International Conference on Computer & Communication Technology (ICCCT)-2011

Fuzzy logic controlled shunt active power filter for reactive power compensation
and harmonic elimination

Nitin Gupta and S. P. Singh S. P. Dubey

Department of Electrical Engineering Department of Electrical Engineering
Indian Institute of Technology (I.I.T.) R.C.E.T.
Roorkee-India Bhilai, India
[email protected], [email protected] [email protected]

Abstract— Active filters are widely used in electrical topology has been considered. Shunt APF harmonic current
distribution system for reactive power compensation and is equal but opposite in-phase with the required harmonic
voltage / current harmonic elimination. In this paper, a fuzzy component of load current. Computation of reference
logic controlled, three-phase shunt active filter to improve current signal and switching pulse generation for voltage
power quality by compensating reactive power and current source inverter (VSI) is main task. Various control
harmonics required by a nonlinear load is presented. PI approaches such as instantaneous reactive power theory [2],
regulator is replaced by fuzzy logic controller to improve the synchronous reference frame theory, closed loop PI control
dynamic performance of shunt active filter under varying load [9], etc. are used for generation of reference current signal
conditions. The advantage of fuzzy control is that it is based on
by sensing combinations of source voltage, load current,
defined linguistic rules and does not require any mathematical
filter current and source current. More number of sensor
model of the system unlike the other traditional controller. The
compensation process is based on source current sensing only, increase complexity and not cost effective solutions. Hence,
an approach different from conventional methods. The control algorithm should be such that which give it’s best
performance of fuzzy logic controller is compared with PI performance under steady-state as well as transient load
controller under dynamic load conditions. Simulated studies conditions with less sensor count. Duke et al. [10] have
show that fuzzy controller is found suitable for steady state and proposed simple synthetic sinusoidal generation technique
transient conditions of load. by sensing load current which further modified by sensing
source current only [11].
Keywords- Power electronics, power quality, shunt active The conventional methods use PI controller to regulate
filter, fuzzy logic controller, hysteresis-band, current controller. the DC bus voltage of VSI. Nemours methods have been
proposed to replace PI control scheme such as optimal
I. INTRODUCTION regulator control [12], sliding mode control [13], and model
reference adaptive control. However, the design of these
The rapid use of power electronic controlled equipment mentioned controllers depends on derived accurate
in electrical distribution system offers highly nonlinear mathematical models which are difficult to obtain. Also,
characteristics and produces voltage and current waveforms these models do not give satisfactory operation under
distortions called as ‘harmonics’ [1]. Some major problems varying load condition with increased nonlinearity [8], [14],
in electrical distribution system due to harmonics are well [15].
enumerated in [2-6]. Distorted current and voltage Artificial intelligence is one of the key area to solve such
waveform further affects other consumers connected to the system complexity and make control more robust for
same point of common coupling (PCC) by propagating transient conditions. Neural network, fuzzy logic, expert
these distortions in their premises. system, various other optimization methods are used for the
In order to overcome these power quality problems, improvement of power quality [15]. Fuzzy logic control
passive and active compensation (filters) approaches is (FLC) is one of the significant tool in control design
used. Passive compensation though simple approach for originated by Zadeh [16]. The advantages of FLC over
achieving above goals but they are associated with several conventional controllers are high robustness, insensitivity to
drawbacks such as, resonance problems, bulky, tuned for parameters variations, handling of non-linearity and
particular frequency component, component ageing, etc. [7], independent on mathematical models [8], [15], [17]-[19].
[8]. For overcoming the drawbacks of passive filters and This paper presents a fuzzy logic controlled shunt APF
reducing power quality problems, a number of attempts for the elimination of current harmonics and reactive power
have been made on the design, analysis and optimal control compensation of nonlinear load. The control scheme is
development of active power filter (APF). APF allows based on indirect current control scheme in which only
compensation of voltage harmonics (series APF), source current is sensed for avoiding switching spikes [20].
compensation of current harmonics (shunt APF) with Three phase voltage and current signal is sensed using
reactive power compensation required by nonlinear load. two voltage and current sensors. The effectiveness and
This paper emphasis on elimination of current harmonics validity of the proposed FLC is verified through MATLAB
and reactive power compensation, hence shunt APF simulation. The total harmonic distortion (THD) results

