IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL.

17, 2024 3409

A Velocity Ambiguity Resolution Algorithm Based

on Improved Hypothetical Phase Compensation for
TDM-MIMO Radar Traffic Target Imaging
Bo Yang , Student Member, IEEE, Siqi Liu , Student Member, IEEE, Hua Zhang , and Yongjun Zhou

Abstract—In principle, the imaging millimeter-wave radar millimeter-wave radar achieves virtual aperture expansion to
based on time-division multiplexing multiple-input multiple- obtain higher angular resolution by modulating the emission
output (MIMO) technologies can provide richer target information method (e.g., time-division multiplexing (TDM) [4], frequency-
for intelligent transportation systems due to the high-density target
point cloud output. However, the chirp repetition interval extension division multiplexing [5], and code-division multiplexing [6])
reduces radar’s inherent maximum detectable velocity, leading to without increasing the number of antennas, which gives MIMO
the unavoidable problem of estimating the target velocity with a radar the capability of target point cloud imaging and environ-
large ambiguity period in imaging radar moving target surveillance mental awareness. For example, the MIMO radar point cloud
applications. To alleviate these problems, we propose an improved
image is utilized to implement sensing of the surrounding envi-
hypothetical phase compensation algorithm. Unlike the original
method of determining the Doppler ambiguity period by compar- ronment of vehicle bodies [7], [8], road curvature estimation [9],
ing the peak amplitude of the angular power spectrum in each and road surface classification [10].
hypothetical case, the proposed algorithm selects the peak of the an- Theoretically, road monitoring using MIMO radar can pro-
gular power signal-to-noise ratio (SNR) spectrum as the processing vide richer and more realistic traffic target information for in-
object and jointly decides the target speed by the average of the two
telligent transportation systems, such as additional information
highest wave peak intervals in the SNR variation curve. Simulations
and practical experiments show that the improved algorithm has on vehicle height, profile, and type [7], [11], [12]. However,
higher anti-interference performance. In particular, the proposed the extended chirp repetition interval makes the maximum un-
algorithm can remain continuously effective when multiple targets ambiguous speed of the MIMO radar just 1/M times that of the
or angle information exist in the same distance Doppler cell, making phased array radar, where M is the actual number of transmitting
it more suitable for MIMO imaging applications.
antennas (TXs). The need for distance in surveillance missions
Index Terms—Hypothetical phase compensation (HPC), also leads to a further reduction in the maximum unambiguous
millimeter-wave radar, time-division multiplexing multiple input speed of the radar. For an MIMO radar system, a vehicle traveling
multiple output (TDM-MIMO), velocity ambiguity resolution. at normal speed on an urban road may produce more than
two times the speed ambiguity period. Unfortunately, incorrect
I. INTRODUCTION target velocity estimation in MIMO radar systems will affect the
accuracy of target angle estimation due to the coupling between
ILLIMETER-WAVE radar has been used in urban traffic
M applications due to its high environmental resistance and
high speed, distance, and angle measurement accuracy charac-
velocity and angle. Therefore, decoupling velocity ambiguity is
one of the crucial techniques for MIMO imaging radar in traffic
surveillance applications.
teristics [1]. For example, traffic detection radar based on phased Resolving Doppler ambiguity is a challenging problem. Uti-
array modulation is used to implement traffic flow [2], speed [3], lizing the multi-pulse-repetition-frequency (PRF) scheme is the
and traffic event detection. With the application and develop- most commonly used method, including the Chinese remainder
ment of multiple-input multiple-output (MIMO) technology, theorem [13], the 1-D set algorithm [14], the lookup table
method algorithm [15], and the multifrequency observation al-
Manuscript received 6 September 2023; revised 9 November 2023 and 20 gorithm [16]. Taking a typical Chinese residual theorem (CRT)
December 2023; accepted 6 January 2024. Date of publication 10 January algorithm as an example, the radar alternately transmits two
2024; date of current version 23 January 2024. This work was supported in frame signals with different chirp periods so that the same target
part by the Technology on Near-Surface Detection Laboratory under Grant
6142414211202, in part by the Civil Aerospace Technology Advanced Research is estimated to have different velocities in different transmitted
Project under Grant D020403, and in part by the Basic Research Project under waveforms, and then, the target velocity ambiguity cycle is
Grant 50236170112202. (Corresponding author: Hua Zhang.) obtained by solving for the maximum common divisor between
Bo Yang, Siqi Liu, and Hua Zhang are with the School of Aerospace Science
and Technology, Xidian University, Xi’an 710071, China (e-mail: byang_18@ the two speeds. In addition, several improved algorithms have
stu.xidian.edu.cn; [email protected]; [email protected]. been proposed to increase the robustness of the CRT algorithm,
edu.cn). including optimizing the remaining PRF by constraining the
Yongjun Zhou is with the Science and Technology on Near-Surface Detection
Laboratory, Wuxi 214035, China (e-mail: [email protected]). minimum sidelobes of the maximum likelihood criterion [17],
Digital Object Identifier 10.1109/JSTARS.2024.3352082 proposing a phase unwrapping algorithm to reduce the effect of

© 2024 The Authors. This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see
https://creativecommons.org/licenses/by-nc-nd/4.0/
3410 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 17, 2024

