sensors

Article
On the Testing of Advanced Automotive Radar
Sensors by Means of Target Simulators
Premysl Hudec and Viktor Adler *
Department of Electromagnetic Field, Faculty of Electrical Engineering, Czech Technical University in Prague,
Technicka 2, 166 27 Prague 6, Czech Republic; hudecp@fel.cvut.cz
* Correspondence: adlervik@fel.cvut.cz; Tel.: +420-22-435-2275




Received: 6 March 2020; Accepted: 6 May 2020; Published: 9 May 2020


Abstract: The rapid development and wide commercial implementation of automotive radar
sensors are strengthening the already considerable interest in matching radar target simulators.
Such simulators boast promising results when used for both essential functional inspections of
active sensors and the high-speed testing of numerous traffic scenarios while examining complex
reactions of automobile electronic systems. For these purposes, advanced versions of target simulators
enabling a generation of multiple targets moving at different velocities and ranges are required.
The design, practical implementation and system programming of advanced sensor simulator setups
require a detailed analytical description concerning all important technical aspects. An abundance of
detailed information on the behavior and parameters of automotive radar sensors can be found in the
references, but similar knowledge on sensor simulator setups is lacking. This article presents detailed
analyses of the all-important RF parameters, where special attention is paid to phase noise, and its
analytical description takes into account an even greater number of simulated targets. The derived
analytical formulas enable both an optimal setup implementation and system programming of a wide
range of practical testing procedures.

Keywords: automotive radar sensor; radar target simulator; sensor testing; sensor simulator setup;
additive RF noise; phase noise

1. Introduction
Automotive radars (ARs) can be rated as sensors with the most intensive development and
expanding applications. At a minimum, three AR units, used as adaptive cruise control or blind
spot sensors, are installed in almost every new vehicle, and in the near future, numerous sensors
will be used to ensure 360-degree coverage [1]. The main task of all these sensors is to monitor a
vehicle’s surroundings [2], i.e., utilize perception algorithms to detect and evaluate targets, such as
neighboring vehicles, pedestrians, traffic signs and guardrails, among others. Based on this information,
the electronic control unit (ECU) adapts the vehicle settings and initiates a suitable response, all of
which should result in substantially improved road traffic safety [2]. In addition, due to their immunity
to rain, dust or darkness, AR sensors will have an irreplaceable role in developed autonomous driving
concepts. An overview of nowadays utilized modulation schemes, target classification and detection
algorithms’ and inter-radars’ immunity can be found in [3]. As all of these electronics relate to the
safety of people, all concerned sensors must be subject to intensive testing. For these purposes, target
simulators are of great benefit.
A radar target simulator (TS) is a “box” placed in front of a tested radar that is capable of
“convincing” a radar that it is detecting a target or multiple targets in required azimuths and ranges,
and of showing the prerequisite velocities and appropriate radar cross-sections (RCSs). Such target
simulators are used to check radar parameters or verify their calibration. In the automotive industry,

Sensors 2020, 20, 2714; doi:10.3390/s20092714 www.mdpi.com/journal/sensors

Sensors 2020, 20, 2714 2 of 22

radar target simulators can play an even more significant role. In practice, automotive electronic
systems equipped with AR sensors must solve a virtually infinite number of traffic scenarios, resulting
in extremely difficult, lengthy and costly test drives. This process becomes additionally cumbersome
following any change to a car’s hardware (HW) or controlling software (SW) when all traffic scenarios
must be tested again. TSs are also essential components in hardware-in-the-loop (HIL) testing and
overall reliability evaluations.
Common AR sensors operate in the 76–77 GHz millimeter-wave frequency band,
use frequency-modulated continuous-wave (FMCW) techniques, and measure ranges (R), radial
velocities (vr ) and azimuths (Ψ) of targets in their view [4,5]. Detailed information on these sensors
including realization in a single-chip technology can be found in [6]. In recent years, a significant
effort has been made to develop advanced automotive sensors. Beyond ensuring higher levels of
immunity against mutual interferences by using more complex modulations, the development is,
above all, focused on reaching a substantially higher accuracy and resolution [7,8]. Both goals can be
met using a wider modulation frequency band in the 77–81 GHz region, and more transmitting (TX)
and receiving (RX) channels. As a result, modern AR sensors detect significantly more targets than
common types. For example, as described in [2], common narrow-bandwidth, low-resolution sensors
detect a nearby vehicle as a single target. However, the surface of any vehicle consists of areas with
low reflectivity and areas with high reflectivity. That is why advanced, wide-band, high-resolution
automotive sensors “see” any neighboring car in its view as a set of many targets, each corresponding
to one high-reflecting area [7,9]. This is true for all other type of targets.
There exist several wider families of TSs utilizing different principles and providing more of less
advanced capabilities. Reference [10] describes a TS based on a reflector mechanically moved alongside
a dielectric waveguide. In [11,12], a very simple TS structure without conversion is presented but
its application is limited to FMCW radars only. The special TS used for the simulation of moving
pedestrians and presented in [13] uses advanced down- and up-conversion circuits but enables only a
Doppler frequency modulation. Solutions according to [14,15] employ switched optical fiber delay
lines. In [16], a TS utilizing purely digital processing of digitalized radar signals is described, whereas
a solution according to [17] depicts a system equipped with both digital and analogue processing
branches. The former branch is supposed to ensure the simulation of far targets, while the latter is
able to simulate nearby targets. Reference [18] describes a flexible TS solution based on commercial
off-the-shelf measurement equipment.
According to Figure 1, a common TS typically consists of one ANT1 receiving antenna, target
simulator circuits (TSC), i.e., all circuits between the TS receiving and transmitting antennas, and one
ANT2 transmitting antenna. The receiving and transmitting antennas are situated in front of the tested
sensor with a suitable distance for RS , ranging from 0.5 to 1 m. In TSCs, the signal from the tested sensor
is received by the receiving antenna, delayed in time, shifted in frequency and amplified/attenuated.
The time delay corresponds to the required range of the simulated target, and the frequency shift
simulates the Doppler frequency shift resulting from the target´s radial velocity, while the amplitude
of the transmitted signal corresponds to the target´s RCS and range. The processed signal is then sent
back to the sensor. Such a sensor simulator setup can simulate a single target (or several targets at
different ranges) at a single azimuth defined by the sensor’s TS transmitting antenna’s interconnecting
line. Simulated target motions can occur at this line only. However, to simulate real traffic scenarios,
it is necessary to generate a higher number of targets at more azimuths while the velocity vectors can
also evince the azimuthal components (components in directions perpendicular to the sensor’s TX
antenna’s interconnecting lines).
As one approach to these requirements, the setup described in [19] employs a single TS receiving
antenna fixed in front of the tested AR and five TS transmitting antennas which can be mechanically
moved alongside sections of circles in view of the tested AR sensor. This structure makes it possible
to simulate up to five independent targets at five different azimuths with independent radial and
alongside a dielectric waveguide. In [11,12], a very simple TS structure without conversion is
presented but its application is limited to FMCW radars only. The special TS used for the simulation
of moving pedestrians and presented in [13] uses advanced down- and up-conversion circuits but
enables only a Doppler frequency modulation. Solutions according to [14,15] employ switched
optical2020,
Sensors fiber
20,delay
2714 lines. In [16], a TS utilizing purely digital processing of digitalized radar signals is
3 of 22
described, whereas a solution according to [17] depicts a system equipped with both digital and
analogue processing branches. The former branch is supposed to ensure the simulation of far targets,
azimuthal
while the velocity
latter is components. This nearby
able to simulate represents an advanced
targets. Reference solution, but it is still
[18] describes far from
a flexible TSthe “raw”
solution
radar pictures “seen” by advanced AR sensors in
based on commercial off-the-shelf measurement equipment.real traffic.
Sensors 2019, 19, x FOR PEER REVIEW 3 of 21

transmitting antenna’s interconnecting PT

Gline.
RT Simulated
GSR target motions can occur at this line only.
PSR NaS
However, to simulateTX real traffic scenarios, itFSLisS necessary to generate τ a higher number of targets at
kT0B
more azimuths while the velocity vectors canL1also evince
ANT1 the azimuthal
delay components (components in
directions perpendicular to the sensor’s A TX antenna’s interconnecting
GS, FS lines).
Doppler fD
As one approach to these kTrequirements, 0 B G RR
the
G ST
setup described
amplitude in [19] employs a single TS
PR
receiving antenna fixed RX
in front Nof
aR the tested AR and five TS transmitting antennas which can be
FSLS
SNRRS TSC
mechanically moved alongside sections of circles in Pview ST of the tested AR sensor. This structure
L2 ANT2
makes it possible to simulateA' up to five independent targets at five different azimuths with
RS
independent radialRadar Sensor
and azimuthal velocity components. Target Simulator
This represents an advanced solution, but it
is still far from the “raw” radar pictures “seen” by advanced AR sensors in real traffic.
Figure 1. Block diagram with basic automotive radar (AR) sensor–target simulator (TS) (TS) relations.
relations.
References [20,21] propose an interesting solution of how to generate really complex radar
pictures. This solution ispropose
References based on an an array of solution
TS transmitting antennas, which can be seen in Figure 2,
According [20,21] to Figure 1, a common interesting
TS typically consists of how to ofgenerate
one ANT1 really complex
receiving radar
antenna, pictures.
target
where
This each TS
solution transmitting
is based on i.e., antenna
an array of TSisbetween
connected to
transmitting a TXMi which
antennas, independent
can be seen signal in processing
Figure 2,and and
where
simulator circuits (TSC), all circuits the TS receiving and transmitting antennas, one
transmitting
each TS module.
transmitting The
antenna high is number
connected of installed
to a TXM array
i
elements
independent enables
signal a high
processing number
and of targets
transmitting
ANT2 transmitting antenna. The receiving and transmitting antennas are situated in front of the
at arbitrary
module. Theranges
high and azimuths to bearray
generated in parallel. Switching targets by ΔΨatfrom one
tested sensor with number
a suitable of distance
installed for RS,elements
ranging from enables 0.5 atohigh1 m.number
In TSCs, of the
targets
signal arbitrary
from the
azimuth
ranges to another
and azimuths simulates
to be motions with
generated in the azimuthal
parallel. Switchingvelocity components.
targets by ∆Ψ Despite
from one having
azimuth been
to
tested sensor is received by the receiving antenna, delayed in time, shifted in frequency and
published
another in 1987 as
simulates motions a tool for testing
withdelay missile
the azimuthal radars,
velocity up to now,
components. this initial
Despite design and the realization
amplified/attenuated. The time corresponds to the required range of thehaving beentarget,
simulated published
and
of
in this type
1987 as a of
tool testing
for setupmissile
testing have not yet matured.
radars, up to now, Tothis
accelerate
initial its development,
design and the it is, above
realization all,
of this
the frequency shift simulates the Doppler frequency shift resulting from the target´s radial velocity,
necessary
type to derive an acceptably simple and economically acceptable HW structure. This primarily
whileofthe testing setup have
amplitude not transmitted
of the yet matured. signal To accelerate
corresponds its development,
to the target´ it is,s above
RCS and all, necessary
range. The to
concerns
derive an the most demanding
acceptably simple and and costly millimeter-wave
economically acceptable HW and digital
structure. This circuits.
primarily The following
concerns the
processed signal is then sent back to the sensor. Such a sensor simulator setup can simulate a single
sections
most indicate thatcostly the required simplification can be derived Thefrom a detailed analysis of the
targetdemanding
(or severaland targets at millimeter-wave
different ranges) and at digital circuits.
a single azimuth following
defined sections
by the indicate
sensor’sthat TS
sensor simulator setup in question. The same analysis is also
the required simplification can be derived from a detailed analysis of the sensor simulator setup in beneficial for system programming,
that meansThe
question. specific
same settings
analysisof is all
alsosystem circuits
beneficial and parameters,
for system programming, whichthat canmeansenablespecific
the generation
settings of
of
the required
all system complex
circuits and radar pictures.
parameters, Considering
which can enable ARthe sensors
generation alone,oftheir structures,
the required parameters,
complex radar
algorithms, applications,
pictures. Considering AR etc.,
sensors can alone,
be found theirinstructures,
a great number parameters, of references,
algorithms, some of which etc.,
applications, are
mentioned
can be found within this text.
in a great numberThisofalso concernssome
references, a detailed
of which analysis of additive
are mentioned RF noise
within this and
text.the phase
This also
noise of FMCW sensors which can be found even in
concerns a detailed analysis of additive RF noise and the phase noise of FMCW sensors which can be somewhat dated references [22,23].
Unfortunately,
found even in somewhat a detaileddatedtheoretical
references analysis of the
[22,23]. AR sensor–target
Unfortunately, simulator
a detailed setup under
theoretical analysis concern
of the
is
ARlacking. This article
sensor–target aims to
simulator setupfill this
under gap.concern is lacking. This article aims to fill this gap.

array holder

TXM2
TXM1 RS
ΔΨ

TXMN

tested AR sensor
AR
Figure 2.
2. Testing
Testingsetup
setupbased
basedon on an array
an array of transmitting
of transmitting modules.
modules. The TSThe TS receiving
receiving antennaantenna is
is situated
anywhere
situated in view of
anywhere inthe ARofsensor.
view the AR sensor.

