PLOS ONE

RESEARCH ARTICLE

Highly efficient photonic radar by

incorporating MDM-WDM and machine
learning classifiers under adverse weather
conditions
Sushank Chaudhary ID1*, Abhishek Sharma2*, Kuldeep Singh ID2, Sunita Khichar3*,
Jyoteesh Malhotra4

1 School of Computer, Guangdong University of Petrochemical Technology, Maoming, China, 2 Department

of Electronics Technology, Guru Nanak Dev University, Amritsar, India, 3 Department of Electrical
Engineering, Chulalongkorn University, Bangkok, Thailand, 4 Department of Electronics and Communication
Engineering, National Institute of Technology, New Delhi, India
a1111111111
a1111111111 * [email protected] (SC); [email protected] (AS);
[email protected] (SK)
a1111111111
a1111111111
a1111111111
Abstract
Photonic radar, a cornerstone in the innovative applications of microwave photonics,
emerges as a pivotal technology for future Intelligent Transportation Systems (ITS). Offering
OPEN ACCESS enhanced accuracy and reliability, it stands at the forefront of target detection and recogni-
Citation: Chaudhary S, Sharma A, Singh K, Khichar tion across varying weather conditions. Recent advancements have concentrated on aug-
S, Malhotra J (2024) Highly efficient photonic radar
menting radar performance through high-speed, wide-band signal processing–a direct
by incorporating MDM-WDM and machine learning
classifiers under adverse weather conditions. PLoS benefit of modern photonics’ attributes such as EMI immunity, minimal transmission loss,
ONE 19(4): e0300653. https://doi.org/10.1371/ and wide bandwidth. Our work introduces a cutting-edge photonic radar system that
journal.pone.0300653 employs Frequency Modulated Continuous Wave (FMCW) signals, synergized with Mode
Editor: Muhammad Nasir Khan, Government Division and Wavelength Division Multiplexing (MDM-WDM). This fusion not only enhances
College University Lahore, PAKISTAN target detection and recognition capabilities across diverse weather scenarios, including
Received: February 1, 2024 various intensities of fog and solar scintillations, but also demonstrates substantial resilience
Accepted: March 1, 2024 against solar noise. Furthermore, we have integrated machine learning techniques, includ-
ing Decision Tree, Extremely Randomized Trees (ERT), and Random Forest classifiers, to
Published: April 1, 2024
substantially enhance target recognition accuracy. The results are telling: an accuracy of
Copyright: © 2024 Chaudhary et al. This is an open
91.51%, high sensitivity (91.47%), specificity (97.17%), and an F1 Score of 91.46%. These
access article distributed under the terms of the
Creative Commons Attribution License, which metrics underscore the efficacy of our approach in refining ITS radar systems, illustrating
permits unrestricted use, distribution, and how advancements in microwave photonics can revolutionize traditional methodologies and
reproduction in any medium, provided the original systems.
author and source are credited.

Data Availability Statement: All relevant data are

within the manuscript and its Supporting
Information files.

Funding: The author(s) received no specific

funding for this work. 1. Introduction
Competing interests: The authors have declared Innovative applications of microwave photonics are revolutionizing the transportation sector,
that no competing interests exist. particularly in the realm of autonomous vehicle (AV) technology. By enhancing the precision

PLOS ONE | https://doi.org/10.1371/journal.pone.0300653 April 1, 2024 1 / 14

PLOS ONE Photonic radar by incorporating MDM-WDM and machine learning classifiers

