IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 16, NO.

1, JANUARY 2001 89

A Novel Prototype Design for a Transformer for High

Voltage, High Frequency, High Power Use
John C. Fothergill, Senior Member, IEEE, Philip W. Devine, and Paul W. Lefley

Abstract—A prototype transformer has been designed and

built which is novel in its combination of high voltage (50 kV),
high frequency (20 kHz) and high power (25 kVA) specifications.
The design technique utilized a spreadsheet approach which
facilitated an iterative design procedure. The transformer used a
ferrite core, nylon insulated secondary bobbins and pressurized
sulfur hexafluoride encapsulation. It was designed as part of a
high-voltage switched-mode power supply for driving electrostatic
precipitators. The transformer was field tested at a large coal-fired
power station and was found to have an efficiency of better than
98%.
Index Terms—AC–DC power conversion, dielectric mate-
rials/devices, EHV transformers, ferrite materials/devices, gas
insulated transformers, HF transformers, HVDC insulation,
power transformers.

I. INTRODUCTION Fig. 1. Conventional high voltage DC power supply. TH1, TH2: Thyristor set.
T1: High voltage, low frequency transformer. D1, D2, D3, D4: High voltage

C ONVERSION of AC mains voltages to a high voltage

( kV), high power ( kW) DC level is a key area of
technology which is a requirement in many industrial processes;
bridge rectifier.

for example in particulate emission control using electrostatic

precipitators. continuous rating for electrostatic precipitation. There are
In the majority of cases this high voltage DC requirement is several improvements that may be expected from the adoption
achieved with a single phase AC regulator (Fig. 1). The regulator of high frequency switched mode operation:
controls the supply to the high voltage transformer. A bridge 1) High frequency switching operation will allow much
rectification network is implemented on the secondary of the more precise control over the operating parameters such
transformer to produce a high voltage DC level. as output voltage level, current level, voltage rise times
Power and voltage control is accomplished by variation of the and response to variations in load demand.
firing angle of the thyristor set. There has been little progress 2) High frequency switching will allow a significant
from this topology over the years due, in part, to the lack of avail- reduction in the size and weight of the high voltage
ability of high frequency, high voltage transformers of sufficient transformer. This reduction in size and weight leads to
power ratings. This has made conventional 50/60 Hz design the a compact design, which minimizes the installation and
more attractive solution. This topology whilst robust and simple
maintenance costs.
has severe drawbacks as far as operation is concerned:
3) High frequency switching will allow the reactance of the
• Low quality input currents and low power factor. transformer core to be much lower and hence the effi-
• Sluggish operating characteristics. ciency of the power supply can be improved.
• Low power supply efficiency. 4) The ability to modulate the output voltage. In some
• Large size, weight and civil engineering costs associated applications the ability to pulse the DC output voltage
with the oil-insulated transformer. of the converter from one level to another at a specific
A high frequency switched mode based power supply has and programmable magnitude, time duration and period
been developed with an output voltage of 50 kV at 0.5 A
has substantial benefits, e.g., in electrostatic precipitators
this method may improve dust/gas particle charging and
Manuscript received September 24, 1999. This work was supported by the collection.
UK Engineering and Physical Sciences Research Council (EPSRC) and the De-
partment of Trade and Industry (DTI) through their LINK PEDDS scheme. The The transformer must be designed to be driven from a voltage
project was in collaboration with National Power (NP) and KH Deakin (ES) sourced H-bridge inverter, Fig. 2. There is sufficient flexibility
Ltd. (later Elequip Project Ltd.). with this type of inverter for using a range of voltage control
The authors are with Electrical Power Engineering Research Group, Depart-
ment of Engineering, University of Leicester, Leicester LE1 7RH, UK. strategies e.g. pulse width or phase shift control. Furthermore,
Publisher Item Identifier S 0885-8977(00)10226-2. the load may be made resonant if required. This flexibility in
0885–8977/00$10.00 © 2001 IEEE

Authorized licensed use limited to: IEEE Xplore. Downloaded on November 10, 2008 at 23:49 from IEEE Xplore. Restrictions apply.
90 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 16, NO. 1, JANUARY 2001

Fig. 3. Transformer with two secondary bobbins driving separate diode

bridges.
Fig. 2. Voltage fed H-bridge output stage of the proposed high voltage DC
power supply.

