3290 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 64, NO.

10, OCTOBER 2016

Impact of Frequency Ramp Nonlinearity, Phase

Noise, and SNR on FMCW Radar Accuracy
Serdal Ayhan, Steffen Scherr, Student Member, IEEE, Akanksha Bhutani, Student Member, IEEE,
Benjamin Fischbach, Mario Pauli, Member, IEEE, and Thomas Zwick, Senior Member, IEEE

Abstract— One of the main disturbances in a frequency- • high-frequency ramp linearity;

modulated continuous wave radar system for range measure- • low phase noise;
ment is nonlinearity in the frequency ramp. The intermediate • high signal-to-noise ratio (SNR).
frequency (IF) signal and consequently the target range accuracy
are dependent on the type of the nonlinearity present in the The influence of the above-mentioned factors on target
frequency ramp. Moreover, the type of frequency ramp nonlin- range estimation is analyzed, and for help in understanding
earity cannot be directly specified, which makes the problem even the novelty of the work presented in this paper, the following
more challenging. In this paper, the frequency ramp nonlinearity references about frequency ramp deviations are considered.
is investigated with the modified short-time Fourier transform The influence of a nonlinear frequency ramp with sinusoidal
method by using the short-time Chirp-Z transform method
with high accuracy. The random and periodic nonlinearities are deviation is described in [5] and the focus is given to moving
characterized and their sources are identified as phase noise and radar targets. The spectrum of sinusoidal nonlinearity is not
spurious. These types of frequency deviations are intentionally derived analytically, but the simulations correspond to the
increased, and their influence on the linearity and the IF-signal is spectrum derived in this paper. The spectrum is investigated
investigated. The dependence of target range estimation accuracy analytically in [6]. In this case, superposition of sinusoidal
on the frequency ramp nonlinearity, phase noise, spurious, and
signal-to-noise ratio in the IF-signal are described analytically nonlinearities is investigated using simulations. Analytical
and are verified on the basis of measurements. results are in agreement with simulation results. Frequency
deviations measured in this paper are smaller than those in the
Index Terms— Cramér–Rao lower bound (CRLB), frequency- simulations, as the output spectrum is strongly influenced by
modulated continuous wave (FMCW) radar, frequency ramp
linearity, phase noise, range measurement, signal-to-noise smaller amplitude deviations, i.e., smaller modulation index.
ratio (SNR). Therefore, the determined influence on range measurement is
smaller than expected (see Section IV).
The approach presented in this paper for describing fre-
I. I NTRODUCTION quency ramp nonlinearities in the intermediate frequency (IF)-
signal can be applied to any frequency ramp deviation in an
T HE usage of radar-based sensors is steadily increasing
in several application areas, which were dominated by
alternative technologies in the recent past. The strong presence
FMCW radar, e.g., square deviation in a free running voltage-
controlled oscillator (VCO). The square deviation is used in
of radar technology in the automotive sector has led to a a similar way in [7] and verified by simulations. In contrast,
tremendous increase in demand for these sensors, causing this paper allows verification with real measurements.
significant reduction in the unit costs [1]. Therefore, radar In [8], frequency deviations are used to compare lineariza-
proves to be an attractive alternative technique for range mea- tion techniques using different hardware topologies. The deter-
surement, especially in industrial applications, in comparison mined frequency ramp phase deviations are compared with
to the generally used optical sensors with high accuracy. The their corresponding theoretical counterparts. An additional
main advantage of radar in comparison to its optical counter- linearization approach is presented in [9]. Nonlinearities based
part is its robustness in a harsh measurement environment, as on polynomial approximation are defined using a closed-
presented in [2]–[4]. form term and a linearization method for rectifying short-term
A frequency-modulated continuous wave (FMCW) radar frequency deviations in a different FMCW radar application
can provide the required high range estimation accuracy, is described.
if the following aspects are considered: In comparison with the state of the art, the methods pre-
sented in this paper clearly describe the frequency ramp non-
Manuscript received January 17, 2016; revised May 6, 2016 and linearities in an FMCW radar and establish their influence on
August 3, 2016; accepted August 4, 2016. Date of publication August 30, range measurement accuracy through mathematical modeling,
2016; date of current version October 4, 2016. This work was supported by
the German Research Foundation (DFG), Germany, under the Context of the analytical study and verification based on real measurement
Collaborative Research Program SPP1476. results. In Section II, the measurement principle and measure-
The authors are with the Institute of Radio Frequency Engineering and ment results for the frequency ramp linearity are shown. The
Electronics, Karlsruhe Institute of Technology, 76131 Karlsruhe, Germany
(e-mail: [email protected]). random and periodic frequency deviations in the frequency
Digital Object Identifier 10.1109/TMTT.2016.2599165 ramp are described mathematically in order to obtain an
0018-9480 © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on June 19,2023 at 12:23:39 UTC from IEEE Xplore. Restrictions apply.
AYHAN et al.: IMPACT OF FREQUENCY RAMP NONLINEARITY, PHASE NOISE, AND SNR ON FMCW RADAR ACCURACY 3291

analytical expression of their influence on the measured IF

signal and the corresponding range accuracy. These deviations
are intentionally increased so that their specific influence can
be analyzed using real measurements. Section III describes the
correlation between random frequency deviation, phase noise,
SNR, and achievable accuracy. In Section IV, the influence
of periodic frequency deviation on the IF-signal and the
corresponding range estimation accuracy is examined, and the
frequency ramp linearity requirements are defined. The last
section concludes this paper.
The novelty of this work can be summarized as follows.
First, it provides a clear description of the relationship between
ramp nonlinearities, SNR, and phase noise. Second, the influ-
ence of these factors on FMCW radar range measurement
accuracy is established. Ultimately, the analytical study carried
out in this paper helps in better system design.

II. L INEARITY A NALYSIS OF THE F REQUENCY R AMP

Evaluation of the IF-signal of an FMCW radar system Fig. 1. Block diagram of the FMCW radar sensor and the measurement
setup (PFD = phase frequency detector, LP = low pass filter, BP = band
is dependent on the frequency ramp linearity. For highly pass IF-filter).
accurate range measurements, this requirement should be met
over the complete measurement duration. A signal is defined
either in time or frequency domain. To analyze ramp linearity,
time/frequency representation (TFR) is required to illustrate
deviations from the expected behavior. In [10]–[12], various
TFR methods are described. In this paper the conventional
short-time Fourier transform approach is used [13] that is
modified for the FMCW radar application by using the short-
time Chirp-Z transform (SCZT) method to reach a higher
frequency estimation accuracy [14], [15].