and range of sensing capabilities, microwave photonics plays a crucial role in the advancement
of self-driving cars. AV’s, also known as self-driving cars, are capable of sensing their environ-
ment and navigating without human input. Autonomous operation of Intelligent Transporta-
tion Systems (ITS) requires real-time decision-making, which necessitates the training of ITS
using a machine learning (ML) approach [1]. Collecting and processing data from various sen-
sors, then training the ITS with a supervised learning model based on that processed data, is
crucial. Thus, it is important to have sensors that can collect data effectively even under chal-
lenging atmospheric conditions and low visibility for AV’s operation to detect, classify, and
recognize surrounding targets. Radio Detection and Ranging (RADAR) is considered as the
main and standard method for target detection, classification, and recognition in all weather
conditions [2]. Conventionally, radars is designed with electronic components and hence
results in limited functions with poor resolution, limited bandwidth and low speed. Hence to
overcome such disadvantages of conventional RADAR, researchers came with Photonics-
based technologies to conventional radar system [3]. These technologies were having discrete
features of modern photonics such as wider bandwidth, multifaceted multiplexing, rapid signal
processing, electromagnetic interference (EMI) immunity, flat response and minimal trans-
mission loss [4–6]. Based upon measurement method of photonic radar is termed as pulsed
radar and continuous wave radar. In pulsed radar system, the detection is carried out using
time of flight technique which is very effective method in remote sensing and aerial navigation.
However, continuous wave radar is preferred which require low input power and light weight
equipment due to use of solid-state devices. Continuous wave radar works with signal modula-
tion in amplitude, phase and frequency domain. In this work, we are primarily concentrating
on evaluating the performance characteristics of photonic radar systems that operate using
Frequency Modulated Continuous Wave (FMCW) technology. FMCW radar offers several
benefits, such as a relatively uncomplicated hardware configuration, lower peak power, and
the ability to obtain both range and velocity information of targets at the same time [7, 8].
FMCW based photonic radar operates in two configurations viz direct detection and Coherent
detection [9]. In direct detection method, backscattered signal is analysed using square law
detection scheme and hence the sensitivity of the system is towards intensity of the received
signal only. Coherent detection is sensitive to the phase, polarization, and intensity of the
incoming signal, as referenced in [10], operating on a principle of linear detection. Coherent
systems are further described as homogeneous and heterogeneous. For extended-distance
telemetry, the FMCW photonic radar system utilizes the Doppler effect alongside the time of
flight method [11]. Further power consumptions of data collecting sensors should be kept low
as limited power is available in AV’s. As FMCW based photonic radar employs solid state
components hence minimal power is needed for their operation as well as small size gives
more flexibility to automobile designer in attaining Low SWaP-C (Size, Weight, Power and
Cost) [12]. In 2020 [13], authors have proposed a microwave photonic radar, which employs a
post bandwidth synthesis technique, is capable of offering a bandwidth of 16 GHz and a range
resolution of 1 cm, according to the authors. In another work [14], The use of wavelength divi-
sion multiplexing (WDM) and radio over fiber technique has been suggested by authors to
create a photonic radar that can generate and distribute radar signals. This proposed micro-
wave photonic radar can achieve a range resolution of 7.3 cm. In another work [15], the
authors have shown the capabilities of a linear frequency-modulated continuous-wave pho-
tonic radar, which can track the radar cross-section of multiple moving targets even in adverse
weather conditions such as fog, cloud, and rain. Furthermore, in real-world traffic scenarios
that involve numerous mobile targets, detecting, associating data, and classifying targets
become even more difficult tasks. In 2021 [16], The authors have suggested the use of an X-
band photonic radar that utilizes a photonic frequency quadrupling scheme for real-time

PLOS ONE | https://doi.org/10.1371/journal.pone.0300653 April 1, 2024 2 / 14

PLOS ONE Photonic radar by incorporating MDM-WDM and machine learning classifiers

detection of low radar cross-section targets and high-resolution imaging. They have also
employed a balanced photo detector and delay interferometer to enhance the signal-to-noise
ratio. The proposed photonic radar has demonstrated the ability to detect targets located 2.7
km away from the radar. Recently, In 2022 [9], a comparison was made by authors between
the performance of direct detection FMCW-based photonic radar and coherent detection
FMCW-based photonic radar. The results indicated that coherent detection-based photonic
radar outperformed direct detection FMCW-based photonic radar, but it is more complex and
costly. Another study [17] proposed microwave photonic radar that utilizes sparse stepped fre-
quency (SSF) chirp signals to achieve ultra-high resolution. This photonic radar was able to
distinguish between two simulated point targets placed 8.3 mm apart. In this proposed work,
photonic radar is developed to detect multiple targets under diverse movement of vehicles. For
operational ease, linearly frequency modulated continuous wave (L-FMCW) based photonic
radar in coherent detection scheme is employed here. LFM chirp is used to modulate the input
signal and de-chirping is intelligible mixing of received echo signal with the local oscillator
(LO) [10, 18]. The multiple targets detection is accomplished by employing mode division
multiplexing (MDM) along with wavelength division multiplexing. MDM improves the secu-
rity and capacity of the optical system [19]. Four targets are considered in this work and their
detections under sever atmospheric conditions is verified. Lastly the data collected using pro-
posed system is further trained using supervised learning model of machine learning. Employ-
ing machine learning classifiers like Decision Tree (DT), Extremely Randomized Trees (ERT)
and Random Forest (RF), we enhanced the target recognition accuracy. The significant contri-
bution of presented work is as follows:
1. Robust Performance Evaluation Under Varied Fog Conditions: Demonstrated the pho-
tonic radar system’s efficacy in diverse fog scenarios, including heavy, medium, and light
fog, highlighting its adaptability and reliability in different meteorological environments.
2. Solar Noise Resilience Analysis: Conducted a thorough assessment of the system’s perfor-
mance under solar noise influences, particularly strong and weak scintillations, proving its
robustness against solar interference, which is critical for real-world ITS applications.
3. Integration of Advanced Mode division multiplexing-Wavelength division multiplexing
(MDM-WDM) and Machine Learning Techniques: Innovatively combined MDM-WDM
with sophisticated machine learning classifiers, significantly enhancing target detection
accuracy and recognition in adverse weather and solar conditions.
4. Benchmarking and Performance Metrics in Adverse Conditions: Achieved impressive
performance metrics such as high accuracy, sensitivity, and specificity, establishing new
benchmarks for photonic radar systems in ITS, especially under challenging environmental
conditions.
This paper firstly discusses the operation principle of photonic radar which is presented in
Section 2. In Section 3 system modelling is presented. Section 4 discusses the findings of pro-
posed model while Section 5 discusses training of the data set attained using proposed model.
Section 6 concludes the papers and discusses future directions in this work.

