IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS 1

A Quasi-Constant On-Time Control for SMPS

With a Nonlinear Inductor Based on Power
Switch Conduction Time Estimation
Daniele Scirè , Member, IEEE, Giuseppe Lullo , Member, IEEE,
and Gianpaolo Vitale , Senior Member, IEEE

Abstract—This article proposes a control method for Ldeepsat Inductance value in deep saturation.
switching-mode power supplies equipped with nonlinear LSR Least square regression.
inductors; the method is based on the estimation of the Tcore Temperature of the inductor.
conduction time (TON ) of the power switch. The TON value
is tuned according to the inductor model to obtain quasi- TON Conduction time of the power switch.
constant ON-time control. The method is devised for ferrite TOFF Blocking time of the power switch.
core power inductor, the nonlinearity is exploited up to Ts Switching period.
saturation, condition where the differential inductance is β0 , β1 , β2 , . . . Thermal coefficients of the inductor model.
reduced to half of its maximum value. This approach allows  Convergence absolute tolerance.
the calculation of the conduction time TON such that the
maximum current of the inductor always corresponds to Φ Magnetic flux of the inductor.
the saturation value while also considering the inductor ΔiL Peak-to-peak current through the inductor in
temperature. This method obtains the current peak versus linear operation.
TON through a proper inductor model and a recursive al- ΔiL−nl Peak-to-peak current through the inductor in
gorithm for a given inductor temperature. The theoretical nonlinear operation.
analysis was verified on a boost converter by comparing
the simulations with experimental data with different loads, vL Voltage across the inductor terminals.
showing that the operating current can be increased by Vs Supply voltage.
approximately 40% avoiding thermal runaway.
Index Terms—Algorithm, inductors, nonlinear circuits, I. INTRODUCTION
nonlinear estimation, saturable core, switched mode power
PTIMIZING the high power density in switching-mode
supply (SMPS).
O power supplies (SMPS) is essential. Researchers have
endeavored to address several aspects, such as improved topolo-
NOMENCLATURE gies [1], new power-switching devices [2], electromagnetic
a0 , a1 , a2 , . . . Coefficients of the polynomial form of the interference filters [3], [4], advanced inductor modeling, and
inductor current. design techniques [5]. In addition, some papers dealing with
D Duty cycle. new control techniques highlight the need for an accurate model
Fs Switching frequency. of the converter components [6], [7], [8].
iL Current flowing through the inductor. The power density of an SMPS can be improved by reducing
iL DC current flowing through the inductor. the core of the inductor [4] and operating it beyond its linear
iL,max Current peak of the inductor. region, up to saturation [9], [10], [11], [12]. The saturation
L Inductance value. zone is defined as the current at which differential inductance
L0 , L1 , L2 , . . . Coefficients of the inductance model. Lsat is reduced to half of its rated value. Beyond the linear
Lsat Inductance value at saturation. region, the inductance value decreases and depends on both
current and temperature. Here, the inductor exhibits nonlinear
behavior; therefore, it is referred to as a nonlinear inductor. By
Received 7 November 2022; revised 10 August 2024; accepted 2 exploiting saturation, a higher dc operating point can be set;
October 2024. Paper no. TII-22-4601. (Corresponding author: Daniele
Scirè.) however, reducing the inductance causes a higher current peak.
Daniele Scirè and Giuseppe Lullo are with the Department of Engi- Consequently, the power switch is more stressed and the losses
neering, University of Palermo, 90128 Palermo, Italy (e-mail: daniele. increase [13].
[email protected]; [email protected]).
Gianpaolo Vitale is with the Institute for high performance comput- This study focuses on the estimation of the current shape
ing and networking, National Research Council (ICAR-CNR), 90146 during the conduction time of the power switch (TON ) because
Palermo, Italy (e-mail: [email protected]). in a basic SMPS, it corresponds to the time interval in which the
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/TII.2024.3477562. current through the inductor increases. Knowledge of this time
Digital Object Identifier 10.1109/TII.2024.3477562 interval is crucial because in a nonlinear inductor, the current

© 2024 The Authors. This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see
https://creativecommons.org/licenses/by/4.0/
2 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS

