Вы находитесь на странице: 1из 12

IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 29, NO.

1, JANUARY 2014

201

Maximum Power Transfer Tracking for


Ultralow-Power Electromagnetic Energy Harvesters
Gyorgy D. Szarka, Stephen G. Burrow, Plamen P. Proynov, and Bernard H. Stark

AbstractThis paper describes the design and operation of


power conditioning system with maximum power transfer tracking (MPTT) for low-power electromagnetic energy harvesters. The
system is fully autonomous, starts up from zero stored energy, and
actively rectifies and boosts the harvester voltage. The power conditioning system is able to operate the harvester at the maximum
power point against varying excitation and load conditions, resulting in significantly increased power generation when the load
current waveform has a high peak-to-mean ratio. First, the paper
sets out the argument for MPTT, alongside the discussion on the
dynamic effects of varying electrical damping on the mechanical
structure. With sources featuring stored energy, such as a resonant
harvester, maximum power point control can become unstable in
certain conditions, and thus, a method to determine the maximum
rate of change of electrical damping is presented. The complete
power conditioning circuit is tested with an electromagnetic energy harvester that generates 600 mVrm s ac output at 870 W under optimum load conditions, at 3.75 ms2 excitation. The digital
MPTT control circuit is shown to successfully track the optimum
operating conditions, responding to changes in both excitation and
the load conditions. At 2 Vd c output, the total current consumption
of the combined ancillary and control circuits is just 22 A. The
power conditioning system is capable of transferring up to 70% of
the potentially extractable power to the energy storage.
Index TermsACDC converter, energy harvesting, low power,
maximum power tracking, rectification.

I. INTRODUCTION
HE output of small electromagnetic energy harvesters typically requires rectification and boosting in order to produce an output voltage that falls within the allowable operating
range of the load electronics. In some applications, there is also
a need to buffer energy in high capacity storage elements, such
as supercapacitors, in order to supply loads with a higher peak
demand than the harvester output [1]. Several circuit architectures have been reported in published literature, which meet
these requirements, including single-stage acdc switch-mode
power converters [2]. Efficiencies up to 75%80% at 500 W

Manuscript received October 11, 2012; revised December 21, 2012 and
January 23, 2013; accepted February 19, 2013. Date of current version July
18, 2013. Recommended for publication by Associate Editor S. Y. (Ron) Hui.
G. D. Szarka, P. P. Proynov, and B. H. Stark are with the Department of Electrical and Electronic Engineering, University of Bristol, Bristol,
BS2 8BB, U.K. (e-mail: G.Szarka.04@bristol.ac.uk; p.proynov@bristol.ac.uk;
Bernard.Stark@bristol.ac.uk).
S. G. Burrow is with the Department of Aerospace Engineering, University
of Bristol, Bristol, BS2 8BB, U.K. (e-mail: stephen.burrow@bristol.ac.uk).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TPEL.2013.2251427

level have been reported [3]. However, in order to achieve the


maximum potential power of an energy harvester, it is important
that the power conditioning system provides the optimum load
for the generator for the particular input and output conditions.
Vibration harvesters have a peak-power-type response:
power variation with changing load damping is not monotonic,
displaying a peak at a damping level determined by harvester parameters [4]. Negative-feedback voltage regulation for switching converters cannot provide a stable operation at the peak
power point as the operating conditions required violate Middlebrooks stability criterion [5]. The optimum damping level of an
ideal harvester driven at resonance is independent of excitation
amplitude; hence, one solution is to employ a converter emulating a fixed impedance to the harvester, as reported in [5] and [6].
However, fixed input impedance places restrictions on converter
design (principally requiring discontinuous conduction), and
this is difficult to maintain over the full range of input and output conditions. The challenge for the power converter is then
to provide the basic functionality of rectification and voltagelevel shifting, while loading the harvester with the optimum
impedance, independently of its input and output conditions. In
this paper, the electromagnetic energy harvester is assumed to
be operating at its mechanical resonance frequency, making its
effective output impedance dominantly resistive [7], and constant over time. By contrast, the apparent input impedance of
the power converter depends on the input and output voltage
conditions. As these vary over time, dynamic control is required
to maintain the desired converter input impedance.
Maximum power point tracking (MPPT) schemes employ an
algorithm, such as gradient descent, to locate a peak power point
for the prevailing operating conditions. They are commonly used
to maintain temperature-dependent photovoltaic cells at their
peak power point [8], [9], and offer the ability to optimally load
an energy harvester, compensating for both variations in optimal load and variations in converter output conditions. However,
when used with systems with significant stored energy (like the
kinetic energy harvesters investigated here) extra care must be
taken to ensure correct operation. The literature on maximum
power point tracking solutions for energy harvesting is relatively
sparse. Elmes et al. [10] reported a maximum energy harvesting
control scheme for an energy harvesting backpack generating
tens of watts of power. In [11], a microcontroller-based power
point tracker is reported for a rotational generator producing dc
output up to 10 mW; the circuit employs a resistance matching
strategy, similar to the approach adapted in [12] for submilliwatt
RF energy harvesting. One of the first authors to publish on peak
power control in the energy harvesting literature was Ottman;
the presented power converter architecture differed from the