The article
The article presents
presents both
both derived
derived analytical
analytical formulas
formulas describing
describing all
all essential
essential system
system parameters
parameters
and recommendations on how they contribute to the optimization of the HW setup
and recommendations on how they contribute to the optimization of the HW setup structure and structure and to
to
system programming.
system programming. Section
Section 22 starts
starts with
with aa signal
signal level
level analysis,
analysis, while
while Section
Section 33 is
is devoted
devoted to
to an
an
additive RF noise analysis. As local oscillators (LOs) are inevitable parts of both advanced sensors
and simulators, Section 4 is dedicated to related phase noise problems. Section 5 studies the same
phenomenon in relation to the generation of multiple targets. Results of the calculated parameters
and performed computer simulations are summarized in Section 6. Section 7 evaluates the obtained
results with respect to the optimal HW structure and system programming.
Sensors 2020, 20, 2714 4 of 22

additive RF noise analysis. As local oscillators (LOs) are inevitable parts of both advanced sensors
and simulators, Section 4 is dedicated to related phase noise problems. Section 5 studies the same
phenomenon in relation to the generation of multiple targets. Results of the calculated parameters and
performed computer simulations are summarized in Section 6. Section 7 evaluates the obtained results
with respect to the optimal HW structure and system programming.

2. Signal Level Analysis

Figure 1 depicts fundamental AR sensor–TS relations. The AR transmitter transmits output power
(PT ) with its output antenna gain equal to GRT . The TS receiving antenna with gain (GSR ) and TS
transmitting antenna with gain (GST ) are situated at distances (RS ) from the AR sensor. The gain (GS )
depicts the overall TSC system gain, excluding the gains of the input and output antennas, while GRT
stands for gain of the sensor’s receiving antenna.
Using the above-defined variables, the AR input power (PR ) obtained through the TS can be
expressed as
G G G G G λ4
PR = PT RT SR S ST4 RR (1)
R4S (4π)
The power (PR ) can be compared with power which corresponds to the AR detecting a real target (PRt )
with the RCS (σ) situated at a range (Rt ), see Figure 3. This power can be evaluated using the standard
radar equation
G G σλ2
PRt = PT RT RR (2)
(4π)3 R4t
Comparing Equation (1) and (2), the RCS of the simulated target or, for design purposes, the overall
TSC system gain (GS ) can be derived as

4πR4S
GS = σ (3)
GSR GST λ2 R4t

As mentioned, GS corresponds to the gain of the overall TS circuits which provide a constant RCS of the
simulated target. This gain includes the gains of all millimeter-wave circuits, intermediate frequency
(IF) circuits, base-band (BB) circuits and, with care, the digital circuits used and the signal processing
applied. It is one of the most important parameters for TS design. When designing TS output circuitry,
the output TS power (PST ) can be useful:

PST = PSR GS (4)

where PSR is the power received by the AR. In specific programming steps described more closely
in Section 7, the solution of Equations (1) and (2), with respect to the RCS, can be necessary, and the
resulting formula can be expressed as

GS GSR GST λ2 R4t

σ= (5)
4πR4S

Using this formula, an effective RCS of the simulated target can be calculated as a function of the
currently set system parameters. Examples of some practically calculated values based on formulas
derived in this section are presented in Section 6.
 GSGSR GST  2 Rt4
 (5)
4 RS4

Using this formula, an effective RCS of the simulated target can be calculated as a function of
the currently
Sensors set
system parameters. Examples of some practically calculated values based
2020, 20, 2714 5 ofon
22
formulas derived in this section are presented in Section 6.

GRT
PT Target
TX
Rt
GRR
PRt
RX
NaR
kT0B
SNRRt
Radar Sensor

Figure 3.
Figure 3. Sensor-target-sensor
Sensor-target-sensor signal
signal path
path with
with signal
signal and
and noise
noise parameters.
parameters.

3. Additive RF Noise Analysis

As with any radio system, the operation of the TS is influenced by both additive RF noise and phase
noise. This section considers the former, but firstly, the signal-to-noise ratio (SNRRt ) corresponding to
a real target must also be evaluated, see Figure 3. Consequently, it must be determined how the TS
influences the AR sensor noise behavior. The SNRRt is related to the input of the AR sensor receiver
and can be expressed using the formula

PRt
SNRRt = (6)
kT0 B + NaR

In this equation, k = 1.38×10−23 J/K represents the Boltzmann constant, while B equals the noise
bandwidth. This bandwidth depends upon the sensor type and signal processing applied, e.g., during
the evaluation of range in common FMCW sensors, it equals a width of one FFT (Fast Fourier Transform)
bin. In Equation (6), PRt stands for the AR input signal power received through a reflection from a real
target and this power can be evaluated using Equation (2). T0 = 290 K is a standard thermodynamic
temperature. Noise power (NaR ) added by the AR, and related to the input of its receiver, is a function
of the AR sensor noise figure (FR ), and is defined by the formula

NaR = (FR − 1)kT0 B (7)

The sensor–TS structure represents a cascade of RF components with two free-space propagation
paths showing high attenuations. Since the surrounding temperature of all involved antennas should be
close to T0 , for the purposes of the RF noise analysis, these signal paths can be modeled using matched
attenuators showing overall attenuations L1 and L2 , see Figure 1. Considering the AR sensor–TS signal
path, L1 attenuation can be expressed as

(4πRS )2
L1 = (8)
λ2 GRT GSR

The L2 attenuator models the TS–AR sensor signal path and its value can be calculated using the
following relation:
(4πRS )2
L2 = 2 (9)
λ GST GRR
According to [24], available noise powers generated by both attenuators L1 and L2 equal kT0 B,
and this relation describes a logical phenomenon that, from a noise generation point of view,
any well-matched attenuator with high loss will behave like a matched load. The noise figure
of the TSC equals FS , while its gain (GS ) is described by Equation (3). Noise added by the TS related to
its input can be evaluated as
NaS = (FS − 1)kT0 B (10)
Sensors 2020, 20, 2714 6 of 22

All noise contributions can be summed in the A-A’ plane at the input of the AR sensor receiver,
and the corresponding noise floor is

GS
NRS = (kT0 B + NaS ) + kT0 B + NaR (11)
L2

The resulting signal-to-noise ratio (SNRRS ) related to the AR sensor input, and corresponding to
an artificial target simulated by the TS, can be evaluated using

PRt
SNRRS = (12)
NRS

For TS design, it may be useful to know what the highest TSC noise figure (FSmax ) is, or the added
noise power (NaSmax ) which degrades the SNRRS , with respect to the SNRRt , by a still acceptable degree.
This can be evaluated using the relation

SNRRS = KRF ·SNRRt (13)

In this condition, KRF stands for the coefficient of the allowed SNRRt drop by additive RF noise,
e.g., KRF = 0.5 for the –3 dB drop considered. If the TS shows the same RCS as the real target, hence PRS
= PRt , the maximum possible noise (NaSmax ) added by the TS, and related to the input of TSC, can be
expressed using Equations (12) and (6) as

(kT0 B + NaR )(1 − KRF )L2

NaSmax = (14)
KRF GS

From Equation (10), the maximum (FSmax ) TSC noise figure can be evaluated as

NaSmax
FSmax = +1 (15)
kT0 B

or, using Equations (3), (7) and (14) as

FR L2 GSR GST λ2 R4t (1 − KRF )

FSmax = +1 (16)
4πR4S KRF σ

It is clear that for an ideal case where the TS circuitry does not worsen the SNRRS (KRF = 1),
the highest TSC noise figure equals FSmax = 1, but quickly increases when even a slight (tenths of dB)
SNRRS degradation is allowed.
Examples of practical values are presented in Section 6. The calculated values of FSmax are
perhaps even surprisingly higher than can be expected from the low range of RS . The high obtained
values of FSmax are naturally advantageous and they can lead to simpler than usual millimeter-wave
down-converting structures. On the other hand, the TS cannot be fully passive, and amplifiers must be
used at suitable system positions. That is why their noise figures must be considered and the noise
figure of the whole TSC carefully evaluated.

4. Phase Noise Influence

References [10–12] are probably the only ones depicting automotive simulators working directly
in the AR sensor operational frequency band. The solution in [10] works in the 76–77 GHz band
and is based on a long section of a dielectric waveguide equipped with a fast sliding reflecting ring.
Its operating principle is simple and it does not require any complex circuitry. Nonetheless, it shows
great dimension, while enabling only a single target to be simulated, with its range limited to 20 m and
the radial velocity a mere ± 30 km/h. The usage of simulators, according to [11], is limited to FMCW
radars, described, e.g., in [5] and [23]. The application of this solution is also troubled by the fact that
Sensors 2020, 20, 2714 7 of 22
Sensors 2019, 19, x FOR PEER REVIEW 7 of 21

manyFrequency
difficult toconversions
ascertain sensor parameters
enable efficient must
signalbe processing
known in advance.
but bring Allathe more versatile
significant and
potential
efficient solutions are based on frequency conversion and signal processing
problem in that a phase noise is introduced into both sensors and simulators from the employed performed at substantially
lowerFor
LOs. intermediate
both sensors frequencies (IF) or one
and simulators, base-band (BB) for
LO is used frequencies, see, e.g.,
both the down- and[13–18].
up-conversions, which
Figure 4 depicts a TS structure using standard millimeter-wave
is why the concerned phase noises can be mutually correlated and time delay-dependent. down-converting and
up-converting
Considering circuits. The signal
practically used ARreceived byitthe
sensors, ANT1 input
is generally veryantenna
difficultisto amplified
make any byassumptions
a low-noise
amplifier
on (LNA)noise
their phase and, [3].
using
Whenthe MIX1 millimeter-wave
evaluating the concerned mixer andnoise
phase the LO1 local oscillator,
influences, converted
it is necessary to
sensor operational frequency band f ± B /2
know many important parameters and a great 0deal Rof information. In addition to the band
from the AR into the intermediate frequency HW
f IF ± BR /2. such
parameters, The asfrequency
the AR sensor BR equals
band output the the
power, ARpower
sensor’s modulation
spectrum densitybandwidth.
(PSD) of its LOIn the IF
phase
frequency band, all signal processing steps necessary for the target simulator operation
noise, cross-talk gain or time delay of this cross-talk, it is necessary to know the modulation scheme are performed.
Consequently,
and the processed
applied signal processing signal is converted
methods. back to the
Unfortunately, ARARsensor
sensormanufacturers
frequency band using
only the
rarely
up-converter
provide consisting ofand
such information theitMIX2
is notmillimeter-wave
usually possible mixer and the
to measure themLO1 in local oscillator.
a reasonable way.Since
Thatthe
is
same LO1 is used for both the down- and up-conversions, the process is coherent
why, to perform a basic estimate of the influence of the simulator's inner LO on all sensor simulators’ and precise output
frequencies
noise behavior,can the
be reached.
influenceThe powernoise
of phase amplifier
on thePA is needed
sensor's if LO
inner an output power to
is considered (PSTbe) exceeding
negligibly
approximately
low. −10 dBm is required, which is nearly always true in the communication field.