978-1-4577-1386-611$26.00©2011 IEEE 82
Authorized licensed use limited to: Malaviya National Institute of Technology Jaipur. Downloaded on May 26,2025 at 06:34:04 UTC from IEEE Xplore. Restrictions apply.
 International Conference on Computer & Communication Technology (ICCCT)-2011

between the conventional PI and proposed FLC are ∞

presented. It is shown by the simulation results that iL (t ) = ∑I f sin (nωt + φ n )
proposed controller based on fuzzy logic is robust that n =1
∞
conventional PI controller for steady-state and transient
conditions of load.
= I1 sin (ωt + φ f ) + ∑I n=2
n sin (nωt + φ n ) (2)

= iLf (t ) + iLh (t )
II. BASIC COMPENSATION PRINCIPLE
Fig. 1 (a) shows the basic APF compensation scheme where iLf and iLh are the fundamental and harmonic
including nonlinear loads with a three-phase supply system. component of load current. I1 and In are the peak values of
A current controlled VSI converter is used as an APF. APF fundamental and harmonic component of load currents,
is controlled to supply / extract compensating current to / respectively. Hence for given supply voltage the
from the utility (PCC). Fig. 1 (b) shows the different instantaneous power can be written as:
waveform of load current, source current and compensating p L (t ) = v s (t ) ∗ iL (t )
current of one of the phase ‘a’. = Vm I1 sin 2 ωt ∗ cos φ f + Vm I1 sin ωt ∗ cos ωt ∗ sin φ f
∞ (3)
+ Vm sin ωt ∗ ∑
n=2
I n sin(nωt + φn )

= pact (t ) + qreact. (t ) + ph (t )
where pact, qreact. and ph are active, reactive and harmonic
power of load. Out of these powers only first component
will be supplied by source. From (3), peak value of
fundamental component can be calculated as:
pact (t ) = Vm I1 sin 2 ωt ∗ cos φ f
(a) = (Vm sin ωt ) ⋅ ( I1 cos φ f ) sin ωt (4)
=v s (t ) ∗ is (t )
where I sm = I1 cos φ f is the peak value of the supply
side current. In order to incorporate switching losses and
converter losses, additional loss component of current will
be supplied by source in addition to the real power of the
load. Hence, the total peak current supplied by supply
source can be written as:
∗
I sm = I sm + I sLoss (5)
(b)
Figure 1. (a) Basic scheme of APF compensation (b) Different waveforms Conventionally, the actual capacitor voltage (Vdc) is
of phase ‘a’ load, source and compensating current
compared with a calculated reference value (Vdc,ref). The
error signal is then fed to a PI controller. The output of PI
A. Estimation of Reference current Supplied by source controller is considered as peak value of supply current
The peak value of the reference source current ∗
( I sm ). The DC side capacitor mainly serve two purpose, (a)
component is calculated by regulating DC link voltage of
it maintain DC bus voltage to set value with some steady-
VSI. Source only supplies active component of current and
state error, and (b) in transient period it serves as energy
rest of the component will be supplied by shunt APF for
storage element to supply real power difference between
power quality improvement. In addition to the active current
source and load. Finally, peak value of this source current
component source will also supplies loss component of
when multiplied with unit templates generated from supply
current. A PI regulator is normally used to maintain DC bus
voltage will give three reference source currents.
voltage constant and the output of this PI controller is
considered as required value of active current component. ∗
isa ∗
= I sm × u saf
An experimental study of shunt APF (Gupta et al. 200x) has
∗ ∗
shown that load current having fundamental and other isb = I sm × u sbf (6)
harmonic components which can be separated by Fourier ∗ ∗
series. Let the source voltage and instantaneous source isc = I sm × u scf
current and load current can be written as [21]:
v s (t ) = Vm sin ωt III. THE PROPOSED FUZZY CONTROLLER
(1)
is (t ) = iL (t ) − ic (t ) Fuzzy logic becomes more popular due to dealing with
problems that have uncertainty, vagueness, parameter

