mark.barton@LIGO.ORG - posted 13:32, Wednesday 06 July 2011 - last comment - 10:51, Friday 22 July 2011(1020)
Saturation of OSEMs
By Mark, for Robert, Richard, Betsy, Jeff and Andres. Per the earlier report # 979, we found that the ITMy (Quad #2) BOSEMs had open light counts of opposite sign (positive) and somewhat larger magnitude (8 of 12 clipped at 32767 by the ADC) compared to earlier measurements on the staging building test stand (X1). We also noticed that the counts for the 4 OSEMs that weren't off-scale didn't seem well-correlated with the X1 values, as would be expected from a simple difference in gains. This led us to wonder if the X1 open light counts hadn't been suspiciously consistent, considering that the 50 BOSEMs that were characterised in detail in T0900496-v4 were evenly distributed in open light current over the full acceptable range of 45-80 µA, whereas the 12 values for the Quad #2 BOSEMs were tightly clustered between 30985 and 31555. We therefore asked Stuart Aston for the open light current measurements for all BOSEMs and made scatter plots against the open light counts recorded at X1 (E1000186-v15) and H2 (Robert's notes). See attached. Sure enough, in plot "X1 vs UK", there is no correlation between the two sets of measurements, even though the X1 data was formerly thought to be good. To make the cause clearer, the same data is replotted in "X1 vs UK (1:1 aspect)" with the aspect ratio forced to an equivalent value of 1:1 considering the different units, and the X1 data is compressed into a narrow band. This suggests saturation at some point in the electronics chain earlier than the ADC. Similarly, in "H2 vs X1" most of the H2 values are railed at 32767 suggesting saturation of the ADC, but the rest are in a narrow range just short of that and don't correlate with the X1 values. This suggests (i) we have saturation of an analog circuit earlier than the ADC as for X1 and (ii) the variation from channel to channel is all about the circuits that are saturating and not at all about the BOSEM that is connected. Therefore we need to set up a measurement where a flag is wound through a BOSEM in a calibrated fashion and the signal is monitored at each point in the chain to discover where the saturation is. (Richard is betting that the signal will already be too large at the output of the satellite amp and that that is where the gain will have to be reduced.) Richard has a dirty BOSEM #005 borrowed from the TipTilters that we can use. This will also be a good opportunity to measure the response of a BOSEM with a new-style flag.
Non-image files attached to this report
Comments related to this report
J. Kissel, P. Fritschel, Mark Barton, for the SUS team --------------- This story has been sent around via email to relevant suspensions people, but I want to stick it here where the complete story will be held for future reference. --------------- The Problem After walking through the schematics involved, Peter and I have traced the problem of the saturating OSEMs. (1) The output impedance of one leg of the satellite amplifier is 160k V/A (as confirmed by Vern, and is evident from the schematic, D0901284-v1, defined by RX02 in each of the four channels of the box). (2) The BOSEMs' photo current can be as large as 80e-6 amps, (See distribution provided by Stuart, now posted to T0900496-v4) (3) The gain of the Anti-Aliasing chassis, D070081-v5 is +1. (4) The ADC, a General Standards 16AI64SSA card, with each channel have a range of +/- 10 Vpp, or | Vmax | = 10 V (see Data Sheet), receives one leg of the differential signal coming in from the satellite amplifier. See attachement 1 for a schematic of the M0/R0 QUAD electronics, which is a portion of version two of T1100378 I'm working on. So, from this, it is clear why we have saturations: 80e-6 [A] * 160e3 [V/A] = 12.8 V input > 10 V allowed. The Proposed Solution Assuming we're to make modifications to the entire lot of aLIGO satellite amps for the 1.7 kHz oscillation modifications anyways (replacing the inductors LX01 with 100 ohm resistors, removing the 100pF cap CX07 and