ANT1 LNA f IF  BR 2
MIX1
RF IF
FIL1 FIL2
LO
f 0  BR 2
f LO1 signal-
LO1 processing
ANT2
LO
PST RF IF
FIL1 FIL2
PLO1
PA f 0  BR 2 MIX2 f IF  BR 2

Figure 4.
Figure TS block
4. TS block diagram
diagram with
with common
common frequency
frequency converting
converting circuits.
circuits.

Frequency
The conversions
simplified analysisenable efficient
considers thesignal
TS with processing
a singlebut bring a significantand
down-converting potential problem
up-converting
in thatand
chain a phase
idealnoise
delayisblock,introducedas caninto bothin
be seen sensors
Figureand 5a. simulators fromthe
In this scheme, thesame
employedLO is LOs.
used For both
for both
sensors and simulators, one LO is used for both the down- and
the down- and up-conversion. In the time domain, phase noise can be described using random up-conversions, which is why the
concerned phase noises can be mutually correlated and time delay-dependent.
phase fluctuations (φ(t)) which add to the general phase of the LO signal. From a frequency point of
view,Considering
phase noisepractically
representsused AR sensors,
random frequency it is generally
fluctuationsvery(Δf).
difficult
In theto case
makeofany assumptions
a zero time delay on
their phase noise [3]. When evaluating the concerned phase noise influences,
(τmem) (practically unachievable) of the TS, the resulting phase noise influences equal zero. The it is necessary to know
many important
concerned frequencyparameters and a great
fluctuations deal of information.
are eliminated In addition
in the down- to the HW parameters,
and immediate such
up-conversions,
as the AR sensor
generally described as output power, the power spectrum density (PSD) of its LO phase noise, cross-talk
gain or time delay of this cross-talk, it is necessary to know the modulation scheme and applied signal
processing methods. Unfortunately, AR   fLO manufacturers
fRFsensor f    fLO  fonly
  frarely
RF provide such information (17) and
it is not usually possible to measure them in a reasonable way. That is why, to perform a basic estimate
of the influences RX of(t)the simulator’s inner LO on all sensor simulators’ noise behavior, the influence of
 t    t ,  mem 
s I F (t)
phase noise on the sensor’s inner LO is considered to be negligibly low.
The simplified analysis considers the TS with a single down-converting and up-converting chain
s L O (t) τmem
and ideal delay block, as can be seen in Figure 5a. In this scheme, the same LO is used for both the down-
and up-conversion. s T X (t) In the time domain, phase noise 1 can be  mem
described using random phase fluctuations
(ϕ(t)) which add to the general phase of the LO signal. From a frequency point of view, phase noise
represents random frequency (a) fluctuations (∆f ). In the case (b)of a zero time delay (τmem ) (practically
unachievable) of the TS, the resulting phase noise influences equal zero. The concerned frequency
Figure 5.are
fluctuations (a)eliminated
Simplified in schematic
the down- of TS
andused for the LO
immediate phase noise propagation
up-conversions, analysis. (b)
generally described as
Equivalent circuit of the LO phase noise propagation.
fRF − ( fLO + ∆ f ) + ( fLO + ∆ f ) = fRF (17)
Considering a real TS, the delay (τmem) consists of a latency of both the analog and digital TS
circuits, and an intentionally set delay simulating a round-trip of the radar signal to the target and
back. The utilized LO signal includes a phase noise described by random phase fluctuations (φ(t)).
Due to the multiplicative nature of phase noise, i.e., phase noise power is always directly
 view, phase noise represents random frequency fluctuations (Δf). In the case of a zero time delay
(τmem) (practically unachievable) of the TS, the resulting phase noise influences equal zero. The
concerned frequency fluctuations are eliminated in the down- and immediate up-conversions,
generally described as
Sensors 2020, 20, 2714 fRF   fLO  f    fLO  f   fRF 8 of 22
(17)

 t    t ,  mem 
s RX (t) s I F (t)

s L O (t) τmem

s T X (t) 1  mem

(a) (b)
Figure
Figure 5. (a)
5. (a) Simplified
Simplified schematic
schematic of used
of TS TS used
for for
the the
LO LO phase
phase noisenoise propagation
propagation analysis.
analysis. (b) (b)
Equivalent
Equivalent circuit
circuit of LO
of the the phase
LO phase noise
noise propagation.
propagation.

Considering
Consideringa real TS, TS,
a real the the
delay (τmem
delay ) consists
(τmem of aoflatency
) consists of both
a latency thethe
of both analog andand
analog digital TS TS
digital
circuits, and an intentionally set delay simulating a round-trip of the radar signal to the target
circuits, and an intentionally set delay simulating a round-trip of the radar signal to the target andand back.
Theback.
utilized
TheLO signal includes
utilized LO signala phase noise
includes described
a phase bydescribed
noise random phase fluctuations
by random phase(ϕ(t)). Due to the
fluctuations (φ(t)).
multiplicative
Due to thenature of phase noise,
multiplicative i.e., of
nature phase noisenoise,
phase poweri.e.,
is always
phase directly proportional
noise power to a power
is always directly
of the deterministic signal, the following analysis neglects signal amplitudes. Hence, the complex LO
signal can be expressed as
sLO (t) = exp(j(2π fLO t + ϕ(t))) (18)

Since phase noise of the AR is supposed to be negligible, the received RF signal can be described
as a pure harmonic signal by
sRX (t) = exp(j(2π fRF t)) (19)

The down-conversion process performed in the TS can be described as a multiplication of the

received and LO signals by

sIF (t) = sRX (t)s∗LO (t) = exp(j(2π fIF t − ϕ(t))) (20)

where * represents the complex conjugate. After the introduction of the time delay (τmem ) and
up-conversion, the signal (sTX ) transmitted back from the TS to the AR sensor can be described as

sTX (t) = sIF (t − τmem )sLO (t) = exp(j(2π fRF t + α0 (τmem ) + ∆ϕ(t, τmem ))) (21)

In this equation, α0 (τmem ) represents a general constant phase shift, while ∆ϕ(t, τmem ) describes
time-dependent random phase fluctuations:

∆ϕ(t, τmem ) = ϕ(t) − ϕ(t − τmem ) (22)

Such a propagation of phase noise from the LO through the down-conversion, time delay and
up-conversion described by Equation (21) can be modeled using the circuit shown in Figure 5b.
The impulse response of this circuit can be expressed as

h(t) = δ(t) − δ(t − τmem ) (23)

where δ represents a Dirac delta function and a time delay (τmem ) of the entire chain generating an
artificial target at the required distance. A voltage transfer function (H (∆f )) of the circuit equals the
Fourier transform of Equation (23):

Z∞
H (∆ f ) = h(t)e−j2π∆ f t dt = 1 − e−j2π∆ f τmem (24)
−∞
Sensors 2020, 20, 2714 9 of 22

Into this equation, the following frequency transform was introduced:

∆ f = f − fb (25)

By considering both the positive and negative values of ∆f, this transform enables the concerned
phase noise frequency dependences in the vicinity of any beat frequency (f b ) corresponding to the
FMCW principles to be described. Based on Equation (24), the power transfer function can be
expressed as
 2
G(∆ f ) = H (∆ f ) = 4 sin2 (π∆ f τmem ) (26)
Sensors 2019, 19, x FOR PEER REVIEW 9 of 21
Frequency dependence of the transfer function G(∆f ) as a parametric function of the simulated
target ranges
positive andfunction
transfer plotted invalues
the dBinscale for the
Figure positive
6 reflect thevalues of ∆finterferences
positive in a logarithmic frequency
of noise scale
in certain
can be seenbands.
frequency in Figure 6.

Sensors 2019, 19, x FOR PEER REVIEW 9 of 21

positive transfer function values in Figure 6 reflect the positive interferences of noise in certain
frequency bands.

Figure
Figure 6.
6. Transfer
Transfer function
function of
of phase
phase noise
noise in
in TS
TS and
and various
various target ranges.
ranges. The
The function
function is
is plotted
plotted in
in
the
the logarithmic
logarithmic frequency
frequency scale
scale and ∆f.
and positive values of Δf.

Owing
The phaseto strong
noise noise correlation,
properties very short
of practical simulatedare
RF oscillators target rangesdescribed
typically (short timebydelays) lead to
the function
a nearly
f Figure
asideal
a 6.PSDsubtraction
of the phaseand suppression
noise related of to
phase
the noise
carrierat power
low frequency offsets. for
and expressed Generally, when
the positive
Transfer function of phase noise in TS and various target ranges. The function is plotted in
simulating targets at higher ranges, time delay values are higher and phase noise at the up-converter
frequencies of Δf only
the logarithmic [25]. scale
frequency It can andbe measured
positive valuesor
of found
Δf. in a datasheet of the LO utilized. An
output is less correlated, therefore, phase noise suppression is not only substantially lower, but in definite
example of the function  f  corresponding to a standard PLL-based mmW (milimeter-wave)
frequency ranges it can be “amplified” up to 6 dB, though it is not a real amplification as the positive
The phase noise properties of practical RF oscillators are typically described by the function
oscillator can be seen
transfer function in Figure
values 7. 6 reflect the positive interferences of noise in certain frequency bands.
in Figure
 f The
as a PSD of the phase noise related to the carrier power and expressed for the positive
phase noise properties of practical RF oscillators are typically described by the function L( f )
frequencies
as a PSD of the of phase
Δf only [25].
noise It cantobe
related themeasured or found
carrier power in a datasheet
and expressed of the LO
for the positive utilized.ofAn
frequencies ∆f
example
only [25]. of the be
It can measured or
function f  found
corresponding to a of
in a datasheet standard PLL-based
the LO utilized. AnmmW
example (milimeter-wave)
of the function
L( f ) corresponding
oscillator can be seen to in aFigure
standard
7. PLL-based mmW (milimeter-wave) oscillator can be seen in Figure 7.

Figure 7. Phase noise spectral density f of a typical mmW LO source described as −65
dBc/Hz@10 kHz, −75 dBc/Hz@100 kHz and −115 dBc/Hz@1 MHz. The half-width of the phase noise
pedestal is 105 Hz.

Since
Figureone
Figure 7. of the
7. Phase
Phase following
noise spectral
noise equations
density
spectral density a typical
f  of mmW
L( f ) ofrequires athe integration
LO source
typical of PSD
mmWdescribed
LO asphase
source−65 noise as
dBc/Hz@10
described inkHz,
an FFT
−65
window, unsymmetrical
−75 dBc/Hz@100 with
kHz and −115respect to MHz.
dBc/Hz@1 the carrier, of the phase noise
the function
The half-width f  was is 105 Hz.
mirror-extended
pedestal to
dBc/Hz@10 kHz, −75 dBc/Hz@100 kHz and −115 dBc/Hz@1 MHz. The half-width of the phase noise
coverpedestal
both the negative
is 10 5 Hz. and positive frequencies of Δf. In the following text, this DSB phase noise
function is indicated as SLO( Δf ).
Sinceresulting
The one of the following
phase equationsdensity
noise spectrum requires(Sthe
Δφ ( integration
Δf )) at theofup-converter's
PSD phase noise in an
output canFFT
be
window, unsymmetrical with respect to the carrier, the function   was mirror-extended to
evaluated as f
S  f   Sof
cover both the negative and positive frequencies LO  f  G  f 
Δf. In the following text, this DSB phase noise
(27)
Sensors 2019, 19, x FOR PEER REVIEW 10 of 21

B usually
Sensors 2020, equals
20, 2714 1/Tsw,
and determines the resolution in a spectrum [26]. The phase noise power
10 of in
22
the digitized IF or BB signal can be evaluated as
f2

PN  PR  S   f  f b  df
Since one of the following equations N requires the integration of PSD phase noise in an FFT window,
(28)
unsymmetrical with respect to the carrier, the function f1 L( f ) was mirror-extended to cover both the
negative and positive frequencies of ∆f. In the following text, this DSB phase noise function is indicated
as SLOIn(∆f
this
). equation, the frequency bounds correspond to the edges of individual FFT bins
following f2 – f1 = B.
The resulting For the
phase target
noise range density
spectrum evaluation,(S∆ϕthe (∆fbin
)) atcontaining the beat frequency
the up-converter’s output can (fb) be
is
decisive; the
evaluated as influences of phase noise in other bins are unimportant.
Figure 8 indicates that any integral S∆ϕ (of
∆ fEquation
) = SLO (∆ (28)
f )Gof(∆the
f ) SΔφ ( Δf ) plot in a common B = 12.5(27)
kHz-wide FFT window, including the beat frequency (fb) at Δf = 0, leads to very low values of NPN,
An example
resulting of the values
in negligible effects of TSSphase
ofthe ∆ϕ (∆ fnoise
) evaluated
on thefor AR thesensor’s
PLL phase noisenoise
overall specified in Figure
behavior. This7
can be seen
situation caninsignificantly
Figure 8. differ if a simulation of multiple targets is required.