molecules, which may degrade the insulation and lead to

inverter control and operation is important to avoid undesirable electrical breakdown.
affects such as ringing. Such effects may occur because of the The transformer was designed to drive a full-wave rectifier
nature of the transformer, exhibiting parasitic capacitance and and to have a modular design so that different maximum
inductance, which affects the voltage and current waveforms in voltages could be specified. The transformer was therefore
the inverter. designed with two secondary bobbins each driving a separate
A thorough and careful design of a high voltage high diode bridge as shown in Fig. 3. In order to ensure equal voltage
frequency transformer is required to ensure that the electrical sharing, a capacitive divider was also used.
and magnetic loadings are optimized, that the electrostatic Bearing these factors in mind a high voltage, high frequency
and thermal stresses are acceptable for the voltage and power transformer for an electrostatic precipitator power supply was
requirements, and that the parasitic parameters are minimized. designed with the following electrical specification:
The design of a high frequency, high voltage transformer for
• V (quasisquare wave).
electrostatic precipitator power supplies differs widely from the
• kV.
standard transformer design methodologies [1]. Several related
issues must be analyzed: • Power rating kVA.
• Switching frequency kHz.
1) Insulation requirements.
The transformer was field tested as part of a HF HV supply
2) Parasitic elements.
for electrostatic precipitators in a large coal-fired power station.
3) Core loss and heat dissipation.
4) Corona effects.
5) Rectification
High voltage transformers generally have a large turns ratio, II. TRANSFORMER DESIGN
typically 600 : 1 to 900 : 1.
Sufficient insulation thickness between the primary and The transformer design required consideration of the insulation
secondary windings is required to avoid electrical breakdown. materials, the magnetic material and the management of elec-
Therefore, the electromagnetic coupling of the primary and trical, magnetic and thermal stresses.
secondary winding will not be as tight as in conventional low
voltage transformers [2]. This results in a parasitic leakage
inductance referred to the primary side that can effect the A. Magnetic Design
maximum power throughput of the transformer. Hence, there is
a trade off between insulation distance and leakage inductance. Because of the high frequency requirement, it was decided to
Furthermore, the high number of turns that is required for use a ferrite core. In order to accommodate two bobbins (one
the secondary winding causes a high distributed capacitance. for the low voltage and one for the high voltage winding) a
When referred to the primary this capacitance value is multi- rectangular core was used with a 50 mm 50 mm
plied by the square of the turns ratio and therefore is not negli- cross-section. This was made from 100 mm 25 mm 25 mm
gible [3], [4]. This parasitic capacitance induces an ineffective sections, since they were the largest available;
current through the secondary winding which results in a loss of these were ground to give good fits between the sections. To en-
transformer efficiency. sure low lossees the flux density was limited to 0.25 T.
Corona discharge can seriously effect the operation and life A method described by McLyman [5] was used for the
expectancy of a high voltage transformer. Any sharp corner magnetic calculations. This method was implemented using
or protrusions may lead to an enhanced electric field and MathCAD; the output of which is in Appendix A. This imple-
corona in this vicinity. A corona will create highly reactive mentation was very useful as it facilitated an iterative design

Authorized licensed use limited to: IEEE Xplore. Downloaded on November 10, 2008 at 23:49 from IEEE Xplore. Restrictions apply.
FOTHERGILL et al.: A NOVEL PROTOTYPE DESIGN FOR A TRANSFORMER FOR HIGH VOLTAGE, HIGH FREQUENCY, HIGH POWER USE 91

process in which the consequences of a parameter change were

immediately reflected on the computer screen.
The important parameters are
• Two primary windings in series, each containing 13 turns
of AWG7 wire giving a primary copper loss of 4.6 W.
• Two secondary windings in parallel, each containing 700
windings of AWG21 wire.
• Total losses (copper and “iron”) of 510 W.

