See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/251964823

Evaluation of a high accuracy range detection algorithm for FMCW/Phase

radar systems

Conference Paper · January 2010

CITATIONS READS

10 1,163

4 authors, including:

Thorsten Kayser M. Pauli

Karlsruhe Institute of Technology Karlsruhe Institute of Technology
36 PUBLICATIONS   450 CITATIONS    84 PUBLICATIONS   1,006 CITATIONS   

SEE PROFILE SEE PROFILE

Thomas Zwick
Karlsruhe Institute of Technology
684 PUBLICATIONS   12,098 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

TR 32054 View project

3D Printed 60-GHz Horn Antennas and Arrays for Next-Generation 5G Applications View project

All content following this page was uploaded by Thomas Zwick on 13 June 2017.

The user has requested enhancement of the downloaded file.

 Proceedings of the 7th European Radar Conference

Evaluation of a High Accuracy Range Detection

Algorithm for FMCW/Phase Radar Systems
Philipp Pahl #1 , Thorsten Kayser #∗2 , Mario Pauli #∗3 , Thomas Zwick #4
#
Karlsruhe Institute of Technology (KIT)
Kaiserstraße 12, D-76131 Karlsruhe, Germany
3
mario.pauli@kit.edu

∗
PKTEC GmbH
Bahnhofstraße 30, D-77746 Schutterwald, Germany

Abstract— The FMCW radar is the most versatile radar measurement line
principle used today. Depending on the system configuration, it
is possible to use an FMCW radar to detect targets in the range
from hundreds of kilometers down to a few centimeters. This directional
paper describes an algorithm, which can be applied to improve coupler
the FMCW range accuracy down to a few mm. Numerical
system simulations are used to evaluate the possibility of using
a combination between an FMCW and a phase radar for a line FM modulator
based range detection at 24 GHz.

I. I NTRODUCTION mixer
Depending on the application different kinds of radar tech- Fig. 1. Blockdiagram of an FMCW radar
niques are used. If the radar has to have an unambiguous
range of a few meters but also a range resolution of less
than one centimeter most of the known radar techniques drop Using a saw tooth modulation with a bandwidth B and a
out. One radar technique, which meets the requirements is the cycle duration T a target at a distance R generates the beat
FMCW radar. Combined with an accurate signal processing frequency
an accuracy of less than 5 mm can be achieved by using only fIF (t, R) = |fTx (t) − fRx (t, R)|
a smart FMCW configuration. The FMCW radar separates B 2R
targets at different distances by transmitting a frequency = · (1)
T c
modulated signal. Depending on the bandwidth and sweep
at the output of the down-converting mixer, where c represents
time of the used modulation signal, an acceptable range of
the propagation velocity on the measurement line. To obtain
unambiguity together with an acceptable accuracy can be
the frequency resolution ΔfIF , which is proportional to the
achieved. If this radar technique is combined with a phase
range resolution ΔR, (1) can be rewritten using ΔfIF and
evaluation technique, the FMCW radar can be used to detect
ΔR to
targets over a distance of a few meters with an accuracy of a
few μm [1], [2], [3]. ΔfIF = |fIF (R2 ) − fIF (R1 )|
In this paper different parameters of an FMCW radar will 2·B
= |R2 − R1 |
be considered in terms of range resolution and unambiguous c·T
range. A FMCW radar system model is introduced to inves- 2·B
= ΔR . (2)
tigate the influence of the used elements (depicted in Fig. 1) c·T
on the range detection. Using the derived equations and the Considering the case that the roundtrip time Δt  T , the IF
system model an improved range detection algorithm with high frequency stays constant during a single sweep. This means
accuracy is described and evaluated. that a single target in front of the radar, induces a single
beat frequency. In time domain the received beat signal is a
II. R ESOLUTION OF AN FMCW R ADAR sinusoidal signal, which transformed to the frequency domain
The FMCW radar transmits a linear frequency chirp to is represented by a Dirac impulse. However, the maximum
determine the distance of targets. Within this paper only measurement period, equivalent to the cycle time of the radar
stationary or slow moving targets are considered. Therefore signal, has to be taken into account. The limited measurement
a saw tooth modulation scheme is applied during the system time distorts the beat signal in frequency domain by convo-
investigation instead of a triangular modulation. luting the Dirac impulse with a sin(x)/x function.

978-2-87487-019-4 © 2010 EuMA 160 30 September - 1 October 2010, Paris, France

 The frequency resolution only depends on the measurement
time during a single cycle