83
Authorized licensed use limited to: Malaviya National Institute of Technology Jaipur. Downloaded on May 26,2025 at 06:34:04 UTC from IEEE Xplore. Restrictions apply.
 International Conference on Computer & Communication Technology (ICCCT)-2011

variation and especially where system model is complex or have to consider while designing number of linguistic
not accurately defined in mathematical terms for the variables in view of accuracy and complexity of FLC.
designed control action. The conception of the fuzzy logic
introduced by Zadeh [16] is a combination of fuzzy set
theory and fuzzy inference system (FIS). Elements of a
fuzzy set belong to it with a certain degree, called degree of
membership. The degree of membership is a result of
mapping the input to certain rules using a membership
function (MF). The progression which maps the specified
input data to the output using fuzzy logic is known as fuzzy
inference. A fuzzy inference system can be classified as: (a)
fuzzification: which is the process of converting any crisp (a) error ‘e’ and change in error ‘ce’
value to analogous linguistic variable based on certain MF,
(b) inference engine: simulates human decision, (c)
knowledge base: consists MF definitions and necessary
rules like IF-THEN or it is combination of condition part
with their associated rules (d) defuzzification: is the
progression of transforming the fuzzy output into a crisp
numerical value. In this paper main control input variable is
the DC-link voltage error and output of FLC is the peak
value of the reference source current. The range of operating
current, normalization and de-normalization is one of the (b) change in reference output current
important design factors of fuzzy controller. The block Figure 3. Triangular shaped membership function used in fuzzification
diagram representation of fuzzy logic controller with
inference system is shown in Fig. 2. The scaling factor Ge, The error e and change of error ce at nth sampling instant
Gce, and Gu are used to scaling the input and outputs as per are used as input of FLC and can be written as [14]:
the designing of FLC.
e = Vdc,ref − Vdc
A. Designing of control rules (7)
ce(n) = e(n) − e(n − 1)
Computational methods determine the computational
efficiency, processor memory requirement and processing The output of FLC with limiter is considered as
time. The fuzzy control rules based on membership function ∗
amplitude of derived reference current ( I sm ). In this paper
defining or relate input variables to output variables. The seven triangular membership functions have been chosen for
number and type of MF determines the computational representing numerical variables into linguistic variables
efficiency of fuzzy control technique. The determination of [15], viz., NL (negative large), NM (negative medium), NS
MFs depends on the designer’s experience and knowledge. (negative small), ZE (zero), PS (positive small), PM
The shape decision of MFs affects how well a fuzzy system (positive medium), PL (positive large). The spacing between
rules approximate a function. MFs may be equal or unequal; it is set here for cover a band
Triangles or triangular membership function (TMF) of load current with good accuracy. After this rules
have been frequently used in several applications of FLC formation as knowledge base, different inference
[22], [23]. TMF are preferred due to simplicity, easy mechanisms have been developed for defuzzify fuzzy rules.
implementation, symmetrical along the axis. Fig. 3 shows In this paper, authors apply Mamdani’s max-min inference
the MFs relating input and output linguistic variables. The method to get an implied fuzzy set of tuning rules. Finally
number of linguistic variables is directly related to the center of mass method is used defuzzify the implied control
accuracy of approximating function and plays an important variables.
role for input-output mapping [24]. However, some limits

Figure 2. Detailed structure of fuzzy logic controller

84
Authorized licensed use limited to: Malaviya National Institute of Technology Jaipur. Downloaded on May 26,2025 at 06:34:04 UTC from IEEE Xplore. Restrictions apply.
 International Conference on Computer & Communication Technology (ICCCT)-2011