B. Insulation Materials

In addition to the solid insulation required for the bobbins, the

insulating encapsulation needed to be defined.
Fig. 4. 50 Hz breakdown strength in a homogeneous field for a 20 mm
It is clear that the secondary bobbin needed to be large enough inter-electrode gap as a function of absolute gas pressure.
to prevent surface tracking from the high voltage winding to
the transformer core or the primary winding. The thickness of
the bobbin also had to be great enough to prevent breakdown
through the insulation. Tracking always occurs when the sur- to the transformer once the encapsulation was in place and,
face electric field exceeds that of either the solid or the encap- if the encapsulation suffered electrical breakdown, it may
sulation surrounding it. It may occur at lower fields because of be necessary to completely re-build the transformer. A fluid
physical and chemical inhomogeneities on the surface and be- encapsulation such as transformer oil or an insulating gas was
cause of geometric field enhancements. Although reliable fig- therefore considered. Transformer oil is a good medium for
ures appear to be hard to find, a distance of at least 1 kV/mm is heat transport. Under ideal natural convection conditions it
commonly allowed for well-controlled surface conditions. (For has heat transfer coefficient of approximately 95 W.m K ;
example Rowland and Nichols [6] found voltages of at least this is equivalent to forced air cooling with a flow velocity of
15 kV were required to sustain dry band arcing over a 10 mm approximately 25 m/s [12]. For prototyping oil causes severe
gap on an “arc resistant thermoplastic compound.”) disassembly problems, as the oil needs to be removed from the
The voltage waveform to be experienced by the high voltage surfaces of all components. Impurities in oil tend to accumulate
bobbin was rather unusual in this application since it was of at points of high field divergence, i.e. at the most critical points,
high frequency (20 kHz) and had a negative DC offset of half due to dielectrophoresis [13] leading to localized discharging.
the peak-to-peak AC voltage. The negative DC offset may give It is therefore usual to continuously pump and filter the oil; this
rise to space charge injection [7] and consequent field distor- is inconvenient for a small transformer.
tion within the insulation. The maximum (“Poissonian”) field For these reasons, it was therefore decided to use a gaseous
within the material may therefore be greater than the average encapsulation. Although it may be possible to use air at normal
(“Laplacian”) applied field. The high frequencies may also atmospheric pressure (this is the case in many high voltage labo-
lead to accelerated aging processes due, for example, to partial ratory supplies), this would make the transformer rather large as
discharges. A polymeric insulation was to be used for ease of the breakdown strength of air is only 2–3 kV.mm even under
machining. Because of the large bobbin size, it was decided ideal parallel plate conditions. The air would also need to be
not to use a very expensive material such as PTFE. Nylon 6, clean and dry and therefore maintained in a sealed container.
6 was chosen as a compromise between ease of manufacture, Sulfur hexafluoride, SF , has a much better dielectric strength,
cost and insulation properties. The breakdown strength of such Fig. 4. It can be seen that, even at a pressure of 1 bar, the break-
materials under these conditions is not readily available but down strength of SF exceeds the maximum design strength for
work by Dissado, Montanari, Crine, Lewis and others (e.g. the bobbin (5 kV.mm ). At 2 bar, the breakdown strength is
[8]–[10]) suggests that they should be able to withstand such comparable with nylon 6, 6 and so tracking resistance is unlikely
conditions indefinitely at fields up to approximately 10 kV/mm. to be improved above this pressure. It was therefore decided to
Because of the unknowns, especially to do with space charge use SF at a minimum of 2 bar as the encapsulation material for
effects, it was decided to limit the electric field in the bobbin to the transformer.
5 kV/mm. Nevertheless, this is still high; typically 0.5 kV/mm
AC or 1.0 kV/mm DC are the maximum used in high voltage
supplies for safety critical applications [11]. C. Thermal Analysis
Given these considerations for the high voltage bobbin,
it was necessary to consider the choice of materials for the The McLyman technique was also used for checking the thermal
transformer encapsulation. A solid encapsulation is likely to design of the transformer. The MathCAD output, shown in Ap-
have many drawbacks, especially during the design phase of pendix A, shows such thermal calculations. A maximum tem-
a prototype. It is likely to be inferior in transporting heat from perature rise of C was specified. The method indicates a
the transformer, it would be impossible to make modifications minimum required area of 533 cm , which is considerably less

Authorized licensed use limited to: IEEE Xplore. Downloaded on November 10, 2008 at 23:49 from IEEE Xplore. Restrictions apply.
92 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 16, NO. 1, JANUARY 2001

Fig. 7. Electrostatic electric field analysis around slots in the high voltage
bobbin. The hashed regions corresponding to the windings which are assumed
Fig. 5. Side-view of high-voltage bobbin showing stress-relief rings and slots to be equipotential regions. The highest fields of 5 kV.mm can be seen to
for layered windings. occur on the corners of the slots.

