TSS Block Diagram (9 kB image)
One of the major challenges facing any SETI program is radio frequency interference (RFI) from our own communications, radar, and navigation signals. To minimize the impact of strong local signals, the MCSA is composed of two layers of digital band pass filters (BPF) followed by a Fourier Transform. Each BPF divides its input bandwidth into approximately 100 smaller bands with adjacent filter bands isolated by more than 100 dB. This high degree of out-of-band rejection prevents strong signals from contaminating the entire observing band. The output samples from the second BPF are Fourier transformed to provide frequency channels with resolutions as narrow as 1 Hz. Such fine resolution is required for detecting continuous (CW) signals such as carriers.
In order to accommodate pulsed signals, the MCSA simultaneously performs multiple FFTs to divide the band into channels with three selectable widths (from six available widths - see the TSS Summary Table at the end of this report). This provides sensitivity to pulses with durations ("on" time) from 0.02 to 1.5 seconds. Since pulses are unlikely to be synchronous with the Earth-based clocks that define the sampling times for the MCSA, successive spectra are overlapped in time by 50%.
The channels in the MCSA overlap each other slightly in frequency, providing near-optimum response to both CW and pulsed signals whether they remain in a channel or drift in frequency by as much as one channel per spectrum.
The MCSA occupies two standard equipment racks with 8 commercial computer boards, 72 custom circuit boards (six types), and includes a custom Digital Signal Processing VLSI chip (384 DSP chips total). The MCSA has a sustained computation rate of approximately 75 GFLOPS (75 billion floating point operations per second).
The SDS Continuous Wave Detector (CWD) The CW signal detector analyzes the finest resolution output from each polarization of the MCSA separately, looking for continuous signals that have a signal to noise ratio in a single channel of 0.25 or greater. At frequencies up to 2 GHz the 1 Hz resolution is used. Above 2 GHz the 2 Hz resolution is processed in order to compensate for the larger Doppler drift range. At the 1 Hz resolution, the CW Detector receives 80 million spectral power measurements per second. (28.74 million channels per polarization every 0.714 seconds.) Since the time-bandwidth product of the data samples is unity, the data rate is the same at the 2 Hz and all other resolutions. The algorithm used to efficiently add the detected powers in each channel along all potential drifting signal paths in the frequency-time plane requires that all of the data be stored during the observation. This is accomplished with a redundant array of inexpensive disks (RAID). After the observation is complete, the stored data are processed through a set of four custom circuit boards that perform 3.2 billion additions per second. While these data are being processed, a second bank of RAIDs store the data from a new observation. The CW Detector reports any signal paths that have a summed power exceeding a pre-set statistical threshold.
The SDS Pulse Detector (PD) Since pulsed signals of the same average power as a CW signal will be relatively strong when they are "on", the data from the MCSA can be "thresholded" before processing. Only channels with power values greater than a predefined threshold are passed to the pulsed signal detector. The threshold is set so that in the presence of noise alone, only about one channel in 10**5 will pass. The reduced data set can be pictured as a sparse matrix in the frequency-time plane. The union of the data from both polarizations is stored on a 1 GB disk and a commercial i860 processor board is used to search for sets of three regularly spaced pulses in the data. Any pulse "triplets" with a summed power greater than a pre-defined statistical threshold are reported to the SCS.
The observer can interactively schedule a series of observations through a graphical user interface. The observer can then configure the subsystems by selecting the resolutions for the MCSA, setting thresholds, choosing subsets of data for display and storage, or accepting default settings. When an observation is about to begin, the SCS requests the observatory control computer to point the telescope at the target star and waits for confirmation that the telescope is tracking the star. The SCS then signals the rest of the subsystems to begin the observation. While the observation is under way, the observer can interactively change the frequency range and resolution of the data displayed. When the signal detection reports are received from the SDS, the SCS compares the reports against a database of known or previously observed interference signals. Signals that cannot be classified as interference are scheduled for further tests with the Follow-up Detection Devices.
In practice, this process is carried out simultaneously on FUDDs at two antenna sites. When the SCS determines that a signal reported by the SDS cannot be ruled out as interference, the signal characteristics are reported to both FUDDs. While the main TSS subsystems go on to a new observing frequency on the target star, the FUDDs tune to the frequency of the candidate signal and observe simultaneously. If the signal is persistent, the FUDD at the main antenna can quickly detect and improve the measurements of the signal characteristics. The improved parameters for the signal and the geometric transformation factors between the two sites are reported to the FUDD at the remote antenna site and used to form a matched filter for that signal. The filter is applied to the data that were collected simultaneously with the main site FUDD. If the signal is detected by both FUDDs, it is a very convincing candidate extraterrestrial signal.
The FUDD is implemented in a commercial Pentium PC chassis. A custom FFT board with Plessey FFT chips (PDSP16510) can process up to sixteen tunable 10 kHz bands within a 10 MHz bandwidth. The high resolution spectra and matched filters for each candidate signal band are generated by the Pentium processor. Two FUDD units are used at each site to cover the 20 MHz bandwidth of the TSS.
The TSS was transported to the Parkes Observatory, NSW, Australia at the end of 1994. A dedicated observing session using the 64-m telescope at Parkes and the 22-m telescope at Mopra began Feb. 2, 1995. With only a few interruptions for time-critical radio astronomy, the Phoenix observations continued tfull-time until June 6. During that time more than 23,000 observations were conducted (an observation is defined as successfully searching a 10 MHz bandwidth).
Since the Arecibo Observatory is still undergoing a major upgrade, the TSS returned to the SETI Institute for hardware and software upgrades. The current status is described elsewhere.
Our next observations will occur later in 1996 at the 140 Foot Telescope of the National Radio Astronomy Observatory in Green Bank, West Virginia. We are currently working with Georgia Tech to upgrade a former communications antenna near Woodbury, Georgia, to serve as the "remote site" for the Green Bank observations.
In subsequent years, we will return to the Arecibo Observatory in Puerto Rico. There, at the world's largest antenna, Project Phoenix will continue observing target stars visible from that site, extending the observations begun by the NASA HRMS in October of 1992.
Later in the decade, the TSS will visit the upgraded Nancay telescope in France and other large northern hemisphere observatories. These sites will allow observations of stars that are beyond the declination limits of Arecibo and Parkes.
Frequency Coverage 1 GHz to 3 GHz
Instantaneous Bandwidth 20 MHz
Polarization Dual Circular
System Temperature < 25 K for all telescopes
Channels per polarization
(any three simultaneously)
CW Signal Sensitivity Limit: S/N > 0.25 for a b*tau =1 sample
Pulse Signal Sensitivity Limit: S/N > (S/N)cw *3/sqrt(bT)
(where T = total observing time)
Signal Frequency Drift Range +/- 1 channel per spectrum
