Professional Documents
Culture Documents
1 Introduction
Whilst Phase-Locked Loops (PLLs) are far from new [1, 2, 3, 4], they continue to
be the workhorse behind modern communications systems, and are a critical com-
ponent of many radar systems. The continual development of PLLs is therefore
as important today as it ever has been. The trend towards Direct Digital Synthe-
sizer (DDS) chips, produced by companies such as Analog Devices [5, 6], has be-
come the norm for large bandwidth RF designs when accompanied by high-M,N
frequency divider feedback networks. However, alternative methods such as the
Delay Line Discriminator (DLD) technique discussed in this paper remain worthy
of consideration.
This paper describes the basic functionality of a PLL circuit capable of linearis-
ing signals with 30dB fluctuation in power with 300MHz bandwidth, which when
passed through a highly-linear 13x frequency multiplier provides 4GHz of linearised
signal output centred at 94 GHz.
David G. Johnson
Australian Centre for Field Robotics, University of Sydney, Sydney, Australia,
e-mail: d.johnson@acfr.usyd.edu.au
Graham M. Brooker
Australian Centre for Field Robotics, University of Sydney, Sydney, Australia,
e-mail: gbrooker@acfr.usyd.edu.au
S.C. Mukhopadhyay, G.S. Gupta (eds.), Smart Sensors and Sensing Technology, 153
c Springer-Verlag Berlin Heidelberg 2008
154 D.G. Johnson and G.M. Brooker
The paper is organised as follows: Sect. 2 outlines the need for the linearisa-
tion of radar transmitters. Section 3 provides an overview of the radars front-end.
Section 4 discusses the specifics of the loop-linearisation process. Section 5 pro-
vides a performance analysis of the linearization process and discusses results for
two closely separated targets.
One of the simplest and most reliable methods of radar operation utilises a Frequency
Modulated Continuous Wave (FMCW) waveform, or Chirp, whereby targets at a
constant range, R, are detected at a unique beat frequency, fb , described by
2R f
fb = . (1)
c T
where f is the extent of the chirp that is generated over time, T, and c is the speed
of propagation.
For an ideal linear chirp, the range resolution, R, is related to the resolution of
the beat frequency, fb , (fb 1/T). Rearranging (1) therefore gives:
c
R = (2)
2 f
The beat frequency can be easily determined by applying the Fast Fourier Trans-
form (FFT) to data digitised at the receiver input. Two factors influence the resolu-
tion with which the beat frequency, and hence range can be measured:
The spectral purity of the transmitted signal at any given moment in time.
The non-linearity of the frequency chirp, as seen in Figure 1.
Noise caused by spectrally impure signals and/or linearity errors broadens the
signal-peaks in the subsequent FFT and hence prevent targets closely located in
range, and hence frequency, from being resolved. The true range resolution, R, is
then a function of both chirp bandwidth and chirp linearity, given by (3) in the case
of a quadratic nonlinearity [7].
%
!2
c
R = + (R.Lin)2 (3)
2 f
where the linearity, Lin, is defined as the change in chirp slope, S = f/T, normalised
by the minimum slope,
S max S min
Lin = (4)
S min
Sensor performance can therefore be improved by minimising the value of Lin.
Whilst the correction of non-linear signals can, to some extent, be performed in
Wide Band Linearization of MMW, SAW, DLD 155
Frequency variation
upon reception
Frequency
Time
post-processing [8, 9], these methods require additional doppler compensation tech-
niques to be applied to each detected target, and in any case it is intuitively more
sensible to ensure the transmitted signal is as pure and as linear as possible. This
can be achieved partially by the choice of transmitter components, but also by the
specific linearisation measures to be discussed in Sect. 4.
Over the past 5 years a number of 94GHz FMCW radars have been developed at the
ACFR (e.g. [10]).
Whilst the antenna, drive circuitry and signal processing have been continually
developed over time, a common front-end, sourced from Elva-1 [11] has been em-
ployed and forms the backbone for this radar, as seen in Fig. 2. The transmit chain
is based on a low frequency VCO followed by an Impatt Diode Active Frequency
Multiplier which is inherently more linear than the alternative: Gunn oscillators; but
is nevertheless limited to a linear bandwidth of 500 MHz at MMW frequencies
when operated in an open-loop arrangement. In order to improve the linearity at
the transmitting antenna, the oscillator must be phase & frequency locked to a stable
reference source.
