Stephen Alland, Wayne Stark, Murtaza Ali, and Manju Hegde
Characteristics, mitigation techniques, and current and future research
Characteristics, mitigation techniques, and current and future research
©ISTOCKPHOTO.COM/TALAJ
Tmechanisms and characteristics of interference and the effects of interference on radar system performance are described. The interference-to-noise ratio (INR) at the output of a detector is a measure of the susceptibility of a radar to interference. The INR is derived from different types of interfering and victim radars and depends on the location of both as well as parameters such as transmit power, antenna gain, and bandwidth. In addition, for victim radar with beamscanning, INR depends on the location of the target the victim radar is attempting to detect. Analysis is presented to show the effects of various interference scenarios on the INR. A review of the current state of the art in interference mitigation techniques previously deployed as well as areas of research currently being addressed is then provided. Finally, important future research directions are suggested. Vehicular sensors Sensors for vehicular (automobiles, buses, trucks, and so on) \
ferent types of currently used sensors being considered for use in future vehicles, such as radars, cameras, lidars, and ultrasonics. Each of these sensors has strengths and weaknesses; good engineering judgment indicates that a combination of sensors, \which complement the strengths and weaknesses of one with ©ISTOCKPHOTO.COM/TALAJ another, is required to maintain the integrity of safety-critical systems. For instance, radars are the best sensors for detecting range and radial velocity and have “all-weather” capability, but are weak for classification and angular resolution. Lidars, in general, have good angular resolution and range, but are limited in field of view (FOV) and have limited ability in adverse weather. Cameras have excellent color perception and classification capabilities but are limited in estimating velocity and range. Cameras also have difficulty in dark or adverse weather. There are several different performance measures to consider when evaluating the vehicular sensors used to detect objects in the environment. A core set includes detection range,
applications are an important area of R&D. There are many different types of currently used sensors being considered for use \
1053-5888/19©2019IEEE detection for reading signs and traffic lights. In addition, the cost of a sensor is important to consider; a vehicle might need, and thus be equipped with, multiple sensors. The cost of an individual
detection for reading signs and traffic lights. In addition, the cost of a sensor is important to consider; a vehicle might need, and thus be equipped with, multiple sensors. The cost of an individual
thus be equipped with, multiple sensors. The cost of an individual sensor is typically affected by a number of factors: the scan type (mechanical, solid state/electronic, or digital), radio-frequency
(RF) circuitry, baseband and digital chipsets, the integration level of fundamental transmit and receive components, digi-
(RF) circuitry, baseband and digital chipsets, the integration level of fundamental transmit and receive components, digi-
level of fundamental transmit and receive components, digital processing subsystem capabilities (bandwidth, throughput, and available memory), the associated packaging and required
and available memory), the associated packaging and required manufacturing processes, and the number of antennas used for transmission and reception. Figure 1 shows the strengths and weaknesses of different sensor systems in relation to different performance measures [1], [2]. Note that the spider plots in the figure illustrate characteristics of radar and vision systems as deployed today in millions of production automobiles, whereas for lidar, they show the potential characteristics—lidars have not been deployed in production automobiles to date.
been deployed in production automobiles to date. The considerable sensor requirements for self-driving cars described previously dictate that multiple sensor modalities
described previously dictate that multiple sensor modalities will be present and that radar will be an important part of that
will be present and that radar will be an important part of that portfolio. Radars can be used to determine location and directly measure the Doppler velocity of objects in the environment. Furthermore, with the ongoing development of RF CMOS and multiple-input, multiple-output (MIMO) radar imaging technologies at 76–77 GHz and 77–81 GHz, automotive radar cost is rapidly decreasing, while overall performance and capability of radars for point cloud imaging, edge detection, and target classification will be substantially improved in the near future. There are multiple automotive applications for radar sensors; consequently, automotive radar is an active research field [3]–[7]. Automatic cruise control, blind-spot detection, and
46
[3]–[7]. Automatic cruise control, blind-spot detection, and collision-warning systems were some of the earliest applications of vehicular radar systems. Recently, other applications, including advanced driver-assistance systems, automatic emergency braking, lane-change assist, and vulnerable user detection have been implemented. Self-driving cars will increase Range × FOV 10
detection for reading signs and traffic lights. In addition, the cost sors on vehicles will increase dramatically in the next 10 years. of a sensor is important to consider; a vehicle might need, and This makes the possibility of radar-to-radar interference in thus be equipped with, multiple sensors. The cost of an individual traffic much greater, as noted in [8] and [9]. sensor is typically affected by a number of factors: the scan type The problem of radar-to-radar interference will be a sig- (mechanical, solid state/electronic, or digital), radio-frequency nificant engineering challenge that the industry will have to (RF) circuitry, baseband and digital chipsets, the integration address. There are several signal processing techniques that
(RF) circuitry, baseband and digital chipsets, the integration address. There are several signal processing techniques that level of fundamental transmit and receive components, digi-mitigate interference and some of these have already been tal processing subsystem capabilities (bandwidth, throughput, implemented in radars deployed in today’s automobiles. With and available memory), the associated packaging and required the increasing number of vehicles being equipped with radar manufacturing processes, and the number of antennas used for and each vehicle having multiple radar sensors—several protransmission and reception. Figure 1 shows the strengths and duction models commercially available in the 2021–2023 time weaknesses of different sensor systems in relation to different frame are investigating the possibility of deploying up to six performance measures [1], [2]. Note that the spider plots in the radar systems per car—the capability of radar systems to operfigure illustrate characteristics of radar and vision systems as ate in the presence of other radar systems in proximity is fast deployed today in millions of production automobiles, whereas becoming a critical performance issue. As a result, interest in for lidar, they show the potential characteristics—lidars have not this area of research has increased significantly. For future radars, the ability to mitigate interference will be as critical as The considerable sensor requirements for self-driving cars detection performance. described previously dictate that multiple sensor modalities will be present and that radar will be an important part of that Background portfolio. Radars can be used to determine location and direct-Radar systems operate by transmitting a signal; this signal is
Furthermore, with the ongoing development of RF CMOS and radar system receiving the reflected signal compares the propmultiple-input, multiple-output (MIMO) radar imaging tech-erties of the reflected signal to the transmitted signal [10]. A nologies at 76–77 GHz and 77–81 GHz, automotive radar cost radar system with a single transmitter and receiver can estiis rapidly decreasing, while overall performance and capability mate the range and velocity of an object in the environment; of radars for point cloud imaging, edge detection, and target with mechanical or electronic scanning, the radar system can classification will be substantially improved in the near future. estimate the angle as well. A radar system with multiple re- There are multiple automotive applications for radar sen-ceiver antennas can estimate the angle of an object via digital sors; consequently, automotive radar is an active research field beamforming. With both multiple transmitter antennas and [3]–[7]. Automatic cruise control, blind-spot detection, and multiple receiver antennas, also known as a MIMO system, a collision-warning systems were some of the earliest applica-radar can estimate the angle of an object via digital beamformtions of vehicular radar systems. Recently, other applications, ing but with enhanced accuracy and resolution compared to including advanced driver-assistance systems, automatic emer-more conventional (non-MIMO) radar [4], [6]. In general, radar gency braking, lane-change assist, and vulnerable user detec-can determine different object angles (e.g., azimuth and/or eltion have been implemented. Self-driving cars will increase evation) depending on the antenna scanning and/or the number and geometry of antenna-receiver channels. Radar systems can be designed for different performance objectives. Some of the performance measures of a typical radar system include detection range, range resolution, maxi-
ly measure the Doppler velocity of objects in the environment. then reflected by an object or target in the environment. The Furthermore, with the ongoing development of RF CMOS and radar system receiving the reflected signal compares the prop-
radar system include detection range, range resolution, maximum unambiguous range, velocity resolution, maximum unambiguous velocity, angular resolution, and FOV. An overview of various types of radars and estimation techniques for range, Resolution velocity, and direction in the absence of interference are given Velocityin [11]. Generally, with a fixed number of antennas, a radar with Detection a broad FOV can be obtained at the expense of less angular resolution, while a narrow FOV can provide better angular resolu- Processingtion. The antenna configuration will determine the estimated target direction in the horizontal plane (azimuth), vertical plane (elevation), or in both. Multiple antennas at the transmitter and receiver can be used to beamform the signal or can be used in a MIMO configuration where different waveforms are transmitted from different antennas and the subsequent receive signals are processed to form a synthetic or virtual receive array. Early IEEE SIGNAL PROCESSING MAGAZINE | September 2019 |
tion capability [12], [13]. Automotive radar is now beginning to incorporate MIMO antennas configured for 2D angle-detection capability (azimuth and elevation), with future systems moving to a greater number of transmit and receive antennas to support 2D point cloud imaging. Fundamentally, the performance of a radar system depends on the bandwidth of the signal, the time duration over which the estimation is performed, and the geometry of the transmit-
the estimation is performed, and the geometry of the transmitter and receiver antennas. A tutorial on its performance in the absence of interference is discussed in detail in [11]. This article is focused on the effect of an interfering radar on a victim radar. Frequency-modulated continuous wave One type of radar, and the most common in current automotive
systems, is frequency-modulated continuous-wave (FMCW) radar. In FMCW radar, the transmitted signal is a sinusoidal
radar. In FMCW radar, the transmitted signal is a sinusoidal signal in which the frequency of the signal varies with time. The transmitted signal is st Tt ()=+ 22 Pf cos (( r cm ft ()))t ,
s_{T}(t)=\sqrt{2P_{t}},\ s(2\pi(f_{c}!+!f_{m}(t))),
is the transmitted power, () is the time-varying frequency. The frequency of the