Fig. 2. Photograph of the measurement setup and the system components.

A. Measurement Setup
A phase-locked loop (PLL)-controlled monostatic K-Band
FMCW radar sensor is used to analyze the frequency ramp The block diagram of the measurement setup for the
linearity. The radar sensor mainly consists of an Integer-PLL SCZT-evaluation and the used components are shown in
chip and radio-frequency (RF) components. The Integer-PLL Figs. 1 and 2. The FMCW radar signal fRF is directly
contains a VCO, and the RF components include a hybrid connected to an oscilloscope and sampled at a sampling rate
coupler, a mixer, and an antenna (see Fig. 1). The VCO of 80 GSa/s. The bandwidth of the oscilloscope is 33 GHz
is controlled by the PLL in order to generate a triangular- (Infiniium 90 000 X-Series), which is sufficient for the task at
modulated FMCW radar signal [16]. The system parameters hand.
are as follows. The SCZT-based TFR evaluation is carried out offline using
• Bandwidth B = 800 MHz. MATLAB. The sampled signal is divided into short time
• Ramp duration T = 500 μs. segments. For each time segment, windowing is performed,
• Start frequency f start = 23.7 GHz. followed by subsequent transformation into the frequency
• PLL division factor N = 1536 (Integer-PLL division domain. The frequency applied in each time segment can be
factor 96 · VCO division factor 16). determined using maximum search in the frequency domain.
• Direct digital synthesizer (DDS) start frequency Windowing of the signal indirectly leads to either a high time
f start,DDS ≈ 15.6 MHz. resolution or a high frequency resolution. An appropriate win-
The reference signal f Ref , required for the Integer-PLL, is dow length is selected based on the given FMCW parameters
generated digitally using a DDS, which supplies a phase incre- and the frequency ramp deviations are determined by building
ment for frequency modulation at a sampling rate of 10 ns. The the difference between the measured, evaluated ramp and the
DDS is controlled by a field programmable gate array (FPGA) theoretical, ideal ramp.
and is presented in detail in [16]. The concept used for The influence of the ramp nonlinearity on the IF-signal is
signal synthesis enables arbitrary signal forms to be generated, investigated by using an identical setup. In this case, instead
i.e., besides ideal frequency ramp, ramps with deliberately of connecting the RF-signal to the oscilloscope, a radar target
added disturbances can also be realized through DDS control. is used. The FPGA controls the synchronized sampling of the

Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on June 19,2023 at 12:23:39 UTC from IEEE Xplore. Restrictions apply.
3292 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 64, NO. 10, OCTOBER 2016

Fig. 5. TFR of the ideal, undisturbed, and disturbed measured frequency

ramp.

Fig. 3. Frequency deviations in the first 10 μs of the measured frequency

ramp with different time windows.

Fig. 6. Disturbance model for normally distributed noise on the frequency

ramp.

in digital circuits or transient response of the PLL.

Since the spurious corresponds to a certain frequency,
it results in a sinusoidal nonlinearity in the frequency
ramp. This sinusoidal nonlinearity is characterized by
a certain frequency deviation fe (in Hz) and deviation
amplitude Ae (in Hz). Fig. 4 shows that the FFT of
Fig. 4. FFT of the measured frequency deviation. the frequency deviations have discrete maxima, which
ranges from 350 kHz to 450 kHz.
bandpass (BP) filtered IF-signal with an external analog digital Based on the above-determined frequency deviations, which
converter (ADC) at the beginning of the FMCW ramp. The are caused by disturbances like noise and spurious, the real fre-
data are again evaluated offline. quency ramp f real (t) with the instantaneous frequency f ideal (t)
and the disturbance e(t) can be described as
B. Measurement Results B
f real (t) = t + f min + e(t) (1)
The frequency deviations  f FMCW evaluated by the SCZT T 
   disturbance
with different time windows are illustrated in Fig. 3 for the ideal
first 10 μs of the frequency ramp. Frequency deviations are with
smoothed by windowing and averaging effect of the SCZT,
if a 200 ns long time window is used. The 20 ns time window e(t) = 2π Ae sin (2π fe t) + An (t) . (2)
along with doubled DDS phase increment rate ensures precise      
periodic random noise
determination of the deviations. The detailed distribution of
the deviations are determined by an FFT (see Fig. 4), such For a more detailed analysis, the impact of these two types
that the frequency ramp deviations can be characterized as of disturbances on the IF-signal spectrum is determined. For
follows. this purpose, the deviations are described mathematically in
• Random deviations in the frequency ramp caused by order to obtain an analytical expression of the frequency error
noise effects cannot be classified directly as nonlinearity. in IF estimation. Then, the corresponding deviations are inten-
These deviations can be termed as random frequency tionally increased by using the DDS and real measurements
ramp variations An (in Hz), caused by increment in are performed. The deviations are investigated separately to
phase noise. The FFT (see Fig. 4) of the frequency devi- analyze their specific influence. When increasing Ae as a
ations clearly show that the deviations are distributed function of f e , the periodic, sinusoidal deviation predominates.
over a wide frequency range and may be described as By increasing An , random frequency deviations can be
randomly distributed noise. enhanced. Fig. 5 shows these parameters on the undisturbed
• Periodic deviations in the frequency ramp caused by and disturbed measured frequency ramp for illustration pur-
spurious. These deviations may result from switching poses, compared with the theoretical, ideal frequency ramp.

Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on June 19,2023 at 12:23:39 UTC from IEEE Xplore. Restrictions apply.
AYHAN et al.: IMPACT OF FREQUENCY RAMP NONLINEARITY, PHASE NOISE, AND SNR ON FMCW RADAR ACCURACY 3293

Fig. 7. Measured frequency deviation in the first 100 μs of the frequency Fig. 9. FFT of the measured frequency deviation with increased noise
ramp with increased noise for A3σ 1 = 5 MHz. for A3σ 1 = 5 MHz.

Fig. 8. Measured frequency deviation in the first 100 μs of the frequency Fig. 10. FFT of the measured frequency deviation with increased noise
ramp with increased noise for A3σ 1 = 10 MHz. for A3σ 1 = 10 MHz.

