Deep learning-based holdover compensation, and related systems, methods and devices

An apparatus, comprising:

• a processor;
• a memory having executable instructions thereon which, when executed by the processor, enable the processor to:
• generate a relationship prediction responsive to a deep learning model selected from a plurality of trained and validated deep learning models, the selected deep learning model to predict a relationship between a fundamental quantity of a first oscillating signal and a fundamental quantity of a second oscillating signal wherein the deep learning model at least partially based on one or more hyperparameters;
• and
• apply, during a period of time defined between a first synchronizing event and a second synchronizing event, an adjustment to the first oscillating signal responsive to the relationship prediction to produce an adjusted oscillating signal, wherein the adjusted oscillating signal replaces the second oscillating signal;
• and
• one or more temperature sources to provide a temperature input to the selected deep learning model, the selected deep learning model to predict the relationship between the fundamental quantity of the first oscillating signal and the fundamental quantity of the second oscillating signal responsive to a sensed temperature associated with the oscillator providing the first oscillating signal.

A method comprising:

• receiving a second clock signal;
• modeling a relationship between a first clock signal and the received second clock signal using a deep learning model selected from among a plurality of trained and validated deep learning models as having a greatest confidence level, wherein the deep learning model comprising an artificial neural network having multiple layers, wherein the deep learning model at least partially based on one or more hyperparameters;
• entering a holdover period to predict an offset with respect to the received second clock signal;
• providing an adjusted clock signal responsive to the modeled relationship and the first clock signal during the holdover period;
• and
• exiting the holdover period.

A method of adjusting a first oscillating signal, comprising:

• generating a function that combines data from a temperature source and time data from an oscillator for input to a deep learning model comprising an artificial neural network having multiple layers;
• training the deep learning model, based at least in part on values of the function and one or more hyperparameters, to predict a relationship between a fundamental property of a first oscillating signal provided by the oscillator and a fundamental property of a second oscillating signal, wherein a hyperparameter is tunable;
• and
• applying, during a period of time defined between a first synchronizing event and a second synchronizing event, an adjustment to the first oscillating signal responsive to the relationship prediction to produce an adjusted oscillating signal, wherein the adjusted oscillating signal replaces the second oscillating signal,
• wherein the fundamental property of the first oscillating signal comprises one or more of phase, wavelength, or amplitude, and
• wherein the fundamental property of the second oscillating signal comprises one or more of phase, wavelength, or amplitude.

An apparatus, comprising:

• a processor;
• and
• a memory having executable instructions thereon which, when executed by the processor, enable the processor to:
• generate a function that combines data from a temperature source and time data from an oscillator for input to a deep learning comprising an artificial neural network having multiple layers;
• generate a relationship prediction responsive to a deep learning model selected from a plurality of trained and validated deep learning models, the selected deep learning model to predict a relationship between a fundamental property of a first oscillating signal and a fundamental property of a second oscillating signal, wherein the deep learning model at least partially based on one or more hyperparameters;
• and
• apply, during a period of time defined between a first synchronizing event and a second synchronizing event, an adjustment to the first oscillating signal responsive to the relationship prediction to produce an adjusted oscillating signal, wherein the adjusted oscillating signal replaces the second oscillating signal,
• wherein the fundamental property of the first oscillating signal comprises one or more of a phase, a wavelength, or an amplitude, and
• wherein the fundamental property of the second oscillating signal comprises one or more of a phase, a wavelength, or an amplitude.