Kostas Doris, Alessio Filippi, and Feike Jansen

The pathway to
higher resolution

Reframing
Fast-Chirp FMCW
Transceivers for Future
Automotive Radar
A
utomotive radar is (mm-wave) and baseband processing The step to conditional driving
evolv ing, leverag- strengths of nanometer (nm) CMOS automation and beyond toward Levels
ing the strengths of processes to enable range and velocity 3–5 (Figure 1) will require substantial
fast-chirp frequency- performance scaling in the 76–81-GHz sensing performance improvements
modulated continuous- band (Figure 1). In combination with in new fronts, such as high angular
wave (FMCW) waveform for distance nm-scaled automotive qualified micro- resolution and optimized fields of
and velocity sensing. Building on uni- processors, this evolution promotes view (FoVs) for azimuth and eleva-
versal adoption of FMCW waveforms, sensor tailoring for short-, mid-, and tion and robustness to radar-to-radar
innovation at the circuit architec- long-range use cases and opens a path- interference. Enhanced sensitivity
ture level exploits millimeter-wave way to high angular resolution and will be needed to support the reli-
radar imaging with the use of multi- able detection of motorcycles at far
Digital Object Identifier 10.1109/MSSC.2022.3167344 ple-input, multiple-output (MIMO) distances, pedestrians, or debris on
Date of current version: 25 June 2022 techniques and sensor cascading. the road in the vicinity of trucks or

44 SPRING 202 2 IEEE SOLID-STATE CIRCUITS MAGAZINE 1943-0582/22©2022IEEE

Authorized licensed use limited to: Richard David. Downloaded on July 27,2022 at 20:18:45 UTC from IEEE Xplore. Restrictions apply.
buildings, and eventually, to enable at time t 0 that increases linearly to A modulator sets the chirp phase
the mapping of the environment, f0 + B over a pulse duration of Tchirp . before amplification and transmis-
while cost-effectiveness and size It employs pulse compression [1] to sion via the antenna. Figure 2 shows
reduction complying to car design decouple range resolution (the abil- a black chirp frame bouncing off a
will remain essential. ity of the radar to distinguish targets target. The reflected chirps in red are
In this article, we take a renewed in range), determined by modulation received, amplified, and mixed with
look at the design approach that bandwidth B, from sensitivity, deter- the transmitted chirp sequence for
nowadays has FMCW waveforms mined by the signal-to-noise ratio correlation. Echo delays x translate
and CMOS transceiver as the epicen- (SNR) for a given range as a func- to beat frequencies (blue) proportion-
ter of distance and velocity detection. tion of the pulse emission duration ally to the modulation bandwidth
We shift our attention to the MIMO Tchirp, emitted power, and receiver over the chirp duration as described
antenna and transceiver array, point- noise. Small frequency variations in (5) in Table 1. The beat tones are
ing out new means to use optimally introduced by the motion of targets filtered and amplified prior to digi-
available physical and IC technology relative to the radar source known tization at a Nyquist sampling rate
resources for angular resolution and as the Doppler effect [1] provide equal to twice the maximum beat
interference before the physical lim- velocity estimation using sequential frequency. Large dynamic range dif-
its imposed by the 76–81-GHz band pulses to measure phase changes of ferences between echoes due to free
necessitate, once more, moving to a reflected waves. The wavelength and space loss in the forward and return
new band. pulse repetition rate determine the paths (for example, 52 dB for two tar-
maximum unambiguous velocity, gets at 10 and 200 m) combined with
FMCW Radar Resolution Space whereas the wavelength and total radar cross-sections varying between
Radar sensing uses modulated elec- measurement time determine the −5 and 30 dB (e.g., for debris com-
tromagnetic pulses to measure dis- velocity resolution (the minimum pared to a truck) in the presence of
tance and velocity simultaneously. speed differences a radar can detect). noise require the use of coherent inte-
The transmitted chirped pulses in Table 1 summarizes the main gration using multiple sequences and
Figure 2 are reflected by an object relationships. Figure 2 also shows optimal matched filtering in the form
and received back with a delay pro- the FMCW transceiver architecture. of FFTs.
portional to the time of flight, which A fractional-N phase-locked loop The range FFT provides quantized
is a measure of distance to the radar. and the frequency translation cir- distance information in range bins
A chirp is a tone with a frequency f0 cuits generate the 76–81-GHz chirp. operating on N R samples from each

Level 1 Level 2 Level 2+ to 3 Level 4–5

Driver Assistance Partial Automation Conditional Automation High/Full Automation

Mapping of the Environment

Seeing Seeing Bicycles Seeing Seeing
Other and Pedestrians Smaller Around
Cars Objects the Car

Imaging Highest
Radar Resolution and
24–24.5 GHz Performance
How to Expand
2D Front and Rear Resolution Further?
Speed, Distance 76–81 GHz Long-Range
Higher
SiGe Radar
Low Performance How Far Before Moving
Resolution 3D Again to a New Band?
+ Azimuth Corner Multiple
Radar Small Modules Which Technologies?
Higher
Resolution 76–81-GHz CMOS + Advanced Processing
4D + Elevation
Resolution Boost

FIGURE 1: Radar resolution expansion in the mm-wave bands.

