Data Collection

1. v å |S [ k ]| E out 10 N N Î( k , floor ( )) F 2 v s v S [ k ] E s [ n ] v x out 100 x [ n ] FFT max { E } v total : past å E |S [ k ]| v total w [ n ] N G Î k ( 0 , floor ( )) 2 Crosstalk Detection in Roadway Sensor Networks Joseph M. Ernst1, DhruvLamba1, James V. Krogmeier1, and Darcy M. Bullock2 1School of Electrical Engineering, Purdue University 2School of Civil Engineering, Purdue University • This study presents a novel signal processing based methodology to detect crosstalk, applies it to over 156 hours of training data and tests the algorithm on 14 hours of test data from 12 sensors. Abstract A C Different Types of Crosstalk in the Time Domain Crosstalk Detection Algorithm 4 Short Period Long Duration Crosstalk Long Period Short Duration Crosstalk Vehicle Signature with No Crosstalk Vehicle Signature with Significant Crosstalk 1 Crosstalk is caused by coupling between electromagnetic sensors operating at similar frequencies. Crosstalk typically leads to false vehicle detections and high noise levels in the signal. Crosstalk has been recognized as a problem since the early 1970’s [1]. 1 2 3 Approach • Characterize crosstalk in the Time Domain • Characterize crosstalk band in the Frequency domain • Develop an algorithm to measure the spectral energy belonging to Crosstalk in the signal • Estimate a threshold to determine the energy-level of the crosstalk band for which crosstalk can be declared likely A D Training the Algorithm B Crosstalk in the Frequency Domain B FFT of signal with vehicle detection and no crosstalk C Crosstalk Unlikely Crosstalk Likely Minimal Out of Band Energy D Relative Frequency Data Collection Northwestern and Stadium, Purdue University West Lafayette Out of band energy (%) 2 FFT of signal with vehicle detection and crosstalk Significant Out of Band Energy Detector Card Array Serial to IP Device Purdue 100Mbit Internet Cloud Purdue Server