The above can be summarized as for implementing FLC: -choice of the energy storage element parameters, a high dc
(1) First, scaling factors consist of the normalization gain voltage Vdc, ameliorate the dynamics of the filter and
for input and de-normalization gain is selected properly. minimize the voltage ripple in the capacitor. As per the
(2) Rules decision based on accuracy and complexity. specification of peak of dc ripple voltage and rated filter
(3) Fuzzification, implication using mamdani’s operator current Ic,rated , the following relation can obtain for (Cdc):
and finally defuzzification to get desired output. π ⋅ I C ,rated
Fig. 4 shows the block diagram of the proposed FLC C dc = (8)
3 ⋅ ω ⋅ Vdcr , p − p
scheme with shunt active filter. The designed rules for
knowledge base are shown in Table 1. The top row and left The value of Cdc depends on the maximum possible
column of the matrix indicate the fuzzy sets of the variables variation in load and not on the steady state value of load
e and ce. current. Hence, proper forecasting in the load variation
reduces the value of Cdc. Further, filter inductor can be
TABLE I. FUZZY CONTROL RULE BASE
calculated as [26]:
e NB NM NS Z PS PM PB maVdc,ref
ce Lc,min = (9)
NB NVB NVB NVB NB NM NS Z (2 2 ) ⋅ ΔI sw, p − p ⋅ K L ⋅ f sw,max
NM NVB NVB NB NM NS Z PS
NS NVB NB NM NS Z PS PM V. HYSTERESIS CURRENT CONTROLLER
Z NB NM NS Z PS PM PB
Among different pulse width modulation (PWM)
PS NM NS Z PS PM PB PVB techniques, hysteresis-band current control technique has
PM NS Z PS PM PB PVB PVB proven to be most suitable for generating switching pulses
PB Z PS PM PB PVB PVB PVB for the switching device of VSI based active filter. It is
broadly used because of its ease and inherent-peak current
IV. DESIGNING OF APF PARAMETERS limiting capability without much information about system
Mainly, selection of active power filter inductor (LC), parameters. The actual source currents are monitored
DC link capacitor (Cdc) and it’s reference value (Vdc,ref) are instantaneously, and then compared to the reference source
the main parameters while designing the power circuit. The currents generated by the proposed fuzzy logic based
output of the bridge is a PWM voltage that has to be filtered control algorithm. The positive group device and the
by an inductance or a high order filter to limit the level of negative group device in one phase leg of VSI are switched
ripple current. The APF filtering inductor is used to reduce in complementary manner for avoiding a dead short circuit.
the ripple of VSI fed converter caused by the switching of The switching logic for ‘phase-a’ is formulated as
∗
the power devices [9]. Hence, the design of filtering follows for hysteresis-band width (HB): if isa < (isa − HB ) ,
inductor is based on principle of harmonic current reduction then upper switch is OFF and lower switch is ON in the
technique [25]. On the dc side of the Shunt APF, capacitor ∗
phase “a” leg then Sa =1 ; if isa > (isa + HB ) , then upper
supplies the dc voltage Vdc.
The quality of distortion compensation is affected by the switch is ON and lower switch is OFF in the phase “a” leg
then Sa=0.

Figure 4. Proposed fuzzy logic scheme based shunt active filter

85
Authorized licensed use limited to: Malaviya National Institute of Technology Jaipur. Downloaded on May 26,2025 at 06:34:04 UTC from IEEE Xplore. Restrictions apply.
 International Conference on Computer & Communication Technology (ICCCT)-2011

Similarly, the switching pattern of rest of phase’s transient due to accurate control of FLC block. The
devices can be derived using HB as the width of hysteresis variation of the dc capacitor voltage during the change in
band. load current can be observed from this figure. When the
load current is reduced or increased, capacitor voltage
VI. SIMULATION RESULTS increase or decrease to compensate for the real power
To simulate the proposed FLC based control scheme supplied by source. This change will be taken into account
with reduced sensors, a model in MATLAB\ SIMULINKTM by the FLC controller to adjust the peak value of reference
and SimPowerSystem Blockset is developed. The complete current.
3-phase active filter system is composed using a supply Under varying load the desired response of the FLC and
source, a voltage source inverter, coupling and smoothing APF system is reasonably fast and steady-state condition is
inductors with highly non-linear characteristic based load. reached within few cycles of AC mains. THD of source
Various simulations are carried out to verify the current is reduced from 26 % to 2.05 %, which shows a
performance of the active power filter using proposed FLC significant reduction in THD. Supply currents are
and conventional PI controller with during steady-sate and sinusoidal, balanced and slightly leading with respect to
transient conditions. The system parameters selected for supply voltage which is necessary to compensate line
simulation studies are given in Table 2. impedance drop.