transmitted signal is then ft Tc ()=+ ff m()t . There are different ways in which the frequency of an FMCW waveform varies. One way is via a ramp or sawtooth signal in which the frequency ramps from a minimum frequency to a maximum and then repeats. Alternatively, a signal that sweeps up in frequency (i.e., up-chirp) and then down in frequency (i.e., down-chirp) could be employed. Slopes for the up-chirp and down-chirp could vary in time as well. The time it takes for the up-chirp signal of the receiver to sweep over a bandwidth B Hz in frequency is sometimes called the chirp duration and is denoted by Tc. In some cases, the frequency changes in a piece-wise linear fashion, although the times for the up-chirp and down-chirp can be different. At the receiver, the received signal is mixed with (multiplied by) the transmitted signal and filtered with a low-pass
B_{r}
P_{t}
f_{c}
tion capability [12], [13]. Automotive radar is now beginning to Additional processing is done after converting from analog to incorporate MIMO antennas configured for 2D angle-detection digital signals. The filter bandwidth limits the unambiguous capability (azimuth and elevation), with future systems moving range of an FMCW radar. The transmitter and receiver block to a greater number of transmit and receive antennas to support diagram for an FMCW radar is shown in Figure 2. The transmitted signal is reflected off of a target and Fundamentally, the performance of a radar system depends received. The received signal, in the case of a stationary taron the bandwidth of the signal, the time duration over which get, is an attenuated and time-shifted version of the transmitted
f_{m}(t)
f_{T}(t)=f_{c}+f_{m}(t)
R=\tau c/2\mathrm{{ o r}}\tau=2R/c
on the bandwidth of the signal, the time duration over which get, is an attenuated and time-shifted version of the transmitted the estimation is performed, and the geometry of the transmit-signal. Based on propagation of signals at the speed of light, ter and receiver antennas. A tutorial on its performance in the the delay x is related to the range by Rc = x /2 or x = 2Rc absence of interference is discussed in detail in [11]. This article This is true for any type of radar signal, not just FMCW. is focused on the effect of an interfering radar on a victim radar. The result of mixing the FMCW-transmitted signal with the received signal will be a signal with frequency proportional to the delay between the radar and the target. Filtering after One type of radar, and the most common in current automotive mixing limits the range of targets that can be detected. The fresystems, is frequency-modulated continuous-wave (FMCW) quency of the signal after mixing and filtering is proportional radar. In FMCW radar, the transmitted signal is a sinusoidal to the delay between the transmitter and receiver and thus prosignal in which the frequency of the signal varies with time. portional to the range of the target. Suppose Rmax is the range of the furthest target to be detected. Then the corresponding maximum delay is xmax = 2Rmax /. c The maximum frequency t , (1) shift is then fB maxm = 2 Rc ax /( Tc). The minimum frequency shift is 0 corresponding to a target at distance 0. The filter must is the transmitted power, have a bandwidth at least as large as this maximum beat fre- () is the time-varying frequency. The frequency of the quency. Equivalently, if the low-pass filter has bandwidth B t . There are dif-the largest range that can be detected is Rc max = TB cr ferent ways in which the frequency of an FMCW waveform For example, if the chirp duration is 30 μs, the sweep bandvaries. One way is via a ramp or sawtooth signal in which width is 300 MHz and the filter bandwidth is Br = 15 MHz, the frequency ramps from a minimum frequency to a maxi-then the maximum range is 225 m. For automotive applicamum and then repeats. Alternatively, a signal that sweeps up tions, this might be considered a long-range radar (LRR). If in frequency (i.e., up-chirp) and then down in frequency (i.e., the sweep bandwidth is 750 MHz, the sweep time is 50 μs and down-chirp) could be employed. Slopes for the up-chirp and the IF bandwidth is Br = 45. MHz, then the maximum range down-chirp could vary in time as well. The time it takes for is 45 m and would be considered a short-range radar (SRR). the up-chirp signal of the receiver to sweep over a bandwidth Generally, a larger bandwidth B will enable better range reso- B Hz in frequency is sometimes called the chirp duration and lution in a radar. An SRR would typically require better range . In some cases, the frequency changes in a resolution than an LRR and thus, use a larger bandwidth. piece-wise linear fashion, although the times for the up-chirp Phase-modulated continuous wave At the receiver, the received signal is mixed with (multi-Another type of radar is phase-modulated continuous-wave
T.
B_{r}
\tau_{\operatorname*{m a x}}=2R_{\operatorname*{m a x}}/c
f_{\mathrm{m a x}}=2B R_{\mathrm{m a x}}/(c T_{c})
R_{\operatorname*{m a x}}=c T_{c}B_{r}/(2B)
IEEE SIGNAL PROCESSING MAGAZINE | September 2019 |
30~\mu\mathrm{s}.
B_{r}=15~\mathrm{M H z},
B_{r}=4.5~\mathrm{M H z},
rGamma,
Transmitter st Tt ()=+ 22 Pf cos (( rz cm tt)),
\begin array}{r}{\mathfrak{s} {T}(t)=\sqrt{2P{t}}\cos(2\pi f_{c}t+\phi_{m}(t)),}\end{array}
t is the modulated phase waveform. The total phase Tc =+ tt (). One way
of generating the phases is to begin with what is known as a spreading code. A spreading code consists of a sequence of chips (e.g., +1, +1, –1, +1, –1, … ) with a chip duration Tc, which is mapped (e.g., +-10 "", 1 r) into a sequence of phases (e.g., 0, 0, r, 0, r, … ), and the phases are used to modulate the RF sinusoidal signal. The phase can be limited to either 0 or r radians (180°) or it can be arbitrary. In the event that the signal phase is only either 0 or r radians, the signal is said to be a binary phase modulated signal. A binary phase modulated signal can also be generated by multiplying a binary (+1 and –1) signal at () with a carrier. In the case of binary spreading codes, the transmitted signal can be written as st Tt ()= 22 Pa()tf cos () r ct . The spreading code at () could be a periodic sequence with a
\phi_{m}(t)
\phi_{T}(t)=2\pi f_{c}t+\phi_{m}(t)
(\mathtt{e.g.},+1,+1,-1,+1,-1,\dots)
T_{c}
+1\to0,-1\to\pi)
(\mathtt{e.g.},0,0,\pi,0,\pi,\dots)
(180^{\circ})
s_{T}(t)=\sqrt{2P_{t}},a(t)\cos(2\pi f_{c}t).
long period so that it appears to be a nearly random sequence. Codes or sequences with good autocorrelation properties are important for use in PMCW-type radar systems. There are many possible spreading codes, including Barker sequences, m-sequences (also known as linear feedback shift register sequences), and gold codes, all of which are binary codes. There are also nonbinary codes such as the Zadoff–Chu codes. More information about spreading codes can be found in [14]. The resulting modulated RF signal has a bandwidth that is proportional to the rate at which the phases change, called the chip rate, which is the inverse of the chip duration, Tc. The receiver, as shown in Figure 3, first mixes the received signal st R() down to baseband and then filters the result to remove unwanted frequency components before converting it to digital signals. The digital signals are processed with a filter matched to
T.
By comparing the return signal to the transmitted signal, )), (2) the receiver can determine the range and velocity of objects in the environment. The digital signal processing block includes a
unwanted frequency components before converting it to digital signals. The digital signals are processed with a filter matched to the transmitted signal (as part of the digital signal process unit
s_{R}(t)
the environment. The digital signal processing block includes a t is the modulated phase waveform. The total phase matched filter, which correlates the received signal to all possi- (). One way ble delays of the transmitted spreading code. For the delay that of generating the phases is to begin with what is known as a matches the delay of the reflected signal, the correlation will spreading code. A spreading code consists of a sequence of be high, and a target at a given distance corresponding to the , which delay will be detected. The wider the bandwidth, the finer the r) into a sequence of phases ability of the receiver to resolve two objects near each other. (e.g., 0, 0, r, 0, r, … ), and the phases are used to modulate the The matched filter will provide correlations to replicas of the RF sinusoidal signal. The phase can be limited to either 0 or r transmitted spreading code of some length. The longer the radians (180°) or it can be arbitrary. In the event that the signal length of the spreading code used to correlate, the greater the phase is only either 0 or r radians, the signal is said to be ability to detect—unambiguously—targets at a long distance. a binary phase modulated signal. A binary phase modulated Although there are other types of radar signals being considsignal can also be generated by multiplying a binary (+1 and ered for automotive applications (e.g., [15]), this article’s focus is () with a carrier. In the case of binary spreading only on these two types of radars (i.e., FMCW and PMCW) and how one or more radars of one type causes interference with
48
() with a carrier. In the case of binary spreading only on these two types of radars (i.e., FMCW and PMCW) and how one or more radars of one type causes interference with another radar of either the same or a different type. t . (3) Interference in automotive radar systems () could be a periodic sequence with a One fundamental reality for automotive radar is the potential
short period or could be a pseudorandom sequence with a very for mutual interference due to multiple radars operating simullong period so that it appears to be a nearly random sequence. taneously in “close proximity” and direct line of sight [16].
long period so that it appears to be a nearly random sequence. taneously in “close proximity” and direct line of sight [16]. Codes or sequences with good autocorrelation properties are Analyses and test results involving automotive radar indicate important for use in PMCW-type radar systems. There are that mutual interference can be substantial unless suitable mitimany possible spreading codes, including Barker sequences, gation is employed. m-sequences (also known as linear feedback shift register se-Figure 4 shows two scenarios where in each, a vehicle with quences), and gold codes, all of which are binary codes. There an “interfering” radar is creating interference for a “victim” are also nonbinary codes such as the Zadoff–Chu codes. More radar. Consider the leftmost example scenario shown in Fig-
are also nonbinary codes such as the Zadoff–Chu codes. More radar. Consider the leftmost example scenario shown in Figinformation about spreading codes can be found in [14]. The ure 4, with a single interfering radar at a distance R mounted resulting modulated RF signal has a bandwidth that is propor-on a vehicle that also acts as a target for the victim radar. In tional to the rate at which the phases change, called the chip this case, the distance from the target and the distance from the interferer are identical. The signal powers received by the The receiver, as shown in Figure 3, first mixes the received victim radar for the vehicle target () Pr and the interference () () down to baseband and then filters the result to remove are given by unwanted frequency components before converting it to digital 22 22 signals. The digital signals are processed with a filter matched to PG t mv, PG t m Pr ==34PI 22, the transmitted signal (as part of the digital signal process unit () 44 r R () r R
(4)
in the direction of the target is G, and the radar cross section (RCS) of the target is v (i.e., the effective reflection area) [17]. The RCS is often stated in units of either square meters or “units” of dBsm, which refers to dB relative to 1 m2 vv in dBsm = 10 log (/m2)]. The received interference power in (4) assumes that the interfering radar has the same transmitter power and antenna gain as the victim radar and is located at the same range as the target. For example, the interferer could be colocated with the target. The signal-to-interference ratio (SIR) at the victim radar receiver is then Pr v PrG p v ==22, SIR G p =. PI 44 rRPI rR
\lambda=c/f_{c}
P_{l}
\sigma
\ 1;mathrm m{{}}^{2}\ ,[{\mathrm{i.e.}},
\mathrm{d B s m}=10\mathrm{l o g}(\sigma/m^{2})]
\frac{P_{r}}{P_{I}}!==,\frac{\sigma}{\ 4\pi R^{2}},\ ~\ \mathrm{S I R}=G_{p},\frac{P_{r}}{P_{I}}!=,\frac{G_{p}\sigma}{4\pi R^{2}}.
