Physical Layer Authentication Techniques based on Machine Learning with Data Compression

Wireless communications employing multi-carrier transmissions, like orthogonal frequency division multiplexing (OFDM) or single-carrier frequency division multiple access (SCFDMA) may involve the use of a large number of subcarriers. In Internet of Things (IoT) contexts, however, the use of such technologies implies the fast management of large amounts of samples on devices with limited memory and computational resources. The adoption of physical layer authentication protocols in IoT may suffer from this fact, especially when they exploit machine learning algorithms yielding a significant computational burden. For instance, the complexity of Nearest Neighbor classifiers strictly depends on the training set dimension, which is directly proportional to the number of used subcarriers. In order to deal with this issue, we start from a naive approach based on random sampling of the input data to extract features, and then consider more advanced data dimension reduction algorithms, such as Principal Component Analysis (PCA) and t-distributed Stochastic Neighbor Embedding (t-SNE). We show that PCA is able to guarantee the best trade-off between authentication performance and complexity, while the application of t-SNE is effective when one wants to reduce data to a very small number of features.