How SETI@home works

What is SETI@home Looking For?

So, what will you be doing for us? What exactly will you be looking for in the data? The easiest way to answer that question is to ask what we expect extraterrestrials to send. We expect that they would want to send us a signal in the most efficient manner for THEM that would allow US to easily detect the message. Now, it turns out that sending a message on many frequencies is not efficient. It takes lots of power. If one concentrates the power of the message into a very narrow frequency range (narrow bandwidth) the signal is easier to weed out from the background noise. This is especially important since we assume that they are far enough away that their signal will be very weak by the time it gets to us. So, we're not looking for a broadband signal (spread over many frequencies), we're looking for a very specific frequency message. The SETI@home screen saver acts like tuning your radio set to various channels, and looking at the signal strength meter. If the strength meter goes up, that gets our attention.

Another factor that helps reject local (earth-based and satellite-based) signals is that local sources are more or less constant. They maintain their intensity over time. On the other hand, the Arecibo telescope is fixed in position. When SETI@home is in operation, the telescope does not track the stars. Because of this, the sky "drifts" past the focus of the telescope. It typically takes about 12 seconds for a target to cross the focus (or "target beam") of the dish. We therefore expect an extraterrestrial signal to get louder and then softer over a 12 second period. Since we are looking for this 12 second "gaussian" signal, we send you about 100 seconds of data. Also, we allow the data in the work-units to overlap a little so we won't miss an important signal by cutting it off early in the analysis.

See the section on RFI (Radio Frequency Interference) for more details.

Let's look at and some examples. If you have RealAudio, you can also listen to simulations of what it might sound like (though remember that these signals are radio waves, not sound waves…)

Click graph for RealAudio sound

This graph (typical of the others below) shows time progressing along the horizontal X-axis. The vertical Y-axis represents the frequency, or pitch, of the signal. Here you see a broadband signal. Many frequencies all mixed together. Note that the signal starts out weak (dim) at the left, and gets louder (brighter), reaching a maximum in the center of the graph 6 seconds later, fading out over the next 6 seconds. This is what we would expect from an extra-terrestrial signal as it drifts past the telescope. Unfortunately, we are not looking for broadband sources. This is probably what a star or other natural astronomical source would look like. Broadband sources are rejected.

This is more what we're looking for. Here you can see the signal is much narrower in frequency range. It also gets stronger and weaker over a 12 second period. We don't know how narrow the bandwidth will be, so we'll check for signals at several bandwidths.

Click graph for RealAudio sound

If our stellar friends are trying to put actual information on their signal (very likely), the signal will almost certainly be pulsed. We'll be looking for this too.

Because planets rotate, both the extraterrestial transmitter and our telescope are movin in circular paths around the axis of their planet. This motion shows up as a changing relative speed during the course of the observation. Because of this, there is likely to be a "doppler shifting" or changing frequency, of the signal because of our relative motions. This might cause the signal to rise or fall in frequency slightly over the 12 seconds. These are called "chirped" signals. We'll check for this too.

Click graph for RealAudio sound

Of course, we'll also be checking for a doppler shifting (chirped) signal that contains pulses too!

Extra Credit Section: More detail on the analysis

The SETI@home software searches for signals about 10 times weaker than the SERENDIP IV search at Arecibo, because it makes use of a computationally intensive algorithm called "coherent integration." No one else (including the SERENDIP program) has had the computing power to implement this method. Your computer performs fast fourier transforms on the data, looking for strong signals at various combinations of frequency, bandwidth, and chirp rates. The following steps are taken on each of the work-units you get from us.

Let's look first at the most computationally intensive portion of the calculation. The first job is to "de-chirp" the data - that is, to remove all the effects of the doppler acceleration. At the finest resolution, we have to do this a total of 20,000 times, from -10 Hz/sec to +10 Hz/sec in steps of .002 Hz/sec. At each chirp-rate, the 107 seconds of data is de-chirped and then divided into 8 blocks of 13.375 seconds each. Each 13.375 second block is then examined with a bandwidth of .07 Hz for peaks (that's 131,072 tests (frequencies) per block per chirp rate!) This is a LOT of calculation! In this first step, you computer does about 200 billion calculations! We drop a few tests between 10 Hz/sec and 50 Hz/sec.

We're not finished, we still have to test other bandwidths too. The next step doubles the bandwidth to 0.15 Hz. We only have to examine 1/4 the number of rates due to the increase in bandwidth. We end up doing about 1/4 the amount of work we did above at the highest resolution narrow bandwidth, or about 50 billion calculations. Piece of cake...

The next step doubles the bandwidth again (from 0.15 to 0.3 Hz) and again reduces the chirps by 1/4. This step (and all successive steps) take 1/4 the calculation of the previous step. In this case only 12.5 billion calculations. This continues for a total of 14 doublings of bandwidth (0.07, 0.15, 0.3, 0.6, 1.2, 2.5, 5, 10, 20, 40, 75, 150, 300, 600, and 1200 Hz) to bring you to a grand total of slightly more than 275 billion operations on the 107 seconds of data. As you can see we actually do most of our work at the narrowest bandwidth (about 70% of the work.)

Finally, signals that show a strong power at some particular combination of frequency, bandwidth and chirp are subjected to a test for terrestrial interference. Only if the power rises and then falls over a 12 second period (the time it takes the telescope to pass a spot in the sky) can the signal be tentatively considered extra-terrestrial in nature. Spikes (shorts radio bursts) above a threshold value are also recorded.

Finding pulsed signals

All the above work is done just to find a continuous extraterrestrial signal - one that's "always on", or an intense pulses. What if our alien friends are trying to signal us with a series of regularly spaced pulses? That means that we must also look for a signal that regularly varies in power over time. SETI@home applies two different tests. One of these tests looks for pulse triplets that are relatively strong. The other test looks for lots of equally spaced, but weak pulses.

The triplet test is pretty simple. For every frequency slice in the spectrum, the computer looks for pulses above a certain threshold value. This threshold is set at a reasonable value so as to reveal a reasonable number of pulses, yet not overwhelm us with noise that would cause us useless calculation. For every pair of pulses above the threshold, the screensaver looks for a pulse exactly in between the two. If it finds one, the screensaver logs it and sends the data back to Berkeley. The screen saver does this intelligently, so that it does not repeat itself in trying ALL pairs. Remember that it must do this for EVERY frequency slice in your 10kHz sample!

The second pulse detection method is a little more complicated. It was developed specially by the SETI@home team just for this purpose. It's called a "fast folding algorithm." Again we do this analysis on every frequency slice in your 10kHz piece of data. This method was designed to detect lots of small repeating pulses in the data. These small pulses may be weak enough to seem lost in the noise and be undetectable. We start by selecting a frequency slice from the data set to examine. We now look at this power vs. time data for the pulses. The screensaver slices up the data into uniform sized time chunks and adds the chunks all together. If the size of the time chunk is the same as (or a multiple of) the period of the pulses, all the pulses will add one on top of the other and we will see the pulses grow out of the noise. The hard part here is that we must guess the correct size for the time slice. Since we have no idea what the frequency of the pulses might be, we have to try ALL the various time periods. Again, the algorithm does this in such a way that it will not repeat work already done. If repeating pulses are found, they are logged and sent back to Berkeley. How long should all these computations take? An average, current model home home computer should take between 10 and 50 hours to complete one work-unit. This assumes that the computer ONLY works on SETI@home.

Depending on how the telescope was moving when the work unit was recorded, your computer will do between 2.4 trillion and 3.8 trillion mathematical operations (flops or floating point operations in technical jargon) to complete its work.

Now you know why we need your help!