Multi-Layer Neural Network for the Stabilization of Ultra-Stable Oscillators

We present results where multi-layer neural networks were used to predict temperature-induced frequency fluctuations in a low size, weight, and power ultra-stable oscillator (USO). Our results demonstrate that a multi-layer USO utilizing inputs from multiple temperature sensors can reduce the modified Allan deviation of the oscillator below 10
^–12
at all timescales between 1 and 1000 seconds. We also show that a direct digital synthesize (DDS) can be used to correct the frequency fluctuations predicted by the neural network while maintaining a low-noise and frequency stable output signal.