Vous êtes sur la page 1sur 4

Proceedings of the 14th European Radar Conference

Modeling of Range Accuracy for a Radar System


Driven by a Noisy Phase-Locked Loop
Frank Herzel1 , Herman Jalli Ng1 and Dietmar Kissinger1,2
1
IHP, Im Technologiepark 25, 15236 Frankfurt (Oder), Germany
2
Technische Universität Berlin, Einsteinufer 17, 10587 Berlin, Germany
Email: herzel@ihp-microelectronics.com

Abstract—We derive analytical expressions for the standard


deviations of distance error and angular error in a radar system ramp
generator PLL PA td=2R/c0
due to PLL phase noise. The standard deviation of the distance
error σR is expressed as a function of the target distance R
and the PLL phase noise spectrum. If the IF radar signal and
noise bandwidth are within the in-band phase noise plateau, the
proportionality σR ∝ R3/2 is valid. By contrast, if signal and (t)
noise bandwidth are in the VCO dominated -6dB/octave region of
the PLL spectrum, we obtain σR ∝ R. The results are compared (t-td)
out (t) LPF LNA
with measurements on a 61 GHz and a 122 GHz FMCW radar
system.
Fig. 1. Simplified block diagram of homodyne radar frontend.
Keywords—radar; FMCW; phase noise; range accuracy; phase-
locked loop.
radar signal round-trip time td is symbolized by a delay line.
I. I NTRODUCTION The mixer multiplies the transmitted signal at time t with the
signal at an earlier time t − td . After the low-pass filter (LPF),
Frequency modulated continuous wave (FMCW) radar has the phase of the radar signal roughly represents the difference
become popular in microwave ranging and imaging systems. of the PLL output phases at these points of time. This is a good
A fractional-N phase-locked loop (PLL) can be used as the approximation, if the noise bandwidth bn of the LPF is much
FMCW generator [1]-[8]. The PLL phase noise may dom- larger than the PLL loop bandwidth fL . Otherwise, additional
inate the range accuracy in a radar system, especially, for low-pass filtering must be included in the calculation when
long-distance radar. For example, in [7] it was found that integrating the spectrum. This results in an expression for the
for distances R > 0.5 m phase noise had the dominating phase variance at the receiver output given by
influence on the signal-to-noise ratio (SNR), whereas for lower  bn
distances phase noise is suppressed to below the thermal noise. 2
Predicting the impact of PLL phase noise is, therefore, helpful σφ = 8 Sφ (f ) sin2 (2πf R/c0 ) df (1)
0
for optimizing a radar system for a low SNR.
as used in [10]. Here, R is the target distance and c0 is the
Range correlation effects can be considered by a weighting velocity of light. Following [11] and [12] , the signal-to-noise
function for the phase noise spectrum as described in [9]. ratio SN R can approximately be calculated from
Basically, the PLL output phases at different points in time
1/SN R = 1/SN R0 + 2 σφ2 (2)
are subtracted in a radar system reducing the phase noise.
Additional low-pass filtering at the mixer output needs to be where SN R0 is the signal-to-noise ratio in the absence of PLL
included for narrowband radar systems [10]. This allows the phase noise. The factor of two in (2) results from the fact that,
rms phase error at the receiver output and the resulting contri- unlike in [11], we normalized the noise voltage to the effective
bution to the signal-to-noise ratio (SNR) to be calculated from signal amplitude rather than to the peak amplitude, see also (7)
phase noise spectrum, target distance and noise bandwidth. in [12]. Expressing the phase error in degrees and the SNR
While it is clear that the SNR is strongly correlated with the in decibels, the maximum achievable SNR in the presence of
ranging accuracy, an analytical description of the rms distance phase noise is given by
error as a function of phase noise and target distance is still √
missing. SN Rmax (dB) = 20 log[(180◦ /π) /( 2 σφ (deg))] . (3)
As an example, for an rms phase error of 3◦ the SNR cannot
This paper presents an analytical model for the rms distance
be larger than 23 dB.
error due to PLL phase noise. Moreover, simplified expressions
for fast and slow ramps are derived.
B. Fast ramps
II. A NALYSIS OF THE PHASE ERROR Let us consider the case of fast ramps where the radar
signal frequency and bn are much above the PLL loop band-
A. General case
width fL . In this case, the PLL phase noise spectrum can be
Fig. 1 shows a simplified model of the radar frontend. The approximated in the relevant frequency region by the VCO

