Multi-task learning neural network framework for RF spectrum sensing and classification

A signal sensing and classification method, comprising:

• receiving a spectrum of radio frequency (RF) signals;
• extracting an RF signal from the spectrum of RF signals;
• and
• performing a plurality of tasks in parallel and simultaneously on the extracted RF signal using a multi-task learning (MTL) neural network model;
• the plurality of tasks including at least three of:
• determining a signal class of the extracted RF signal;
• determining a modulation class of the extracted RF signal;
• determining at least one signal descriptor of the extracted RF signal;
• performing a classification task;
• performing a regression task, wherein the regression task comprises at least one of estimating a bandwidth of the extracted RF signal and estimating a pulse-width of the extracted RF signal.

A signal sensing and classification method, comprising:

• receiving a spectrum of radio frequency (RF) signals;
• extracting an RF signal from the spectrum of RF signals;
• and
• performing a plurality of tasks in parallel and simultaneously on the extracted RF signal using a multi-task learning (MTL) neural network model, wherein the MTL neural network model comprises a hard parameter sharing deep convolutional neural network model;
• the plurality of tasks including at least three of:
• determining a signal class of the extracted RF signal;
• determining a modulation class of the extracted RF signal;
• determining at least one signal descriptor of the extracted RF signal;
• performing RF fingerprinting on the extracted RF signal to identify an RF device that produced the extracted RF signal;
• performing a classification task;
• and
• performing a regression task, wherein the regression task comprises at least one of estimating a bandwidth of the extracted RF signal and estimating a pulse-width of the extracted RF signal.