International Journal of Remote Sensing

ISSN: (Print) (Online) Journal homepage: www.tandfonline.com/journals/tres20

High-precision 1D distance monitoring with radar

data

Christian Radach, C. Köhler & J. Benndorf

To cite this article: Christian Radach, C. Köhler & J. Benndorf (2021) High-precision 1D distance
monitoring with radar data, International Journal of Remote Sensing, 42:24, 9212-9229, DOI:
10.1080/01431161.2021.1987577

To link to this article: https://doi.org/10.1080/01431161.2021.1987577

© 2021 The Author(s). Published by Informa

UK Limited, trading as Taylor & Francis
Group.

Published online: 10 Nov 2021.

Submit your article to this journal

Article views: 1542

View related articles

View Crossmark data

Citing articles: 1 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=tres20
INTERNATIONAL JOURNAL OF REMOTE SENSING
2021, VOL. 42, NO. 24, 9212–9229
https://doi.org/10.1080/01431161.2021.1987577

High-precision 1D distance monitoring with radar data

Christian Radach, C. Köhler and J. Benndorf
TU Bergakademie Freiberg, Institute of Geodesy and Mine Surveying, Freiberg, Germany

ABSTRACT ARTICLE HISTORY

A measurement setup based on a high-precision, low-cost FMCW Received 6 June 2021
radar sensor has been developed for radar-based 1D deformation Accepted 20 September 2021
monitoring of engineering structures. The method enables contact­
less and time-continuous scans, but has lower acquisition costs
compared to commercially available terrestrial laser and radar tech­
nologies. The article contains the instrument-related description of
the sensor with accuracy tests for 1D distance measurements on
a laboratory scale as well as a case study for monitoring the move­
ments of a headframe over the course of the day. The detectability
of deformations was statistically verified by means of a hypothesis
test and applied to the case study under real conditions. With
a defined target object (corner reflector), measuring accuracies of
up to 10 μm with a 99,7% confidence level could be achieved. The
case study shows absolute movements of the headframe during
the day of 1 mm. Changes of 0,2 mm can be detected in time
intervals of 9 min with a statistical significance level of 95%.

Abbreviations : CA-CFAR: Cell-averaging constant false alarm rate; CR-

I, CR-II: Corner reflectors; CUT: Cell Under Test; DFT: Discrete Fourier
transformation; EXP: Experiment; FMCW: Frequency-modulated con­
tinuous-wave; iSDR-C:indurad ScanningDynamicRadar-Compact; LOD:
Limit of detection; PI: Peak interpolation

1. Introduction
Both airborne and terrestrial radar sensors have a wide range of possible applications in
geomonitoring. The technology Interferometric Synthetic Aperture Radar (InSAR) as
a satellite-based method has been established for decades in remote sensing for the detec­
tion and monitoring of in particular slow ground movements on the earth’s surface, caused
by glacier movements, mining-related subsidence or different types of landslides (Butt et al.
(2020); Caduff et al. (2015); Joughin et al. (1996); Läufer, Lehmann, and Rödelsperger (2017);
Lehmann (2015); Bürgmann, Rosen, and Fielding (2000); Tapete et al. (2013)). The area to be
monitored is recorded several times by repeated overflights and from different viewing
angles and evaluated interferometrically, whereby changes can be detected Pieraccini and
Miccinesi (2019); Moreira et al. (2013). Due to the comparably low frequency of overflights,
satellite-based InSAR is more suitable for observing slow phenomena rather than rapid
ground movements.

CONTACT Christian Radach [email protected] TU Bergakademie Freiberg, Institute of Mine

Surveying and Geodesy, Freiberg, Germany
© 2021 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group.
This is an Open Access article distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivatives License
(http://creativecommons.org/licenses/by-nc-nd/4.0/), which permits non-commercial re-use, distribution, and reproduction in any med­
ium, provided the original work is properly cited, and is not altered, transformed, or built upon in any way.
 INTERNATIONAL JOURNAL OF REMOTE SENSING 9213

In terrestrial radar interferometry, also known as ground-based InSAR (GBInSAR), the

measuring principle of the InSAR technology is largely adopted in order to be able to
measure long-range geometric changes of objects directly from the earth’s surface. The
difference and advantage to airborne and spaceborne radars, in which the direction of
movement often is unfavourable to the viewing direction of the satellite, consists primarily in
the adaptable observation direction of the radar relative to the expected movement of the
object (Lehmann (2015); Läufer, Lehmann, and Rödelsperger (2017)). In addition, terrestrial
radar sensors offer the possibility of a timely continuous and extensive detection of defor­
mations, even without direct access to the measuring object (Lehmann (2015)). Though,
commercially available GBInSAR measuring systems are currently still very expensive.
A different type of ground-based radar devices are also used at close range for distance
measurement in the automotive sector or other technical applications, such as level or
door sensors. Frequency-modulated continouis-wave (FMCW) radar sensors are usually
used for this purpose. Such low-cost FMCW-devices have extremely high distance accura­
cies in the sub-mm range down to below 1 μm and are therefore an order of magnitude
more precise than laser measurement technology (Scherr (2017); Scherr et al. (2017)).
However, the acquisition costs of an FMCW sensor are around 10 times lower than
commercially available terrestrial radar interferometers.
The aim of this contribution is to investigate the suitability of FMCW radar sensors from
industrial measurement technology for high precision and short-range measuring geo­
monitoring and engineering surveying. A newly adapted non-contact and time-
continuous measurement method is described that includes an FMCW radar sensor for
1D distance measurements. The measurement setup developed for this sensor is also in
the low-cost segment, and makes its application attractive compared to the high-
precision deformation measurement technology available on the market, such as terres­
trial laser scanners. The accuracy in 1D distance measurements is examined under
laboratory conditions and for use in 1D monitoring of engineering structures. As a case
study, the movement of a mining shaft headframe is observed over the course of a day.
Absolute movements of the headframe in the range of about 1 mm could be observed.
Using a statistical test procedure, statistically significant movements of around 0,2 mm
could be detected with a confidence level of 95%.