The linearity of the VCO in open-loop is shown in Fig. 3, where it can be
seen that an IF bandwidth of 600 MHz is achievable, which when multiplied to
MMW frequencies gives a theoretical bandwidth of 7.8 GHz. However, the useful
156 D.G. Johnson and G.M. Brooker
Injection
13x Freq. Bandpass Power Tx / Rx
Isolator Locked Circulator
Multiplier Filter Divider Antenna
Amplifier
Key:
Power 5.83 GHz UHF IF
Amplifier LO Amplifier Amplifier Common
'Backbone'
Components
The accepted method for implementing a Frequency & Phase Locked Loop for
FMCW radar is via Direct Up-Conversion as depicted in Fig. 5. A high-speed DDS
PLL CHIP
Phase /
Charge
Frequency
Pump
Detector
DDS CHIP
chip, pre-programmed with the required frequency slope parameters and contain-
ing a high dynamic range Digital to Analogue Converter (DAC), outputs a series
of incremental frequency reference signals to the Phase/Frequency Detector (PFD)
within the PLL chip. The reference signals are then compared to a sub-sampled
version of the VCO output within the PFD, which in combination with a charge
pump and a Low Pass Filter outputs a corrected voltage to the VCO, resulting in a
frequency closer to that desired.
The obvious drawback of the DDS approach in terms of the linearity of the fre-
quency slope is the fact that the frequency is stepped rather than swept linearly.
However for modern, fast DDS chips, the duration of each frequency step is typ-
ically less than a cycle, hence providing phase transients between frequency-steps
are well catered for [12], the overall sweep linearity can be quite reasonable.
In our method, shown in Fig. 6, a linear voltage ramp is passed to a VCO, combined
with a PLL correction voltage. Further linearisation is then achieved by tapping o
VCO (RF)
1.4GHz
150MHz
7.23GHz 2
150MHz 700MHz
75MHz
Differentiate
DELAY LINE DISCRIMINATOR
1.22 s
LPF
(Integrate)
Phase /
Frequency
Detector
VCO (PLL)
183 kHz 183kHz Nonlinearities
Fig. 6 PLL Employing DLD
Wide Band Linearization of MMW, SAW, DLD 159
5 Results
linear, as can be seen from Fig. 3. A subset of this region, corresponding to a MMW
range of 92.596 GHz, was then chosen such that transient eects within the PLL
loop filter generated by changes in slope polarity were minimised. By sampling at
a rate of 5 MHz, a 4 k FFT could then by applied to the IF signal within this region
for later range profile generation.
Figure 7b displays the output of the PLL loop filter Integrator on both up and
down sweeps. Employing a positive edge-triggered phase comparator with a trans-
fer characteristic given by (5), the deviation in frequency slope from the ideal is
essentially given by the magnitude of this curve when the phase dierence between
the two input signals (DLD REF ) is truly less than 180 .
VCC
VPC = (DLD REF ) (5)
2
It can be seen that the linearization begins to break down towards the end of the
up-sweep. This is due to a small dierence (5%) in the sweep time on the up-sweep
causing the change in phase comparator output to exceed the relatively narrow band-
width of the loop filter (which is still attempting to lock to a reference signal fixed
at 203 kHz), allowing the relative phase to pass over the 2 phase-boundary.
The final two parts of Fig. 7 display the DLD output in both the temporal and fre-
quency domain. Figure 7d was formed by sliding a Hanning window and 512 pt FFT
Wide Band Linearization of MMW, SAW, DLD 161
over the raw DLD signal and applying a peak detection and interpolation algorithm
to the result. The mirroring between the results from up- and down-sweeps is made
slightly less obvious by the change in up-sweep duration previously mentioned caus-
ing a step change in DLD frequency output around 1 ms and a slight scale-reduction
of up-sweep data in the un-linearised case. However, whilst lock is maintained on
the first part of the up-sweep the sweep rate is almost identical to that of the down-
sweep. It should be noted that the sharp deviation in linearity observed at around
1.7 ms is more an artefact of the peak detection algorithm rather than a kink in the
frequency slope associated with the phase transition seen in Fig. 7b. For the bet-
ter performing down-sweep an improvement in linearity from 0.12 to <0.02 can be
observed by applying (4).
As stated previously, a radars true range resolution is best measured by its ability
to discriminate between two closely separated targets. Range measurements were
therefore made against a pair of closely located targets in both linearised and open-
loop modes. The trihedral corner-reflector targets (3 m2 at 94 GHz), were mounted
on a tripod (7 m 5 cm) down-range from the radar within the 1-degree beamwidth
of the MMW antenna. The tripod was then placed 2 m in front of a brick wall.
Figure 8 depicts the apparatus setup:
It was later realised that the 1.22 s delay line corresponded coincidentally to a
target at 7.1 m for the bandwidth and sweep-time employed. Fortunately this did not
impact on results due to the range oset caused by waveguide lengths within the
radar.