978-2-87487-049-1 © 2017 EuMA 521 11–13 Oct 2017, Nuremberg, Germany


phase noise Sφ,VCO (f ) = Sφ,VCO (Δf )(Δf /f )2 , where f is Assuming σTout << Tout the error propagation law can
the frequency offset from the carrier and Sφ,VCO (Δf ) is the be applied. Then, the output frequency fout = 1/Tout has the
VCO phase noise at a specific offset Δf in the -6 dB/octave same relative error as Tout . This yields with (7)
region, e.g., at Δf =1 MHz. Here, we assumed that the quan- σTout fout
tization noise in the PLL can be suppressed to a negligible σfout = fout = √ σφ . (8)
level. In view of the high input frequencies in fractional-N Tout 2π
PLLs, this can typically be achieved by using a high-order In the case of white noise, averaging n uncorrelated measure-

low-pass filter in the PLL. Performing the integration in (1) ments reduces the standard deviation by a factor of 1/ n.
yields for bn → ∞ One frequency sweep contains n = Ts fout oscillations, where
Ts is the sweep duration. As an approximation, we disregard
σφ2 = 8π 2 Sφ,VCO (Δf )(Δf )2 (R/c0 ) . (4) cross-correlation between different IF output periods. Then,
the mean frequency
√ error for a sweep is lower than (8) by a
For example, with a VCO phase noise of -80 dBc/Hz at factor of 1/ Ts fout . We obtain
Δf =1 MHz we obtain for a distance of R = 25 m an rms phase 
error of σφ =0.256, which corresponds to 15◦ . For a distance 2 fout σφ
σfout = . (9)
of R = 1 m the phase error is as small as 3◦ . Ts 2 π
The receiver output frequency is related to the ramp slope
C. Slow ramps df0 /dt = bs /Ts and the target distance R by

Next, we consider slow ramps where the radar signal and bs 2R


fout = (10)
the noise bandwidth are much below the loop bandwidth, that T s c0
is, bn << fL . Assuming that the PLL phase noise can be where bs is the sweep bandwidth (in hertz). From (10) and (9)
described by a constant in-band noise floor Sφ,fl in the relevant we obtain the rms distance error resulting from phase noise
frequency region, we obtain from (1) given by
 √
bn T s c0 Ts fout c0
σφ2 = 8 Sφ,f l sin2 (2πf R/c0 ) df . (5) σR = σfout = √ σφ . (11)
2 bs 8 π bs
0
Substituting (10) into (11) we obtain the final result
III. A NALYSIS OF THE DISTANCE ERROR 
c0 R σφ
A. General case σR = (12)
bs 2 π
In an FMCW radar system the accuracy of the distance where σφ is given
√ by (1). Note that this expression is larger
CRLB
measurement is related to the rms frequency error at the re- by a factor of n than the Cramér-Rao lower bound σR
ceiver output. A higher accuracy can be obtained by evaluating in [10] for SNR >> 1, where n = Ts fout is the number of
the phase of the radar signal as well [12],[13], but this is receiver output periods in one frequency sweep. From n ∝ R
CRLB
beyond the scope of this analysis. From the rms phase error we conclude that σR (R) has a larger slope than σR (R)
at the LPF output σφ we obtain the standard deviation of the in agreement with the measurements in [10]. From n ≥ 1 we
CRLB
corresponding timing error given by find σR ≥ σR as it should be. In order to reach the CRLB,
the phase of the IF radar signal must be evaluated as in [13].
Tout
σtout = σφ . (6)
2π B. Fast ramps
The momentary output period Tout (t) is defined as the timing For bn >> fL the general result (12) can be simplified. By
difference between two consecutive minus-to-plus zero cross- combining (4) and (12) we find for the variance of the distance
ings of the receiver output signal. Disregarding the initial time error measurement for one sweep
interval 0 < t < ts ≈ 10/ωL of a linear frequency sweep 2 R2
2
where the PLL settles, Tout (t) can be considered a stationary σR = Sφ,VCO (Δf )(Δf )2 . (13)
stochastic process [14]. Neglecting 1/f noise, we find for the bs
standard deviation of the receiver output period Tout Fig. 2 shows the rms distance error as a function of the target
distance R.
√ Tout
σTout = 2 σtout = √ σφ . (7)
2π C. Slow ramps

Here, we assumed that in the case of white noise sources two By substituting (5) into (12) we obtain for the variance of
successive zero crossings of the output signal are statistically the distance measurement for bn << fL
independent. This assumption is motivated by the observation  bn
2 2 R c0
that the phase noise spectrum of the PLL is high-pass filtered σR = 2 Sφ,f l sin2 (2πf R/c0 ) df . (14)
π bs 0
according to (1). This corresponds to a small auto-correlation
time of the phase, since the auto-correlation function (ACF) of For short and medium distances we have
the phase represents the inverse Fourier transform of the phase c0
noise spectrum. R << (15)
2πbn

522
target

1 R2 ~ (y-d) / 
antenna2
y - d
Rms distance error (m)

R1 ~ y /  y
0.1
d 


0.01 antenna1 x

Fig. 4. Illustration of angle calculation from two distance measurements.


