PROVAS DE DISSERTAÇÃO DO MIEEC—OUTUBRO DE 2015 1

Performance Improvement of a Buck Converter

Abstract—This dissertation aims at developing a predictive A generalized method that uses the inductor current slopes
current control algorithm for a synchronous buck converter using equations to control a buck converter in [3],[5]. Results show
a Kalman filter algorithm that has no need for current sensing. a step response of 200 us and a current error elimination in
This method requires a model of the system, its application in
the Kalman filter and the current loop tuning, which will be two switching cycles. An Extended Kalman Filter is used in
described and the algorithm will be implemented in a state-of-the- [4] as a current observer for a boost converter, proposing a
art hardware provided by Infineon. Results show that this method load estimation algorithm to improve the results. An accuracy
provides better results than standard methods when Gaussian of 5 % is achieved in steady state while in steady state the
white noise is present in the output voltage, maintaining a good load error is never bigger than 10 %.
step response time.
Index Terms—Synchronous buck converter, Kalman filter,
predictive current control, DC-DC converter, microcontroller,
II. EKF BASED P REDICTIVE C URRENT C ONTROL
white noise A LGORITHM
Analysing a synchronous buck converter circuit, the equa-
tions during ton can be written in state space form as:
I. I NTRODUCTION
WITCHING mode power supplies are widely used in
S
"   #
RC Rout
− RLL + − L (RR out
 1
power management systems due to the high efficiency L (Rout +RC ) out +RC )
iL (t)
+ L vin(1)
(t)
that they can achieve. When comparing the two methods of
Rout
− C (Rout1+RC ) vC (t) 0
C (Rout +RC )
controlling an SMPS, i.e. voltage-mode and current control,
Writing the same equations for the buck converter during
the latter offers an easier compensation loop implementation
tof f and merging both using a state-space average method, it
and faster response to load changes. However, a current sensor
is possible to write a discrete averaged state space transition
is required, at the inductor or the input. This sensing scheme
matrix for the buck converter given by:
increases the overall bill of materials of the system, the  
" #
converter footprint, and makes the system more sensitive to 1 − RLL + L (RRC Rout
Ts − Rout Ts
iL (k)

out +RC ) L (Rout +RC )
several noise sources. .
(2)
Rout Ts
1− Ts vC (k)
The objective of this dissertation is to implement a pre- C (Rout +RC ) C (Rout +RC )

dictive current control algorithm to address the mentioned After the model is derived, the extended Kalman filter must
problems. This algorithm will not need a current sensing be designed and tuned. Adding to the system model derived
circuit and will estimate the inductor current value measuring above, an equation for the measurement output needs to be
the input and output voltages, merging this data with the circuit derived. Since the measured variable is the output voltage
equations, trough a Kalman Filter. Vout and it depends on the system states, iL and vC the
Many authors addressed the subject of predictive algorithms measurement equation will be:
for DC-DC converters, either using Kalman filtering or simply
RC Rout (k) RC
modelling the slopes of the inductor current. A model for vout (k) = iL (k)+vC (k)− vC (k).
the variations of a boost converter parameters is presented in Rout (k) + RC Rout (k) + RC
(3)
[1] and experimental results show that the control algorithm
Finally after the model of the system and the Kalman
achieves better results than a voltage mode control algorithm
Filter tuning, the control loop needs to be designed. This loop
only if the input voltage is feedforwarded.
consists in two parts: the voltage loop and the current loop.
Two different methods of controlling DC-DC converters are
The voltage loop is a simple PI controller, while the current
presented in [2]: in the first the inductor current is derived from
loop takes into account the inductor current slopes, defined
the system known or measured variables, such as the duty
by:
cycle and the input voltage and output voltage; the second
one considers the whole system model and is implemented vin (k) − vout (k) − RL iL (k)
s1 (k) = (4)
using Extended, showing better accuracy and noise immunity L
in the current estimation. vout (k) + RL iL (k)
s2 (k) = . (5)
1 Student,
L
FEUP, [email protected]
2 AssistantProfessor at DEEC, FEUP, [email protected] The inductor current rises during the on time of the buck
3 Concept Engineer, Infineon Technologies AG, [email protected] converter, i.e., during δ and falls during the off time, or 1 − δ.
PROVAS DE DISSERTAÇÃO DO MIEEC—OUTUBRO DE 2015 2

