COMPARATIVE EVALUATION OF AVERAGING METHODS FOR FILTERING MEASUREMENT SIGNALS

Abstract

To improve the quality of manufactured products, it is necessary to improve all technological
processes, which requires increasing the accuracy of the entire measuring path as a whole.
For this it is necessary to carefully analyze systematic, random and fluctuating errors in the measurement
channel and take all measures to reduce them. Digital filtering or averaging of intermediate
measurements (observations) according to certain rules is a radical means of improving the
accuracy of measurements performed. The aim of this work is to compare the quality of suppression
of near-real noise interference using the eight most well-known averaging methods. A model of the measurement path and a general block diagram for modeling the measurement process on a
computer under the influence of random interference are proposed for eight averaging algorithms.
As a criterion for evaluating the quality of averaging methods, the ratios of absolute error variances
and mean square deviations before the computing device and after applying the specified averaging
algorithm are taken. Based on the simulation results, the following conclusions are made. 1. All averaging
algorithms provide suppression of random error components of complex interference to the
level of 40–60 dB. Three algorithms are the best: arithmetic mean AR, a-truncated mean AU5 and atenderized
mean AB5, which provide for the suppression of 5 % of anomalous results. With an increase
in the number of observations, the suppression coefficients increase proportionally. 2. The
sampling time must be a multiple of the duration of the 50 Hz AC mains period (20 ms). The optimal
number of observations (measurements) is 100–128; with 128 measurements, the division operation
is reduced to a simple shift, and the averaging result can be obtained in 1–2 μs. 3. When experimentally
applying the AR averaging method for filtering a highly noisy measurement signal in a communication
line with a length of 800 m, a decrease in the spread of ADC output codes was observed
from ± 3.5 % to ± 0.1 % after filtering (AR, 64 measurements in 40 ms).