III. R ANDOM F REQUENCY D EVIATIONS from the SNR of the IF-signal, which will be described as
IN THE F REQUENCY R AMP follows.
Random frequency deviations An (t) in the frequency ramp
are modeled as a normally distributed noise with an expec- A. Methods for Estimating the SNR of the IF-Signal
tation value of zero and a 3σ -standard deviation of A3σ for a Monostatic FMCW Radar
(see Fig. 6). 1) SNR Caused by Phase Noise: The basic parameter
The frequency deviations of the disturbed frequency describing phase noise is the spectral power density of phase
ramp of A3σ 1 = 5 MHz and A3σ 2 = 10 MHz are plotted in variations, which is given by the squared effective phase
Figs. 7 and 8 over the first 100 μs of the frequency ramp. variation relative to the measurement bandwidth [17]. Integra-
A3σ 1 and A3σ 2 are compared with initial undisturbed fre- tion over the spectral power density of the phase noise may
quency ramp given by A3σ 0 = 0 MHz. The effective deviation also be interpreted as noise power with respect to the carrier
amplitude Ae of A3σ 0 is 1.3 MHz, and it arises due to the signal [18]. This allows to describe phase noise at the output
superposition of a periodic disturbance around 400 kHz and of the mixer by means of the SNR definition, with the mixing
random noise of the frequency ramp, caused by nonideal process being taken into account. The SNR η is derived for
system. The normally distributed disturbance A3σ , generated the FMCW radar in [19] and can be given as follows:
by controlling the DDS output, is superposed with the existing 1
disturbance. The FFT of the frequency deviations shows the η =  BIF PIF ( f m ,R)
. (3)
resulting increase of the deviations as a function of A3σ over 2· 0 10 10 d fm
the complete bandwidth (see Figs. 9 and 10). The single side band (SSB) phase noise spectrum
Random frequency deviations in the frequency ramp may PIF ( fm , R) in the IF-signal is composed of the SSB
be interpreted as an influence on the phase of the radar phase noise spectrum PRF ( f m ) of the RF signal at the
signal. This results in a direct relationship between phase frequency offset fm and an additional term that acts like a
noise and random frequency deviations in the frequency ramp. high-pass filter in the frequency range, which depends on
Therefore, a criterion can be derived for the impact of random the range R [19]. When using an FMCW radar for range
frequency deviations on range measurements using the phase measurement, this effect is caused by the coherence between
noise. For this purpose, it is necessary to determine the the transmitted and the received signal. Noise sources at the
influence of phase noise on the IF-signal. This can be obtained mixer output are correlated, such that phase noise is partly

Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on June 19,2023 at 12:23:39 UTC from IEEE Xplore. Restrictions apply.
3294 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 64, NO. 10, OCTOBER 2016

compensated at short distances [20], which minimizes its

influence on the IF-signal. The remaining fraction causes the
SNR to deteriorate. For evaluation of the influence on the
BP-filtered IF-signal, the range within the known IF-filter
bandwidth BIF from 8 kHz to 100 kHz of the FMCW-radar
is of interest, which defines the integration bandwidth in (3).
2) SNR of the IF-Signal: The expression in (3) describes
the SNR of the IF-signal as a function of phase noise.
No additional noise impacts are considered. For this reason,
it is required to determine the total SNR for verification.
Therefore, the spectrum of the IF-signal is evaluated and
variance of the noise PR in the spectrum is calculated with
respect to the maximum amplitude PS over BIF . The dis- Fig. 11. Measured SSB phase noise with increased noise.
tance to the radar target, i.e., the frequency of the IF-signal
can be chosen arbitrary. Multiple or additional reflections TABLE I
in the range of BIF may distort the result. To exclusively SSB P HASE N OISE AND SNR E STIMATION OF THE
obtain noise with a random power density spectrum, the IF-S IGNAL AND D UE TO P HASE N OISE
noise PR outside of BIF with identical bandwidth is chosen.
Based on the known filter characteristics, amplitude differ-
ences PFilter (Bnoise) are compensated over the selected noise
bandwidth Bnoise . When calculating the SNR outside of BIF ,
the noise fraction PSSB (Bnoise, R) caused by phase noise has
to be considered as well. The spectrum of the IF-signal is
calculated using FFT and a Blackman window. When calcu-
lating the radar SNR in time domain for comparison with the
Cramér–Rao lower bound (CRLB) (see Section III-C), the
processing gain Fgain = 10 log(Ns /2) with Ns sampled
points [21] and amplitude difference by the window function
Wloss have to be considered [22]. The SNR ηIF in dB can be
estimated as
ηIF = PS − PR (Bnoise) − Fgain + Wloss
+PFilter (Bnoise) − PSSB (Bnoise, R). (4)
Now, the estimated SNR η according to (3) caused by phase
noise can be compared with the estimated SNR ηIF in the
spectrum of the IF-signal. Fig. 12. Measured IF-signal in the frequency domain with averaging.
3) Quantization Noise of the ADC: Quantization noise of
the ADC may also limit the maximum achievable SNR.
In the present application, the voltage range of the 16-b 40 MHz is plotted for the undisturbed and disturbed output sig-
ADC is set to ±1 V. The IF-signal reaches maximum values nals of the radar in Fig. 11. The phase noise values determined
of about ±250 mV. The ADC is not driven to full output. at an offset of 100 kHz and the SNR η calculated using
Effective accuracy is reduced by (3) are listed in Table I. The deviations cause the expected
  deterioration of phase noise and of the SNR of the IF-signal.
1V
log2 = 2 bit. (5) For A3σ 0 , an SNR η of 32 dB is obtained. The SNR
250 mV
is reduced by 11.2 dB and 14.4 dB for A3σ 1 and A3σ 2 ,
The maximum achievable SNR of the ADC with an effective respectively. This directly explains the relationship of ran-
bit width of 14 is calculated to be 86 dB according to [23]. dom frequency deviations, phase noise, and the SNR of the
IF-signal. Additionally, an impact of the ADC can also be
B. Evaluation of the SNR in the IF-Signal Due to excluded, as the SNR of the IF-signal ηIF is far below the
Random Frequency Deviations by Using Phase estimated quantization noise of 86 dB (see Section III-A3).
Noise Measurements For verification, the total SNR ηIF in the spectrum of the
To establish the relationship of random frequency deviations IF-signal is determined for an arbitrary radar target distance
in the frequency ramp and phase noise, a monotone output of 0.45 m and the same noise values A3σ 1,2 . The spectrum
signal of the radar at 24 GHz is disturbed with normally dis- of the IF-signal is plotted in Fig. 12 for the averaging factor
tributed noise, which was also used for the analysis of the fre- AFn with n = 1 (no measurement repetition) and n = 100
quency ramp linearity. Next, the phase noise is determined and (measurement repetitions). It is clearly evident from the spec-
the SNR is evaluated with the method in Section III-A. As an trum that the noise is reduced by averaging in the frequency
example, the SSB phase noise measured at a bandwidth of range from 200 to 300 kHz. At about 400 kHz, the influence

Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on June 19,2023 at 12:23:39 UTC from IEEE Xplore. Restrictions apply.
AYHAN et al.: IMPACT OF FREQUENCY RAMP NONLINEARITY, PHASE NOISE, AND SNR ON FMCW RADAR ACCURACY 3295

TABLE II TABLE III

PARAMETER FOR SNR-E STIMATION OF THE M EASURED C OMPARISON OF THE M EASURED S TANDARD D EVIATION σ R W ITH THE
IF-S IGNAL IN THE F REQUENCY D OMAIN T HEORETICAL CRLB S TANDARD D EVIATION σCRLB FOR R ANGE
M EASUREMENT D EPENDENT ON THE SNR

of the existing periodic deviations of the frequency ramp can

be seen. For this reason, the noise bandwidth Bnoise is set
between 208 and 300 kHz to have the same bandwidth as the
IF-filter bandwidth BIF . The SNR values determined for the
undisturbed ramp and the disturbed ramps are listed in Table I.
In the undisturbed case, an SNR ηIF of about 30.5 dB was
observed without averaging using (4) with the parameters ranges between 1.27 and 1.65. The σ R -values determined with
of Table II. respect to the CRLB hardly differ from each other.
Averaging in time domain for AF100 gives ηIF of 49.5 dB. Apart from this absolute relationship to the CRLB, the
The expected improvement in SNR as a function of the number relative differences of the SNR values and standard devia-
of measurement repetitions can be represented as 10 log(AFn ). tions of the range estimation are compared for verification.
With AF100 , the deviation from the theoretical value is 1 dB. To this end, the measured standard deviations and theo-
Averaging shows that no fixed disturbance frequencies are retical CRLB standard deviations are related to the stan-
found within the chosen noise bandwidth. For A3σ 1 , a ηIF1 dard deviation with the lowest SNR value at the highest
of about 27.9 dB was observed without averaging. For disturbance A3σ 2 = 10 MHz. Depending on the SNR,
n = 100 averaged measurements, the distance error is a theoretically improvement by (σCRLB,10 MHz /σCRLB,5 MHz ) =
improved by about 20 dB. This influence may also be deemed 1.89 results for the CRLB with the smaller disturbance
noise in this case. With increasing disturbance A3σ 2 , the SNR A3σ 1 = 5 MHz. For the undisturbed frequency ramp, this fac-
is reduced to ηIF2 = 23.6 dB. Compared with A3σ 0 , the SNR tor increases to (σCRLB,10 MHz /σCRLB,0MHz ) = 5.13. The factor
is reduced by 2.6 and 6.9 dB for A3σ 1 and A3σ 2 , respectively. (σR,10 MHz /σ R ) calculated from the measurements increases
According to the measurement results obtained with addi- from 1.45 to 4.42 with increasing SNR. Under the given noise
tional randomly disturbed noise, the relative SNR differences conditions, the absolute deviation of the standard range error
for η and ηIF are in good agreement, as the range outside from the CRLB and the relative deviation of the range errors
the IF-filter bandwidth BIF is influenced due to the deteriorated from each other are in agreement, such that the determined
phase noise. The absolute values differ, as the increased SNR values can be confirmed. In addition, the measurement
noise fraction resulting from phase noise within BIF is not results show that the K-band radar used here with a relative
considered when estimating the SNR in the spectrum of the low bandwidth of 800 MHz and low SNR of approximately
IF-signal. Consequently, phase noise is the parameter limiting 30 dB delivers a standard range error of 112 μm. According
the SNR, which is directly given by the SNR η . to the CRLB, a standard range error of 79 μm is possible [14].

C. Evaluation of the SNR Influence on Range IV. N ONLINEARITIES IN THE F REQUENCY R AMP
Estimation Using the CRLB From the frequency deviations determined in the frequency
Having determined the SNR, the CRLB allows us to ramp, the real frequency ramp with the instantaneous fre-
describe the influence of the SNR on the range estimation quency f real (t) is obtained by (1). To describe the influence of
as a function of the IF-signal. This bound can be applied the nonlinear frequency ramp on the IF-signal, the following
for error estimation of FMCW radar systems, if the signal to signal model is established for simplification. The amplitudes
be evaluated is a single sine oscillation with white, Gaussian and random frequency deviations are neglected. The emitted
noise [24]. Under real conditions, other limits, such as multiple frequency ramp sTx (t) is given as
reflections, nonlinearities, and nonwhite noise, occur. The sTx (t) = cos(ψ(t)). (6)
CRLB has to be understood as a reference for the achievable
accuracy at a certain SNR only. For the given radar, this bound The phase ψ(t) contains the frequency ramp and is given by
is derived as standard deviation σCRLB of the range evaluation the instantaneous frequency f Tx (t)
in [14].
For comparison with the CRLB, the standard deviation ψ(t) = 2π f Tx (t) dt (7)
of frequency and of the distance σ R , is determined over
1000 measurement values for a target distance of 0.45 m. The with
differences of the individual SNR conditions of the radar are B
t + f min .
f Tx (t) = (8)
summarized in Table III. They are calculated using the CRLB T
and determined by measurements. The ratio (σ R /σCRLB ) Upon ideal mixing with the delayed signal sTx (t−τ ) and ideal
is 1 in the ideal case. For the different SNR values, this factor filtering using the propagation time τ of the received signal,

Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on June 19,2023 at 12:23:39 UTC from IEEE Xplore. Restrictions apply.
3296 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 64, NO. 10, OCTOBER 2016

Fig. 13. Disturbance model for periodic nonlinearity on the frequency ramp.