is the processing gain of a matched filter in the victim radar, which improves the SIR. Still, the SIR can be low enough
to inhibit target detection. For example, given a 10-dBsm RCS typically assumed for a small-to-midsize passenger car, and a processing gain of 50 dB, the SIR falls below 10 dB for a range greater than approximately 90 m. In practice, the RCS for a vehicle can vary substantially with the aspect angle. For example, the RCS can vary from 0 dBsm to as high as 30 dBsm at 77 GHz for midsize passenger cars, as shown in [18] for a Mazda 6. Even small changes in the aspect angle seen on a frame-toframe basis can lead to fluctuation in the observed RCS. Hence, statistical RCS models (e.g., the well-known Swerling models) are often used to predict automotive radar performance. In general, the parameters of the victim and interfering radar are different, and the victim radar is required to detect targets of varying range and the RCS over a defined
G_{p}
G() DK (),
is the antenna gain for the interfering (victim)
in the direction of the target is G, and the radar cross sec-For radar, a relevant performance metric is the INR after tion (RCS) of the target is v (i.e., the effective reflection area) processing in the victim radar. In general, the INR depends [17]. The RCS is often stated in units of either square meters on the parameters of the victim and interfering radars, signal
[17]. The RCS is often stated in units of either square meters on the parameters of the victim and interfering radars, signal
[i.e., modulation characteristics of the interfering radar, and demod-
)]. The received interference power ulation/downconversion processing employed by the victim
in (4) assumes that the interfering radar has the same transmit-radar. For simplicity, consider noise-like interference spread
ter power and antenna gain as the victim radar and is located at uniformly over the passband of the victim radar. This is often
the same range as the target. For example, the interferer could the case and will result in the lowest overall INR. If the interbe colocated with the target. The signal-to-interference ratio fering radar has bandwidth B, then the power spectral density
() PSD int of the interference at the victim radar is given by
PG tT mLL TX fT NG XR mLL RX fR N X
PSDint = ;2E; E() DK F ()
22. (5) 4r
44 BR () 4r
12 4 444444444312 4444 4444 3
Interfering Radar Victim Radar
\begin{aligned}{\operatorname{P S D_{i n t}}=}&{{}\underbrace{\bigg[\frac{P_{l}G_{T}\lambda L_{T}}L_{f}N_{T}} {\operatorname{l n c o c c i n g}R d r}\underbrace{\bigg[\frac{G{R}\lambda L_{R X}L_{f} {R X}}{4\pi}\bigg]}{\operatorname{V i c i i i m a d d d r}}(D_{F})(K),}\ {=}&{{}\bigg \frac{P_{l}N_{T G} T{T}L{f}}_{J}\lambda^{2}G{L_{X}} L{}{J}{N_{X}}\bigg(K),}\ \end{aligned}
P_{l}
G_{p}
N_{T X}(N_{R X})
range greater than approximately 90 m. In practice, the RCS for mitted power of the interfering radar, NN TX( RX) is the number
a vehicle can vary substantially with the aspect angle. For ex-of transmitting (receiving) antennas for the interfering (victim)
ample, the RCS can vary from 0 dBsm to as high as 30 dBsm at radar, LL TX () RX is the transmit (receive) loss for the interfer-
77 GHz for midsize passenger cars, as shown in [18] for a Maz-ing (victim) radar, and Lf is the loss due to the fascia (e.g., the
da 6. Even small changes in the aspect angle seen on a frame-to-auto’s bumper) of both radars. The duty factor parameter D
frame basis can lead to fluctuation in the observed RCS. Hence, accounts for the fraction of time the interfering radar operates
statistical RCS models (e.g., the well-known Swerling models) within the dwell time and band of the victim radar; hence, the
value of DF varies from 0 to 1.
In general, the parameters of the victim and interfer-The parameter K generally applies to the case of FMCW
ing radar are different, and the victim radar is required to modulation for both the victim and interfering radars and is given
detect targets of varying range and the RCS over a defined by the inverse ratio of the interference PSD in the victim radar
\mathrm{(P S D)_{i n t}}
L_{T X}(L_{R X})
IEEE SIGNAL PROCESSING MAGAZINE
P_{t}
D_{F}
Interfering Radar on Target Vehicle
\
- *
K=\frac{\mathrm{P S D} {I}^{\mathrm{B B}}}{\mathrm{P S D}{I}^{\mathrm{R F}}}=\frac{\Delta F_{I}^{\mathrm{R F}}}{\Delta F_{I}^{\mathrm{B B}}},
is the PSD of interference at RF prior to down-
is the PSD
of interference after downconversion in the victim radar
receiver, DFIRF is the RF sweep bandwidth of the FMCW
interfering radar, and DFIBB is the interference bandwidth
in the FMCW victim radar receiver after downconversion
to baseband. The interference bandwidth at baseband after
downconversion in the victim radar receiver depends on the
FMCW slopes of the interfering and victim radars (as well
as their time and frequency alignment) and, for similar
FMCW slopes, can be significantly less than the RF sweep
bandwidth of the FMCW interferer. In this situation,
as the FMCW interference is concentrated into a narrow
bandwidth in the FMCW victim radar receiver, thereby increasing its PSD.
For situations with, e.g., PMCW employed by either the
victim or interfering radar, K is generally equal to unity.
Interference mechanisms and characteristics for both
FMCW and PMCW modulations are discussed later in the
\mathrm{P S D}_{I}^{\mathrm{R F}}
F_{n}
\mathrm{P S D}_{I}^{\mathrm{B B}}
\mathrm{P S D} {\mathrm{n o i s e}}=k T{0}F_{n},
\Delta F_{I}^{\mathrm{R F}}
K\gg1
Without mitigation, the resulting INR can be substantial
depending on the range and RF bandwidth of the interfer-
\Delta F_{I}^{\mathrm{B B}}
Interference mechanisms and characteristics for both
FMCW and PMCW modulations are discussed later in the
Mechanisms and Characteristics of Interference section.
In the case of FMCW modulation used by both the interferer and victim, the Mechanisms and Characteristics of
Interference section includes equations and results for
parameter K for two different examples of time and fre
quency alignment.
is the noise factor of the receiver. Finally, the INR is
is the PSD of interference at RF prior to down-
PSDint
is the PSD INR =.
PSDnoise
of interference after downconversion in the victim radar
T_{0},
\mathrm{I N R}=\frac{{Smathrm D_{i\mathrm{n n t}}}}{{\mathrm{S S}}_{\mathrm{n o i s e}}}.
ing radar.
In a dynamic on-road encounter, interference seen by a victim radar will vary depending on a number of factors, including the relative position of the interfering radar (range and
ing the relative position of the interfering radar (range and
cross range) and the orientation and shape of the victim and
interfering radar antenna patterns, as illustrated in Figure 5.
K & 1 The evaluation of performance versus the geometry of victim
as the FMCW interference is concentrated into a narrow and interferer is also addressed in [17].
bandwidth in the FMCW victim radar receiver, thereby in-Figure 5 shows contours of constant INR depending on the
location of an interfering radar relative to a victim radar. The
For situations with, e.g., PMCW employed by either the simulation is based on the theoretical equation for INR, i.e.,
victim or interfering radar, K is generally equal to unity. (9). The situation depicted considers a victim radar facing an
For situations with, e.g., PMCW employed by either the simulation is based on the theoretical equation for INR, i.e.,
victim or interfering radar, K is generally equal to unity. (9). The situation depicted considers a victim radar facing an
Interference mechanisms and characteristics for both opposing interfering radar. The closer the interfering radar is
FMCW and PMCW modulations are discussed later in the to the victim radar, the larger the INR, and the worse the per-
Mechanisms and Characteristics of Interference section. formance is. Parameters typical of MIMO automotive medi-
In the case of FMCW modulation used by both the inter-um-range radar (MRR) were used. Both victim and interfering
ferer and victim, the Mechanisms and Characteristics of radars are assumed to be FMCW radars with “dissimilar fast-
Interference section includes equations and results for crossing” slopes such that K = 1. When the interfering radar
parameter K for two different examples of time and fre-is approximately 150 m downrange from the victim radar, the
interference is roughly 10 dB above the thermal noise level. As
50 IEEE SIGNAL PROCESSING MAGAZINE | September 2019 |
\
- *
- *
the interfering radar, as shown in Figure 5 for two beam positions of the MIMO victim radar, i.e., 0 and 2°, respectively. Of
course, with the potential for multiple interfering radars to be
spread across the FOV, the mitigation offered by beamforming
may be substantially diluted.
As seen in Figure 5, the INR will increase as the interferer
range decreases. Additionally, the interference level can be
substantially greater for situations with interferers of higher
substantially greater for situations with interferers of higher
radiated power density such as automotive LRR or traffic
control radar. The MOre Safety for All by Radar Interference
Mitigation (MOSARIM) project [16], [19], [20] concluded
that: “For automotive radars, without any mitigation technique
applied, the interference power can exceed the noise level by
20 to 50 dB. The achieved results show that for typical antenna
and modulation parameters, an increase of noise in the victim
receiver and thus a reduction of the usable measurement range
is very likely, while the occurrence of ghost targets seems to
be rather unlikely.” Note that, even in dense traffic with many
interferers, the occurrence of ghost targets remains unlikely
because any individual ghost target detections would, over
time, be sporadic/random in nature and thereby mitigated by
the tracking function in the victim radar. Mitigation can be further improved by the victim radar dither of relevant waveform
timing parameters that help randomize the range and Doppler
of ghost detections on a frame-to-frame basis. This technique
has been implemented by a number of radar suppliers.
