Application of reinforcement learning for NQR excitation sequence optimization

Abstract

Nuclear quadrupole resonance is a spectroscopy technique well-known for its high specificity and its downsides, like very low signal-to-noise ratio, susceptibility to radio frequency interference and temperature dependence of the resonance lines. Multiple optimization techniques have been applied in nuclear resonance spectroscopy to improve the excitation sequence and Bayesian optimization has been shown to give good results. This paper proposes a different approach to finding the optimized excitation sequence by applying reinforcement learning algorithms. Six solutions are proposed, validated and tested in both simulated and real environments. The solutions are shown to be inappropriate for this application and to provide no information gain, similar to the random algorithm. Bayesian optimization is considered better suited for nuclear quadrupole resonance sequence optimization.