Based on "A DSP Engine for a 64-Element Array"
By Steven W. Ellingson,
Ohio State University ElectroScience Laboratory
7/13/99

[Note that the above link is to a PDF file requiring the Adobe Acrobat reader (available at no charge from www.adobe.com).]

Supported by a grant from the SETI Institute, the ElectroScience Laboratory at The Ohio State University has been conducting research on a new, state-of-the-art digital processor which will analyze data from an array of antennas. The antenna array will be part of the proposed Argus system, the next-generation radio telescope which will be able to view the entire visible sky at once through the use of digital beamforming.

In digital beamforming, the signal output from each of the antennas in the array is converted from analog format (i.e. continuously varying electrical voltage) to digital format (computer ones and zeroes), and an extremely accurate digital time stamp is added. (The time stamp allows calculation of the direction of arrival, since radio waves from a particular signal arrive at some of the antennas just fractions of a second before they arrive at others.) The digital samples may then be added together in any number of combinations to form digital beams -- the digital equivalent of where the telescope is "pointing".

With today's supercomputer-on-a-chip technology, it is possible to perform millions of these digital beam combinations very quickly, thus generating a radio picture of the entire visible sky without having to physically move, or point, the antennas. Because traditional radio telescopes must be pointed in only one direction at a time, it is possible for a transient (short-lived) radio signal in another part of the sky to be overlooked. The Argus telescope, using software-defined signal processing, will greatly improve the odds of transient signal detection, and will cost less to build than traditional steel structures due to the falling cost of computer hardware. (Labor costs to build massive steel structures are also on the rise.)

Applications for the Argus system can be found in radio astronomy, the search for extraterrestrial intelligence (SETI), and parasitic bistatic radar (i.e., imaging objects passing overhead via a bounced radar signal). In SETI, for example, it is desirable to detect very weak, intermittent narrowband (narrow frequency range) signals with no advance knowledge of either the frequency or the direction of arrival.

The digital signal processing experiment set up at The Ohio State University simulated the output from an 8 by 8 rectangular array of antennas (64 antennas in total). The simulated data was fed into an electronic circuit board containing two digital signal processing chips, called "SHARC"s, manufactured by Analog Devices, Inc. The onboard digital signal processing (DSP) chips have what is called a "multiprocessor memory space", a memory area which is accessible by both of the DSP chips on the board. The DSP chips also have a great deal of internal memory, which allows complete sets of data to be brought into the chip at one time for processing without the chip having to spend a lot of time waiting for parts of the data to arrive.

The researchers at OSU's ElectroScience Lab quickly realized that it took a lot of time for a chip to acquire a simulated data set, in fact so much time that there would be little time left over to process the data before the next data set arrived. Therefore, they came up with the idea of "rotating acquisition". In this method, one chip acquires the data, and another chip processes the data, and vice versa. One chip acquires the data and places it in the common memory area. The other chip then retrieves the data from the common memory area, processes it, and sends it out for storage. The chips take turns acquiring and processing the data.

The Ohio State researchers also figured out very rapidly that the DSP chips had a finite amount of processing ability within a given time due to the length of time required for performing the complex mathematical computations necessary for analyzing the data. They discovered that if the antennas were sending data at the rate of, for example, ten million samples per second, then the frequency range in the analysis would have to be cut way back to perhaps only 20 kHz (20 kiloHertz) to allow time for the processing. This means that a transient radio signal at a frequency outside of the frequency range being analyzed would be overlooked. Conversely, if the antennas sent only thousands of samples per second, then a much larger frequency range, perhaps 10 MHz (megaHertz) could be used. However, this would mean that there was a wait time, or rest phase, for each antenna between samples. This would allow a short-lived signal to be overlooked during the wait time. Clearly, there was a tradeoff between how often the antennas were taking samples, which the researchers called "duty cycle", and frequency range. The only way to allow millions of samples per second over as wide a frequency range as possible would be to add many more DSP boards at increased cost. Assuming that a large amount of money for such a system might be obtained, this processing ability is theoretically feasible.