Temperature Sensitivity of RFML Algorithms

A variety of hardware, including radios, are made to operate for extended periods of time in extreme environments where temperatures can vary greatly. Both ambient temperature and temperature changes caused by extended device operation have known effects on signals transmitted by a device. These signal variations could impact the performance of Radio Frequency Machine Learning (RFML) techniques, such as Specific Emitter Identification (SEI) and Automatic Modulation Classification (AMC). This work summarizes an experiment to evaluate the sensitivity of temperature on SEI and AMC by examining both ambient temperature and temperature rise created during transmissions. SEI and AMC models are trained and tested on data collected in various temperature combinations. The experimental results confirm that ambient temperature has a material impact on the performance of SEI models, while AMC models are resilient, highlighting that designers must consider the fundamental source of learned behaviors during real world deployment of RFML algorithms.