Closed hay-k closed 3 months ago
All modified and coverable lines are covered by tests :white_check_mark:
Project coverage is 66.66%. Comparing base (
c109c51
) to head (061aa18
).
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
Looks like the rapid changes in amplitude (when the amplitude sweep is the inner one) result into some vertical shadows.
I can understand that this is also solved by rearranging sweepers, but if there were other reasons to keep the second order, it should be also solved by a sufficient relaxation time. Isn't it?
should be also solved by a sufficient relaxation time. Isn't it?
Yes!
EDIT: added a plot related to that to the main description of the PR for completeness.
@hay-k rebased with the flux pulse hotfix so the issue I found is gone and we can merge it.
Currently, the ZI driver may change the order of real-time sweepers in certain situations. This was most probably historically done for the following two reasons:
Recently, we discovered that (and confirmed by laboneq developers), that point 1 does not exist. As for point 2, with @Jacfomg we observed that order of real-time sweeps may affect the quality of acquired data (for physical reasons, and not specific to instrument). In light of this it seems that the better decision is that the driver not change the order of sweeps. It is better to let qibocal decide which order works best for which experiment.
Below is output from qubit flux dependence experiment, where flux sweep is done as an amplitude sweep for a long square flux pulse. So we have nested 2 real-time sweeps (flux pulse amplitude and qubit drive frequency). The first plot corresponds to the case where frequency sweep is the inner loop, and for the second plot the frequency sweep is the outer loop.
Looks like the rapid changes in amplitude (when the amplitude sweep is the inner one) result into some vertical shadows. The two experiments above were done using
relaxation_time = 2000
. Increasing it mitigates the shadowing issue, as seen from the below plot done withrelaxation_time=20_000
:Checklist: