Differentiating ethernet devices using normal link pulse with efficient computation and the impacts on performance.

摘要

The tolerances in manufacturing Ethernet devices cause detectable differences in the signals sent by two different devices. Here, the design space is examined for using the IEEE 802.3 Normal Link Pulse (NLP) as the signal to use for differentiating devices. A previously collected set of NLP records as well as new sets of NLP data are used for testing the detection algorithm. Further tests have been run to determine the possibility of reducing the sampling rate to the point where Analogue-to-Digital Converters (ADCs) are more readily available and inexpensive. Reduced precision at each decimation was also tested. The design space survey indicates that trimming the time domain NLP records is beneficial to a certain point, and tracking the changes or drift of the signal has a great benefit. The design space survey also showed both wavelet-based filtering and noise spectra density scaling are beneficial on their own, but noise spectra density scaling can impair our algorithm when wavelet filtering is also being used. The tests on reducing sample rate and precision of the collected NLP records yielded results showing that sample rate effected false negative (device falsely unauthenticated) rates noticeably at decimation factors 8 and 16. Furthermore, false positive (devise falsely authenticated) rates were mostly effected by reduced precision. It is also apparent that performance of the algorithm, as determined by the impostor minimum to authentic maximum power mean squared error ratio, decreases with increasing data decimation before there is an increase in false negatives.