AUTOMATIC TUNING OF A PROPORTIONAL-INTEGRAL-DERIVATIVE CONTROLLER FOR DC-DC BUCK SWITCHING POWER CONVERTER

Abstract

A method of automatically tuning a proportional-integral-derivative (PID) controller (102) controlling a DC-DC converter (100) includes substituting the PID controller (102) with a modified relay feedback test (MRFT) (104) controller having a threshold parameter (β) and a magnitude parameter (h), iteratively determining a current error signal (e [k]) by measuring an output voltage (Vo[k]) of the DC-DC converter (100) and comparing it to a desired reference output voltage (Vo-ref), input the error signal (e[k]) to the MRFT (104), determining an output value (u[k]) of the MRFT, update the duty cycle (D) based on the output value (u[k]), operating the DC-DC converter (100) using the updated duty cycle (D), exciting oscillations in the loop containing the MRFT (104) and the DC-DC converter (100), measuring a frequency (Ω0) and amplitude (α0) of the output voltage (Vo[k]), updating PID controller parameters based on the frequency (Ω0) and amplitude (α0), and substituting the MRFT (104) with the PID controller (102).

Description

FIELD OF INVENTION

The present invention relates to systems and methods for automatically tuning a proportional-integral-derivative (PID) controller for DC-DC buck switching power converter.

BACKGROUND

Fueled by the development of low-cost microcontrollers with high clock speeds, the autotuning of digitally-controlled DC-DC switched-mode converters has attracted the attention of researchers in the last two decades. Even though analog controllers are still faster and cost less, the digital controller offers some unmatched advantages such as the immunity to controller parameter variation, the ease of retuning the controller, and the ability to implement highly sophisticated control algorithms

Autotuning in the present context refers to the automatic online tuning of a digital controller at pre-set intervals or in response to a user command or a triggering event. Because components in converters have a tolerance and so their exact values are unknown, and because they are prone to ageing, the ability of a controller to retune itself based on actual system parameters rather than a nominal mathematical model can largely improve the stability and dynamic performance of a converter. Autotuning generally includes a test stage in which the system dynamics are excited, and a tuning stage in which the controller is retuned using measurements from the test. Some autotuning methods use an intermediate stage of system identification.

The relay feedback test (RFT) is classically known in the field of control. It involves placing a relay in series with the plant or process, and operating the system in closed-loop, which leads to limit-cycle oscillations occurring at the phase cross-over)(−180°) frequency of the plant. The amplitude and time period of the oscillations are measured and used for tuning a PID controller, for example, using the Ziegler and Nichols (Z-N) PID tuning rules. While this usually yields satisfactory performance, stability is not guaranteed, nor are gain or phase margin be specified. A form of RFT-based autotuning of a DC-DC buck converter has been previously implemented. However, different from the conventional RFT, an integrator is placed in series with the RFT relay. This forces oscillations at a frequency corresponding to a −90° phase shift, thereby identifying the LC resonant frequency of the converter. The first PID zero is placed at the resonant frequency, while the other zero is iteratively tuned until the desired phase margin is met. The PID gain is finally tuned (iteratively) to achieve the desired bandwidth. A disadvantage of this method is that since oscillation amplitude is large at the resonant frequency, the test is performed at a voltage below the nominal, during the startup ramp. This means autotuning must be performed during regular operation. Also, the tuning process is somewhat complex and long as it consists of several stages, some with iterations. In other prior implementations, an improvement to the same RFT-based work has been developed in which only time-period measurements are used in order to avoid errors related to amplitude measurement. In other prior art, limited-cycle oscillations are induced in a DC-DC buck converter by temporarily reducing the resolution of the digital PWM rather than using a relay, which allows for continued regulation of the system even during the test stage, although with poorer tracking.

