...I would, perhaps incorrectly, expect at least some types of obvious overflow signal [tasks] to be fairly 'white'. It's been a long time since I looked at an autocorrelation function, from vague memory they involve a single convolution. Something like that would be able to judge the whiteness of the signal against a decided threshold dirac figure/function. I've used them in signal processing many years ago for analysis of buried periodic signals. Subtracting the autocorrelation of white noise from that of the source [ had interesting results].... but that was on a 1k node torus so algorithmic complexity and other practical considerations weren't an issue under much consideration . [though they should've been]...Jason
I've always thought of the WUs as consisting of a lot of white noise out of which we try to extract what isn't. Although the ALFA receivers are recent state of the art with 300 MHz. bandwidth, what goes on the Multibeam recorder is 2 bit complex values representing only the signs of the real and imaginary complex data points of the 2.5 MHz. bandwidth we're examining. That digitization represents a lot of noise, no? Then after the recording is received at Berkeley, the Splitter expands the data into complex single floats, does forward FFTs at 2048 length and inverse FFTs at length 8 and repacks the output in 2 bit form to be placed in the WUs.
As a practical matter, most overflows have been on Spikes until the recent radar RFI problem caused a lot of Triplets. The record for my Willamette P4 shows 1682 Gaussians, 1807 Pulses, 11764 Spikes, and 10718 Triplets. 8162 of those Triplets are from one data set we crunched at SETI Beta before the radar RFI was understood. That's from 3524 results. The thresholds are nominally adjusted to give about an even chance of finding a signal or not if the noise level is normal, the Gaussian and Pulse counts match that but the excess Spike count probably reflects cases where random RFI caused overflows. Note that Spikes aren't exactly what the name implies, they are found by examining the power spectrum of a single FFT; if there's one frequency that's more than 24 times more powerful than the average it's counted as a Spike. A continuous narrow band interfering signal will cause a sequence of Spike reports as the FFTs are stepped through the duration of the data. Joe