IEEE SOLID-STATE CIRCUITS MAGAZINE SPRING 202 2 45

Authorized licensed use limited to: Richard David. Downloaded on July 27,2022 at 20:18:45 UTC from IEEE Xplore. Restrictions apply.
 10 MHz Field of View, ange
tion R
Windowing ADC Filter e.g., 45° Detec 200 m Range
e.g.,
Resolution
Range FFT 76–81 GHz
Frac-N PLL υ
Doppler FFT I
Up-Converter υ
Q
MIMO Processing Azimuth Resolution,
Phase Control e.g., 1°

Synchronization
Doppler

Doppler

Doppler
FFT

FFT

FFT

Sensitivity
f0 + B Transmit Received Echo
Chirp

Frequency
Chirp
Number of Chirps (256 Typical)

Frame
Range Gate Slice

υres ∆f ∆f
g φres ∆t
sin f0
ces Rres
ro FFT Time
rP 10–40 µs Time
s

a
na

l
gu
en

An
nt

Sampling Moments
fA
ro
be
um

Range
Range Gates (1,024 Typical)
N

FFT

FIGURE 2: The fast-chirp FMCW signal processing and transceiver architecture. Frac-N: fractional-N.

single chirp. Windowing improves processing gain of 10 log 10 N R N D dB. exceed a certain threshold value for
the peak-to-sidelobe ratio for the Multiple antenna signal streams form reliable detection. Radar signal pro-
easier detection of weak targets at a third dimension for the detection cessing algorithms operate on the
the cost of a loss in the range resolu- of the angle of arrival, as will be dimensions of the cube.
tion. The output from all the chirps detailed in later sections. The result- Radar design is subject to multiple
in the frame is stored to perform a ing data cube is a quantized repre- tradeoffs within and across the wave-
second Doppler FFT on N D samples, sentation of the observation scene of form, antenna, and circuit domains.
corresponding to equal distance to the radar in space and velocity. Pixel For example, assuming a long-range
the extract velocities, which boosts dimensions represent resolution radar scenario, chirps need to be far
target over noise power with a total with a given sensitivity, which must apart to avoid range ambiguities for a
long range in (3), but for high maxi-
mum unambiguous velocity, they
TABLE 1. BASIC RADAR EQUATIONS. need to be close together. Long chirps
and long measurement times improve
PARAMETER EQUATION COMMENTS
sensitivity and velocity resolution,
Range resolution R res = c 0 . (1) co: speed of light respectively, but this makes chirps
2B B: modulation bandwidth
from the same sequence measure
Maximum range c fN Tchirp fN = fs /2: Nyquist frequency for max range the same object at different ranges,
R max = . (2)
2B fs: sampling rate resulting in range migration and
Tchirp: chirp duration,
the loss of sensitivity. For example,
Velocity resolution v res = m
(3) m: wavelength detecting a motorcycle with a relative
2Tmeas Tmeas: measurement time
speed of 200 km/h with a 20-ms chirp
Maximum velocity v max = m PRF(4) Pulse repetition frequency (PRF) sequence using a 500-MHz modula-
4 tion bandwidth (0.3-m resolution)
Beat frequency fbeat = B x x: propagation delay makes it move almost four range bins
(5)
Tchirp within the same sequence, limiting
Angular resolution 50.7 (6) Assuming ! 90c FoV and a virtual array the sensitivity benefits of long chirps.
z res =
N TX N rx N TX: transmit antennas The keystone transformation could
N RX: receive antennas mitigate range migration at the cost

46 SPRING 202 2 IEEE SOLID-STATE CIRCUITS MAGAZINE

Authorized licensed use limited to: Richard David. Downloaded on July 27,2022 at 20:18:45 UTC from IEEE Xplore. Restrictions apply.
of latency, memory, and computa- architecture defines the distance Unlocking the Potential
tional power to perform velocity to the waveform front performing of Fast-Chirp FMCW
before range processing. Increasing a matching function (in conceptual Redundancy and elegant demodu-
the transmitter output power enables analogy to power transfer matching) lation underline FMCW’s suitability
a reduced chirp sequence duration of technology resources to the appli- for ranging and velocity detection
for the same signal-to-noise ratio cation. The ramification of this gen- in automotive conditions. Fast-chirp
(SNR), but the associated power dis- eralization is that since today’s radar sequences with large modulation
sipation limits the simultaneous use architectures are optimizing physi- bandwidths offer high range and
of multiple transmitters in the same cal and technology resources mainly velocity resolution for multiple targets
IC, at the cost of angular resolution. with FMCW for range and velocity simultaneously without range-veloc-
Performing a power combination in detection, they have to be reconsid- ity ambiguities. High scan rates create
the air with multiple lower-power ered in view of new requirements a redundancy that allows resolving
transmitters increases the radiation for angular resolution and radar-to- velocity ambiguities in certain use
power at a given direction, reducing radar interference. cases. The demodulation process in
the need for digital processing gain,
at the cost of scanning speed and
more antennas.
To develop a sense of direction
in this multidimensional design
space that extends far beyond the
few examples given in the last para-
graph, let us first generalize the ADC
Radar Baseband

single FMCW transceiver unit with its

RX
copies in Figure 2 into one integral ADC
radar transceiver architecture from
antenna array to digital baseband DAC
output, as in Figure 3(a), which gen- TX
erates the radar cube at a given frame DAC
rate, in analogy to an analog-to-digi-
Freq.
tal converter converting an analog Chirp I/Q Split
Control
signal into bits. The cube perfor-
mance boundaries in Figure 3(b)
are ultimately limited by avail-
(a)
able physical resources, such as the
available spectrum, safety-defined Waveform Front at 76–81 GHz
maximum cycle time for scanning
h
gt
ion n

and processing, and physical sensor

at ele

Waveform Optimization
loc av

space. A waveform front is defined by

Al l/W

fundamental limitations imposed by

tra

Baseband Technology Front

ec

wavelength and noise. Performance Bandwidth RF Performance

Sp

improvements at the waveform front and Efficiency

Velocity

require spending more time, space,

Faster Chirps
power, and spectral resources: a larger Range Innovation
cycle time for integration; sensor place- Gap
e
gl
An

ment offering more space for anten- Directivity

nas; and a smaller wavelength and
Radar Signal
more bandwidth, e.g., moving beyond Processing
e
Siz

100 GHz. Interference

Ra (Sa

or

Mitigation
da fety

The technology front is defined

ns

In-Package
rC )

Se

accordingly by limitations from sili-

yc

Integration
le

con, antenna, and packaging tech-

nologies and can be overcome with
architectural innovation until the (b)
waveform front is reached. In this
conceptual analogy to the Schreier FIGURE 3: (a) The generalized radar transceiver architecture. (b) ­Radar waveform and tech-
FoM in data converters [2], the radar nology fronts.

