RFedRNN: An End-to-End Recurrent Neural Network for Radio Frequency Path Fingerprinting

Abstract

Radio frequency (RF) fingerprinting is a commonly used indoor positioning method. However, the random fluctuation of radio signals is a major challenge for RF positioning. Traditional methods only match the signal patterns at a single time or space point without regard to the spatial or temporal pattern of the signal. Inspired by the application of neural networks in the field of natural language processing, we presents an end-to-end RF fingerprint positioning model, named RFedRNN. The model consists of two Recurrent Neural Networks (RNNs). The first RNN encodes the RF sequence into a vector from which the second RNN decodes the corresponding target path. Training this neural network is to learn a mapping from a sequence of RF fingerprints to a path, called path fingerprinting. This method has low labor cost and the parameters do not increase with the expansion of data set, which is suitable for mobile devices. Simulation and the real-world dataset experiments show that the proposed method is superior to the existing methods in positioning accuracy and robustness.