Several other autotuning methods using tests besides RFT have also been implemented, for example, methods that use the injection of a pseudo-random binary sequence (PRBS) in the test stage. But almost all known methods, whether based on the RFT, PRBS injection, or others, have some level of complexity, mainly because the full burden of test and tuning is to be done online. One way to simplify the test and tuning stages, and thus make them more suitable for practical implementation, is to utilize general knowledge of the system's behavior to do part of the tuning preparation offline. The modified relay feedback test (MRFT) autotuning method is one such method. It is built on a coordinated test and tuning concept that guarantees a specified gain margin or phase margin. Further, the MRFT autotuning method can also be extended to include optimized non-parametric tuning, which in addition to the guaranteed gain or phase margin, provides near-optimal transient performance for a wide range of designs of a given system. The MRFT autotuning method consists of a single-step test and a single-step tuning stage, which are described below.

TEST STAGE: A general block diagram of the test stage of the MRFT autotuning method is shown in FIG. 1a, where W_p(s) is the transfer function of the plant to be controlled. The logic of the MRFT block is given below, where e(t) is the error signal and u(t) is the control command produced by the MRFT block. u(t−) is the control output immediately prior to the time t. b₁ and b₂ are the relay switching conditions, given by b₁=−βe_min and b₂=βe_max, where e_min and e_max are the last (negative) minimum and last (positive) maximum of e(t), respectively, and where β is a “threshold parameter” of the test set by the user in the range −1<β<1.

$\begin{array}{cc}u\left(t\right)=\{\begin{array}{cc}& \left\{e\left(t\right)\ge b_1\&u\left(t-\right)=-h\right\}\\ +h,\mathrm{if}& \mathrm{or}\\ & \left\{e\left(t\right)\ge -b_2+u\left(t-\right)=+h\right\}\\ & \\ & \left\{e\left(t\right)\le -b_2\&u\left(t-\right)=+h\right\}\\ -h,\mathrm{if}& \mathrm{or}\\ & \left\{e\left(t\right)\le b_1\&u\left(t-\right)=-h\right\}\end{array}& \mathrm{Eq}.\left(1\right)\end{array}$

h is the “MRFT magnitude”, which is the amount of the control effort provided by the MRFT block to the process, or in other words the command provided to the actuator. The test stage of MRFT autotuning method is performed when the subject system is in steady-state. FIG. 1b shows illustrative waveforms of the test stage. e_max and e_min are initially set to zero, making b₁=b₂=0. Oscillations start to develop in the output y(t), and consequently in error signal e(t). Every time a minimum (e_min) or maximum (e_max) is recorded, the switching condition (b₁ or b₂) for the upcoming half-cycle is updated. After a few transient cycles, the oscillations stabilize, and |e_max| and |e_min| are equal. This is when amplitude and frequency measurements are taken. Let Ω₀ (in rad/s) be the frequency of the resulting oscillations and let a₀=e_max=−e_min be the amplitude of the oscillations. T_u=2π/Ω₀ is called the ultimate time period. For improved accuracy, T_u and a₀ are best computed as an average over a few oscillation cycles. The ultimate gain K_u refers to the gain from the fundamental component of e(t) to the fundamental component of u(t). In other words, it is the ratio of the amplitudes of the first harmonics of u(t) and e(t) and is calculated using the equation below.

$\begin{array}{cc}{K}_u=\frac{4h}{\pi a_0}& \mathrm{Eq}.\left(2\right)\end{array}$

TUNING STAGE: A PID controller of the format

$W_c\left(s\right)=K_c\left(1+\frac{1}{T_{i}s}+T_{d}s\right)$

is used in the MRFT autotuning method. The tuning stage is to simply calculate the updated PID parameters using the following equations, where c₁, c₂, c₃ are predetermined constants:

K_c=c₁K_u, T_i=c₂T_u, T_d=c₃T_u   Eq. (3)

SUMMARY

According to one or more aspects of the present disclosure, a method of automatically tuning a digitally-implemented proportional-integral-derivative (PID) controller for the purpose of controlling a DC-DC buck switching power converter includes the steps of

• a) Using a PID controller of the form

$W_c\left(s\right)=K_c\left(1+\frac{1}{T_{i}s}+T_{d}s\right),$

where s is the Laplace variable, K_c is the proportional gain, T_i is the integral time, and T_d is the derivative time, to drive the system to a steady state;