TABLE II. SYSTEM PARAMETERS

System parameters Values
Supply voltage and frequency 230 V, 50Hz
Source impedance (Rs , Ls) 0.05 Ω, 0.5 mH
Filter impedance (Rc , Lc) 0.25 Ω, 4 mH
Smoothing inductor (Rsm , Lsm) 0.5 Ω, 1.5 mH
Load impedance (RL1, RL2, LL1, LL2) 25 Ω, 75 Ω,
10 mH, 15 mH
Reference DC link voltage (Vdc,ref) 680 volts
DC link capacitance (Cdc) 1650 μF
Switching frequency (fsw) 10-12 KHz
Figure 8. Transient response of fuzzy logic controlled shunt APF
Fig. 7 shows the performance results using fuzzy
controller of source voltage (VS), source, load and filter The same MTALAB model of three phase shunt APF is
current (IS, IL and IC) and DC side capacitor voltage (Vdc). tested for conventional PI controller by replacing fuzzy
For the clarity purpose, only one phase filter current is logic controller by PI controller. The value of proportional
shown. The filter is switched on at 0.06 s. and integral gains (Kp and Ki) are selected as 0.59 and 2.35,
Initially the three-phase load is 11-12 kW. The instant respectively. Fig. 9 shows the performance results using
the filter is switched on the source current becomes conventional PI controller of source voltage (VS), source,
sinusoidal having THD as per specified by IEEE standard load and filter current (IS, IL and IC) and DC side capacitor
from the stepped wave shape. The THD in the load current voltage (Vdc).
is 27%. THD of source current is reduced from 27 % to 2.36
%, which proves the effectiveness of proposed FLC.

Figure 9. Transient response of PI controlled shunt APF

Fig. 9 depicts that DC-link voltage rise at load change

Figure 7. Switch-on response of system parameters based on FLC instant is much slower using PI controller. Also, settling
time for transient case is also more as compared to FLC
To show the performance during transient condition load controller. It is clear from simulation results that the
is increased to 11 kW to 15 kW at t=0.12 s. Fig. 8 shows the transient performance of supply current and DC-link voltage
various system parameters waveform of the fuzzy logic is good for FLC as compared to PI controller. However,
controlled shunt APF during load change. Fig. 8 shows that THD performance for both FLC and PI are well with-in the
there is smooth change in source current during this limits specified by IEEE 519 standards as shown in Table 3.

86
Authorized licensed use limited to: Malaviya National Institute of Technology Jaipur. Downloaded on May 26,2025 at 06:34:04 UTC from IEEE Xplore. Restrictions apply.
 International Conference on Computer & Communication Technology (ICCCT)-2011

