Machine learning enhanced resolver systems
Processing circuitry comprising an analog-to-digital converter configured to perform the techniques of this disclosure. The analog-to-digital converter may obtain, based on electrical interactions with a resolver sensor, a noisy modulated digital cosine signal (NMDCS) and a noisy modulated digital sine signal (NMDSS). The processing circuitry includes a filter configured to process the NMDCS and the NMDSS to output a filtered NMDCS and a filtered NMDSS, and a rectifier configured to process the filtered NMDCS and the filtered NMDSS to output a rectified NMDCS and a rectified NMDSS. The processing circuitry includes a digital signal processor that executes a first trained machine learning model to process the rectified NMDCS and the rectified NMDSS to obtain a demodulated DCS and a demodulated DSS, and compute, based on the demodulated DCS and the demodulated DSS, an approximate angle of the rotating element.
1 . Processing circuitry comprising:
an analog-to-digital converter electrically coupled to a resolver sensor attached to a rotating element controlled by the processing circuitry, the analog-to-digital converter configured to obtain, based on electrical interactions with the resolver sensor, a noisy modulated digital cosine signal for an indirectly sensed angle of the rotating element and a noisy modulated digital sine signal for the indirectly sensed angle of the rotating element;
one or more filters configured to process the noisy modulated digital cosine signal and the noisy modulated digital sine signal to output a filtered noisy modulated digital cosine signal and a filtered noisy modulated digital sine signal;
one or more rectifiers configured to process the filtered noisy modulated digital cosine signal and the filtered noisy modulated digital sine signal to output a rectified noisy modulated digital cosine signal and a rectified noisy modulated digital sine signal; and
a digital signal processor configured to:
execute a first trained machine learning model that processes the rectified noisy modulated digital cosine signal and the rectified noisy modulated digital sine signal to obtain a demodulated digital cosine signal and a demodulated digital sine signal; and
compute, based on the demodulated digital cosine signal and the demodulated digital sine signal, an approximate angle of the rotating element.
2 . The processing circuitry of claim 1 , wherein the one or more filters comprises a cyclic integrating comb filter that downsamples the noisy modulated digital cosine signal and the noisy modulated digital sine signal to output a downsampled digital cosine signal as the filtered noisy modulated digital cosine signal and a downsampled digital sine signal as the filtered noisy modulated digital cosine signal.
3 . The processing circuitry of claim 1 , further comprising one or more integrators,
wherein the one or more rectifiers output the rectified noisy modulated digital cosine signal and the rectified noisy modulated digital sine signal to the one or more integrators, and
wherein the one or more integrators process the rectified noisy modulated digital cosine signal and the rectified noisy modulated digital sine signal to output an integrated noisy modulated or demodulated digital cosine signal and an integrated noisy modulated or demodulated digital sine signal to the digital signal processor as the rectified noisy modulated or demodulated digital cosine signal and the rectified noisy modulated or demodulated digital sine signal.
4 . The processing circuitry of claim 1 , wherein the first trained machine learning model comprises an autoencoder neural network or a transformer.
5 . The processing circuitry of claim 1 , wherein the digital signal processor is further configured to execute a second trained machine learning model that processes the demodulated digital cosine signal and the demodulated digital sine signal to obtain an approximate velocity of the rotating element.
6 . The processing circuitry of claim 5 , wherein second trained machine learning model comprises a multi-layer perceptron.
7 . The processing circuitry of claim 1 , wherein the processing circuitry resolves the approximate angle of the rotating element in less than one millisecond.
8 . The processing circuitry of claim 1 ,
wherein the rectified noisy demodulated digital cosine signal includes first in band noise,
wherein the rectified noisy modulated digital sine signal includes second in band noise, and
wherein the first trained machine learning model removes the first in band noise from the rectified noisy modulated or demodulated digital cosine signal and the second in band noise from the rectified noisy modulated or demodulated digital sine signal to obtain the demodulated digital cosine signal and the demodulated digital sine signal.
9 . The processing circuitry of claim 1 , wherein the digital signal processor, to compute the approximate angle of the rotating element, computes an arctangent of the demodulated digital cosine signal and the demodulated digital sine signal to obtain the approximate angle of the rotating element.
10 . The processing circuitry of claim 1 , wherein the processing circuitry comprises a microcontroller.
