Methods and apparatus to discriminate authentic wireless internet-of-things devices

Abstract

Methods and apparatus are described for automatically discriminating authentic wireless Internet-of-Things (IoT) devices using a trained machine-learning module. In a training phase, the machine-learning module is trained to identify authentic IoT devices based on data in frame headers of wireless data emitted by the IoT devices. The trained machine-learning module may identify authentic IoT devices without analysing data from the payload of the frames to which the frames headers belong, and thus the privacy of data in the payload of the frame is not compromised and encryption of the payload data does not adversely affect performance of the trained machine-learning module in a subsequent production phase. Each training data sample may consist of header data from a sequence of successive frames of wireless data from authentic wireless IoT devices and, to enhance accuracy, may exclude address data.