IEEE SOLID-STATE CIRCUITS MAGAZINE SPRING 202 2 47

Authorized licensed use limited to: Richard David. Downloaded on July 27,2022 at 20:18:45 UTC from IEEE Xplore. Restrictions apply.
(3) separates targets in the frequency what can be offered in today’s prod- and matching, simultaneously reduc-
domain as a function of distance, uct platforms [3]. ing losses and process variability
reducing dynamic range require- Another important aspect is that impact, with profound benefits in the
ments: near and far-away objects it has been made possible to achieve output power or noise figure (NF).
with large size differences (thus, sig- breakthrough output power, noise Process variability, automotive
nal power) are easily separated, and figure, and phase noise with low temperature ranges, reliability, life-
strong reflections from large nearby power and area. This enabled high time operation, and built-in-self-tests
objects and antenna coupling are eas- sensitivity and broke the SiGe ver- (BISTs) for functional safety and
ily removed with filtering. sus CMOS technology partitioning supply isolation constrain transmit
Radar circuit architectures exploit barriers between the front end and lineup power efficiency compared to
the potential of FMCW, maximally baseband, allowing large-scale inte- the state-of-the-art figures reported
leveraging the strengths of CMOS tech- gration and adding more benefits. in literature for power amplifiers [7].
nologies through innovative design. The architectural and performance This makes attention to signal losses
The RF-to-baseband bandwidth com- footprints of highly integrated FMCW critical. To achieve, for example, the
pression t/Tchirp allows for resolution radar architectures [3], [4] driv- required 12-dBm output power per
scaling through the gradual expan- ing market deployment today are transmitter at the antenna port at
sion of the baseband bandwidth and reviewed next and used as pivoting high temperatures with NF 1 13 dB
radar digital processing capabilities point in the following sections. at ADC output in 40-nm CMOS, the
boosted with nm CMOS technologies. RFCMOS offers power efficiency redundancy of constant envelope FM
This is appreciated with help of (2) for advantages compared to SiGe HBT modulation is used, which allows sat-
the maximum unambiguous range technologies for the automotive urated voltage swings for better toler-
using (5). Increasing the modulation radar link budget. Figure 4(a) shows ance to process variations, amplitude
bandwidth B increases the range res- that the NMOS in 40- and 28-nm noise, and nonlinearity. Capacitance
olution, ultimately limited by band CMOS achieves a similar peak FT and neutralization [7] reduces the impact
allocation, while reducing the chirp FMAX as the SiGe HBT devices used in from transistor capacitances, while
duration improves velocity detection, production for mm-wave infrastruc- on-die power combiners and digital
limited by the wavelength and safety- ture products [5] at a much lower calibrations offer power efficiency
defined cycle time. To satisfy the con- current density. Noise performance and precision phase control down
stant maximum range, the Nyquist has improved similarly. Geometric to a couple of degrees, exploiting
frequency needs to keep increasing, tailoring allows the optimization smaller devices. Passive mixers with
which links to CMOS ADC evolu- of mm-wave performance at lower linear MOS switches combined with
tion. For example, completely filling current densities against electro- linear baseband amplifiers achieve
the 76–77-GHz long-range band to migration with partitioning and high linearity to handle strong reflec-
achieve a 15-cm range resolution at staggered connections [5], [6], while tions, and noise is lowered with high-
250 m with an 8.2-µs chirp duration the transformer and power splitter/ resolution ADCs.
translates to the 200-MHz baseband, combiner geometries in Figure 4(b) The ADC challenge in Figure 5 is
which is 10 times more compared to perform impedance transformations to achieve low-power weak-signal

450
40 nm
400 12
28 nm
350 SiGe-HBT
10
FT, FMAX (GHz)

300
Pout (dBm)

250 Source Drain 8

200 FMAX
150 6
#via’s

Matching and Impedance

100 Gate Transformation With Transformer [7]
Bulk 4
50 FT
[US10381447] Matching With LCL Lines
Retraction
2
10–6 10–4 10–2 70 72.5 75 77.5 80 82.5 85
ID or IC (A/µm) Frequency (GHz)
(a) (b)

FIGURE 4: Examples of technology tailoring for active (a) and passive devices (b) in RFCMOS. LCL: inductor capacitor inductor.

48 SPRING 202 2 IEEE SOLID-STATE CIRCUITS MAGAZINE

Authorized licensed use limited to: Richard David. Downloaded on July 27,2022 at 20:18:45 UTC from IEEE Xplore. Restrictions apply.
detection at tens of MHz with spurs reliability for single-layer wafer- array (eWLB) packages, reducing
far below −90 dBFS for very small level chip scale packaging (WLCSP) mm-wave losses from the die to the
input signals and achieve very low and embedded wafer level ball grid PCB further.
1/f noise to avoid ghost targets and
Doppler shifts, respectively. This
requires ADCs highly optimized
to the radar application exploiting
SAR as well as sigma delta concepts,
analog-to-digital domain transforma- Target Cross-Section
tions, dithering, chopping and noise Varies Around 30 dB
shaping, nonlinear transformations,
and multiple calibrations. Such ADCs
are not captured in popular FoMs [8]
but follow the architectural evolution
taking place in this field.
Figure 6(a) shows sensitivity limita-
dBFS

tions stemming from the phase noise 40

dB Small Target Far Away
of the chirp generator at long ranges. pe
rD
eca
In addition to the large size differ- de ADC Spurs
Large Target (Ghost Targets)
ences between targets, phase noise
correlation between transmitted and Noise Floor
received waveforms is reduced, mak-
ing phase noise up to 1-MHz offsets
from the carrier a critical system per- Beat Frequency (= Distance)
formance parameter. Improving phase
noise and making faster and more lin- FIGURE 5: The ADC challenge.
ear chirps brings major design chal-
lenges. Process technology-limited
chirp generation is addressed with
Large Target
frequency transformations exploiting
architectural redundancy. For exam-
ple, in Figure 6(b) [3], chirp generation
Power (dBm)