11 . A method comprising:
obtaining, by a delta-sigma analog-to-digital converter of a microcontroller, and based on electrical interactions with a resolver sensor, a noisy modulated digital cosine signal for an indirectly sensed angle of a rotating element and a noisy modulated digital sine signal for the indirectly sensed angle of the rotating element;
processing, by one or more filters of the microcontroller, the noisy modulated digital cosine signal and the noisy modulated digital sine signal to output a filtered noisy modulated digital cosine signal and a filtered noisy modulated digital sine signal;
processing, by one or more rectifiers of the microcontroller, the filtered noisy modulated digital cosine signal and the filtered noisy modulated digital sine signal to output a rectified noisy modulated digital cosine signal and a rectified noisy modulated digital sine signal;
executing, by a digital signal processor of the microcontroller, a first trained machine learning model that processes the rectified noisy modulated digital cosine signal and the rectified noisy modulated digital sine signal to obtain a demodulated digital cosine signal and a demodulated digital sine signal; and
computing, by the digital signal processor, and based on the demodulated digital cosine signal and the demodulated digital sine signal, an approximate angle of the rotating element.
12 . The method of claim 11 ,
wherein the one or more filters comprise a cyclic integrating comb filter, and
wherein processing the noisy modulated digital cosine signal and the noisy modulated digital sine signal comprises downsampling, by the cyclic integrating comb filter, the noisy modulated digital cosine signal and the noisy modulated digital sine signal to output a downsampled digital cosine signal as the filtered noisy modulated digital cosine signal and a downsampled digital sine signal as the filtered noisy modulated digital cosine signal.
13 . The method of claim 11 , further comprising processing, by one or more integrators, the rectified noisy modulated digital cosine signal and the rectified noisy modulated digital sine signal to output an integrated noisy modulated or demodulated digital cosine signal and an integrated noisy modulated or demodulated digital sine signal to the digital signal processor as the rectified noisy modulated or demodulated digital cosine signal and the rectified noisy modulated or demodulated digital sine signal.
14 . The method of claim 11 , wherein the first trained machine learning model comprises an autoencoder neural network or a transformer.
15 . The method of claim 11 , further comprising executing a second trained machine learning model that processes the demodulated digital cosine signal and the demodulated digital sine signal to obtain an approximate velocity of the rotating element.
16 . The method of claim 15 , wherein second trained machine learning model comprises a multi-layer perceptron.
17 . The method of claim 11 , wherein resolving the approximate angle comprises resolving the approximate angle of the rotating element in less than one millisecond.
18 . The method of claim 11 ,
wherein the rectified noisy demodulated digital cosine signal includes first in band noise,
wherein the rectified noisy modulated digital sine signal includes second in band noise, and
wherein executing the first trained machine learning model comprises executing the first trained machine learning model that is configured to remove the first in band noise from the rectified noisy modulated or demodulated digital cosine signal and the second in band noise from the rectified noisy modulated or demodulated digital sine signal to obtain the demodulated digital cosine signal and the demodulated digital sine signal.
19 . The method of claim 11 , wherein computing the approximate angle of the rotating element comprises computing an arctangent of the demodulated digital cosine signal and the demodulated digital sine signal to obtain the approximate angle of the rotating element.
20 . A system comprising:
a rotating element;
a resolver sensor attached to the rotating element; and
a microcontroller communicatively coupled to the resolver sensor and configured to control the rotating element, the microcontroller comprising:
a delta-sigma analog-to-digital converter electrically coupled to the resolver sensor, the delta-sigma analog-to-digital converter configured to obtain, based on electrical interactions with the resolver sensor, a noisy modulated digital cosine signal for an indirectly sensed angle of the rotating element and a noisy modulated digital sine signal for the indirectly sensed angle of the rotating element;
one or more filters configured to process the noisy modulated digital cosine signal and the noisy modulated digital sine signal to output a filtered noisy modulated digital cosine signal and a filtered noisy modulated digital sine signal;
one or more rectifiers configured to process the filtered noisy modulated digital cosine signal and the filtered noisy modulated digital sine signal to output a rectified noisy modulated digital cosine signal and a rectified noisy modulated digital sine signal; and
a digital signal processor configured to:
execute a first trained machine learning model that processes the rectified noisy modulated digital cosine signal and the rectified noisy modulated digital sine signal to obtain a demodulated digital cosine signal and a demodulated digital sine signal; and
compute, based on the demodulated digital cosine signal and the demodulated digital sine signal, an approximate angle of the rotating element.