increases faster when approaching the end of the TON . For a fixed a shape similar to the inductance variation to be reproduced, such
voltage across the inductor, the current derivative is inversely as the arctangent [9] and polynomial models [17].
proportional to the inductance, which also depends on inductor Considerable attention has been paid to ferrite as a core
temperature. Overestimation of the TON produces an overcurrent material owing to its low losses over a wide frequency range and
that can damage the power switch. Conversely, for a short TON , high specific resistivity [18]. However, ferrite exhibits an abrupt
the operation occurs in the linear zone, nullifying the advantage decrease in the inductance when saturation is achieved [10],
of saturation exploitation. [19], [20].
Some studies dealing with constant ON-time control Several analytical models have been proposed that show dif-
(COT) [14], [15] confirm the interest in this control technique; ferent tradeoffs between the accuracy and calculation resources.
however, they can consider TON as a constant because the adopted The improved arctangent-based method [12] provides good
inductor is linear. Calculating the TON for the conventional opera- accuracy for the entire inductor curve, including the temperature
tion of an SMPS with a linear inductor is straightforward because variations. Such a model can be implemented in a simulation to
it does not depend neither on the temperature of the inductor nor reliably reproduce the current waveform in a dc–dc converter
on its current. In contrast, when the inductor operates outside the at the operating temperature [21]. The piecewise-affine formu-
linear zone, the behavior is nonlinear and the temperature must lation [22] improves the use of the computational resources
also be considered, emphasizing a dependence on the current. of Oliveri et al. [12]. Hammerstein modeling can also provide
The novelty of this study is that an algorithm allows the tuning nonlinear functions in power electronics, as in [23], where it is
of the TON value based on temperature to obtain quasi-constant applied to characterize different ferrite inductors using a pulse
on-time control (QCOT). compression identification technique. In [24], a neural network
In principle, the temperature can be retrieved using a suitable model (E-αNet) was developed to reproduce the ferrite power
loss model [16]; however, knowledge of environmental temper- inductors inductor behavior up to saturation also considering
ature is required. The core temperature was measured using a the core temperature. The so-called local approach generally
thermocouple. In addition, the temperature distribution on the adopts different approximations depending on the current. The
inductor is assessed and confirmed using thermal images. main disadvantage is the discontinuity of the inductance curve,
The design of a QCOT requires suitable characterization and which can only be used for small current variations [11]. The
a proper model of the power inductor to describe the satura- solution proposed in this article takes advantage of a polynomial
tion [16]. The proposed recursive algorithm allows the calcula- model that approximates the inductance variation, including
tion of the current peak based on TON , and vice versa. Starting temperature dependence [25]. The inductor temperature can be
from the maximum current imposed on the inductor and tuning considered constant during the switching period. Therefore, a
the TON , it will always operate in saturation, avoiding thermal model of the inductor at a given temperature is suitable for re-
runaway regardless of the temperature. Although a practical producing the current waveform when the temperature is known.
application is discussed, the proposed method is general; it can However, during operation, the inductor modifies its behavior
be used to analyze the current through a nonlinear inductor when owing to a thermal transient, whose duration is of the order of
a constant voltage is applied for a given time interval. tens of seconds; consequently, the inductance variation versus
The rest of this article is organized as follows. First, a short temperature has been considered. This implies that different core
review of the models adopted for inductors in the literature is temperatures require different TON values to exploit saturation,
presented in Section II. Section III details the model adopted whereas all the other parameters remain unchanged. The devised
in this study. Section IV explains the characterization of the QCOT is useful since it varies the switching frequency, main-
inductor used for experimental verification. Section V describes taining the duty cycle imposed by the converter control, always
the main theoretical contribution, that is, the algorithm for the assuring a proper TON value. The goodness of the model consists
current estimation. Section VI proposes validation using a boost of reproducing the current profile under operating conditions,
converter with two different loads, and the operations of the including the temperature increase due to losses [21], [26] and
converter are based on the TON estimation according to core dangerous operations, potentially causing thermal runaway [16],
temperature. Section VII presents a discussion on the quality [27], [28], [29], [30].
and the limitation of the proposed approach, paving the way for
future developments. Finally, Section VIII concludes this article.
III. INDUCTOR MODELING BY POLYNOMIAL
The proposed approach considers an inductor modeled using
II. REVIEW OF THE INDUCTOR MODELING
a polynomial function. This model was successfully applied to
The literature considers both the behavior of the inductor in the behavior of saturable reactors in [25]; here, it was adapted
saturation and the consequences of the SMPSs operation. to ferrite power inductors, including temperature [16], [31]. It
Analytical models, rather than physical models, are preferred reproduced the sharp transition from linear to saturation, and
for power electronics applications. Analytical models provide proved to be computationally lightweight, providing a good ap-
an external representation of the inductor using a relationship proximation with a third-order polynomial [16], [17]. In contrast
that gives the inductance versus current (with temperature as a to saturable reactors operated in ac, the power inductors used in
parameter). Many approaches use general and well-known mod- SMPSs operate with a dc offset and superimposed ac signal;
eling techniques; two of them are based on a function that shows therefore, a different characterization system is required [32].
SCIRÈ et al.: QUASI-CONSTANT ON-TIME CONTROL FOR SMPS WITH A NONLINEAR INDUCTOR 3

TABLE I
MODEL COEFFICIENT OF THE INDUCTOR

The polynomial coefficients are considered to be linearly de-

pendent on the temperature, implying that the magnetic flux
increases linearly with temperature for a given magnetizing
current. The characterization of the work in [33] in terms of the
Fig. 1. Inductance versus current versus core temperature of the Coil-
saturation flux shows that a linear trend can approximate it for craft 5010H-334.
the temperature range 25–125 ◦ C. Moreover, we experimentally
verified this behavior in the studied inductor.
The inductance is modeled as