Mechanisms and characteristics of interference
Here, we consider the two main modulation techniques described previously, i.e., FMCW and PMCW. Because the inter-
scribed previously, i.e., FMCW and PMCW. Because the inter-
FMCW Interferer, f2(t)
Spatial discrimination of MIMO beamforming offers miti-interference characteristics for the different types of interfergation of interference for beams not pointed in the direction of ing and victim radars.
the interfering radar, as shown in Figure 5 for two beam positions of the MIMO victim radar, i.e., 0 and 2°, respectively. Of FMCW–FMCW
course, with the potential for multiple interfering radars to be Consider a victim radar and interfering radar both using FMCW
spread across the FOV, the mitigation offered by beamforming modulation. Figure 6(a) shows the mechanism of interference
and the resulting time-domain and frequency-domain respons-
and the resulting time-domain and frequency-domain respons-
As seen in Figure 5, the INR will increase as the interferer es. For downconversion in the receiver, FMCW radar uses a reprange decreases. Additionally, the interference level can be lica (coupled version) of the transmitted FMCW signal. For the
substantially greater for situations with interferers of higher situation where the interfering FMCW signal crosses the victim
radiated power density such as automotive LRR or traffic FMCW signal, the interference appears as a linear chirp signal
control radar. The MOre Safety for All by Radar Interference after downconversion in the victim radar receiver which, assum-
Mitigation (MOSARIM) project [16], [19], [20] concluded ing “dissimilar, fast-crossing” slopes, covers a wide bandwidth
that: “For automotive radars, without any mitigation technique as it sweeps through the victim radar passband. After bandpass
applied, the interference power can exceed the noise level by filtering in the victim radar, the interference signal resembles an
20 to 50 dB. The achieved results show that for typical antenna impulse-like signal in the time domain. The resulting frequency
and modulation parameters, an increase of noise in the victim spectrum is broadband and often well above the background
receiver and thus a reduction of the usable measurement range noise floor [Figure 6(b)] when representative automotive MRR
is very likely, while the occurrence of ghost targets seems to parameters are used (i.e., 15–20 dB above noise).
be rather unlikely.” Note that, even in dense traffic with many The position and width of the impulse-like interference
interferers, the occurrence of ghost targets remains unlikely signal in the time domain following downconversion and
because any individual ghost target detections would, over bandpass filtering in the victim radar depends on the relative
time, be sporadic/random in nature and thereby mitigated by timing and slopes of the FMCW modulation of the interfering
because any individual ghost target detections would, over bandpass filtering in the victim radar depends on the relative
time, be sporadic/random in nature and thereby mitigated by timing and slopes of the FMCW modulation of the interfering
the tracking function in the victim radar. Mitigation can be fur-and victim radars. The resulting frequency spectrum characther improved by the victim radar dither of relevant waveform teristics of the interference in the victim radar, including the
timing parameters that help randomize the range and Doppler PSD, depend on the relative timing and FM slopes as well. For
of ghost detections on a frame-to-frame basis. This technique example, with slower FM rates and/or similar FM slopes for
the victim and interfering radars (i.e., “slow-crossing” slopes),
the time extent of interference in the victim radar passband
and the associated PSD can increase significantly.
Here, we consider the two main modulation techniques de-In (8), the parameter K for FMCW-to-FMCW interference
scribed previously, i.e., FMCW and PMCW. Because the inter-is fundamentally the ratio of the chirp bandwidth transmitted
Bandpass Filter
\
IEEE SIGNAL PROCESSING MAGAZINE | September 2019 |
\
- *
- *
the bandwidth after downconversion in the victim radar
depends on the difference in the FM sweep (i.e., modulation)
rate between the interfering and victim radars. Assuming that
the FM modulation rates of the interfering and victim radars
produce “broadband” interference following downconversion
in the victim radar, i.e., interference spread over no less than
the baseband bandwidth of the victim radar, K generally ranges from a minimum value of 0.5 to a maximum value equal to
the sweep bandwidth of the interfering radar divided by the
baseband bandwidth of the victim radar.
The parameter K for FMCW interference in (8) depends on
the FMCW sweep rates of the interfering and victim radars as
well as their time and frequency alignments, as demonstrated
in the following examples.
well as their time and frequency alignments, as demonstrated
in the following examples.
■■Case 1: Interfering and victim radar sweeps with the same
duration TS, start time, and start frequency
RF
DF ST S
Case 2: Interfering and victim radar sweeps with the same
T_{S}
S_{I}.
K!=!\frac{\Delta F_{I}^{\mathrm{R F}}}{\Delta F_{I}^{\mathrm{B B}}}!=!\left\ \frac{S_{I}T_{S}}{\left(S_{V}-!_{I}\right)T_{S}}\right|!=!\left|\frac{{ }S{I}}{\left(S_{V}!-!S_{I}\right)}\right|.
K=2\bigg|frac S{ I I}T{\ S{}}{{,S}_{}}\ ((S{ V}-S{ \ }),T{ S})\bigg|=2\bigg|\frac{S{ I I}}{{\ \ }(S{ V}-S{_I})}\bigg|,
K=10
Figure 7 shows parameter K and the corresponding interference in the time domain after downconversion and bandpass
ST IS SI
K = 22 =,
() SS VI-TS () SS VI-
the bandwidth after downconversion in the victim radar the same center frequency (i.e., case 2). Two examples are shown
depends on the difference in the FM sweep (i.e., modulation) in Figure 7 ; the FM sweep of the victim radar is shown in green.
rate between the interfering and victim radars. Assuming that One example corresponds to K = 1 (i.e., the FM sweep of interthe FM modulation rates of the interfering and victim radars fering radar (blue) with a sweep rate, SI, equal in magnitude
produce “broadband” interference following downconversion but opposite in sign to the sweep rate of the victim radar, S
in the victim radar, i.e., interference spread over no less than a second example corresponds to K = 10 (i.e., the FM sweep
the baseband bandwidth of the victim radar, K generally rang-of interfering radar (red) with a sweep rate similar to the sweep
es from a minimum value of 0.5 to a maximum value equal to rate of the victim radar). Compared to a “fast” (i.e., high) crossthe sweep bandwidth of the interfering radar divided by the ing rate for “dissimilar” FM sweeps (e.g., K = 1), as the FM
sweeps become more similar (e.g., K = 10), the crossing rate
The parameter K for FMCW interference in (8) depends on decreases, resulting in interference with a longer time duration
the FMCW sweep rates of the interfering and victim radars as and higher PSD after downconversion and bandpass filtering
well as their time and frequency alignments, as demonstrated in the victim radar. All else being equal, the K = 10 example
results in interference with 10 times the PSD (and, correspond-
Case 1: Interfering and victim radar sweeps with the same ingly, 10 times the INR) compared to the K = 1 example.
Early automotive FMCW radar typically used “slow-chirp”
waveforms with one or several linear FM sweeps transmitted
. (10) during a dwell or update interval and each chirp using a rela-
()
(11)
K=1
\S_{I}=\mathrm{F M}
Figure 7 shows parameter K and the corresponding interference in the time domain after downconversion and bandpass
\ _V!!=!\mathrm{F M}
as well as its duration in the time domain.
FIGURE 7. The influence of victim and interfering radar FMCW sweeps on interference in the victim radar passband.
. (10) during a dwell or update interval and each chirp using a rela-
()
tively slow sweep rate. Automotive radars using “fast-chirp”
Case 2: Interfering and victim radar sweeps with the same FMCW waveforms consisting of many identical linear FM
chirps during a dwell, where each chirp has a relatively fast
sweep rate, are becoming more prevalent. Fast-chirp FMCW
, (11) radar converts the sampled time-domain data from each chirp
to a 2D range-Doppler frequency spectrum, typically via a 2D
-fast Fourier transform (FFT) process.
The 2D frequency spectrum for a fast-chirp FMCW victim
K and the corresponding interfer-radar and slow-chirp FMCW interfering radar is shown in Figence in the time domain after downconversion and bandpass ure 8 using typical automotive MRR parameters. The vertical
ence in the time domain after downconversion and bandpass ure 8 using typical automotive MRR parameters. The vertical
Time-Domain Response of Interference fast-chirp victim radar sweeps over the frequency in a linear
fashion with respect to time. The corresponding interference
frequency spectrum for each chirp of the victim radar (ranging
frequency domain) sweeps through Doppler frequency in a linear fashion and then folds into the ambiguous Doppler interval
of the victim radar. The resulting 2D range-Doppler frequency
spectrum exhibits a “noise-like” response. For MRR, the simulation shows that the resulting INR is roughly 15–20 dB.
To help illustrate the parameters DF (and K) in (6) (i.e.,
the equation for determining the PSD of interference and
ultimately the INR), consider the case of a fast-chirp FMCW
ultimately the INR), consider the case of a fast-chirp FMCW
victim radar with dwell parameters in Figure 9 and an interfering
\
D_{F}
fast-chirp victim radar sweeps over the frequency in a linear factor of 67%. There are two fast-chirp dwell types, A and B,
fashion with respect to time. The corresponding interference sweeping 500 and 250 MHz, respectively. Each dwell type
frequency spectrum for each chirp of the victim radar (ranging has two complementary dwells, i.e., 1 and 2, with 512 and
frequency domain) sweeps through Doppler frequency in a lin-450 chirps, respectively. The four-dwell sequence therefore
ear fashion and then folds into the ambiguous Doppler interval uses four different fast-chirp sweeps.
of the victim radar. The resulting 2D range-Doppler frequency For the dwell sequence and timing shown in this section,
spectrum exhibits a “noise-like” response. For MRR, the simu-the probability of the victim radar encountering interference
with a different chirp slope (leading to crossing FMCW slopes
with a different chirp slope (leading to crossing FMCW slopes
K) in (6) (i.e., that produce a wideband interference spectrum) is effectively
the equation for determining the PSD of interference and unity with the different cases and their respective K factors.
ultimately the INR), consider the case of a fast-chirp FMCW ■■Victim radar dwell A1—interfering radar dwell B1 (K = 4)
victim radar with dwell parameters in Figure 9 and an interfering ■■Victim radar dwell A1—interfering radar dwell A2
(K = 16)
(K = 16)
■■Victim radar dwell A1—interfering radar dwell B2
■■Victim radar dwell A1—interfering radar dwell B2
(K = 35.)
■■Victim radar dwell B1—interfering radar dwell A2
■■Victim radar dwell B1—interfering radar dwell A2
(K = 4)
■■Victim radar dwell B1—Interfering Radar Dwell B2
■■Victim radar dwell B1—Interfering Radar Dwell B2
(K = 16)
■■Victim radar dwell A2—interfering radar dwell B2
■■Victim radar dwell A2—interfering radar dwell B2
(K = 4).