2. Operating principal
The efficiency of AV’s is defined by its ability to distinguish between different targets that are
closely spaced under adverse atmospheric conditions. This efficiency is termed as range

PLOS ONE | https://doi.org/10.1371/journal.pone.0300653 April 1, 2024 3 / 14

PLOS ONE Photonic radar by incorporating MDM-WDM and machine learning classifiers

resolution (LRes) and can be expressed arithmetically as in Eq 1 [20]:

c
LRes ¼ ð1Þ
2B
where c is speed of light in free space and B is bandwidth of the system. Hence it can be
deducted from the above equation that with high bandwidth, resolution will be much narrower
and hence more closely spaced targets can be distinguished by the system. The existing LIDAR
uses frequency band of 24 GHz which is having very limited bandwidth of 250 MHz also
known as Industrial Scientific and medical (ISM) band and thus does not make a suitable can-
didate for AV applications [21]. Instead of ISM band, short range radar (SRR) band (77 GHz)
that offers 4 GHz bandwidth is much suited for AV applications [22]. This higher bandwidth
of SSR band offers 20 times improved range resolution and 3 times velocity resolution in com-
parisons to ISM band. In the proposed photonic radar, Doppler-effect is considered for detec-
tion and ranging of targets. Doppler effects have been used in RADARs for target detection.
The transmitted radio signal is reflected from the target in terms of echoes and collected by the
receiver to extract data such as distance and velocity of the target. The efficacy of the RADAR
system is constrained at higher bandwidths in the microwave band due to the diminishing
impact of harmonics. Enhancing performance can be achieved by overlaying the microwave
signal onto an optical signal. Likewise, RADARs have larger antenna size and high aperture
diameters thus offer high beam divergence whereas photonic radar utilizes laser source which
is having narrow line-width and low aperture offers low beam divergence and are more suited
for AV applications [23].
The frequency of the echo signal received from the target, referred to as the range frequency
(fr), is mathematically represented as shown in Eq 2 [24]:
2�B�R
fr ¼ ð2Þ
Ts � c

Where B is bandwidth of the system, R is range of the target from photonic radar equipped
vehicle, Ts is sweep time and c is speed of light in vacuum. The system is further tested for
adverse atmospheric attenuation and scintillation effects. Lastly the data is utilised for training
of autonomous system using supervised learning model.

3. System modelling
The modeling of the proposed photonic radar system, which incorporates MDM-WDM, is
illustrated in Fig 1 of the manuscript. This system was simulated using OptiSystemTM and
MATLABTM software.
As shown, four transmitters are transmitting signal over a single channel and two wave-
lengths that is 1550 nm and 1550.1 nm. Every transmitter contains three sections: Linear Fre-
quency Modulator (LFM), CW source with corresponding mode generator and dual port
Mech Zehnder modulator (DMZM) modulator. In LFM section, pseudorandom signal is
encoded into triangular waveform. As triangular waveform provides fast sweep rate hence it is
chosen over other waveform [25, 26]. This encoded triangular signal is further modulated with
microwave frequency in LFM with 77 GHz radio signal having 4 GHz of bandwidth. The radio
frequency-modulated LFM signal is divided into two parts; one part is directed to the modula-
tor, while the other serves as a reference signal for the receiver. A continuous wave laser source
is used as carrier signal with low input power of -10 dBm. The optical carrier is the fed into
transverse mode generator where mode profiles are attached to the optical signal. In this work,
we have engaged Laguerre-Gaussian (LG) mode profile and is expressed mathematically as in

PLOS ONE | https://doi.org/10.1371/journal.pone.0300653 April 1, 2024 4 / 14

PLOS ONE Photonic radar by incorporating MDM-WDM and machine learning classifiers

Fig 1. Proposed 4 targets based photonic radar system (a) MDM-WDM Photonic Radar Configuration (b) Photonic
Radar Transmitter Configuration.
https://doi.org/10.1371/journal.pone.0300653.g001