the residual error on the integer solution [18], and presenting the prevent the interference of anomalous shock points com-
closed-form robust CRT to reduce algorithm complexity [19]. pared to selecting only the peak value as a decision.
However, the need for at least two sets of frequency-modulated 3) Numerical simulations show that the proposed algorithm
continuous-wave signals makes the radar system less real time, has higher accuracy and better velocity and angle estima-
and the matching between targets in multitarget scenarios in- tion stability than those of the HPC algorithm. The superi-
creases the complexity and error of the algorithm [20]. ority of the proposed algorithm becomes more significant
To alleviate the above problems, some single-frame velocity as the number of targets in the same distance–velocity
deambiguity schemes have been proposed. In [21], a resolution Doppler cell increases.
scheme via exploiting carrier frequency multiplexing is pro- 4) Two radar physical platforms and real traffic scenarios
posed. However, the method only requires a set of primary linear are demonstrated for performance verification and com-
frequency-modulated continuous-wave (LFMCW) signals but parison of the proposed algorithm. The experiments show
increases the signal bandwidth, which puts higher demands on that the proposed algorithm has a better imaging effect
the radar analog-to-digital converter sampling rate. To avoid than that of the HPC algorithm in both planar imaging
the modification of the radar transmit waveform, literature [22] and spatial stereo imaging, and the experimental results
utilizes overlapping elements in the antenna array to achieve are consistent with the simulation results.
the Doppler ambiguity period solution, but it also limits the The rest of this article is organized as follows. Section II
antenna array layout. In addition, some researchers have utilized provides a millimeter-wave radar signal model and theoretically
nonlinear phase components of the linear frequency-modulated analyzes the impact of velocity ambiguity on MIMO radar angle
signal echo to resolve this ambiguous estimation [23], but it also estimation. Sections III and IV elaborate on the improved HPC
increases the computational effort. algorithm model and the unified conditions for algorithm perfor-
In [24] and [25], the hypothetical phase compensation (HPC) mance verification. Sections V and VI validate and analyze the
algorithm was proposed based on the relationship between the functionality and performance of the proposed algorithm from
velocity-induced Doppler phase and the peak of the target an- both digital simulation and actual scenarios. Finally, Section VII
gular power spectrum, i.e., the more accurate the Doppler phase concludes this article.
compensation is, the higher the peak of the criterion angular
power spectrum. Theoretically, in addition to the low-latency II. RADAR SIGNAL MODELING AND INFLUENCE OF VELOCITY
characteristics common to single-frame velocity ambiguity res- ON ANGLE
olution scheme, the HPC algorithm increases the maximum
estimated speed of the radar system by MTX (transmitter antenna A. Radar Signal Model
number) times via searching for the hypothetical case where In the LFMCW radar system, the radar transmits a sawtooth
the peak of the target angular power spectrum is the largest signal (called chirp) through the TX, which can be expressed as
among all the Doppler phase compensation assumptions, which   
1 2
makes it more suitable for dynamic scenarios of urban vehicle s(t) = exp j2π fo t + kt (1)
surveillance. 2
Although the HPC algorithm is easy to implement, the peak where fo is the starting frequency and k = B/Tc is the slope,
angular power spectrum is highly influenced by noise making where B is the bandwidth and Tc is the duration of chirp. Here,
the HPC algorithm less stable. In particular, the algorithm fails we consider the initial phase to be 0 and do not consider energy
when there are multiple targets within the same distance Doppler amplitude changes.
cell, which is detrimental to the implementation of vehicle The transmitted signal is received by the radar receiving
surveillance via planar or spatial stereo imaging using MIMO antenna (RX) after being reflected by the target, and the echo
radar. Therefore, we propose an improved HPC algorithm and signal can be expressed as
verify its performance through simulation and practical tests.   
The contributions of this article are summarized as follows.  1 2
s (t) = exp j2π fo (t − τ ) + k (t − τ ) (2)
1) First, we select the highest value of the angular power 2
spectrum signal-to-noise ratio (SNR) in each hypothetical
where τ is the electromagnetic wave travel time. The beat
case instead of the peak power to form the SNR variation
frequency signal is obtained when the echo signal is mixed with
curve. Compared to the peak power spectrum, the peak of
the transmitted signal, which can be expressed as
SNR is more stable, which avoids the false detection of
 
spurious peaks due to noise fluctuations or the superposi- 1 2
y = exp j2π(fo τ + ktτ − kτ . (3)
tion of multiple target flap energies. 2
2) Next, a comparison is made between the first and second
peaks in the SNR curves. If the first peak is much larger To obtain the distance and speed of the target, the radar emits
than the second peak, the Doppler phase corresponding multiple chirp signals in one frame processing time. Assuming
to the first peak is considered the actual target phase. that the initial distance between a target and the radar is R and
Otherwise, the Doppler ambiguity period is judged by the radial velocity is v, the expression for τ is
the magnitude of the mean value in the waveform inter- R + vt + v(l − 1)Tc
val. Introducing the second spectral peak can effectively τ (t) = 2 ∗ (4)
c
YANG et al.: VELOCITY AMBIGUITY RESOLUTION ALGORITHM BASED ON IMPROVED HYPOTHETICAL PHASE COMPENSATION 3411

where c is the speed of light, l = 1, 2, 3, . . . , L, and L is the total

number of chirp in a data frame. Then, the expression of (4) can
be changed to
2kR 2vfo
fb ≈ + (5a)
c c
2vfo
fd = (5b)
c
2fo R
φl = fd (l − 1)Tc + (5c)
c
y(l) = exp(j2π(fb )t)exp(j2πφl ) (5d)
where fb denotes the frequency of the beat signal, which is re-
lated to the distance of the target (assuming that the displacement
of the target in one frame time is ignored), φl denotes the phase
of the lth echo, which is related to the speed of the target, and
fd denotes Doppler.
Then, expression (6) can be approximated by the
two-dimensional fast Fourier transform (2-D FFT) Fig. 1. TDM-MIMO model.
 LTc  Tc
S= exp (j2π (f1 − fb ) t1 ) dt1 The target angle can be obtained by processing the phase
0 τ difference between the channels utilizing 3-D FFT (uniform line
∗exp (j2π (f2 − fd ) t2 ) dt2 . (6) array) or digital beam synthesis methods.

Obviously, the signal power is maximum when f1 = fd and

C. Influence of Velocity on Angle
f2 = fb . Therefore, in practical engineering applications, the
2-D FFT is used to obtain the range–Doppler power spectral However, (7) only applies to stationary targets. When there is a
matrix (RDM), and the RDM is fed into the constant false alarm velocity on the target, there is also a target Doppler-related phase
detector to obtain frequency points (fb and fd ) related to the factor between the signals due to the time difference between
target distance and velocity. the signals emitted by the TX. Then, the signal expression is

ϕ = 4πvtarget M Tc /λ (8a)
B. TDM-MIMO Model  ϕ 
In a radar system, at least two or more RXs are required to Ym = S ∗ exp (j2πΦm ) ∗ exp j ∗ (m − 1) . (8b)
M
estimate the target angle information. In the TDM-MIMO mode,
the radar realizes the RX aperture expansion using time-sharing Therefore, Doppler phase compensation must be applied to
signal transmission at the transmitter side, and its effect is the data from different channels before estimating the target
consistent with the result of increasing physical antennas. When angle.
the number of TXs and RXs of the radar system is MTX and NRX , To more intuitively understand the effect of the target speed
respectively, the maximum virtualizable number is MTX ∗ NRX . when it exceeds the maximum unambiguous speed of the radar,
For the convenience of description, we assume that there are we give the results of a simulation experiment (shown in Fig. 2).
M TXs and one RX and illustrate the TDM-MIMO mode of Assume that the target speed is V
operation in Fig. 1. Each TX sequentially transmits a chirp
signal and is received by the antenna, and M RDMs are ob- V = V0 + N ∗ Vmax (9)
tained through data separation, recombination, and 2-D FFT where V0 ∈ [0, Vmax ] and Vmax is the radar maximum unambigu-
operations. ous speed in the positive, because positive and negative speeds
There is a phase difference (Φ) in the signal between the give the same effect. N = 1, 2, . . . , 9 denotes the multiplier,
receiving channels due to different positions of the receiving which represents how many times the current speed exceeds
channels. Assuming RX1 and TX1 as reference antennas, the the maximum unambiguous speed of the radar. As N increases
signal expression received by each channel is (target velocity increases in multiples of Vmax ), the difference
Ym = S ∗ exp (j2πΦm ) = S ∗ exp (j2πdm sinθ/λ) (7) between the target velocity detected by constant false alarm rate
detector (CFAR) and the actual target velocity keeps growing [as
where dm represents the position between the mth antenna and shown in Fig. 2(a)]. At the same time, the radar angle estimation
the reference antenna, m = 1, 2, . . . , M , and d1 = 0. Obviously, error also increases gradually [as shown in Fig. 2(b)]. In addition,
with the same range–Doppler unit, there is only a one-phase as the angle of the target with respect to the radar center is larger,
difference between the signals received from different channels the error between the value of the estimated angle and the actual
due to different antenna positions. angle is larger. Therefore, resolving velocity blur is important
3412 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 17, 2024

Fig. 3. Three different phase hypotheses for the same detection velocity with
three TXs.