3
m=0 Lm (1+βm Tcore )iL , L(iL ) > Ldeepsat
m
L(iL , Tcore ) =
Ldeepsat , otherwise
(1)
where Lm are polynomial coefficients, and the linear depen-
dence of the inductor temperature Tcore (◦ C) is expressed by
the proportionality factors βm . The parameter Ldeepsat is the
inductance value when deep saturation (also known as hard
Fig. 2. Electrical scheme of a boost converter.
saturation) is reached, where the flux density is almost constant.
The first term in (1) corresponds to the thermal behavior of a
traditional linear inductor. Once the coefficients are known, (1)
can be easily implemented and requires minimal computational varies with temperature, resulting in different current curves. It
resources. However, the coefficients Lm , βm , and Ldeepsat need should be remarked that, from the characteristic curves shown in
to be previously identified; they were obtained experimentally Fig. 1, the use of the Coilcraft 5010H-334 inductor in linear zone
using the method explained in Section IV. Once the model is can be performed with a current up to 1 A. Instead, by nonlinear
identified, the practical limit or saturation point (Lsat ), where operation, the operating point can be increased by 40%, meaning
the differential inductance is halved, can be retrieved. a current of 1.4 A with a temperature core of 85 ◦ C. At this point,
the rated inductance is reduced to approximately 200 μH. The
IV. INDUCTOR CHARACTERIZATION same current with linear operation would require a larger core
inductor, such as 5040H-330, which exhibits a larger volume
The main issues related to inductor model identification for of approximately 70%. This demonstrates that a design with a
linear inductors are discussed in [34]. However, as explained nonlinear inductor improves power density. An increase of 40%
above, a nonlinear inductor requires a different characterization in the rated current can be assumed as a safe tradeoff between
method. This was performed using the test system described limiting the nonlinear operations and achieving a high-density
in [32] and [35] to obtain a representation of the inductor as SMPS design and can be considered as a practical target by the
in (1). designer.
The operating principle of the characterization system can be
summarized as follows. The inductor under test was inserted into
a dc/dc converter with a variable active load. It is automatically V. ESTIMATION OF THE MAXIMUM CONDUCTION TIME AND
CURRENT PEAK
controlled by a LabVIEW-based instrument that imposes 1) the
switching frequency, 2) the duty cycle of the power switch, 3) the After the inductor characterization, there is need to calculate
dc bias current by varying the load value, and 4) the inductance its current when it is operated in a converter. The inductor current
L, which is calculated by the ratio between the voltage applied to estimation is performed considering a boost converter as a case
the inductor (maintained constant) and the slope of the current. study; the analysis can be extended to other SMPS without losing
The measurement system was applied to a 330 μH inductor generality because the inductor is supplied by a constant voltage
delivered by a Coilcraft model DO-5010H-334. The identified for a given time interval as in basic dc/dc converters and in many
coefficients corresponding to (1) are given in Table I, and their other topologies. The time interval depends on the switching
characteristic curves are shown in Fig. 1, considering the temper- frequency and the duty cycle. The voltage and time intervals
ature. For low currents, the inductance is constant, and the tem- were chosen based on the converter.
perature does not affect this value, which corresponds to a linear In a boost converter (see Fig. 2) with a linear inductor operated
inductor operation. By increasing the current, the inductance in a continuous current mode, the current flowing through the
4 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS

TON , with Δt = TON /n. The starting value is iL (0), that is, the
minimum value of the current iL,min .
Initially, (4) is solved, starting with iL,min = 0 for a time
interval equal to TON , thus generating a vector containing n
samples of the current. Next, the current samples were fitted
with a polynomial function of time using the coefficients aj
iL (t) = a0 + a1 · t + a2 · t2 + · · · + a5 · t5 . (5)
In this study, a fifth-order polynomial for (5) was used to
achieve a reasonable accuracy. To retrieve an interpolating
polynomial curve, an LSR is applied to the discrete waveform
obtained by (4) such that the coefficients aj of (5) can be found.
The LSR can be applied because the error in the measured time
is much lower than the error in the current measurement, which
follows the Gauss theorem because it is retrieved experimentally,
Fig. 3. Comparison of current waveforms with linear and nonlinear
inductors.
and all events are independent (no memory effect occurs) with
the same variance.
To perform LSR, (5) is written in a matrix form as follows:
matrix A has as many columns as the number of coefficients
inductor exhibits a triangular shape, and the current variation to be identified (corresponding to the degree of the polynomial
can be easily calculated. Instead, when the inductor is exploited plus one) and a number of rows equal to n = TON /Δt. Each
outside the linear region, the current differs from the well-known row contains the time raised to the power corresponding to the
triangular waveform and its time derivative increases with time. coefficient of the current
Fig. 3 compares the current waveforms for linear and saturated ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
inductors with the same average current. The mean value iL 1 0 ... 0 a0 iL (0)
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
of the current does not vary because it depends on the load ⎢1 Δt . . . Δtm ⎥ ⎢ a1 ⎥ ⎢ iL (Δt) ⎥
⎢
A = ⎢. ⎥ ⎢ ⎥ ⎢
.. ⎥ p = ⎢ .. ⎥ y = ⎢ .. ⎥
(neglecting losses in first approximation); consequently, the two .. .. ⎥.
areas defined by the nonlinear curve under and over the mean ⎣ .. . . . ⎦ ⎣ . ⎦ ⎣ . ⎦
value iL must be the same. Instead, because of the upward 1 TON . . . TON m
am iL (TON )
concavity of the current waveform for the nonlinear inductor, (6)
the current peak iL−nl,max increases, and the current variation in Due to the error distribution, a solution as y T = p · A + e
the nonlinear case ΔiL−nl is higher than the value in the linear can be obtained, where e denotes an error vector. Under the
case ΔiL . hypothesis of a normal distribution of the error and null mean, the
The following section presents an analysis of the nonlinear Gauss Markov theorem gives the solution vector of coefficients
inductor. If the inductance is not constant, the general relation- as
ship between inductor voltage vL and flux Φ must be considered −1
p = AT A AT y. (7)

δΦ(iL , Tcore ) The coefficients aj must satisfy the following condition:

