Date: Tue, 26 Aug 2008 15:11:12 +0200
Subject: Re: FEIC band 9 beam pattern measurement discussion
Re: Andrey's report FEIC-b9-beam-report-3.pdf.
General comment: nice work
. Only a few quibbles as follows.
- Fig. 2. The scanned area is just a bit too small. The truncation at maxY results in ghost images in the far-field. At the limit of significance, granted, but if you aim for an automated pipeline, you want some buit-in margin, not having to assess visually intermediate processing steps. Even without enlarging the FOV, the situation could be improved by better centering. At IRAM, we aim for -40dBr at field edges for the near-field data.
- Top of page 3: "The beam pointing angles were calculated by a center of mass formula". OK for a first verification. However, that might differ from other determinations because, a.o.: + asymmetry of the beam (in the rx's ff, i.e. at the subref); + influenced by ghost images (mentioned above) or the noise in the "empty" part of the ff map. Especially when the nf beam fills most of the nf scan area (as is the case here), this results in the ff beam occupying only a small fraction of the ff array, and noise flucutations could become significant. In IRAM software, we first isolate the significant portion of the beam.
- Probe geometry. I feel that esp. for a production setup, the impact of the transmitter pattern should be evaluated; in other words, results looking good is maybe not quite good enough a validation. An edge falloff of order -3dB from the transmitter pattern (in the corners!) can be corrected and the correction need not be accurate to the 1/100 dB.
Re: Richard's comments to my comments. "Response to Bernard's
After one iteration, we agree on most points, kind of nice. Remaining
comments (only one, actually, on the latter two, I just concur with
- [BL] Concerning the X-Y locations of the phase center... [RH] At present their fitting technique involves looking at the plots of phase and trying to decide what coordinates of the phase centre gives the flattest result.
- This involves manual tweaking. Again something you would like to avoid in production. Assuming the Z coordinate (focus) can also be automated (certainly someone will come up with a solution) I recommend: DFT from nf to ff (subref); truncation by subref (or, more correctly, by the equivalent parabola), and another DFT to the edge of the Universe (EOU). The location of the beam at the EOU can be easily found (barycenter or other). From the telescope's equivalent focal length, one can go back to the offsets in the Cass focal plane. That procedure is both automatic and non-iterative.
- [RH] It [NSI softwar] also interpolates the data onto uniform grid in angle, using a polynomial or some such, which I am less keen happy about.
- Indeed!!. What lends itself to proper transforms by DFT is sampling on a uniform linear grid in the nf plane. And the proper coordinates in the ff are the x- and y-coordinates of the unit vector in the current aiming direction NOT the angles in some spherical coordinate system.
- [RH]...using Excel...what is being done at the NA FEIC...
- Oh No! Here we use MathCad just because of ease of WYSIWYG programming. But MathCad was caught red-handed giving plain wrong answers, or crashing unexpectedly. Should move to Matlab, that has proven its reliability in other tasks at IRAM. Other choices certainly are suitable, like Mathematica used (I believe) by Andrey.
- 26 Aug 2008