2. Materials and methods

2.1. Measurement and evaluation
2.1.1. Radar sensor specification and measuring principle
The indurad ScanningDynamicRadar-Compact (iSDR-C) used in this study is a 360° scan­
ning FMCW sensor manufactured from indurad GmbH (indurad GmbH (2021)). The entire
360° angle range is resolved to 2000 angle settings, corresponding to an angular resolu­
tion of 0,18°. Although angular ranges can be set, the sensor is used with a fixed angular
position throughout the paper.
The FMCW principle realizes a linear frequency modulation, i.e. the frequency νðtÞ
changes during the measurement at a constant rate kν . The bandwith B ¼ νend νstart of
the transmitted pulse is covered within the adjustable ramp duration τ ramp (Läufer,
Lehmann, and Rödelsperger (2017)):
9214 C. RADACH ET AL.

Figure 1. Principle of linear frequency modulation with exemplary values for τramp and B (Läufer,
Lehmann, and Rödelsperger (2017)).

B νend νstart
kν ¼ ¼ ; (1)
τ ramp τramp
where νstart and νend are the start and end frequencies. The time-dependent frequency
with linear chirp can be represented as follows and is shown in Figure 1:
νðtÞ ¼ νstart þ kν � t: (2)
The iSDR-C sensor works with a maximum bandwidth of 10 GHz in the range from
72 GHz to 82 GHz, the parameters being adjustable within this range. Thus the average
radar wavelength λ is 3,9 mm. τramp can be set in the range from 1 μs to 1 s.
The signal reflected from a target at a distance r travels at the speed of light c0 and is
received at the antenna with a time delay Δt ¼ 2r=c0 , with the lower frequencies arriving
first. The frequency ramp leads to a difference between the received and the transmitted
instantaneous frequency Δν (Figure 1):
2B � r
Δν ¼ kν � Δt ¼ : (3)
τramp � c0
The further processing contains the multiplication of the instantaneous transmitted signal
by the signal received at the same time, which gives the raw data. Thus τ ramp has to enable
the overlap of the transmitted and reflected signal and physically limits the longest
possible range for distance measurement. The multiplied raw data are sampled in time
domain at a rate νsampling , resulting in Nsamples data points:
Nsamples ¼ τ ramp � νsampling : (4)

2.1.2. Signal processing for obtaining distances

2.1.2.1. Fourier transformation. After the multiplied raw signal has been sampled in
the time domain with νsampling , the data is transformed into the frequency domain via the
discrete Fourier transform (DFT) (Scherr (2017)). Sampling takes place within the fixed
window from 0 to Nsamples .
Before the actual transformation, the sample values are weighted using a window
function. In this case, the so-called Von Hann window is specifically used, which is
particularly helpful in suppressing possible sidelobes of the radar signal (Blackman and
Tukey (1959); Dahlen (1982)).
 INTERNATIONAL JOURNAL OF REMOTE SENSING 9215

^ which have an amplitude

The result of the DFT are the complex Fourier coefficients C,
and a phase. These form a spectrum in the frequency domain. The range defined by
Nsamples in the time domain is transformed by the DFT into an equidistant Fourier grid with
discrete frequencies. These frequencies correspond to the shifts Δν that arise due to the
time delay Δt of the transmitted and received signal (section 2.1.1). The distances r are
obtained from Equation (3):
Δν � τramp � c0
r¼ : (5)
2B

2.1.2.2. Distance resolution of DFT. The discrete Fourier grid with distances r specifies
the minimum theoretical range resolution δr of possible peaks, which corresponds to the
distance between two grid points. δr is given by c0 and the bandwidth B (Pieraccini and
Miccinesi (2019); Richards et al. (2010); Läufer, Lehmann, and Rödelsperger (2017)):
c0
δr ¼ : (6)
2B
An adjustable bandwith of B ¼ 10GHz results in a theoretical range resolution of δr ¼
14; 99 mm.
In addition to the distances the phase ϕ of the Fourier coefficients also provides
information about distance differences and is periodic in the range π < ϕ < π. The
absolute phase does not allow the direct determination of a measured distance (Scherr
(2017)). Only the phase shift Δϕ ¼ ϕ2 ϕ1 as the difference between two phase values
can be evaluated as a change in distance Δr. Δr results from the following equation with
the radar wavelength or the start frequency (Moreira et al. (2013)):

λ c0
Δr ¼ � Δϕ ¼ � Δϕ: (7)
4π 4π � νstart

With the known νstart , a phase period 2π corresponds to a change in distance of approxi­
mately 2,08 mm.

2.1.2.3. CFAR-algorithm and peak interpolation. After DFT the discrete spectrum of
an incoming signal must be distinguished into peaks, i.e. a signal reflected from the target
and noise. Therefor a thresholding method with variable threshold T, specifically a cell-
avering constant false alarm rate (CA-CFAR) algorithm with constant false alarm rate Pfa ,
was used in this work (Figure 2 (a)) (Rohling (1983); The MathWorks). Pfa indicates the
number of false targets detected in a certain time. For a real signal, the amplitude must
exceed T. The setting of T influences the number of wrong targets detected and correctly
identified targets.
Each discrete distance from the Fourier grid is processed, with the current point of the
grid being referred to as the Cell Under Test (CUT). The CFAR detector checks for each CUT
whether the threshold value is exceeded or not. The number of CUT is equal to the
number of data. T is calculated for each CUT using a definable number of Training Cells
9216 C. RADACH ET AL.