Eq 3 [27]:
� �jn2j � � � 2� � �( )
2r2 n 2r2 r pr2 SinðjnjφÞ; n � 0
cm;n ðr; φÞ ¼ L m exp 2 exp j 2 ð3Þ
w2o w2o wo lRo CosðjnjφÞ; n � 0

where, m and n represent azimuthal and radial indexes. Ln,m represents Laguerre Polynomial, r
is radius and wo is spot size. Two modes has been used here that is LG 00 and LG 01 such that
channel 1 uses LG 00 and Channel 2 uses LG 01 on same wavelength of 1550 nm. Likewise
channel 3 uses LG 00 and channel 4 uses LG 01 on same wavelength of 1550.1 nm. Mode pro-
files of LG 00 and LG 01 is shown here in Fig 2.
The optical signal, created in a particular mode, undergoes modulation with the LFM signal
using a DMZM. The output signal from the DMZM is mathematically expressed as shown in
Eq 4 [24]:
!
Ein ðtÞ ðg � eðjpv2 ðtÞ=VpRF þjpvbias2 =VpDC Þþ
EðtÞ ¼ IL=20 � ð4Þ
10 ðð1 gÞ � eðjpv1 ðtÞ=VpRF þjpvbias1 =VpDC Þ

where, Ein (t) is intensity of optical signal, IL represents insertion loss, VπDC & VπRF signify
switching bias voltage and modulation voltage respectively. Signal from transmitter 1 and 2 is
combined with signal of transmitter 3 and 4 using wavelength division multiplexer. This

PLOS ONE | https://doi.org/10.1371/journal.pone.0300653 April 1, 2024 5 / 14

PLOS ONE Photonic radar by incorporating MDM-WDM and machine learning classifiers

Fig 2. Mode profiles of LG modes (a) LG 00 (b) LG 01.

https://doi.org/10.1371/journal.pone.0300653.g002

multiplexed signal is further amplified using an optical amplifier. Modeling of the channel,
taking into account severe atmospheric conditions, is conducted through a Gamma-Gamma
fading channel, and this is depicted mathematically in Eq 5 [28]:

dR 2 R
a10
PReceived ¼ PTransmitted � 2 10 ð5Þ
ðdT þ yRÞ

where, transmitter and receiver aperture diameters are given by dT & dR are respectively, beam
divergence is given by θ, effective distance between transmitter and receiver is given by R and
atmospheric turbulences are represented by α.
For acute atmospheric conditions we have considered fog conditions as visibility reduces
drastically with increasing fog thereby affecting the efficiency of AV. Different models has
been utilized for calculating attenuation due to rain and fog. Rain attenuation is calculated
using Kim model and is expressed in Eq 6 [29]:
Arain ¼ k � ro a ð6Þ

where extent of rain in mm/hr is given by ro while k and α are variables calculated by Marshall
Palmer distribution [30]. Likewise attenuation due to fog is calculated using Mie scattering
model and is mathematically expressed as in Eq 7 [31]:
� � r
3:91 l
ðlÞ ¼ ð7Þ
V 550

where scattering coefficient is given by ρ, operational wavelength of system is given by λ and

visibility in kilometres is given by V. In this work, we have adopted International visibility
code [32] as 0.1 dB/km for clear weather and 75 dB/km for heavy fog conditions. The received
signal in form of echoes is detected using PIN diode. The power received by PIN diode is
given as in Eq 8 [33]:
8
>
> rt D2 topt t2atm
>
< Pt for extended target
4R2
Pr ¼ ð8Þ
>
> rt At D2 topt t2atm
>
: Pt for any target
4R2 Aill

where, ρt is target reflectivity, τopt and τatm the optical and atmospheric transmission loss

PLOS ONE | https://doi.org/10.1371/journal.pone.0300653 April 1, 2024 6 / 14

PLOS ONE Photonic radar by incorporating MDM-WDM and machine learning classifiers

respectively. R is range of target from system, receiver aperture diameter is given by D, while
At is target area and Aill is illuminated target area. The photo detector output is mixed with
local oscillator signal and the signal is further passed through a low pass filter to obtain beat
signal. The beat signal is expressed mathematically as in Eq 9 [9]:
� �
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pb 2
Sb ðtÞ ¼ R � Alo � Plo � Pr cos 2pfstart t þ ðtÞ þ 2pfr ðtÞ sin½od ðtÞ þ ðyoðtÞ yloðtÞ ð9Þ
Tm

where, amplitude of FM modulator is given by Alo and input power of low pass filter is given
by Plo, < is the responsivity of photo detector, β is the modulation index, range frequency is
given by fr and initial frequency is given by fstart.