where q represents the ambiguity period, q ∈ Z. Only when

the target speed is less than the maximum unambiguous speed,
ϕcfar = ϕtrue ; otherwise, there is a phase difference of 2π integer
multiples. Assuming that TX1 is the reference antenna, the phase
generated by the other TXs is evenly divided by 2qπ
i
ϕtrue − ϕcfar = (2qπ), i ∈ [1, MTX − 1] . (11)
M
After compensating only ϕcfar , all the channels have individ-
ual constant phase shifts, given by (11), which depend on the
number of TXs and form unique patterns.
Furthermore, we have expanded the number of antennas to 5,
i.e., the compensated phase patterns with q range from −3 to
Fig. 2. Effect of target velocity on radar velocity estimation and angle estima- 3 for three different antenna arrays having three, four, and five
tion when the target velocity is greater than the maximum unambiguous radar transmitters are shown in Fig. 4, respectively. In the case of TX =
velocity. (a) Relationship between the maximum unambiguous speed multiplier 3, when the target Doppler ambiguity period is not greater than
(N ) and velocity estimation error. (b) Relationship between the maximum
unambiguous speed multiplier (N ) and angle estimation error. 1 (|q| <= 1), the channel phases in each hypothetical case are
different from each other (H0 = H1 = H−1 ), i.e., the Doppler
phase caused by the target velocity corresponds uniquely to H.
for MIMO radar imaging, especially in cases with large blur When the ambiguity period is greater than 1 (|q| > 1), there is
periods. a situation where H is the same (H2 = H−1 , H−2 = H1 ), i.e.,
Doppler phase does not correspond uniquely to H. To eliminate
III. IMPROVED HPC ALGORITHM velocity ambiguity, the channel phases between different H must
be different from each other. Therefore, the effective value of q
A. Phase Relationship and the HPC Algorithm Overview is |q| <= 1. Similarly, when TX = 4 and TX = 5, the effective
Usually, we use the Doppler phase obtained by the CFAR range of values for q is |q| <= 1 and |q| <= 2, respectively.
detector to implement compensation. According to the trans- Then, some conclusions can be obtained.
mission mode of TDM-MIMO, it represents the phase difference 1) As the number of TXs increases, the nonambiguous phase
between two transmitted signals from the same TX. However, range becomes larger (as shown in the green area in Fig. 4),
this only applies to situations where the target velocity is less which means that the range of q values becomes wider.
than the maximum unambiguous velocity of the radar. 2) In theory, the relationship between q and the TX is:
We define Ha as a hypothetical case regarding the target if MTX is odd, then |q| <= (MTX − 1)/2; otherwise,
velocity, where a represents the value of q, i.e., the Doppler |q| <= (MTX /2) − 1.
ambiguity period. Fig. 3 shows the relationship between the Different H corresponds to different Doppler compensation
target true Doppler phase (ϕtrue ) and the detection phase (ϕcfar ) values; according to (8), the Doppler phase compensation results
obtained by the CFAR detector with q from −1 to 1 (H−1 , H0 , will also have an impact on the peak of the target angular power
and H1 ). The relationship between phases can be expressed as spectrum, i.e., the closer the compensated Doppler value is to
the true target velocity, the higher the peak of the target angle
ϕtrue = ϕcfar + 2qπ (10) power spectrum. Based on the amplitude change characteristics,
YANG et al.: VELOCITY AMBIGUITY RESOLUTION ALGORITHM BASED ON IMPROVED HYPOTHETICAL PHASE COMPENSATION 3413

are in the same range–Doppler cell, the HPC algorithm has a high
probability of failure, which is unfavorable for high-resolution
MIMO radar moving target imaging. To alleviate the above is-
sues, we propose an improved HPC algorithm named HPC-SNR
(in this article, the original algorithm is called the HPC-Peak).
A complete HPC-SNR model mainly includes the parameter
input part, Doppler phase compensation and angle estimation
part, angle spectrum (SNR) part, SNR peak value curve part,
and decision making (as shown in Fig. 5). Moreover, the pro-
cessing flow of using the HPC-SNR algorithm to resolve speed
ambiguity to obtain the actual velocity and angle of the target is
as follows.
Step 1 (Parameter input): Before implementing the algorithm,
the detection phase (φcfar ), channel dataset (S), and hypothetical
number (q) need to be determined. Through the CFAR detector,
the distance and the Doppler index of the target in the RDM can
be obtained, and φcfar can be calculated based on the Doppler
index. At the same time, we extract data from the same in-
dex position in each channel RDM to form a channel dataset
(S = s(1,1) , s(2,1) , . . . , s(MTX ,1) , s(2,1) , . . . , s(MTX ,NRX )) for an-
gle estimation. The specific q value can be calculated based on
the number of TXs in the radar system.
Step 2 (Doppler phase compensation and angle estimation):
First, based on the q value, list all the HPC cases
H = H−q , . . . , Hk , . . . , Hq . (13)
Any H represents a hypothetical case where the compensation
Fig. 4. Compensated phase pattern for three, four, and five transmitters. phase is (φcom )
φkcom = φcfar + 2kπ, k = −q, −q + 1, . . . , q. (14)
the HPC algorithm is proposed. The algorithm principle is: in Next, phase compensation is applied to the channel data under
the nonambiguous interval, the Doppler phase generated by the each hypothetical scenario
real target uniquely corresponds to H, and the peak value of the
S K = f (S, Hk ) (15)
target’s angular power spectrum is highest after compensation.
 
Then, the target speed can be obtained by comparing the peak φkcom
values of the angle power spectrum at each H. skm,n = sm,n ∗ exp −j ∗ (m − 1) (16)
MTX
In addition, there is a fixed relationship between the Doppler
ambiguity period and the target velocity where f (S, Hk ) represents the compensation function of chan-
nel data S under the Hk hypothesis, and its compensation
Vtrue = Vcfar + 2qVmax (12) method is shown in (15). Then, FFT or DBF operations are
performed on the compensated channel data (S k ) to obtain the
where Vtrue represents the true speed of the target, Vcfar repre-
angular power spectrum. Unlike the HPC-Peak algorithm, we
sents the estimated speed obtained from the detection results
convert the angular power spectrum into an SNR spectrum.
of the CFAR detector, and Vmax represents radar maximum
Finally, an angle SNR spectrum will be obtained for each H
unambiguous speed. If the speed defuzzification algorithm is not
case, with a total of 2q + 1.
implemented, the maximum detectable speed range of the radar
Step 3 (SNR peak spectrum mapping and processing): A
is only [−Vmax , +Vmax ]. However, using the HPC algorithm,
curve about the change in the magnitude of the SNR am-
it is possible to extend the maximum detectable speed of the
plitude is formed by extracting the maximum value of the
radar to [−MTX Vmax , +MTX Vmax ] (if MT X is odd), which greatly
angular SNR spectrum in each H in turn, where the hori-
improves the ability of MIMO radar to detect moving targets.
zontal coordinate indicates which H the SNR value belongs
to and the vertical coordinate indicates the magnitude of the
B. Improved HPC Algorithm Model SNR value. Then, the largest peak value (Dk ), the second
Although the HPC algorithm can highly expand the speed largest peak value (Dp ), and the neighboring values in the
detection range of MIMO radar systems and is easy to implement curve (Dk−1 , Dk+1 , Dp−1 , and Dp+1 ) were found by spectral
in engineering, its stability is poor as the q interval increases peak search. Dk , Dp , AV Ek , and AV Ep were sent to the de-
due to the susceptibility of power spectrum peaks to noise, cision part, where SU Mk = (Dk−1 + Dk + Dk+1 ), SU Mp =
interference, and other targets. Especially, when multiple targets (Dp−1 + Dp + Dp+1 ), and A denotes the mean value.
3414 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 17, 2024

Fig. 5. HPC-SNR algorithm flow.

