Aerospace and Electronic Systems - August 2018 - 33

Clausen et al.
numerical results of the calculations (bluedashed box in Figure 4). The second part,
situated on the left-hand side of the lower
half of the window (green-dashed box in
Figure 4), is used when the user wants to
choose the dataset containing the error
measurements. Different datasets are included in the package and serve as examples to learn about the features of the platform. The user may also upload their own
dataset that must be arranged by columns
in a txt-file. The empirical WV of the
selected data can then be calculated and
compared to the one of the datasheet. In
the third part situated in the middle of the
GUI (red-dashed box in Figure 4) the user
will choose a subset of the available models mentioned in Table 1. The GMWM
will then calculate the parameters depending on the user-selected model. The fourth
part on the right side allows the user to
adapt and tune the optimization process
as well as to define a datasheet when the
user decides to upload their own dataset
(orange-dashed box in Figure 4).
The data included in the software
are error measurements from different
kinds of sensors ranging from low-cost
MEMS-IMUs to navigational-grade Figure 4.
high-end IMUs. The data consists of Overall view of the GUI with all the information as seen online. The upper part displays the graphical
and numerical results. The lower parts allow for data-uploading and selection, noise-modeling, and setstatic measurements of the two types of
ting of options related to the computations.
sensors (gyroscopes and accelerometers)
where each are made by three axes. The
lengths of the error measurements vary between datasets; but in
general, they should be taken over a period of at least 30 minutes
at a relatively high frequency. In general, the longer the dataset,
the better the estimation should be; however, there is no minimal
length of the dataset imposed. The mean of the dataset is irrelevant
for the calculations as it is removed automatically.

EMPIRICAL WV AND DATASHEET VALUE
Once the data source is selected, the user can overlay the datasheet
values by activating the checkbox Add Datasheet WV in the fourth
part of the GUI. A click on the button Plot WV launches the calculations of the empirical WV vˆ and, if necessary, also the theoretical
WV ν based on the datasheet parameters. The result is shown in
Figure 5 and will be plotted in the upper half of the actual GUI.

MODEL IDENTIFICATION (ITERATIVE)
The log-log plot of the WV is useful to understand what kind of
models underlie the provided error signal. An example of how this
plot is useful in this sense is given in Figure 3, where different
slopes and characteristics of the plot can indicate the presence or
absence of certain stochastic models. If we compare the latter plot
AUGUST 2018

Figure 5.

Empirical WV representation (blue line/dots) of the gyroscope y-axis
with its confidence interval. Datasheet specified theoretical WV displayed by red line/circles.

with the one given in Figure 5 it is clear that the signal is composed
of at least a QN, a WN, and several GMs.
To check which combination of models appear to fit best, we
select the desired models in the middle part of the GUI and use
the Fit Model button. The result of selecting the composite model
made by the sum of a QN + WN + GM noise (error) models is shown
in Figure 6a and is represented by the orange line. Given that the

IEEE A&E SYSTEMS MAGAZINE

Aerospace and Electronic Systems - August 2018

Table of Contents for the Digital Edition of Aerospace and Electronic Systems - August 2018