Ntrain and Guard Cells Nguard , which are around the CUT. The cell averaging method used in
this process averages the absolute values of the signal amplitudes of the Ntrain Training
Cells:

� �
1
Ntrain 1 NX train

T ¼ α � Pn ¼ Ntrain � Pfa 1 � � xm ; (8)

Ntrain m¼1

where α is the so-called threshold factor, xm the absolute amplitude values and Pn the noise
power estimate. The parameters Pfa , Ntrain used by the CA-CFAR algorithm can be adjusted.
The theoretical range resolution (Equation (6)) limits the differentiation of neighbouring
signal peaks of different target objects in the Fourier grid and the distance determination.
^ ¼ a � ðr rint Þ2 þ b is fitted at the peak of
For more precise localization, a function jCj
a recognized target (Figure 2 (b)). The amplitude jCj^ at the peak maximum position rpeak
and at the two neighbouring points (C1 , C2 and C3 in Figure 2 (b)) are used for a quadratic
fit. The maximum of this parabola indicates the position of the peak more precisely (Smith
(2011)). The interpolated distance rint is obtained as follows:

C1 C3
rint ¼ δr � þ rpeak ; (9)
2C1 4C2 þ 2C3

where C1;2;3 are the signal amplitudes, rpeak is the peak distance of the Fourier grid and δr
is the distance between two points of the Fourier grid. The parameter rint calculated by the
quadratic peak interpolation (PI) indicates the more precise peak position and this
significantly improves the range resolution from the theoretical value of about 15 mm
to less than 0,1 mm, as shown in Chapter 3.

Figure 2. (a) Schematic representation of the CA-CFAR algorithm for the 1D case. The number of Guard
and Training Cells is adjustable, but equal on both sides of the Cell Under Test (Rohling (1983); The
MathWorks, Inc. (2020)). (b) Principle of the quadratic peak interpolation with the signal (black curve)
and the quadratic function through three points nearest the peak (blue curve).
 INTERNATIONAL JOURNAL OF REMOTE SENSING 9217

2.2. Experiments
With regard to a use for 1D geomonitoring, a characterization of the FMCW radar sensor
used and a verification of its accuracy was carried out.

2.2.1. Lab experiments: precision and trueness assessment

The iSDR-C sensor was examined for its accuracy (precision and trueness) with regard to
1D distance measurements under laboratory conditions. For this purpose, static distance
and phase measurements with a fixed angular position were recorded. In order to test the
influence of materials with different surface properties on the accuracy of distance
measurements, the experiments are carried out with different target objects (Figure 4).
The distances obtained from the experiments are evaluated and compared for the
methods of PI from the Fourier grid and the phase difference Δϕ.
The general measurement setup is shown in Figure 3 and consists of the sensor and the
reflector to be tested. The radar is attached to a linear rail, which enables the systematic
change of the antenna-target-distance in increments of 0.1 mm. Thus, series of measure­
ments with a distance change of 20 mm could be recorded.
As targets with defined reflection properties regarding the radar waves, a flat alumi­
nium plate and two different trihedral triangular reflectors made of aluminium with an
edge length of approximately 20 cm, so-called corner reflectors (CR), were used.
A triangular CR consists of three electrically conductive triangular surfaces arranged at
right angles to one another, which reflect microwave radiation precisely back in the
direction of incidence through multiple reflections (Garthwaite et al. (2015); Garthwaite
(2017); Sarabandi and Chiu (1996)). For comparison, the measurements were also carried
out on objects with an irregular surface and undefined backscatter properties (rock
sample and drywall). The tested targets are shown in Figure 4.

Figure 3. Schematic representation of the measurement setup with adjustable distance from the
sensor to the target object. The coloured beam of the radar shows the opening angle of the antenna
radiation with a loss of intensity. The loss of intensity is – 3 dB at an angle of � 4 °.

Figure 4. Images of five different target objects for accuracy studies. The corner reflectors have an
edge length of approximately 20 cm (CR-I and CR-II).
9218 C. RADACH ET AL.

Figure 5. Headframe of the visitor mine ”Reiche Zeche”.

2.2.1.1. Experiment A: precision. In order to evaluate the precision of the iSDR-C sensor
in 1D distance measurements, the measurement error for the different targets was
estimated by 100 repeated measurements each (Norm (1990)). Both the evaluation via
PI and Δϕ were considered. The tests were carried out in a fixed angular position at
a constant distance, the selected distances being in the range of 2 m to 4,5 m and thus on
a laboratory scale. Two separate series of experiments were realized, of which the first
measurements were carried out indoors and the second outdoors on a house wall.
External influences such as wind are therefore added to the tests on the house wall.

2.2.1.2. Experiment B: trueness. To prove detectable changes and to examine the

sensor for its trueness, a series of measurements was made with a controlled change in
distance in steps of 0,4 mm. For each set distance, the values of the PI and the phase were
recorded at least three times. EXP B aims to demonstrate the linearity of measured
changes in distance as a function of the distances set using the linear rail.

2.2.2. Case study: monitoring movement of a mining shaft headframe

As realistic case study, 1D monitoring was carried out using the FMCW radar sensor iSDR-C
on the specific example of the headframe of the Reiche Zeche mine in Freiberg (Figure 5).
The aim was to track the movement of the headframe over the course of an entire day
using continuous distance measurements. Due to the various movement processes
caused by the weather and the mine operation, the headframe of a mine is ideally suited
as a test object for deformation monitoring.
The measurement with the sensor was carried out with a ratio of about 30 values/min
in a restricted horizontal angular range 3,6 °, which is sufficient for the detection of the
headframe. The sensor distance to the headframe was about 36 m. For the evaluation, all
measured distances of each angular position are considered separately.
The signal (peak) with the highest intensity is used for each measurement. This ensures
that the same point on the headframe is always viewed with the radar, although the object
offers a large number of possible backscatterers due to its structure made of steel girders.
The recorded distances or changes are to be used to test the movements for significance.
In other words, it should be proven if the winding tower has moved during the day or not.
 INTERNATIONAL JOURNAL OF REMOTE SENSING 9219

2.3. Evaluation of experiments

2.3.1. Precision
For the repeated measurements of EXP A, the empirical standard deviation of the
individual measurements sx was used, which characterizes the variance around the
arithmetic mean value �x. The values were calculated using both the phase difference
and the PI. The random largest error ΔX was estimated for a number of individual
measurements n ¼ 100 with a confidence level of 99,7% corresponding to the range
pffiffiffi
�x � 3sx . The empirical standard deviation of the mean value s�x ¼ sx = n as an estimate for
the accuracy of the arithmetic mean was also used (Norm (2010); Pruscha (2006)).
For the phase analysis, the standard deviation of the absolute phase ϕ was used as Δϕ.
Using Equation (7) these differences Δϕ were translated into changes in distance Δr.

2.3.2. Trueness
The measurement trueness of EXP B was assessed by means of a linear fit using the
interpolated distance values. Ideally, the measured values should have a linear relation­
ship with a coefficient of determination of R2 ¼ 1; 0 and a slope of 1,0. In a linear
regression with n given value pairs ðxi ; yi Þ, the empirical mean �y and an estimator ^y
according to the following equation, R2 is calculated as the ratio of the sums of squares:
Pn
2 ð^yi �yÞ2
R ¼ Pi¼1 n
: (10)
i¼1 ðyi ^yÞ2

The slope m of the regression line yi ¼ m � xi þ c results from:

Pn
ððxi �xÞ � ðyi �yÞÞ
m ¼ i¼1Pn : (11)
i¼1 ðxi �xÞ2

2.3.3. Statistical evaluation

The standard deviations determined by multiple static measurements can be used to
calculate a minimum limit of detection (LOD) for the measured 1D distances by means of
a statistical hypothesis test. For this purpose the so-called two-sample Gaussian test, also
known as the z-test, is used as a test for standard normally distributed test variables with
known standard deviations. Based on the 1D accuracy measurements carried out in this
study, the sample mean values �x0 and �x1 of two series of measurements recorded at
different times t0 and t1 can be compared with one another. In this test, hypotheses about
the expected values of the basic populations are drawn up and tested using the sample
mean values. The difference of the mean values is considered to test whether they differ
statistically significantly from one another:
Δ ¼ μ1 μ0 ; (12)
where μ0 and μ1 are the true mean values or expected values of the measuring distances.
The combined standard deviation of the mean values of both series of measurements
results from:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
s�x1 �x0 ¼ s�2x1 þ s�2x0 : (13)
9220 C. RADACH ET AL.

The null hypothesis H0 and the corresponding alternative hypothesis HA for the test
whether the difference between the two mean values is systematic or random can be
determined as follows:

H0 : Δ ¼ μ1 μ0 ¼ 0; (14)

HA : Δ�0: (15)

Certain wrong decisions exist for both hypotheses, the so-called errors of type I and type
II. A type I error is present when the null hypothesis H0 is incorrectly rejected even though
it is true (HA becomes accepted). On the other hand, a type II error means the assumption
of H0 , although the alternative hypothesis HA would be correct. The quality of the
hypothesis test is largely given by the selected, acceptable error probabilities α (type
I error) and β (type II error).
The hypothesis H0 assumes that there is a zero difference between the averaged
distances at both measurement times t0 and t1. This results in the statistical test variable
z for H0 :

�x1 �x0 Δ
z ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : (16)
s�2x1 þ s�2x0

The variable z is compared with the quantile of the standard normal distribution u, which
corresponds to the desired error probability α. In this case the test is applied to a general
change of μ, regardless of the direction. Therefore the α=2 quantile u1 α=2 is used and H0 is
assumed if jzj < u1 α=2 applies. Such a two-sided problem is shown in Figure 6:
In addition to a significance test, the z-test can be used to determine a minimum ΔLOD
as limit of detection. With the given error probabilities α and β and the associated
quantiles u1 α=2 and uβ , the smallest significantly detectable distance between the
expected values μ1 μ0 will be calculated. ΔLOD consists of two parts for the type I and
type II errors (Figure 6). These are the confidence intervals for 1 α=2, based on H0 , and β,
based on the alternative HA . These confidence intervals (red and blue areas in Figure 6)
can be calculated from Equation (16). The assumption μ1 μ0 ¼ 0 for H0 applies to the
difference in the sample mean values:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
�x1 �x0 ¼ u1 α=2 � s�2x1 þ s�2x0 : (17)

ΔLOD is obtained for the assumption Δ ¼ ΔLOD �0 for H1 by using Equation (16), where uβ is
a negative number:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
�x1 �x0 ΔLOD ¼ uβ � s�2x1 þ s�2x0 : (18)

ΔLOD results as the smallest value, which conforms both requirements for z with the error
probabilities α and β, by setting both Equation (17) and (18) equal:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ΔLOD ¼ ðu1 α=2 uβ Þ � s�2x1 þ s�2x0 (19)
 INTERNATIONAL JOURNAL OF REMOTE SENSING 9221

Figure 6. Two-sided hypothesis test for change detection. The limit of detection results from the areas
for the type I and type II errors in the graphic.