As previously noted, K = 1 if the slopes are the same
As previously noted, K = 1 if the slopes are the same
magnitude but opposite in sign. As the slopes become more
similar, the crossing rate decreases and K increases. The
similar, the crossing rate decreases and K increases. The
Bi factor DF equals the fractional overlap of the dwells and
varies from roughly 0.002 for the overlap of a single chirp
(1/512), to 1 for a complete overlap. For each of the afore
mentioned cases, the probability of at least one chirp over
\ \ Doppler Frequency Binlap is 33% and the probability of at least a 50% overlap
( DF $ 05.) is 16.7%. Considering all of the cases, the com-
FIGURE 8. A 2D range-Doppler spectrum for fast-chirp FMCW victim radar posite probability of at least two dissimilar slopes with at
least a 50% overlap is 83%.
B2
D_{F}
FIGURE 9. An example of a fast-chirp FMCW dwell sequence and its associated parameters.
IEEE SIGNAL PROCESSING MAGAZINE
FIGURE 9. An example of a fast-chirp FMCW dwell sequence and its associated parameters.
FIGURE 9. An example of a fast-chirp FMCW dwell sequence and its associated parameters.
\
- *
- *
The mechanism and characteristics of FMCW-to-FMCW the background noise floor, as illustrated in Figure 10(b) using interference in Figure 6(a) are shown for the single dwell/chirp representative parameters for automotive MRR (i.e., 15–20 dB of an interfering radar. In practice, the interference charac-above noise). teristics seen by an FMCW victim radar in the presence of multiple FMCW interfering radars of different types (e.g., PMCW–FMCW (or FMCW–PMCW) SRR, MRR, LRR, or multimode using fast- and/or slow-chirp Consider a victim radar with biphase PMCW modulation FMCW waveforms) can be quite complex with many impulse-and interfering radar with FMCW modulation or vice versa. like signals of different amplitudes and widths spread across Figure 11 illustrates the interference mechanism and the re- the time domain. sulting time-domain and frequency-domain responses. In both situations (i.e., PMCW victim/FMCW interferer or FMCW PMCW–PMCW victim/PMCW interferer), the interference is noise like in the Consider a victim and interfering radar, both of which uti-time and frequency domains and, all else being equal, the INR lize PMCW modulation. Figure 10(a) and (b) shows the is the same. mechanism of interference and the resulting time-domain and frequency-domain responses. PMCW interference with Comments on interference for PMCW versus random, noise-like biphase coding using chip rate DfT ic = 1/ FMCW modulation is assumed and appears as a spread-spectrum noise-like sig-Considering situations with FMCW or PMCW modulation for nal with bandwidth DfT ic = 1/ centered at carrier frequen-either the victim and/or interfering radars, the INR scales are cy f c. In the example chip rate DfT ic = 1/ with bandwidth
DfT vc = 1/, the PMCW victim radar is likewise assumed to INR? Piv,, ... PMCW victim or PMCW interferer
transmit a PMCW biphase-coded noise-like signal with the Bi Piv,
same chip rate, bandwidth, and carrier (center) frequency as INR? K, ... FMCW victim and FMCW interferer,
Bi the interfering PMCW radar but with an independent, uncor-(12) related spreading code. The victim PMCW radar downcon- verts the received signal with a constant local oscillator fre-where P iv, is the power of the interferer received at the victim quency at the common carrier frequency (shown as ff 1 = c) radar. Downconversion/demodulation and subsequent signal and demodulates the received signal with a delayed copy of processing in the victim radar generally result in spreading of the PMCW biphase code (chip rate DfT ic = 1/ and bandwidth the interference in a noise-like fashion over the passband and/ DfT vc = 1/, assumed to be the same as the corresponding or detection band. The resulting INR is then given by the PSD parameters of the PMCW interfering radar in the example of interference divided by the PSD of noise in the victim radar. shown). Following downconversion, demodulation, and band-The PSD of interference in the victim radar depends on the pass filtering in the victim radar, the interference appears as a bandwidth of the interferer, B i (i.e., the frequency spread of noise-like signal in both time and frequency domains. The re-interference), and the interference power received by the vic- sulting frequency spectrum is broadband and often well above tim radar, P iv,, which is determined by using the “one-way”
Interference at baseband is a Demodulated noise-like signal spread over ±∆f/2 PMCW Interference = f1– f2(t) (total spread = ∆f, K = 1). PMCW Interferer f2(t) Victim Radar PMCW Code (∆f v) ∆f i/2 ∆f/2 Bandpass f c PMCW Victim LO, f1 Downconversion –∆f/2 Filtering Frequency∆f i/2 and Demodulation Frequency Amplitude Time Time Bandpass Filter Time (a)
1 –25 –35 Interference Interference –45 –55 –65 –75 Noise –85 Noise –1 Relative Power (dB) –95 Normalized Amplitude01 02 46 81 01 21 41 61 82 0 Normalized Time Baseband Frequency (MHz)
(b)
FIGURE 10. (a) An PMCW-to-PMCW interference mechanism with biphase noise coding and (b) its simulated time–frequency domain characteristics.
IEEE SIGNAL PROCESSING MAGAZINE | September 2019 |
\
- *
- *
band in the victim radar, and thus its PSD at baseband, depends
on the relative FM sweep rates (FM slopes) of victim and interfering radars (reflected in the parameter K). If the FMCW
victim and interfering radar slopes are similar, the interference
power is downconverted into a narrow frequency band increasing the PSD compared to that of dissimilar slopes ( K 2 1).
Hence, all else being equal, situations with phase modulation
(PMCW) for either the victim or interfering radar generally
results in lower INR levels.
Interference mitigation
Interference analysis and interference mitigation techniques
(K>1)
in radar systems have been investigated in a number of projects and reported in a number of papers. Recently, significant
ects and reported in a number of papers. Recently, significant
research has focused on a victim radar that employs FMCW
modulation subject to interference from radars using FMCW
modulation as well (see [17] and [21] –[24]). In this section, we
focus on interference mitigation techniques.
Techniques that mitigate interference in automotive radars
include transmission techniques (e.g., frequency hopping and
timing jitter) and receiver techniques (e.g., time-domain exci-
timing jitter) and receiver techniques (e.g., time-domain excision). Generally, transmission techniques rely on ensuring that
different radars transmit in such a way that the signals are nearly orthogonal to each other in some domain (e.g., polarization,
time, and frequency). Most of the studies on interference in
automotive radar are focused on interference mitigation at the
receiver for both the interfering radar and the victim radar (e.g.,
the FMCW interferer and victim). The MOSARIM project
[19], [20] completed a comprehensive study of interference in
automotive radar systems that focused on interference mitigation. Interference mitigation techniques were grouped into six
PMCW Victim-FMCW Interferer
band in the victim radar, and thus its PSD at baseband, depends beamforming), and strategic approaches. Strategic approaches
on the relative FM sweep rates (FM slopes) of victim and in-included detecting interference and changing waveform paramterfering radars (reflected in the parameter K). If the FMCW eters and/or beamscanning in response, as well as detecting and
victim and interfering radar slopes are similar, the interference excising interference with a subsequent repair of the received
power is downconverted into a narrow frequency band increas-signal in either the time, frequency, or joint-time-frequency
ing the PSD compared to that of dissimilar slopes ( K 2 1). domains. Another strategic technique considered was the gen-
Hence, all else being equal, situations with phase modulation eral concept of intervehicle communication that negotiates
(PMCW) for either the victim or interfering radar generally noninterfering radar parameters (e.g., time or frequency slots).
With the exception of the polarization domain, many of the
techniques described in the MOSARIM project involved substantial signal processing for processing complex waveforms,
stantial signal processing for processing complex waveforms,
Interference analysis and interference mitigation techniques adaptively nulling interference, and/or detecting and excising
in radar systems have been investigated in a number of proj-interference. Techniques with the highest level of signal proects and reported in a number of papers. Recently, significant cessing complexity include digital beamforming with adaptive
research has focused on a victim radar that employs FMCW nulling, time-frequency transforming with detection and excimodulation subject to interference from radars using FMCW sion of interference, and space-time adaptive processing.
modulation as well (see [17] and [21] –[24]). In this section, we The MOSARIM project performed modeling, simulations,
and tests of interference to automotive radar of the aforemen-
Techniques that mitigate interference in automotive radars tioned mitigation techniques and concluded that [16]: “ To
Techniques that mitigate interference in automotive radars tioned mitigation techniques and concluded that [16]: “ To
include transmission techniques (e.g., frequency hopping and assure an I/N level of 0 or −10 dB, reliable mitigation techtiming jitter) and receiver techniques (e.g., time-domain exci-niques in the order of (a minimum of) 50 dB mitigation marsion). Generally, transmission techniques rely on ensuring that gin are needed.” In assessing the capability of interference
different radars transmit in such a way that the signals are near-mitigation, MOSARIM concluded that individual mitigation
ly orthogonal to each other in some domain (e.g., polarization, techniques are not adequate, multiple techniques will need to
time, and frequency). Most of the studies on interference in be applied and, as automotive radar volumes increase, it may
automotive radar are focused on interference mitigation at the be beneficial to include, via regulatory means, the assignment
receiver for both the interfering radar and the victim radar (e.g., of polarization and frequency bands depending on the radar
the FMCW interferer and victim). The MOSARIM project application/type (e.g., SRR, MRR, or LRR) and on-vehicle
[19], [20] completed a comprehensive study of interference in mounting location. An example of this is using different subautomotive radar systems that focused on interference mitiga-bands in various directions (front, back, and side) and using
tion. Interference mitigation techniques were grouped into six different polarizations.
PMCW Victim-FMCW Interferer
IEEE SIGNAL PROCESSING MAGAZINE
\
- *
- *
“Mechanisms and Characteristics of Interference” section, an
interfering radar with the same structure as the victim radar
can create a “ghost target” if that signal, when received at the
victim radar, begins a sweep within a small window of time
proportional to the bandwidth of the filter. However, it is much
more likely that the interfering radar creates a noise-like signal. In [17], a single FMCW interferer with a victim FMCW
radar was considered. The SIR was derived as a function of the
distance between the interferer and the victim radar and the
distance between the target and the victim radar. Using these
parameters, the region where the SIR was above some threshold (e.g., 10 dB) for a given target size was calculated. The
effect of the FMCW chirp slopes on the SIR was determined.