-80 dBc/Hz at 1 MHz
-90 dBc/Hz at 1 MHz
0.001 -100 dBc/Hz at 1 MHz
approximate sin(θ) ≈ θ and cos(θ) ≈ 1. In other words, the
1 10 100 1000 Taylor series with respect to θ is truncated after the linear term.
Target distance (m) For d << Δy we find
Fig. 2. Calculated standard deviation of distance measurement for one
frequency sweep according to (13) for different VCO phase noise levels at
1 MHz offset (bs =1 GHz). R1 = Δy/θ, R2 = (Δy − d)/θ (17)
and
d
and we can replace sin(x) with x in (14). This yields
θ= . (18)
8 R3 R1 − R2
2
σR ≈ Sφ,f l b3n . (16)
3 c0 b s For small distance errors, the standard deviation of θ can be
For a noise bandwidth of bn =100 kHz and a sweep bandwidth calculated from (18) using the error propagation law. However,
bs =1 GHz we obtain the standard deviation due to PLL phase this requires the relative changes of θ to be small. Only then,
noise shown in Fig. 3. The approximation (15) resulting in the angle θ can be linearized with respect to the distances R1
and R2 . This requires a small target distance error compared
to the antenna distance d. Let us consider the case
σR << R1 − R2 (19)
1
which implies σR << d. In this case, the relative changes of θ
Rms distance error (m)

0.1 due to circuit noise are small and we can use error propagation.
This yields with σR1 ≈ σR2 = σR
0.01
    √
 dθ 2  dθ 2 2d
0.001
σθ =   σ +
2  σ =
2 σR .
dR1  R1  dR2  R2 (R1 − R2 )2
0.0001 noise floor -70 dBc/Hz (20)
noise floor -80 dBc/Hz
noise floor -90 dBc/Hz By substituting (18) into (20) we obtain
1e-05
√ √ 2
1e-06 2θ 2θ
1 10 100 1000 σθ = σR = σR . (21)
Target distance (m) R1 − R2 d
Fig. 3. Calculated standard deviation of distance measurement for one
frequency sweep according to (14) (solid lines) and according to (16) Note that the angular accuracy can be enhanced by phase
(dashed lines). The phase noise floor levels are -70 dBc/Hz, -80 dBc/Hz and measurements, i.e., interferometry. The purpose of this simple
-90 dBc/Hz. consideration was to illustrate how the more-dimensional error
propagation law (20) can be applied for angular accuracy
σR ∝ R3/2 is very accurate up to a target distance of 500 m estimation.
for our example, provided that the phase noise spectrum is
constant as assumed in this subsection. Summarizing this section, an angular measurement based
on two distance measurements requires the antenna distance
IV. A NGULAR ACCURACY to be much larger than the rms distance error.
By using more than one antenna, an angular probability V. M EASUREMENTS
density function for the target can be estimated, provided that
the distance accuracy is high enough. In order to obtain a first We performed accuracy measurements on a scalable sensor
understanding, we assume a situation, where two antennas with platform with 61 and 122 GHz transceivers described in [8].
the distance d between them are used as illustrated in Fig. 4. The distance measurements were repeated 1000 times. A
For simplicity, we consider only small angles θ where we can Gaussian distribution was then fitted to the error distribution

523
R=1m R=2m would require to include noise in the receiver as well as
quantization noise.

VI. C ONCLUSIONS
We have presented a model for the ranging error of radar
systems resulting from PLL phase noise. The standard devia-
tion in a distance measurement was calculated as a function
of PLL phase noise, target distance and receiver bandwidth.
For a large number of receiver output periods per frequency
R=3m R=4m sweep our estimate is significantly larger than the Cramér-Rao
lower bound. Experiments have been performed on a 61 GHz
/ 122 GHz radar system to support the theoretical results.

