Initiating A Spectrometer Development Program for GALFA

GALFA Meeting, Arecibo, March 21-23 2003

Executive Summary

The GALFA Consortium requests that NAIC immediately commence development of a prototype spectroscopic system for Galactic astronomy using ALFA. The starting point should be the software Fourier Transform PC based system being developed by Jeff Hagen at NAIC. We suggest that such a prototype system will resolve a number of issues related to this technical approach, and clarify cost and path issues for the full system that will allow commensal observing with ALFA. We here also outline some of the necessary steps required for developing a system for spectroscopy of HI and of radio recombination lines.


The Galactic ALFA Consortium (GALFA) has a range of interests falling under the umbrella ofGalactic astronomy. These include atomic hydrogen (HI), radio recombination lines (RRL), and continuum observations. For all of these, ALFA will have a dramatic impact, if the system is equipped with data taking back ends (spectrometers) appropriate for the various scientific areas. During the first meeting of the GALFA Consortium at Arecibo in March 2003, this problem was discussed at length. Di Li assembled the various "spectrometer requirements" into a compact Table, but in various discussions, culminating in a telecon on 6 May, it became evident that there was a high degree of agreement on the urgency of having NAIC get started on a plan to develop a prototype of a proposed "all-software" spectrometer system. Here, we briefly describe this concept and outline the steps that should be taken in the immediate future to assess the viability of this approach.

[Note: we do not discuss the continuum backend, as this will require a 300 MHz bandwidth system. The unit developed for pulsars could certainly be used, but likely not in a commensal fashion, given that the pulsar observations will require a coherent integration time of several minutes, while continuum ISM work benefits from very rapid telescope scanning, with integration times on the order of seconds.]

One of the distinguishing aspects of the HI portion of the GALFA science is that the Galactic HI emission is completely contained in a bandwidth of a few MHz (1 MHz corresponds to 210 km/s). This makes GALFA an ideal venue for exploiting the enormous development of general purpose computers for signal processing. In discussions at the Arecibo GALFA Consortium meeting, it became clear that Jeff Hagen's data acquisition system for ionospheric radar showed that a modest PC can carry out sampling and direct Fourier transform on an input signal having a bandwidth of a few MHz. It was not entirely clear exactly what overhead results, nor how far the bandwidth can be extended, but the basic reasonableness of the approach seems clear.

The ALFA Consortium proposes that NAIC initiate a program to develop a prototype test-bed system for assessing how well this design approach could satisfy the need for (1) a HI spectrometer for GALFA, and (2) serve as a building block of a RLL GALFA spectrometer. We feel that it is imperative that this be done in the very near future, so that there is the possibility of effective commensal observations. One great virtue of this approach for a GALFA spectrometer system is that the purchased hardware includes a sampler board and PC for each pixel, but there is no special- purpose equipment that needs to be designed or constructed. Thus, once the prototype has been proved to work satisfactorily, it should be very easy to replicate it. This means that even at this late date it is realistic to imagine a GALFA spectrometer system working commensally with the WAPP or other system, and that achieving this would place only relative modest demands on the NAIC technical staff. Also, the total data rate is sufficiently low that there will not be a huge impact either on moving data around or on archiving it.

Development Process and Schedule

The outline of what we feel should happen is as follows

(1) May - October 2003. During this period, a prototype unit will be assembled, programmed, and tested. Among the key parameters to be quantified are:

   a) efficiency (fraction of dead time) as function of bandwidth
   b) digitizer options and implications
   c) input filtering options
   d) possibility of handling multiple subbands
   e) stability
   f) cost model

(2) November 2003 - January 2004. During this period the following tasks will be carried out

   a) The implications of phase 1 will be analyzed and overall cost for the full GALFA system developed.
   b) The prototype unit will be used for astronomy.
   c) The software will be refined and updated based on experience with the unit being used for observing.

(3) February 2004. The full complement of hardware required will be ordered.

(4) February - May 2004. The design of a multiple frequency-channel unit for RRL research will be pursued. There are approx. 11 RRLs in the ALFA front end frequency range, and system sensitivity would be maximized by observing them all simultaneously. Especially for studies of the diffuse ISM, this may well be essential in order to allow commensal observing with pulsar searching. Each RRL will require a 3 MHz bandwidth, so it is conceivable that increases in PC processing power will allow two (or more) RRLs to be handled by one computer. The RRL unit will likely involve a number of down converters and local oscillators (but at least the latter can be shared among the 7 pixels), and will likely also require a modest factor larger number of computers for the raw spectrometer power needed.

(5) April 2004. At this time the hardware should arrive, and the actual assembly of the single-line multi-PC GALFA spectrometer undertaken. At present, it appears that a single PC can cover required bandwidth for two polarizations, so the spectrometer would include 7 PCs. It would thus occupy (more or less) one large equipment rack.

If a bandwidth of 8 MHz (1680 km/s) is achieved with 1 kHz frequency resolution in each of two polarization, each PC outputs approximately 8000 (channels) x 2 (polarizations) x 4 (bytes/dump) x 1 (dump/s) = 64 kbyte/s. The 7 PCs produce an aggregate data rate of 448 kbyte/s. This data will have to be collected and sent to appropriate central storage unit. This task, while modest compared to that for other ALFA backends, must be planned for and executed in a timely manner so that the data reduction software can be properly tested.

(6) May - August 2004. During this period, the HI GALFA spectrometer system will be tested initially with inputs from the single pixel L-band receiver initially, but shifting to using the ALFA front end as the signal source when this becomes available. During this period, the results of test observations carried out with single pixel receivers, can evolve into the first pilot projects for debugging the entire multibeam system.

(7) September 2004 - January 2005. The multi-line version of the system for RRL work will be designed, and prototype units will be tested. Much of this effort will focus on straightforward channelization and down conversion tasks, but this system will have to be appropriately designed and built to minimize self-generated RFI, as well as other problems. Ordering PCs, programming them, and integrating the whole system together will be non-trivial, but will not require any new technology development.


The above is still a relatively schematic task outline, but we present to demonstrate that in order to get the basic HI system ready for the ALFA system first light, prototype testing must begin this month. Much time has already passed since the need for multiple spectrometers became evident. For GALFA, this has not been entirely wasted, inasmuch as Jeff Hagen has been continuing work on the ionospheric system which is, to a significant degree, the test bed for this concept. However, we must now turn the focus to the GALFA system needs, and embark on the significant list of tasks indicated here. The hardware cost is expected to be very moderate, less than $50,000. If the prototype test results (October 2003) are positive, we see the PC/software spectrometer system as an efficient and very cost-effective approach to get first class data from the multibeam system as soon as the front end is ready to use for astronomy.