(a) (b)
8. Phase noise spectral density (S
Figure 8. (S∆ϕ
Δφ (Δf
(∆f ))))of
of the
the TS
TS up-converted
up-converted output
output signal
signal for
for an LO
source with L(ff) shown
shownininFigure
Figure7.7.(a)
(a)Plotted
Plottedforforthe
thepositive
positivevalues
valuesof of ∆f
Δf in a log scale, frequency
range
range 0–10
0–10 MHz.
MHz. TheThe marked
marked frequency
frequency range
range is
is crucial
crucial when
when simulating
simulating more
more artificial
artificial targets
targets at
at aa
single
single azimuth.
azimuth. (b)
(b) For
For both
both the
the positive
positiveandandnegative frequenciesofof∆fΔfininaalinear
negativefrequencies linearfrequency
frequencyscale,
scale,atata
frequency
a frequencyrange
range ±500
ofof ±500kHz.
kHz.

The impact
5. Simulation of TS phase
of Multiple noise on an AR sensor’s overall performance depends on the radar
Targets
modulation scheme and applied signal processing. For example, as mentioned earlier, FMCW radars
The target
evaluate generation
rangesoffrom
complex radar frequencies
beat signal pictures may (f brequire thepurpose,
). For this simulation of being
before M > 1processed
targets (but
by
typically M ≤ 3) at a single azimuth, i.e., at a single sensor’s TS transmitting antenna’ connecting
the FFT, the IF or BB signals at the outputs of radar receivers are sampled by suitable ADCs throughout line.
Theoretically,
almost an entireonly the nearest
duration target chirps
of frequency should(Tbe detected, while the more distanced reflections
sw ). The corresponding noise bandwidth B usually
should1/T
equals beswshaded. In practice,
, and determines theautomotive
resolution insensors often[26].
a spectrum detect
Thetaller
phasetargets standing
noise power ordigitized
in the moving
behind the nearest target. In
IF or BB signal can be evaluated asthis case, the phase noise problems described in Section 4 can become
more serious. This phenomenon results from the overlapping of the phase noise spectra around the
beat signals corresponding to nearby targets. Zf2
According to, e.g., [6], the rangeNPN (Rti)=ofPRthe i-th
S∆ϕ (target
f − fb )measured
df by the FMCW sensor can(28) be
evaluated from the detected i-th beat frequencyf (fbi) as
1

f c Tsw
Rti  bi 0 to
In this equation, the frequency bounds correspond the edges of individual FFT bins following (29)
2 BR
f 2 – f 1 = B. For the target range evaluation, the bin containing the beat frequency (f b ) is decisive; the
influences
where c0 is of phase
the speednoiseof
in other
light. bins are unimportant.
In this equation, BR represents the FMCW radar modulation
Figure 8 indicates that any integral
bandwidth (chirp bandwidth), while Tsw stands of Equation
for the (28) the S∆ϕ The
chirpofduration. (∆f ) total
plot phase
in a common B=
noise power
12.5
(NPNikHz-wide
) degrading FFTthewindow, including
beat signal the simulated
of the i-th beat frequency b ) at ∆f
target(ftakes = account
into 0, leads the
to very
PSDlow values
of its own
of N
phase , resulting in negligible effects of TS phase noise on the AR sensor’s overall
PNnoise and the phase noise of M–1 neighboring targets simulated at the same sensor simulator’s noise behavior.
This situation can
TX connecting significantly
line. The phasediffer
noiseifpower
a simulation of multiple
(NPNi) can targets as
be calculated is required.
a summation of the phase
noise powers of Equation (28) from of all the individual targets in a B spectrum frequency bin width
5. Simulation of Multiple Targets
as
The generation of complex radar pictures may
f2 i require the simulation of M > 1 targets (but
 
M
N PNati a 
typically M ≤ 3) at a single azimuth, i.e., PRj  sensor’s
single S j f  TS
fbj transmitting
df antenna’ connecting line.
(30)
j 1
Theoretically, only the nearest target should be detected,
f1i while the more distanced reflections should
be shaded. In practice, automotive sensors often detect taller targets standing or moving behind the
Sensors 2020, 20, 2714 11 of 22

nearest target. In this case, the phase noise problems described in Section 4 can become more serious.
This phenomenon results from the overlapping of the phase noise spectra around the beat signals
corresponding to nearby targets.
According to, e.g., [6], the range (Rti ) of the i-th target measured by the FMCW sensor can be
evaluated from the detected i-th beat frequency (f bi ) as

fbi c0 Tsw
Rti = (29)
2BR

where c0 is the speed of light. In this equation, BR represents the FMCW radar modulation bandwidth
(chirp bandwidth), while Tsw stands for the chirp duration. The total phase noise power (NPNi )
degrading the beat signal of the i-th simulated target takes into account the PSD of its own phase
noise and the phase noise of M–1 neighboring targets simulated at the same sensor simulator’s TX
connecting line. The phase noise power (NPNi ) can be calculated as a summation of the phase noise
powers of Equation (28) from of all the individual targets in a B spectrum frequency bin width as

M
X Zf2i  
NPNi = PR j S∆ϕ j f − fbj d f (30)
j=1 f1i

Considering the worst case f bi positions, the maxima of the PSD of the phase noise of the
neighboring targets fall at the beat frequency of the middle target. According to Figure 8, the PSD
maxima occur around the frequency offset ∆f =f bc = 100 kHz, and by using Equation (29), the critical
distance (∆Rtc ) between the simulated targets can be evaluated as

fbc c0 Tsw
∆Rtc = (31)
2BR

According to Equation (27), in addition to the transfer function G(∆f ), the maxima of the function
S∆ϕ (∆f ) strongly depend upon the phase noise properties of the microwave frequency source used.
Modern LOs are based on the PLL structure which, within a bandwidth of its low-frequency loop,
reduces the PSD of the phase noise around the carrier to an approximately constant value (LP ) shaped
into a so-called pedestal ([26,27]). In an example presented in Figure 7, only the right end of this phase
noise pedestal determines the maxima of the function S∆ϕ (∆f ), and therefore the critical beat frequency
(f bc ) and the critical distance (∆Rtc ) between the two targets, respectively.
The common values Tsw = 80 µs and BR = 600 MHz used in Section 6 lead to ∆Rtc = 2 m which
may be quite a realistic inter-target distance. In Figure 9, the simulated spectra of the IF or BB radar
signal, with M = 3-detected targets with identical RCSs, are plotted. All considered targets share the
same LO characterized by the PSD of its phase noise, as defined in Figure 7. The middle target is
exactly 2 m from the outer targets, i.e., about 100 kHz on the frequency axis. The beat signal (SIF2 )
corresponding to the middle target at the distance Rt2 = 15 m is degraded by a phase noise originating
from both outer beat signals. Considering the approximately constant values of S∆ϕi (∆f ) in a frequency
band B around f bc , the worst case phase noise power (NPNw ) in the single FFT bin containing the
middle beat frequency can be estimated as

NPNw = B(PR1 S∆ϕ1 ( fbc ) + PR3 S∆ϕ3 ( fbc )) (32)

where PR1 and PR3 depict the powers of beat signals of the two outer targets, while S∆ϕ1 (f bc )
and S∆ϕ3 (f bc ) represent the maximal values of the PSD of the correlated phase noise according to
Equation (29). Due to the generally small value of S∆ϕ for small frequency offsets (see Figure 8),
the degradation of the beat signal of the middle target by its own phase noise is considered to be
negligible. The above-described behavior is illustrated in Figure 9.
source used. Modern LOs are based on the PLL structure which, within a bandwidth of its
low-frequency loop, reduces the PSD of the phase noise around the carrier to an approximately
constant value ( P ) shaped into a so-called pedestal ([26,27]). In an example presented in Figure 7,
only the right end of this phase noise pedestal determines the maxima of the function SΔφ(Δf ), and
therefore the critical beat frequency (fbc) and the critical distance (ΔRtc) between the two targets,
Sensors 2020, 20, 2714 12 of 22
respectively.

(a) (b)
Figure 9.9. (a)
(a)MATLAB
MATLAB simulation
simulation of the
of the power
power spectrum
spectrum of received
of received radar
radar beat beat of
signals signals of three
three artificial
artificial
targets attargets
a singleatazimuth
a singlewith
azimuth
a radarwith
cross-sections (RCS) σ = 100
a radar cross-sections m2 and
(RCS) σ = 100 m2 and
distances of distances
13, 15 andof1713,
m
(mutual critical distance equals 2 m). Applied noise bandwidth equals B = 12.5 kHz and all simulated
15 and 17 m (mutual critical distance equals 2 m). Applied noise bandwidth equals B = 12.5 kHz and
all simulated
targets share thetargets
same share the same LO
LO characterized by characterized by thedensity
the power spectrum power (PSD)
spectrum
of itsdensity (PSD)
phase noise of its
defined
in Figure
phase 7. The
noise frequency-modulated
defined in Figure 7. The continuous-wave (FMCW)continuous-wave
frequency-modulated radar parameters(FMCW)
considered are
radar
defined
parametersin Section 6. (b)are
considered Detail of theinapproximately
defined Section 6. (b) Detail of the−125
calculated dBm phase calculated
approximately noise floor.−125 dBm
phase noise floor.
This graph depicts beat signals, together with the frequency dependences of the phase noise
powers
The(N PN ) integrated
common values T inswthe = 12.5
= 80B μs and kHz windows,
BR = 600 MHz used and inrelated
Sectionto 6three
lead targets
to ΔRtc simulated
= 2 m which at
the
mayranges
be quiteequal to 13, 15
a realistic and 17 m.
inter-target The phase
distance. noise 9,
In Figure powers related to
the simulated the individual
spectra beats
of the IF or were
BB radar
calculated
signal, with M = 3-detected targets with identical RCSs, are plotted. All considered targets sharewas
using Equation (28) and the phase noise power corresponding to the phase noise floor the
calculated using Equation
same LO characterized by (30).
the PSDIt can be phase
of its clearlynoise,
seen that the worst
as defined case phase
in Figure 7. The noise power
middle level
target is
corresponds to thethe
exactly 2 m from 15-m-target, but significant
outer targets, i.e., about phase
100 kHznoise
onlevels are apparent
the frequency axis.even
The around both(S
beat signal side
IF2)
targets. If Rt1,3 >>
corresponding to R tc , the
the received
middle targetpowers
at theand the PSDRof
distance t2 the
= 15correlated phase noise
m is degraded by from
a phasebothnoise
side
targets are almost
originating identical,
from both outerand
beatassuming an approximately
signals. Considering constant value
the approximately for LP , values
constant anotherofdegree
SΔφi(Δfof)
simplification
in a frequencycan be derived:
band B around fbc, the worst case phase noise power (NPNw) in the single FFT bin
containing the middle beat frequency can be estimated as
NPNw = B(PR1 S∆ϕ1 ( fbc ) + PR3 S∆ϕ3 ( fbc )) ≈ 2BPR S∆ϕ ( fbc ) ≈ 2BPR G( fbc )LP (33)
NPNw  B  PR1SΔφ1  f bc   PR3 SΔφ3  f bc   (32)
From the resulting value of NPNw , corresponding to the 15-m-target, requirements placed on the
where simulator
target PR1 and PLO
R3 depict the powers of beat signals of the two outer targets, while SΔφ1(fbc) and
phase noise qualities can be determined. Since no practical data on AR sensor
S Δφ3(fbc) represent the maximal values of the PSD of the correlated phase noise according to Equation
phase noise are generally available, the influence of TS phase noise was related to the AR sensor’s
(29). DueRFtonoise
additive the properties,
generally small
whichvalue
can beofestimated
SΔφ for small much frequency
more easilyoffsets (see more
and much Figureprecisely.
8), the
degradation
As of the
with the case beatadditive
of the signal of
RF the middle
noise analysis target by its in
presented own phase3, noise
Section is considered
an allowed drop (KPNto) be
of
negligible. The above-described behavior is illustrated in Figure 9.
the SNRRt of the detected beat signal occurs at the AR sensor’s input and this can be evaluated using
This graph depicts beat signals, together with the frequency dependences of the phase noise
the relation
powers (NPN) integrated in the B = 12.5 SNR kHz windows, and related to three targets simulated at(34) the
RP = KPN ·SNRRt