Firstly, in order to obviate the divergent electric stress that

would otherwise occur at the ends of the windings, stress relief
rings are used. These rings, which are split to prevent shorted
turns around the transformer core, are connected to the ends
of the windings; indeed they form the connections. The usual
way to relieve stress is to alter the winding spacing at the ends
but this can be very wasteful of space. The stress relief rings
decrease the divergent field due to the high voltage, ,
by increasing the effective radius of the wire, . In this way the
maximum field, on the inside of the stress relief ring, is limited
to 5 kV.mm .
The other unusual feature is a slotted bobbin that allows
layered windings. In this way it was possible to increase the
number of turns considerably without placing large demands on
the insulation of the winding wire; indeed this was only rated at
600 V. Furthermore the parasitic capacitance between the ends
of the windings is reduced. A single cylindrical winding along
the surface of the bobbin, which only had 300 turns, was found
to have a parasitic capacitance of 130 nF at 1 kHz. However the
arrangement shown here, which had more windings (700 turns),
had a reduced capacitance of 64 nF. Such an arrangement does
give rise to locally high divergent fields at the corners at the
bottom of the slots and an electrostatic analysis was made to
ensure that a field of 5 kV.mm was not exceeded. The results
of this analysis are shown in Fig. 7.

III. TEST RESULTS

Fig. 6. Photograph of the high voltage bobbin. The transformer was tested as part of the high frequency,
high voltage power supply described earlier. It was installed
than the area of approximately 1500 cm that was actually used in National Power’s 2035 MW coal-fired “Didcot A” power
(although only about 50% of this area was exposed). A temper- station at Didcot (near Oxford) UK where it replaced a conven-
ature rise of less than C was therefore to be expected. tional 50 Hz transformer rectifier and supplied an electrostatic
precipitator field. Precipitators are difficult loads since they are
prone to arcs and sparks, during which time they act as virtual
D. Design of High Voltage Bobbin and Electrostatic Analysis short circuits, and the current they draw is very unsteady due
to the corona discharge and continuously changing coal dust
Fig. 5 shows a drawing of the high-voltage bobbin on which burden. The power electronics contained protection circuits
the high-voltage secondary winding was wound, Fig. 6 shows to ensure that the supply voltage was rapidly reduced during
a photograph of the bobbin. The bobbin contains two unusual arcing and sparking. The load was therefore a useful test to
features. ensure that the unit was sufficiently powerful and robust but it

Authorized licensed use limited to: IEEE Xplore. Downloaded on November 10, 2008 at 23:49 from IEEE Xplore. Restrictions apply.
FOTHERGILL et al.: A NOVEL PROTOTYPE DESIGN FOR A TRANSFORMER FOR HIGH VOLTAGE, HIGH FREQUENCY, HIGH POWER USE 93

Fig. 9. Primary voltage (lower trace) and current (upper trace) for 38 kV
output.

