How NASA Salvaged Galileo's Jupiter Mission After Its Main Antenna Failed

In April 1991, Galileo was eighteen months into its voyage toward Jupiter when mission control sent the command to unfold the spacecraft's main antenna. The 4.8-meter dish, built like an umbrella with wire mesh stretched across eighteen ribs, was supposed to spring open and become the spacecraft's primary link to Earth. Instead, the motors stalled. Three of the ribs had stuck fast to the antenna's central tower, and no amount of coaxing would free them. Engineers would later determine that a dry lubricant, likely shaken loose during cross-country truck transport before launch, had left the ribs' pins and sockets vulnerable to friction. Over the following years, the team tried everything: they warmed and cooled the antenna by rotating the spacecraft toward and away from the Sun, they cycled the deployment motors on and off more than thirteen thousand times, they spun the craft up to ten rotations per minute. Nothing worked. Galileo would have to complete its mission with a backup antenna never meant to carry the full weight of the science.

The consequences were immediate and severe. The high-gain antenna had been designed to transmit data at up to 134,400 bits per second—roughly one image per minute. The low-gain antenna, by contrast, broadcast in a wide, weak beam. NASA's engineers compared it to the difference between a spotlight and a bare bulb. The signal reaching Earth would be ten thousand times fainter. Without any improvements, the low-gain antenna could manage only eight to sixteen bits per second from Jupiter's distance. To grasp what that meant: an uncompressed image from Galileo's camera was 800 by 800 pixels at eight bits each, totaling about 5.1 million bits. Transmitting a single image at ten bits per second would take nearly six days, and that was before any other instrument or engineering data got a turn. The spacecraft's computers made the problem harder still. They were late-1970s designs with 8-bit and 16-bit microprocessors and less than one megabyte of total memory. Any software solution had to fit inside that constraint.

NASA formally approved and funded a rescue plan in January 1993, which engineers called the S-band mission. No single fix was sufficient, so the team stacked multiple solutions on top of one another. First, they rewrote Galileo's flight software, loaded in stages during 1995 and 1996, to let the spacecraft edit and compress data before sending it. Second, they deployed stronger error-correction coding, a more powerful scheme that allowed ground receivers to recover data from signals too weak and noisy to read otherwise. Third, they upgraded Earth's listening apparatus, combining signals from up to six radio dishes—including antennas at Canberra and Goldstone and the 64-meter Parkes telescope in Australia—and fitting the Canberra 70-meter dish with a quieter receiver. Fourth, they replanned the science itself, having Galileo record data from each close moon flyby onto its tape recorder, then spend the rest of each orbit slowly editing, compressing, and transmitting it home. JPL's engineers estimated that the communications improvements alone would lift the average transmission rate about tenfold, and compression would add roughly another tenfold gain in useful information.

The compression method was the heart of the solution. JPL engineers chose an integer cosine transform, a close relative of the discrete cosine transform behind the JPEG format. The image was cut into eight-by-eight pixel blocks, and each block was rewritten as a mix of simple patterns, from smooth gradients to fine stripes. Smooth areas needed only a few patterns to describe them. The encoder then rounded off the fine-detail patterns that mattered least—that was where the savings came from, and where the losses did too. A standard cosine transform required multiplication by awkward decimal numbers, expensive for Galileo's processors. The integer version swapped those numbers for small integers that were powers of two or the sum of two powers of two. A computer could multiply by eight by shifting a binary number three places, multiply by six by adding a value shifted twice to the same value shifted once. The transform avoided multiplication entirely, using only shifts and additions. The results were indistinguishable from a standard cosine transform.

But lossy compression had traditionally been avoided on deep-space missions, because each image was irreplaceable. Galileo's scientists needed a way to decide how much loss was acceptable before the spacecraft started throwing information away. In a 1994 NASA study, fifteen evaluators from twelve institutions involved in Galileo's camera experiment viewed pairs of compressed images side by side and rated how useful each would be. The study found no single acceptable compression ratio. Scene content and simulated radiation noise—the kind of speckling Jupiter's harsh environment caused—changed how much compression evaluators considered useful. A fairly uniform dark scene tolerated about eighty-five to one. In the most extreme case, adding simulated noise cut the acceptable compression for one image from fifty-seven to one to less than three to one. The team built in safeguards: scientists could mark a ninety-six-by-ninety-six-pixel area anywhere in an image to be sent without lossy compression, preserving key details and giving them a reference for checking how faithful the rest of the image was. Synchronization markers every eight image lines limited error propagation and allowed decompression to resume at the next marker. The software could also drop less valuable data, such as the dark background of space around a moon, before compression even began.

Galileo began sending compressed images and data from Jupiter orbit in 1996. Before arrival, NASA had expected the reworked mission to meet at least seventy percent of its original science goals. JPL's technical account later confirmed that the expected gains were achieved and that the orbiter completed seventy percent of the original primary mission's objectives. One of its most important results didn't come from a camera at all. Galileo's magnetometer measured the strength and direction of magnetic fields—a stream of numbers tiny next to an image. As Europa orbited inside Jupiter's magnetic field, that field could induce an electric current in anything conductive near the moon's surface, which in turn created a weak magnetic field of Europa's own. The direction of Jupiter's field at Europa changed predictably. During a January 2000 flyby, it pointed the opposite way from earlier passes in 1996 and 1998. UCLA's Margaret Kivelson and her team predicted that if a conductive layer such as salty water existed beneath the ice, Europa's induced field should flip too. Galileo's measurements matched that prediction. Project scientist Torrence Johnson called the magnetometer data "the only indication we have that there's an ocean there now," while noting that the evidence was still indirect. A failed antenna had forced the mission into an exercise in priorities, deciding in advance what each bit was worth. Among the results returned through that narrow link were magnetic-field measurements that strengthened the case for a salty ocean beneath Europa's ice.