• b) substituting the PID controller with a modified relay feedback test (MRFT) block;
• c) the input to the MRFT block is an error signal (e[k]), which is obtained by measuring an output voltage (V_o[k]) of the DC-DC buck switching power converter and comparing it to a desired reference output voltage (V_o-ref);
• d) the output of the MRFT block is the new modulation index that is input to the pulse-width modulator (PWM); the MRFT block output is set to either D+h or D−h, where D is the nominal duty cycle, i.e., the modulation index that is normally provided by the PID controller to the pulse-width modulator (PWM) to produce an output voltage equal to the set point (V_o-ref); This is in contrast to the prior implementations of the MRFT in prior art, where the output of the MRFT block is only +h or −h.
• e) repeating steps a) through c) until a number of cycles (N_cycles) of sustained oscillations are produced;
• f) measuring the frequency (Ω₀) and amplitude (a₀) of the sustained oscillations resulting from the application of the MRFT, where a₀ is the incremental amplitude around the steady-state output voltage; for improved accuracy, Ω₀ and a₀ are taken as an average over several cycles of MRFT oscillations;
• g) updating the PID controller parameters based on the measured Ω₀ and a₀; using the following equations

$K_c=c_1\frac{4h}{\pi a_0},T_i=c_2\frac{2\pi }{{\Omega }_0},T_d=c_3\frac{2\pi }{{\Omega }_0};$

and

• h) replacing the MRFT block back with the PID controller.

In one or more embodiments of the method according to any one of the previous paragraphs, the threshold parameter of the MRFT block is given by β=−0.2±5%.

In one or more embodiments of the method according to any one of the previous paragraphs, c₁=0.69±5%, c₂=1.14±5%, and c₃=0.19 ±5%.

According to one or more aspects of the present disclosure a non-transitory computer readable storage medium has data stored therein representing software executable by a computer. The software includes instructions to switch a controller circuit of a DC-DC buck switching power converter from a proportional-integral-derivative (PID) controller to an MRFT block; determine an error signal (e[k]) by measuring an output voltage (V_o[k]) of the DC-DC buck switching power converter and comparing it to a desired reference output voltage (V_o-ref); input the error signal (e[k]) to the MRFT block; depending on the value of e[k], the MRFT block will output either D+h or D−h, where D is the nominal duty cycle described earlier and h is the MRFT magnitude; repeat the preceding instructions until a predetermined number of oscillation cycles (N_cycles)have developed; measure a frequency (Ω₀) and amplitude (a₀) of the oscillations as an average over the N_cycle oscillations, but disregarding an initial number of cycles (N_discard); update the PID controller parameters based on Ω₀ and a₀ using the equations

$K_c=c_1\frac{4h}{\pi a_0},T_i=c_2\frac{2\pi }{{\Omega }_0},T_d=c_3\frac{2\pi }{{\Omega }_0};$

and finally switch the controller circuit of the DC-DC buck switching power converter from the MRFT block back to the (updated) PID controller.

In one or more embodiments of the non-transitory computer readable storage medium according to any one of the previous paragraphs, the software further includes instructions to continuously track the error signal (e[k]) and record every occurring maximum (e_max) and minimum (e_min).

In one or more embodiments of the non-transitory computer readable storage medium according to any one of the previous paragraphs, the software further includes instructions to measure the frequency (Ω₀) and amplitude (a₀) of the oscillations as an average based on a number of recorded cycles (N_MRFT)—which equals a total number of cycles (N_cycles) minus the initial number of cycles (N_discard)—and an elapsed time period (T_MRFT) of the number of recorded cycles (N_MRFT); Ω₀ and a₀ are then calculated using

${\Omega }_0=\frac{2\pi }{\frac{T_{MRFT}}{N_{\mathrm{MRFT}}}}\mathrm{and}{a}_0=\frac{{\sum }_{N_{MRFT}}{e}_{\max }+{\sum }_{N_{MRFT}}\left|e_{\min }\right|}{2N_{MRFT}}.$