vL (t) = (2)
δt 1 TON
iL = iL (t)dt. (8)
where Tcore is the core temperature, which is practically constant TON 0
during the switching period.
The value iL is the dc current through the inductor. This is
Equation (2) can be expanded as
imposed by the load. By substituting (5) into (8), the mean value
dΦ(iL ) diL (t) diL (t) of the current is given by
vL (t) = · = L(iL ) (3)
diL dt dt 1 1 1
iL = a0 + a1 · TON + a2 · TON 2
+ · · · + a5 · TON
5
. (9)
where inductance is considered a polynomial function of the 2 3 6
current, as in our case, using (1). Equation (3) can be solved From (5), it can be noted that coefficient a0 represents the
numerically in a time interval equal to the TON , where vL is a minimum value of the current iL,min obtained for t = 0.
constant Once interpolation was performed, it was possible to invert
vL Δt (9) to calculate a new value of a0 . By substituting a0 in (4) as
iL (k + 1) = iL (k) + . (4) a new starting point iL,min , the procedure can be repeated until
L(iL (k))
convergence is obtained, that is, when the difference between
Discretization (4) follows the Euler method to simplify the de- the new starting point and the previous starting point is less than
scription. However, the fourth-order Runge–Kutta method was the tolerance . In the discontinuous current mode operation of
used in our simulations to achieve a better accuracy. Equation the SMPS, iterations are unnecessary because the value of iL,min
(4) requires n steps corresponding to n + 1 points from t = 0 to is always null.
SCIRÈ et al.: QUASI-CONSTANT ON-TIME CONTROL FOR SMPS WITH A NONLINEAR INDUCTOR 5

Algorithm 1: Estimation Algorithm for Calculating iL,max

and iL,min .
1: iL,min ← 0 Initialize with a zero-initial current
2: iL (0) ← iL,min Set the initial current i(0)
3: repeat
4: for k ← {0, . . . , n − 1} n steps are required for
calculation of (4)
5: L = L0 (1+β0 Tcore )+L1 (1+β1 Tcore )i(k)+. . .
Inductance calculation (1)
6: if L < Ldeepsat Inductance saturation evaluation Fig. 4. Proposed block diagram of the control system.
7: L ← Ldeepsat Inductance set to the saturation
value (1) in order to leave the possibility to scale-up the power rating in
8: end if future works. The operation of the control system is illustrated
9: iL (k + 1) ← iL (k) + vL Δt/L Step for in the block diagram, as shown in Fig. 4. The QCOT module
calculation of (4) is drawn by gray lines; it represents an add-on compatible
10: end for with the existing control. The prototype is shown in Fig. 5(a).
TON
11: iL = TON 1
0 iL (t)dt Mean current evaluation The inductor under test was a ferrite core inductor (Coilcraft
(8) DO-5010H-334, LNOMINAL 330 μH). This component is charac-
12: LSR algorithm Algorithm for the aj evaluation terized as explained in Section IV. A UT70B digital multimeter
13: iL (0) ← a0 Set the initial current iL (0) from Lafayette equipped with a K-type thermocouple (range
14: until |iL,min − iL,min | <  Repeat the process until of −40÷260 ◦ C; accuracy of ±0.75%) is used to measure the
tolerance  is reached core temperature of the inductor. The temperature distribution
15: iL,max = a0 + a1 · TON + a2 · TON 2
+ ... of the inductor [see Fig. 5(b)] was measured through thermal
Maximum current evaluation (10) imaging using a camera model FLIR SC660; its temperature
measurement capability ranges from −40 to 1500 ◦ C with 1
◦
C accuracy. For better visualization in each thermal image, the
After convergence, the curve that reproduces the current is color palette was started at room temperature and ended at the
identified using polynomial (5). The maximum current value highest temperature of each measurement. The thermal analysis
can be obtained by (5) and evaluated for t = TON verifies that during operation, 1) the temperature of the core is not
influenced by the heatsink of the MOSFET, and 2) the temperature
iL,max = a0 + a1 · TON + a2 · TON
2
+ · · · + a5 · TON
5
. (10)
measurements made with a thermocouple in contact with the coil
The calculation method explained here is concisely schema- core are coherent with thermal imaging.
tized in Algorithm 1. It provides the peak current (iL,max ) through The classical pulsewidth modulation (PWM) control loop is
an inductor for a given TON . updated with information from the LUT, providing the TON based
Therefore, the evaluation of the maximum current was re- on both the maximum current peak imposed on the inductor
peated by applying Algorithm 1 to the different core temper- and the core temperature. The switching frequency in the PWM
atures. Each iteration yields a triplet of iL,max , TON , and Tcore . modulator was calculated from the TON and required duty cycle.
Considering that this study aims to derive the maximum allowed The entire test rig is shown in Fig. 5(c). The photograph and the
conduction time for a fixed maximum current, the dataset of thermal images have been performed removing the converter box
the collected triplets can be arranged in a lookup table (LUT). only for the time needed to take the picture. In the following,
Complementary, the dataset can be used to obtain the regression three situations are described.
f and the relationship between the maximum conduction time, 1) Operation of the converter in an open loop with a con-
peak current, and core temperature as in (11). In the practical im- stant TON , to show how the current peak increases as the
plementation, it allows retrieving the TON based on the measured temperature of the inductor increases.
temperature and the maximum current imposed to the inductor 2) Starting from the same operating point of the last test,
an operation with the proposed QCOT, showing that
TON,max = f (iL,max , Tcore ) . (11) the current peak remains constant owing to the control
system, which avoids thermal runaway.
3) An example of the thermal runaway due to the excessive
VI. VALIDATION
mean inductor’s current.
A. Boost Converter and Measurement Setup Description
The boost converter used for the validation test was supplied B. Operation With Constant TON
with Vs = 24 V. It adopts an FDP12N60NZ MOSFET as the The main waveforms of the boost converter operated with a
switch, and an STTH806 rectifier as the diode. The component constant TON and switching frequency of 46 kHz are shown in
selection was driven by the need for a reliable and available Fig. 6(a) for different core temperatures. This shows the evolu-
prototype setup with a significant safety margin, but also flexible tion of temperature during operation in an open loop. Fig. 6(b)
6 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS

Fig. 5. (a) Prototype of the boost converter used for test. The free-wheel diode is mounted on the backside of the PCB. (b) Thermal image of the
converter in standby. (c) Photograph of the test rig.