1 1
ΔfIF = ≈ . (3)
Tmess T

beat signal
The range resolution only depends on the bandwidth B and is
independent of the sweep time T . Combining (2) and (3) the
range resolution can be expressed as

c
ΔR = . (4)
2·B

For instance, using (4) a bandwidth of B = 1 GHz corre-

sponds to a range resolution of ΔR = 15 cm in case of free
space propagation. This maximum range resolution can be
further improved by using an appropriate signal processing t in μs
algorithm, which is discussed in detail in section IV. For a
Fig. 2. Beat signal at the mixer output in time domain
high frequency resolution it is important that the measurement
time is much longer than the signal propagation delay. Hence
the following condition has to be redeemed
can be evaluated by frequency (6) and phase (7)
4 B 2·R
T  · Rmax . (5) fB (τ ) = (6)
c T c 
1 2·R
ϕB (τ ) = 2π fmin − fB (τ )
III. S YSTEM S IMULATION M ODEL 2 c
2·R
≈ 2πfmin . (7)
The given simulation model is based on the block diagram c
shown in Fig. 1 and valid for any kind of line based radar Combining both, a frequency and a phase analysis, a high
measurement. The VCO is modeled as an ideal up-chirp from range accuracy as well as a large unambiguous range can
23.5 to 24.5 GHz. The generated time domain vector is used to be achieved. To be able to use the phase investigation the
evaluate the passive elements between the VCO and the ADC frequency based range determination has to come up with an
at the output of the low-pass filter. The coupler is modeled accuracy better than λ/2 of the highest used VCO frequency.
as an ideal 3-dB branch-line coupler. For the down-converting If this coarse detection can be achieved, the additional phase
mixer an ideal mixer, generating only the desired baseband and determination provides all the necessary information for an
its image signal is assumed. The transmission from the coupler accuracy of a few μm. Fig. 4 depicts the system simulation
to the line is modeled with an input matching of −15 dB. A
standard second order Bessel filter with a cut-off frequency of 0
1 GHz is implemented as low-pass filter behind the mixer’s
IF output. -10
Fig. 2 depicts the simulated beat signal in time domain for
the following configuration: B = 1 GHz, fcenter = 24 GHz, -20

single target at a distance of R = 2 m and a reflection

-30
coefficient of rtarget = −3 dB. It can be seen that there is
beat signal in dBm

no simple relation between the distance and the simulated -40

beat signal. This is caused by the superposition of the beat
signal and additional signals at much lower frequencies due -50
to unwanted reflections, e.g. the non-ideal matching of the
line, which mixes down to a very low frequency. The very -60

high frequency at the beginning is due to the saw tooth FM

-70
modulation. At the beginning of an up-chirp the reflected RF
signal oscillates at about 24.5 GHz and is mixed with the -80
current frequency of the chirp generator (VCO) at 23.5 GHz 60 70 80 90 100 110 120 130 140
resulting in a beat frequency of approximately 1 GHz. Fig. 3 frequency in MHz

shows the Fourier transformed of the beat signal over one cycle
time T . The beat signal as it is shown in Fig. 2 and Fig. 3 Fig. 3. Beat signal at the mixer output in frequency domain (baseband)

161
 results for the phase evaluation by (7). In this case a distance the DFT can be calculated to
sweep of 1 mm is shown. Despite the influence of the simu- N
 −1
2π
lated system, the form of the phase is mostly linear over the Yn = yk e−jkn N (10)
distance variation, which means that a high range accuracy of k=0
a few μm is possible. fB
1 jϕB 1 − ej2π Δf
= e “
fB
” n = 0, . . . , N − 1
2 j 2π Δf −n
⎧ 1 1 − e N

jϕB
⎨ 2e N
⎪ for n = m, q = 0
= .
⎪
⎩ 1 jϕB 1 π
ejπq(1− N ) ej N (n−m) sin πsin πq
2e [ N (m−n+q)]
detected phase in °

It is convenient to divide the beat frequency in a multiple m

position in mm
Fig. 4. Phase over varying target distance between 500 mm and 501 mm
IV. H IGH R ESOLUTION A LGORITHM
The range resolution by frequency evaluation depends on
the available bandwidth and is therefore limited to a certain
value. However, to employ the phase evuluation the accuracy
of the frequency evaluation has to be better than λ/2. Thus,
to combine frequency and phase evaluation, a smart algorithm
has to be used to improve the resolution of the frequency
analysis in a way that its accuracy becomes better than λ/2.
Afterwards an accurate fine detection by a phase detection
algorithm can be performed, which however is not subject of
of the sampling frequency Δf and the residual part q, as it is
done in (10). In the trivial case the beat frequency is a multiple
of Δf , which means that there is only one corresponding peak
in the frequency domain. Hence the beat frequency detected
by the discretized spectrum is correct.
Usually the beat frequency is not a multiple of Δf , which
means that there is no single maximum. In this case m and
q have to be determined as good as possible. The maximum
of |Yn | still is the discrete frequency closest to the actual beat
frequency. Therefore fB can be expressed as follows
1 1
fB = mΔf +qΔf with |Ym | = max |Yn | q ∈ − , +
n 2 2
(11)
and m can be always determined. To determine q, the ampli-
tudes of the neighboring discretized frequencies (|Ym−1 | and
|Ym+1 |) are investigated. The sign of q can be defined, since
the real beat frequency is located between the two discretized
frequencies with the highest amplitudes. For the numerical
evaluation of q the ratio
Ym − Ym±1 Ym±1
A±1 := =1− (12)
Ym Ym
can be used. With (10) and (12) the absolute value of this
equation can be rewritten to
sin π
|A±1 | =   π N
 , (13)
this paper. The frequency resolution has to be improved by at sin (1 − q) 
N
least fmin /B for a maximum measurement time of T . In a
which can be further simplified if N  π to
realistic system the measurement time Tm is only a part of T .  
Therefore the following equation can be used to calculate the 1  Ym±1 
|A±1 | = 
= 1− (14)
required improvement 1±q  Ym 
ΔfB,requ B T and then be solved for q to
= · . (8)
Δf1/T fmin Tm |A+1 | − 1 1 − |A−1 |
q= and q= . (15)
|A+1 | |A−1 |
A discrete Fourier transformed (DFT) is used to calculate
the discrete spectrum of the beat signal. The sample frequency With the determined values of q and m, (11) is used to de-
within the DFT depends on the measurement time and the termine the real beat frequency fB in between the discretized
number of points of the DFT. Using the complex representa- frequency spectrum.
tion of the beat signal Though the algorithm is quite simple, there is one problem
 which should be considered if a real signal is sampled. Real
1 jϕB j2πfB t  signals are characterized by the fact that the absolute value
yk : = e e  (9)
2 t=kΔt
of their Fourier transformed is always symmetrical around
1 jϕB j2πfB kΔt the f = 0 axis. Therefore for a low beat frequency which
= e e k = 0, . . . , N − 1 is not a multiple of Δf the sin(x)/x function superposed
2