Fig. 8. Measured current–voltage characteristic under electrostatic there is a current path through their flywheel protection diodes.
precipitator load. The current during this period is otherwise uncontrolled and
the small current oscillation during the dead time probably
corresponds to a resonance of the magnetizing inductance with
was difficult to obtain accurate results because of the unsteady the parasitic capacitance.
current drawn. It was difficult to estimate the efficiency of the transformer
Fig. 8 shows the measured current-voltage characteristics accurately because of the measurement uncertainty in the
of the electrostatic precipitator load; the line is an exponential continuously fluctuating secondary current and unfortunately
curve drawn to assist the eye. The current increased rapidly the experimental arrangement precluded measurement of the
above a corona inception voltage of approximately 30 kV. instantaneous secondary current. All our estimates lead us to
The circled groups of measurements at approximately 0.2 A believe that the efficiency of the transformer with rectifier was
( kV) and 0.4 A ( kV) were almost certainly made better than 97%. The total rectifier voltage drop was 200 V
as the supply was recovering from a spark in the precipitator. which, at 0.4 A, would give an 80 W power loss and contribute
The voltage is increasing at this point but little corona current % to the overall power lost. The best indication of power
is being produced. By extrapolation of the exponential, it may loss was the temperature rise of the transformer core, which
be estimated that the transient arcing current was several amps was measured using a thermistor. The temperature rose from
at the higher voltages. The dotted exponential curve shows the C to C, a rise of C, when the output was
average “steady-state” characteristics of the precipitator. 45 kV, 0.4 A, a power of 18 kW. The design calculations,
Conventional transformer-rectifier sets have ripples that are Appendix A, suggest that with a loss of 510 W, a temperature
typically 50% of the mean voltage [14]. Arcing occurs if the rise of C would be found for a core surface area of 533 cm .
peak of the voltage ripple exceeds a critical value and so for The actual exposed surface area was approximately 800 cm .
much of the cycle, when the voltage is much less than this crit- An estimate of the actual power lost is therefore
ical value, little corona current is produced. Typically the mean
current is much less than half that produced at the peak. Further-
more the ripple produces a considerable capacitive displacement
C cm
current which increases the size of the secondary transformer W W
windings. In this high frequency supply, very little ripple was C cm
observed. At 30 kV the ripple was V, at 44 kV it was
V; i.e. it was much less than 1% under all conditions. It For an output power of 18 kW this corresponds to an efficiency
is therefore possible for the supply to produce more than twice of 98.5%. The overall efficiency of the power supply was
as much current without arcing occurring. estimated to be % which compares very favorably with
Fig. 9 shows the primary current and voltage waveforms for conventional 50 Hz transformer–rectifier sets with efficiencies
an output of 38 kV. As the primary voltage is provided by a of only %.
phase-controlled H-bridge, there is a dead time between both
positive and negative cycles when the duty cycle is less than
100%. By considering the voltage waveform in Fig. 9 it can IV. CONCLUSION
be seen that the period is approximately 50 s corresponding
to a switching frequency of 20 kHz. The duty cycle was set A highly efficient high-voltage high-frequency transformer
to 40% with the dead times equal to 10 s and the positive has been designed, built and field tested. The transformer
and negative on times being 15 s. The primary voltage was formed part of a switched-mode power supply for electrostatic
V. During the positive (negative) on times the current precipitators. The design, which is still at a prototype stage, was
increased (decreased) reasonably linearly, due to the magne- found to be much more efficient and smaller than conventional
tizing inductance, to a value of approximately A. 50/60 Hz transformer rectifier sets. The unit will now be
During the dead times all four H-bridge transistors are off but developed for manufacture.

Authorized licensed use limited to: IEEE Xplore. Downloaded on November 10, 2008 at 23:49 from IEEE Xplore. Restrictions apply.
94 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 16, NO. 1, JANUARY 2001

APPENDIX
MAGNETIC AND THERMAL DESIGN USING MCLYMAN DESIGN METHOD

Authorized licensed use limited to: IEEE Xplore. Downloaded on November 10, 2008 at 23:49 from IEEE Xplore. Restrictions apply.
FOTHERGILL et al.: A NOVEL PROTOTYPE DESIGN FOR A TRANSFORMER FOR HIGH VOLTAGE, HIGH FREQUENCY, HIGH POWER USE 95

Authorized licensed use limited to: IEEE Xplore. Downloaded on November 10, 2008 at 23:49 from IEEE Xplore. Restrictions apply.
96 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 16, NO. 1, JANUARY 2001

Authorized licensed use limited to: IEEE Xplore. Downloaded on November 10, 2008 at 23:49 from IEEE Xplore. Restrictions apply.
FOTHERGILL et al.: A NOVEL PROTOTYPE DESIGN FOR A TRANSFORMER FOR HIGH VOLTAGE, HIGH FREQUENCY, HIGH POWER USE 97

ACKNOWLEDGMENT REFERENCES
[1] P. J. Devine, “The requirements of a high voltage DC power supply,”
Leicester University Engineering Department, 1995.
The authors would like to thank R.I. Deakin, M. Selby (NP), [2] G. Liberati, “High voltage, high power switched mode power supply for
ion production,” ENEL, CRTN.
D.A. Rose (DTI), K.R. Parker and D.F. Warne for their consid- [3] M. A. Perez, “A new topology for high voltage, high frequency trans-
erable technical assistance and support throughout the project. formers,” IEEE, 0-7803-2482-X/95, 1995.
They would also like to thank A. Smith and A. Joseph for their [4] H. Takano, “Feasible characteristic evaluations of resonant inverter
linked DC-Dc power converter using high voltage transformer parasitic
cooperation and assistance with the field trial at Didcot A power components,” in Power Electronics and Variable Speed Drives: IEE,
station. Sept. 23–25, 1996, pp. 525–533.