0885-8993/$31.00 2013 IEEE

202

work here as the rectifier of the two-stage topology fed into a


stiff dc link, which itself was controlled to provide the power
tracking capability. Additionally, the paper focused on piezoelectric transducers. Low-power solutions offering regulation
of the buffered output voltage of the rectification stage for maximum power point operation have been demonstrated in the literature: for piezoelectric harvesters using an off-the-shelf buck
converter operated intermittently [14], or for electromagnetic
generators using a low-power integrated boost converter [15].
Recently, an integrated maximum power point tracking circuit
has been presented in [16] that operates down to sub-100 W
levels, using an approach referred to as power-optimal point of
charging for the control of a dcdc charge-pump.
These solutions are designed to maximize the power generated by the harvester without considering the losses of power
conversion or the quiescent power overheads of any control and
ancillary circuits. Derivatives of the MPPT technique, referred
to as maximum power transfer tracking (MPTT) [8], are designed to maximize the power transferred to the load and the
energy storage element. Dayal and Parsa [17] proposed a lowpower implementation for an MPTT scheme, but only simulated
results are presented for the control circuit. In the reported paper, the output voltage is maximized by varying the duty ratio
of the pulse width modulation (PWM) driving signal of a splitcapacitor acdc converter. This technique assumes that the load
is near constant and purely resistive; otherwise, maximizing
the voltage would not yield maximum output power conditions.
Also, it requires the energy storage element to be small in order
to be able to monitor the effect of varying duty ratio. These
assumptions pose impractical limitations on small-scale energy
harvesting that is typically characterized by very low generated
power levels, large energy storage elements, and load circuits
with highly dynamic power consumption.
The effects of the stored energy within the mechanical oscillator of the harvester on MPPT are alluded to in [10] and
[11], by stating that the control loop has to be slow in order to avoid instability. However, the behavior of the harvester under varying damping and the implications regarding
the design of the control circuit has not been discussed in the
literature.
In this paper, the maximum response rate of the MPTT
algorithm resulting in stable operation is investigated and a
complete MPTT harvesting system is presented. Section II describes an electromagnetic energy harvester with the governing
equations of motion, investigating the harvesters response to
dynamic electrical damping experimentally. Next, an analysis
of the harvesters response to a step change in the damping is
described, which enables the derivation of the minimum settling
time required between perturbations of the control parameter.
Section III provides a description of the power conditioning
circuit. Section IV presents the operating principles for the
perturb-and-observe algorithm and describes the control circuit.
Section V presents experimental results that show the steadystate performance of the MPTT and the behavior under transient
load and excitation conditions. Finally, Section VI summarizes
the key findings and concludes with suggestions for future
work.

IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 29, NO. 1, JANUARY 2014

Fig. 1. Small-scale electromagnetic energy harvester with a 3.4 g NdFeB


magnet acting as the moving mass within a coil wound using 600 turns of 100 m
diameter copper wire. The resonant frequency is 43.8 Hz. A piezoelectric thin
film is bonded to the top of the cantilever for displacement monitoring.

II. ENERGY HARVESTER: STEADY-STATE AND


TRANSIENT CHARACTERISTICS
A. Electromagnetic Energy Harvester
A cantilever-type vibration harvester is used in this paper,
shown in Fig. 1. It features a BeCu beam with a 3.4 g NdFeB magnet acting as the tip mass. Energy from the motion
is transferred to the electrical domain via the electromagnetic
coupling between the moving magnet and a wound coil. The tip
mass is large compared to the beam; hence, the generator can
be modeled as a base excited, second-order, velocity-damped
massspring system, where the response to external forcing is
described by
m
z + cz + kz = m 2 Y sin t.

(1)

The base displacement has an amplitude Y at an angular


frequency of . The equivalent mass at the tip is given as m,
while c denotes the viscous damping within the system that is the
combination of the mechanical and the electrical damping. The
spring constant of the compliant beam is k, and z refers to the
motion of the moving mass relative to the frame. The schematic
illustrations of the electromagnetic energy harvester and the
lumped element model of the spring-mass-damper system are
presented in Fig. 2.
The steady-state solution, referring to the condition where the
amplitude of the periodic motion is constant, is given as [4]
z(t) = Z sin(t )

(2)

where the amplitude is given by


Z=

m 2 Y

(3)

(k 2 m)2 + c2 2

and the phase angle between the base and tip displacement is
given by
= tan1

c
.
k 2 m

(4)

At the natural resonance frequency, which is described by



k
(5)
= n =
m

SZARKA et al.: MAXIMUM POWER TRANSFER TRACKING FOR ULTRALOW-POWER ELECTROMAGNETIC ENERGY HARVESTERS

203

Fig. 2. (a) Schematic illustration of the electromagnetic energy harvester


shown in Fig. 1, and (b) illustration of the lumped elements of the spring
massdamper system.

Fig. 4. Generated output power versus time corresponding to the 0.5, 1, 2, 4,


and 8 s load resistance sweep profiles of Fig. 3. Excitation amplitude is actively
held at 3.75 ms2 and the frequency is 43.8 Hz.

Fig. 3. Load resistance profiles with various gradients, resulting in linear load
resistance sweeps between 1000 and 100 in 0.5, 1 s, 2, 4, and 8 s. Both
(left) decreasing and (right) increasing resistance profiles are considered.