In [17], the conclusion noted a that interference can cause a
victim radar to “lose a target.” As an example, an FMCW radar
with a processing gain (time-bandwidth product) of 50 dB and
an interferer that is 10 m away will cause the SIR to drop below
10 dB when the target is 30 m away. Because FMCW radars
have been the dominant type of radar used in automotive applications, there are quite a few papers that analyze the performance of an FMCW radar interferer on a victim FMCW radar.
1
Target + Noise
“Mechanisms and Characteristics of Interference” section, an [23], [25]–[27]. The results of a basic simulation with timeinterfering radar with the same structure as the victim radar domain excision for a slow-chirp FMCW interferer and a fastcan create a “ghost target” if that signal, when received at the chirp FMCW victim radar are shown in Figures 12 and 13. As
victim radar, begins a sweep within a small window of time previously discussed, with an interferer FMCW sweep crossing
proportional to the bandwidth of the filter. However, it is much a victim radar FMCW sweep, the interference appears as a linear
more likely that the interfering radar creates a noise-like sig-chirp signal after downconversion in the victim radar receiver.
nal. In [17], a single FMCW interferer with a victim FMCW The linear chirp interference signal sweeps through the victim
radar was considered. The SIR was derived as a function of the radar passband and, assuming “fast-crossing” sweeps, produces
distance between the interferer and the victim radar and the an “impulse-like” signal in the time domain after bandpass fildistance between the target and the victim radar. Using these tering in the victim radar receiver. Prior to 2D range-Doppler
parameters, the region where the SIR was above some thresh-FFT-matched filtering in a fast-chirp victim radar, the target
old (e.g., 10 dB) for a given target size was calculated. The signal is typically well below the noise level while the impulseeffect of the FMCW chirp slopes on the SIR was determined. like interference signal is well above the noise level. Matched
In [17], the conclusion noted a that interference can cause a filtering provides substantial integration gain for a target-like
victim radar to “lose a target.” As an example, an FMCW radar constant frequency signal, while the impulse-like interference
with a processing gain (time-bandwidth product) of 50 dB and signal spreads in a noise-like fashion over the range-Doppler
an interferer that is 10 m away will cause the SIR to drop below frequency spectrum.
10 dB when the target is 30 m away. Because FMCW radars Basic time-domain excision uses a threshold above the
have been the dominant type of radar used in automotive appli-background noise level to remove interference. In other words,
cations, there are quite a few papers that analyze the perfor-time-domain samples above the threshold are set to zero. Sim-
cations, there are quite a few papers that analyze the perfor-time-domain samples above the threshold are set to zero. Simmance of an FMCW radar interferer on a victim FMCW radar. ulated results before and after time-domain excision are shown
in Figure 12 for one chirp of the fast-chirp victim radar. Note
that the simulated example includes target, interference, and
noise. Time-domain excision is repeated for each chirp.
Simulated results for the range-Doppler frequency response
of the fast-chirp victim radar are shown in Figure 13. Results
correspond to the cases of target + noise (no interference) and
correspond to the cases of target + noise (no interference) and
for target + interference + noise, first without time-domain
excision and then with time-domain excision. As shown, with-
Excision Threshold
out any mitigation, the interference substantially raises the
“noise” floor and masks the target. Time-domain excision
is able to remove the FMCW interference while preserving
1
the target signal; however, as expected, some signal loss and
potential for artifacts occur depending on the amount and pattern of excision required.
FIGURE 12. The simulated time-domain response before and after time-
The simulated example illustrates time-domain excision for
the case of a single interferer. Overall effectiveness degrades
Target + Interference + Noise tial. Although mitigation of a single FMCW interferer on an
FMCW victim is fairly well understood, there is much ongoing
research into other scenarios.
Ongoing and future research
Although simple interference mitigation techniques are gen-
erally well understood—with some having been implemented
in automotive radars—interference mitigation is an important
in automotive radars—interference mitigation is an important
and ongoing area of R&D. Typically, a radar system will have
a receiver structure, as shown in Figure 14. The RF front end
includes analog components such as a low-noise amplifier,
mixers, and filters. The output of the RF front end is converted
to digital by an analog-to-digital converter (ADC). The first
process step is range processing, which is an FFT in an FMCW
radar system and a matched filter in a PMCW radar system.
Range processing is followed by Doppler processing, which
is followed by beamforming, object detection, and tracking.
Interference mitigation may be added to the block diagram,
potentially at different points in the processing steps. One technique used to mitigate the effect of an interferer (either FMCW
or PMCW) is isolating the part of the signal that does not have
interference; this may be implemented using a time-domain
notch filter in an FMCW system or a “zero-forcing” type of detector in a PMCW system (as discussed further in this section).
However, these techniques also remove a portion of the desired
signal and can become problematic as the number of interferers grow. Time excision would typically happen between the
output of the ADC and the input of the range-matched filter.
In [23], the interference cancellation of a single FMCW interferer on an FMCW victim by identifying the interference location using techniques from image processing and then “zeroing
ferer on an FMCW victim by identifying the interference location using techniques from image processing and then “zeroing
out” those time locations with additional smoothing to avoid
ringing effects, shows that without interference cancellation, the
victim radar would not be able to detect a certain target due to
the increase in noise level, but with interference cancellation,
the radar is able to detect the target. In [25], the characteristics
of an FMCW-interfering signal, which are much stronger than
the desired signal reflected from the target on an FMCW victim
radar, are estimated and then subtracted from the overall signal,
thereby improving the receiver sensitivity. This technique would
also work for multiple radars as long as the interference from
one radar did not overlap in time with the interference from
another radar. This interference mitigation would also occur
before the range-FFT processing. In [7], interference mitigation not requiring a threshold (compared to normal time-domain
excision) was considered and shown to be effective, even for
tial. Although mitigation of a single FMCW interferer on an ferer. Simulations with a CW signal as the interference source
FMCW victim is fairly well understood, there is much ongoing and an 800-MHz bandwidth FMCW signal for the victim radar
show that the interference reduced by 40 dB. Here, four receiving antennas were used with optimal weighting to remove the
interference. Measurements corresponding to the simulation
Although simple interference mitigation techniques are gen-show that the resulting INR was under 2 dB [24, Table 1] for a
erally well understood—with some having been implemented single interferer. This interference mitigation technique would be
in automotive radars—interference mitigation is an important applied in the beamforming processing unit. In [27], an FMCW
and ongoing area of R&D. Typically, a radar system will have interferer and victim were considered, where again, the interfera receiver structure, as shown in Figure 14. The RF front end ence is detected and zeroed out. Although this does reduce the
includes analog components such as a low-noise amplifier, noise level because of the interferer, it also removes the desired
mixers, and filters. The output of the RF front end is converted signal over a certain time period when the frequency of the interto digital by an analog-to-digital converter (ADC). The first ferer falls within a certain frequency range of the victim radar. To
process step is range processing, which is an FFT in an FMCW mitigate this effect, [27] considered a technique that regenerates
radar system and a matched filter in a PMCW radar system. the desired signal during the time when the received signal was
Range processing is followed by Doppler processing, which zeroed out. An iterative algorithm was used for that purpose and
is followed by beamforming, object detection, and tracking. allowed for smaller targets to be detected than would have oth-
Interference mitigation may be added to the block diagram, erwise been detected with just the zeroing-out approach (with or
potentially at different points in the processing steps. One tech-without additional smoothing).
nique used to mitigate the effect of an interferer (either FMCW Interference in cellular communication systems has been
or PMCW) is isolating the part of the signal that does not have the subject of considerable investigation. Code-division multiinterference; this may be implemented using a time-domain ple access (CDMA)—the communication version of a PMCW
FIGURE 14. A generic receiver structure for radar.
interference; this may be implemented using a time-domain ple access (CDMA)—the communication version of a PMCW
notch filter in an FMCW system or a “zero-forcing” type of de-radar—has been widely deployed in 2G and 3G cellular systector in a PMCW system (as discussed further in this section). tems. The processing gain associated with PMCW signals,
However, these techniques also remove a portion of the desired similar to the CDMA signals used for communications, allows
signal and can become problematic as the number of interfer-for multiple radars to be used simultaneously.
ers grow. Time excision would typically happen between the In a PMCW–PMCW scenario, the large number of spreadoutput of the ADC and the input of the range-matched filter. ing codes generally ensures the interference will be a wide-
In [23], the interference cancellation of a single FMCW inter-band, noise-like signal because each radar can use a different
In [23], the interference cancellation of a single FMCW inter-band, noise-like signal because each radar can use a different
ferer on an FMCW victim by identifying the interference loca-spreading code. There are a number of techniques that can be
tion using techniques from image processing and then “zeroing used in PMCW–PMCW situations to improve the interferout” those time locations with additional smoothing to avoid ence mitigation capability. Although some of these techniques
ringing effects, shows that without interference cancellation, the require knowing the spreading codes of other radars, there are
victim radar would not be able to detect a certain target due to also “blind” techniques that work without that knowledge [28].
the increase in noise level, but with interference cancellation, These are the same techniques that are useful in a communicathe radar is able to detect the target. In [25], the characteristics tions context (e.g., CDMA). These techniques do not completeof an FMCW-interfering signal, which are much stronger than ly eliminate interference but may drastically reduce its effect,
the desired signal reflected from the target on an FMCW victim especially in a near-far scenario similar to multiple interferers
radar, are estimated and then subtracted from the overall signal, versus a victim receiver, and work best when the interference is
thereby improving the receiver sensitivity. This technique would periodic in nature, i.e., the spreading codes repeat after a certain
also work for multiple radars as long as the interference from number of chips (in much the same way an FMCW type of radar
one radar did not overlap in time with the interference from would have a repetitive signal). The interference mitigation in
another radar. This interference mitigation would also occur CDMA systems (i.e., PMCW) is based on the cyclostationary
before the range-FFT processing. In [7], interference mitiga-structure of the interfering signal. These interference mitigation
tion not requiring a threshold (compared to normal time-domain techniques are based on estimating the correlation matrix of the
excision) was considered and shown to be effective, even for received signal, then employing an “orthogonalizing matched problem, there is an additional unknown (i.e., the data).