Authorized licensed use limited to: IEEE Xplore. Downloaded on November 10, 2008 at 23:49 from IEEE Xplore. Restrictions apply.
98 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 16, NO. 1, JANUARY 2001

[5] W. T. McLyman, Transformer and Inductor Design Handbook. New Philip Devine was born in 1971. He graduated from
York, NY: Dekker, 1978. the University of Leicester in 1994 and is currently
[6] S. M. Rowland and I. V. Nichols, “Effects of dry-band arc current on studying for a Ph.D. by part-time study. Prior to his
ageing of self-supporting dielectric cables in high fields,” IEE Proc., degree he worked for Heenan Drives Ltd. He was em-
Science, Measurement and Technology, vol. 143, no. 1, pp. 10–14, Jan. ployed as a Research Assistance on the project de-
1996. scribed in this paper and is now a Teaching Company
[7] J. C. Fothergill and L. A. Dissado, Eds., Space Charge in Solid Di- Associate. His main area of interest is in high voltage
electrics: The Dielectrics Society, 1998, pp. 259–272. power electronics.
[8] L. A. Dissado, G. Mazzanti, and G. C. Montanari, “The role of trapped
space charges in the electrical aging of insulating materials,” IEEE
Trans. DEIS, vol. 4, no. 5, pp. 496–506, Oct. 1997.
[9] J.-P. Crine, “A molecular model to evaluate the impact of aging on space
charges in polymer dielectrics,” IEEE Trans. DEIS, vol. 4, no. 5, pp.
487–495, Oct. 1997.
[10] P. Connor, J. P. Jones, J. P. Llewellyn, and T. J. Lewis, “Mechanical
origin for electrical ageing and breakdown in polymeric insulation,” in Paul Lefley was born in Ipswich, UK in 1964.
Proceedings of the 1998 IEEE 6th International Conference on Con- He graduated from the University of Nottingham
duction and Breakdown in Solid Dielectrics, Conf. code 49 244, pp. in 1985, and was awarded a Ph.D. from the same
434–438. institution in 1989. He worked for Mobicom Ltd.
[11] J. W. Williams, “Navy/industry ‘NAVMAT’ high voltage guide- from 1988–89. In 1989 he returned to Nottingham
lines—An overview,” in 1995 High Voltage Workshop, USA: University where he was awarded a three year
University of Utah. Post-Doctoral Fellowship. In 1992, he moved to
[12] Conti-Elektro-Berichte, July/Sept. 1966, p. 189. the University of Leicester where he took up the
[13] H. A. Pohl, Dielectrophoresis. Cambridge: Cambridge University post of lecturer. His main area of interest is power
Press, 1978. electronics and machines. Whilst at Leicester
[14] K. R. Parker, Ed., Applied Electrostatic Precipitation: Chapman and he has been chairman and treasurer of the IEE
Hall, 1997. East Midlands Power Division, and chairman of the 34th University Power
Engineering Conference.

John Fothergill was born in Malta, in 1953. He grad-

uated from the University College of North Wales,
Bangor where he also completed an M.Sc. and then
a Ph.D. in 1979. Following this he worked at STL
(Harlow), the research laboratories for the Standard
Telecommunication Company, on power cable insu-
lation. He is now a Senior Lecturer in the Department
of Engineering at the University of Leicester, UK. His
main area of interest is in high voltage dielectrics on
which he has co-authored a book. He is a Fellow of
the IEE and a Senior Member of the IEEE. He presented the invited opening
lecture at the High Voltage Workshop in Daytona Beach in 1998.

Authorized licensed use limited to: IEEE Xplore. Downloaded on November 10, 2008 at 23:49 from IEEE Xplore. Restrictions apply.