is compressed around 9 GHz to allow 100-KHz Offset

the use of mode-optimized high-volt-
Thermal Noise 1-MHz Offset
age swing VCOs and thick-oxide tran-
sistors. This optimizes phase noise
Phase Noise
despite the tank quality factor and
Masked
Target

reliability-related voltage swing limi-

tations at mm-wave frequencies.
Amplitude compression thanks
Distance (m)
to FM modulation creates strong
(a)
third harmonics used to reso-
nate and expand the signal back to Frequency Multipliers
76–81 GHz with transformers [9] Charge Pump Filter VCO Tanks
3× 3×
and inherent MOST nonlinearities, 76–81-GHz
reducing 1/f noise penalties. Reduc-
ing power dissipation in combination
with good package thermal properties 8.45–9-GHz
simplifies thermal management at the PFD
Divider
benefit of higher duty cycle, and thus, 25.3–27-GHz
improves sensitivity. Moreover, for
as long as the area remains small, Modulator
Ref.
expensive Teflon-based soft RF sub-
(b)
strates, such as RO3003 and under-
fill or edge bond techniques, are not FIGURE 6: (a) Phase noise sensitivity for long ranges. (b) Architecture of an nm CMOS
required to guarantee board-level fast-chirp synthesizer. Ref.: reference.

IEEE SOLID-STATE CIRCUITS MAGAZINE SPRING 202 2 49

Authorized licensed use limited to: Richard David. Downloaded on July 27,2022 at 20:18:45 UTC from IEEE Xplore. Restrictions apply.
Angular Resolution Expansion receiver. The resolution depends on signal orthogonality, as shown in Fig-
Angular resolution defines the abil- the aperture of the virtual instead ure 8, requiring the sharing of physical
ity of a radar to resolve targets with of the physical array, which is much resources (bandwidth, cycle time, and
different angular positions. Shown smaller. This reduces system com- so on), introducing tradeoffs. TDMA
in Figure 7, it is estimated by apply- plexity substantially. The array pat- ensures transmit orthogonality but
ing reflection path length differ- tern is determined by convolving trades angular resolution with maxi-
ences between the targets and the transmit and receive arrays. Mul- mum unambiguous velocity; it limits
antennas in the array. Resolution is tiple transmitters also improve the maximum velocity within the fixed
inversely proportional to the aper- SNR. Angular resolutions well below cycle time and compromises average
ture of the array. Coherent MIMO 1c require virtual arrays of 192 ele- radiation power using one transmit-
[10] exploits transmit waveform ments or more, e.g., 12 transmitters ter only per FMCW ramp [11]. Dop-
orthogonality to construct a virtual and 16 receivers operating concur- pler division multiple access (DDMA)/
array with a larger aperture than rently at the cost of complexity. slow-time CDMA (ST-CDMA) and FDMA
the physical array. Orthogonality FMCW requires encoding in the [11] allow for simultaneous transmis-
entails that signals from different time, frequency, Doppler, range, or sion, trading angular resolution for
transmitters can be separated at the code domains to achieve transmit maximum velocity or range, and
introduce respective domain ambigui-
ties. Range division MA (RDMA) [12]
Physical Array encodes signals in the range domain,
Direction trading angular resolution for maxi-
of Arrival Different Distance to the
Reflector Results in Phase Shift mum range, and introduces range
ambiguities. Channel coherency limits
nt waveform performance.
e Fro
φ v In other words, waveforms need
Wa
to stay close together in time, fre-
quency, and space; otherwise, targets
Transmit
are not observed in the same way,
Antenna
similar to the range migration that
was introduced earlier. Phase dif-
Receive Antenna Virtual Array ferences are corrupted, which limits
angular dynamic range. In TDMA,
FIGURE 7: The detection of the angle of arrival using a virtual array. this occurs when targets move during

Time Division Multiple Access Doppler and Slow-Time Code Division

Multiple Access
TX Sequentially Activated: 1, 2, 3, ... N
e
od
Frequency

N
e
s N
Frequency

a 1 –1 –1 1
Ph
X
T 2 1 –1 1 –1 1 –1
1 1 1 1 1

Time Time

Frequency and Range Division Multiple Access Fast-Time Code Division (Phase Coded)

TX: 1, 2, 3, ... N

de N 10101..01 10101..01 10101..01

Co
Frequency
Frequency

Frequency

2 110010..0 110010..0 110010..0

1
00111..10 00111..10 00111..10

Time Time Time

FIGURE 8: MIMO-encoding waveforms.

