Feedforward Quantum Control and Coherence Protection of Single Electron Spin in Diamond using Deep Learning

Measurement-based realtime control is an important strategy in quantum information processing, which is applied in fields from qubit readout to error corrections. However, the time cost of quantum measurement inevitably induces a latency in the control process and limits its performance. Here we introduce the deep learning approach to relax this restriction by predicting and compensating the latency-induced control error. We experimentally implement feedforward quantum control of a single-spin system of nitrogen-vacancy (NV) center in diamond to protect the coherence of the electron spin. The new approach enhances the decoherence time as well as the the spectrum resolution of Ramsey interferometry about three times comparing with conventional scheme. This enables resolving optically indistinguishable NV centers from their magnetic spectrum with a frequency difference less than 20 kHz. A theoretical model is proposed to explain the improvement, where we show that the low-frequency magnetic noise is perfectly reduced. This scheme could be applied in general measurement schemes and extended to other quantum control systems.