(2) simplifies to


mY
sin t
.
(6)
c
2
Depending on the application, both of the excitation frequency and amplitude Y can vary over time. Furthermore,
damping arising from the interfacing electronics affects the total damping given by [4]
z (t) =

c = cm +

Rcoil

2
+ Rload

(7)

where cm represents the mechanical damping, the electromagnetic coupling coefficient, and Rcoil is the parasitic resistance
of the coil. Rload is the load resistance, which is synthesized by
the input impedance of the power converter in a practical system. In typical small-scale electromagnetic energy harvesters,
the impedance of the parasitic coil inductance at the excitation
frequency is several orders of magnitude lower than the combined equivalent mechanical resistive output impedance and ac
coil resistance. Therefore, it is assumed that close to the theoretical, maximum power can be extracted using a purely resistive
load at resonance.
B. Illustration of Response to Dynamic Damping
The transient response of the mechanical system is related
to the energy stored in the oscillator and the total damping of
the system. To illustrate this experimentally, the load applied to
the harvester of Fig. 1 is swept from 100 to 1000 at differing
rates (see Fig. 3), while the excitation is kept constant. Prior to

Fig. 5. Measured generated power profiles of Fig. 4 mapped onto corresponding load resistance sweeps and compared against steady-state measurements.

each sweep, the harvester is in steady state. The instantaneous


generated power is measured and averaged over one cycle. The
resulting output powertime profiles are presented in Fig. 4.
The output power mapped onto the corresponding load resistance is shown in Fig. 5, resulting in hysteretic trajectories
of output power as the load is swept up and down at a particular rate. The cycle-averaged power recorded under steady-state
conditions is also shown for comparison. This reveals that the
maximum power that can be sustained by the energy harvester
at the optimum resistance of 400 is around 870 W; however, during a sweep, much higher powers are available if the
damping is rapidly increased and much lower powers while the
damping is being reduced.
This behavior can be understood by considering the energy
stored in the mechanical components of the harvester. Most
harvesters feature significant amplification of source vibrations,
analogous to the quality factor or Q of electrical resonant circuits. Q also defines the ratio of energy stored in an oscillator to
that dissipated each cycle. Thus, for high-Q systems, a change in
the operating point requires significant energy to be either added
or dissipated. At high load resistance, corresponding to low

204

IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 29, NO. 1, JANUARY 2014

In the steady-state solution of (6), the amplitude of the tip


displacement before the step change is
Zo =

mY
co

(9)

and after the step change, the oscillation should settle to the new
steady-state solution with a tip displacement amplitude of
Zs =

mY
.
cs

(10)

The initial conditions for the nonhomogeneous second-order


differential equation of (1) are obtained by finding the steadystate displacement and velocity at time t = 0. Hence, the initial
displacement is z (0) = Zo , and velocity is z (0) = 0. The
solution of the differential equation is
z (t) = (Zs Zo ) e(c s /2m )t cos (t)
(Zs Zo ) cs (c s /2m )t
e
sin (t) Zs cos (t) (11)
2m

where is given as 1 2 , and the damping ratio is
+

Fig. 6. (a) Generated rms current and voltage waveforms during the 4 s sweep
transients, and (b) current and voltage excursions mapped to the corresponding
load resistance.

damping, significant energy is stored in the system. Increasing


the damping reduces the displacement of the mass, thereby reducing the stored energy, which is seen as transient additional
output power. Conversely, reducing the damping leads to an
increase in stored energy, which although is supplied by the
vibration source, limits the rate at which the new steady state
is approached. As the rate of change of the electrical damping is reduced, the measured output power levels converge to
the steady-state solution. The generated rms current and voltage transients are illustrated in Fig. 6 for the 4-s sweep case.
The curves show that during the increase of the damping, the
corresponding transient voltage and current travels are higher
than during decreasing damping, as some of the initial kinetic
energy of the energy harvester is dissipated. The behavior illustrated here is of great importance when implementing a power
tracking scheme. If the controller does not take into account the
transient component of the output power by allowing sufficient
settling time of the harvester, the system can become unstable.
C. Response to Step Change in Damping
In this section, a step change in the damping is considered,
as might be the case when a digital system steps a control reference signal. The harvester is modeled as a highly underdamped,
single-degree-of-freedom massspringdamper system, excited
at its natural frequency with constant acceleration amplitude.
Also, prior to the change, the mechanical structure is assumed
to be in steady state. The damping as a function of time can be
described as

c(t) =

co ,
cs ,

t0
.
t>0

(8)

c
=
.
2 mk

(12)

In a highly underdamped system ( < 0.1), such as typical


small-scale electromagnetic energy harvesters, is assumed to
be 1. This simplifies the solution to


z (t) = Zs + (Zo Zs ) e(c s /2m )t cos (t)
(Zo Zs ) cs (c s /2m )t
e
sin (t) .
(13)
2m
This equation shows that the underlying sinusoidal oscillation
with initial amplitude of Zo decays exponentially to a sinusoidal
oscillation with amplitude of Zs , with a rate that is determined
by the total damping and the moving mass.
This time-domain solution is validated (see Fig. 7) using
measured output power for several step changes in the load
resistance. The measurement results show a good correlation
with the calculated waveforms.
The second exponential term of (13) can be considered negligible in a highly underdamped system, where the damping coefficient is very small, as the denominator of multiplying fraction
is typically greater than 1, and the product of (Zo Zs ) cs is
small, in the order of 105 . Hence, the amplitude of the sinusoidal oscillation is dominated by
+

|z|
= Zs + (Zo Zs ) e(c s /2m )t

(14)

an exponential decay to the new tip displacement amplitude.


Differentiating this solution yields the approximate solution for
the velocity
cs (c s /2m )t
e
cos (t)
v (t) = Zs sin (t) + (Zo Zs )
2m
+ (Zo Zs ) e(c s /2m )t sin (t) .

(15)

Considering that the velocity amplitude in the steady-state


solution is given as V = Z and that  c/2m, the velocity

SZARKA et al.: MAXIMUM POWER TRANSFER TRACKING FOR ULTRALOW-POWER ELECTROMAGNETIC ENERGY HARVESTERS

205

where Is and Io are given according to


I=

m 2 Y
c (Rcoil + Rload2 )

(20)

with c=cs and c=co , respectively. The instantaneous generated


power dissipated in the load resistance is given as
P (t) = i2 (t) Rload2 .

