Frequency-Multiplexed Qubit Readout Kernel Training for Crosstalk

The patent segments the frequency-multiplexed readout signal into individual qubit readout signals by training separate kernels for each qubit. Each kernel is specialized to extract the readout signal of a specific qubit from the composite multiplexed signal, effectively dividing the complex multi-qubit readout task into manageable individual qubit extraction tasks. This segmentation enables high-fidelity readout of multiple qubits simultaneously through a shared readout chain.

The patent introduces trained kernels as intermediary processing elements between the shared readout bus and the qubit state determination. These kernels act as mediators that filter and extract the specific signal components corresponding to each qubit from the composite multiplexed signal, blocking crosstalk interference while preserving the desired qubit readout information. This intermediary processing layer enables accurate qubit state discrimination despite the presence of multiple coupled resonators on a shared bus.

The patent performs preliminary kernel training during the system initialization or calibration phase, before actual quantum computations begin. The trained kernels are stored and reused for subsequent readout operations, eliminating the need for repeated training during quantum experiments. This preliminary action approach ensures high readout fidelity is achieved while minimizing the time lost to training during productive quantum computing operations.