Step 4 (Decision making): The true Doppler phase of the target

is obtained based on the judgment conditions
⎧
⎪
⎨Hk (φk ), if Dk  Dp
Htrue (φtrue ) = Hk (φk ), else if SU Mk ≥ SU Mp (17)
⎪
⎩
Hp (φp ), otherwise
where Htrue (φtrue ) denotes the assumption that corresponds to
the real situation (Doppler generated by the true velocity of the
target), based on which the true velocity of the target can be
obtained.

IV. SIMULATION AND REAL EXPERIMENT CONDITION

OVERVIEW
In this article, MATLAB-based numerical simulations and
real experiments based on actual physical platforms and traffic
scenarios are implemented to validate the performance of the
proposed algorithm and compare it with the HPC-Peak algo-
rithm. In order to make the results of the simulation more
credible, as well as to maintain all the experiments uniform
and facilitate analysis, all the parameter configurations in the
simulation experiments are kept consistent with those of the
actual radar system, including the radar antenna array layout, the
radar beam parameters, and the signal processing flow. However,
the results of the simulation experiments can be considered as
the results in an ideal environment since the objects are ideal
point targets and the background noise is Gaussian white noise.
First, in our experiments, we verify the performance of the
proposed algorithm in terms of both planar imaging and spatial
stereo imaging of millimeter-wave radar. Fig. 6 shows two an-
tenna array configurations for simulation and real-world testing,
both consisting of 12 TXs and 16 RXs but with different array
layouts. In array A [as shown in Fig. 6(a)], most antennas
are distributed in the azimuthal dimension resulting in a high
directional angular resolution, and thus, the array is used for Fig. 6. Antenna array layout of radar systems. (a) Antenna array A for planar
planar imaging simulation and testing. In array B [as shown in imaging. (b) Antenna array B for spatial imaging.
Fig. 6(b)], the antennas are uniformly distributed in the azimuth
and pitch dimensions, and thus, the array is used for spatial
imaging simulation and testing.
Next, the simulation model and the actual radar configuration etc. The parameters are shown in Table I. In the table, superscript
parameters are the same, i.e., the radar has the same performance, 1 denotes a planar imaging radar system parameter, and super-
including distance resolution, speed resolution, angle resolution, script 2 denotes a spatial imaging radar system parameter; a
YANG et al.: VELOCITY AMBIGUITY RESOLUTION ALGORITHM BASED ON IMPROVED HYPOTHETICAL PHASE COMPENSATION 3415

TABLE I
RADAR SYSTEM PARAMETERS

Fig. 7. Radar signal processing process.

Fig. 9. (a)–(d) Schematic diagram of amplitude or SNR changes under differ-

ent H conditions in the planar imaging system.

signal transmission, receive, and target raw data acquisition. In

the IF board, we use the physical architecture of FPGA+RAM to
provide a platform for software development. At the same time,
six pieces of DDR memory (two on the FPGA side and four
on the RAM side) are used to support the data processing and
data flow of the algorithm. The final processing result will be
transmitted through the network port for statistics and analysis.

V. NUMERICAL SIMULATIONS AND ANALYSIS

A. Single-Angle Signal Simulation Experiment
Fig. 8. Radar system physical platform. (a) Planar imaging radar system.
(b) Space imaging radar system. Based on two antenna arrays and radar signal process-
ing models, the HPC-SNR algorithm function is implemented
parameter without a superscript indicates that the value is the for simulation validation and comparison with the HPC-Peak
same in both the radar systems. algorithm.
Finally, based on the antenna array and radar parameters, Fig. 9 shows the test results of both the HPC algorithms
we constructed a MATLAB simulation program and the radar with the planar imaging antenna array. The peaks of the angular
system with the same signal processing process for algorithm power spectrum and the angular SNR spectrum are highest only
performance verification. Fig. 7 shows the signal processing when the compensated phase coincides with the Doppler phase
flow in the simulation experiment, mainly including the target produced by the target’s actual velocity (H true case), as shown
data setting and raw data acquisition part, the signal processing in Fig. 9(a) and (b). The peaks of the angular spectra at each H are
part, including the proposed algorithm, and the result statistical extracted to form a peak change curve; then, H corresponding
analysis part. In actual experiments, the radar system platform to the highest point of the curve is the correct case, as shown in
provides a carrier for implementing the algorithm, as shown Fig. 9(c) and (d). Similarly, the same test results are obtained
in Fig. 8. The radar system consists of a radio frequency (RF) under spatial imaging antenna array, as shown in Fig. 10. Only
board and an intermediate frequency (IF) board. The RF board if the correct phase compensation (H true case) is performed,
based on four radar chips and antenna arrays is used for radar the target position and the correct number of targets are clearly
3416 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 17, 2024

Fig. 10. (a)–(f) Schematic diagram of amplitude or SNR changes under

different H conditions in the spatial imaging system.

presented in the target azimuth–pitch angle spectrum or SNR

spectrum. The normalized curve shows that the peak or SNR
value at the target is the highest at the H true case.
Multiple independent repetitions of the experiment were im- Fig. 11. Functional simulation results of the HPC-SNR algorithm. (a) Velocity
estimation error under different maximum unambiguous velocity multipliers
plemented with the target at different angles, as shown in Fig. 11. (N ). (b) Angle estimation error under different maximum unambiguous velocity
Although the target velocity is increasing, the error between the multipliers (N ).
target actual velocity and the estimated velocity value does not
exceed 0.6˜m/s, and the error between the target real angle and
the estimated angle value does not exceed 0.15◦ .
Through several repetitive simulation experiments, it can be
concluded that our proposed improved algorithm has the ability
of velocity ambiguity, can correctly solve for the correct velocity
and angle of the target under planar imaging and spatial imaging,
and has the same processing cycle as the original algorithm.
Moreover, the performance of the proposed HPC-SNR algo-
rithm is similar to that of the HPC-Peak algorithm when there is
only a single target or single angle signal in the range–Doppler
cell.