In all further considerations of this study, an acceptable risk for the errors of the type I and
II of α ¼ β ¼ 5% is assumed. With these probabilities the quantile u1 α=2 takes the value
1,9600 and the quantile uβ the value −1,6449 (De Gruijter et al. (2006)). For the estimated
standard deviations of the measurement series considered here, s�x0 ¼ s�x1 applies. Based
on the measurement errors determined in EXP A, the associated limits of detection ΔLOD
were calculated using Equation (19).

3. Results
3.1. Experiment A: precision
A comprehensive overview of all determined measurement errors from the static multiple
measurements and the calculated limits ΔLOD (n ¼ 100) can be found in Table A1 in the
appendix for the various target objects and measurement distances from 2 m to 4,5 m.
The estimated errors 3sx determined at a constant distance from the sensor to the target
are graphically shown in Figure 7 for the evaluation by means of interpolation and phase.
For all tested target objects, there are errors 3sx in the calculated distances of less than
one millimetre with an unchanged sensor-target distance.
Differences between the materials can be identified. But there is no clear trend that the
standard deviation increases or decreases with target distance, according to the laws of
physics for radar waves.
Overall, the two corner reflectors clearly show the smallest errors 3sx of around 5 μm
for peak interpolation and less than 1 μm for phase evaluation. The largest values have
been measured for the targets rock sample, house wall and the reflector CR-II on a steal
surface with over 500 μm for interpolation and about 20 μm for phase evaluation. The
errors of the phase evaluation are generally smaller compared to peak interpolation, on
average by a factor of five.
As Figure 7 shows, the measurement accuracy generally depends on the type of target
object and its arrangement. Corner reflectors as point targets provide up to ten times
higher precision for static distance measurements than planar targets such as a wall or
rock surface. Furthermore, the errors for a stand-alone reflector are smaller compared to
a target object in front of a planar background which influences the backscatter signal.
9222 C. RADACH ET AL.

Figure 7. Bar plots of the 3sx values for the peak interpolation method (left) and the phase evaluation
(right). The measurement errors are shown for distances from 2 m to 4,5 m and different target
objects, respectively.

Comparing methods, the phase evaluation provides smaller measurement errors com­
pared to the peak interpolation, which can contribute to the measurement of smaller
changes. However, the consideration of the phase with regard to the measurement setup
is also more susceptible to external influences.
The limits ΔLOD (Table A1) are calculated for n ¼ 100 single measurements. The values
show that with the tested corner reflectors a limit of detection for one-dimensional
distance measurements with the iSDR-C radar sensor of about 1 μm to 2 μm for PI can
be achieved. With the phase evaluation it was even possible to determine values for ΔLOD
below 0,1 μm with given errors α ¼ β ¼ 5%. It should be noted that this detection limit is
a theoretical value that was determined under ideal laboratory conditions using the
statistics mentioned. Practically detectable changes are discussed in Chapter 3.3.

3.2. Experiment B: trueness

The results of the accuracy study based on controlled changes in distance are graphically
shown in the plots of Figure 8 exemplarily for the corner reflector CR-II and the drywall.
Depending on the changed distance in mm, the discrete peak distance (black curve), the
quadratic interpolated distance (red curve) and the phase shift (blue curve) are plotted
(Figure 8 (a) and (c)). The unwrapped phase and the calculated distances changes from Δϕ
using Equation (7) (blue curve) are plotted too (Figure 8 (b) and (d)).
The black curve with a stepped course shows the distance resolution of the Fourier grid
with about 15 mm. A better resolution can be seen from the red curve of the interpolated
data. Changes in phase ϕ from the Fourier transformation are shown with the blue curve
and can be translated into distance differences as well. For the radar sensor used with
νstart ¼ 72 GHz, a phase period 2π corresponds to a distance of approximately 2 mm.
Based on the results of EXP A and the blue curve in Figure 8, the phase evaluation proves
to be the method with the highest resolution for changes in distance.
 INTERNATIONAL JOURNAL OF REMOTE SENSING 9223

Figure 8. Plot of the measured discrete peak positions (black curve), interpolated distances (red curve)
with linear fit (green line) and absolute phase ϕ (blue curve) depending on the change in distance Δr
for the targets CR-II (a) and drywall (c). Plot of the unwrapped phase and the calculated distances (blue
curve) using the phase with linear fit (green line) depending on the change in distance Δr for the
targets CR-II (b) and drywall (d).

The coefficients of determination R2 from the linear fit are compared with one another.
The reflector CR-II has the highest value with R2 ¼ 0; 9999 (Figure 8(a)). This coincides with
the error estimation from EXP A and shows that this reflector has the best reflection proper­
ties in terms of accuracy. The reflector CR-I has a slightly worse coefficient of determination
with 0,9991, as does the aluminium plate with 0,9996. Accordingly, both objects also have
good accuracies for reflection, the properties of a flat surface strongly depending on the
angle of incidence of the radar waves. Only if the irradiation is perpendicular does it reflect
like a special corner reflector. Therefore, a reflector is preferred as a radar target.
Equivalent to the individual measurements at a fixed distance (EXP A), the surfaces of
the wall and the rock sample show the largest inaccuracies and deviations from linearity.
The R2 values of 0,9985 (drywall) and 0,9979 (rock sample) confirm this, as does Figure 8(c).
The phase evaluation results in R2 values for all tested targets of 0,9999 or higher
(Figure 8(b) and (d)), which proves the high trueness regarding the phase evaluation
method. The slopes of all fitted curves are approximately 1,0.
9224 C. RADACH ET AL.

