[Marc, Erik]
H1IOPISCEX and H1IOPOMC0 were rebuilt with custom RCG software, tagged "h1iopomc0_h1iopiscex_20230221" to support new LIGO timing card firmware.
Firmware version 2 was installed in both IO chasses. This firmware removes 7 usec of latency on the duotone signal, affecting the IOP's dutotone zero-crossing measurement. The process involved restarting the IO chasses and the front end servers.
H1OMC duotone zero-crossing measurement averaged 7.21 usec before the firmware upgrade and now is between 0.1 and 0.2 usec.
H1ISCEX was at 7.10 usec before the upgrade and now is between -0.1 and 0 usec. DAC duotone dropped from 68.6 usec to 61.6.
The custom software uses Daniel Sigg's non-linear correction to the fit and revised sampling windows that are centered near the expected crossing point.
Summary: All observed time shifts make sense.
Until today the reported DuoTone delay for the h1iopomc0 front-end was reported as 14.50μs. This included a 4 cycles offset at 2-19 Hz due to the ADC starting the sampling 4 samples ahead of the 1PPS within each IOP cycle. Ths brings the delay to 6.87μs. The algorithm improvement then brought this to 7.21μs, a +0.34μs shift. The new firmware corrects for the delay in the DuoTone generation to align the zero crossing with the 1PPS. The programmed shift is –7.09μs. This brings the reported shift to 0.12μs, which agrees with the measured 0.1-0.2μs.
For h1iopiscex, the old reported DuoTone delay was 6.82μs, which shifted to 7.10μs with the improved algorithm, a +0.28μs shift. The new firmware again shifts by –7.09μs, bringing the reported delay very close to 0μs.
The DuoTone algorithm is a linear regression over a set number of points (12) starting ahead of the 1PPS (5 points). The old algorithm used to start 6 points ahead. The choice of total and ahead points changes the result for the zero crossing by as much as 0.25μs. The main reason for this is that the underlying sine wave is not excatly linear within the selected window. The improved algorithm adds a linearization step before calculating the linear regression using
Equation
With y the measured ADC value, Dy the estimated slope, and f = 960.5Hz. The slope is estimated using the last and first data point of the selected analysis window. This reduces the dependes on the window choice by an order of magnitude.