email_160607

BMX Status update

We haven't had telecon for a while, but there were two meetings, first on the tails of CCS16 between Anze, Paul S and Jeff and the second one a week later between myself, POC, Sergio and Joe. I also spend a two days at Perimeter, where I talked with Kendrick Smith about CHIME. This is the compressed outcome of these interactions:

• Neelima and I won the SB/BNL SEED grant, basically funding for one year of SB grad student. In addition to Chris Sheehy winning the Goldhaber fellowship, this means we're getting really quite impressive lab support for the shit. Let's not squander it! Neelima has yet to identify a suitable drudge, but i do expect her or him to work with Chris to make the thing work.

• Site selection: I have looked again at the two sets of RFI data that we have. The basin is indeed the quitest of all in the quiet part of the spectrum. However, it is louder that SW at around 1.2 and 1.4 GHz. But I think we really aim to use the quietest 100 MHz existing, so I vote for basin. F&O people are hassling me, so I will just unilatery decide on the basin, unless someone objects strongly.

• Dish design: PS keeps insisting that he can indeed get the full dish ready by end of July. PS, hit me! We need ladder and some sort of assembly for feed horn. For storage of equipment we can use dry boxes sold for sailing industry

• OMT/Feedhorn: Jeff will work on having something in hand by end of July and a cardboard model way before then.

• If we are going to do differencing, we need to have some sort of cryo system. Doesn't need to be very big as we can keep just the amplifier, hybrid and cold load in there, some 0.5 ft^3. Justine identified a couple of commercial that could do this for not very much money:

http://www.ucryo.com/

http://www.infraredlaboratories.com/Home.html

• For the demo, the back-end is the current show-stopper, see below

• For real experiment, if we attempt to do z>2, we need 150 m dishes, becomes scary, especially when you also want to put in large secondaries. Seems you are forced back to inteferometers. However, ideally you want to do both very large scales (i.e. SD part for non-Gaussianity) and very small scales at the same time. Note the SKA can't do this (i.e. they operate as a collection of single dishes OR as an interferometer, but not both really) and if we crack this problem, it would be interesting. The reason, why SKA can't do it, in my understanding is that you need to choose between hybridising for SD or straight correlations for interferometer. You can just correlate hybridised signal, but you would really need to correlate everything with everything, i.e. 4x more baselines in order to get it back. Tone injection might help with this. Moreover, tone injection could help with tracking both gain and phase changes, hopefully allowing FFT telescope approaches to work. From talking to Kendrick I've learned (obvious in retroscpect) that doing the FFT telescope vs cross-correlating all baselines is equivalent to just stacking all redundand baselines before correlating, which normally doesn't work because you haven't yet calibrated phases. There is a lot of interesting basic radio astronomy research that could be done here.

• Kendrick also told me that the biggest screw up with CHIME is that they massively underestimated the amount of software engineering time to get all the FPGAs, and GPU correlators working. They have a system in which the signal is channelised and routed around in FGPAs, but correlated in GPUs. In principle, one could also do channelisation in GPUs, but they simply don't have a way of routing those signals around in time. One could do it, but routers that you need cost millions of dollars today (but will probably go down in time).

• Another useful thing he told me that their dish surface is really crap. This is consistent with what Laura told me last week, Peterson ordered 20 dishes from china on the cheap and their surface are utter shite. So, I think our way of debugging building the dishes here and then just oursourcing the production to china is probably the correct one. He also told me that he would actually go with dishes. So our pair of dishes can act as a prototype of 2SD + 1 baseline.

• speaking to Sergio, Joe and POC it became clear that we cannot realistically implement polyphase detection in time for end of July where other pieces will be coming together. We can do simpler things, though:

• Option A is to order XFFTS which should just work off-the shelf. Somewhat problematic, cannot do more than one channel (so forget about hybridisation), delivery is in September. I also think it is just somewhat lame buying other people's equipment. But it would be useful to have something for debugging.

• Option B is to implement something minimal in FPGA: collect stream, downconvert to 400 MHz band of interest, pipe it to PC, to evertyhing else in GPUs, where it is easy to do. There are some issues about bandpass filtering without FFTS, need to do some numerical experiments. For a joke I tried to code up polyphase channeliser in straight C and on my desktop CPU it does seem like you could do it even in CPU for a single channel at the full 2.5GSampling rate using something like 3 cores (one doing network buffers, one doing the multiplications, third doing FFTs and spitting out). This can be implemented on a much shorter time-scales. But Joe first needs to work out how to actually correctly collect the stream.

• Option B2, which I like a lot, is to buy a ~$1.5k demo board that samples AD signal and pushes it out straight into your PC. Using current bandwidth of USB3 or 10Gps network, you get some 50% dead time (maybe not even that if you can convince it to spit just most significant bits), but this doesn't matter as a first order demonstrator. I like this option a lot.

• We also want some lab set up to be able to test this. Sergio is going to look into getting some DDS (direct digital synthesis) boards to get produce some tones and measure them. We also want to have two temperature controlled 50Ohm resistors to measure known temperature differences in the lab to characterise gain fluctuations etc. (again for a benchtop test, one could be 300K and the other 320K). One should also replicate the setup weinreb did to measure gain fluctuations.

Finally, it was decided that rather than have weekly telecons, we will have a once a month longer tele-meeting, giving people to do some real focused work, rather than little chunks every week. I'll send the doodle for that next week.