In this formula, i.e., KPN = 0.5 for a −3 dB drop, KPN = 0.1 for a −10 dB drop, etc., the SNRRP
represents the signal-to-noise ratio of the power of the detected beat signal of the 15-m-target, with
respect to the total influence of the additive RF noise power and worst case phase noise evaluated by
Equation (30), and related to the input of the AR sensor’s receiver. From the FMCW sensor point of
view, both additive RF noise and phase noise show similar behavior. Both noises are uncorrelated,
increase the noise floor and vary the phase of the detected beat signals. That is why their total influence
can be evaluated as the sum of the above stated noise powers:

PR
SNRRP = (35)
kT0 B + NaR + NPNw
Sensors 2020, 20, 2714 13 of 22

Using Equations (2), (7), (26), (33), (34) and (35), it is possible to express the boundary phase noise
pedestal value (LPb ) of the LO source within the TS which KPN -times degrades the signal-to-noise
ratio (SNRRt ) in a single B-wide FFT bin as

(kT0 B + NaR )(1 − KPN )(4π)3 R4t

LPb =   (36)
2(R −R )
8KPN PT GRT GRR σλ2 B sin2 π fbc tc0 S

It should be emphasized that this equation corresponds to a simplification of Equation (33),

meaning that the evaluation of Equation (36) for near targets provides less accurate results.
The practical values of the allowed phase noise properties of a TS´s LO source, evaluated
according to Equation (36) and presented in the next section, indicate that the influence of phase noise
on a sensor simulator’s behavior is still relatively low and that even relatively “dirty” LOs can be
acceptable. As stated in Section 7, this fact simplifies the design and construction of the concerned
sensor simulator setups.

6. Application and Verification of Derived Formulas

This section presents practically calculated values corresponding to the TS suitable for the testing
of FMCW medium-range radars (MRRs). As a reference unit, the Bosch MRR1 device operating in
the 76–77 GHz band, and described in [6] and [28], is considered. Many of its parameters are not
available to the public, however, they were estimated from an open sample of this AR sensor and
spectrum-analyzer measurements. The obtained parameters necessary for the calculations presented
in this section are summarized in Table 1.

Table 1. Automotive medium-range radars (MRR) parameters.

MRR Parameter Value

Start frequency of FMCW chirps f c =76 GHz
Bandwidth of FMCW chirps BR = 500 MHz
Sweep time FMCW chirps Tsw = 80 µs
Radar transmitted power PT = 10 dBm
Radar transmitting antenna gain GRT = 20 dB
Radar receiving antenna gain GRR = 10 dB
Noise figure of radar receiver FR = 15 dB
Sampling frequency of IF signal f s = 10 MHz

Values presented in Table 1 were read from the performed AR measurements or estimated from
an opened AR sample. Valuable information can also be found in the references, especially [6].
The measurements were performed using a W-band receiving horn antenna, mixer with a 75.9 GHz LO
signal and oscilloscope with 5 GHz sampling. Figure 10 depicts the measured spectrogram showing the
bandwidth of the frequency chirps and active sweep time. The transmitting antenna is formed by an 8
× 10 array and its gain was estimated to 20 dB, while the gain of the receiving 1 × 8 array was estimated
to be 10 dB. The AR sensor receiver noise figure was assessed according to noise figures of similar
80 GHz LNAs, and significant reserves were added to cover the additional losses. The presented
sampling frequency value is based on consultations with experts involved in the AR problems.
 Sweep time FMCW chirps Tsw = 80 µ s
Radar transmitted power PT = 10 dBm
Radar transmitting antenna gain GRT = 20 dB
Radar receiving antenna gain GRR = 10 dB
Sensors 2020, 20, 2714
Noise figure of radar receiver FR = 15 dB 14 of 22
Sampling frequency of IF signal fs = 10 MHz

Figure 10.
Figure 10. Measured
Measuredspectrogram
spectrogram of of received
received andand down-converted
down-converted signal
signal from from the MRR1
the MRR1 radar
radar sensor
sensor
with thewith the external
external LO frequency
LO frequency set to 75.9set to 75.9
GHz. GHz. Transmitted
Transmitted chirps
chirps range fromrange fromGHz,
76 to 76.5 76 towhile
76.5
GHz, while the active time of the chirp sweep is approximately
the active time of the chirp sweep is approximately 80 µs long. 80 μs long.

Let us suppose
suppose aa TSTSequipped
equippedwith withtwo
twoidentical
identicalhorn
hornantennas
antennaswithwith gains
gains GSR
GSR = G=STG=ST
14=dB14and
dB
and AR sensor–TS
AR sensor–TS distance
distance equal
equal to R toS R = 0.5
= S0.5 m.m. With
With respecttoto[6],
respect [6],all
all further
further calculations
calculations consider
targets with an RCS σσ = = 1, 10 and 100 m22, and simulated ranges of Rtt == 3,
simulated target ranges 3, 10,
10, 30
30 and
and 100
100 m.
m.
These values should sufficiently cover the requirements for MRR testing involving pedestrians, pedestrians, cars,
cars,
trucks, traffic signs or guardrails.
guardrails.
The RF power received by the TS reaches PSR = –20.1 dBm behind the TS’s receiving antenna,
whileTable 2. Calculated
the calculated valuesofofthe
values overall TS system
TS system’s gain(G(G
gain S )S)providing
(dB) necessary to simulate
a constant RCS,anasartificial
defined by
target(3),
Equation witharea constant
stated inRCS and2.specific
Table range from –60 dB to +20 dB which represents an 80 dB
ranges.
The values
required dynamic range. In practice, this dynamic range can be significantly lower. Usually, AR
σ
sensors show relatively narrow antenna radiation 1 m2 10 m2
patterns 100 m
and 2
large targets in near ranges are not
Rt
fully illuminated. In addition, in near ranges, for proper target detection, it is not necessary to simulate
3m 0.0 10.0 20.0
targets with very high RCSs. Therefore, a Gsmax not higher than 0 dB and a dynamic range lower than
10 m –20.9 –10.9 –0.9
60 dB can be considered as satisfactory. This dynamic range can be easily realized, for example, by a
30 m –40.0 –30.0 –20.0
combination of digitally controlled analog attenuators.
100 m –60.9 –50.9 –40.9
Table
The RF2. Calculated values of
power received byoverall
the TSTSreaches
system gain
PSR (G S ) (dB)dBm
= –20.1 necessary to simulate
behind the TS’sanreceiving
artificial target
antenna,
with a constant RCS and specific ranges.
while the calculated values of the TS system’s gain (GS) providing a constant RCS, as defined by
Equation (3), are stated in Table σ
2. The values range from –60 dB to +20 dB which represents an 80 dB
Rt 1 m2 10 m2 100 m2
3m 0.0 10.0 20.0
10 m −20.9 −10.9 −0.9
30 m −40.0 −30.0 −20.0
100 m −60.9 −50.9 −40.9

The ability of the TS, featuring a passive millimeter-wave up-converter (Figure 4), to simulate
targets at near ranges with a high RCS, is practically limited by the maximal TSC transmitted power
(PST ) according to Equation (4), see Table 3.

Table 3. Calculated values of the TS transmitted signal power (PST ) (dBm) necessary to simulate an
artificial target with a constant RCS at specific ranges.

σ
Rt 1 m2 10 m2 100 m2
3m −20.1 −10.1 −0.1
10 m −41.0 −31.0 −21.0
30 m −60.1 −50.1 −40.1
100 m −81.0 −71.0 −61.0

The developed miniature millimeter-wave converters, described briefly in Section 7, are based
on HMC1058 (Analog Devices) mixers, see [29]. Considering the +13 dBm assumed LO power,
the maximum IF input power (up-converter) should not be, due to linearity requirements, higher than
Sensors 2020, 20, 2714 15 of 22

+3 dBm. Taking into account the 13 dB mixer conversion loss, the maximum up-converter output
power reaches −10 dBm in front of the FIL1 and typically −13 dBm at the input of the ANT2 (Figure 3).
So, a GSmax equal to 0 dB and PST equal to −20 dBm can easily be ensured. Further, for proper target
detection at a 3-m-range, an RSC equal to 1 m2 should be fully sufficient.
During system programming, it can also be useful to evaluate the topical RSC of the simulated
target for the given TS setting. This is possible using Equations (3) and (4) (see Table 4) when
PST = −20 dBm is considered. In near ranges, the values correspond well to the expected reduced
RCS values. In far ranges, these achievable target RCSs are, with many dB to spare, sufficient for the
simulation of all practical traffic scenarios.

Table 4. Calculated values of an achievable RCS of targets at various ranges with a limited maximal TS
output power equal to –20 dBm.

Rt (m) 3 10 30 100
σ (m2 ) 1.02 126 10,200 1,260,000

A high probability of the accurate detection of targets by AR sensors is ensured by sufficient

signal-to-noise ratios. The SNRRt values, defined by Equation (6), correspond to the AR sensor detecting
real targets, and are referred to the input of its receiver. Calculated results are presented in Table 5.
The noise bandwidth equals B = 1/Tsw = 12.5 kHz, which relates to standard FMCW radar chirp rates
and signal processing. As a threshold for a reliable target detection, an SNRRtmin = 10 dB, defined
in [6], was considered. This threshold corresponds to a target with an RCS equal to 20 m2 at a 100-m
range, which corresponds well with the expected MRR capabilities.

Table 5. Calculated values of the signal-to-noise ratio (SNRRt ) (dB) at the AR sensor receiver input for
various RCS values and ranges considering real targets.

σ
Rt 1 m2 10 m2 100 m2
3m 57.8 67.8 77.8
10 m 36.9 46.9 56.9
30 m 17.8 27.8 37.8
100 m −3.1 6.9 16.9

The detection of targets simulated by the TS is affected by the additive RF noise generated by
the TS and described by its noise figure (FS ). The relation in Equation (16) enables the maximum FS
values (FSmax ) to provide a KRF -times degradation of the SNRRS , with respect to the SNRRt values, to be
calculated. Table 6 includes the calculated FSmax values corresponding to a mere 1 dB signal-to-noise
ratio drop (KRF ≈ 0.794). Assuming the fully passive millimeter-wave down-converter structure,
according to Figure 4 with the LNA and PA left out, and IF LNA at the input of the following IF circuits,
the noise figure (FS ) can easily attain approximately 20 dB (5 dB input filter loss, 13 dB conversion loss,
2 dB IF LNA noise figure). As can be deduced from Table 6, the allowed noise figures are, mostly, many
tens of dB higher. Moreover, according to Table 5, the SNRRt values at low ranges are so high that
even the concerned 1 dB drop cannot noticeably influence target detection. These parameters generally
allow the TSs to be designed and realized in a relatively simple way by utilizing components with
higher losses and noise figures, antennas with lower gains, etc.
Sensors 2020, 20, 2714 16 of 22

Table 6. Calculated values of maximal noise figures (FSmax ) (dB) of TS circuits causing the degradation
of the SNRRt by 1 dB.

σ
Rt 1 m2 10 m2 100 m2
3m 49.2 39.2 29.2
10 m 70.1 60.1 50.1
30 m 89.2 79.2 69.2
100 m 110.1 100.1 90.1

As mentioned at the end of Section 4, when generating only a single target, the influence of
the phase noise introduced by the TS LO, due to the correlation effects and integration in close
vicinity to the beat frequency, is negligible. In the case of the TS enabling simulation of more targets
at a single azimuth, the phase noise of the LO source within the TS can become more important.
Equation (36) makes it possible to compute an LO’s phase noise pedestal value (LPb ), providing a
KPN -times degradation of the SNRRt , which is a signal-to-noise ratio corresponding to a real target in
the defined ranges. Table 7 shows the results of Equation (36) for the allowed degradation of the SNRRt
equal to 1 dB, i.e., KPN = 10−1/10 ≈ 0.794. The evaluation assumes f bc = 100 kHz, which corresponds to
a mutual critical distance between two neighboring targets Rtc = 2 m. In the Rt = 3 m and 10 m lines,
the condition Rt >> Rtc does not hold and certain deviations can be expected.

