Dr. Frank wrote before that the 0.43 ppm/day number is based on the assumption of random walk (0.43 ppm/day random walk gives about 8 ppm after a year which is the basic instrument accuracy spec). If the drift is continuous and persistent for more than a year as in Midis case, that assumption does not hold. In other words: The observation of a long term continuous ADC drift is far away from what Agilent engineers had seen and means that an ADC board maybe in risk.
Regards, Dieter
Sorry again, Dieter, but this 0.43 ppm/day criterion is not based on 'Random Walk' phenomenom, and I have never said so.
Again and again, there was no , and there is no specified criterion for the annual drift of the resistor array inside U180.
When this serial batch problem occurred in about 2006, the agilent engineers have used the timely DCV drift specification of the 3458A, which is mostly caused by the LTZ1000A drift, and boiled that down to its theoretical 24h drift. The typical behavior of any thermally induced drift is a square-root one, see article from hp journal 4/89, with an annual drift of 8ppm/yr.
So the drift function can be written down as: dU(t)/U
ref = 8ppm/yr * SQRT(t(hours)/8742h).
For t = 24h you get these ominous 0.42ppm.
First, this crude criterion has another flaw. If you have a 4ppm/yr. reference built in, this criterion is two times too high.
Second, this usual SQRT law will vanish after some years, and instead effects like popcorn noise, or random walk effects will get dominant.
So the drift of such references will probably never come to an end.
For comparison, I show the timely drift of one of my LTZ1000 references (at 50°C), from 'First Light' to the actual status.
The measurement is done relative to the group of the other 7 references, using the 3458A only for comparing them.
So this measurement is an approximation only for the absolute timely drift, but supported by several baseline comparisons to calibrated standards (e.g. on MM2019).
The comparison is done on a monthly basis, so you see a lot of small jumps and humps on the order of 0.2ppm.
These are caused by the intermittent measurements, maybe also by other drift phenomenons.
In the end, this is my worst drifting reference.. if hp would implement the LTZ1000 as intended, i.e. at 50..60°C oven temperature only, for metrology purposes in a metrology grade environment, an annual drift rate of typically -0.8ppm/year could easily be achieved.
To demonstrate that such stability measurements can also be performed much smoother, I also show a continuous 24h experiment, comparing my 3458A versus an ultrastable Datron 7000 reference at constant room temperature (after 2h). There's no ACAL during these 24h. A combined noise of about 280nV
eff, and small random jumps of +/- 0.1ppm can be observed. Both effects are caused mostly by the LTZ1000 / LTZ1000A references inside both instruments.
In the end, these are practical and realistic use cases for the 3458A, showing the general limitations of DCV stability measurements.
Another hint, my 3458A obviously does its job nicely, as no practical relevant drift of its U180 can be detected in this case.
Looking at these diagrams, this discussion, whether the theoretical 0.03ppm/day drift for the U180 of MiDi's 3458A, is' healthy', or not, to me seems to be really academic.
Frank