Fig. 6. (a) Example of an open-loop thermal runaway: duty cycle (top), output voltage (middle), and current through the inductor (bottom) of the
boost converter increasing the temperature at a constant frequency of 46 kHz corresponding to a fixed TON of 10.9 μs. The mean current is 1.4A.
(b) Thermal image of the converter when the inductor reaches the temperature of 105 ◦ C.

shows thermal images obtained at the highest temperature. The peak further increases, increasing the losses and, therefore, the
converter starts at environmental temperature (25 ◦ C), and dur- temperature, thus generating the thermal runaway.
ing operation, it reaches 105 ◦ C. At Tcore = 60 ◦ C, the inductor
is still in linear operation; after 150 s, its temperature is equal
to 80 ◦ C, and an increase in the current peak can be observed. C. Operation With QCOT Technique
This increase is further evident after 813 s (Tcore = 105 ◦ C). By applying Algorithm 1, triplets (iL,max , TON , and Tcore )
Photographs and thermal images were obtained by removing the were retrieved for a specific mean current of the inductor
converter box only for the time required to capture the images. corresponding to a given load value and output voltage. Two
It should be noted that an increase in the temperature modifies case studies were presented by choosing an output voltage of
the current waveform. The power inductor operates in the lin- 48 V and two different mean inductor current values (1.4 and
ear zone only at low temperatures and exhibits a well-known 1.9 A) obtained with two loads of 63 and 50 Ω, respectively.
triangular shape. Subsequently, the temperature increase drives Fig. 7 depicts a 3-D representation of the triplets (iL,max , TON ,
it toward saturation, with a consequent increase in the current and Tcore ) obtained by running Algorithm 1 when the load is
peak. It demonstrates that a constant TON operation cannot ensure 63 Ω. It can be noted that for high temperature and conduction
safe operation in saturation because the current peak depends time, the maximum achievable current highly exceeds the rating
on the temperature. It assesses the need to estimate TON versus current of the inductor, limiting, therefore, the operating region
temperature. Finally, it is worth noting that the traditional voltage of the converter. However, reducing either the temperature or
closed-loop feedback worsens the thermal runaway, which is the conduction time reduces the current peak to a safer value.
inherently unstable. This increases the duty cycle to maintain a Fig. 8 presents the current peak as a function of conduction
constant output voltage against losses. Consequently, the current time, which was parameterized to the core temperature. It is
SCIRÈ et al.: QUASI-CONSTANT ON-TIME CONTROL FOR SMPS WITH A NONLINEAR INDUCTOR 7

Fig. 7. Surfaces representing maximum current peak versus core tem- Fig. 9. Surfaces representing maximum current peak versus core tem-
perature versus conduction time obtained for the 63 Ω load. perature versus conduction time obtained for the 50 Ω load.

Fig. 8. Relationship between peak current and the conduction time for Fig. 10. Relationship between peak current and the conduction time
different core temperatures obtained for the 63 Ω load. for different core temperatures obtained for the 50 Ω load.

worth noting that the current peak presents a linear trend as a correctly exploited, and the temperature stabilizes at about 90
◦
function of the conduction time (see Fig. 9), except when the C [see Fig. 11(b)]. The maximum current peak was set to 1.8
core temperature and high conduction time bring the inductor A and the current waveform was not linear for temperatures
to deep saturation. In the latter case, the current slope versus higher than 60 ◦ C. The control system regulates the output
conduction time increased, rapidly increasing the current peak. voltage (operational safety is assured by TON control) by varying
Similar results are shown in Fig. 10 for the case study with a the switching frequency in the range of 46–74 kHz. For the
load of 50 Ω, where the converter has a smaller region in which chosen iL,max , an appropriate conduction time is set based on
the current peak is below the maximum rated inductor current. It the measured inductor temperature (see Fig. 8). Similar results
is worth noting that for the highest and lowest temperatures, the were obtained in another case study (see Fig. 12) with an output
maximum current exhibited a linear trend as a function of the load of 50 Ω, corresponding to an inductor mean current of 1.9
conduction time. Instead, for intermediate temperatures (e.g., A. In this case, the current peak was set to a higher value (2.6
45 and 55 ◦ C), the slope increased after exceeding a specific A), overcoming saturation, and the conduction time regulation
conduction time. was obtained by varying the switching frequency in a broader
The two case studies were first simulated to address an ade- range of 40–105 kHz. Although the current peak was maintained
quate load value, and then implemented on the boost converter constant, a temperature runaway was experienced, and the mea-
previously described to validate the proposed method. The surements were stopped to prevent any damage to the converter
simulation resulted in LUTs being used to set the conduction after the safe temperature threshold was reached at 105 ◦ C. From
time. The first case is presented in Fig. 11(a) (load of 63 Ω the tests reported above, it can be deduced that, by using an
corresponding to a mean inductor current of 1.4 A with a 40% inductor up to saturation to reduce its core size in an SMPS, the
increase to the rated value), where the inductor saturation is TON time of the power switch must be tuned during operation to
8 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS

Fig. 11. (a) Example of stable operation exploiting nonlinearity: duty cycle (top), output voltage (middle), and current through the inductor (bottom)
of the boost converter with variable TON and frequency as the temperature rises. It can be noted that the peak current remains constant despite
the temperature variation because the ON time is reduced by the control system. The temperature stabilizes at about 90 ◦ C after about 7 min from
startup. (b) Thermal image of the converter when the inductor reaches the temperature of 90 ◦ C.

