Time series measurements of vehicle speeds are often performed by measuring the travel time of each vehicle between 2 given points in units of pulse counts from a pulse generator of known frequency. The travel time is then measured in discrete equidistant values, but 2 discrete values are possible for each actual travel time because of end effects in the counting technique. Such a measurement technique effects a transformation of the continuous spectrum of actual vehicle speeds into a discrete spectrum of observed speed values. Because speed is inversely proportional to the measured time, the separation of the discrete speed values increases hyperbolically with increasing speeds. A computer program was developed to simulate such a measurement technique. The data simulated by this program were used for a statistical analysis to quantify the errors and special effects of such a measurement technique on selected distributions, of speeds, distributions of relative speeds, and space parameters computed from the time series data. Among the results of this analysis are that the effects of the measurement technique can be generally ignored for speed distributions, if a certain set of intervals is used for classifying the speed values, but they cannot be made negligible for relative speed distributions. (Author/publisher)
Abstract