Going back to Singapore almost a year later, this time to give a talk about new machine learning models developed by ReachRF for the latest Doppler sensor: ADS. See you at the 2017 IEEE Global Communications Conference: Ad Hoc and Sensor Networks
Cepstral Analysis for Classifying Car Collisions in LOS/NLOS Vehicle-to-Vehicle Networks
In this work the Doppler effect is exploited for identifying a collision between two vehicles under Line-of-Sight (LOS) and Non-Line-of-Sight (NLOS) conditions. A measurement campaign at 5.9GHz is conducted to observe the Doppler effect for Vehicle-to-Vehicle (V2V) networks. In our previous work, we developed a collision avoidance estimator based on tracking the LOS Doppler shift in the Doppler domain. We now report real-world captures of the Doppler phenomenon in terrestrial V2V networks during a collision and present a method for collision identification by tracking the NLOS Doppler spread in the channel. Through machine learning, classification of collision feature observations is accurate with an accuracy of nearly 100% and reliable with a misclassification rate less than 1%.