3.3. Case study: monitoring movement of a mining shaft headframe

The time courses of the measured distances at one specific angular position with the
largest amount of available data are plotted in Figure 9 as changes for the evaluation of PI
and Δϕ in relation to the start value. A total of 4500 individual measurements were
recorded for this position, which corresponds to a rate of around 3,5 values per minute.
The graphs in Figure 9 show that the measured distances from the sensor position to
a certain reflective point on the headframe change over time. These changes have an
absolute magnitude of about 0,8 mm for the PI evaluation and 1 mm for the phase based
on the blue averaged curve. The single measurements (green points) are subject to
a certain amount of noise sx (red area). sx is significantly smaller in the phase evaluation
compared to the interpolation, which confirms the results from EXP A.
It is generally observed that fluctuations in the measured values occur especially in the
afternoon between 12:00 and 16:00. During this period of time, the solar radiation comes
from the direction in which the distance from the sensor to the headframe is also
measured. This causes an increased, measurable movement of the object in the measur­
ing direction. Other possible causes of short-term movements of the headframe are man-
rides in the mine or wind, which may have caused the target object to vibrate.
Two corner reflectors were attached to the target object in order to achieve the
accuracies of individual measurements in Chapter 3.1 measured under laboratory condi­
tions. However, the evaluation of the case study shows, that the two reflectors do not
stand out from the overall structure of the headframe and the reflected signal cannot be
clearly assigned to a corner reflector. The reason for this is on the one hand the steel
structure of the object, which has a large number of reflective surfaces for the radar
waves. On the other hand, the antenna footprint has a high influence on the back­
scattered signal. With an opening angle of the antenna radiation of � 4 ° at a distance
of 36 m, the antenna footprint corresponds to an illuminated circular area with a diameter

Figure 9. Comparing plot of all single distance measurements as changing course for the angular
position – 0,54° over the time of day. The blue curve shows the moving averages of the distances and
the red area the standard deviations sx .
 INTERNATIONAL JOURNAL OF REMOTE SENSING 9225

of approximately 5 m. In relation to the size of the reflectors used, the footprint is larger by
a factor of about 1000. Therefore, the evaluated signal cannot be assigned to an exact
position on the headframe or to a reflector.
In order to be able to statistically assess a possible movement of the headframe, time
intervals with corresponding mean values and their scatter must be evaluated and com­
pared with one another. By means of the hypothesis test described and Equation (19), a LOD
can be calculated using s�x for each time interval. The differences between the mean values
of successive intervals can then be compared with the calculated LOD. With given prob­
abilities for the errors type I and II, a statistically reliable statement can be obtained as to
whether the headframe has moved significantly in certain time segments or whether the
distances have remained the same within the scope of the measurement uncertainty.
The observed interval durations can be motivated by using Equation (19).
sx ΔLOD ðnÞ ΔLOD nsx ¼ 0; 22 mm ΔLOD ¼ 0; 2 mm By choosing a desired detection limit
of ΔLOD ¼ 0; 2 mm and using the experimentally determined standard deviation
sx ¼ 0; 22 mm, approximately 30 measurement values or time inervals of 9 min result.
In Figure 10 the statistical evaluation is shown based on the mean values in 9 min i­
ntervals for both evaluation with PI and Δϕ. The red areas in both representations show
the time intervals at which the mean sensor-headframe distances have changed signifi­
cantly from the previous interval. The statistical certainties α and β specified in chapter 2.3
were used for this purpose. According to Figure 10, the phase evaluation shows signifi­
cantly more time intervals with statistically significant distance changes compared to the
PI. In addition, the mean value (black curve) has fewer fluctuations and the course of
the day with a maximum difference of around 1 mm is more clearly recognizable
compared to the interpolation. The error range of s�x supports this statement.
The results of the case study are of a similar order of magnitude to those of (Martienßen
(2017)), who observed a maximum deflection of the headframe of about 5 mm using
tachymetry with a Trimble S8 total station (Braun (2015)). These tachymetry measure­
ments were carried out in June. Indeed, the movement of the headframe is heavily

Figure 10. Plot of time-averaged distances for 9 min intervals (black curve) for the angular position –
0,54 ° with interval standard deviations sx (Orange area). The blue and red areas show the result of the
statistical evaluation corresponding to a limit of detection of 0,2 mm.
9226 C. RADACH ET AL.

dependent on the day and the weather, since solar radiation in particular influences the
magnitude of the displacement. A comparison of both measurement methods is there­
fore only possible to a limited extent. Additionally, the data from the radar sensor was
recorded with a higher time resolution. This higher measuring frequency enables short-
term movements, such as vibrations, of the headframe to be better recorded.

4. Summary
In summary, the suitability of an FMCW radar sensor for high-precision geomonitoring
could be verified as an overall objective in this contribution. With the reviewed FMCW
radar sensor, 1D distance measurements could be realized on a laboratory scale and in
a real case study of the headframe mointoring with resolutions in the sub-mm range. The
precision (3sx ) of measurements with a corner reflector as a point target was determined
to be less than 10 μm, whereby the evaluation via phase differences has proven to be
more sensitive and precise. The associated limits of detection for changes in distance were
in the range of 2 μm. Under real conditions, 1D monitoring was used to detect move­
ments of a headframe of around 0,2 mm over the course of a day with a statistical
significance level of 2sx (α ¼ β ¼ 0; 05). Due to the increasing radar footprint with
distance, a spatial averaging of the reflected signal is to be expected, unlike in the case
laboratory measurements of defined targets without a background. This is also a possible
explanation of the smaller measured changes (1 mm) in comparison to the tachymetric
survey of (Martienßen (2017)), which was carried out with fixed laser targets.
In conclusion, the method can be used to investigate further case studies in which
reflectors stand out more clearly in the footprint and the movement rates are in the range of
millimetres per day. These include dams or underground subsidence due to mining.