B. Multiangle Signal Simulation Experiment

The most significant advantage of HPC-SNR is that when
multiple targets are in the same range–Doppler unit, it can stably
and correctly give all targets’ true speed and angle informa-
tion, which is needed for traffic imaging applications. Fig. 12
shows the velocity disambiguation performance of HPC-Peak Fig. 12. (a)–(i) Simulation results of the HPC-Peak and the HPC-SNR algo-
and HPC-SNR algorithms when multiple targets are in the same rithms with different number of targets.
YANG et al.: VELOCITY AMBIGUITY RESOLUTION ALGORITHM BASED ON IMPROVED HYPOTHETICAL PHASE COMPENSATION 3417

range–Doppler cell. In the simulation, multiple angle values

were randomly selected from −40◦ to 40◦ . Angles with different
values are randomly combined to form angle test groups, such
as (θi , θj ), (θi , θj , θp ) or (θi , θj , θp , θq ), where (θi = θj = θp =
θq ). In addition, the difference between angular values belonging
to the same angle test group is larger than the angular resolution
of the radar.
We use the error between the estimated angle obtained after
velocity compensation and the actual value to measure the
algorithm’s performance, as shown in Fig. 12(a), (b), (d), (e),
(g), and (h). Fig. 12(c), (f), and (i) shows the angular power
spectrum waveform after HPC algorithm compensation and FFT
operation. The FFT points represent the angle frequency points
of the target, i.e., targets at different angles are located at different
frequency locations. The number of spectrum peaks represents
the number of targets.
It can be seen that some angle groups may make the HPC-Peak
algorithm fail when multiobject velocity disambiguation. Once
HPC-Peak fails, there will be a large error between the estimation
angle of the target and the real one (the maximum error can reach
5◦ ), as shown by the red dot area in Fig. 12(a), (d), and (g). The
number of peaks in the angular power spectrum does not match
the true number of targets (as shown in the red angular spectrum
waveform in Fig. 12(c), (f), and (i), i.e., the number of targets
is misestimated. On the contrary, the HPC-SNR algorithm has Fig. 13. (a)–(f) Comparison and analysis of algorithmic failure results in
always maintained good performance in Doppler compensation. planar imaging.
Throughout the experiment, the error between the estimated
◦
angle and the actual value does not exceed 0.5 , and the number
of peaks of the power spectrum is consistent with the number of
actual targets, as shown by the blue dot area and waveform in
Fig. 12(b), (e), (h), (c), (f), and (i).
Fig. 13 shows the results of one of the repeated experiments.
The parameters of the simulation experiment are: the number of
targets is three, the target speed is v ∈ (5vmax , 6vmax ), and the tar-
get angle is θ = [0◦ , 3◦ , 6◦ ]. According to the phase relationship,
the correct compensation should be H3 . Fig. 13(a) and (b) shows
the detection results of the two HPC algorithms. In the HPC-Peak
algorithm, the highest peak point is H5 , so the algorithm cannot
get the correct phase compensation value. However, the HPC-
SNR algorithm’s SNR curve is highest only at H3 so that the
target data can get the correct Doppler phase compensation. The
fundamental reason for the HPC-Peak algorithm’s failure is that
the angular spectrum’s power variation is unstable. Fig. 13(c)
and (d) shows the angular power spectrum for the H5 and H3 Fig. 14. (a)–(d) Comparison and analysis of algorithmic failure results in
cases. In H3 , the angular power spectrum has three clear target spatial imaging.
peaks, while in H5 , the angular power spectrum has the wrong
number of heights and a distorted waveform. However, the value [0.6◦ , 3.1◦ , 5.6◦ , 8.1◦ ]. Utilizing the HPC-Peak algorithm does
of the highest peak in H5 is larger than that in H3 due to the effect not give the correct compensation results, and there are only
of the bottom noise, so the correct result cannot be obtained three targets in the azimuth–pitch angle power spectrum [as
using the HPC-Peak algorithm. However, the problem can be shown in Fig. 14(a) and (c)]. However, the correct target Doppler
effectively avoided by utilizing the HPC-SNR algorithm because compensation value and the number of targets can be obtained
the magnitude of the SNR is unaffected despite the elevated using the HPC-SNR algorithm [as shown in Fig. 14(b) and (d)].
target peak, as shown in Fig. 13(e) and (f).
Similarly, the same situation occurs in planar arrays. The pa-
rameters of the simulation experiment are: the number of targets C. Discussion and Summary Under Simulation Experiments
is four, the target speed is v ∈ (5vmax , 6vmax ), the target azimuth Through the experiment, the following conclusions can be
angle is [0.6◦ , 3.1◦ , 5.6◦ , 8.1◦ ], and the target elevation angle is obtained.
3418 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 17, 2024

Fig. 16. Schematic diagram of algorithm performance evaluation methods.

(a) Algorithm performance evaluation method based on point cloud image
location. (b) Algorithm performance evaluation method based on point cloud
Fig. 15. Traffic test scenarios. (a) Planar imaging traffic scene. (b) Spatial image location trajectory.
imaging traffic scene.

1) The proposed improvement points do not affect the algo- Fig. 15(b). The radar center is taken as the coordinate origin,
rithm’s ability to deal with speed ambiguity. In particular, the direction of the radar beam (parallel to the lane direction) as
the performance of the HPC-SNR and HPC-Peak algo- the Y -axis, and the perpendicular to the direction of the radar
rithms is similar under single-angle signal or single-target beam (perpendicular to the direction of the lane) as the X-axis.
conditions. The target speed away from the radar is defined as positive
2) When multiple targets are in the same range–Doppler cell, speed, otherwise negative speed. To facilitate the experimental
the performance of the HPC-Peak algorithm is unstable. comparison with the simulation results, the parameters of the
Once the algorithm fails, there is a large error between the two radar systems in the actual experiment are the same as in
estimated angle and the actual value, and the number of Table I.
targets is incorrectly estimated. However, the HPC-SNR It is inconvenient to accurately measure the speed of vehicles
algorithm performs well in multiobject situations, and the traveling on the road in a practical test. Therefore, to better
angular power spectrum can correctly reflect the number of characterize and compare the performance of the algorithms, we
targets, which is crucial for millimeter-wave radar imaging define an evaluation method that is easy to statistically and oper-
applications. ationally evaluate before testing. According to the relationship
3) The angle estimation error of the HPC-SNR algorithm between target velocity error and angle, it can be seen that the
is smaller under multiangle signals, even when both the wrong velocity compensation will make the angle estimation
algorithms are valid. fail, which makes the target’s position in space necessarily
4) The proposed algorithm is independent of the antenna deviate from the actual position. The trajectory of a usually
array. It is suitable for both planar imaging under the radar traveling vehicle is bound to coincide with the lane. Therefore,
1-D array and spatial imaging under the radar 2-D array. in this article, we utilize the locations and trajectories of vehicle
point cloud images formed by radar imaging or continuous imag-
VI. PRACTICAL SCENARIO EXPERIMENTS AND ANALYSIS ing operations to evaluate the performance of the algorithms.
As shown in Fig. 16(a), if the position of the vehicle’s point
A. Test Scenario and Evaluation Method Definition
cloud image should be the same as the actual lane where the
We verify and compare the performance of the proposed algo- vehicle is located, then the algorithm is effective; otherwise,
rithm on real roads by testing on vehicles traveling at different the algorithm fails. At the same time, we use the target’s point
speeds, in different directions, and on different lanes. Fig. 15 cloud image trajectory to assess the algorithm’s stability. As
illustrates two algorithmic test scenarios. In scenario A, the radar shown in Fig. 16(b), if the trajectory formed by the target’s point
with antenna array A is set up on an overpass to perform planar cloud image overlaps with the lane during continuous imaging,
images of vehicles in the lane in a top-down view, as shown in then the algorithm has good stability. If the trajectory is discrete
Fig. 15(a). In scenario B, the radar with antenna array B is set and fluctuating, the algorithm has poor stability. Similarly, the
up on the side of the roadway to perform the spatial imaging of algorithm accuracy can also be evaluated by fitting the center of
vehicles on the road from a side-view perspective, as shown in the target trajectory.
YANG et al.: VELOCITY AMBIGUITY RESOLUTION ALGORITHM BASED ON IMPROVED HYPOTHETICAL PHASE COMPENSATION 3419