(21)

Defining a common factor as


A=

2 m2 4 Y 2
Rload2
(Rcoil + Rload2 )2

allows us to write the current and power amplitudes as






1
1
A
1
(c s /2m )t
|I| =
+

e
Rload2 cs
co
cs
Fig. 7. Calculated, based on (13), and measured average output power during
the transient response of the system that occurred after a step change in the
load resistance. Initial load resistances are shown on the right; after the step
change, the resistance is constant at 400 . Excitation is a constant 3.75 ms2
acceleration at 43.8 Hz.

amplitude during the transient response can be approximated


with great fidelity in a highly underdamped systems as


(16)
|v| = Vs + Vo Vs e(c s /2m )t .

and



1
1
2
1
+

e(c s /2m )t
c2s
cs co
cs


2
1
1
(c s /m )t
+

e
.
co
cs

U (t) = Blv(t) = v(t)

where B is the magnetic field strength and l is the effective length


of the conductor within the magnetic field. Thus, the amplitude
of the induced sinusoidal voltage is the coupling coefficient
times the velocity amplitude given in (16). The steady-state
instantaneous current is dependent on the total load resistance
of the circuit and is equal to
i(t) =

U (t)
Rcoil + Rload2

(18)

where Rload2 denotes the load resistance after the step change,
and during the settling time. During the settling time, the amplitude of the generated current will decay exponentially toward
the steady-state value
I (t) = Is + (Io Is ) e(c s /2m )t

(19)

(24)

According to the definition of the settling time, at t = ts the


amplitude of the power for increasing damping should be
|P | = Ps (1 + )

(17)

(23)

|P | = A

D. Minimum Settling Time Selection


The assumption is made that in order to avoid instability from
any system condition, a digital control system should initiate a
step change in damping c only once the system has settled. The
settling time in this paper is defined as the time required for the
output power to reach within 1% of its final value. In a typical MPPT system, adjusting of the control parameter varies the
apparent input resistance of the interfacing power electronics,
resulting in discrete step perturbation of the generator damping. Considering these steps, a harvester settling time can be
defined and used as the lower limit on the time period between
adjustments.
The counterelectromotive force induced in the coil is proportional to the velocity of the moving magnet according to

(22)

(25)

where Ps = A/c2s , and the error is 0.01. Substituting (25) into


(24), then dividing both sides by A and rearranging to one side
yields


1
2
1
1

e(c s /2m )t s
2 +
cs
cs co
cs

2
1
1
+

e(c s /m )t s = 0
(26)
co
cs
cs

Multiplying both sides by e m ts c2s co / (cs co ) gives a


quadratic equation in the form of

co
cs co
x2 2x
=0
cs co
co

(27)

where x = exp[(cs /m)ts ]. Only the positive root of the


quadratic equation yields a real solution

1+ 1+
(c s /2m )t s
(28)
=
e
co /(cs c0 )
Isolating the settling time ts gives

2m 1 + 1 +
ts =
ln
|cs co | .
cs
co

(29)

Equation (29) gives a conservative estimate that typically falls


within 0.5% of the exact solution derived numerically from (13).
Therefore, this formula provides a convenient means of estimating the minimum interval between perturbations in digital
maximum power tracking control systems. It also shows the
dependence on harvester parameters and tolerances.

206

IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 29, NO. 1, JANUARY 2014

Fig. 8.

System-level block diagram showing the main power electronics blocks and the power definitions within the main power flow of the system.

III. POWER CONVERTER AND ANCILLARY CIRCUITRY


The power conditioning system architecture is shown in
Fig. 8. It builds on the low-power system reported by Szarka
et al. [6], with the addition of low-power measurement and control circuits for the MPTT control, and a modified low-power
gate drive circuit design.
The output voltage across the supercapacitor varies, thereby
allowing the stored energy to be maximized; at the same time it
is kept within the allowable dc supply voltage range of the load
electronics, which avoids the losses of additional voltage regulation. Zero-energy start-up is provided by a passive voltage
multiplier circuit that is automatically disconnected when the
active power converter circuit becomes operational. The main
power converter is a nonsynchronous full-wave boost rectifier,
as shown in Fig. 9, which provides rectification and voltage
boosting in a single stage. The parasitic coil impedance, although considered to be negligible at the mechanical resonance
frequency, is significant at the switching frequency of the power
converter and can be used to eliminate the need for an additional
boost inductor. The MPTT control circuit adjusts the duty ratio
of the PWM gate drive signals. This controls the apparent
input impedance of the converter, preserving maximum output
power Pstore. Consequently, near-optimum damping conditions
for the energy harvester are maintained independently of variations in the output voltage, the conduction mode, and the excitation magnitude; all of which are time-varying factors that
influence the apparent input impedance of the power converter.
The particular converter topology selected here does not support
bidirectional power flow, thus restricting the control to resistive
impedance matching only. This approach may be suboptimal
when the generator is excited off resonance; however, it offers

Fig. 9. Circuit schematic of a nonsynchronous, full-wave boost rectifier that


uses the parasitic coil inductance as the boost inductor. The semiconductor
devices used are M1/M2PMF-280UN and D1/D2 1PS79SB30.

high utilization when the source excitation is at the mechanical


resonance frequency [3]. An example of typical input current
and voltage waveforms are shown in Fig. 10.
The gate drive circuit generates two-variable duty ratio PWM
signals for the two low-side transistors. The switching frequency
is constant 32.768 kHz, determined by the output of the lowpower oscillator chip (OV-7604), as shown in Fig. 11. Polarity
of the input voltage is sensed by a thin-film piezoelectric sensor
bonded to the cantilever beam.
A saw-tooth signal is created using a one-shot circuit and a
modified RC filter that provides a slow discharge of the 15 pF
capacitor. The duty ratio is determined by the Reference
input, which is the common interface to the control circuits.
The logic output stage, which directly drives the gates of the
MOSFETS, receives polarity information from the detection
circuit. At any moment in time, only one of the outputs is toggled at the switching frequency, while the other PWM signal is
kept high. The only exception to this is during a blanking period
around the zero-crossing of the generators output voltage, when
both outputs are low.