R EFERENCES
[1] T. Musch, I. Rolfes, and B. Schieck, “A highly linear frequency ramp
generator based on a fractional divider phase-locked-loop,” IEEE Trans.
Instrum. Meas., vol. 48, no. 2, pp. 634-637, Feb. 1999.
[2] M. Pichler, A. Stelzer, P. Gulden, C. Seisenberger, and M. Vossiek,
Fig. 5. Histograms for the measured error values at R =1 m, 2 m, 3 m and “Phase error measurements and compensation in PLL frequency synthe-
4 m for the 61 GHz system for 1000 frequency sweeps each. sizers for FMCW sensors-I: context and application,” IEEE Transactions
on Circuits and Systems I: Regular Papers, vol. 54, pp. 1006-1017, May
2007.
for four distances. Fig. 5 shows the histograms for the 61 GHz [3] J. Lee, Y.-A. Li, M.-H. Hung, and S.-J. Huang, “A fully-integrated
radar system. The same measurement was performed on the 77-GHz FMCW radar transceiver in 65-nm CMOS technology,” IEEE
Journal of Solid-State Circuits, vol. 45, pp. 2746-2756, Dec. 2010.
122 GHz system. Fig. 6 shows the measured rms distance error
[4] N. Pohl, T. Jaeschke, and K. Aufinger, “An ultra-wideband 80 GHz
as a function of the target distance R for both the 61 GHz and FMCW radar system using a SiGe bipolar transceiver chip stabilized
the 122 GHz radar system. The values from Fig. 3 are also by a fractional-N PLL synthesizer,” IEEE Transactions on Microwave
Theory and Techniques, vol. 60, pp. 757-765, Mar. 2012.
[5] H. J. Ng, A. Fischer, R. Feger, R. Stuhlberger, L. Maurer, and A. Stelzer,
“A DLL-supported, low phase noise fractional-N PLL with a wideband
VCO and a highly linear frequency ramp generator for FMCW radars,”
1000
Rms distance error (micrometer)

IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 60,


pp. 3289-3302, Dec. 2013.
[6] W. Wu, R. B. Staszewski, and J. R. Long, “A 56.4-to-63.4 GHz multi-
rate all-digital fractional-N PLL for FMCW radar applications in 65 nm
100 CMOS,” IEEE Journal of Solid-State Circuits, vol. 49, pp. 1081-1096,
May 2014.
[7] G. Hasenaecker, M. van Delden, T. Jaeschke, N. Pohl, K. Aufinger,
and T. Musch, “A SiGe fractional-N frequency synthesizer for mm-wave
wideband FMCW radar transceivers,” IEEE Trans. on Microwave Theory
10 measured at 122 GHz and Techniques, vol. 64, no. 3, pp. 847-858, Mar. 2016.
measured at 61 GHz [8] H. J. Ng, M. Kucharski and D. Kissinger, “Scalable sensor platform
noise floor -60 dBc/Hz
noise floor -70 dBc/Hz with multi-purpose fully-differential 61 and 122 GHz transceivers for
noise floor -80 dBc/Hz MIMO radar applications,” in 2016 IEEE Bipolar / BiCMOS Circuits
and Technology Meeting (BCTM), New Brunswick, NJ, Oct. 2016, pp.
1 170-173.
1 10
Target distance (m) [9] M. C. J. Budge and M. P. Burt, “Range correlation effects in radars,” in
Proc. Record of the 1993 IEEE National Radar Conference, Lynnfield,
Fig. 6. Measured standard deviations of the two radar systems as a MA, Apr. 1993, pp. 212-216.
function of target distance (symbols); calculated standard deviation of distance
measurement with one frequency sweep according to (16) (lines). [10] K. Thurn, R. Ebelt and M. Vossiek, “Noise in homodyne FMCW
radar systems and its effects on ranging precision” 2013 IEEE MTT-
S international microwave symposium digest, Seattle, WA, USA, Jun.
shown for comparison. The modeled curves grow by 9 dB with 2013, pp. 1-3.
distance doubling according to (16). The mean slope of the [11] D. Banerjee, PLL Performance, Simulation, and Design, 4th ed. Indi-
measured errors is similar. In addition the PLL phase noise, anapolis: Dog Ear Publishing, 2006.
the radar distance measurement is also affected by thermal [12] P. V. Brennan, L. B. Lok, K. Nicholls, and H. Corr, “Phase-sensitive
noise in the receiver antenna amplified by the receiver noise FMCW radar system for high-precision Antarctic ice shelf profile moni-
figure. Moreover, the finite resolution of the analog-to-digital toring,” IET Radar, Sonar & Navigation, vol. 8, no. 7, pp. 776-786, Aug.
2014.
converter results in an error in the distance measurement.
[13] S. Ayhan, P. Pahl, T. Kayser, M. Pauli, and T. Zwick, “Frequency
Referring to [7], the thermal noise results in a dependency estimation algorithm for an extended FMCW radar system with addi-
σR ∝ R and the quantization noise contribution is independent tional phase evaluation,” Proc. Sixth German Microwave Conference,
of R. The similarity of the measured slope with the modeled Darmstadt, Germany, Mar. 2011.
behavior of σR ∝ R3/2 might be an indication that PLL [14] F. Herzel, A. Ergintav, and Y. Sun, “Phase noise modeling for integrated
phase noise is relevant in our measurements. However, a sound PLLs in FMCW radar,” IEEE Trans. Circuits Syst-II: Express Briefs, vol.
comparison of measurements with the theoretical predictions 60, no. 3, pp. 137-141, Mar. 2013.

524

Vous aimerez peut-être aussi