4. Results and discussion

The section discusses simulation results for a proposed MDM-WDM based photonic radar
system. Firstly, the system was tested under clear atmospheric conditions, where attenuation
was considered to be 0.1 dB/km. The detection of multiple targets under these conditions is
depicted in Fig 3.
In the depicted scenario, it is posited that four separate targets are positioned at varying
lengths from the photonic radar system: the first target at 25 meters, the second at 30 meters,
the third at 75 meters, and the fourth at 60 meters. It is observed that the range frequency for
target 1 is at 66.66 MHz, for target 2 it is at 80 MHz, for target 3 it is at 200 MHz, and for target
4 it is at 160 MHz. The range frequencies obtained from the simulation results depicted in
Fig 3 match with the theoretically calculated values of range frequency using Eq 2, indicating
error-free detection of multiple targets. The proposed photonic radar system is further tested
under adverse atmospheric conditions.

Fig 3. Maximum received power for multi-target detection at varying distances in clear atmospheric condition.
https://doi.org/10.1371/journal.pone.0300653.g003

PLOS ONE | https://doi.org/10.1371/journal.pone.0300653 April 1, 2024 7 / 14

PLOS ONE Photonic radar by incorporating MDM-WDM and machine learning classifiers

Fig 4. Multiple Target detection under heavy attenuation of (a) Low Fog at 12.5 dB/km (b) Medium Fog at 25 dB/km
and (c) Heavy Fog at 75 dB/km.
https://doi.org/10.1371/journal.pone.0300653.g004

Firstly, the system is tested for varying fog conditions. According to the international visi-
bility code, three conditions have been considered, namely low fog with an attenuation of 12.5
dB/km, medium fog with an attenuation of 25 dB/km, and heavy fog with an attenuation of 70
dB/km. Fig 4(A) shows the impact of low fog conditions when attenuation is kept at 12.5 dB/
km, as per the international visibility code. It can be observed that there is a loss of power
when comparing the maximum received power with clear weather conditions. However, suc-
cessful detection of all four targets has been reported. Similarly, in Fig 4(B), with medium fog
conditions, although further power loss is observed, all the targets have been successfully
detected. Finally, with heavy attenuation, it was observed that all targets are detected with less
power as compared to light and medium fog conditions.
Hence, the results indicate that the suggested model achieves accurate detection in all low,
medium and heavy fog conditions. Lastly, the system is tested for the impact of scintillation
effects. Two different scenarios are considered, with strong scintillation of 10−6 and weak scin-
tillation of 10−22. Fig 5 depicts the impact of scintillation on the proposed system. As shown in
Fig 5, the scintillation has a minimal impact on the signal intensity, and the results obtained in
weak as well as strong scintillation are comparable.
Table 1 shows the measured power levels of all detected targets for different atmospheric
conditions. It shows that all the targets have been successfully detected in all atmospheric
conditions.

PLOS ONE | https://doi.org/10.1371/journal.pone.0300653 April 1, 2024 8 / 14

PLOS ONE Photonic radar by incorporating MDM-WDM and machine learning classifiers

Fig 5. Multiple Target detection under (a) weak scintillations and (b) strong scintillations.
https://doi.org/10.1371/journal.pone.0300653.g005

5. Training with ML
In this section, we investigate the efficacy of various machine learning classifiers for enhancing
the target detection capability of our photonic radar system. We compared the performance of
six classifiers: DT, ERT, RF, Histogram-Based Gradient Boosting (HGB), Extreme Gradient
Boosting (XGB), and Adaptive Boosting (AB). DT is known for its simplicity and interpretabil-
ity, often used in various applications due to its straightforward decision-making process. ERT
is a type of ensemble learning technique that combines multiple decision trees to improve pre-
diction accuracy and control over-fitting. RF is another ensemble technique that uses a collec-
tion of decision tree classifiers to improve predictive performance and robustness. HBG
Boosting is a fast, scalable machine learning algorithm that builds an ensemble of decision
trees in a gradient boosting framework. XGB is an efficient and scalable implementation of
gradient boosting that is popular for structured or tabular data while AB is a technique that
combines multiple weak classifiers to form a strong classifier, adapting by focusing more on
instances that previous classifiers misclassified.
Our comparative analysis focused on key performance metrics: accuracy, sensitivity, speci-
ficity, F1 Score, and false discovery rate (FDR). A dataset was created for training and testing
the model in various conditions, including clear weather, strong and weak attenuation effects,

PLOS ONE | https://doi.org/10.1371/journal.pone.0300653 April 1, 2024 9 / 14

PLOS ONE Photonic radar by incorporating MDM-WDM and machine learning classifiers

