Detecting rf emissions and classifying rf sources using spectrum transformer models

A method comprising:

• capturing first data regarding at least a first transmission of RF signals over a period of time, wherein the first data is captured in the time domain by a first receiver;
• transforming the first data regarding the first transmission of the RF signals to a frequency domain;
• generating at least a first two-dimensional representation based at least in part on the first data transformed to the frequency domain;
• extracting at least a first set of features from the first two-dimensional representation;
• encoding at least the first set of features with second data representing at least one range of frequencies associated with the first data;
• identifying a first plurality of sets of features, wherein each one of the first plurality of sets of features was extracted from one of a plurality of two-dimensional representations generated based at least in part on data regarding RF signals associated with one of a plurality of RF events transformed to the frequency domain;
• generating at least a first embedding based at least in part on the first set of features encoded with the second data and at least some of the first plurality of sets of features;
• and
• identifying at least one attribute of the first transmission of the RF signals based at least in part on the first embedding.

An edge computing unit comprising:

• a signal receiver;
• and
• at least one server rack comprising at least one processor unit and at least one data store,
• wherein the at least one server rack is programmed with one or more sets of instructions that, when executed, cause the edge computing unit to perform operations comprising:
• receiving a query of information from a first computer system over one or more networks;
• transforming first data representing a plurality of RF signals captured by the signal receiver to a frequency domain;
• generating a first spectrogram based at least in part on the transformed first data;
• extracting a first set of features from the first spectrogram;
• identifying second data representing a plurality of RF events, wherein each one of the plurality of RF events includes a transmission of at least one RF signal;
• extracting a first plurality of sets of features from the second data, wherein each one of the first plurality of sets of features represents one of the plurality of RF events;
• executing a spectrum encoder to generate a first plurality of embeddings, wherein each one of the first plurality of embeddings is generated based at least in part on the first set of features and one of the first plurality of sets of features;
• generating at least one embedding based at least in part on the query for information;
• identifying a second plurality of embeddings, wherein each one of the second plurality of embeddings represents at least a portion of one of a plurality of resources including semantic information regarding RF signals, sources of RF signals or RF events including transmissions of RF signals;
• executing a spectrum decoder to generate a response to the query for information based at least in part on the first plurality of embeddings, the at least one embedding generated based at least in part on the query for information, and the second plurality of embeddings;
• and
• providing a set of words generated based at least in part on the output to one of the first computer system or a second computer system.