Fig. 18. Small vehicle planar point cloud image trajectory comparison and
analysis.

target point cloud images. First, since the maximum detection

range of the radar system used in the TDM-MIMO mode is 50 m,
the results of the target point cloud image in the 20–40 m range
are selected for the experiment to be counted and analyzed. On
the one hand, choosing the middle range allows the vehicle to
always be within the effective imaging range of the radar. On
the other hand, when the target is too close to the radar, strong
reflections from the target can cause the point cloud image to
extend. When the target is too far away from the radar, the weak
Fig. 17. (a)–(o) Functional testing of speed disambiguation algorithms in field reflection of the target’s energy results in a sparse point cloud.
scenarios.
All these affect the statistics, and analysis of the radar point cloud
imaging results should be avoided. Next, the target point cloud
B. Functional Verification image is obtained through radar system reception signals and
Based on two types of millimeter-wave radar platforms and signal processing with the HPC algorithm, and multiple frame
real targets, the importance of velocity ambiguity resolution images are aggregated to form the target point cloud trajectory.
algorithms is demonstrated in both planar imaging and stereo- Finally, the performance and practicality of the algorithm are
scopic imaging, as shown in Fig. 17. Fig. 17(a)–(e) verifies in evaluated by analyzing the point cloud trajectory, including
a real data way that the Doppler generated by the target speed trajectory position, trajectory fluctuation size, target velocity
affects the correctness of the radar angle estimation. When the consistency, and the number of abnormal point cloud images.
target speed exceeds the radar maximum detectable speed, the Figs. 18–21 show the effectiveness of the proposed HPC-SNR
point cloud image position of the vehicle obviously deviates algorithm in continuous planar imaging and continuous spatial
from the original lane if the speed disambiguation process is imaging applications of vehicles and compare it with the original
not taken. The greater the velocity, the more obvious the point HPC-Peak algorithm. We mark the parts where the results of the
cloud image position deviation. When the speed disambiguation two algorithms differ with red boxes or circles and show the point
algorithm is applied, the radar’s point cloud imaging correctly cloud image of that frame independently for easy comparison.
reflects the actual position and shape of the vehicle, i.e., the At the same time, some imaging frames that are correct for both
vehicle’s point cloud image is always in the correct lane position the algorithms are marked with purple boxes.
regardless of the speed and direction of travel. Figs. 18 and 19 show the imaging results of the planar imaging
As demonstrated in Fig. 17, our improvements to the HPC- radar at a top-view angle. When the object of radar planar
Peak algorithm do not affect the algorithm’s ability to solve imaging is a small vehicle (as shown in Fig. 18), the imaging
the speed ambiguity. The proposed HPC-SNR can improve the results obtained by the HPC-SNR algorithm and the HPC-Peak
radar’s maximum detectable speed range, and the improved algorithm are almost the same, both for the vehicle point cloud
range is basically consistent with the theory. Under the radar trajectory formed by continuous imaging and for the target point
system in this article, the radar-detectable target speed has been cloud image in a single frame. The point cloud image trajectory
higher than the radar maximum unambiguous speed [as shown in is consistent with the actual trajectory of the target. According
Fig. 17(f) and (o)], and the maximum speed can be up to 18˜m/s to statistics, in planar imaging experiments on small vehicles,
[e.g., Fig. 17(g) and (i)]. In addition, the proposed algorithm is the proportion of abnormal frames in the HPC-SNR algorithm
applicable in both the planar and spatial imaging applications. is 2.5%. The proportion of abnormal frames in the HPC-Peak
algorithm is 3.5%.
However, as the size of the vehicle target increases, points
C. Performance Verification
with abnormal positions and velocities appear in the target point
We evaluate the algorithm’s stability and practical application cloud trajectory, which makes the point cloud trajectory discrete.
effectiveness through the trajectories formed by continuous Compared to the HPC-SNR, there are more abnormal points in
3420 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 17, 2024

Fig. 19. Large vehicle planar point cloud image trajectory comparison and analysis.

Fig. 20. Small vehicle spatial point cloud image trajectory comparison and analysis.

the results of the HPC-Peak algorithm, as shown in the red- side view, and top view for easy observation. Fig. 20 shows the
circled area in Fig. 19. When we show the imaging results of continuous spatial imaging results of the spatial imaging radar
one of the frames with discrepancies separately (as shown in the system for small vehicles. Although the objects imaged by the
red box in Fig. 19) and compare them, we can see that the vehicle radar system are both small vehicles, there are large differences
point clouds are more concentrated, and the velocities between between the trajectory images formed by the two algorithms.
the point clouds are approximately the same (with the same The HPC-SNR algorithm remains effective almost continuously
orange velocity points) in the imaging results of the proposed throughout the imaging section, and it can correctly solve the
HPC-SNR algorithm. In contrast, in the imaging results of the target’s actual velocity so that the target point cloud image is in
HPC-Peak algorithm, there are some points with completely the correct position, and the point cloud trajectory is consistent
wrong velocities (green velocity points), impacting the imaging with the target’s actual trajectory. However, the performance
quality. According to statistics, in planar imaging experiments of the HPC-Peak algorithm decreases significantly. When the
on large vehicles, the proportion of abnormal frames in the HPC- vehicle is far from the radar, the performance of the HPC-Peak
SNR algorithm is 8.5%. The proportion of abnormal frames in algorithm remains the same as that of the HPC-SNR, but as the
the HPC-Peak algorithm is 13.5%. distance between the car and the radar decreases, the HPC-Peak
Figs. 20 and 21 show the imaging results of spatial imaging begins to fail. Compared to the results of planar imaging, the
radar in roadside scenes. For the point cloud trajectories formed error in spatial imaging is more significant when HPC-Peak
by the same target, we present them in three ways: front view, fails. The radar completely misestimates the vehicle’s speed
YANG et al.: VELOCITY AMBIGUITY RESOLUTION ALGORITHM BASED ON IMPROVED HYPOTHETICAL PHASE COMPENSATION 3421

Fig. 21. Large vehicle spatial point cloud image trajectory comparison and analysis.

and direction of travel, and the location of the point cloud

image is far from the correct lane. According to statistics, in
spatial imaging experiments on small vehicles, the proportion
of abnormal frames in the HPC-SNR algorithm is 10.5%. The
proportion of abnormal frames in the HPC-Peak algorithm is
26.5%.
Fig. 21 shows the results of continuous imaging of a large
vehicle by the spatial imaging radar. The performance of the
HPC-Peak algorithm deteriorates further, and the algorithm
suffers from failures even when the target is far from the radar.
For the HPC-SNR algorithm, the overall results are significantly
better than the HPC-Peak algorithm, even though there are some
anomalies in the point cloud image. According to statistics, in
spatial imaging experiments on large vehicles, the proportion
of abnormal frames in the HPC-SNR algorithm is 18.5%. The
proportion of abnormal frames in the HPC-Peak algorithm is
55.5%.