Acknowledgements
The investigation has been performed as part of the BMBF-funded project MURadIn, reference
number 033RK066B. The used FMCW radar sensor was provided by the company indurad GmbH
(indurad GmbH (2021)).

Data availability statement

Raw data were generated at TU Bergakademie Freiberg. Derived data supporting the findings of this
study are available from the corresponding author C. R. on request.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Funding
This work was supported by the Federal Ministry of Education and Research - BMBF [033RK066B].
 INTERNATIONAL JOURNAL OF REMOTE SENSING 9227

References
Blackman, R. B., and J. W. Tukey. 1959. “Particular Pairs of Windows.” The Measurement of Power
Spectra, from the Point of View of Communications Engineering 37, Dover, New York: 98–99.
Braun, J. 2015. “Testing of the Automatic Targeting of Total Station Trimble S8 on Reflective
Targets.” SGEM 20 (2): 491.
Bürgmann, R., P. A. Rosen, and E. J. Fielding. 2000. “Synthetic Aperture Radar Interferometry to
Measure Earth’s Surface Topography and Its Deformation.” Annual Review of Earth and Planetary
Sciences 28 (1): 169–209. doi:10.1146/annurev.earth.28.1.169.
Butt, J., Z. Gojcic, L. Schmid, and A. Wieser. “Terrestrische Radarinterferometrie und terrestrisches
Laserscanning zur überwachung alpinen Geländes: Praxiserfahrungen und Theorie.” 2020.
Caduff, R., F. Schlunegger, A. Kos, and A. Wiesmann. 2015. “A Review of Terrestrial Radar
Interferometry for Measuring Surface Change in the Geosciences.” Earth Surface Processes and
Landforms 40 (2): 208–228. doi:10.1002/esp.3656.
Dahlen, F. A. 1982. “The Effect of Data Windows on the Estimation of Free Oscillation Parameters.”
Geophysical Journal International 690 (2): 537–549. doi:10.1111/j.1365-246X.1982.tb04964.x.
De Gruijter, J., D. J. Brus, M. F. P. Bierkens, and M. Knotters. 2006. Sampling for Natural Resource
Monitoring 28. Berlin: Springer Science & Business Media.
Garthwaite, M. C. 2017. “On the Design of Radar Corner Reflectors for Deformation Monitoring in
Multi-frequency Insar.” Remote Sensing 9 (7):648. doi:10.3390/rs9070648. June.
Garthwaite, M. C., S. Nancarrow, A. Hislop, M. Thankappan, J. H. Dawson, and S. Lawrie. 2015.
“Design of Radar Corner Reflectors for the Australian Geophysical Observing System.”
Geoscience Australia 3: 490.
indurad GmbH. https://www.indurad.com/technology/sensors/isdr/ 09.April.2021.
The MathWorks, Inc. Constant False Alarm Rate (CFAR) Detection. https://www.mathworks.com/
help/phased/examples/constant-false-alarm-rate-cfar-detection.html 27.July.2020.
Joughin, I., S. Tulaczyk, M. Fahnestock, and R. Kwok. 1996. “A Mini-surge on the Ryder Glacier,
Greenland, Observed by Satellite Radar Interferometry.” Science 274 (5285): 228–230. doi:10.1126/
science.274.5285.228.
Läufer, G., M. Lehmann, and S. Rödelsperger. 2017. “Terrestrische Mikrowelleninterferometrie.” In
Ingenieurgeodäsie, 213–233. Springer Spektrum, Berlin, Heidelberg.
Lehmann, M. 2015. “Schwingungen von Ingenieurbauwerken–Bestimmung mittels terrestrischer
Radarinterferometrie.” Conference: GeoMonitoring 171–187.
Martienßen, T. 2017. “Ein Beitrag zum Monitoring von Schachtanlagen im Raum Freiberg.” BHM
Berg-und Hüttenmännische Monatshefte 162 (10): 434–439. doi:10.1007/s00501-017-0633-4.
Moreira, A., P. Prats-Iraola, M. Younis, G. Krieger, I. Hajnsek, and K. P. Papathanassiou. 2013.
“A Tutorial on Synthetic Aperture Radar.” IEEE Geoscience and Remote Sensing Magazine 10 (1):
6–43. doi:10.1109/MGRS.2013.2248301.
Norm, D. I. N. 18723-7:1990-07,Feldverfahren zur Genauigkeitsuntersuchung geodätischer instru­
mente; Vermessungskreisel, 1990.
Norm, D. I. N. 18709-4:2010-09-00,Begriffe, Kurzzeichen und Formelzeichen in der Geodäsie - Teil 4:
Ausgleichsrechnung und Statistik, 2010.
Pieraccini, M., and L. Miccinesi. 2019. “Ground-based Radar Interferometry: A Bibliographic Review.”
Remote Sensing 11 (9): 1029. doi:10.3390/rs11091029.
Pruscha, H. 2006. Statistisches Methodenbuch: Verfahren, Fallstudien, Programmcodes. Springer-
Verlag, Berlin.
Richards, M. A., J. Scheer, W. A. Holm, and W. L. Melvin. 2010. Principles of Modern Radar. SciTech
Publisher, Raleigh, NC.
Rohling, H. 1983. “Radar Cfar Thresholding in Clutter and Multiple Target Situations.” IEEE
Transactions on Aerospace and Electronic Systems AES-19 (4): 608–621. doi:10.1109/
TAES.1983.309350.
Sarabandi, K., and T.-C. Chiu. 1996. “Optimum Corner Reflectors for Calibration of Imaging Radars.”
IEEE Transactions on Antennas and Propagation 44 (10): 1348–1361. doi:10.1109/8.537329.
9228 C. RADACH ET AL.