50 SPRING 202 2 IEEE SOLID-STATE CIRCUITS MAGAZINE

Authorized licensed use limited to: Richard David. Downloaded on July 27,2022 at 20:18:45 UTC from IEEE Xplore. Restrictions apply.
antenna multiplexing. In FDMA, the new direction for radar architecture The overlapping FoVs allow for
radar cross-section might change optimization. increased antenna spacing, achieving
with frequency. Alternatives to FMCW 4 # resolution gain that compensates
waveforms, like phase-modulated Closing the Innovation Gap the use of four transmitters per scan
continuous wave (PMCW) [1] (Fig- Pushing the technology front requires out of 12 available. Three-range reso-
ure 8) using binary zero-correlation- optimizing the antenna and trans- lutions and FoVs become simultane-
zone, almost perfect autocorrelation ceiver array in Figure 3(a) holisti- ously available and independent of
sequence [13], or polyphase coding cally with its MIMO waveforms and radar ego-velocity in the radar cycle.
sharing similarities to FMCW [1], [14], signal processing functions also for Time multiplexing multiple DDMA
also have their auto- and cross-corre- the angle of arrival detection. The waveforms optimized for range and
lation properties compromised (long combination with new technologies FoV enabled by fast chirps and high-
codes needed for multiple antennas creates a multitude of innovation resolution phase control generates
and processing gain) due to the time- opportunities, as exemplified later. information that allows resolving
varying nature of the channel, espe- The imaging radar prototype in angular and velocity ambiguities.
cially at high speeds [14]. Figure 9 [15], [16] demonstrates the This leads to up to 1.8c angular reso-
All waveforms inevitably entail use of waveform and radiation pat- lution for long, mid, and short ranges
tradeoffs combining range, velocity, tern redundancy with four 3 # 4 ICs in elevation and 16c
. in azimuth with
and angular domain performance. [3] optimized for TDMA and DDMA up to 30-cm resolution, ! 78 m/s
TDMA, ST-CDMA, DDMA, and RDMA MIMO versatility. Twelve transmit maximum (max) unambiguous veloc-
multiply the need for baseband antennas are partitioned in three sub- ity, and 0.06 m/s resolution with
bandwidth and the reduction of arrays optimized for directivity and fewer than 20 W for the complete sys-
Tchirp, compared to when only range/ FoV for long, mid, and short ranges, tem that delivers the target list.
velocity domains are considered, respectively, combined with 16 receive Several technology and archi-
whereas RDMA also requires active antennas. Directivity minimizes the tecture directions emerge upon
echo cancellation. Besides that, all need for transmitter output power, the inspection of the prototype in
MIMO waveforms require better RF and thus, power dissipation. Energy Figure 9. Operating many transmit-
performance and simultaneous oper- is focused where it is needed in the ters from a single IC could expose
ation of many transmitters. All these observation scene within the radar the sensor board size wavelength
additional requirements together cycle in comparison to using larger limits. This is very challenging
define today’s angular resolution arrays with small gain and wide FoV due to power dissipation and auto-
technology barriers, and thus, set a supporting all three use cases [17]. motive package and IC reliability

Transceiver MIMO Waveform System Integration

Phase Noise: –94 dBc/Hz at 1 MHz TDMA 12 TX × 16 RX

Imaging Radar Prototype
DDMA
Frequency (MHz)

Frequency
4 GHz

TX Antennas
LRR MRR SRR

16.8 µs Time Time RX Antennas

Antenna Array Waveguide

INL (1 LSB = 5.625°) Launchers
2 75 m 150 m 250 m
76 GHz 79 GHz 81 GHz IC Array
1
LSB

0
Phase Code 0 to 64 S32 µ Cs
–1

–2
0.3 m 0.6 m 1.2 m
Chirp Synthesizer/Modulator
Example

FIGURE 9: An example of a multimode TDMA/DDMA imaging radar [15].

IEEE SOLID-STATE CIRCUITS MAGAZINE SPRING 202 2 51

Authorized licensed use limited to: Richard David. Downloaded on July 27,2022 at 20:18:45 UTC from IEEE Xplore. Restrictions apply.
constrains. Wireless optimized tech- radar signal processing in the need of surface waves achieving 19-dBm
nologies able to improve mm-wave for aggressive CMOS scaling. Similar EIRP per antenna while eliminating
performance and power efficiency to waveforms and circuit architec- the antenna-to-mm-wave die inter-
while offering enhanced digital inte- tures, the solution may then be in face losses. Achieving high antenna
gration capabilities could enable the combination of multiple tech- efficiency within the boundaries of
steps in this direction. For example, nologies in the same package. packaging technologies constrains
FD-SOI [18] offers device stacking, New packaging technologies offer these solutions to short ranges. For
back-gate control, and potentially multiple metal layers that can allow longer ranges, integrated waveguide
lower noise that could help improve the integration of passive compo- launchers in packages (LiPs), shown in
efficiency and manage process nents, antennas, and waveguide Figure 10(b), can couple the RF signal
variations. mm-Wave-optimized Fin- launchers in the package. This can directly from the die to the metallic
FET technologies also offer a high- lead to the further reduction of losses waveguide antenna. This allows for
frequency potential [19]; however, in the path between the die to the high directivity and pattern shaping
thermal self-heating, increased gate antenna and ease system manufactur- optimization, reducing transition and
noise, and higher losses due to fin ing. For example, the 3 # 4 antenna- alignment losses, increasing band-
capacitances have to be addressed. in-package (AiP) prototype shown width, and simplifying the footprint
At the same time, larger arrays and in Figure 10(a) [16] uses artificial area around the IC, as shown in Fig-
higher resolution translate to more dielectric layers to reduce the impact ure 10(b). Optimizing directivity can

Artificial Dielectric Concept

Multilayer PCB Frontside RDL

M1 Mold
M2 Si
M3

Backside RDL

External PCB
(a)

TX Array RX Array
Waveguide WG Antenna
Reference
Plane
Die
Reference RO3003
Plane eWLB, External Waveguide Launcher

Waveguide TX Array RX Array

Reference WG Antenna
Plane
Die
Reference
Plane FR4

eWLB-LiP, Embedded Waveguide Launcher

(b)

FIGURE 10: In-package integration examples. (a) An AiP utilizing a multilayer eWLB package and (b) an embedded waveguide launcher (LiP)
in the eWLB package. RDL: redistribution layer; WG: waveguide.