Variations on the current in two consecutive switching cycles the voltage takes about 200 µs to stabilize, showing the ability
can be derived using (4) and (5), giving of the algorithm to correct the load change and maintain a
constant voltage of 1.8 V.
iL (k + 1) − iL (k) = s1 (k) δ Ts − s2 (k) (1 − δ) Ts . (6)
The next experimental test was done to evaluate the line
Solving (6) in order to the duty cycle represents the value regulation capabilities of the proposed algorithm. To perform
that has to be given to the PWM in order to reach the desired this evaluation the buck converter board was supplied with a
level of current. Assuming this fact, iL (k +1) is the next value DC power supply, were the input voltage was intentionally
of the current, i.e, the value that the controller has to achieve changed to evaluate the algorithm response. As stated in the
– the reference value of the current IREF given by the voltage beginning of the section, the input voltage of this synchronous
control loop. The duty cycle that adjusts the current error can buck converter is usually 12 V, but in this test the voltage
be calculated by at some point of the simulation intentionally dropped down
to 10 V and the results confirm that the algorithm is able to
IREF − iL (k) + s2 (k) Ts
δ= . (7) adapt to input voltage changes, showing a response time of
(s1 + s2 ) Ts
about 100 µs.
III. I MPLEMENTATION
IV. C ONCLUSION
In order to prove the theory the algorithm has been de-
veloped in a XMC4200 microcontroller unit and tested on a The objective of this work was improving the performance
prototype board with the characteristics presented in table I. of a synchronous buck converter using a KF. The nonlinear
version of KF, EKF, can be used to design a predictive current
Input Voltage 15 V control loop that do not need current sensors, reducing also the
Output Voltage 1.8 V impact of the noise in the control loop. The developed method
Maximum Output Current 5A is does not need current sensing, reducing then the bill of ma-
Switching Frequency 100 kHz terials of the synchronous converter, achieving an estimation
error or approximately 1 % for the current estimation. Using
Table I: Buck converter design specifications. an extended Kalman filter approach, it is possible to reduce
the influence of Gaussian white noise in the output voltage,
The voltage and current loop, as well as reading conversion while maintaining a good transient response of the algorithm.
values and updating the duty cycle take 8 µs, which added
to the EKF execution time (14 µs) make a total of (22 µs), R EFERENCES
meaning that the whole control algorithm will run every third
[1] P. Mattavelli, “Digital control of DC-DC boost converters with inductor
switching cycle. current estimation,” in IEEE Nineteenth Annual Applied Power Electron-
In the first experiment a sweep load was applied to evaluate ics Conference and Exposition (APEC’2004), vol. 1, 2004, pp. 74–80.
the accuracy of the estimation in steady state. The applied [2] A. Beccuti, S. Mariethoz, S. Cliquennois, S. Wang, and M. Morari,
“Explicit model predictive control of dc-dc switched-mode power sup-
sweep was form 0.5 Ω to 2 Ω with a step of 0.1 Ω and the plies with extended kalman filtering,” IEEE Transactions on Industrial
results are depicted in Fig. 1. As can be observed, both current Electronics, vol. 56, no. 6, pp. 1864–1874, June 2009.
and load estimations are below 1 %, proving the accuracy of [3] M. Run, C. Chen, Z. X. Dong, and Z. X. Cheng, “Optimal current
observer for predictive current controlled buck dc-dc converter,” in Inter-
the implemented algorithm for different loads. national Symposium on Computer, Consumer and Control (IS3C’2014),
Jun 2014, pp. 896–899.
[4] Q. Zhang, Q. Tong, and H. Zhang, “An inductor current observer based
1.00 0.24
iL
on improved EKF for DC/DC converter,” in International Symposium on
Computer, Consumer and Control (IS3C’2014), Jun 2014, pp. 892–895.
Estimated Inductor Current Error (%)

0.95 R 0.22
[5] Q. Zhang, Q. Tong, H.Zhang, IEEE International Conference on Com-
puter Vision, Object Recognition fromLocal Scale-Invariant Features,
Estimated Load Error (%)

0.90 0.20
1999.Y. Qiu, H. Liu, and X. Chen, “Digital average current-mode control
0.85 0.18 of PWM DC-DC converters without current sensors,” IEEE Transactions
on Industrial Electronics, vol. 57, no. 5, pp. 1670–1677, May 2010.
0.80 0.16

0.75 0.14

0.70 0.12

0.65 0.10

0.60 0.08
0.5 1 1.5 2
Load (+)

Figure 1: Accuracy of the inductor current and load resistor

To evaluate the overall performance of the buck converter a