SZARKA et al.: MAXIMUM POWER TRANSFER TRACKING FOR ULTRALOW-POWER ELECTROMAGNETIC ENERGY HARVESTERS

207

Fig. 10. Measured typical input current Iin and voltage V in waveforms. A period of 100 s at 5 ms point is shown on the right for increased resolution. The duty
ratio of the converter is 67% and the output voltage is 3.2 V. The energy harvester is excited with 43.8 Hz, 3 ms2 acceleration, producing a little over 450 W.

The printed circuit board implementation of the power conditioning system is shown in Fig. 12, corresponding to the block
diagram presented in Fig. 12.
IV. CURRENT-SHUNT-BASED MAXIMUM POWER
TRANSFER TRACKING
A. Operating Principles and Measurement
A typical low-power energy harvesting system requires large
capacity storage in order to be able to supply the high peak-tomean ratio power demand required by load electronics such as
wireless sensor nodes. The combination of low output power
and large storage capacitance results in slow charge up of the
supercapacitor. This enables the MPTT control to rely solely on
output current measurement. The reference signal that sets the
duty ratio of the boost rectifier is perturbed, and the effect of
this on the power transferred to the supercapacitor is observed
by measuring the output current while the output voltage is assumed to be near constant. This implementation requires the rate
of change of the output voltage to remain low enough to ensure
a near-constant voltage between successive measurements. In
the presented system, this limits the minimum capacitor size to
20 mF, when the worst-case voltage increase between measurements is 5 mV.
The measurement circuit is an operational amplifier based
circuit (see Fig. 13) that is designed to measure the voltage
drop across a 150 precision shunt resistor. The current ripple,
both at the switching and at the excitation frequency, must be
minimized to ensure correct measurement.
This is achieved by a combination of parallel capacitance
introduced before the shunt resistor in the circuit, and an active low-pass SallenKey [18] architecture employed in the design of the amplifier circuit. The 30 F capacitor smoothes the
switching frequency current ripple without adding any significant delay to the response of the system to the perturbation.

Fig. 11. Gate drive circuit generating 32.768 kHz output PWM signals. Pos,
Neg, and Blank are digital signals from the polarity detection circuit [6].
Reference is an analogue signal from the control circuit to set the duty ratio.

Fig. 12. Printed circuit board implementation of the main power conditioning
system blocks, corresponding to Fig. 12.

The second-order Butterworth filter has a cutoff frequency of


20 Hz, approximately half the excitation frequency, in order to
provide a close to dc current measurement at the output of the
amplifier. The micropower operational amplifier MCP6031 is
selected because of its low input offset voltage that is typically

208

IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 29, NO. 1, JANUARY 2014

Fig. 13. Shunt-resistor-based current-sense amplifier using a low-pass Sallen


Key architecture [18].

Fig. 15. Steady-state characterization of power conditioning system in openloop configuration. Voltage reference is stepped using an external signal at
>5 s intervals. Excitation is held at 3.75 ms2 and 43.8 Hz. Output voltage is
regulated using a shunt regulator circuit as a load (LM4041).

Fig. 14. Digital MPTT circuit using a low-power MSP430 microcontroller


that samples the output current using a 10-b ADC and controls the reference
voltage of the duty ratio by an 8-bit 0450 mV output R2R ladder. The sample
readings are synchronized to the generators displacement cycle using the output
of the piezoelectric displacement sensor.

The reference voltage is therefore changed in discrete steps of


1.75 mV, corresponding to a 0.6% change in the duty ratio.
The worst-case minimum time required between perturbations,
using (29), is approximately 350 ms, roughly 15 displacement
cycles. The output of a level-crossing detection circuit that monitors the piezoelectric sensor output is used as a nonmaskable
interrupt (NMI) that wakes the microcontroller up from sleep
mode. This approach not only circumvents the need for a power
hungry timer circuit but also ensures that the sample readings are
synchronized to the displacement, which was shown to improve
the control circuits stability [10].
Under this configuration, the microcontroller is active for a
total of 90 s, plus an additional 25 s when only the internal
ADC is ON, in every 342.5 ms period. This results in an average
power consumption of approximately 750 nW.
V. EXPERIMENTAL RESULTS

around 150 V [19], its low current consumption, and rail-torail input/output signal support. The gain of 30 V/V is selected,
in order to maintain a sufficient signal-to-noise ratio.
B. Digital Implementation of MPTT
The well-known perturb-and-observe algorithm is implemented on a MSP430F1132 microcontroller: the PWM duty
ratio of the boost converter is altered and the consequent change
in the output power is measured when the mechanical structure has settled. Based on the measured outcome, the direction
of the subsequent alteration is determined and the process repeated. The controllers CPU is clocked at 5 MHz frequency
in order to minimize the power hungry on-time. The 10-b onboard analogue-to-digital converter (ADC) converts the current
amplifiers output, and the reference voltage for the duty ratio is
set by an R2R ladder, shown in Fig. 14. The top two bits of the
ladder are connected to ground, while the remaining eight bits
are controlled via a full output port of the controller. This provides a 1.75 mV resolution with a maximum output of 450 mV.