TABLE III. HARMONIC SPECTRUM OF LOAD AND SOURCE CURRENT controlled current active filter,” IEEE Trans. Power Electron., vol.
FOR PI AND FUZZY LOGIC CONTROLLER 8, pp. 140-146, 1993.
[11] B. Singh, K. Al-Hadded and A. Chandra, “Computer aided
Order of Load Source current Source current
modeling and simulation of active power filters,” Electr. Mach.
harmonics current THD with PI THD with Fuzzy
Power Sys., vol. 27, pp. 1227-1241, 1999.
THD controller controller
[12] T. Murata, T. Tsuchiya, and I. Takeda, “Vector control for
1 100 100 100
induction machine on the application of optimal control theory,”
5 21.89 1.1 .84
IEEE Trans. on Ind. Electron., vol. 37, no. 1, pp. 283–290, 1990.
7 10.64 .75 1.25 [13] V.I. Utkin, “Sliding model control design principles and
11 7.63 .64 .85 applications to electric drives,” IEEE Trans. on Ind. Electron., vol.
13 5.27 .50 .58 40, no. 1, pp. 23–36, 1993.
17 3.76 .39 .27 [14] P. Kirawanich and R. M. O. Connell, “Fuzzy logic control of an
19 2.89 .1 .20 active power line conditioner,” IEEE Trans. on Power Electron.,
THD 26.66 % 2.85 % 2.26 % vol. 19, no. 6, pp. 1574-1585, 2004.
[15] B. K. Bose, Modern Power Electronics and AC drives, Pearson
Education Asia, 2002.
VII. CONCLUSION [16] L. A. Zadeh, “Fuzzy stes,” Informat. Contr., vol. 8, pp. 338-353,
1965.
A simple fuzzy logic based three-phase shunt active [17] B. K. Bose, “Expert system, fuzzy logic, and neural network
power filter for current harmonic elimination and reactive applications in power electronics and motion control,” in Proc.
power compensation is presented in this paper. The IEEE PEMC’94, pp. 1303-1323, 1994.
performance of fuzzy logic controlled shunt APF has been [18] R. C. Bansal, “Bibliography on the fuzzy set theory applications in
power systems (1994-2001),” IEEE Trans. on Power Sys., vol. 18,
studied and compared with the conventional PI controller. n. 4, pp. 1291-1299, 2003.
The steady state performance is comparable to the PI [19] R. C. Bansal, T. S. Bhatti and D. P. Kothari, “Artificial
controller whereas transient response is found better than intelligence techniques for monitoring and control of power
the PI controller. The system has fast dynamic response for systems: an overview,” in Proc. Contr. Instrumentation Inform.
varying load condition and harmonic spectrum is found well Commun. Conf., Calcutta, India, 2001, pp. 91–95.
[20] B. N. Singh, B. Singh, A. Chandra, P. Rastgoufard and K. Al-
below 5%, harmonic limit imposed by the IEEE-519 Hadded, “An improved control algorithm for active filters,” IEEE
standard. The compensation process is simple with less Trans. Power Delivery, vol. 22, no. 2, pp. 1009-1020, Apr. 2007.
sensor count and is based on sensing source current only. [21] N. Gupta, S. P. Dubey and S. P. Singh, “Neural network based
Experimentation of the proposed scheme using shunt active filter with direct current control for power quality
TMS320F28335 floating point digital signal processor is in conditioning,” Int. J. Power Electron., in press.
[22] W. Pedrycz, ‘‘Why triangular membership functions,’’ Fuzzy Sets
process and will be reported in near future. Syst., vol. 64, pp. 21---30, 1994.
[23] A.I. Dounis and D.E. Manolakis, ‘‘Design of a fuzzy system for
living space thermal comfort regulation,’’ Appl. Energy, vol. 69,
REFERENCES no. 1, pp. 119-144, 2001.
[24] H. Ying, Fuzzy Control and Modeling: Analytical Foundations and
[1] W. M. Grady, M. J. samotyi and A. H. Noyola, “Survey of active
Applications, IEEE Press, 2000.
power line conditioning methodologies,” IEEE Trans. on Power
[25] S. J. Chaing and J. M. Chang, “Design and implementation of the
Delivery, vol. 5, no. 3, pp. 1536-1542, 1990.
paralleable active power filter,” IEEE general meeting, pp. 406-
[2] H. Akagi, E. H. Watanabe, and M. Aredes, Instantaneous Power 411, 1999.
Theory and Applications to Power Conditioning. Piscataway, NJ: [26] R. D. Patidar and S. P. Singh, “Digital signal processor based
IEEE Press, 2007. shunt active filter controller for customer-generated harmonics
[3] A. Chandra, B. Singh, B. N. Singh and K. Al-Haddad, “An and reactive power compensation,” Elect. Power Comp. and Syst.,
improved control algorithm of shunt active filter for voltage vol. 38, pp. 937-959, 2010.
regulation, harmonics elimination, power-factor correction, and
balancing of nonlinear loads,” IEEE Trans. Power Electron., vol.
15, no. 3, pp. 495-507, May 2000.
[4] E. F. Fuchs and M. A. S. Masoum, Power Quality in Power
System and Electrical Machines, Elsevier Academic Press, 2008.
[5] J. S. Subjek and J. S. Mcquilkin, “Harmonic causes, effects,
measurement and analysis,” IEEE Trans. on Ind. Appl., vol. 26,
no. 6, pp. 1034-1041, 1990.
[6] N. Gupta, S. P. Singh and S. P. Dubey, “Digital signal processor
based single phase power quality compensator under distorted
supply voltage,” Elektrika J. of Elect. Engg., in press.
[7] J. C. Das, “Passive filter-potentialities and limitations,” IEEE
Trans. Ind. Applicant., vol. 40, no. 1, pp. 232-241, Jan/Feb. 2004.
[8] G. K. Singh, A. K. Singh and R. Mitra, “A simple fuzzy logic
based robust active power filter for harmonic minimization under
random load variation,” Elect. Power Sys. Res., vol.77, pp. 1101-
1111, 2007.
[9] S. K. Jain, P. Agarwal and H. O. Gupta, “Design simulation and
experimental investigations on a shunt active power filter for
harmonics and reactive power compensation,” Elect. Power Comp.
and Sys., vol. 31, pp. 671-692, 2003.
[10] R. M. Duke and S. D. Round, “The steady state performance of a

87
Authorized licensed use limited to: Malaviya National Institute of Technology Jaipur. Downloaded on May 26,2025 at 06:34:04 UTC from IEEE Xplore. Restrictions apply.