replace both AD797s IC552 with OP27s), would be to change the output impedance to 120k V/A, such that 80e-6 [A] * 120e3 [V/A] = 9.6 V input < 10 V allowed. (The other choice would be to only use OSEMs with 60e-6 amps, but Stuart's distribution (T0900496) shows that the majority of the OSEMs show open light current above 60e-6 amps.) Justification of Proposed Solution and An Explanation of the Safety Margin (i.e. why 120k V/A is OK): The counts as measured by the ADC are the differential voltage between the two legs of the differential driver. Hence, BOSEM #638 (now installed as H2SUSITMY's M0 F2 BOSEM) would yield 57.39e-6 A * 160e3 V/A = 9.182 V on *one* leg of the differential output of the sat amp, which is what's fed into *one* pin of the differential ADC. The other leg of the differential driver / ADC pin would receive -9.182 V, making the differential voltage measured 2 * 9.182 = 18.364 V (Here in lies the (my) prior confusion between saying the "gain" of the sat amp is 160e3 V/A vs. 320e3 V/A -- because the input to the sat amp is single ended, and is split via a differential driver and we measure the signal differentially, the "gain" of the sat amp could be treated as though Differential Voltage OUT / Single Ended Current IN is 160e3*2 = 320e3 Ohms, *but* each leg of the driver (which is what a given pin on the ADC would see) does not see the factor of two.) On the 16-bit ADCs, which have a 40 Vpp differential swing over 2^16 cts, this would mean that the count value seen at the input of the digital filters is 18.364 V * 2^16 cts / 40 V = 30088 cts. Now, we cannot in anyway assume that the photo current from this guy has stayed exactly 57.3900000e-6 A. From the above eLOG entry, H2 SUS ITMY's M0 F2 has a open light value of 30840 cts, implying an open light current of 30840 cts * (40 V / 2^16 ct) * (1/2 * 1/160e3 A/V) = 58.82e-6 A, a (58.82 - 57.39) / 57.39 = 2% change. Using the data from T0900496, If we include a safety margin of 2%, then only 1 (one) of the 700 BOSEMs would exceed the 32000 ct saturation limit, if we converted to a transimpedance of 120 V/A: threshold = 32000 cts * (40 V / 2^16 ct) * (1/2 V/V) * (1/120e3 A/V) = 81.38e-6 A margin = 0.02 safeTreshold = threshold - margin * threshold = 79.75e-6 A find(T0900496 > safeThresh) = 1 (at 79.96e-6 A) Even if we have a safety margin of 5%, that would only exclude 37 of the 700. Further -- bare in mind that this is merely to measure the open light current such that we might operate in the mid range of the LED. It is by no means a show stopper if a few of the OSEMs exceed the 32k limit by 5%. Not a show stopper, once we replace the gain resistor Further Details explaining the difference between test stands and the production electronics (from Peter): Looking at the circuits a little more carefully, the problem is that the opamps in the satellite amp (OP2177) do not have adequate current drive capability when connected to the v1 (non-buffered) AA filters. The OP2177 is spec'd at a current drive of +/-10 ma, and with the v1 AA, it's driving close to 500 ohms, so it's well in its non-linear region when being asked to put out ~10 V. On the other hand, the small-signal gain in the two cases is actually the same. So we really shouldn't be using the v1-AA units with the sat amps. The whole story points out why we went to buffered inputs and outputs with v2+ on the AAs ... The ADC Downsampling filter Note that the only piece of DC Gain in the puzzle we have yet to included in the above analysis is the down-sampling filter from the native ADC sampling rate 65k Hz to the model's rate of 16k Hz. This *used* to have some small non-unity gain at DC (at the 5% level), but I have confirmed on the MIT system here that in RCG 2.3 (which is the same as at the sites, also confirmed) that this bug has been fixed, and the DC Gain is now unity to within 0.6% see second attachment for transfer function between a 65k test point in the IOP model (M1:IOP-HAMX_MADC3_TP_CH0) and the corresponding channel in the suspension model (M1:SUS-MCTS_M1_OSEMINF_T3_IN1)
Non-image files attached to this comment