A. Steady-State Performance
In order to evaluate the effectiveness of the MPTT circuitry,
the power Pstore as a function of the duty ratio is measured. This
is the useful output power after subtracting losses incurred due to
nonideal loading of the generator, during the power conversion
process, and the quiescent power overheads of the ancillary and
control circuits. In this test, the output voltage is kept constant
using a micropower shunt regulator (LM4041) and the reference
voltage is stepped using an external source in 5 s intervals to
create quasi-steady-state measurements. During all of the tests
presented in this section, a 68 mF supercapacitor is used as the
main energy storage element. The combined current consumption of the ancillary circuits and the leakage current drawn by
the start-up circuits is 19 A at 2 V output voltage and increases
to 22 A at 4.5 V output. The digital control circuit adds an
additional 3 A, amounting to a total of 44 W of minimum
quiescent power loss.
Fig. 15 presents the results measured at 3.75 ms2 excitation
magnitude, corresponding to a maximum extractable power of

SZARKA et al.: MAXIMUM POWER TRANSFER TRACKING FOR ULTRALOW-POWER ELECTROMAGNETIC ENERGY HARVESTERS

Fig. 16. Measured duty ratio samples and their modal values compared with
the range of duty ratios (shaded region) that correspond to over 99% of the
maximum useful output power based on the measurements of Fig. 15.

870 W under optimum load conditions. The maximum transferred power is 615 W at 2 V at the optimum duty ratio. This
corresponds to a power conversion efficiency of 78.5%, and an
overall system efficiency of 70.7%, where 100% is the maximum extractable power from the energy harvester. As the output
voltage is increased, the optimum duty ratio is also increased.
The maximum useful power is, however, reduced due to the
increased power overheads of the power conditioning circuits.
It is worth noting that a small dip in the power level can be
observed in the 4.5 V line around 80% duty ratio, creating two
local maxima that could result in the failure of a perturb-andobserve MPTT algorithm. During normal operation however,
this situation would not arise, as the output voltage rises slowly
across the large capacity storage element, allowing the algorithm
to track the correct peak. Starting the tracking at high duty ratios
would also reduce the risk of finding the wrong local maximum.
Fig. 16 shows duty ratio at discrete output voltages, held constant by the adjustable shunt regulator circuit. As the duty ratio is
constantly changing, even in these steady-state conditions, 100
recordings of the duty ratio are plotted, as dots, for each voltage,
showing a range of around 2.5%. The solid black line is fitted
over the statistical modal value of each group of 100 samples,
representing the most frequent duty ratio. The shaded region
in Fig. 16 represents the duty ratio range over which in excess
of 99% of the maximum transferred power is obtained. This
region widens toward the low output voltage levels, as can also
be seen in Fig. 15. The modes of the duty ratio measurements
fall within the 99% power region for most of the output voltage
points, which shows that the control circuit can effectively track
the optimum power point.
During the measurements of the duty ratio samples, the energy
harvesters generated power, the transferred power, and the quiescent power overhead of the power conditioning system were
also recorded. The 100 measurement points were averaged, to
take into account that the duty ratio ranges around an optimum
value. These values are plotted in Fig. 17 against output volt-

209

Fig. 17. Average power obtained from 100 measurements per output voltage.
Output voltage is regulated by an adjustable shunt regulator. Constant frame
acceleration of 3.75 ms2 at 43.8 Hz. P m a x is the maximum extractable
harvester power under optimum load conditions.

age, along with the maximum extractable power from the energy
harvester under optimum load conditions for comparison.
The difference between the measured generated power and
the transferred power are due to a combination of three major
loss mechanisms: 1) a nonideal power conversion process; 2)
quiescent power overheads of the control and ancillary circuits;
and 3) the conduction loss in the shunt resistor used for the
monitoring of the output current.
The ratio between the generated power of the nonideally
loaded harvester and the maximum potentially extractable
power under optimum load conditions is referred to as the utilization factor. The utilization of the energy harvester peaks
above 89% at 1.8 V output, and remains over 86% over the entire output voltage range. Less than 100% utilization is primarily
due to the increased conduction losses within the coil that result
from switching frequency current ripple. The power conversion
efficiency is calculated as the ratio of the useful output power
(Pstore ) of the converter to the generated power of the harvester,
while the overall system effectiveness is defined as the useful
output power normalized to the maximum extractable power
from the energy harvester under optimum load conditions. The
conversion efficiency is at its maximum of 76.5% at low voltage
levels where the quiescent power overhead is at its minimum,
dropping down to 66% at 4.5 V. The peak overall effectiveness reaches almost 70%, which when compared against the
maximum of 70.7% recorded under the steady-state characterization (see Fig. 15) shows a highly effective control.
B. Transient Response to Step Change in Excitation and
Output Voltage
The dynamic performance and stability is evaluated by
recording the transient behavior of the power conditioning circuit, monitored by the duty ratio of the converter, in response
to a step change in the frame excitation magnitude and in the
output voltage under the worst-case considerations.

210

Fig. 18. Output current and duty ratio in response to a step change in excitation
magnitude from 3 to 4 ms2 at a constant 2.5 V output. The frequency is
constant at 43.8 Hz. The current is inferred from the output of the current-sense
amplifier circuit.

