Edoardo-Pedicillo
commented
1 month ago The current implementation of single shot classification does not distinguish between state e and any higher states.
This is expected. According to the theory, considering the pulses we are sending, we can just discriminate between zero and non-zero states.
When used to optimise the amplitude of the readout pulse, it then favours high powers that separate the clouds but that excite higher levels.
Could you please provide a report with the results. This happens because the input parameters are set incorrectly. https://github.com/qiboteam/qibocal/blob/f869f280a966fe8f93ab17c62268ca22c0fa940d/src/qibocal/protocols/characterization/readout_optimization/resonator_amplitude.py#L20-L30
To avoid it, it is enough to lower amplitude_stop and sweep the amplitude around the operational point found with the punchout. The purpose of this protocol is to fine-tune the amplitude, so we could change the input parameters, inserting amplitude_width and amplitude_step, and scan around the readout amplitude in the platform.
The current implementation of single shot classification does not distinguish between state e and any higher states. This results in an inaccurate evaluation of the readout fidelity. When used to optimise the amplitude of the readout pulse, it then favours high powers that separate the clouds but that excite higher levels. We need a routine that can accurately distinguish at least g, e, and f and calculate readout fidelities more accurately.
As an example, this is the classification obtained after updating the readout amplitude with the best amplitude generated by the readout optimization routine.