52 SPRING 202 2 IEEE SOLID-STATE CIRCUITS MAGAZINE

Authorized licensed use limited to: Richard David. Downloaded on July 27,2022 at 20:18:45 UTC from IEEE Xplore. Restrictions apply.
further be done with feeding networks substantial sensitivity loss that can The combination with the waveform
[20], whereas mm-wave switches [17] prove critical in certain situations. A diversity offered by today’s regulatory
can better utilize the available anten- real-life measurement example dem- framework introduces several degrees
nas, as is already practiced in wireless onstrating this effect in the range of freedom. The radar receiver in Fig-
infrastructure fields. Doppler domain is shown in Figure ure 2 provides, at the output of the
At the circuit architecture level, 11(a). An aggressor chirp with a dif- down-converting mixer, the instanta-
reducing the receiver noise figures ferent chirp profile compared to the neous frequency difference between
and 1/f noise with quadrature archi- victim one causes an increase of the any aggressor and own (victim) trans-
tectures and better analog baseband noise floor, limiting the dynamic mit signal (typically uncorrelated).
circuit topologies can extend long- range (blue area in the plot becom- This demodulation mechanism is
range and Doppler performance. Bet- ing lighter), and introduces two ghost shown in Figure 11(b). For two FMCW
ter noise figures can also be used to Doppler lines at 30 km/h across all signals (e.g., red and blue), the result-
reduce transmit power. Functional ranges. For example, this could be an ing baseband signal is a burst that
integration with mixer-DAC transmit urban scenario with a car no longer resides inside or outside the radar
modulators and more calibrations [21] able to see a bicycle approaching. passband dependent on chirp slopes,
offers a pathway for versatile MIMO There are several attributes mak- relative chirp starting timing, starting
waveform synthesis combining time, ing radar-to-radar interference a chirp frequencies, passband width,
frequency, and phase domains, which complex problem that requires solu- antialiasing properties, and chirp digi-
is a prerequisite for advanced MIMO tions at multiple layers of physical tal encoding [27].
schemes. As the array grows larger, hierarchy. In contrast to radar reflec- Adapting these properties with
IC-to-IC synchronization with chirps tions, radar interference undergoes versatile waveform generators can
generated centrally forms power, one-way path loss and can occur thus reduce interference probability
noise, and linearity bottlenecks. More for an arbitrary time duration with or position it out of band. By posi-
integration and decoupling mm-wave chirps having any frequencies within tioning, for example, the blue chirp
carrier synchronization from radar the 76–81-GHz band at any repetition in Figure 11(b) in another subband
modulation through digital wave- rate relative to the victim radar. This (green chirp), the interference stem-
form synthesis, as is done in commu- maps to scenarios of vehicles and ming from the red chirp can be
nication fields, could be the way to other targets changing their relative avoided for as long as there is suf-
address it. positions and chirp radiation pro- ficient modulation bandwidth avail-
Finally, advances in the CMOS ADC files on the scene. Mitigating interfer- able. The unlikely case of correlated
frontier with wider bandwidths and ence at the circuit/antenna domain aggressor victim transmissions
lower spurs [22], combined with novel requires a very high dynamic range results in false targets that can be
chirp synthesizers, e.g., [23], will con- at the receiver and the ability to per- handled by the tracking filter. The
tinue to push resolution expansion. As form the fast and reliable detection impact of interference in the range
the chirp duration keeps decreasing, of aggressors [26] within the radar Doppler map also relates to modu-
offering higher and higher numbers cycle to drive adaptive gain control, lation [27], leading to major coexis-
of simultaneously operating transmit- active cancellations, and spatial tence challenges between PMCW and
ters within coherence limits, the lim- antenna radiation filtering, in anal- FMCW in the same band.
ited time overlap between transmit ogy to techniques used in wireless A fundamentally different strategy
and receive chirps (e.g., zero overlap infrastructure domains. at a higher abstraction layer avoids
with a 2-µs chirp for 300-m detection) These techniques introduce new interference in the first place with
will bring new sensitivity limitations, tradeoffs on sensitivity, scanning the standardization of the channel
introducing the final frontier of the speed, additional antenna count, access while guaranteeing fairness.
FMCW waveform. and so forth. Removing corrupted A possible approach could be to stati-
samples in the digital domain can be cally allocate bandwidth and time
What About Radar-to-Radar very effective as well when the inter- resources. For instance, front and
­Interference? ference duration is very short. More rear radars could use nonoverlapping
Radar-to-radar interference is a grow- advanced techniques analyze signal parts of the spectrum or different
ing challenge [24], [25] for the auto- characteristics, e.g., using the short- polarization to mitigate interference
motive industry. Radar radiation of time Fourier transform or wavelet to each other. Access rules borrowed
around 10 dBm or more from a single transform to separate the signal and by telecom could also be used as a
aggressor transmitter at short ranges interferer more accurately. starting point. Deterministic channel
with typical antenna gains of 10–20 dB Reliable detection with listen- access requires centralized coordi-
directed at the main radiation lobe ing modes around the car can also nation, e.g., via telecom infrastruc-
into the passband of a victim receiver allow interference-free transmis- ture in 6G telecom systems defining
may blind it temporarily or cause sion at the right time and frequency. what to do and receiving access to

IEEE SOLID-STATE CIRCUITS MAGAZINE SPRING 202 2 53

Authorized licensed use limited to: Richard David. Downloaded on July 27,2022 at 20:18:45 UTC from IEEE Xplore. Restrictions apply.
 45 45

25 25

Speed (km/h)

Speed (km/h)
No Interferer Interferer
0 0

Target Target
–25 –25

–45 –45
0 10 20 30 0 10 20 30
Distance (m) Distance (m)
r (a)

Victim Chirp
ere

Passband ir p
Ch
erf

Aggressor Chirp
d
Int

itte et Victim Chirp in New Band

sm
Frequency

n rg
ar
y
Tra Ta
Chirp

d
un

Chirp Frequency
Bo
sing
lia y
rA da
r
pe un
Up Bo
ing
Passband

il as
A
er
Low

Time Time

Target
Baseband
Amplitude

Time

Out of Band Out of Band

(Filtered) In Band (Filtered)
(b)

FIGURE 11: (a) An interference example and (b) an interference demodulation mechanism with chirp frequency versus time plan.