D. Discussion and Summary Under Practical Experiments

We analyzed the results of the point cloud images that ap-
peared in the actual experiments, and there were four cases.
Fig. 22. Four cases in the actual test.
1) When both the HPC-Peak and HPC-SNR algorithms are
effective, the wave peak value in the correct H case is the
highest and much larger than the other positions both in the radar systems, spatial radar systems can form more point
normalized angular power spectrum curve and normalized clouds due to the planar array antennas allowing the radar
SNR curve, as shown in case 1 in Fig. 22. to detect target height information, further increasing the
2) The advantage of the proposed HPC-SNR algorithm is probability of multiple angular information (or multiple
that it can still implement the correct velocity estimation targets) within the same range–Doppler cell. Multiple
stably and efficiently when there are multiple targets or angular information can exist in the same distance Doppler
angular information in the same range–Doppler cell. In cell when imaging large vehicles compared to small ones,
practice, increasing target size increases the probability especially in spatial imaging. At this point, using angular
that different parts of the target itself fall in the same spectral peaks for decision making is unreliable since
distance Doppler region. Compared to planar imaging the angular signals interfere with each other. As shown
3422 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 17, 2024

in case 2 in Fig. 22, the peak at the true H position in

the normalized angular power spectrum curve is not the
highest, at which point the HPC-Peak algorithm fails. The
more angular information there is, the larger the algorithm
error, which is consistent with the previous numerical sim-
ulation results (shown in Fig. 12). However, the HPC-SNR
algorithm can remain effective because the bottom noise is
less affected by the SNR. Therefore, the advantage of the
HPC-SNR algorithm in experiments with large vehicles is Fig. 23. Small vehicle planar point cloud image trajectory in long range.
more obvious.
3) Relying only on the highest SNR value for decision
making is unstable. As in case 3 in Fig. 22, the actual
experiment also shows that the location of the maximum
SNR value in the normalized SNR curve is not the actual H
location due to the nonideal environment, which cannot be
found in the simulation experiment. To solve this problem,
when the value of the highest wave crest is similar to
that of the second-highest wave crest, the mean value of
the waveform interval is used for the secondary decision. Fig. 24. Large vehicle planar point cloud image trajectory in close range.
Obviously, the mean value of the true H position interval is
higher than the mean value of the false H position interval,
M times (when M is an odd number), where M is the
since the higher the phase compensation accuracy, the
number of transmission antennas.
higher the angular power spectrum is maximized.
2) Compared to the original HPC algorithm, the proposed
4) However, adding the wave interval mean judgment does
HPC-SNR algorithm is more accurate and stable, espe-
not completely solve all the problems. As in case 4 in
cially when there are multiple targets or angle signals in
Fig. 22, there exists an SNR curve in which the highest
the same range–Doppler cell, and is also more suitable for
point is not the correct H position, and the mean value of
MIMO radar imaging applications.
the wrong H interval is also larger than the mean value of
3) Experiments have shown that the proposed algorithm is
the correct H position interval. The HPC-SNR algorithm
suitable for both radar planar imaging and spatial stereo
will fail at this point, corresponding to the experiment’s
imaging, i.e., it is not affected by the antenna array layout.
outlier part. Fortunately, the number of anomalies and the
However, there are also failures of HPC-SNR, which can
percentage of occurrence probability are very small and
affect the imaging quality. In the future, we would like to
have a small impact on imaging.
correct or remove outliers in the imaging traces by using a joint
The overall planar and spatial imaging experiments show that
clustering algorithm.
the proposed HPC-SNR algorithm is significantly better than
the HPC-Peak algorithm and is more suitable for the imaging
applications of MIMO radar systems, which is consistent with APPENDIX
the simulation conclusions. The HPC-SNR algorithm maintains EXPLANATION OF TARGET POINT CLOUD IMAGES IN CLOSE
a good velocity disambiguation capability, which is not the RANGE AND LONG RANGE
case with the HPC-Peak algorithm, despite the fact that there Here, we add two additional experiments (shown in Figs. 23
are multiple angular signals in the same range–Doppler cell and 24) to illustrate two main points: first, the proposed algo-
due to the increase of the target size or the increase of the rithm is not limited by distance and is applicable throughout the
angular dimensional information of the radar system. The two detection range of the radar system; second, the core purpose
radar systems with different antenna arrays also show that the of choosing the intermediate distance segment of the radar
proposed algorithm is not limited by the antenna arrays. system in the experiments in the main text is to make the target
present a better point-cloud image to facilitate the observation
and comparison of the algorithm performance.
VII. CONCLUSION When the target is too far from the radar, the target point
This article proposes a velocity estimation algorithm based cloud density presented by the radar system will decrease due
on improved hypothetical phase compensation (HPC-SNR) for to the influence of radar angular resolution and target reflection
TDM-MIMO radar traffic target imaging. Numerical simula- intensity, leading to sparse point cloud trajectories, especially
tions and real experiments are also provided and analyzed. The for small-volume targets. The target in Fig. 23 is the same as the
algorithm has the following characteristics and advantages. target in Fig. 19, but due to distance differences, the trajectory’s
1) The HPC-SNR algorithm still retains the advantages of density and area vary greatly. Obviously, the trajectory of the
the original HPC algorithm, i.e., in a single data frame target point cloud at a long distance is difficult to observe and
(without changing the radar transmission waveform), the analyze. However, the proposed HPC-SNR algorithm remains
maximum detectable speed of the radar is increased by effective.
YANG et al.: VELOCITY AMBIGUITY RESOLUTION ALGORITHM BASED ON IMPROVED HYPOTHETICAL PHASE COMPENSATION 3423