An FMCW interferer on a victim PMCW radar is very similar to a jammer in a spread-spectrum system. This type of interference, as well as effective mitigation techniques, has been
ference, as well as effective mitigation techniques, has been
well studied. An FMCW interferer signal to a PMCW victim is
the same as that of a “swept-tone jammer” in spread-spectrum
communication systems discussed in [30] and [31]. The performance measure in a spread-spectrum communication system
is typically bit error rate, rather than the typical performance
measures used in a radar system. Nevertheless, the mitigation
techniques would be similar.
A PMCW interferer on a victim FMCW radar system
can appear as just additional noise that might seem difficult
to mitigate in the time domain. However, the PMCW signal’s
to mitigate in the time domain. However, the PMCW signal’s
spectral characteristics can be estimated and used to improve
the filtering that may reject wideband color noise [32]. Certain
short-term time-frequency processing techniques may be able
to mitigate this interference [33]. As with other classes of interferer and victim radars, transmission techniques such as polarization or frequency separation can be applied here.
Because both FMCW and PMCW types of radar are essentially spread-spectrum types of systems, interference mitigation techniques applicable to spread-spectrum communication
systems may potentially be of use in radar systems. One tech-
tion techniques applicable to spread-spectrum communication
systems may potentially be of use in radar systems. One technique for mitigating strong interference in the presence of a
weak signal is based on locally optimum Bayesian detection.
For example, [34] considers a spread-spectrum signal in the
presence of different types of interference and noise. Although
the focus of these techniques is on communication systems,
they have potential for application in radar systems as well.
Future research that addresses interference to automotive
radar sensors includes joint radar/communication systems [35],
decentralized multiple-access protocols, and alternative modu-
of inherent resistance to interference by virtue of a large time-
58 IEEE SIGNAL PROCESSING MAGAZINE
decentralized multiple-access protocols, and alternative modulation techniques (and the corresponding matched-filter signal
processing) that limit the potential for, and subsequent level of,
interference. Further development of joint, multiple-domainadaptive signal processing algorithms that excise/null interference within the polarization-spatial-temporal-frequency
domains must be explored as well.
One aspect of interference mitigation to consider is multiple-access techniques at the transmitter. In other words, by
coordinating transmission (e.g., in the time domain, frequency
domain, and polarization domain), interference can be prevent-
coordinating transmission (e.g., in the time domain, frequency
domain, and polarization domain), interference can be prevented from occurring. We note that this type of problem has been
studied extensively in the context of communication systems,
where information-theoretic formulations can be used to determine the possible rates of data transmission for different users.
Of course, the radar problem is different in that data are not
transmitted and targets are detected. Nevertheless, approaches
to multiple access must be investigated.
Conclusions
In general, automotive radar systems include a substantial level
of inherent resistance to interference by virtue of a large time-
is related to its ability to reject interference). Regardless, with
An FMCW interferer on a victim PMCW radar is very simi-an increasing number of radars deployed per vehicle and an
lar to a jammer in a spread-spectrum system. This type of inter-increase in the number of vehicles having radars, interference
ference, as well as effective mitigation techniques, has been levels, especially in certain situations such as rush-hour traffic,
well studied. An FMCW interferer signal to a PMCW victim is will likely be quite severe. Automotive radar manufacturers
the same as that of a “swept-tone jammer” in spread-spectrum have been active in developing and implementing many of the
communication systems discussed in [30] and [31]. The perfor-mitigation techniques described in this article that reduce the
mance measure in a spread-spectrum communication system impact of mutual interference.
is typically bit error rate, rather than the typical performance Because radars are becoming pervasive and ubiquitous on
measures used in a radar system. Nevertheless, the mitigation automobiles and perform safety-critical functions, there is a
need to optimize interference mitigation both at the trans-
need to optimize interference mitigation both at the trans-
A PMCW interferer on a victim FMCW radar system mitter and receiver by limiting the amount of interference so
can appear as just additional noise that might seem difficult that victim radar performance can be affected only up to a
to mitigate in the time domain. However, the PMCW signal’s prescribed amount. To this end, developing standards will
spectral characteristics can be estimated and used to improve make the engineering of interference mitigation easier and
the filtering that may reject wideband color noise [32]. Certain more effective.
short-term time-frequency processing techniques may be able
to mitigate this interference [33]. As with other classes of inter-Authors
ferer and victim radars, transmission techniques such as polar-Stephen Alland ( swalland@uhnder.com) received his B.S.
Because both FMCW and PMCW types of radar are essen-Polytechnic Institute, Troy, New York, in 1976 and 1977,
tially spread-spectrum types of systems, interference mitiga-respectively. He is a radar consultant with more than 40 years
tion techniques applicable to spread-spectrum communication of experience in radar systems, including 22 years in automosystems may potentially be of use in radar systems. One tech-tive radar systems. Previously, he was a technical fellow and
nique for mitigating strong interference in the presence of a manager for advanced radar development with Delphi
weak signal is based on locally optimum Bayesian detection. Electronics and Safety in Malibu, California. From 1995 to
For example, [34] considers a spread-spectrum signal in the 2014, he worked for Delphi to develop automotive radar syspresence of different types of interference and noise. Although tems. Prior to Delphi, he worked for Hughes to develop air
the focus of these techniques is on communication systems, defense radar systems, beginning in 1977. He holds 21 U.S.
they have potential for application in radar systems as well. patents related to radar systems. His technical areas of exper-
Future research that addresses interference to automotive tise include radar system design and simulation, radar waveradar sensors includes joint radar/communication systems [35], form design, radar signal processing, and target tracking.
decentralized multiple-access protocols, and alternative modu-Wayne Stark ( stark@eecs.umich.edu) received his B.S.,
lation techniques (and the corresponding matched-filter signal M.S., and Ph.D. degrees in electrical engineering from the
processing) that limit the potential for, and subsequent level of, University of Illinois, Urbana, in 1978, 1979, and 1982,
and M.S. degrees in electrical engineering from Rensselaer
Because both FMCW and PMCW types of radar are essen-Polytechnic Institute, Troy, New York, in 1976 and 1977,
of inherent resistance to interference by virtue of a large time-of Engineering and Technology, Dhaka, in 1989, and his
IEEE SIGNAL PROCESSING MAGAZINE | September 2019 |
processing) that limit the potential for, and subsequent level of, University of Illinois, Urbana, in 1978, 1979, and 1982,
interference. Further development of joint, multiple-domain-respectively. Since 1982, he has been a faculty member in the
adaptive signal processing algorithms that excise/null inter-Department of Electrical Engineering and Computer Science
ference within the polarization-spatial-temporal-frequency at the University of Michigan, Ann Arbor. He was selected by
the National Science Foundation as a 1985 Presidential
One aspect of interference mitigation to consider is mul-Young Investigator. He received the IEEE Military
tiple-access techniques at the transmitter. In other words, by Communications Conference (MILCOM) Board 2002
coordinating transmission (e.g., in the time domain, frequency Technical Achievement Award for sustained contributions to
domain, and polarization domain), interference can be prevent-military communications. In 2009, he was corecipient of the
ed from occurring. We note that this type of problem has been IEEE MILCOM Ellersick Prize for best paper in the
studied extensively in the context of communication systems, Unclassified Technical Program. In 2010, he received the
where information-theoretic formulations can be used to deter-Journal of Communications and Networks Best Paper Award.
mine the possible rates of data transmission for different users. He is on the editorial board of IEEE Journal on Selected
Of course, the radar problem is different in that data are not Areas of Communications. His research interests include the
transmitted and targets are detected. Nevertheless, approaches areas of coding and communication theory, spread-spectrum
systems, wireless communication networks, and radar systems. He is a Fellow of the IEEE.
Murtaza Ali ( murtaza@uhnder.com) received his B.S.
In general, automotive radar systems include a substantial level degree in electrical engineering from Bangladesh University
of inherent resistance to interference by virtue of a large time-of Engineering and Technology, Dhaka, in 1989, and his of systems engineering at Uhnder, Inc. Prior to joining
Uhnder, he was a Distinguished Member of Technical Staff,
and manager of the Perception and Analytics Lab at Texas
Instruments (TI). At TI, he led R&D teams for millimeterwave radar, mobile WiMAX, and asymmetric digital subscriber line systems. He also represented TI in various
standards organizations including Telecommunications
Industry Association, the International Telecommunication
Union, HomePlug, Home Phone line Networking As
sociation, and the IEEE. His research interests include development of novel communications and signal processing
systems. He holds 32 U.S. patents and has published more
than 40 papers in refereed and invited forums. He is a Senior
Member of the IEEE.
Manju Hegde ( manju@uhnder.com) received his B.S.
degree in electrical engineering from the Indian Institute of
Technology, Bombay, and his Ph.D. degree in computer infor-
Technology, Bombay, and his Ph.D. degree in computer information and control engineering from the University of
Michigan, Ann Arbor, in 1987 and 1979, respectively. He is the
chief executive officer (CEO) of Uhnder Inc. Previously, he
was the corporate vice president at Advanced Micro Devices
(AMD), where he was responsible for driving and executing
AMD’s strategy across technology and marketing functions
into the client and server computing ecosystems. He also
cofounded and led AMD Ventures before joining AMD. He
was vice president of Compute Unified Device Architecture
technical marketing at Nvidia, where he focused on training
and enabling researchers and developers to leverage the parallel architecture and performance of global processing units for
general purpose applications. He was cofounder and CEO of
AGEIA Technologies from its inception in 2002.
References
[1] B. Schoettle, “Sensor fusion: A comparison of sensing capabilities of human
drivers and highly automated vehicles,” Univ. of Michigan Transportation Res.
Inst., Ann Arbor, Tech. Rep. SWT-2017-12, 2017.
Inst., Ann Arbor, Tech. Rep. SWT-2017-12, 2017.
[2] G. Rudolph and U. Voelzke, “Three sensor types drive autonomous vehicles,”
2017.
[6] F. Engels, M. Wintermantel, and P. Heidenreich, “Automotive MIMO radar
angle estimation in the presence of multipath,” in Proc. IEEE Radar Conf.
(EURAD), 2017, pp. 82–85.
[7] M. Wagner, F. Sulejmani, A. Melzer, P. Meissner, and M. Huemer, “Threshold-free
[2] G. Rudolph and U. Voelzke, “Three sensor types drive autonomous vehicles,”
Sensors Mag., Nov. 2017. [Online]. Available: https://www.sensorsmag.com/
components/three-sensor-types-drive-autonomous-vehicles
[3] I. Bilik, O. Bialer, S. Villeval, H. Sharifi, K. Kona, M. Pan, D. Persechini, M.
