Summary: Repeating the Pcal timing signals measurements made at LHO (aLOG 28942) and LLO (aLOG 27207) with more test point channels in the 65k IOP model, we now have a more complete picture of the Pcal timing signals and where there are time delays. Bottom line: 61 usec delay from user model (16 kHz) to IOP model (65 kHz); no delay from IOP model to user model; 7.5 usec zero-order-hold delay in the DAC; and 61 usec delay in the DAC or the ADC or a combination of the two. Unfortunately, we cannot determine from these measurements on which of the ADC or DAC has the delay. Details: I turned off the nominal high frequency Pcal x-arm excitation and the CW injections for the duration of this measurement. I injected a 960 Hz sine wave, 5000 counts amplitude in H1:CAL-PCALX_SWEPT_SINE_EXC. Then I made transfer function measurements from H1:IOP-ISC_EX_ADC_DT_OUT to H1:CAL-PCALX_DAC_FILT_DTONE_IN1, H1:IOP-ISC_EX_MADC0_TP_CH30 to H1:CAL-PCALX_DAC_NONFILT_DTONE_IN1, and H1:CAL-PCALX_SWEPT_SINE_OUT to H1:CAL-PCALX_TX_PD_VOLTS_IN1, as well as points in between (see attached diagram, and plots) The measurements match the expectation, except there is one confusing point: the transfer function H1:IOP-ISC_EX_MADC0_TP_CH30 to H1:CAL-PCALX_DAC_NONFILT_DTONE_IN1 does not see the 7.5 usec zero-order-hold DAC delay. Why? There is a 61 usec delay from just after the digital AI and just before the digital AA (after accounting for the known phase loss by the DAC zero-order-hold, and the analog AI and AA filters). From these measurements, we cannot determine if the delay is in the ADC or DAC or a combination of both. For now, we have timing documentation such as LIGO-G-1501195 to suggest that there are 3 IOP clock cycles delay in the DAC and 1 IOP clock cycle delay at the ADC. It is important to note that there is no delay in the channels measured in the user model acquired by the ADC. In addition, the measurements show that there is a 61 usec delay when going from the user model to the IOP model. All this being said, I'm still a little confused from various other timing measurements. See, for example, LLO aLOG 22227 and LHO aLOG 22117. I'll need a little time to digest this and try to reconcile the different results.
By looking at the phase of the DuoTone signals we can constrain whether there is any delay in ADC side (like Keita's analysis here). The DuoTone signals are desgined such that the two sinusoidal signals 960 Hz and 961 Hz will be maximum at the start of a GPS second (and also in phase with each other). To be presice, the maximum will be 6.7 µs delayed from the integer GPS boundary (T1500513). The phase of 960 Hz signal at IOP (L1:IOP-ISC_EX_ADC_DT_OUT) is -92.52 degrees with respect to GPS integer boundary (LLO a-log 27207). Since the DuoTone signal is supposed to be maximum at GPS integer boundary i.e, it is a cosine function, this corresponds to -2.52 degrees (estimate of 92.52 assumes it is a sine function) phase change. Converting this phase change to time delay we get 7.3 µs. Since there is an inherent 6.7µs delay by the time the DuoTone signals reaches the ADC, we are left with only 0.6 µs delay possibly from ADC process (or some small systematic we haven't accounted for yet). This is what Keita's measurements were showing. Combing this measurment and above transfer function measurments we can say that we understand the ADC chain and there are no time delays more than 0.6 µ in that chain. This also suggest that the 61 µs delay we see in ADC-DAC combination exist completely in DAC side.
The DuoTone signals are sine waves, so a minor correction to Shivaraj's comment above, the zero-crossing corresponds to the supposed GPS integer second. I looked at a time series and observe that the zero-crossing occurs at ~7.2 usec. Since the analog DuoTone signal lags behind the GPS second by ~6.7 usec, I can confirm that the ADC side has essentially no delay. Thus, the 61 usec seen through the DAC-ADC loop is entirely on the DAC side. Attached is a time series zoom showing the zero crossing of the DuoTone signal.
When using dtt to make a transfer function measurement between an IOP model and a user model, one has to keep in mind that dtt does another decimation silently. This is due to dtt trying to match the number of data points between two models. Fortunately, this does not seem to affect the phase, see my note at https://dcc.ligo.org/T1600454.
Updated the timing diagram for consistency with other timing measurements (LHO aLOG 30965). See attached PDF to this comment.