the IF-signal sIF (t) results as

sIF (t) = cos(ψ(t) − ψ(t − τ )). (9)
Using the standard definition of a difference quotient, this Fig. 14. Amplitude of the carrier J0 and the side bands J1 and J2 dependent
on the sinusoidal deviation amplitude Ae on the frequency ramp.
term can be approximated as a derivative, if small propagation
time τ in the meter range (as given for radar sensor applica-
tions) is assumed However, this term can be compared with frequency modula-

dψ(t) tion [18] and can be described with the help of the Bessel
sIF (t) ≈ cos τ . (10)
dt function of first type Jn (z). First, the nonlinear expression is
expressed by the Bessel function and a series expansion
Equation (10) represents an expression for the frequency and
phase of the IF-signal. Inserting (7) into (10) results in ∞

e j z sin θ = Jn (z)e j nθ . (15)
sIF (t) = cos(2π fTx (t)τ ). (11)
   n=−∞
φIF (t ) The IF-signal is given by substitution of θ = 2π f e t = ωe t
The signal model used for the description of the IF-signal and z = 2π Ae τ by
shows that the instantaneous phase φIF (t) of the IF-signal is ∞

composed of the instantaneous frequency f Tx (t) of the ramp sIF (t) = e j φIF (t )
· Jn (2π Ae τ ) e j nωe t . (16)
and multiplication by the propagation time τ of the radar n=−∞
signal. This relationship can be used for investigating the
The FFT of this expression is calculated to get the spectrum
impact of nonlinearity on the IF-signal, as any disturbance
of the IF-signal
e(t) in the frequency ramp is now defined in the phase of the
IF-signal as well. ∞

FFT{sIF (t, τ )} = Jn (2π Ae τ )Sk (ω − nωe ). (17)
n=−∞
A. Sinusoidal Nonlinearity
Sk (ω) is the spectrum of the undisturbed IF-signal sIF (t) that
To describe periodic nonlinearity in the frequency ramp,
ideally has a maximum (±(B/T )τ ). The spectrum Sk (ω) is
e(t) in (1) is assumed to be sinusoidal (see Fig. 13) with the
additionally described by the Bessel function for the sinusoidal
effective deviation amplitude Ae and deviation frequency f e
nonlinearity in the frequency ramp at the frequencies ±nωe .
e(t) = Ae sin(2π f e t). (12) Their amplitudes correspond to the Bessel function value
Jn (2π Ae τ ). Above all, the maximum in the spectrum of
With (1), (11), and (12), the disturbed IF-signal sIF (t)
IF-signal is relevant for analysis. For this reason, the IF f IF
results as
⎛ ⎞ is referred to as the carrier and the replicates of f IF around
the frequencies ±nωe are designated side bands. Only the
⎜ B ⎟
sIF (t) = cos ⎜
⎝2π T tτ + 2π fmin τ + 2π Ae sin(2π f e t)τ ⎟ . upper side bands will be considered below, as the negative
  ⎠ frequencies in the spectrum are neglected.
   disturbance 2πe(t )τ
φIF (t )
(13) B. Influence of the Deviation Amplitude Ae
The signal contains the ideal instantaneous phase φIF (t) and Fig. 14 shows the amplitude J0 of the carrier and the
the additional disturbance e(t) resulting from nonlinearity. amplitudes J1 and J2 of the first two upper side bands. They
To estimate the influence on the IF-signal, the spectrum of the depend on the sinusoidal frequency deviation in the frequency
IF-signal has to be analyzed. For simplification, the IF-signal ramp with amplitude Ae at a target distance of 0.45 m.
is written in complex form as It can be observed that at an amplitude Ae = 17.9 MHz,
the amplitude of the carrier J0 is decreased by about 3 dB.
sIF (t) = e j φIF (t,τ ) · e j 2π Ae τ sin(2π f e t ) . (14)
According to the CRLB, variance is increased and accuracy
The undisturbed IF-signal is multiplied by the term of range measurement reduced. At an amplitude of 22.8 MHz,
e j 2π Ae τ sin(2π f e t ) . This term can be expressed nonlinearly the carrier amplitude J0 is more strongly attenuated than that
as e j z sin θ , such that the spectrum cannot be calculated directly. of the first side band J1 . Therefore, range evaluation is no

Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on June 19,2023 at 12:23:39 UTC from IEEE Xplore. Restrictions apply.
AYHAN et al.: IMPACT OF FREQUENCY RAMP NONLINEARITY, PHASE NOISE, AND SNR ON FMCW RADAR ACCURACY 3297

Fig. 16. Measured frequency deviation with sinusoidal disturbance in the

frequency ramp using f e = 200 kHz and Ae = 2.3 MHz.
Fig. 15. Schematic picture for description of the sinusoidal disturbance
influence on the IF-signal for two different sinusoidal deviation amplitudes Ae
on the frequency ramp.

TABLE IV
A MPLITUDE OF THE C ARRIER AND THE S IDE BANDS FOR A S INUSOIDAL
D ISTURBANCE D EPENDENT ON THE D EVIATION A MPLITUDE Ae

longer possible. The side bands with indices larger than n = 2

can be neglected due to the large frequency offset. Fig. 17. Measured frequency deviation with sinusoidal disturbance in the
frequency ramp using f e = 200 kHz and Ae = 23 MHz.
The frequency deviations in the additional disturbed fre-
quency ramp by the DDS control with deviation frequency
f e = 200 kHz and deviation amplitude Ae = 2.3 MHz
C. Influence of Deviation Frequency f e
in (12) are shown in Fig. 16. The chosen deviation ampli-
tude Ae is 10 times smaller than the maximum allowed It is evident from the investigation that apart from the
disturbance for sinusoidal nonlinearity in the frequency ramp. deviation amplitude Ae in the frequency ramp, the deviation
The chosen deviation frequency f e in the frequency ramp frequency f e plays an important role in evaluation as well.
is located in a noise-dominated frequency range of the At frequencies larger than 100 kHz, which corresponds to the
IF-signal (see Section III). The influence of nonlinearity can upper cutoff frequency of the IF-filter in the radar, the first
be illustrated exclusively for the selected values Ae and fe . side band has to be assumed to have a small influence due to
The measured frequency deviation in the frequency ramp is the large frequency offset from the carrier and therefore f IF .
Ae,meas = 2.8 MHz. The difference from the set value is Taking into account these boundary conditions, the deviation
attributed to the already nonideal frequency ramp without frequency f e is varied in steps of 0.5 kHz over the IF-filter
any additional disturbance. For the determined amplitude bandwidth and the distance error of frequency evaluation in
Ae,meas , the expected amplitude of the carrier in the IF-signal an FMCW radar system is measured as a function of Ae for
is −0.1 dB, that of the first side band is −19.1 dB three deviation amplitudes (1 MHz, 5 MHz, 10 MHz). The
(see Table IV). Fig. 18 shows the disturbed and the undisturbed result is shown in Fig. 20. For a distance of 0.45 m used
IF-signal in the frequency domain. The maximum at about in the analysis, the IF is about f IF = 21.5 kHz. The curve
22 kHz is the IF fIF corresponding to the chosen distance. The of the distance error shows increased, sinusoidal errors up to
first upper side band at f e = 200 kHz is clearly visible. The about f e < 5 kHz. Depending on Ae , they cause errors of up
relative power difference between f IF and f e + f IF is 22.4 dB to 50 mm.
and deviates from the expected value of 19 dB by about 3 dB The same error behavior is encountered symmetrically
only. The above described effect is illustrated schematically around f e = 43 kHz, which corresponds to twice f IF .
in Fig. 15. In this case, an approximately linear relationship between the
With the maximum allowed amplitude of Ae = 23 MHz resulting error and the increasing amplitude Ae is obtained.
the measured frequency deviation is Ae,meas = 21.6 MHz. The maximum error for the amplitudes of 1, 5, and 10 MHz is
In Figs. 17 and 19, the corresponding IF-signal is shown in about 4, 20, and 40 mm, respectively. In addition, frequencies
time and frequency domain. Due to strong attenuation of the from f e > 70 kHz are not found to cause any deterioration of
carrier, fIF cannot be detected in this case. This test case is the measurement results irrespective of the deviation ampli-
illustrated in Fig. 15. tude Ae .

Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on June 19,2023 at 12:23:39 UTC from IEEE Xplore. Restrictions apply.
3298 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 64, NO. 10, OCTOBER 2016

Fig. 18. Measured IF-signal in the frequency domain with sinusoidal

frequency deviation in the frequency ramp using f e = 200 kHz and
Ae = 2.3 MHz.

Fig. 21. Schematic for description of the convolution between the original
spectrum and the spectrum in the first side band.

considered that the spectrum in the side band also contains

the image frequencies. At a frequency of f e < 5 kHz, the
sinusoidal distance error is generated as a function of f e , as the
spectrum contains several maxima and minima that more or
less influence the IF fIF in case of a convolution of the spectra.
A deviation frequency around f e = 2 f IF ± 2.5 kHz causes
Fig. 19. Measured IF-signal in the frequency domain with sinusoidal the image frequencies to be located near f IF in the spectrum
frequency deviation in the frequency ramp using f e = 200 kHz and of the first side band. The deviation frequency f e = 2 f IF at
Ae = 23 MHz. the exactly doubled IF causes no or minimum errors only,
as the negative part of the spectrum in the first side band
exactly lies on the positive part of the original spectrum.
At about f e = 2 f IF ± 2.5 kHz, a maximum distance error
results from the convolution of the spectra. Identical distance
error behaviors dependent on f e can be found for other
distances.

D. Sources of Sinusoidal Nonlinearity

Sinusoidal nonlinearity on the basis of a real interference
scenario, which exists in DDS-based signal generation, can be
modeled. Apart from typical error sources, such as spurious
caused by the DDS as a function of the generated frequency,
spurious in the supply voltage or in the DDS-clock also leads
to spurious in the DDS output signal. These spurs usually have
a fixed frequency with a certain amplitude, which modulate
Fig. 20. Range evaluation using the measured IF-signal frequency dependent
on the frequency deviation f e .
the frequency of the DDS output signal. Analogous to the
sinusoidal nonlinearity in the frequency ramp, this spurious in
the DDS output signal can be described by the Bessel functions
as follows:
The error behavior determined can be explained by the 
∞
influence of the first upper side band. Sinusoidal nonlinearity j 2π f DDS,M t Ae,DDS
sDDS (t) = e · Jn e j n2π f e,DDSt .
does generate not only fIF but also the complete original f e,DDS
n=−∞
spectrum of the IF-signal (see Fig. 21) in the side bands ±nωe . (18)
The specific error behavior resulting from the variation of
f e may be interpreted as convolution between the original The first term gives the DDS output signal at a set frequency
spectrum and the identical spectrum in the first side band, f DDS,M . The second term describes the modulation at the DDS
with the amplitudes differing according to (17). It has to be output by the Bessel function with the spurious frequency

Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on June 19,2023 at 12:23:39 UTC from IEEE Xplore. Restrictions apply.
AYHAN et al.: IMPACT OF FREQUENCY RAMP NONLINEARITY, PHASE NOISE, AND SNR ON FMCW RADAR ACCURACY 3299

frequency conversion by measurements on an FMCW radar,

it can also be applied to other PLL or DDS signal generation
solutions.
In the existing system, the DDS generates spurious with
−62 dBc difference from the carrier at a frequency offset of
400 kHz ± 50 kHz (see Fig. 22). As obvious from the linearity
studies, the spurious causes an effective frequency deviation
of about 1.3 MHz in the frequency ramp (see Section III).
The frequency deviation Ae in the frequency ramp caused
by this spurious with an power amplitude (Ae,DDS / f e,DDS ) =
−62 dBc is shown in Fig. 23. The theoretical curve illustrates
that the expected deviation of Ae = 1.4 MHz is very
Fig. 22. DDS output signal dependent on set radar output signal close to the measured deviation of 1.3 MHz (see Fig. 7).
f RF = f DDS · 1536. The small difference is caused by the selected spurious fre-
quency f e of 400 kHz, as spurious is distributed in a range
of ±50 kHz for the real DDS output.