Table 1. Measured power values for all targets under different atmospheric conditions.
Atmospheric Conditions Target Detected Frequency (MHz) Power (dBm)
Clear Weather 1 66.66 -49.60
2 80 -54.47
3 200 -55.42
4 160 -52.06
Low Fog 1 66.66 -50.08
2 80 -55.04
3 200 -56.84
4 160 -53.20
Medium Fog 1 66.66 -50.33
2 80 -55.34
3 200 -57.59
4 160 -53.80
Heavy Fog 1 66.66 -51.35
2 80 -56.57
3 200 -60.67
4 160 -56.26
Weak Scintillations 1 66.66 -49.60
2 80 -54.47
3 200 -55.42
4 160 -52.06
Strong Scintillations 1 66.66 -49.98
2 80 -54.85
3 200 -56.14
4 160 -52.85
https://doi.org/10.1371/journal.pone.0300653.t001

and low, medium, and heavy fog. The dataset includes four target vehicles located at different
positions on the road, as shown in Table 2.
Table 2 shows that in 6 different weather conditions, 4 different vehicles have been
attempted to be identified simultaneously. Therefore, a total of 24 samples have been generated
for training the model, which has 4096 elements of different frequency levels as input. The
input elements and output have been represented in numerical form. Similarly, the output vec-
tor has been created with 4 class categories of targets, representing Target 1 to Target 4.
Figs 6 and 7 depict the performance of proposed system by comparing key performance
metrics using different classifiers based upon inputs of Table 2.
Fig 6 shows the overall accuracy of each classifier. Among the classifiers, the Decision Tree
stands out with the highest accuracy of 91.51%, indicating its robustness in correctly identify-
ing targets under various conditions. This high accuracy demonstrates the effectiveness of the
DT classifier for the photonic radar system in terms of overall target classification accuracy.

Table 2. Data summary.

Weather Conditions Target Vehicles Category
Clear Weather Target1
Attenuation: Strong Target2
Weak Target3
Heavy Fog Target4
Medium Fog
Low Fog
https://doi.org/10.1371/journal.pone.0300653.t002

PLOS ONE | https://doi.org/10.1371/journal.pone.0300653 April 1, 2024 10 / 14

PLOS ONE Photonic radar by incorporating MDM-WDM and machine learning classifiers

Fig 6. Performance comparison of given ML classifiers in terms of accuracy for target classification.
https://doi.org/10.1371/journal.pone.0300653.g006

Fig 7 provides a detailed analysis of the performance of the given machine learning classifiers
in terms of sensitivity, specificity, F1 score, and FDR values for different targets. In Fig 7(A), the
DT and ERT classifiers demonstrate high sensitivity, with the Decision Tree achieving a
sensitivity of 91.47%. This suggests their effectiveness in correctly identifying true targets,
which is vital for avoiding missed detections in radar applications. Fig 7(B) assesses the clas-
sifiers’ ability to correctly identify non-targets (true negatives) in terms of specificity. The
DT shows high specificity of 97.17%, indicating a lower false alarm rate, crucial for radar
systems to avoid unnecessary responses to non-target objects. Similarly, Fig 7(C) highlights
the F1 Score, with the Decision Tree achieving a high score of 91.46%, indicating a balanced
and effective performance in both precision and sensitivity aspects. Moreover, Fig 7(D)
presents the FDR, with the Decision Tree maintaining a relatively low FDR of 8.54%, indi-
cating a higher reliability of the detections made by the classifier. The results of the pro-
posed method, as demonstrated by the DT classifier, are compared with other methods to
highlight its superior performance across multiple parameters. This comparison under-
scores the potential of the Decision Tree classifier as a reliable tool for our photonic radar
system, particularly beneficial for precise target detection and classification. The

PLOS ONE | https://doi.org/10.1371/journal.pone.0300653 April 1, 2024 11 / 14

PLOS ONE Photonic radar by incorporating MDM-WDM and machine learning classifiers

Fig 7. Performance Comparison of given ML classifiers in terms of (a) Sensitivity (b) Specificity (c) F1 Score (d) FDR
for different targets.
https://doi.org/10.1371/journal.pone.0300653.g007

incorporation of the DT classifier into our photonic radar system significantly enhances its
target detection proficiency, aligning with our goal of developing a highly accurate and effi-
cient system. Future work will explore the integration of these classifiers in real-world sce-
narios, further validating their effectiveness in operational environments.

6. Conclusion
In this study, we have showcased the innovative applications of microwave photonics through
the development of a MDM-WDM based photonic radar system, specifically tailored for
advancements in intelligent transportation. Our approach involved rigorous numerical simu-
lations for detecting four distinct targets under various atmospheric conditions, both clear and
adverse. The results demonstrated successful target detection, effectively aligning with the the-
oretical range-frequency predictions and highlighting the strengths of microwave photonics in
practical applications. Significantly, the integration of machine learning, particularly using
classifiers like DT, ERT, and RF, has markedly enhanced target classification. Achieving an
impressive accuracy of 91.51% with high sensitivity and specificity, this integration exemplifies
the synergy between advanced photonics and artificial intelligence. This marks a substantial
improvement over the previously reported SVM-based model accuracy. As we look to the
future, our focus will include expanding the dataset for further training enhancement and
implementing real-time test beds. These efforts aim to validate and refine our model in opera-
tional scenarios, further demonstrating the transformative impact of microwave photonics in
intelligent transportation systems.