Table 7. Calculated values of the LO phase noise pedestal (LPb ) (dBc/Hz) causing degradation of the
SNRRt by 1 dB; the degradation relates to the middle target when generating three targets with identical
RCSs and a critical Rtc = 2 m distance.

σ
Rt 1 m2 10 m2 100 m2
3m −68.9 −78.9 −88.9
10 m −59.0 −69.0 −79.0
30 m −49.6 −59.6 −69.6
100 m −39.1 −49.1 −59.1

The interpretation of the calculated LPb values presented in Table 7 is slightly more sensitive
than the interpretation of the values of FSmax included in Table 6. In the given 76–81 GHz frequency
band, the common phase noise PSD pedestal values of LPb of the available LOs range from −65 to
−70 dBc/Hz. This explains why Table 7 indicates potential problems, especially in the 10 m and 3 m
lines where relatively high PSD ratios are required. However, even at 30 m, the concerned values of the
SNRRt are, according to Table 5, so high that a 1 dB drop cannot represent any substantial degradation
of the sensor’s detection capabilities. What is more, the evaluated values of LPb corresponding to the
decisive Rt = 100 m range can easily be met, even by relatively “dirty” LOs.
A verification of the formulas derived for the analysis of additive RF noise and phase noise was
performed using the Cadence AWR Visual System Simulator [30]. This solution was enforced by the
fact that common AR sensors offer, at their digital busses, neither digitized internal signals nor any
information on individual detected targets. These data are exclusively internal, while at the CAN
busses, only the resulting instructions for the ECU (“start braking”, “turn to the left”, etc.) are usually
available. Under such circumstances, computer simulations appeared to be the most feasible and
reliable way of how to verify the developed analytical description.
The simulation schematic set for proper RF noise simulation, containing the AR sensor with
parameters stated in Table 1, is shown in Figure 11. The generated FMCW radar chirp is noiseless
and the first stage adding RF noise kT0 B is the TS receiving antenna. Both mixers in TS are considered
as noiseless and the overall TS noise figure (FS ) is assigned to the A7 amplifier. This amplifier, with
variable gain (GS ), also ensures the simulation of an artificial target with an appropriate RCS in
accordance with Equation (3). The influence of free-space losses between the AR sensor and TS are
 Sensors 2019, 19, x FOR PEER REVIEW 16 of 21

degradation of the sensor’s detection capabilities. What is more, the evaluated values of
Sensors 2020, 20, 2714 17 of 22
Pb corresponding to the decisive Rt = 100 m range can easily be met, even by relatively “dirty” LOs.

A verification of the formulas derived for the analysis of additive RF noise and phase noise was
performed
included in theusing
ARthe Cadence
sensor’s AWR
and TS’sVisual System
receiving Simulator
antennas. [30].
The Thisdelay
time solution was enforced
between by the
the transmitted
fact that common AR sensors offer, at their digital busses, neither digitized internal signals
and the received signal, resulting in an appropriate beat frequency in the radar receiver, is ensured nor any
information on individual detected targets. These data are exclusively internal, while at the CAN
by RFDELAY blocks including free-space (FS) delays on the distance (RS ) as well. The total noise
busses, only the resulting instructions for the ECU (“start braking”, “turn to the left”, etc.) are
contribution of the radar to the received signals is introduced by the A14 amplifier with a noise figure
usually available. Under such circumstances, computer simulations appeared to be the most feasible
(FR ) and 0 dB gain.
and reliable way of how to verify the developed analytical description.
CO NSTANTS RADAR PARAM ETERS TS PARAM ETERS TARGET PARAM ETERS OTHER EQ UATIONS

c 0 = c ons t ant s ( " c 0" ) f 0 = 76e9
k = c ons t ant s ( " bol t z mann") BR = 600e6
pi = c ons t ant s ( " pi " ) Ts w = 80e- 6
f s = _SMPFRQ PT = 10^ ( 10/ 10) * 1e- 3
f I F0 = 1. 1e9 GRT = 10^ ( 20/ 10)
GRR = 10^ ( 10/ 10)
FR = 10^ ( 15/ 10)
GSR = 10^ ( 14/ 10) Rt = 30 l ambda = c 0/ ( f 0 + BR/ 2)
GST = GSR RCS = 10 nRt Del ay = r ound( 2* ( Rt - RS) / c 0* f s , 1)
RS = 0. 5 nRSDel ay = r ound( RS/ c 0* f s , 1)
K = 1 FSL = ( 4* pi * RS/ l ambda) ^2
L2 = FSL/ ( GST* GRR)
GS = RCS* 4* pi * RS^ 4/ ( GSR* GST* l ambda^ 2* Rt ^4)
FSmax = FR* L2* GSR* GST* l ambda^ 2* Rt ^ 4* ( 1 - K) / ( 4* pi *RS^ 4* K* RCS) + 1

C HIRP_LFM CHI RP IN OUT

ID=A1 GENERATOR 2 1 1 2
SIGPW R=PTdB dBW 1
+f
PRF=1/T sw
fc MIXER_B2
D UTYCYCLE=100 ID=A5
RIS E PCT=100 -f ANTEN NA RFDELAY ANTEN NA RFDELAY
R ID=A2 MODE=DIFF LO
FSTART=f0 Hz ID=A3 ID=A4 ID=A6
P ATHLOS S=0 dB GCONV=0 dB
FSTOP=f0 + BR Hz DELA Y=nRSDelay/fs s P ATHLOS S=FSLdB dB DELA Y=nRtDelay/fs s PARAM ETERS IN dB
ANTGAIN=GRTdB dB LOS S=0 dB NO IS E=Noiseless
CTRFRQ=f0 + BR/2 Hz 2 ANTGAIN=GSRdB dB TONE LOS S=0 dB
NO IS E=Auto NO IS E=Noiseless NO IS E=Auto NO IS E=Noiseless
ID=A9 GSRdB = 10* l og10( GSR)
FRQ=f0 + BR/2 - fIF0 Hz GSTdB = 10* l og10( GST)
PW R=0 dBm
NO IS E=Noiseless GRTdB = 10* l og10( GRT)
PNMA S K={0.1e3, -65, 10e3, -65, 100e3, -75, 1e6, -115} GRRdB = 10* l og10( GRR)
LINAMP ANTEN NA RFDELAY ANTEN NA PNO IS E=No phase noise LINAMP FSLdB = 10* l og10( FSL)
MIXER_B2 ID=A11 ID=A10 MIXER_B2 ID=A7
ID=A14 ID=A13 ID=A12
P ATHLOS S=FSLdB dB DELA Y=nRSDelay/fs s P ATHLOS S=0 dB ID=A8 GAIN=GSdB FRdB = 10* l og10( FR)
PORT GAIN=0 MODE=DIFF LO
P=1 NF=FRdB ANTGAIN=GR RdB dB LOS S=0 dB ANTGAIN=GSTdB dB LO MODE=SUM NF=FSmaxdB GSdB = 10* l og10( GS)
GCONV=0 dB NO IS E=Noiseless NO IS E=Auto GCONV=0 dB NO IS E=Auto
Z=_Z0 Ohm NO IS E=Auto NO IS E=Noiseless NO IS E=Auto FSmax dB = 10* l og10( FSmax)
NO IS E=Noiseless PTdB=10* l og10( PT)
2 1 1 2
OUT IN OUT IN

F M C W SENSOR F S D E LA Y T A R G E T S IM U LA T O R
Figure
Figure 11.11.
AWRAWR VisualSystem
Visual SystemSimulator
Simulator schematic
schematic used
used for
for the
theanalysis
analysisofofthe
theFMCW
FMCW radar
radarwith
with
the TS simulating a single artificial target. The radar model consists of the chirp generator
the TS simulating a single artificial target. The radar model consists of the chirp generator A1, mixer A1, mixer
A13A13
andand transmitting
transmitting and and receiving
receiving antennas
antennas A2 andA2A12.
and The
A12.radar–TS
The radar–TS round-trip
round-trip delay isdelay is
simulated
simulated by blocks A3 and A11. The TS model consists of receiving and transmitting
by blocks A3 and A11. The TS model consists of receiving and transmitting antennas A4 and A10, antennas A4
andand A10, and
down- and up-converting
down- and up-converting
mixers A5 mixers
and A8,A5 andthe
with A8,shared
with the shared
local local oscillator
oscillator A9. Range A9.
ofRange
a single
of a single target is set by delay block A6 and its
target is set by delay block A6 and its RCS by amplifier A7. RCS by amplifier A7.

The simulation
Figure 12 depicts schematic
the powerset for proper
spectrum RF noise
of the simulation,
beat signal containing
present the AR
at the radar sensor output
receiver’s with
parameters
PORT1. The beatstated in Table
signal with 1, is shown in
a frequency ofFigure
1.5 MHz 11.isThe generated
caused by theFMCW radarofchirp
simulation is noiseless
an artificial target
and the first stage adding RF noise kT 0B is the TS receiving antenna. Both mixers in TS are
set to Rt = 30 m and an RCS σ = 10 m . The PRt = −90.2 dBm power of the beat signal is in accordance
2
considered as noiseless and the overall TS noise figure (FS) is assigned to the A7 amplifier. This
with
SensorsEquation
2019, 19, x(2),
FORwhile the plotted noise floors correspond to the allowed SNRRt drops (KRF ) equal
PEER REVIEW of to
amplifier, with variable gain (GS), also ensures the simulation of an artificial target with17an 21
0, 3 and 10 dB. To mimic the sampling of the beat signal during the duration of the chirp, the noise
appropriate RCS in accordance with Equation (3). The influence of free-space losses between the AR
bandwidth
frequency of the
plot of simulation
phase noise wasat set B = 1/T
thetoradar's BBand= 12.5 kHz.
output (PORT1) FromisEquation
shown in(12), the signal-to-noise
Figure 13. The spikes
sensor and TS are included in the AR sensor’ssw TS’s receiving antennas. The time delay between
ratios (SNR
repeating RS ) at1/T
every theswradar receiver corresponding to thediscontinuities
considered values of K RF areinat
equal to 27.8,
the transmitted and=the12.5received
kHz aresignal,
causedresulting
by periodic
in an appropriate beat in beat signals
frequency theradar
the start of
24.8
every and 17.8
frequency dB. Figure
chirp. 12
Aside also
from includes
these noise
spikes, levels
very (N
good ) calculated
agreement
RS according
between to
Equation
receiver, is ensured by RFDELAY blocks including free-space (FS) delays on the distance (RS) as well. Equation
(27) and (11),
the
and
VSS the
The agreement
simulation
total with the
noiseiscontribution
clearly simulated
observed. values
of the radar is very
to the good.
received signals is introduced by the A14 amplifier
with a noise figure (FR) and 0 dB gain.
Figure 12 depicts the power spectrum of the beat signal present at the radar receiver’s output
PORT1. The beat signal with a frequency of 1.5 MHz is caused by the simulation of an artificial target
set to Rt = 30 m and an RCS σ = 10 m2. The PRt = −90.2 dBm power of the beat signal is in accordance
with Equation (2), while the plotted noise floors correspond to the allowed SNRRt drops (KRF) equal
to 0, 3 and 10 dB. To mimic the sampling of the beat signal during the duration of the chirp, the noise
bandwidth of the simulation was set to B = 1/Tsw = 12.5 kHz. From Equation (12), the signal-to-noise
ratios (SNRRS) at the radar receiver corresponding to the considered values of KRF are equal to 27.8,
24.8 and 17.8 dB. Figure 12 also includes noise levels (NRS) calculated according to Equation (11), and
the agreement with the simulated values is very good.
A validation of the phase noise correlation effects described in Section 4 was also performed
using the VSS (Virtual System Simulator) schematic depicted in Figure 11. Small changes consisting
of Figure 12.
turning12.onSimulated
the phase and calculated
noise spectra
generation of theofTS's radar
BB radar
LO signals
signals
and with
with
turning additive
offadditive RF
RF
all other noise.
noise.
noise The beat
contributions
signal
were at frequency
applied. 1.5 MHz is LO
The considered caused by an
phase artificial
noise target set
parameters aretothose
a 30-mfrom
rangeFigure
and an7.RCS
Theequal to
resulting
10 m
10 m22. The
The amplitude
amplitude of
of the beat signal is constant independently of the TS’s noise figure. Values of
KRF =1,1, 0.5
RF = 0.5 and correspond to TS’s noise figures FSS == 0,
and 0.1 correspond 0, 85
85 and
and 95
95 dB,
dB, respectively. The
The “theory”
“theory”
lines represent
represent the
the theoretical
theoretical additive
additive noise
noise floors
floors evaluated
evaluated according
accordingto
toEquation
Equation(14).
(14).
Sensors 2020, 20, 2714 18 of 22