V. C ONCLUSION
The effects of frequency deviations from an ideal frequency
ramp were comprehensively studied, mathematically analyzed,
and thoroughly verified using real measurements. On the same
basis, it was established that phase noise and spurious in the
reference signal result in frequency deviations, which tend
to have a significant impact on radar range measurement.
Thereby, using the methods presented here design limits for
nonlinearity of frequency ramp, phase noise, and spurious can
be defined dependent on a range accuracy target.
Fig. 23. Theoretical frequency deviation Ae with spurious at the
DDS output dependent on the sideband amplitude ( Ae,DDS / f e,DDS ) and
The negative influence of phase noise on the ideal frequency
spurious frequency f e,DDS . ramp was analyzed and verified as random frequency devi-
ations in the frequency ramp, which leads to an increased
noise floor in the IF-signal. Furthermore, several methods were
f e,DDS and spurious amplitude Ae,DDS . The spectrum of presented to determine the SNR of a monostatic radar, which
sDDS (t) results as in turn helped in estimating the influence of SNR on range
∞  estimation accuracy. This was verified in theory using the
Ae,DDS Cramér-Rao bound. These methods can also be transferred to
FFT{sDDS (t)} = Jn Sk ( f − n · f e,DDS ).
n=−∞
fe,DDS other systems.
(19) It was shown that using an SNR of approximately 30 dB
leads to an accuracy of 112 μm in practice, which is higher
The spectrum consists of the carrier and the side bands that are than the ideal CRLB value of 79 μm. The assessment of the
arranged in the grid of the spurious frequency ±n f e,DDS and achieved accuracy depends on the application of the FMCW
have an amplitude of Jn (Ae,DDS / f e,DDS ) each. The spurious radar sensor. A higher SNR is required for further accuracy
frequency f e,DDS and the power amplitude J1 (Ae,DDS / f e,DDS ) improvement. The higher SNR can be achieved by averaging
of the first side band can be determined directly with a spectral or by using an improved sensor; therefore, a tradeoff between
analyzer. The spurious frequency f e,DDS and the spurious measurement time and cost is necessary. The FMCW range
amplitude Ae,DDS of the DDS output signal finally result in estimation accuracy can be further enhanced by using phase
the sinusoidal deviation in the frequency ramp with Ae and fe . evaluation [25]. In this case, the frequency estimation accuracy
Additionally, frequency conversion to high frequency by should be better than (λ/2) due to the unambiguous range of
the PLL with the division factor N has to be taken into the phase. In this case, it was proved that the range estimation
account accuracy is not limited by the SNR value.
Ae = Ae,DDS · N. (20) Finally, the source of sinusoidal deviations in the frequency
ramp was investigated and tracked down to spurious in
The spurious frequency f e,DDS is maintained during frequency the DDS-based reference signal of PLL-radar. This spurious
conversion and can be equated to the deviation frequency f e in was mathematically analyzed, thus establishing equations to
the frequency ramp. Requirements for the purity of the voltage describe its influence on range accuracy. The approach used
source and the reference signal for radar can be defined. Since in this paper can be further extended to describe the influence
the methods in this paper are used to determine the influence of any frequency ramp deviation on the IF-signal of an FMCW
of spurious sources at the output of a signal generator with radar.

Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on June 19,2023 at 12:23:39 UTC from IEEE Xplore. Restrictions apply.
3300 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 64, NO. 10, OCTOBER 2016

R EFERENCES Serdal Ayhan was born in Nußloch, Germany,

in 1981. He received the Dipl.-Ing. degree in com-
[1] H. Winner, S. Hakuli, and G. Wolf, Eds., Handbuch Fahrerassisten- munications and electronics and the M.Sc. degree
zsysteme, 2nd ed. (in German). Wiesbaden, Germany: Vieweg+Teubner, in information technology from the Mannheim Uni-
2011. versity of Applied Sciences, Mannheim, Germany,
[2] S. Ayhan et al., “Millimeter-wave radar distance measurements in in 2009 and 2010, respectively. He is currently
micro machining,” in Proc. IEEE Topical Conf. Wireless Sensors Sensor pursuing the Ph.D. degree in radar technology at
Netw. (WiSNet), Jan. 2015, pp. 65–68. the Institute of Radio Frequency Engineering and
[3] S. Ayhan, S. Scherr, P. Pahl, T. Kayser, M. Pauli, and T. Zwick, “High- Electronics, Karlsruhe Institute of Technology, Karl-
accuracy range detection radar sensor for hydraulic cylinders,” IEEE sruhe, Germany.
Sensors J., vol. 14, no. 3, pp. 734–746, Mar. 2014.
[4] S. Ayhan, S. Scherr, P. Pahl, S. Wälde, M. Pauli, and T. Zwick,
“Radar-based high-accuracy angle measurement sensor operating in the
K-band,” IEEE Sensors J., vol. 15, no. 2, pp. 937–945, Feb. 2015.
[5] S. O. Piper, “Homodyne FMCW radar range resolution effects with
sinusoidal nonlinearities in the frequency sweep,” in Proc. IEEE Int.
Radar Conf., May 1995, pp. 563–567.
[6] M. Pichler, A. Stelzer, P. Gulden, and M. Vossiek, “Influence of
systematic frequency-sweep non-linearity on object distance estimation
in FMCW/FSCW radar systems,” in Proc. 33rd Eur. Microw. Conf.,
vol. 3. Oct. 2003, pp. 1203–1206. Steffen Scherr (S’09) was born in Speyer,
[7] P. V. Brennan, Y. Huang, M. Ash, and K. Chetty, “Determination of Germany, in 1985. He received the Dipl.-Ing.
sweep linearity requirements in FMCW radar systems based on simple (M.S.E.E.) degree in electrical engineering from
voltage-controlled oscillator sources,” IEEE Trans. Aerosp. Electron. the Karlsruhe Institute of Technology, Karlsruhe,
Syst., vol. 47, no. 3, pp. 1594–1604, Jul. 2011. Germany, in 2010, where he is currently pursuing
[8] S. Scheiblhofer, S. Schuster, and A. Stelzer, “Signal model and lineariza- the Ph.D. degree in radar technology at the Institute
tion for nonlinear chirps in FMCW radar SAW-ID tag request,” IEEE of Radio Frequency Engineering and Electronics.
Trans. Microw. Theory Techn., vol. 54, no. 4, pp. 1477–1483, Jun. 2006. His current research interests include system
[9] A. Stelzer, E. Kolmhofer, and S. Scheiblhofer, “Fast 77 GHz chirps with design and digital signal processing for radar sys-
direct digital synthesis and phase locked loop,” in Proc. Asia–Pacific tems, in particular, for frequency-modulated contin-
Microw. Conf., vol. 3. Dec. 2005, pp. 1–4. uous wave radar systems.
[10] F. Hlawatsch and G. F. Boudreaux-Bartels, “Linear and quadratic time-
frequency signal representations,” IEEE Signal Process. Mag., vol. 9,
no. 2, pp. 21–67, Apr. 1992.
[11] L. Cohen, Time-Frequency Analysis. Upper Saddle River, NJ, USA:
Prentice-Hall, 1995.
[12] V. C. Chen and H. Ling, Time-Frequency Transforms for Radar Imaging
and Signal Analysis. Norwood, MA, USA: Artech House, 2001.
[13] W.-K. Lu and Q. Zhang, “Deconvolutive short-time Fourier transform
spectrogram,” IEEE Signal Process. Lett., vol. 16, no. 7, pp. 576–579, Akanksha Bhutani (S’16) was born in Dehradun,
Jul. 2009. India, in 1985. She received the B.Tech. degree
[14] S. Scherr, S. Ayhan, B. Fischbach, A. Bhutani, M. Pauli, and in electronics and communication engineering from
T. Zwick, “An efficient frequency and phase estimation algorithm with the Maulana Azad National Institute of Technol-
CRB performance for FMCW radar applications,” IEEE Trans. Instrum. ogy, Bhopal, India, in 2007, and the M.Sc. degree
Meas., vol. 64, no. 7, pp. 1868–1875, Jul. 2015. in information and communication engineering
[15] L. Rabiner, R. Schafer, and C. Rader, “The chirp z-transform algorithm,” from the Karlsruhe Institute of Technology (KIT),
IEEE Trans. Audio Electroacoust., vol. 17, no. 2, pp. 86–92, Jun. 1969. Karlsruhe, Germany, in 2012.
[16] S. Ayhan et al., “FPGA controlled DDS based frequency sweep gen- She was involved in high precision frequency
eration of high linearity for FMCW radar systems,” in Proc. German detection algorithms for radar applications with the
Microw. Conf. (GeMiC), Mar. 2012, pp. 1–4. Institute of Radio Frequency Engineering and Elec-
[17] M. Thumm, W. Wiesbeck, and S. Kern, Hochfrequenzmesstechnik: tronics (IHE), KIT, during her master’s thesis. Since 2012, she has been with
Verfahren und Messsysteme. (in German), Stuttgart, Germany: Teubner,
IHE as a Research Associate. Her current research interests include LTCC
1998. technology-based antenna-in-package concepts.
[18] D. Banerjee, PLL Performance, Simulation and Design, 4th ed.
Indianapolis, IN, USA: Dog Ear Publishing, 2006.
[19] K. Thurn, R. Ebelt, and M. Vossiek, “Noise in homodyne FMCW
radar systems and its effects on ranging precision,” in IEEE MTT-S Int.
Microw. Symp. Dig. (IMS), Jun. 2013, pp. 1–3.
[20] M. C. Budge and M. P. Burt, “Range correlation effects in radars,” in
Proc. IEEE Nat. Radar Conf., Apr. 1993, pp. 212–216.
[21] C. Sturm, Y. L. Sit, M. Braun, and T. Zwick, “Spectrally inter-
leaved multi-carrier signals for radar network applications and multi-
input multi-output radar,” IET Radar, Sonar Navigat., vol. 7, no. 3, Benjamin Fischbach was born in Pforzheim,
pp. 261–269, Mar. 2013. Germany, in 1987. He received the B.Sc. degree
[22] F. J. Harris, “On the use of windows for harmonic analysis with the in electrical engineering from the University of
discrete Fourier transform,” Proc. IEEE, vol. 66, no. 1, pp. 51–83, Stuttgart, Stuttgart, Germany, in 2011, and the
Jan. 1978. M.Sc. degree in electrical engineering from the
[23] Analog Devices Inc., Data Conversion Handbook, 1st ed. London, U.K.: Karlsruhe Institute of Technology, Karlsruhe,
Newnes, 2005. Germany, in 2013.
[24] S. Kay, Fundamentals of Statistical Signal Processing: Practical Algo- He was with Robert Bosch, Buyong, South
rithm Development, vol. 3. Upper Saddle River, NJ, USA: Prentice-Hall, Korea, from 2012 to 2013 as an Intern. Since
2013. 2013, he has been with Rohde & Schwarz, Munich,
[25] S. Ayhan, M. Pauli, T. Kayser, S. Scherr, and T. Zwick, “FMCW Germany, as a DSP Engineer. His current research
radar system with additional phase evaluation for high accuracy range interests include digital signal processing, ASIC/FPGA algorithm design,
detection,” in Proc. Eur. Radar Conf. (EuRAD), Oct. 2011, pp. 117–120. and microwave engineering.

Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on June 19,2023 at 12:23:39 UTC from IEEE Xplore. Restrictions apply.
AYHAN et al.: IMPACT OF FREQUENCY RAMP NONLINEARITY, PHASE NOISE, AND SNR ON FMCW RADAR ACCURACY 3301

Mario Pauli (S’04–M’10) was born in Lahr, Thomas Zwick (S’95–M’00–SM’06) received the
Germany, in 1975. He received the Dipl.-Ing. Dipl.-Ing. (M.S.E.E.) and Dr.-Ing. (Ph.D.E.E.)
(M.S.E.E.) degree in electrical engineering and the degrees from the Karlsruhe Institute of Technol-
Dr.-Ing. (Ph.D.E.E.) degree from the Karlsruhe Insti- ogy (KIT), Karlsruhe, Germany, in 1994 and 1999,
tute of Technology (KIT), Karlsruhe, Germany, respectively.
in 2003 and 2011, respectively. He was a Research Assistant with the Institute of
He spent four months as an Intern with the IBM Radio Frequency Engineering and Electronics (IHE),
T. J. Watson Research Center, Yorktown Heights, KIT, from 1994 to 2001. In 2001, he joined as a
NY, USA, in 2002, where he was involved in time Research Staff Member with the IBM T. J. Watson
and frequency domain measurement systems for the Research Center, Yorktown Heights, NY, USA. From
characterization of the 60 GHz indoor radio channel. 2004 to 2007, he was with Siemens AG, Lindau,
From 2004 to 2011, he was with the Institute of Radio Frequency Engineering Germany. During this period, he managed the RF Development Team for
and Electronics, KIT, as a Research Assistant, where he has been a Research automotive radars. In 2007, he became a Full Professor with KIT, where he
Associate since 2011. For the Carl Cranz Series for Scientific Education, is currently the Director of IHE. Since 2008, he has been the President of
he served as a Lecturer for radar and smart antennas. He is currently the the Institute for Microwaves and Antennas. He has authored or co-authored
Co-Founder and Managing Director of PKTEC GmbH, Paderborn, Germany. over 200 technical papers and holds 20 patents. His current research interests
His current research interests include electromagnetic field theory, radar and include wave propagation, stochastic channel modeling, channel measurement
sensor systems, antennas, millimeter-wave packaging, and microwave heating. techniques, material measurements, microwave techniques, millimeter-wave
antenna design, wireless communication, and radar system design.
Dr. Zwick was the recipient of over ten Best Paper Awards in international
conferences. He served on the Technical Program Committees (TPCs) of
several scientific conferences. In 2013, he was the General TPC Chair of the
International Workshop on Antenna Technology, and the European Microwave
Conference in 2013. He was selected as a Distinguished Microwave Lecturer
from 2013 to 2015 with his lecture on QFN-based packaging concepts for
millimeter-wave transceivers.

Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on June 19,2023 at 12:23:39 UTC from IEEE Xplore. Restrictions apply.