PLOS ONE | https://doi.org/10.1371/journal.pone.0300653 April 1, 2024 12 / 14

PLOS ONE Photonic radar by incorporating MDM-WDM and machine learning classifiers

Supporting information
S1 File.
(ZIP)

Author Contributions
Conceptualization: Sushank Chaudhary.
Data curation: Abhishek Sharma, Sunita Khichar.
Methodology: Sushank Chaudhary.
Resources: Abhishek Sharma.
Software: Jyoteesh Malhotra.
Supervision: Jyoteesh Malhotra.
Validation: Kuldeep Singh.
Visualization: Kuldeep Singh, Sunita Khichar.
Writing – original draft: Abhishek Sharma.
Writing – review & editing: Sushank Chaudhary.

References
1. Yuan T., da Rocha Neto W., Rothenberg C. E., Obraczka K., Barakat C., and Turletti T., "Machine learn-
ing for next-generation intelligent transportation systems: A survey," Transactions on emerging tele-
communications technologies, vol. 33, no. 4, p. e4427, 2022.
2. Skolnik M., "An introduction and overview of radar," Radar Handbook, vol. 3, pp. 1.1–1.24, 2008.
3. De S. and Bazil Raj A., "A survey on photonics technologies for radar applications," Journal of
Optics, pp. 1–30, 2022.
4. Capmany J. and Novak D., "Microwave photonics combines two worlds," Nature photonics, vol. 1, no.
6, p. 319, 2007.
5. Urick V. J., Williams K. J., and McKinney J. D., Fundamentals of microwave photonics. John Wiley &
Sons, 2015.
6. Marpaung D., Roeloffzen C., Heideman R., Leinse A., Sales S., and Capmany J., "Integrated micro-
wave photonics," Laser & Photonics Reviews, vol. 7, no. 4, pp. 506–538, 2013.
7. J. A. Solano-Pérez, J.-M. Molina-Garcia-Pardo, J.-V. Rodrı́guez, L. Jofre, and A. Mateo-Aroca, "Com-
parison of using mm-wave FMCW radar IF signals against frequency-domain VNAs in the application of
the multifrequency bifocusing (MFBF) imaging algorithm," AEU—International Journal of Electronics
and Communications, vol. 117, p. 153103, 2020/04/01/ 2020, https://doi.org/10.1016/j.aeue.2020.
153103.
8. J. Shan, K. Rambabu, Y. Zhang, and J. Lin, "High gain array antenna for 24 GHz FMCW automotive
radars," AEU—International Journal of Electronics and Communications, vol. 147, p. 154144, 2022/04/
01/ 2022, https://doi.org/10.1016/j.aeue.2022.154144.
9. Sharma A. and Malhotra J., "Simulative investigation of FMCW based optical photonic radar and its dif-
ferent configurations," Optical and Quantum Electronics, vol. 54, no. 4, p. 233, 2022.
10. Adany P., Allen C., and Hui R., "Chirped lidar using simplified homodyne detection," Journal of light-
wave technology, vol. 27, no. 16, pp. 3351–3357, 2009.
11. P. Feneyrou et al., "Frequency-modulated multifunction lidar for anemometry, range finding, and veloci-
metry–1. Theory and signal processing," Appl. Opt., vol. 56, no. 35, pp. 9663–9675, 2017/12/10 2017,
https://doi.org/10.1364/AO.56.009663
12. Sharma A., Chaudhary S., Malhotra J., Saadi M., Al Otaibi S., and Wuttisittikulkij L., "Speed-Direction
Sensing under Multiple Vehicles Scenario Using Photonic Radars," CMC-COMPUTERS MATERIALS
& CONTINUA, vol. 73, no. 3, pp. 5399–5410, 2022.
13. Ma C., Chen H., Ye X., Wang X., and Pan S., "Ultra-high resolution microwave photonic radar with post-
bandwidth synthesis," Chinese Optics Letters, vol. 18, no. 7, p. 072501, 2020.