time and frequency slots to be used, ms by 250 MHz) and exploiting syn- regulators and incumbent stakehold-
also implying waveform restrictions. chronization from GPS signals. ers to define the access rules for auto-
Alternative stochastics approaches At a certain moment, there will be motive radars above 100 GHz.
deployed at the medium access a need for more spectral resources and
control (MAC) layer of the communi- a step to a new band beyond 100 GHz. Acknowledgments
cations stack [ALOHA, carrier sense Smaller wavelengths pave a way to The authors would like to express
multiple access (CSMA), CSMA-colli- sub-THz imaging crossing radar, their appreciation to numerous col-
sion avoidance, and so on] are distrib- camera, and lidar boundaries and leagues at NXP Semiconductors who
uted and could still allow sensors to offer new classes of information and have contributed to the activities
follow the same transmission proto- potential to reduce sensor size. Along supporting this article, with special
col that ensures fair channel access. with all that come many technology mention to W. Syed, G. Carluccio, T.V.
One could also think to standardize challenges requiring the extension Dinh, A. de Graauw, both E. Janssen
only the way the channel is accessed, of the presented design approaches. and E. Janssen, S. Thuries, M. Lont,
organizing the channel in integer Currently, the automotive industry F. Tiemeijer, C. Vaucher, F. Laghezza,
time and frequency blocks (e.g., 20 is working actively with European and A. Turley.

