---------------------------------------------------------------------- UPDATE: 23 May 02 (RMG) ----------------- * misc notes: - for some reason IOTA IDL display inverts order of pixels w.r.t order in data files, in my plots I keep same order as data files, which is physical order? - when recording PICNIC data w.o. the rest of the VxWorks stuff running, headers have first two categories screwed up (target info & array geometry), so remember to edit by hand if expect to use standard data readers * PICNIC Background ----------------- - IMPORTANT: Error in above comparison with NICMOS background: at K' it is quoted in RMG thesis as 586 e/ms not per second! - Therefore I re-do the correct comparison. This is based on scan data recorded: NICMOS: Nov97, 4R4L, tint=1.03ms, looking out at sky through filters PICNIC: May02, 4R4L, tint=3.125ms, cold plug or looking out at shutters Numbers in table below are average of point-by-point average of N scans, computed for each pixel (2 for NICMOS, 6 for PICNIC) and then averaged over the N pixels (2 or 6). See accompanying plots. cold-plug J H K' ----------------------------------------------------- NICMOS cam ... 2.5 5.2 189.6 (du/ms) PICNIC cam 0.9 ... 1.2 44.7 (du/ms) (in the PICNIC plots, ignore pixel 4, that day it decided to show the weird t-domain artifacts) Conclusions ----------- - The previous claim of unusually high PICNIC cam background disappears - Using H and K' for a direct comparison, PICNIC cam seems to have about 4x less background than NICMOS. Could this factor be a combination of (a) bigger solid angle of PICNIC pixels looking out (due to lack of baffle, bad) and (b) smaller magnification of IONIC focusing optics (recall that in NICMOS background became noticable after adding the 3x magnification inmersion lens) ? - Using the preliminary PICNIC gain measurement above at 4R4L: 3.7 e/du, in that mode the H-backround is 1.2du/ms*3.125ms*3.7e/du = 13.9 electron. The shot noise associated with that is sqrt(13.9) = 3.7 elec. The read-noise in that mode according to all the above estimates is about 2du = 7.4 elec. Therefore in this slow mode appropriate for faint sources, we are read-noise limited -- not background limited -- by a factor of 7.4/3.7 = 2 * PICNIC Noise Power Spectrum --------------------------- - After I left, EP stayed at IOTA site (May14-May19) and improved further on the grounding situation, with good results He should explain in writing what he did and why, but the conclusion is that the winning combinations is: PICNIC power supply connected to isolation transformer (maybe not needed) through cheater plug (no Earth pin) Connect optical table (and therefore PICNIC electronics, through contact with dewar) to Earth ground using the copper strap as usual, but added a choke (ferrite bead with wire windings) to filter out high freq noise - I illustrate the current improvement and current situation with the following set of plots for each condition: 1. typical set of scans and power spectra (ps) 2. scan_rms, maximum power spectrum noise and corresponding freq, for each scan 3. scan cascade plots 4. average power spectra (all data is NLoops=1, NReads=1) - Results: A. PICNIC/VxWorks before improvements (data of UT 2002May10): scan_rms is 7 - 9du Has mean max ps noise of 3.5 - 4 with worst cases of 8 Features are clearly persistent in frequency Main ps feature drifts linearly in freq inside each set of scans! Average ps have peaks of about 2.0 and 1.0 B. PICNIC/VxWorks after ground improvements (data of 2002May20): scan_rms is 4 - 7du, factor 1.7 - 1.3 improvement average max ps noise is down to 0.5 - 1.6, factor 2 - 8 improvement worst cases are down to 3, factor of 2.7 improvement still detectable persistent feature, which still drifts linearly inside each set of scans, so must be a weaker version of what we had before peaks in average ps are about 0.2, factor 5 & 10 improvement => much improved, but not perfect (now compare with NICMOS and with Pentium control) C. PICNIC/Pentium after partial ground improvements (data of 2002May13) this is with optical table not connected to Earth, before EP scheme described above (data after EP scheme not available) scan_rms is 4du, same as best VxW case average max ps noise is 0.5, as good as best VxW behaviour worst cases are 1.0, 3 times better than VxW hint of a frequency-persistent feature in ps cascade plot peak in average ps is about 0.14, 1.4 times better than VxW => looks better by all measures, and might even be better (or not?) after EP improvements But, is that just a result of differences in readout code between Pentium and VxWorks ? => RMG and EP will compare notes and RMG will make Pentium code identical to CPLD readout to be sure comparison is fair Also should re-take this data in same conditions as (B) i.e. improved EP grounding D. NICMOS/Pentium archival data from FLUOR (2000Feb09) scan_rms is 6 - 7du, similar but worse than PICNIC/VxW average max ps noise is 1.3 - 1.6, similar but worse than PICNIC/VxW worst cases ps noise are 3.0, same as PICNIC/VxW nothing very persistent in ps, at least at the high freqs where this was the case for VxW => except for persistence aspect, new PICNIC system is slightly better by all most measures than NICMOS system Conclusion on PICNIC noise -------------------------- - PICNIC/VxWorks system is good - The fact that most of the improvement has come from grounding manipulations leaves open the possibility that we might be able to make it even better Open Questions, for someday --------------------------- * Make a comparison with Pentium readout under identical readout condition * Look for cause of the t-domain artifacts that sometimes appear in pixel 4 for some readout modes (likely a CPLD artifact) * Can we clock faster? * Why do our pixels need so much time to settle after addressing? * Can we reduce the settling time after RESET? * Why does our noise increase at higher frequencies?