Signal detection using machine learning models
Methods, systems, and apparatus, including computer programs encoded on computer-storage media, for signal detection using machine learning models. In some implementations, a method includes obtaining communications data comprising one or more radio signals; providing the communications data to a first machine learning model that is trained to detect frequency bands that likely include radio signals; obtaining information representing one or more frequency bands that correspond to likely radio signals in the communications data as an output of the first machine learning processing the communications data; providing at least a portion of the communications data corresponding to the one of the one or more frequency bands to a second machine learning model that is trained to detect one or more features of radio signals; and obtaining a signal classification as an output of the second machine learning model generated by processing the portion of the communications data provided to the machine learning model.
1 . A method for processing radio signals using one or more neural networks, the method comprising:
obtaining communications data corresponding to one or more radio signals;
providing the communications data to a first machine learning model that is trained to detect one or more frequency bands that likely include radio signals;
obtaining, as an output of the first machine learning model, information representing one or more frequency bands that correspond to likely radio signals in the communications data, wherein the information representing the one or more frequency bands is generated by the first machine learning model by processing the communications data;
providing at least a portion of the communications data corresponding to the obtained one or more frequency bands to a second machine learning model that is trained to detect one or more features of radio signals; and
obtaining, as an output of the second machine learning model, a signal classification corresponding to the obtained one or more frequency bands, wherein the signal classification is generated by the second machine learning model based on processing the portion of the communications data provided to the second machine learning model.
2 . The method of claim 1 , wherein generating the information representing the one or more frequency bands by the first machine learning model by processing the communications data comprises:
processing, by the first machine learning model, the communications data using one or more regression heads.
3 . The method of claim 2 , wherein processing, by the first machine learning model, the communications data using the one or more regression heads comprises:
processing, by the first machine learning model, the communications data using one or more fully connected layers in the one or more regression heads to predict a likelihood that a particular frequency band likely include radio signals.
4 . The method of claim 2 , wherein training the one or more regression heads comprises:
comparing a predicted set of frequency bands that likely include radio signals to a known set of frequency bands that include radio signals using a set matching algorithm.
5 . The method of claim 4 , wherein training the one or more regression heads comprises:
minimizing a difference between the predicted set of frequency bands that likely include radio signals to the known set of frequency bands that include radio signals.
6 . The method of claim 1 , further comprising:
determining, by the first machine learning model using one or more classifier heads, a class of each radio signal of the radio signals.
7 . The method of claim 1 , further comprising:
providing features associated with the one or more radio signals to the second machine learning model with the portion of the communications data, wherein the features include one or more of a center frequency, bandwidth, confidence value, or signal strength.
8 . The method of claim 1 , wherein obtaining the communications data corresponding to the one or more radio signals comprises:
obtaining data representing in-phase and quadrature signals demodulated from a received radio signal of the one or more radio signals.
9 . The method of claim 1 , further comprising:
generating the portion of the communications data by performing one or more of the following signal processing on the communications data: normalization, scaling, filtering, tuning, decimation, resampling, adjustment, transformation, channelization, or feature extraction.
10 . The method of claim 1 , wherein generating the signal classification by the second machine learning model by processing at least the portion of the communications data comprises:
processing, by the second machine learning model, the portion of the communications data using one or more classifier heads to determine a signal classification of each radio signal in the one or more radio signals.
11 . One or more non-transitory computer storage media encoded with computer program instructions that when executed by one or more computers cause the one or more computers to perform operations comprising:
obtaining communications data corresponding to one or more radio signals;
providing the communications data to a first machine learning model that is trained to detect one or more frequency bands that likely include radio signals;
obtaining, as an output of the first machine learning model, information representing one or more frequency bands that correspond to likely radio signals in the communications data, wherein the information representing the one or more frequency bands is generated by the first machine learning model by processing the communications data;
providing at least a portion of the communications data corresponding to the obtained one or more frequency bands to a second machine learning model that is trained to detect one or more features of radio signals; and
obtaining, as an output of the second machine learning model, a signal classification corresponding to the obtained one or more frequency bands, wherein the signal classification is generated by the second machine learning model based on processing the portion of the communications data provided to the second machine learning model.
12 . The media of claim 11 , wherein generating the information representing the one or more frequency bands by the first machine learning model by processing the communications data comprises:
processing, by the first machine learning model, the communications data using one or more regression heads.
13 . The media of claim 12 , wherein processing, by the first machine learning model, the communications data using the one or more regression heads comprises:
processing, by the first machine learning model, the communications data using one or more fully connected layers in the one or more regression heads to predict a likelihood that a particular frequency band likely include radio signals.
14 . The media of claim 12 , wherein training the one or more regression heads comprises:
comparing a predicted set of frequency bands that likely include radio signals to a known set of frequency bands that include radio signals using a set matching algorithm.
15 . The media of claim 14 , wherein training the one or more regression heads comprises:
minimizing a difference between the predicted set of frequency bands that likely include radio signals to the known set of frequency bands that include radio signals.
16 . The media of claim 11 , wherein the operations comprise:
determining, by the first machine learning model using one or more classifier heads, a class of each radio signal of the radio signals.
17 . The media of claim 11 , wherein the operations comprise:
providing features associated with the one or more radio signals to the second machine learning model with the portion of the communications data, wherein the features include one or more of a center frequency, bandwidth, confidence value, or signal strength.
18 . The media of claim 11 , wherein obtaining the communications data corresponding to the one or more radio signals comprises:
obtaining data representing in-phase and quadrature signals demodulated from a received radio signal of the one or more radio signals.
19 . The media of claim 11 , wherein the operations comprise:
generating the portion of the communications data by performing one or more of the following signal processing on the communications data: normalization, scaling, filtering, tuning, decimation, resampling, adjustment, transformation, channelization, or feature extraction.
20 . A system, comprising:
one or more computers and one or more storage devices on which are stored instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising:
obtaining communications data corresponding to one or more radio signals;
providing the communications data to a first machine learning model that is trained to detect one or more frequency bands that likely include radio signals;
obtaining, as an output of the first machine learning model, information representing one or more frequency bands that correspond to likely radio signals in the communications data, wherein the information representing the one or more frequency bands is generated by the first machine learning model by processing the communications data;
providing at least a portion of the communications data corresponding to the obtained one or more frequency bands to a second machine learning model that is trained to detect one or more features of radio signals; and
obtaining, as an output of the second machine learning model, a signal classification corresponding to the obtained one or more frequency bands, wherein the signal classification is generated by the second machine learning model based on processing the portion of the communications data provided to the second machine learning model.