[time-nuts] Replacement A9 boards for the HP 5065A
Charles Steinmetz
csteinmetz at yandex.com
Sun Mar 4 14:17:24 EST 2018
Poul-Henning wrote:
> The input signal to the integrator is continuous and jump-free and
> the relevant time constant is sub-second. Dielectric absorption
> doesn't matter when there are no voltage jumps.
That is a very common misconception, probably fostered by the usual test
methodology for DA. But it is not correct.
DA is a charge-migration phenomenon, so it occurs *whenever the charge
on a capacitor changes.* A step function in the charge very obviously
represents such a change (and for this reason is used for testing the
parameter in most circumstances), but so do gradual changes in the
driving charge. These can be less obvious (than step changes),
depending on the time scale of the change relative to the time scale of
the DA process in the particular capacitor.
There are several dimensions to DA -- not just how much charge isn't
returned immediately, but what the "tail" looks like (i.e., the time
profile of charge return, generally measured as current vs. time).
Different dielectrics (and, to a certain extent, different construction
techniques) exhibit different charge-return profiles.
Long-TC integrators do tend to mask the effects of DA, because the time
scale of changes in the driving charge is closer to the time scale of
the delayed charge return caused by DA. But that does not mean that DA
isn't important, because slow integrators are frequently used where
extremely high precision is required (such as integrators for
high-resolution DACs and EFC servos for precision oscillators), so the
tolerable amount of delayed charge return is extremely low.
As with any circuit, the A9 integrator is "good enough" when the errors
it causes are swamped by other accumulated errors. But I'm no fan of
"just good enough" design, particularly in a case like the 6065A where
time nuts are diligently working to improve the performance of the
instrument. Some times, "just because it's better" is a sufficient
reason to overdesign, particularly where the incremental cost is low and
especially where the projected number of units is low, both of which are
true WRT the improved A9 board.
Best regards,
Charles
More information about the time-nuts
mailing list