First, a step change in the acceleration magnitude from 3


to 4 ms2 is considered (see Fig. 18) potentially providing
twice the generated power. The output voltage is maintained at
a constant 2.5 V for the test. The increase in excitation amplitude results in increased generated power and voltage, which,
in turn, affects the emulated resistance of the power converter.
The control adjusts the duty ratio of the rectifier, and is seen to
reestablish optimum damping conditions within around 4.5 s.
At the instant of the step change in the frame acceleration magnitude (just before the 2 s mark), the available generated power
and thus the useful output power increases at a rate that is dominated by the mechanical response of the energy harvester, i.e.,
the time required by the tip mass to build up its momentum.
The duty ratio of the power converter, as a means of monitoring the control trajectory of the algorithm, is also presented
in Fig. 18. The results show the clearly distinguishable duty
ratio steps that characterize the digital implementation of the
perturb-and-observe algorithm, and the tracking around a new
optimum point.
A worst-case step change in the load conditions is also considered in this study: a rapid discharge of the energy storage
element from its maximum rated output voltage of 4.5 to 2 V, as
shown in Fig. 19. This situation mimics the burst release of energy from the storage that may occur as a result of the execution
of a power hungry task by the load electronics. Small dynamic
variations in the power consumption of the load do not affect
the output power performance of the power conditioning circuit
as the 68 mF supercapacitor provides a buffer between the load
and the power converter that smoothes these changes.
The output current waveform is also presented in Fig. 19.
During the discharge of the output capacitor, the output of the
current-sense amplifier is saturated, and then the current starts
to increase as a result of the changing duty ratio of the power
converter. The duty ratio drops from around 84% to below 70%
and converges to a new optimum condition in approximately
8.5 s after the start of the discharge.

IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 29, NO. 1, JANUARY 2014

Fig. 19. Rapid discharge of the 68 mF supercapacitor from its maximum


rated voltage of 4.5 to 2 V via an adjustable shunt regulator (LM4041), thus
mimicking a burst release of the stored energy for the load application.

Fig. 20. Charging of a 68-mF supercapacitor over 500 s, showing zero-energy


start-up and active MPTT. The overall system efficiency is normalized to the
maximum extractable power of 870 W. Measurements were taken at constant
frame acceleration of 3.75 ms2 at 43.8 Hz.

This reaction time represents the worst-case scenario as it


corresponds to the largest optimum duty ratio difference that
can occur under normal operation conditions and is a function
of the maximum rate of change of the duty ratio of the control
circuit and, thus, of the calculated settling time. An energy
harvesting system with smaller tip mass, for example, would
have a smaller settling time and the control could be set to react
more quickly to rapid changes in the environmental conditions
or in the power drawn by the load electronics.
C. Overall System
The last test is aimed to present the capability of the power
conditioning system to start-up from zero-energy conditions
and then track the maximum power transfer point actively while
charging a 68 mF supercapacitor.
At a constant excitation of 3.75 ms2 , the charging of the
supercapacitor is recorded over 500 s (see Fig. 20) along with

SZARKA et al.: MAXIMUM POWER TRANSFER TRACKING FOR ULTRALOW-POWER ELECTROMAGNETIC ENERGY HARVESTERS

211

TABLE I
KEY SYSTEM METRICS

Fig. 21. Comparison of charge up times of a 68 mF supercapacitor using: 1)


passive voltage quadrupler circuit; 2) full-wave boost rectifier in an open-loop
with constant duty ratio of = 0.67; and 3) full-wave boost rectifier with MPTT
control. Excitation is regulated to be 3.75 ms2 at 43.8 Hz.

the duty ratio (inferred from the reference voltage of the control
circuit). The measured output voltage and useful output current
of the power converter are used to calculate the transferred
power. The ratio of this to the maximum extractable power under
optimum load conditions provides the overall system efficiency
over (see Fig. 20).
The passive quadrupler circuit provides zero-energy startup, charging the supercapacitor to 1.85 V at which point the
active boost rectifier circuit starts to operate. The duty ratio,
starting from a high initial value, quickly finds and tracks the
optimum power transfer point over the charging of the capacitor.
The overall efficiency is highest at low output voltages with an
average value of close to 70%, as presented in Section V-A.
The recorded instantaneous efficiency may exceed this due to
the inertial effects of the mechanical structure resulting in short
burst of power when the damping is increased.
Fig. 21 presents the transient accumulation of energy in the
68 mF supercapacitor when charged by differing power conditioning solutions: 1) passive voltage quadrupler; 2) the full-wave
nonsynchronous boost rectifier in open loop with a constant duty
ratio of = 0.67; and 3) full-wave boost rectifier with MPTT
control. The start-up phase is common for all approaches, provided by the passive voltage multiplier circuit, during which the
capacitor is charged from 0 to 1.85 V.
The results illustrate that the overall output power gain of the
system when MPTT control is employed: The total charge-up
time, corresponding to 0.5 J of energy stored, is reduced by
26.5% and by 8.6% as compared to the passive circuit and the
open-loop systems, respectively. A summary of the key system
metrics is presented in Table I.
VI. CONCLUSION
The work presented in this paper aimed to address the challenges that arise from implementing MPTT for low-power, kinetic electromagnetic energy harvesters. The transient response
of the single-degree-of-freedom mechanical system is presented

and discussed using experimental results and analytical derivations. A method that aids the design of perturb-and-observe
algorithm-based control with discrete perturbations of the control parameter is presented: the minimum time required between
perturbations in order to allow the mechanical structure to settle is calculated for highly underdamped massspringdamper
systems under the assumption of a constant, sinusoidal, nondirect excitation that occurs at the natural resonance frequency of
the mechanical structure.
A complete power conditioning circuit for a low voltage, submilliwatt electromagnetic energy harvester has been presented
that requires no external power and is capable of self-starting
from zero-energy conditions. In contrast with previous work,
the transferred power is maximized instead of the generated
power, thus accounting for the losses suffered during the power
conversion process and the due to the quiescent power overheads. Good peak power tracking effectiveness is demonstrated
despite of the lack of accurate regulation of the apparent input
impedance of the boost rectifier, thus allowing the use of a slow
feedback control, and consequently, a reduced quiescent power
implementation. The total power consumption of the power conditioning system is 44 W at 2 V output.
Steady-state and transient measurements show that the system
is stable and capable of tracking the maximum power transfer
point by optimizing the duty ratio of the PWM signals of the
boost converter over the entire output voltage range. A harvester
utilization of up to 89% is achieved. Overall system effectiveness up to 70% is recorded, corresponding to approximately
600 W of useful output power. The transient responses show
that the calculated settling time provides a reasonable choice for
the minimum time between perturbations. The power converter
topology used in this study can only synthesize resistive load
conditions, limiting the maximum achievable utilization when
the harvester is not in resonance as the optimum load required

