error distribution of range measurements
TRANSCRIPT
-
7/30/2019 Error Distribution of Range Measurements
1/2
Error Distribution of Range Measurements in
Wireless Sensor Networks (WSNs)
In this paper the main objective is to scrutinize this assumption for different environments.
Experiments are performed in both outdoor and indoor environments with Line-of-Sight (LOS)
and Non-Line-of-Sight (NLOS) conditions using IEEE 802.15.4 compliant devices. These
devices consist of a low-cost 2.45 GHz chipset and use Time-of-Flight (TOF) measurement
technique for range estimation. Our results and analysis are based on four different statistical
Goodness-of-Fit (GOF) tests i.e., a graphical technique, Linear correlation coefficient, Anderson-
Darling, and Chisquared for investigating the Gaussianity of range measurements.
In several sensor networks mainly for environmental applications such as water quality
monitoring, precision agriculture, and indoor air quality monitoring, the on hand sensing data
may be rendered useless by a lack of accurate sensor location estimation. The availability of
accurate sensor location estimates can help reduce configuration requirements and device cost.
Further, accurate sensor location estimates enable applications such as inventory management
,intrusion detection, detecting and monitoring car thefts, and vehicle tracking and detection .The
assumption of Gaussianity is prevalent and fundamental to many statistical theories and
engineering applications. In recent years a lot of research has been done in the field oflocalization in WSNs and all the existing localization methods are based on the Gaussianity
assumption such that either the range estimates r or the noise/error distorting the range
estimates are assumed to be Gaussian distributed.
GOODNESS OF FIT (GOF) TESTS-
A. Graphical TechniqueB. Linear Correlation Coefficient ()C. Anderson-Darling Test (A2)D. Chi-Squared Test (_2)The experiments have been performed in an indoor and an outdoor environment with both LOSand NLOS conditions. For each condition three different sets of experiments have been
performed and for each set of experiments the transmitter and receiver nodes are mounted on a
tripod at a specific height. The three different heights used are 0.5 m, 1.0 m and 1.5 m. At each
-
7/30/2019 Error Distribution of Range Measurements
2/2
height five experiments were performed with different ranges of 10 m, 20 m, 30 m, 40 m and 50
m. For each range 500 iterations are executed giving 500 range estimates.
GRAPHICAL AND NUMERICAL RESULTS-
It is observable from the results that each method is powerful in scrutinizing the hypothesis of
Gaussianity however the graphical and techniques are less consistent but simpler to use in
comparison with the A2 and _2 tests. The end results acquired from A2 and _2 appears more
valid and authentic as they can be compared with standard critical values for these tests whereas
in the case of graphical technique the outcomes are to be visually examined and for linear
correlation coefficient, , technique some generalization has to be made for validating the
hypothesis. The results encourage further investigation into the topic. The central limit theorem
(CLT) states that the sum of large numbers of independent, statistically more or less identical,
random variables has an approximately Gaussian distribution, so in future the effect of the
number of range estimates n on Gaussianity will be examined and the outdoor NLOS condition
will be studied in more detail. It will also be investigated why 75% of the result negates the
hypothesis of Gaussianity. If Gaussian distribution is not the perfect model for the error
distribution in range estimates, then which statistical model can best describe the observational
error distribution in ranging.
Submitted byPriti Singh Tanwar
(1125938)