PLOS ONE | https://doi.org/10.1371/journal.pone.0300653 April 1, 2024 13 / 14

PLOS ONE Photonic radar by incorporating MDM-WDM and machine learning classifiers

14. Dong J., Zhang F., Jiao Z., Sun Q., and Li W., "Microwave photonic radar with a fiber-distributed
antenna array for three-dimensional imaging," Optics Express, vol. 28, no. 13, pp. 19113–19125, 2020.
https://doi.org/10.1364/OE.393502 PMID: 32672195
15. Sharma V. and Kumar L., "Photonic-radar based Multiple-Target Tracking under complex traffic-envi-
ronments," IEEE Access, vol. 8, pp. 225845–225856, 2020.
16. Bae Y., Shin J., Lee S.-G., and Kim H., "Field experiment of photonic radar for low-RCS target detection
and high-resolution image acquisition," IEEE Access, vol. 9, pp. 63559–63566, 2021.
17. Ma C. et al., "High-Resolution Microwave Photonic Radar With Sparse Stepped Frequency Chirp Sig-
nals," IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1–10, 2022.
18. Pierrottet D., Amzajerdian F., Petway L., Barnes B., Lockard G., and Rubio M., "Linear FMCW laser
radar for precision range and vector velocity measurements," MRS Online Proceedings Library (OPL),
vol. 1076, 2008.
19. S. Chaudhary and A. Amphawan, "Selective excitation of LG 00, LG 01, and LG 02 modes by a solid
core PCF based mode selector in MDM-Ro-FSO transmission systems," Laser Physics, vol. 28, no. 7,
p. 075106, 2018/05/25 2018, https://doi.org/10.1088/1555-6611/aabd15
20. Zhang F., Guo Q., and Pan S., "Photonics-based real-time ultra-high-range-resolution radar with broad-
band signal generation and processing," Scientific reports, vol. 7, no. 1, pp. 1–8, 2017.
21. K. Ramasubramanian and K. Ramaiah, "Moving from Legacy 24 GHz to State-of-the-Art 77-GHz
Radar," ATZelektronik worldwide, vol. 13, no. 3, pp. 46–49, 2018/06/01 2018, https://doi.org/10.1007/
s38314-018-0029-6
22. P. Zhao, N. Li, Y. Shan, and J. Bao, "Design and analysis of microstrip transceiver array antennas with
the function to suppress beam deflection for 77 GHz automotive radar," AEU—International Journal of
Electronics and Communications, vol. 162, p. 154587, 2023/04/01/ 2023, https://doi.org/10.1016/j.
aeue.2023.154587.
23. Piatek S. and Li J., "A photonics guide to the autonomous vehicle market," Laser Focus World, vol. 53,
no. 11, pp. 28–31, 2017.
24. Elghandour A. H. and Ren C. D., "Modeling and comparative study of various detection techniques for
FMCW LIDAR using optisystem," in International Symposium on Photoelectronic Detection and Imag-
ing 2013: Laser Sensing and Imaging and Applications, 2013, vol. 8905: International Society for Optics
and Photonics, p. 890529.
25. SHARMA A. et al., "A Cost-Effective Photonic Radar under Adverse Weather conditions for Autono-
mous Vehicles by incorporating Frequency Modulated Direct Detection Scheme," Frontiers in Physics,
p. 467, 2021.
26. Chaudhary S. et al., "Coherent detection-based photonic radar for autonomous vehicles under diverse
weather conditions," PLoS one, vol. 16, no. 11, p. e0259438, 2021. https://doi.org/10.1371/journal.
pone.0259438 PMID: 34780504
27. GHATAK A. A., Ghatak A., Thyagarajan K., and Thyagarajan K., An introduction to fiber optics. Cam-
bridge university press, 1998.
28. S. Chaudhary and A. Amphawan, "High-speed MDM-Ro-FSO system by incorporating spiral-phased
Hermite Gaussian modes," Photonic Network Communications, vol. 35, no. 3, pp. 374–380, 2018/06/01
2018, https://doi.org/10.1007/s11107-017-0752-6
29. Rashidi F., He J., and Chen L., "Spectrum slicing WDM for FSO communication systems under the
heavy rain weather," Optics Communications, vol. 387, pp. 296–302, 2017.
30. Yangang L., "Statistical theory of the Marshall-Palmer distribution of raindrops," Atmospheric Environ-
ment. Part A. General Topics, vol. 27, no. 1, pp. 15–19, 1993.
31. Olsen R., Rogers D., and Hodge D., "The aRbrelation in the calculation of rain attenuation," IEEE Trans-
actions on Antennas and Propagation, vol. 26, no. 2, pp. 318–329, 1978, https://doi.org/10.1109/TAP.
1978.1141845
32. Awan M. S., Csurgai-Horváth L., Muhammad S. S., Leitgeb E., Nadeem F., and Khan M. S., "Character-
ization of fog and snow attenuations for free-space optical propagation," JCM, vol. 4, no. 8, pp. 533–
545, 2009.
33. A. Sharma, J. Malhotra, S. Chaudhary, and V. Thappa, "Analysis of 2 × 10 Gbps MDM enabled inter sat-
ellite optical wireless communication under the impact of pointing errors," Optik, vol. 227, p. 165250,
2021/02/01/ 2021, https://doi.org/10.1016/j.ijleo.2020.165250.

PLOS ONE | https://doi.org/10.1371/journal.pone.0300653 April 1, 2024 14 / 14