162
 frequency of approximately 1 GHz at the beginning of each
sweep evaluation, which is due to the signal propagation delay.
Time gating can be used to overcome this problem, but usually
in a real FMCW radar the ratio between TS and TD is big
enough to neglect this error. Despite the small sweep time of
Mirror / Sidesignal in dB

TS = 100 · TD the green curve already shows a range accuracy

of ΔR ≤ 5 mm. Hence, using the proposed range estimation
algorithm the range accuracy can be drastically improved by a
factor of approximately 30. The predicted range resolution is
better than 5 mm which is less than λ/2 of a 24 GHz signal.
Hence, an additional phase detection algorithm can be applied.

TS = 10 TD
TS = 100 TD

detected position in m (~beat frequency)

frequency deviation in Δf

Fig. 5. Ratio between the image signal and the desired beat signal

with the negative beat frequency −fB influences the detection

of the beat frequency and can cause a wrong detection of
the beat frequency in the worst case. The effect gets worse
for lower beat frequencies since the positive and the negative
image frequencies come closer. In Fig. 5 the ratio between the
absolute value of the beat frequency Y±1 and the amplitude
of its negative image at the frequency of Y±1 is plotted for
two different beat frequencies and all possible values of q. As position in m
mentioned before the ratio gets worse for low beat frequencies.
If the beat frequency is close to a frequency step n · Δf and Fig. 6. Position detection according to the detected beat frequency
therefore q is close to zero or one, the ratio is worse either. It
becomes best for q = ±0.5 and high beat frequencies.
V. S UMMARY & C ONCLUSIONS
The described position estimation algorithm is now applied
on the FMCW radar system simulator introduced in the Within this paper a high range resolution algorithm was
previous section. Besides the target position all parameters presented and evaluated by numerical simulations. The results
like the center frequency (f = 24 GHz) and the bandwidth predict that it is possible to build an FMCW radar for line
(B = 1 GHz) stay the same as defined in section III. For the based radar applications which uses a combination of phase
simulation a single target is moved from 1.5 m to 1.8 m and and frequency detection to achieve an accuracy of a few μm
the beat frequency is analyzed at each distance. Note that the by a phase detection algorithm and an unambiguous range
simulation assumes that the target is stationary and therefore of several meters by a frequency detection algorithm. Further
the received signal does not suffer from any Doppler shift. In investigations focusing on the phase resolution, the necessary
Fig. 6 the results for two different cycle times are plotted SNR and the assembly of a prototype are currently in progress
containing effects like additional time delays or frequency and will be used to validate the simulation results.
dependent group delays within the different building blocks. R EFERENCES
The blue line represents a cycle time TS of 10 times the
[1] Q. GuoQing, “Digital signal processing in fmcw radar marine tank
signal propagation delay (TD = 16.5 ns), the green line uses gauging system,” vol. 1, pp. 7–10 vol.1, Oct 1996.
a TS = 100 · TD and the red line is the real position of the [2] A. Stelzer, C. Diskus, K. Lubke, and H. Thim, “A microwave position
target. The offset between the actual position of the target and sensor with submillimeter accuracy,” Microwave Theory and Techniques,
IEEE Transactions on, vol. 47, no. 12, pp. 2621–2624, Dec 1999.
the range estimation is due to the additional time delay in the [3] G. Woods, D. Maskell, and M. Mahoney, “A high accuracy microwave
different radar building blocks of the simulated system. For a ranging system for industrial applications,” Instrumentation and Measure-
proper estimation of the position this has to be removed by ment, IEEE Transactions on, vol. 42, no. 4, pp. 812–816, Aug 1993.
the algorithm. The large steps of the blue line predict a rather
large uncertainty of the distance evaluation. For the green
line the steps and therefore the range uncertainty becomes
smaller. Obviously the steps in range depend on the cycle
time TS . These steps appear because of the very high beat

163
View publication stats