Location Estimation of RF Emitting Source Using Supervised Machine Learning Technique

Supervised machine learning algorithms are dealing with a known set of input data and a pre-calculated response of that set (output or target). In the present work, supervised machine learning is applied to estimate the x-y location of an RF emitter. Matlab Statistical and Machine Learning Tool Box 2019b is used to build the training algorithms and create the predictive models. The true emitter position is calculated according to the data gathered by two sensing receivers. Those data are the training data fed to the learner to generate the predictive model. A linearly x-y moving emitter-sensors platform is considered for generality and simplicity. Regression algorithms in the toolbox regression learner are tried for the nearest prediction and better accuracy. It is found that the three regression algorithms, Fine tree regression, Linear SVM regression, and Gaussian process regression (Matern 5/2) achieve better results than other algorithms in the learner library. The resulted location error of the three algorithms in training phase are about 1%, 2.5%, and 0.07% respectively, and the coefficient of determination is about 1.0 for the three algorithms. Testing new data, errors reach about, 4%, 5.5%, and 1%, and the coefficient of determination is about 0.9. The technique is tested for near and far platforms. It is proved that emitter location problem is solved with good accuracy using supervised machine learning technique.