Scherr, S. 2017. FMCW-Radarsignalverarbeitung zur Entfernungsmessung mit hoher Genauigkeit.

Volume 83. KIT Scientific Publishing, Karlsruhe.
Scherr, S., R. Afroz, S. Ayhan, S. Thomas, T. Jaeschke, S. Marahrens, A. Bhutani, M. Pauli, N. Pohl, and
T. Zwick. 2017. “Influence of Radar Targets on the Accuracy of Fmcw Radar Distance
Measurements.” IEEE Transactions on Microwave Theory and Techniques 65 (10): 3640–3647.
doi:10.1109/TMTT.2017.2741961.
Smith, J. O. Spectral Audio Signal Processing. http://ccrma.stanford.edu/jos/sasp///ccrma.stanford.
edu/jos/sasp/, 2011. online book, 2011 edition.
Tapete, D., N. Casagli, G. Luzi, R. Fanti, G. Gigli, and D. Leva. 2013. “Integrating Radar and Laser-based
Remote Sensing Techniques for Monitoring Structural Deformation of Archaeological
Monuments.” Journal of Archaeological Science 400 (1): 176–189. doi:10.1016/j.jas.2012.07.024.
 INTERNATIONAL JOURNAL OF REMOTE SENSING 9229

Appendices
Appendix A. EXP A: Precision

Table A1. Results 3sx , s�x and ΔLOD of the indoor and outdoor precision measurements of EXP A for
n ¼ 100 single measurements. The estimated errors, calculated by PI and phase error Δϕ of the static
multiple measurements are shown for target distances from 2 m to 4,5 m and the different sample
materials. All calculated numerical values are given in μm.
r Al Rock CR-II on steal House wall,
(m) Evaluation Parameter CR-I CR-II plate sample Drywall surface plastered
2,0 PI 3sx 5,716 5,484 8,373 61,375 88,184 62,394 138,839
s�x 0,198 0,186 0,301 2,026 2,882 2,080 4,628
ΔLOD 1,007 0,946 1,534 10,327 14,695 10,603 23,594
2,0 Δϕ 3sx 0,332 0,490 0,346 1,045 2,431 15,771 3,944
s�x 0,011 0,017 0,012 0,034 0,079 0,526 0,131
ΔLOD 0,059 0,085 0,063 0,176 0,405 2,680 0,670
2,5 PI 3sx 5,979 5,499 7,801 50,421 55,411 41,319 311,006
s�x 0,224 0,204 0,289 1,664 1,820 1,370 10,367
ΔLOD 1,143 1,038 1,473 8,484 9,278 6,987 52,851
2,5 Δϕ 3sx 0,378 0,276 0,557 1,184 1,771 6,560 12,814
s�x 0,014 0,010 0,021 0,039 0,058 0,218 0,427
ΔLOD 0,072 0,052 0,105 0,199 0,297 1,109 2,178
3,0 PI 3sx 4,918 5,059 7,779 272,741 56,100 32,616 184,824
s�x 0,180 0,183 0,288 9,002 1,843 1,087 6,130
ΔLOD 0,917 0,932 1,469 45,892 9,394 5,543 31,252
3,0 Δϕ 3sx 0,264 0,173 0,894 4,872 2,017 18,623 4,940
s�x 0,010 0,006 0,033 0,161 0,066 0,621 0,164
ΔLOD 0,049 0,032 0,169 0,820 0,338 3,165 0,835
3,5 PI 3sx 7,338 10,355 4,689 352,852 108,144 13,327 226,536
s�x 0,275 0,388 0,174 11,589 3,501 0,444 7,514
ΔLOD 1,403 1,980 0,885 59,083 17,850 2,265 38,306
3,5 Δϕ 3sx 0,539 0,299 0,474 6,454 2,931 2,667 5,884
s�x 0,020 0,011 0,018 0,212 0,095 0,089 0,195
ΔLOD 0,103 0,057 0,090 1,081 0,484 0,453 0,995
4,0 PI 3sx 14,316 10,818 18,042 65,264 191,137 57,739 656,023
s�x 0,530 0,384 0,672 2,154 6,371 1,925 21,867
ΔLOD 2,703 1,960 3,428 10,981 32,481 9,812 111,482
4,0 Δϕ 3sx 1,046 0,834 1,228 19,276 3,309 21,875 13,952
s�x 0,039 0,030 0,046 0,636 0,110 0,729 0,465
ΔLOD 0,198 0,151 0,233 3,243 0,562 3,717 2,371
4,5 PI 3sx 11,265 12,474 7,338 33,897 199,597 21,064 247,233
s�x 0,422 0,462 0,246 1,130 6,588 0,702 8,241
ΔLOD 2,154 2,355 1,253 5,760 33,585 3,580 42,014
4,5 Δϕ 3sx 1,152 0,691 0,351 0,914 4,609 10,209 9,125
s�x 0,043 0,026 0,012 0,030 0,152 0,340 0,304
ΔLOD 0,220 0,130 0,060 0,155 0,775 1,735 1,551