Hi Jonathan/Dan,
Yes, at the ATA we use a stripped-down correlator for calibration for the beamformer. Each beamformer has 12 leaf-nodes of 8 antpols each, each leaf node has a 1x8 correlator to calibrate phases/delays within that node. Successively downstream nodes (branch nodes) have 1x3 correlators to correlate subbeams. This method provides a simple way to calibrate on strong sources, but it does take longer than it would if you had all baselines present, and also it's vulnerable to reference-antenna problems (e.g., in the "1x8", if the one reference antenna is bad for some reason, all cross correlations will be bad). I've slimmed down the ATA beamformer firmware quite a bit, and now we do have a model that compiles for 8x8 in the leaf nodes and 3x3 in the branch nodes. This ought to improve calibration quite a bit, but I've not yet had time to redo the control software to implement these changes. Billy ________________________________ From: [email protected] [mailto:[email protected]] On Behalf Of Dan Werthimer Sent: Saturday, January 08, 2011 5:31 PM To: [email protected] Subject: Re: [casper] Correlating subsets of antenna inputs hi jonathan, as you pointed out, for a phased array feed, where the correlator is used only to calibrate the beamformer coefficients, you don't need to compute all the possible cross correlations, and you don't have to do the computation it in real time. the calibration time scale requirement depends on wavelength, and atmospheric and instrument fluctuations. for the ATA i think billy barrot implemented a 1 by 8 correlator that could switch around to correlate any 8 antennas against a master antenna (a 1 by 8 correlator). it took a while to get all the calibration cross correlation measurments done this way, but it worked well. i think now, billy has implemented a slightly larger correlator to speed the calibration up. best wishes, dan On 1/8/2011 2:28 PM, Jonathan Landon wrote: Has anyone created a correlator that only correlates overlapping subsets of the antenna inputs? Suppose I have 32 inputs and want to correlate 1-16, 8-24, 16-32. For a large phased array feed it may only be necessary to correlate relatively near neighbors. The benefit of correlating subsets comes if it's easier to build three 16-input correlators than one 32-input correlator. If it's true that only subsets are needed for a large phased array feed, then complexity and resource usage grows linearly as the number of inputs is increased, instead of exponentially. The drawback of course is that such an architecture could be so complicated that it just translates computational burden into longer development time and complexity. Does anyone have a sense if this may be a bad idea or if anyone has done it? Thanks, Jonathan Landon