212

IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 29, NO. 1, JANUARY 2014

can have a significant reactive component. A natural extension


of the work would be to apply MPPT control to converters
capable of synthesizing complex load impedances.
REFERENCES
[1] B. H. Stark, G. D. Szarka, and E. D. Rooke, Start-up circuit with low
minimum operating power for microwatt energy harvesters, IET Circuits,
Devices Syst., vol. 5, no. 4, pp. 267274, 2011.
[2] G. D. Szarka, B. H., Review of power conditioning for kinetic energy
harvesting systems, Trans. Power Electron., vol. 27, no. 2, pp. 803815,
2012.
[3] G. D. Szarka, P. P. Proynov, B. H. Stark, S. G. Burrow, and N. McNeill,
Experimental investigation of inductorless, single-stage boost rectification for sub-mW electromagnetic energy harvesters, in Proc. Int. Symp.
Low Power Electron. Design (ISLPED), Fukuoka, Japan, 2011, pp. 361
366.
[4] N. G. Stephen, On energy harvesting from ambient vibration, J. Sound
Vibration, vol. 293, pp. 409425, 2006.
[5] S. G. Burrow and L. R. Clare, Open-loop power conditioning for vibration energy harvesting, IEEE Electron. Lett., vol. 45, no. 19, pp. 999
1000, Sep. 2009.
[6] G. D. Szarka, S. G. Burrow, and B. H. Stark, Ultra-low power, fullyautonomous boost rectifier for electromagnetic energy harvesters, vol. 28,
pp. 33523362, Jul. 2013.
[7] C. B. Williams, C. Shearwood, M. A. Harradine, P. H. Mellor, T. S. Birch,
and R. B. Yates, Development of an electromagnetic micro-generator,
IEE Proc. Circuits Devices Syst., vol. 148, no. 6, pp. 337342, 2001.
[8] Y. Kim, C. Naehyuck, W. Yanzhi, and M. Pedram, Maximum power transfer tracking for a photovoltaic-supercapacitor energy system, in Proc.
2010 ACM/IEEE Int. Symp. Low-Power Electron. Des., Austin, TX, USA,
2010, pp. 307312.
[9] A. S. Weddell, G. V. Merrett, and B. M. Al-Hashimi, Ultra low-power
photovoltaic MPPT technique for indoor and outdoor wireless sensor
nodes, in Proc. Design, Autom. Test Europe, Grenoble, France, Mar.
1418, 2011, pp. 905908.

[10] J. Elmes, V. Gaydarzhiev, A. Mensah, K. Rustom, and J. Shen, Maximum


energy harvesting control for oscillating energy harvesting systems, in
Proc. IEEE Power Electron. Specialists Conf., Orlando, FL, USA, 2007,
pp. 27922798.
[11] T. T. Toh, P. D. Mitcheson, A. S. Holmes, and E. M. Yeatman, A continuously rotating energy harvester with maximum power point tracking, J.
Micromech. Microeng., vol. 18, no. 10, 104008, 2008, doi: 10.1088/09601317-18-/10/104008.
[12] T. Paing, J. Shin, R. Zane, and Z. Popovic, Resistor emulation approach to
low-power RF energy harvesting, IEEE Trans. Power Electron., vol. 23,
no. 3, pp. 14941501, May 2008.
[13] G. K. Ottman, H. F. Hofmann, and G. A. Lesieutre, Optimized piezoelectric energy harvesting circuit using step-down converter in discontinuous
conduction mode, IEEE Trans. Power Electron., vol. 18, no. 2, pp. 696
703, Mar. 2003.
[14] C. K. Lee, P. K. Chan, and S. W. Or, Loosely power flow control scheme
for piezoelectric energy harvesting, IEEE Electron. Lett., vol. 46, no. 25,
pp. 16891691, Dec. 2010.
[15] J. Leicht, D. Maurath, and Y. Manoli, Autonomous and self-starting efficient micro energy harvesting interface with adaptive MPPT, buffer monitoring, and voltage stabilization, in Proc. ESSCIRC, Bordeaux, France,
2012, pp. 101104.
[16] D. Maurath, P. F. Becker, D. Spreeman, and Y. Manoli, Efficient energy harvesting with electromagnetic energy transducers using active
low-voltage rectification and maximum power point tracking, IEEE J.
Solid-State Circ., vol. 47, no. 6, pp. 112, Jun. 2012.
[17] R. Dayal and L. Parsa, Low power implementation of maximum energy
harvesting scheme for vibration-based electromagnetic microgenerators,
in Proc. IEEE Appl. Power Electron. Conf. Expo., Fort Worth, TX, USA,
2011, pp. 19491953.
[18] J. Karki. (2000, Oct.). Active Low-Pass Filter Design [Online]. Available:
http://www.science.unitn.it/bassi/Signal/TInotes/sloa049.pdf
[19] Microchip Technology, Inc. (2008). [Online]. Available MCP6031
Datasheet: http://ww1.microchip.com/downloads/en/DeviceDoc/22041b.
pdf
Authors photographs and biographies not available at the time of publication.

Вам также может понравиться