Interleaved Coherence

[jevillegasd]

Would be very useful to run a T1 and a Ramsey simultaneously (and maybe an echo sequence as well) get an experiment where the measurements are correlated. The idea is just to interleave the sequence of the multiple experiments.

Background

Currently time base characterizations are run sequentially, which means there is significant difference in time between two measurements (10s of seconds). Because of this, any low frequency effects on the qubit coherence affects the measurements and generate variations that make it difficult to correlate qubit performance (e.g. we sometimes I've measured $T_{2,echo}$ that exceed $2 T_1$, which is only possible if the T1 that I am comparing to is not up to date).

Proposed Pulse Sequence

A version of this is in the example implementation in 20c9a88

seq = PulseSequence()

for shot in shots:
  for delay in delays:
    # Sequence for T1
    seq.append(reset) # This reset is not really needed since is added by the qibo exceution
    seq.append(r180)
    seq.append(delay)
    seq.append(measure)

    # Sequence for Ramsey
    seq.append(reset)
    seq.append(r90)
    seq.append(delay)
    seq.append(r90)
    seq.append(measure)

Parameters and Outputs

These are identical input and result parameters as the t1 and ramsey routines. We could import them from there instead.

[alecandido]

To avoid repeating the implementation of existing experiments, the current proposal was to change the Routine.acquisition interface, to make the sequences accessible, when possible. Cf. #1280

When this will be implemented, we will be able to get access to the T1 and Ramsey sequence.
However, to generally unroll in a single execution, we should make the sweepers local to the sequences. Which is not yet allowed to Qibolab.
But since in this case they can share the sweeper, unrolling should just work out of the box.

The general case should also become available, with a larger rewriting planned in qiboteam/qibolab#917.
There would even be a faster alternative, described in qiboteam/qibolab#1404, which would be possible with extremely limited driver's rewriting (since it does not alter the compilation itself). But this one is not useful in this case, since it would not allow interleaving experiments, but just stacking them.

[alecandido]

Btw, for an even faster implementation, I would support your experiment, including the duplication, but more as an external plugin.
The PR for that is ready #1402, and it will be merged as soon as I will be able to provide docs (and have a last pass with tests - which are passing, but I still want to better check, since I changed a bit of the Executor, which is deeply ingrained in the core mechanisms).