When the target is too close to the radar, the strong reflection [15] S. Zhang, “An improved method of range ambiguity resolution using look-
from the target causes the point cloud image to expand as up table,” Inf. Res., vol. 41, no. 2, pp. 23–26, 2015.
[16] X. Hu, Y. Li, M. Lu, Y. Wang, and X. Yang, “A multi-carrier-frequency
well as out multipath effect, and the occlusion of the target random-transmission chirp sequence for TDM MIMO automotive radar,”
itself causes the target image to be incomplete, especially for IEEE Trans. Veh. Technol., vol. 68, no. 4, pp. 3672–3685, Apr. 2019.
large volume targets. The target in Fig. 24 is the same as the [17] X.-G. Xia, “Doppler ambiguity resolution using optimal multiple pulse
repetition frequencies,” IEEE Trans. Aerosp. Electron. Syst., vol. 35, no. 1,
one in Fig. 20, but the target’s proximity to the radar leads pp. 371–379, Jan. 1999.
to multipath effects (as shown by the red circle in Fig. 24) [18] X.-G. Xia and G. Wang, “Phase unwrapping and a robust Chinese remain-
and incompleteness in the point cloud image. Obviously, these der theorem,” IEEE Signal Process. Lett., vol. 14, no. 4, pp. 247–250,
Apr. 2007.
additional interferences are not conducive to the observation and [19] W. Wang and X.-G. Xia, “A closed-form robust Chinese remainder theo-
comparison of the algorithm performance and are not the focus rem and its performance analysis,” IEEE Trans. Signal Process., vol. 58,
of this article. Therefore, the point cloud trajectories of targets no. 11, pp. 5655–5666, Nov. 2010.
[20] I. Bilik et al., “Automotive MIMO radar for urban environments,” in Proc.
that are particularly close to the radar are not considered in the IEEE Radar Conf., 2016, pp. 1–6.
experiments in the main text. However, the proposed HPC-SNR [21] C. Zhang, M. Cao, Y. Li, Y. Gong, and Y. Huang, “Velocity ambiguity
algorithm is still effective in close-range imaging. resolution for wideband automotive millimeter wave radar: A carrier
frequency multiplexing framework,” J. Electromagn. Waves Appl., vol. 34,
Improving the radar point cloud density at long distances and no. 3, pp. 375–389, 2020.
reducing the scattering of close-range targets is a hot research [22] C. M. Schmid, R. Feger, C. Pfeffer, and A. Stelzer, “Motion compensation
topic in radar imaging. These phenomena can occur in both and efficient array design for TDMA FMCW MIMO radar systems,” in
Proc. 6th Eur. Conf. Antennas Propag., 2012, pp. 1746–1750.
the planar and spatial imaging radar systems, but these are not [23] M. Dikshtein, O. Longman, S. Villeval, and I. Bilik, “Automotive radar
the focus of this article and do not affect the application of the maximum unambiguous velocity extension via high-order phase compo-
proposed HPC-SNR algorithm. nents,” IEEE Trans. Aerosp. Electron. Syst., vol. 58, no. 1, pp. 743–751,
Feb. 2022.
[24] H. A. Gonzalez, C. Liu, B. Vogginger, and C. G. Mayr, “Doppler ambiguity
REFERENCES resolution for binary-phase-modulated MIMO FMCW radars,” in Proc.
Int. Radar Conf., 2019, pp. 1–6.
[1] A. Prabhakara et al., “High resolution point clouds from mmWave radar,” [25] C. Liu, H. A. Gonzalez, B. Vogginger, and C. G. Mayr, “Phase-based
in Proc. IEEE Int. Conf. Robot. Automat. (ICRA), 2023, pp. 4135–4142, Doppler disambiguation in TDM and BPM MIMO FMCW radars,” in
doi: 10.1109/ICRA48891.2023.10161429. Proc. IEEE Radio Wireless Symp., 2021, pp. 87–90.
[2] H. Liu, N. Li, D. Guan, and L. Rai, “Data feature analysis of non-scanning
multi target millimeter-wave radar in traffic flow detection applications,”
Sensors, vol. 18, no. 9, 2018, Art. no. 2756.
[3] E. Klinefelter and J. A. Nanzer, “Automotive velocity sensing using
millimeter-wave interferometric radar,” IEEE Trans. Microw. Theory
Techn., vol. 69, no. 1, pp. 1096–1104, Jan. 2021.
[4] X. Li, X. Wang, Q. Yang, and S. Fu, “Signal processing for TDM
MIMO FMCW millimeter-wave radar sensors,” IEEE Access, vol. 9,
pp. 167959–167971, 2021.
[5] B. Liu, “Orthogonal discrete frequency-coding waveform set design with Bo Yang (Student Member, IEEE) received the B.S.
minimized autocorrelation sidelobes,” IEEE Trans. Aerosp. Electron. Syst., degree in telecommunications engineering from the
vol. 45, no. 4, pp. 1650–1657, Oct. 2009. Civil Aviation University of China, Tianjin, China,
[6] H. Sun, F. Brigui, and M. Lesturgie, “Analysis and comparison of MIMO in 2018. He is currently working toward the Ph.D.
radar waveforms,” in Proc. Int. Radar Conf., 2014, pp. 1–6. degree in navigation guidance and control with Xidian
[7] H. Cui, J. Wu, J. Zhang, G. Chowdhary, and W. R. Norris, “3D detection University, Xi’an, China.
and tracking for on-road vehicles with a monovision camera and dual Currently, he is undertaking and researching some
low-cost 4D mmWave radars,” in Proc. IEEE Int. Intell. Transp. Syst. scientific research projects on millimeter wave radar
Conf., 2021, pp. 2931–2937. imaging algorithm and their application in transporta-
[8] G. Li et al., “Pioneer study on near-range sensing with 4D MIMO-FMCW tion. His research interests include radar detection and
automotive radars,” in Proc. 20th Int. Radar Symp., 2019, pp. 1–10. radar imaging and their applications.
[9] T.-Y. Lee, V. Skvortsov, M.-S. Kim, S.-H. Han, and M.-H. Ka, “Application
of w-band FMCW radar for road curvature estimation in poor visibility
conditions,” IEEE Sens. J., vol. 18, no. 13, pp. 5300–5312, Jul. 2018.
[10] S. M. Sabery, A. Bystrov, P. Gardner, A. Stroescu, and M. Gashinova,
“Road surface classification based on radar imaging using convolu-
tional neural network,” IEEE Sens. J., vol. 21, no. 17, pp. 18725–18732,
Sep. 2021.
[11] M. Lei, D. Yang, and X. Weng, “Integrated sensor fusion based on 4D
MIMO radar and camera: A solution for connected vehicle applications,”
IEEE Veh. Technol. Mag., vol. 17, no. 4, pp. 38–46, Dec. 2022. Siqi Liu (Student Member, IEEE) received the B.S.
[12] M. Elbeialy, S. You, B. J. Jeong, and Y. Kim, “Target classification using degree in mechanical and electrical engineering in
frontal images measured by 77 GHz FMCW radar through DCNN,” Appl. 2021 from Xidian University, Xi’an, China, where
Sci., vol. 12, no. 20, 2022, Art. no. 10264. she is currently working toward the master’s degree
[13] X. Huang, H. Wang, G. Huang, and J. Luo, “Doppler shift detection based in control science and engineering.
on Chinese remainder theorem and spectrum correction,” in Proc. IEEE Currently, she is undertaking and researching some
23rd Int. Conf. Digit. Signal Process., 2018, pp. 1–5. scientific research projects on millimeter-wave radar
[14] M. Li and M. Li, “A high efficiency algorithm of PD radar for range imaging algorithms and point cloud condensation
ambiguity resolution based on the one-dimension method,” Electron. Inf. algorithms. Her research interests include radar data
Warfare Technol., vol. 25, no. 5, pp. 22–25, 2010. processing and 4-D millimeter wave radar imaging.
3424 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 17, 2024

Hua Zhang received the Ph.D. degree in circuits Yongjun Zhou received the M.D. degree in com-
and systems from Xidian University, Xi’an, China, munications engineering from Shanghai Jiao Tong
in 2011 . University, Shanghai, China, in 2010 .
In 2019, he became a Professor with the School of He is currently a Senior Engineer with the Science
Aerospace Science and Technology, Xidian Univer- and Technology on Near-Surface Detection Labo-
sity, where he is the Director of the Department of ratory, Wuxi, China. His research interests include
Guidance, Navigation and Control. In recent years, Radar detection, near-ground target detection tech-
he has undertaken and completed more than 20 major nology, and ammunition smart fuze technology.
national and ministerial scientific research projects
on radar system design, millimeter-wave radar navi-
gation, and protection technology. He has edited three
textbooks and authored or coauthored more than 40 SCI papers. He holds more
than 25 parents. He is a peer reviewer of many significant journals, such as IEEE
JOURNAL OF QUANTUM ELECTRONICS and IET Radar Sonar and Navigation.
His research interests include radar imaging detection and application, quantum
timing positioning technology, time-varying astronomical signal processing, and
navigation mechanisms.