Fig. 12. Example of thermal runaway due to the excessive mean current: duty cycle (top), output voltage (middle), and current through the inductor
(bottom) of the boost converter with variable TON and frequency as the temperature rises. It can be noted that the peak current remains constant
despite the temperature variation because the ON time is reduced by the control system. The temperature is not stable and exceed 105 ◦ C, therefore,
the measurement was stopped to prevent any damage.

ensure that the current has a constant peak value. Therefore, the frequency ranging from 40 to 105 kHz. This converter has been
constant TON technique was modified into a QCOT technique. initially conceived to test and characterize nonlinear inductors
This implies a frequency variation whose range depends on the based on LabVIEW. Then, we devised the QCOT control as an
load value. The higher the load current, the wider the frequency add-on requiring low computational resources (represented by
variation interval. The QCOT ensures stable operation with gray color in Fig. 4). In this way, the dynamics of the system
an increase of up to 40% in the inductor mean current with is always governed by the PI controller that imposes the duty
respect to the maximum allowed by the manufacturer. The power cycle whereas the QCOT manages the switching frequency for
switch is commutated at the end of the TON , without the need given temperature. The QCOT module introduces a negligible
for a dedicated circuit for current-threshold detection. Even if delay since it requires reading the LUT and a division. The
the proposed approach requires temperature measurement, the adopted setup has the following limits: it uses a relatively
related circuit is simpler than the current detection and does not high 330 μH inductance, the maximum switching frequency
require fast intervention because the thermal evolution of the is limited by the MOSFET conduction resistance and by the
inductor is slow and can take tens of seconds. Furthermore, the LabView implementation. On the other hand, this experimental
temperature measurement prevents a thermal runaway. Finally, setup assesses using a smaller and lighter SMPS through the
it can be observed that the proposed QCOT allows a higher mean exploitation of inductors up to the saturation point. A modern
current through the inductor (up to 40%), which can be chosen implementation would benefit of higher switching frequencies
with a smaller core compared to a linear operation. In our case, allowing for smaller inductances and of the use of a low-cost
the volume of the inductor was reduced by 70% compared with microcontroller for the QCOT algorithm implementation, e.g.,
that of the 5040H-330 inductor that operates in the linear zone. the TMS320F28379D from Texas Instrument (clock frequency
up to 200 MHz), allowing for the implementation of the QCOT
at much higher switching frequency with the use of smaller
VII. DISCUSSION AND FUTURE OUTLOOK inductors.
The aim of this article is to demonstrate the feasibility of using The proposed QCOT has been tested with a constant input
saturable inductors in SMPS applications, highlighting the po- voltage; however, the variable input can be managed by a family
tential benefits in terms of inductor size reduction. The adaptive of LUTs calculated in the expected range of input voltages. For
ON time control based on the thermal model of the inductor is a given input voltage, the corresponding LUT is selected.
implemented as an effective strategy to prevent thermal runaway With the implementation of the LUT family, the QCOT can
caused by saturation. This feasibility has been assessed by a address significant variations in input voltage, which is of partic-
low-voltage, low-power dc–dc boost converter with a switching ular interest in applications, such as LED drivers and automotive
SCIRÈ et al.: QUASI-CONSTANT ON-TIME CONTROL FOR SMPS WITH A NONLINEAR INDUCTOR 9