[3] I. Bilik, O. Bialer, S. Villeval, H. Sharifi, K. Kona, M. Pan, D. Persechini, M.
Musni et al., “Automotive MIMO radar for urban environments,” in Proc. IEEE
Radar Conf., 2016. pp. 1–6. doi: 10.1109/RADAR.2016.7485215.
[4] I. Bilik, S. Villeval, D. Brodeski, H. Ringel, O. Longman, P. Goswami, C. Y.
[7] M. Wagner, F. Sulejmani, A. Melzer, P. Meissner, and M. Huemer, “Threshold-free
interference cancellation method for automotive FMCW radar systems,” in Proc. IEEE
Int. Symp. Circuits and Systems, 2018, pp. 1–4.
[8] H. H. Meinel and J. Dickman, “Automotive radar: From its origin to future
of systems engineering at Uhnder, Inc. Prior to joining [12] M. Schneider, “Automotive radar—status and trends,” in Proc. German
Uhnder, he was a Distinguished Member of Technical Staff, Microwave Conf., 2005, pp. 144 –147.
and manager of the Perception and Analytics Lab at Texas [13] E. G. Hoare and R. Hill, “System requirements for automotive radar antennas,”
in Proc. IEEE Colloq. Antennas for Automotives, Mar. 2000, pp. 1–11.
Instruments (TI). At TI, he led R&D teams for millimeter-
[14] E. Garcia, J. A. Paredes, F. J. Álvarez, M. C. Pérez, and J. J. García,
wave radar, mobile WiMAX, and asymmetric digital sub-“Spreading sequences in active sensing: A review,” Signal Process., vol. 106, pp.
88–105, Jan. 2015.
scriber line systems. He also represented TI in various
[15] N. Levanon, “Multifrequency complementary phase-coded radar signal
standards organizations including Telecommunications
Proc.—Radar, Sonar Navigation, vol. 147, no. 6, pp. 276, 2000.
Industry Association, the International Telecommunication [16] M. Kunert, “Final report,” European Commission: MOre safety for all by radar
Union, HomePlug, Home Phone line Networking Asinterference mitigation (MOSARIM), Luxembourg, Tech. Rep. 248231, 2010.
[Online]. Available: https://cordis.europa.eu/project/rcn/94234/reporting/en
sociation, and the IEEE. His research interests include devel-
[17] T. Schipper, M. Harter, T. Mahler, O. Kern, and T. Zwick, “Discussion of the
opment of novel communications and signal processing operating range of frequency modulated radars in the presence of interference,” Int.
systems. He holds 32 U.S. patents and has published more J. Microw. Wireless Technol., vol. 6, no. 3–4, pp. 371–378, 2014.
[18] J. Hasch, E. Topa k, R. Schnabel, T. Zwick, R. Weigel, and C. Waldschmidt-,
than 40 papers in refereed and invited forums. He is a Senior
“Millimeter-wave technology for automotive radar sensors in the 77 GHz frequen
cy band,” IEEE Trans. Microw. Theory Techn., vol. 60, no. 3, pp. 845–860, 2012.
Manju Hegde ( manju@uhnder.com) received his B.S. [19] M. Ahrholdt, F. Bodereau, C. Fischer, M. Goppelt, R. Pietsch, A. John, A.
Ossowska, and M. Kunert, “D12.1-Study report on relevant scenarios and applicadegree in electrical engineering from the Indian Institute of
tions and requirements specification,” European Commission: MOre Safety for All
Technology, Bombay, and his Ph.D. degree in computer infor-by Radar Interference Mitigation (MOSARIM), 2010. [Online]. Available: https://
cordis.europa.eu/project/rcn/94234/reporting/en
mation and control engineering from the University of
[20] M. Kunert, F. Bodereau, M. Goppelt, C. Fischer, A. John, T. Wixforth, A.
Michigan, Ann Arbor, in 1987 and 1979, respectively. He is the Ossowska, T. Schipper et al., “D1.5-Study on the state-of-the-art interference mitichief executive officer (CEO) of Uhnder Inc. Previously, he gation techniques,” Technical report, European Commission: MOre Safety for All
by Radar Interference Mitigation (MOSARIM), 2010. [Online]. Available: https://
was the corporate vice president at Advanced Micro Devices cordis.europa.eu/project/rcn/94234/reporting/en
(AMD), where he was responsible for driving and executing [21] M. Umehira, T. Nozawa, Y. Makino, X. Wang, S. Takeda, and H. Kuroda, “A
AMD’s strategy across technology and marketing functions novel iterative inter-radar interference reduction scheme for densely deployed automotive FMCW radars,” in Proc. 19th Int. Radar Symp. (IRS), 2018, pp. 1–10. doi:
into the client and server computing ecosystems. He also 10.23919/IRS.2018.8448223.
cofounded and led AMD Ventures before joining AMD. He [22] M. Goppelt, H.-L. Blöcher, and W. Menzel, “Analytical investigation of mutual
interference between automotive FMCW radar sensors,” in Proc. 6th German
was vice president of Compute Unified Device Architecture
Microwave Conf., 2011, pp. 1–4.
technical marketing at Nvidia, where he focused on training
[23] M. Barjenbruch, D. Kellner, K. Dietmayer, J. Klappstein, and J. Dickmann,
and enabling researchers and developers to leverage the paral-“A method for interference cancellation in automotive radar,” in Proc. IEEE MTT-S
Int. Conf. Microwaves for Intelligent Mobility (ICMIM), 2015, pp. 1–4.
lel architecture and performance of global processing units for
[24] J. Bechter, K. Eid, F. Roos, and C. Waldschmidt, “Digital beamforming to mitgeneral purpose applications. He was cofounder and CEO of igate automotive radar interference,” in Proc. IEEE MTT-S Int. Conf. Microwaves
for Intelligent Mobility (ICMIM), 2016, pp. 1–4.
[25] J. Bechter and C. Waldschmidt, “Automotive radar interference mitigation by
reconstruction and cancellation of interference,” in Proc. IEEE MTT-S Int. Conf.
Microwaves for Intelligent Mobility (ICMIM), 2015, pp. 1–4.
[1] B. Schoettle, “Sensor fusion: A comparison of sensing capabilities of human
[26] Z. Xu and Q. Shi, “Interference mitigation for automotive radar using orthogonal
drivers and highly automated vehicles,” Univ. of Michigan Transportation Res.
noise waveforms,” IEEE Geosci. Remote Sens. Lett., vol. 15, no. 1, pp. 137–141, 2018.
[27] J. Bechter, F. Roos, M. Rahman, and C. Waldschmidt, “Automotive radar
[2] G. Rudolph and U. Voelzke, “Three sensor types drive autonomous vehicles,”
interference mitigation using a sparse sampling approach,” in Proc. European
Sensors Mag., Nov. 2017. [Online]. Available: https://www.sensorsmag.com/
Radar Conf., 2017, pp. 90–93.
[28] U. ,” Madhow, “Blind adaptive interference suppression for direct-sequence
[3] I. Bilik, O. Bialer, S. Villeval, H. Sharifi, K. Kona, M. Pan, D. Persechini, M.
CDMA Proc. IEEE, vol. 86, no. 10, pp. 2049–2069, 1998.
Musni et al., “Automotive MIMO radar for urban environments,” in Proc. IEEE
[29] K. Fukawa and H. Suzuki, “Orthogonalizing matched filter (OMF) detection
for DS-CDMA mobile radio systems Proc. Global Telecommunications Conf.,
[4] I. Bilik, S. Villeval, D. Brodeski, H. Ringel, O. Longman, P. Goswami, C. Y.
Communications: The Global Bridge,” in ., 1994, pp. 385–389.
Kumar, S. Rao et al., “Automotive multi-mode cascaded radar data processing
[30] G. Saulnier, P. Das, and L. Milstein, “An adaptive digital suppression filter for
direct-sequence spread-spectrum communications,” IEEE J. Sel. Areas Commun.,
[5] F. Engels, P. Heidenreich, A. M. Zoubir, F. K. Jondral, and M. Wintermantel,
vol. 3, no. 5, pp. 676 –686, 1985.
“Advances in automotive radar: A framework on computationally efficient high-resolution frequency estimation,” IEEE Signal Process. Mag., vol. 34, no. 2, pp. 36–46, [31] L. B. Milstein, “Interference rejection techniques in spread spectrum communications,” Proc. IEEE, vol. 76, no. 6, pp. 657–671, 1988.
[6] F. Engels, M. Wintermantel, and P. Heidenreich, “Automotive MIMO radar [32] L. Milstein and P. Das, “An analysis of a real-time transform domain filtering
angle estimation in the presence of multipath,” in Proc. IEEE Radar Conf. digital communication system—part II: Wide-band interference rejection,” IEEE
Trans. Commun., vol. 31, no. 1, pp. 21–27, 1983.
[7] M. Wagner, F. Sulejmani, A. Melzer, P. Meissner, and M. Huemer, “Threshold-free [33] V. C. Chen and H. Ling, Time-Frequency Transforms for Radar Imaging and
interference cancellation method for automotive FMCW radar systems,” in Proc. IEEE Signal Analysis, Norwood, MA: Artech, 2002.
[34] S. N. Batalama, M. J. Medley, and I. N. Psaromiligkos, “Adaptive robust
[8] H. H. Meinel and J. Dickman, “Automotive radar: From its origin to future spread-spectrum receivers,” IEEE Trans. Commun., vol. 47, no. 6, pp. 905–917,
1999.
[9] H. H. Meinel, “Evolving automotive radar: From the very beginnings into the [35] A. R. Chiriyath, B. Paul, G. M. Jacyna, and D. W. Bliss, “Inner bounds on perfuture,” in Proc. European Conf. Antennas and Propagation (EuCAP), 2014, pp. formance of radar and communications co-existence,” IEEE Trans. Signal Process.,
vol. 64, no. 2, pp. 464–474, 2016.
IEEE SIGNAL PROCESSING MAGAZINE | September 2019 |
IEEE SIGNAL PROCESSING MAGAZINE
[10] M. I. Skolnik, Introduction to Radar. New York: McGraw-Hill, 1962.
IEEE SIGNAL PROCESSING MAGAZINE
- *