to be moving with respect to our receiver in such a way as to introduce a drift in frequency due to the Doppler effect.

“Seeing a drifting signal is unusual for most Earth-based transmitters,” explains Sofia Sheikh, a radio astronomer at the SETI Institute who collaborates with the Breakthrough Listen team. So in 2020, when Sheikh, and Breakthrough Listen summer intern Shane Smith, spotted a signal from the Parkes telescope that was drifting in frequency, and only appeared when the telescope was pointed at the star Proxima Centauri, they dubbed it “BLC1” — Breakthrough Listen Candidate 1.

“BLC1 had a few features that made it stand out as unique from the millions of signals that we'd seen before,” says Sheikh. “As well as being narrowband and drifting, the signal persisted over several hours, which gave us lots of data, and a high confidence that we weren’t seeing something like a fast-moving satellite. The signal only seemed present when we were pointed at Proxima Centauri, so it appeared to be localized in the sky.”

Ultimately, though, the team determined that blc1 was RFI. “The key to solving the blc1 mystery was associating it with a group of similar signals at other frequencies, which all had the same shape and general properties,” explains Sheikh. “These signals showed up no matter where the telescope was pointed, implying that the transmitter causing them was on Earth's surface and that the drifts were being caused by some sort of electronic malfunction instead of true motion.”

When weird signals pass through the filters, the search for lookalikes that are obviously RFI can be laborious. But artificial intelligence might help vet future candidates. “Machine learning algorithms are really good at tasks like classification, so an ML classifier could likely find those mirror signals much faster than we did with traditional database methods,” says Sheikh. “Alternatively, if there were a set of true technosignatures in the data and we only found one with traditional methods, an ML classifier could find similar true signals with the same technique.”

University of Toronto student Peter Ma, an undergraduate intern with Breakthrough Listen, has developed a tool to do just that. Using a machine learning algorithm known as an autoencoder, Ma studies the way signals cluster in “latent space” — an abstract representation of signal morphology. Ma has used this technique to implement what amounts to a “reverse image search” that can be applied to radio SETI data, seeking lookalike signals that can help confirm or reject a candidate. Ma has also trained his autoencoder on simulated signals, and then set it loose on 120 terabytes of Green Bank Telescope data. The algorithm found eight signals of interest that had been missed by the classical approach.

But might there be more effective ways of rejecting RFI in the first place? While big single-dish telescopes like Green Bank have some unique capabilities, an increasing number of SETI experiments are using arrays of dishes for their searches. 

“Telescope arrays are perfect for localizing signals — and localization is a key feature of a true technosignature,” says Sheikh. “Arrays allow you to monitor multiple nearby stars at once with a technique called `beamforming’, taking data much faster than single dishes. And, if a signal is found in any one target, it can be compared to a simultaneous control sample by looking at the other beams on other targets. This is great for getting rid of any non-localized RFI (the majority of the signals that we detect in a SETI search), and extracting signals that could be true technosignatures.”

Chenoa Tremblay is spearheading this approach using the new “COSMIC” system on the Very Large Array in New Mexico. COSMIC — the Commensal Open Source Multimode Interferometer Cluster — combines signals from the VLA’s 27 antennas to form beams on targets of interest within the array’s primary field of view on the sky.

“Our SETI system on the VLA operates alongside regular operation of the telescope, without interfering with the scientists’ primary science goals and data. This is known as a commensal system,” explains Tremblay. “The advantage of this commensal operation is that we record and process a search for technosignatures whenever the telescope is operational. Many previous technosignature searches required dedicated time on large telescopes. This meant that surveys of a few thousand stars could take years of observation.

Now, with COSMIC, we are observing up to 3000 stars per hour and can operate when the telescope does; up to 24 hours a day, 7 days a week. The disadvantage is that we don’t control where the telescope is pointing or what frequency it is observing. As a commensal system, we are along for the ride.”

Systems like COSMIC, and a similar instrument deployed by Breakthrough Listen on the MeerKAT telescope in South Africa, are helping to sift through the RFI haystack more and more effectively, placing ever tighter constraints on the number and power of any extraterrestrial radio transmitters. But so far, a real technosignature remains elusive. Peter Ma’s eight signals of interest had gone away when the stars were reobserved. Ultimately, confirmation of a technosignature would require an independent team to point their telescope at the same target and confirm that the signal was indeed coming from that location on the sky, rather than from some pernicious source of local RFI.

But what if ET is using technologies other than radio? Although many modern SETI searches still use radio telescopes, a variety of other strategies are also employed, from looking for anomalous changes in brightness of stars that might indicate engineered megastructures in orbit, to searching for the characteristic signatures of interstellar laser communication.

Perhaps, though, they are not using electromagnetic waves at all, or have some technology that is so advanced as to be incomprehensible to humans. Be that as it may, we’ve sampled such a small region of our galaxy, over a short span of time, and a limited range of frequencies, that there is still much searching to be done using radio.

As Cocconi and Morrison said in their 1959 paper, “The probability of success is difficult to estimate; but if we never search, the chance of success is zero.”

For now, we need to keep “listening”.