systems, where large voltage surges are common. Indeed, in [3] M. L. Heldwein and J. W. Kolar, “Impact of EMC filters on the power
case of SMPS exploiting nonlinear inductors, also the maximum density of modern three-phase PWM converters,” IEEE Trans. Power
Electron., vol. 24, no. 6, pp. 1577–1588, Jun. 2009.
input current must be regulated by the QCOT to avoid the deep [4] D. Scirè, G. Lullo, and G. Vitale, “EMI filter re-design in a SMPS with
saturation of the inductor core. inductor in saturation,” in Proc. IEEE 15th Int. Conf. Compat., Power
On this basis, for any given inductor, its operating range can Electron. Power Eng., 2021, pp. 1–7.
[5] T. Guillod, P. Papamanolis, and J. W. Kolar, “Artificial neural network
be extended up to the 40% of the rated current, the proposed (ANN) based fast and accurate inductor modeling and design,” IEEE Open
characterization allows defining the operating frequency range; J. Power Electron., vol. 1, pp. 284–299, 2020.
the constraint for the choice of the microprocessor is represented [6] H. Komurcugil, S. Biricik, and N. Guler, “Indirect sliding mode control
for DC–DC SEPIC converters,” IEEE Trans. Ind. Inform., vol. 16, no. 6,
by the maximum switching frequency for the PWM modulator. pp. 4099–4108, Jun. 2020.
Once the inductor characterization is performed offline, the [7] H. Du, C. Jiang, G. Wen, W. Zhu, and Y. Cheng, “Current sharing
implementation as an add-on is simple. control for parallel DC–DC buck converters based on finite-time control
technique,” IEEE Trans. Ind. Inform., vol. 15, no. 4, pp. 2186–2198,
Apr. 2019.
VIII. CONCLUSION [8] S. Lucia, D. Navarro, B. Karg, H. Sarnago, and O. Lucia, “Deep learning-
based model predictive control for resonant power converters,” IEEE
The exploitation of the nonlinearity of a power inductor up to Trans. Ind. Inform., vol. 17, no. 1, pp. 409–420, Jan. 2021.
the saturation zone in SMPSs requires that the conduction time of [9] G. D. Capua and N. Femia, “A novel method to predict the real operation
of ferrite inductors with moderate saturation in switching power supply
the power switch is tuned based on the inductor behavior, which applications,” IEEE Trans. Power Electron., vol. 31, no. 3, pp. 2456–2464,
depends on both current and temperature. However, it allows a Mar. 2016.
current threshold higher than that of a linear inductor with the [10] M. S. Perdigão, J. P. F. Trovão, J. M. Alonso, and E. S. Saraiva, “Large-
signal characterization of power inductors in EV bidirectional DC–DC
same rated value, but with a reduced volume, thus improving converters focused on core size optimization,” IEEE Trans. Ind. Electron.,
the power density. vol. 62, no. 5, pp. 3042–3051, May 2015.
This article proposes a method for retrieving the current flow- [11] A. Oliveri, M. Lodi, and M. Storace, “Nonlinear models of power induc-
tors: A survey,” Int. J. Circuit Theory Appl., vol. 50, no. 1, pp. 2–34, 2022.
ing through a nonlinear inductor that operates up to saturation. [12] A. Oliveri, G. Di Capua, K. Stoyka, M. Lodi, M. Storace, and N. Femia, “A
This method was exploited by an algorithm to estimate the power-loss-dependent inductance model for ferrite-core power inductors
time required for the current to reach the maximum value cor- in switch-mode power supplies,” IEEE Trans. Circuits Syst. I: Reg. Papers,
vol. 66, no. 6, pp. 2394–2402, Jun. 2019.
responding to the conduction time of the power switch. It is a [13] A. Abramovitz and S. Ben-Yaakov, “RGSE-based SPICE model of ferrite
recursive algorithm from which a complete characterization of core losses,” IEEE Trans. Power Electron., vol. 33, no. 4, pp. 2825–2831,
the inductor in terms of TON versus current peak and versus core Apr. 2018.
[14] X. Duan and A. Q. Huang, “Current-mode variable-frequency con-
temperature is obtained. These curves, which were implemented trol architecture for high-current low-voltage DC–DC converters [let-
in a LUT, were used to implement a quasi constant TON technique ters],” IEEE Trans. Power Electron., vol. 21, no. 4, pp. 1133–1137,
in the boost converter. The QCOT implementation is an add-on Jul. 2006.
[15] W.-H. Yang et al., “A constant-on-time control DC–DC buck converter
tool that preserves the dynamic of the PI control and imposes with the pseudowave tracking technique for regulation accuracy and load
the switching frequency maintaining the duty cycle. Two case transient enhancement,” IEEE Trans. Power Electron., vol. 33, no. 7,
studies were proposed with two different loads to demonstrate pp. 6187–6198, Jul. 2018.
[16] G. Vitale, G. Lullo, and D. Scirè, “Thermal stability of a DC/DC converter
that an increase of 40% in the rated current corresponding to with inductor in partial saturation,” IEEE Trans. Ind. Electron., vol. 68,
saturation exploitation can be managed by the QCOT algorithm. no. 9, pp. 7985–7995, Sep. 2021.
The QCOT increased the switching frequency to tune a suitable [17] D. Scirè, G. Lullo, and G. Vitale, “Non-linear inductor models comparison
for switched-mode power supplies applications,” Electronics, vol. 11,
TON value. A current exceeding saturation leads to thermal run- no. 15, 2022, Art. no. 2472.
away, which is stopped by temperature detection. The increase [18] C. W. T. McLyman, Transformer and Inductor Design Handbook, 4th ed.
in the operating current of 40% resulted in a 70% reduction in Boca Raton, FL, USA: CRC Press, 2011.
[19] A. Van den Bossche and V. C. Valchev, Inductors and Transformers for
the volume and a cheaper inductor. Power Electronics, 1st ed. Boca Raton, FL, USA: CRC Press, 2005.
[20] N. Femia, K. Stoyka, and G. D. Capua, “Impact of inductors saturation
ACKNOWLEDGMENT on peak-current mode control operation,” IEEE Trans. Power Electron.,
vol. 35, no. 10, pp. 10969–10981, Oct. 2020.
The authors would like to thank the technicians of ICAR-CNR [21] S. Musumeci, L. Solimene, and C. S. Ragusa, “Identification of DC thermal
steady-state differential inductance of ferrite power inductors,” Energies,
Giampiero Rizzo for their support in setting up the computa- vol. 14, no. 13, 2021, Art. no. 3854.
tional system to carry out the simulations and Giovanni Ruggieri [22] A. Oliveri, M. Lodi, and M. Storace, “A piecewise-affine inductance model
to arrange the hardware to perform the experimental test given for inductors working in nonlinear region,” in Proc. 16th Int. Conf. Synth.,
Model., Anal. Simul. Methods Appl. Circuit Des., 2019, pp. 169–172.
in this article. [23] P. Burrascano, G. Di Capua, N. Femia, S. Laureti, and M. Ricci, “A pulse
compression procedure for power inductors modeling up to moderate non-
REFERENCES linearity,” Integration, vol. 66, pp. 16–23, 2019.
[24] G. Pilato, G. Vitale, G. Vassallo, and D. Sciré, “Neural modeling of power
[1] N. Lee, J.-Y. Lee, Y.-J. Cheon, S.-K. Han, and G.-W. Moon, “A high- nonlinear inductors by the E-αNet network,” Nonlinear Dyn., vol. 112,
power-density converter with a continuous input current waveform for no. 19, pp. 17069–17086, 2024.
satellite power applications,” IEEE Trans. Ind. Electron., vol. 67, no. 2, [25] A. Barili, A. Brambilla, G. Cottafava, and E. Dallago, “A simulation
pp. 1024–1035, Feb. 2020. model for the saturable reactor,” IEEE Trans. Ind. Electron., vol. 35, no. 2,
[2] D. Shahzad, S. Pervaiz, N. A. Zaffar, and K. K. Afridi, “GaN-based pp. 301–306, May 1988.
high-power-density AC–DC–AC converter for single-phase transformer- [26] S. F. Roberto, D. Sciré, G. Lullo, and G. Vitale, “Equivalent circuit
less online uninterruptible power supply,” IEEE Trans. Power Electron., modelling of ferrite inductors losses,” in Proc. IEEE 4th Int. Forum Res.
vol. 36, no. 12, pp. 13968–13984, Dec. 2021. Technol. Soc. Ind., 2018, pp. 1–4.
10 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS

[27] K. Górecki and K. Detka, “Application of average electrothermal models in Giuseppe Lullo (Member, IEEE) was born
the SPICE-aided analysis of boost converters,” IEEE Trans. Ind. Electron., in Palermo, Italy, in 1965. He received the
vol. 66, no. 4, pp. 2746–2755, Apr. 2019. M.S. (Hons.) and Ph.D. degrees in electronic
[28] F. Bizzarri, M. Lodi, A. Oliveri, A. Brambilla, and M. Storace, “A engineering from the University of Palermo,
nonlinear inductance model able to reproduce thermal transient in SMPS Palermo, Italy, in 1990 and 1995, respectively.
simulations,” in Proc. 2019 IEEE Int. Symp. Circuits Syst., 2019, pp. 1–5. From 1993 to 1994, he was a Visiting Re-
[29] D. Scirè, G. Lullo, and G. Vitale, “Assessment of the current for a search Fellow within the Optoelectronics Labo-
non-linear power inductor including temperature in DC-DC converters,” ratory, Glasgow University, Glasgow, U.K., and
Electronics, vol. 12, no. 3, 2023, Art. no. 579. in 1998, a Visiting Scientist within the Depart-
[30] D. Scirè, G. Lullo, and G. Vitale, “Design and modeling of an interleaving ment of Electronics and Computer Science,
boost converter with quasi-saturated inductors for electric vehicles,” in Massachusetts Institute of Technology, Cam-
Proc. 2020 AEIT Int. Conf. Elect. Electron. Technol. Automot., 2020, bridge, MA, USA. In 1999, he became a Researcher and in 2005, an
pp. 1–6. Associate Professor of electronics with the University of Palermo. More
[31] D. Scirè, S. Rosato, G. Lullo, and G. Vitale, “Characterization of non-linear recently, he has been involved in research in the field of power electronic
inductors including thermal effects for power applications,” Renewable systems. He holds a patent on an optical diffractometer. He also founded
Energy Power Qual. J., vol. 1, no. 16, pp. 728–734, 2018. and comanaged for twenty years, as a Chief Technical Officer, Microtech
[32] D. Scirè, G. Vitale, M. Ventimiglia, and G. Lullo, “Non-linear inductors s.r.l., a high-tech spin-off company devoted to the realization of laser
characterization in real operating conditions for power density optimiza- direct writing equipments for microlithographic processes. His research
tion in SMPS,” Energies, vol. 14, no. 13, 2021, Art. no. 3924. interests include several areas, regarding the design and fabrication
[33] M. Kachniarz, J. Salach, R. Szewczyk, and A. Bieńkowski, “Temper- of optoelectronic and electronic devices, the design of custom optical
ature influence on the magnetic characteristics of Mn-Zn ferrite mate- systems, the design of mixed-signal electronic systems, and electronics
rials,” in Progress in Automation, Robotics and Measuring Techniques, for controlling X-ray satellite telescopes.
R. Szewczyk, C. Zieliński, and M. Kaliczyńska, Eds. Cham, Switzerland: Dr. Lullo is a Member of the Italian National Institute for Astrophysics.
Springer, 2015, pp. 121–127.
[34] C. Chen et al., “Online inductor parameters identification by small-signal
injection for sensorless predictive current controlled boost converter,”
IEEE Trans. Ind. Inform., vol. 13, no. 4, pp. 1554–1564, Aug. 2017.
[35] M. Ventimiglia, D. Sciré, G. Lullo, and G. Vitale, “A measurement system
for power inductors in non-linear operating conditions,” in Proc. IEEE
30th Int. Symp. Ind. Electron., 2021, pp. 1–6.

Daniele Scirè (Member, IEEE) was born

in Palermo, Italy, in 1991. He received the
M.S. degree in electronic engineering and the
Ph.D. degree in information and communication
technologies from the University of Palermo, Gianpaolo Vitale (Senior Member, IEEE) was
Palermo, Italy, in 2017 and 2021 respectively. born in Palermo, Italy, in 1964. He received the
From 2018 to 2019, he was a Visiting Ph.D. M.S. degree in electronic engineering from the
Student with the Photovoltaic Material and De- University of Palermo, Palermo, Italy, in 1988.
vices group, Delft University of Technology, He is the Research Director with the Institute
Delft, The Netherlands. Since 2022, he has for High Performance Computing and Network-
been an Assistant Professor with the University ing belonging to the National Research Council
of Palermo. His current research interests include power electronics of Italy. He teaches “industrial electronics” with
converter and their control, nonlinear magnetic components, modeling the Department of Engineering, MD Electron-
and design of solar cells, and thin film deposition and characterization. ics Engineering, University of Palermo. He has
Dr. Scirè was the recipient of the Graduate Student Award from coauthored two books, two edited books, four
the European Material Research Society in 2019 and the Best Paper book chapters, and more than 100 scientific works. His current research
Award “Salvatore (Enzo) Piazza” from the Department of Engineering, interests include the fields of power electronics, power generation from
University of Palermo, in 2022. He is a Reviewer for several journals and renewable sources, electromagnetic compatibility, and robotics.
conferences. Prof. Vitale is reviewer for several journals and conferences.