Signal intelligence system to integrate spectrum prediction with emitter classification and PNT
Embodiments of the disclosure provide a signal intelligence system for integrating radio frequency (RF) spectrum prediction with emitter classification and position, navigation, and timing (PNT). Methods of the disclosure include detecting a signal from an emitter operating within a wireless network. Continued processing includes forecasting, via a multi-task learning component of a machine learning module, one of a signal modulation, an emission protocol, and an identification of the emitter based on the detected signal. Additional processing includes calculating a positioning, navigation, and timing (PNT) profile for the emitter using a recurrent graph neural network (ReGNN) component of the machine learning module. Further processing includes predicting, via a recurrent neural network (RNN) of the machine learning module, a spectrum occupancy of the emitter within the wireless network.
1 . A method comprising:
detecting a signal from an emitter operating within a wireless network;
forecasting, via a multi-task learning component of a machine learning module, at least one of a signal modulation, an emission protocol, and an identification of the emitter based on the detected signal;
calculating a positioning, navigation, and timing (PNT) profile for the emitter using a recurrent graph neural network (ReGNN) component of the machine learning module, wherein the ReGNN component determines a time-of-flight (ToF) of the signal and a position of the emitter using parameters including at least two of:
a time difference of arrival (TDoA) of the signal,
a direction-of-arrival (DoA) of the signal,
a differential amplitude of the signal,
an angle-of-arrival (AoA) of the signal, and
a phase of the signal; and
predicting, via a recurrent neural network (RNN) of the machine learning module, a spectrum occupancy of the emitter within the wireless network.
2 . The method according to claim 1 , wherein the signal includes raw in-phase (I) and quadrature (Q) signals and spectrogram signals.
3 . The method according to claim 2 , further comprising generating the spectrogram signals by:
applying a short-time Fourier transform (STFT) to the I and Q signals over multiple time slots to provide a two-dimensional (2D) data set for each time slot; and
aggregating the 2D data sets for the multiple time slots to generate three-dimensional (3D) spectrogram signals.
4 . The method according to claim 1 , wherein the parameters are measured at a plurality of different locations.
5 . The method according to claim 1 , wherein the RNN predicts the spectrum occupancy of the emitter at least partially based on a regression analysis of the detected signal and the PNT profile for the emitter.
6 . The method according to claim 5 , wherein the radio environment map includes a predictive spectrum occupancy of the emitter in future time slots.
7 . The method according to claim 1 , further comprising generating a radio environment map in real-time based on the PNT profile for the emitter.
8 . A signal intelligence (SIGINT) system comprising:
a sensor configured to detect a signal from an emitter operating within a wireless network; and
a computing device coupled to the sensor and including a machine learning module, wherein the computing device is configured to:
forecast, via a multi-task learning component of the machine learning module, one of a signal modulation, an emission protocol, and an identification of the emitter based on the detected signal,
calculate at least one of a positioning, navigation, and timing (PNT) profile for the emitter using a recurrent graph neural network (ReGNN) component of the machine learning module, wherein the ReGNN determines a time-of-flight (ToF) of the signal and a position of the emitter using parameters including at least two of:
a time difference of arrival (TDoA) of the signal,
a direction-of-arrival (DoA) of the signal,
a differential amplitude of the signal,
an angle-of-arrival (AoA) of the signal, and
a phase of the signal; and
predict, via a recurrent neural network (RNN) of the machine learning module, a spectrum occupancy of the emitter within the wireless network.
9 . The system according to claim 8 , wherein the signal includes raw in-phase (I) and quadrature (Q) signals and spectrogram signals.
10 . The system according to claim 9 , wherein the machine learning module is further configured to generate the spectrogram signals by:
applying a short-time Fourier transform (STFT) to the I and Q signals over multiple time slots to provide a two-dimensional (2D) data set for each time slot; and
aggregating the 2D data sets for the multiple time slots to generate three-dimensional (3D) spectrogram signals.
11 . The system according to claim 9 , wherein the parameters are measured at a plurality of different locations.
12 . The system according to claim 8 , wherein the RNN is further configured to predict the spectrum occupancy of the emitter at least partially based on a regression analysis of the detected signal and the PNT profile for the emitter.
13 . The system according to claim 8 , wherein the machine learning module is further configured to generate a radio environment map in real-time based on the PNT profile for the emitter.
14 . The system according to claim 13 , wherein the radio environment map includes a predictive spectrum occupancy of the emitter in future time slots.
15 . A system including a wireless network and a plurality of emitters sharing a spectrum of the network, comprising:
a sensor configured to detect a signal from an emitter operating within the wireless network; and
a computing device coupled to the sensor and including a machine learning module, wherein the computing device is configured to:
forecast, via a multi-task learning component of the machine learning module, one of a signal modulation, an emission protocol, and an identification of the emitter based on the detected signal,
calculate at least one of a positioning, navigation, and timing (PNT) profile for the emitter using a recurrent graph neural network (ReGNN) component of the machine learning module, wherein the ReGNN component determines a time-of-flight (ToF) of the signal and a position of the emitter using parameters including at least two of:
a time difference of arrival (TDoA) of the signal,
a direction-of-arrival (DoA) of the signal,
a differential amplitude of the signal,
an angle-of-arrival (AoA) of the signal, and
a phase of the signal; and
predict, via a recurrent neural network (RNN) of the machine learning module, a spectrum occupancy of the emitter within the wireless network.
16 . The system according to claim 15 , wherein the signal includes raw in-phase (I) and quadrature (Q) signals and spectrogram signals, and wherein the machine learning module is further configured to generate the spectrogram signals by:
applying a short-time Fourier transform (STFT) to the I and Q signals over multiple time slots to provide a two-dimensional (2D) data set for each time slot; and
aggregating the 2D data sets for the multiple time slots to generate three-dimensional (3D) spectrogram signals.
17 . The system according to claim 15 , wherein:
the RNN is further configured to predict the spectrum occupancy of the emitter at least partially based on a regression analysis of the detected signal and the PNT profile for the emitter; and
the machine learning module is further configured to generate a radio environment map in real-time based on the PNT profile for the emitter, wherein the radio environment map includes a predictive spectrum occupancy of the emitter in future time slots.