54 SPRING 202 2 IEEE SOLID-STATE CIRCUITS MAGAZINE

Authorized licensed use limited to: Richard David. Downloaded on July 27,2022 at 20:18:45 UTC from IEEE Xplore. Restrictions apply.
References [17] V. Giannini et al., “9.2 A 192-virtual-receiv-
er 77/79GHz GMSK code-domain MIMO ra-
performance data converters for wire-
[1] M. A. Richards et al., Principles of Modern
Radar. Edison, NJ: SciTech, 2010. dar system-on-chip,” in Proc. IEEE Int. Sol- less and cable modem applications.
[2] G. Manganaro, Advanced Data Convert- id-State Circuits Conf., 2019, pp. 164–166, Since 2012, he has focused on mm-
ers. Cambridge, U.K.: Cambridge Univ. doi: 10.1109/ISSCC.2019.8662386.
Press, 2012. [18] S.N. Ong et al., “22nm FD-SOI technology wave CMOS transceivers for automo-
[3] “Radar transceivers,” NXP Semiconductors, with back-biasing capability offers excel- tive radars. He is the (co-) author of
Eindhoven, The Netherlands, TEF82xx, Apr. lent performance for enabling efficient,
2022. [Online]. Available: www.nxp.com/ ultra-low power analog and RF/millime- multiple papers, patents, and books
products/radio-frequency/radar-transceivers ter-wave designs,” in Proc. IEEE Radio on data converters and RFCMOS radar.
[4] K. Dandu et al., “2.2 high-performance and Freq. Integr. Circuits Symp., 2020, pp. 323–
small form-factor mm-wave CMOS radars 326, doi: 10.1109/RFIC.2019.8701768. He was an associate editor for the
for automotive and industrial sensing in [19] Q. Yu et al., “mmWave and sub-THz tech- IEEE Transactions of Circuits and Sys-
76-to-81GHz and 57-to-64GHz bands,” nology development in Intel 22nm FinFET
in Proc. IEEE Int. Solid State Circuits Conf. (22FFL) process,” in Proc. Int. Electron De- tems, a Technical Program Committee
(ISSCC), 2021, pp. 39–41, doi: 10.1109/ vices Meeting, 2020, pp. 17.4.1–17.4.4, doi: member, and a European Union (vice)
ISSCC42613.2021.9365838. 10.1109/IEDM13553.2020.9372105.
[5] P. Magnee, D. Leenaerts, M. Van der Hei- [20] J. Xu, W. Hong, H. Zhang, G. Wang, Y. Yu, chair and forum committee member.
jden, T. V. Dinh, I. To, and I. Brunets, “The and Z. H. Jiang, “An array antenna for Alessio Filippi (alessio.filippi@
future of SiGe BiCMOS: Bipolar amplifiers both long- and medium-range 77 GHz
for high-performance millimeter-wave automotive radar applications,” IEEE nxp.com) received his telecommuni-
applications,” in Proc. IEEE BiCMOS Com- Trans. Antennas Propag., vol. 65, no. 12, cation engineering degree from the
pound Semicond. Integr. Circuits Tech- pp. 7207–7216, Dec. 2017, doi: 10.1109/
nol. Symp. (BCICTS), 2021, pp. 1–7, doi: TAP.2017.2761549. University of Padova, Italy, in 2002,
10.1109/BCICTS50416.2021.9682483. [21] M. Neofytou, P. Athanasiadis, M. Ganzerli, and received his Ph.D. degree from
[6] L. F. Tiemeijer, R. J. Havens, N. Pavlovic, M. Lont, G. I. Radulov, and K. Doris, “A
and D. M. W. Leenaerts, “Record Q sym- novel 2-dimensional correction method the Technical University of Kaiser-
metrical inductors for 10-GHz LC-VCOs in for mm-wave cartesian I/Q modulators,” slautern under the supervision of
0.18-μm Gate-length CMOS,” IEEE Electron in Proc. IEEE Int. Symp. Circuits Syst.
Device Lett., vol. 23, no. 12, pp. 713–715, (ISCAS), 2021, pp. 1–5, doi: 10.1109/IS- Prof. P.W. Baier in 2005 as an exter-
Dec. 2002, doi: 10.1109/LED.2002.805740. CAS51556.2021.9401408. nal Ph.D. candidate from Siemens,
[7] D. Zhao and P. Reynaert, “An E-band [22] M. Bolatkale et al., “A 28nm 6GHz 2b con-
power amplifier with broadband par- tinuous-time ΔΣ ADC with –101dBc THD Munich, Germany. In 2005, he joined
allel-series power combiner in 40-nm and 120MHz bandwidth using digital DAC Philips Research to continue work-
CMOS,” IEEE Trans. Microw. Theory Techn., error correction,” in Proc. Int. Solid State
vol. 63, no. 2, pp. 683–690, Feb. 2015, doi: Circuits Conf., 2022, pp. 416–417. ing on modulation and demodulation
10.1109/TMTT.2014.2379277. [23] H. Shanan et al., “A 0-to-12GHz coupled- techniques for wireless communica-
[8] E. Janssen, “Analog-to-digital converters RTWO FMCW ADPLL with 97fs RMS jitter,
for automotive radar,” in Forum F6 The -120dBc/Hz PN at 1MHz Offset, and with tions, with a particular focus on syn-
Right Tool for the Job: Application-Opti- retrace time of 12.5nm and 2μs chirp set- chronization and channel estimation
mized Data Converters, Int. Solid State Cir- tling time,” in Proc. Int. Solid State Circuits
cuits Conf. (ISSCC). Conf., 2022, pp. 146–147. algorithms for broadcast multi-car-
[9] Z. Zong, M. Babaie, and R. B. Staszewski, [24] W. Buller et al., “Radar congestion study,” rier systems. Since 2012, he has been
“A 60 GHz frequency generator based on National Highway Traffic Safety Admin-
a 20 GHz oscillator and an implicit multi- istration, Washington, DC, USA, Rep. with NXP Semiconductors, Eind-
plier,” IEEE J. Solid-State Circuits, vol. 51, No. DOT HS 812 632, Sep. 2018. [Online]. hoven, 5600 KA, The Netherlands,
no. 5, pp. 1261–1273, May 2016, doi: Available: https://www.nhtsa.gov/sites/
10.1109/JSSC.2016.2528997. nhtsa.dot.gov/files/documents/13790 and has extended his research inter-
[10] J. Li and P. Stoica, “MIMO radar with co- _radarstudy_092518_v2b-tag.pdf ests to car-to-car communications
located antennas,” IEEE Signal Process. [25] M. Kunert, “MOre safety for all by radar
Mag., vol. 24, no. 5, pp. 106–114, 2007, interference mitigation,” European Com- and automotive radar.
doi: 10.1109/MSP.2007.904812. mission, Luxembourg City, Luxembourg, Feike Jansen (Feike.Jansen@nxp.
[11] F. Jansen, “MIMO radar waveforms for au- Final report, Tech. Rep. 248231, 2010. [On-
tomotive,” in Proc. Automot. Forum, Eur. line]. Available: https://cordis.europa.eu/ com) received his degree in electri-
Radar Conf., 2021. project/rcn/94234/reporting/en cal engineering from the Technical
[12] G. Babur, O. A. Krasnov, A. Yarovoy, and [26] F. Laghezza, F. Jansen, and J. Overdevest,
P. Aubry, “Nearly orthogonal waveforms “Enhanced interference detection method University of Eindhoven, Eindhoven,
for MIMO FMCW radar,” IEEE Trans. Aero- in automotive FMCW radar systems,” in The Netherlands, in 2006. In 2006,
sp. Electron. Syst. (1965–present), vol. 49, Proc. Int. Radar Symp. (IRS), 2019, pp. 1–7,
no. 3, pp. 1426–1437, Jul. 2013, doi: doi: 10.23919/IRS.2019.8767459. he joined Philips, working in the field
10.1109/TAES.2013.6557996. [27] J. Overdevest, F. Jansen, F. Laghezza, F. of 24-GHz radar antennas. In 2007,
[13] J. Overdevest, F. Jansen, F. Uysal, and A. Uysal, and A. Yarovoy, “Uncorrelated in-
Yarovoy, “Doppler influence on waveform terference in 79 GHz FMCW and PMCW he joined NXP Semiconductors, NXP
orthogonality in 79 GHz MIMO phase- automotive radar,” in Proc. Int. Radar Semiconductors, Eindhoven, 5600 KA,
coded automotive radar,” IEEE Trans. Veh. Symp. (IRS), 2019, pp. 1–8, doi: 10.23919/
Technol., vol. 69, no. 1, pp. 16–25, Jan. IRS.2019.8768181. The Netherlands, where he defined
2020, doi: 10.1109/TVT.2019.2951632. the signal processing techniques for
[14] M. Jamil, H. Zepernick, and M. I. Pet-
tersson, “Performance assessment of
About the Authors a single-carrier, 60-GHz wireless com-
polyphase pulse compression codes,” in Kostas Doris (Kostas.doris@nxp. munication system with frequency-
Proc. IEEE Int. Symp. Spread Spectr. Techn.
Appl., 2008, pp. 166–172, doi: 10.1109/
com) received his Ph.D. degree from domain equalization. Since 2012, he
ISSSTA.2008.37. Technical University of Eindhoven has been working on MIMO coding
[15] F. Jansen et al., “SimultaneousRemove
IMR3 multi-mode automotive imaging ra-
in 2004. He is a fellow at NXP Semi- techniques and direction of arrival
dar using cascaded transceivers,” in Proc. conductors, Eindhoven, 5600 KA, estimation algorithms for automotive
Eur. Radar Conf., London, U.K., 2021. [On-
line]. Available: https://www.eumw2021.
The Netherlands, and professor at radar systems. He has been involved
com/docs/2021_programme.pdf Technical University of Eindhoven, in the definition and development of
[16] K. Doris et al., “Mm-wave automotive
radar: From evolution to revolution,” in
5600 KA, The Netherlands. He worked multiple radar reference designs and
Proc. IEEE Int. Electron Devices Meeting, at Philips Research Laboratories IC architectures.
2021, pp. 25.7.1–25.7.4, doi: 10.1109/ 
IEDM19574.2021.9720646.
and NXP Research, developing high

IEEE SOLID-STATE CIRCUITS MAGAZINE SPRING 202 2 55

Authorized licensed use limited to: Richard David. Downloaded on July 27,2022 at 20:18:45 UTC from IEEE Xplore. Restrictions apply.