A validation of the phase noise correlation effects described in Section 4 was also performed
using the VSS (Virtual System Simulator) schematic depicted in Figure 11. Small changes consisting of
turning on the phase noise generation of the TS’s LO and turning off all other noise contributions were
applied.
FigureThe12.considered
Simulated LO
andphase noisespectra
calculated parameters are those
of BB radar from
signals Figure
with 7. The
additive RF resulting
noise. Thefrequency
beat
plot of phase noise at the radar’s BB output (PORT1) is shown in Figure 13. The spikes
signal at frequency 1.5 MHz is caused by an artificial target set to a 30-m range and an RCS repeating
equal toevery
1/Tsw10=m 12.5 kHz
2. The are caused
amplitude bybeat
of the periodic
signaldiscontinuities in beat signals
is constant independently of the at thenoise
TS’s startfigure.
of every frequency
Values of
chirp.KRF
Aside from these spikes, very good agreement between Equation (27) and the VSS simulation
= 1, 0.5 and 0.1 correspond to TS’s noise figures FS = 0, 85 and 95 dB, respectively. The “theory” is
clearly observed.
lines represent the theoretical additive noise floors evaluated according to Equation (14).

Figure 13. Simulation

Figure 13. Simulation ofof phase
phase noise
noise atat the
the radar
radar BB
BB output
output for
for targets
targets at
at ranges
ranges of of 33 and
and 100
100 mm
producing beat frequencies (f ) equal to 150 kHz and 5 MHz, respectively. The frequency
producing beat frequencies (f b ) equal to 150 kHz and 5 MHz, respectively. The frequency axis of ∆f bb
b axis of Δf
represents onlythe
represents only thepositive
positiveoffset
offset from
from thethe corresponding
corresponding beatbeat frequency.
frequency. The spikes
The spikes repeating
repeating every
every =
1/Tsw 1/Tsw = 12.5
12.5 kHz arekHz are by
caused caused by periodic
periodic discontinuities
discontinuities in beatatsignals
in beat signals the startatofthe startfrequency
every of every
frequency
chirp. The chirp. The theoretical
theoretical dependency dependency of phase
of phase noise noise was computed
was computed using(27).
using Equation Equation (27).

To verify
verifythethedeveloped
developedanalytical
analytical description
description presented
presented in Section
in Section 5, a simulation
5, a simulationof an LOof anphase
LO
noise influence when generating more targets at a single azimuth was
phase noise influence when generating more targets at a single azimuth was also performed. The also performed. The resulting
spectrum spectrum
resulting of signals ofcorresponding to the AR to
signals corresponding sensor
the ARinput can be
sensor seencan
input in Figure
be seen14. in The simulation
Figure 14. The
correspondscorresponds
simulation to three artificial targets
to three with antargets
artificial RCS σ with= 100an m2RCS
at distances
σ = 10013, m215,
at and 17 m. The
distances concerned
13, 15, and 17
targets
m. Theproduce
concerned beattargets
signalsproduce
with frequencies
beat signals(f b ) with
equalfrequencies
to 650, 750 and (fb) 850
equalkHz to and
650,powers
750 and (P850
Rt ) equal
kHz
to −65.6,
and powers −68.1
(PRtand −70.3
) equal to dBm.
−65.6, The
−68.1 phase noise dBm.
and −70.3 parameters
The phaseof thenoise
LO signal are shown
parameters of theinLOFigure 7.
signal
To
areproduce
shown in a spectrum
Figure 7. To with higher aresolution,
produce spectrum the withsimulation was performed
higher resolution, with a noise
the simulation was bandwidth
performed
with
equalato B =bandwidth
noise 1.25 kHz. equal to B =side
To reduce 1.25lobes,
kHz. To thereduce side lobes,
Hann window wastheapplied
Hann window
to the timewas domain
applied
to the time
signals domain
before the FFT signals before the
processing. FFT processing.
In addition, to smooth In addition,
the noise to smooth
floor, the noisespectrum
10 individual floor, 10
individual
realizationsspectrum realizations
were averaged. During were
the averaged. During
first simulation the firstassimulation
(marked “noiseless”), (marked
all noiseas sources
“noiseless”),
were
all noiseoff,
turned sources
whereas were turned
during theoff, whereas
next during
simulation the next
(“PN”), onlysimulation
the phase (“PN”),
noise ofonlyan LO thewas
phase noise of
introduced.
an LO
The was introduced.
spectrum in Figure 14 The spectrum
should in Figure
correspond to that14 in
should
Figurecorrespond
9, with the onlyto that in Figure
difference 9, with
being that the
only
noisedifference being
floor is about 10 that the because
dB less noise floor is about
of the ten-times10 dB less because
narrower noiseof the ten-times
bandwidth used,narrower noise
as mentioned,
bandwidth used, resolution.
to obtain a higher as mentioned, to obtainthis
Considering a 10higher resolution.
dB shift, Considering
the agreement between thisFigures
10 dB9 shift,
and 14the is
agreement
very good. between Figure 9 and Figure 14 is very good.
Sensors 2020, 20, 2714 19 of 22
Sensors 2019, 19, x FOR PEER REVIEW 18 of 21

Figure 14. Simulation of beat signals at the AR sensor’s receiver coming from three targets at
distances of 13, 15 and 17 m with an RCS 100 m2. In the “noiseless” case, the AR and TS are
considered to be noiseless, and the only factors worsening the detection of targets is a main lobe of
the Hann window utilized in the FFT processing. The “PN” case introduces phase noise to the LO
signal in TS according to Figure 7. Thanks to the summation of noise contributions from both side
targets, the center target is degraded, as predicted in Figure 9.

7. Optimized HW Structure, System Programming

Results of the analyses presented in this article enable a significant simplification of the tested
sensor simulator setup. Moreover, the definite derived formulas can also be claimed as a benefit for
systemFigure 14. Simulation
14.
programming.Simulationofof beat
beat signals
signals at theat AR
thesensor’s
AR sensor’s
receiver receiver
comingcoming
from three from threeattargets
targets distancesat
distances
of 13, 15 of
and 13,
17 m15 and
with an17 m
RCS with
100 man2 . In
RCSthe 100 m 2. In the
“noiseless” case,“noiseless”
the AR andcase,
TS the
are AR and
considered TStoare
be
Results obtained in Sections 3 and 5 enable the structure depicted in Figure 4 to be simplified.
considered
noiseless,
Firstly, due to tothe
and be
thenoiseless,
only
highfactors andworsening
FSmax the only derived
values factors worsening
the detection
in Section the detection
of targets 3is and
a main oflobe
targets
evaluated isin
of the a main
Hann window
Sectionlobe of6, the
utilized
the Hann in the FFT
window processing.
utilized in The
the “PN”
FFT case introduces
processing. The phase
“PN”
millimeter-wave LNA (MMIC chip, chip capacitors, non-trivial power supply, together noise
case to the
introduces LO signal
phase in TS
noise according
to the LO with
to Figurein 7. Thanks to the
to summation of noise tocontributions from of both side targets, the from
signal
approximately TS15according
wire-bonds) Figure
can be 7. Thanks
eliminated. theSecondly,
summation takingnoise
into contributions
account thecenter target
both
simulator is
side
output
degraded,
targets, theascenter
predicted
targetinisFigure
degraded,9. as predicted in Figure 9.
power (PST) values derived in Section 2 and evaluated in Section 6, the power amplifier (PA) can also
usually
7. be omitted.
Optimized The resulting
HW Structure,
Structure, System down- and up-converters consist only of chip MMIC mixers and
Programming
7. Optimized HW System Programming
planar filters (see Figure 15).
Results
Results of
of the
the analyses
analyses presented in
in this
presentedincludes this article
article enable
enable a significant simplification of
of the
the tested
The developed mmW converter a built-in 14 adB significant
horn antenna simplification
and subharmonically tested
sensor
sensor simulator
simulator setup.
setup. Moreover,
Moreover, the
the definite
definite derived
derived formulas
formulas can
can also
also be
be claimed
claimed as
as aa benefit
benefit for
for
pumped balanced mixer, and, together with an input RF filter reducing the mirror signals, IF filter
system
system programming.
and LO programming.
and IF connectors, occupies only an 80 × 40 × 8 mm volume. Identical converters can be used
Results obtained
obtained inSectionsSections 3 and 5 enable the structure depicted
in Figurein4 to Figure 4 to be
both Results
for the RX and TXinbranches. 3Aand
low 5priceenable
andthelow structure
dimensions depicted
are extraordinarily be simplified.
important for
simplified.
Firstly, due Firstly, due to the high F Smax values derived in Section 3 and evaluated in Section 6,
the design andtorealization
the high ofFSmax values derived
the advanced simulator in setups
Sectionenabling
3 and aevaluated
generationin of aSection 6, the
high number
the millimeter-wave LNA
millimeter-wave (MMIC chip, chipcapacitors,
capacitors,non-trivial
non-trivialpower powersupply,
supply, together
together with
of targets at moreLNA (MMIC
azimuths using chip,
arrayschip of target simulating modules (see Figure 2). Small mmW with
approximately
approximately 15 wire-bonds)
15 wire-bonds) can
can be eliminated.
be eliminated. Secondly,
Secondly, taking
taking into account
intowhich
account the simulator
the simulator output
units can be easily and densely placed in view of the tested sensor enables complexoutput radar
power
power (P (PST ) values derived in
ST) values derived
in Section
Section 22 andand evaluated
evaluated in in Section 6, 6, the power
power amplifier
amplifier (PA)(PA) cancantoalso
pictures corresponding to the practical radar pictures Section
seen by the AR sensors in real traffic also
be
usually be omitted.
usually be omitted. The
Thetheresulting
resulting down- and up-converters consist only of chip MMIC mixers and
generated. Results of phase down- and up-converters
noise analysis also enableconsist only of
the usage ofchip MMIC simple
relatively mixers and
planar
planar filters
filters (see
(see Figure
Figure 15).
15).
therefore low-cost TS local oscillators.
The developed mmW converter includes a built-in 14 dB horn antenna and subharmonically
pumped balanced mixer, and, together with an input RF filter reducing the mirror signals, IF filter
and LO and IF connectors, occupies only an 80 × 40 × 8 mm volume. Identical converters can be used
both for the RX and TX branches. A low price and low dimensions are extraordinarily important for
the design and realization of the advanced simulator setups enabling a generation of a high number
of targets at more azimuths using arrays of target simulating modules (see Figure 2). Small mmW
units can be easily and densely placed in view of the tested sensor which enables complex radar
pictures corresponding to the practical radar pictures seen by AR sensors in real traffic to be
generated. Results of the phase noise analysis also enable the usage of relatively simple and
therefore low-cost TS local oscillators.

Figure 15. Fabricated

Figure 15. FabricatedmmW
mmWdown-down- andand up-converter
up-converter containing
containing a built-in
a built-in 14 dB14 dBantenna,
horn horn antenna,
MMIC
MMIC balanced mixer, planar filters and LO and IF connectors.
balanced mixer, planar filters and LO and IF connectors.

