Flux (pulse) arc from time-traces

[alecandido]

The current experiment consists of an amplitude sweep of the flux pulse, tracked through the accumulated phase evolution.
Since the detuning can be quite large (up to ~GHz) this imposes quite a dense amplitude scan, since otherwise the phase would wildly jump, and the detuning reconstruction can simply fail.
Indeed, right now the algorithm is the following:

0. (for a fixed user-specified flux pulse duration)
1. compute the phase from the X and Y projections
2. unwrap the phase accumulated for increasing amplitudes
3. assuming a constant detuning, infer it just dividing the phase by the pulse duration

However, step 2 heavily relies on the density of the amplitude scan wrt the flux pulse duration, since missing a full rotation would make it impossible for the unwrapping algorithm to properly reconstruction the full accumulated phase.
Moreover, the pulse can not be too short, since step 3 assumes a constant detuning, and a short pulse would be strongly affected by flux distortions. Which require this experiment as a prerequisite to be estimated and compensated (turning it into a chicken-and-egg problem).

In https://arxiv.org/abs/2504.17082 (ch 4, sec 2.3), an alternative procedure is proposed, estimating the detuning from time-traces and their power spectral density (PSD). This can increase noise resilience, and lifts the requirement of a dense scan, accelerating the experiment.

Relevant text and figures from reference

My colleague and I significantly expedited the calibration by avoiding the uploading of redundant waveforms to AWG memory and by measuring fast Cryoscope traces with a low number of points in 20ns and low acquisition averages of $2^7$. Instead, we achieved high precision by fitting the power spectral density (PSD) of the measured data [Figure d], which is more resilient to measurement noise. This new flux arc measurement completes in 5min per qubit, using four different flux pulse amplitudes to achieve the required accuracy.

Each data point is obtained via fitting the power spectral density of the measured time traces in (c), resulting in frequency detunings (red dots in (b)).

[alecandido]

Btw, it would be nice to have it as an additional experiment, such that it can be compared to the existing one, and keep that available as a fallback.