A 4 k FFT was then applied to the IF signal on the down sweep as described
in Sect. 5.1. According to the measurements of DLD frequency deviation made
previously, a spectral peak bandwidth of 20 kHz was expected for the un-linearised
case and this can indeed be seen in the FFT of both the DLD and IF signals in Fig. 9.
Comparing the upper graphs in Figs. 9 and 10 the narrowing of the spectral peak
of the false-target generated by the DLD can be easily observed.
Examining the lower graphs in Figs. 9 and 10 it would appear at first glance
that the two targets can be discerned just as easily without additional linearization.
7m
7 cm
Radar (NOT TO SCALE)
10 cm
9m
Fig. 8 Experimental setup
162 D.G. Johnson and G.M. Brooker
However, applying (1) to the highlighted peaks in the un-linearised case, with the
appropriate bandwidth/sweep-duration ratio, places the targets 19 cm apart, com-
pared to 10 cm for the linearised case, as required. The wall is also more clearly
discernable in the linearised case.
6 Conclusion
A method for linearising an FMCW radar signal over a large bandwidth, without the
need for a DDS, has been developed and tested experimentally. Results show a better
than 6 fold improvement over the already highly-linear open-loop configuration,
with a range resolution of <10 cm readily achievable.
One remaining drawback of this method is that whilst the frequency-slope gradi-
ent may be kept constant over a chirp, the upper and lower frequency bounds tend
to drift over time. This will be fixed in the next iteration of the PLL loop filter by
means of a secondary frequency locked loop.
References
1. S.C. Gupta, Phase-Locked Loops, Proceedings of the IEEE, 63 (2), pp 291306 (1975)
2. M. Curtin, P. OBrien, Phase-Locked Loops for High Frequency Receivers and
Transmitters-Part 1, Analog Devices Inc., Analog Dialogue 333 (1999)
3. M. Curtin, P. OBrien, Phase-Locked Loops for High Frequency Receivers and
Transmitters-Part 2, Analog Devices Inc., Analog Dialogue 335 (1999)
4. M. Curtin, P. OBrien, Phase-Locked Loops for High Frequency Receivers and
Transmitters-Part 3, Analog Devices Inc., Analog Dialogue 337 (1999)
5. AD9954-400 MSPS, 14-Bit, 1.8V CMOS, Direct Digital Synthesizer, Analog Devices,
Norwood, MA, Datasheet, Rev. A, Jan 2007 [Online]. Available: http://www.analog.com
6. AD9956-2.7 GHz DDS-based AgileRFTM Synthesizer, Analogue Devices, Norwood, MA,
Datasheet, Rev. A, Sep 2004 [Online]. Available: http://www.analog.com
7. G.M. Brooker, Understanding Millimetre Wave FMCW Radars, Proceedings International
Conference on Sensing Technology, pp 152157 (2005)
8. S. Scheiblhofer, S. Schuster, A. Stelzer, Signal Model and Linearization for Nonlinear
Chirps in FMCW Radar SAW-ID Tag Request, IEEE Transactions on Microwave Theory
and Techniques, 57 (4), pp 14771483 (2006)
9. M. Vossiek, T. v. Kerssenbrock, P. Heide, Novel Nonlinear FMCW Radar for Precise
Distance and Velocity Measurements, Microwave Symposium Digest, 1998 IEEE MTT-S
International, pp 511514, vol 2 (1998)
10. G.M. Brooker, S. Scheding, M.V. Bishop, R.C. Hennessy, Development and Application
of Millimeter Wave Radar Sensors for Underground Mining, IEEE Sensors Journal, 5 (6),
pp 12701280 (2005)
11. FMCW Radar Front-End Module, Datasheet, [Online]. Available: http://www.elva-1.com
12. M. Pichler, A. Stelzer, P. Gulden, C. Seisenberger, M. Vossiek, Frequency-Sweep Lineariza-
tion for FMCW Sensors with High Measurement Rate, Microwave Symposium Digest, 2005
IEEE MTT-S International, pp 16931696 (2005)
164 D.G. Johnson and G.M. Brooker
13. M. Feldmann, J. Hena, Surface Acoustic Waves for Signal Processing, Artech House, Boston
(1989)
14. D.G. Johnson, G.M. Brooker, 4GHz Bandwidth Closed Loop Linearisation of a Millimetre-
Wave, Linear Frequency Modulated Radar, Proceedings 2nd International Conference on
Sensing Technology, Palmerston North, New Zealand, pp 534539 (2007)