The developed mmW converter includes a built-in 14 dB horn antenna and subharmonically
pumped balanced mixer, and, together with an input RF filter reducing the mirror signals, IF filter
and LO and IF connectors, occupies only an 80 × 40 × 8 mm volume. Identical converters can be used
both for the RX and TX branches. A low price and low dimensions are extraordinarily important for
the design and realization of the advanced simulator setups enabling a generation of a high number
of targets at more azimuths using arrays of target simulating modules (see Figure 2). Small mmW
units can be easily and densely placed in view of the tested sensor which enables complex radar
Figure 15. Fabricated mmW down- and up-converter containing a built-in 14 dB horn antenna,
MMIC balanced mixer, planar filters and LO and IF connectors.
Sensors 2020, 20, 2714 20 of 22

pictures corresponding to the practical radar pictures seen by AR sensors in real traffic to be generated.
Results of the phase noise analysis also enable the usage of relatively simple and therefore low-cost TS
local oscillators.
The derived formulas are useful for system programming as well. In particular, for setting specific
sensor simulator parameters enabling the generation and measurement of specific testing sequences
and definite traffic scenarios. This concerns setting a suitable time delay, a Doppler frequency shift and
the amplitude of a signal transmitted back from the TS to the AR sensor.
The amplitude of the signal transmitted by the simulator back to the AR sensor should be set
according to Equation (4). If the gain (GS ) is set according to Equation (3), the sensor detects the target
at any range as a target with a constant RCS. This is an advantageous, but not necessarily mandatory,
setting. In near ranges, from the radar sensor point of view, even slightly lower signal levels are still
strong enough and ensure perfect detection. That is why, in this case, practically set values of GS can
be, without any difficulties, lower than those defined by Equation (3). Lower values of Gs also mean a
lower system gain is required and minor intermodulation problems, even in the case of M, in parallel
simulated targets can occur. An effective RCS value, corresponding to this modified setting, can be
evaluated using Equation (5) and the procedure can check if such an RCS still ensures a reliable target
detection. The time delay of the i-th artificial target should be set according to

2(Rti − RS )
τi = − τL (37)
c0

where τL describes the total latency of the simulator circuits, including the latency of both the analog
and digital circuits. The Doppler frequency shift corresponding to a target’s radial velocity can be
evaluated as
2 f0 vi → → 
fDi = v 0i ·Ψ0i (38)
c0
→
In this formula, vi describes the instantaneous velocity of the i-th simulated target, v 0i represents
→
its unity velocity vector, while Ψ0i stands for its unity azimuth vector defined by the AR sensor
simulator’s TX antenna’s interconnecting line. In the digital domain, the Doppler frequency shift can
be easily performed by multiplying the received and time-shifted signal by a complex exponential.

8. Conclusions
The employment of target simulators represents the most comprehensive way of how to test
modern AR sensors. In addition to mere functionality checking, it also enables the evaluation of
corresponding ECU reactions so crucial for passenger safety. Since advanced AR sensors offer high
target resolution, the matching testing simulator should offer the simulation of a high number of targets
at arbitrary ranges and azimuths, a feature still lacking in most available simulating devices.
This work indicates that these requirements cannot be met without a detailed theoretical analysis
of the AR sensor simulator setup in question. This explains why the presented article includes a set
of formulas derived for the description of all crucial system parameters. Especially detailed is the
analysis of phase noise behavior indicating the range and LO quality dependences. For the design
of the modern TSs, a description of the influences of this noise source for the generation of multiple
targets can be important. All equations were verified by computer simulations using an AWR VSS,
a professional CAD tool oriented to the analysis and design of RF systems. Agreements between
calculated results and CAD simulations are, in all cases, very good.
Even relatively simple commercially available target simulators enabling the generation of targets
at a single azimuth, with no azimuthal velocity component option, are exceptionally costly devices.
HW simplification, enabled by the results of this work, can reduce the price of such sensor simulator
setups by nearly an order and open the way for testing key automotive sensors, not only in the
laboratories of automotive research departments but also in standard automotive testing practices.
Sensors 2020, 20, 2714 21 of 22

The reduction in the complexity and price of the circuitry used in sensor simulator setups also
opens the way to solutions based on arrays of a large number of TS transmitters offering a generation
of high numbers of targets including movements in both radial and azimuthal directions. Such setups
make it possible to generate targets radically close to practical radar pictures seen by AR sensors in a
real operation. This should pave the way for enhanced quality in advanced automotive sensor testing.

Author Contributions: Conceptualization, P.H.; methodology, P.H. and V.A.; software, V.A.; validation, P.H. and
V.A.; formal analysis, V.A.; investigation, P.H. and V.A.; resources, P.H.; data curation, V.A.; writing—original
draft preparation, P.H.; writing—review and editing, P.H and V.A.; visualization, V.A.; supervision, P.H.; project
administration, P.H.; funding acquisition, P.H. All authors have read and agreed to the published version of
the manuscript.
Funding: This work was supported by the European Structural Funds Operational Program Research, Development
and Education, grant “Advanced Testing of Automotive Radars” No. CZ.02.1.01/0.0/0.0/16_025/0007318.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Dickmann, J.; Appenrodt, N.; Bloecher, H.-L.; Brenk, C.; Hackbarth, T.; Hahn, M.; Klappstein, J.;
Muntzinger, M.; Sailer, A. Radar contribution to highly automated driving. In Proceedings of the 11th
European Radar Conference, Rome, Italy, 8–10 October 2014. [CrossRef]
2. Dickmann, J.; Klappstein, J.; Hahn, M.; Appenrodt, N.; Bloecher, H.-L.; Werber, K.; Sailer, A. Automotive Radar
the Key Technology for Autonomous Driving: From Detection and Ranging to Environmental Understanding.
In Proceedings of the 2016 IEEE Radar Conference (RadarConf), Philadelphia, PA, USA, 2–6 May 2016.
[CrossRef]
3. Gerstmair, M.; Melzer, A.; Onic, A.; Huemer, M. On the Safe Road Toward Autonomous Driving: Phase
Noise Monitoring in Radar Sensors for Functional Safety Compliance. IEEE Signal Process. Mag. 2019, 36,
60–70. [CrossRef]
4. Brooker, G.M. Understanding Millimetre Wave FMCW Radars. In Proceedings of the IEEE 1st International
Conference on Sensing Technology, Palmerston North, New Zealand, 21–23 November 2005.
5. Winkler, V. Range Doppler Detection for Automotive FMCW Radars. In Proceedings of the 2007 European
Radar Conference, Munich, Germany, 10–12 October 2007. [CrossRef]
6. Hasch, J.; Topak, E.; Schnabel, R.; Zwick, T.; Weigel, R.; Waldschmidt, C. Millimeter-Wave Technology for
Automotive Radar Sensors in the 77 GHz Frequency Band. IEEE Trans. Microw. Theory Technol. 2012, 60,
845–860. [CrossRef]
7. Roos, F.; Bechter, J.; Knill, C.; Schweizer, B.; Waldschmidt, C. Radar Sensors for Autonomous Driving:
Modulation Schemes and Interference Mitigation. IEEE Microw. Mag. 2019, 20, 58–72. [CrossRef]
8. Bourdoux, A.; Parashar, K.; Baudui, M. Phenomenology of Mutual Interference of FMCW and PMCW
Automotive Radars. In Proceedings of the 2017 IEEE Radar Conference (RadarConf), Seattle, WA, USA,
8–12 May 2017. [CrossRef]
9. Andres, M.; Feil, P.; Menzel, W. 3D-Scattering Center Detection of Automotive Targets Using 77 GHz UWB
radar sensors. In Proceedings of the 6th European Conference on Antennas and Propagation (EuCAP),
Prague, Czech Republic, 26–30 March 2012. [CrossRef]
10. Suzuki, H.; Saito, K. Moving Target Using Dielectric Waveguide for Radar-Based Pre-Crash Safety Systems.
In Proceedings of the 2013 European Radar Conference, Nuremberg, Germany, 9–11 October 2013;
ISBN 978-2-87487-033-0.
11. Scheiblhofer, W.; Feger, R.; Haderer, A.; Stelzer, A. A Low-Cost Multi-Target Simulator for FMCW Radar
System Calibration and Testing. In Proceedings of the 2017 European Radar Conference (EuRAD), Nuremberg,
Germany, 11–13 October 2017. [CrossRef]
12. Scheiblhofer, W.; Feger, R.; Haderer, A.; Stelzer, A. Low-Cost Target Simulator for End-of-Line Tests of
24-GHz Radar Sensors. In Proceedings of the 22nd International Microwave and Radar Conference (MIKON),
Poznan, Poland, 14–17 May 2018. [CrossRef]
13. Iberle, J.; Mutschler, M.A.; Scharf, P.A.; Walter, T. A Radar Target Simulator Concept for Close-Range
Targets with Micro-Doppler Signatures. In Proceedings of the 12th German Microwave Conference (GeMiC),
Stuttgart, Germany, 25–27 March 2019. [CrossRef]
Sensors 2020, 20, 2714 22 of 22

14. Lutz, S.; Erhart, C.; Walte, T.; Weigel, R. Target Simulator Concept for Chirp Modulated 77 GHz Automotive
Radar Sensors. In Proceedings of the 11th European Radar Conference, Rome, Italy, 8–10 October 2014.
[CrossRef]
15. Engelhardt, M.; Pfeiffer, F.; Biebl, E. A High Bandwidth Radar Target Simulator for Automotive Radar
Sensors. In Proceedings of the 2016 European Radar Conference (EuRAD), London, UK, 5–7 October 2016;
ISBN 978-2-8748-7045-3.
16. Roszkowski, P. Carrier- and Doppler-tunable FPGA-based Active Reflector for Radar Calibration.
In Proceedings of the 17th International Radar Symposium (IRS), Krakow, Poland, 10–12 May 2016.
[CrossRef]
17. Gruber, A.; Gadringer, M.; Schreiber, H.; Amschl, D.; Bösch, W.; Metzner, S.; Pflügl, H. Highly Scalable Radar
Target Simulator for Autonomous Driving Test Beds. In Proceedings of the 2017 European Radar Conference
(EuRAD), Nuremberg, Germany, 11–13 October 2017. [CrossRef]
18. Heuel, S.; McCarthy, D. Real Time Radar Target Generation. Microw. J. 2015, 58, 92–106.
19. Radar Test Bench. Available online: www.dspace.com/en/pub/home/products/hw/test_benches/radar_test_
bench.cfm (accessed on 4 March 2020).
20. Russel, M.E.; Drubin, C.A. Automotive Forward Looking Sensor Test Station. European Patent EP 1034442B1,
3 June 1999.
21. Maples, V.H.; Eastman, G.A. Radar Scene Simulator. U.S. Patent No. 4,660,041, 21 April 1987.
22. Budge, M.C.; Burt, M.P. Range Correlation Effects in Radars. In Proceedings of the 1993 IEEE National Radar
Conference, Lynnfield, MA, USA, 20–22 April 1993. [CrossRef]
23. Stove, A.G. Linear FMCW Radar Techniques. IEE Proc. F Radar Signal Process 1992, 139, 343–350. [CrossRef]
24. Pozar, D.M. Microwave Engineering, 4th ed.; John Wiley & Sons, Inc.: New York, NY, USA, 2011;
ISBN 978-0-470-63155-3.
25. IEEE Standards Coordinating Committee 27. IEEE Standard Definitions of Physical Quantities for Fundamental
Frequency and Time Metrology–Random Instabilities; IEEE: New York, NY, USA, 2009. [CrossRef]
26. Siddiq, K.; Watson, R.J.; Pennock, S.R.; Avery, P.; Poulton, R.; Dakin-Norris, B. Phase noise analysis in
FMCW radar systems. In Proceedings of the 2015 European Radar Conference (EuRAD), Paris, France,
9–11 September 2015. [CrossRef]
27. Schlecht, E.; Mehdi, I. Effects of Local Oscillator Phase Noise on Submillimeter-Wave Spectrometer
Performance. In Proceedings of the 39th International Conference on Infrared, Millimeter, and Terahertz
waves (IRMMW-THz), Tucson, AZ, USA, 14–19 September 2014. [CrossRef]
28. Mid-Range Radar Sensor (MRR). Available online: www.bosch-mobility-solutions.com/en/products-and-
services/passenger-cars-and-light-commercial-vehicles/driver-assistance-systems/automatic-emergency-
braking/mid-range-radar-sensor-(mrr)/ (accessed on 4 March 2020).
29. HMC1058 71–86 GHz Sub-Harmonic Mixer. Available online: https://www.analog.com/en/products/hmc1058.
html (accessed on 4 March 2020).
30. Visual System Simulator. Available online: https://www.awr.com/software/products/visual-system-simulator
(accessed on 4 March 2020).

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).