In one or more embodiments of the non-transitory computer readable storage medium according to any one of the previous paragraphs, the software further includes instructions to update the PID controller parameters based on the amplitude (a₀) where

$a_0=\frac{{\sum }_{N_{\mathrm{MRFT}}}{e}_{\max }+{\sum }_{M_{\mathrm{MRFT}}}\left|e_{\min }\right|}{2N_{MRFT}}.$

In one or more embodiments of the non-transitory computer readable storage medium according to any one of the previous paragraphs, the software further includes instructions to calculate a proportional gain (K_c), T_i an integral time (T_i), and a derivative time (T_d) for the PID controller, where:

$K_c=c_1\frac{4h}{\pi a_0},T_i=c_2\frac{2\pi }{{\Omega }_0},T_d=c_3\frac{2\pi }{{\Omega }_0}.$

In one or more embodiments of the non-transitory computer readable storage medium according to any one of the previous paragraphs, the non-transitory computer readable storage medium has data stored therein representing a value of C₁ as 0.69±5%, a value of c₂ as 1.14±5%, and a value of c₃ is 0.19±5%.

In one or more embodiments of the non-transitory computer readable storage medium according to any one of the previous paragraphs, the non-transitory computer readable storage medium has data stored therein representing a value of β as −0.2±5%.

DESCRIPTION OF THE DRAWINGS

The present invention will now be described, by way of example with reference to the accompanying drawings, in which:

FIG. 1a shows a schematic block diagram of the MRFT (which represents the test stage of the MRFT autotuning method) according to the prior art;

FIG. 1b shows illustrative waveforms of the operation of the MRFT according to the prior art;

FIG. 2 shows a flowchart for an implementation of the MRFT test stage on a DC-DC switching buck converter according to some embodiments;

FIG. 3 shows a schematic block diagram of the MRFT block with the DC-DC switching buck converter according to some embodiments;

FIG. 4 shows a graph of oscillations resulting from the application of the MRFT to a DC-DC switching buck converter, using scales of 20 mV/division for the vertical axis and 200 μs/division for the horizontal axis, according to some embodiments;

FIG. 5 shows a graph of oscillations resulting from the application of the MRFT to a DC-DC switching buck converter overlaid on the relay status according to some embodiments;

FIG. 6a shows the transient response of the output voltage of a DC-DC switching buck converter due to a step in the commanded set-point voltage from 2V to 2.2V, where the converter has an inductance value of 10 μH and a capacitance value of 660 μF, and where the PID controller of the converter is tuned with the MRFT autotuning method, according to some embodiments;

FIG. 6b shows a transient response of the output voltage of a DC-DC switching buck having the same inductance and capacitance values as in FIG. 6a and for the same step in set-point like in FIG. 6a, but where the PID controller of the converter, unlike FIG. 6a, is tuned using an optimal but non-autotuning method;

FIG. 7a shows the transient response of the output voltage of a DC-DC switching buck converter due to a step in the commanded set-point voltage from 2V to 2.2V, where the converter has an inductance value of 10 μH and a capacitance value of 440 μF, and where the PID controller of the converter is tuned with the MRFT autotuning method, according to some embodiments;

FIG. 7b shows a transient response of the output voltage of a DC-DC switching buck having the same inductance and capacitance values as in FIG. 7a and for the same step in set-point like in FIG. 7a, but where the PID controller of the converter, unlike FIG. 7a, is tuned using an optimal but non-autotuning method;

FIG. 8a shows the transient response of the output voltage of a DC-DC switching buck converter due to a step in the commanded set-point voltage from 2V to 2.2V, where the converter has an inductance value of 4.8 μH and a capacitance value of 660 μF, and where the PID controller of the converter is tuned with the MRFT autotuning method, according to some embodiments;

FIG. 8b shows a transient response of the output voltage of a DC-DC switching buck having the same inductance and capacitance values as in FIG. 8a and for the same step in set-point like in FIG. 8a, but where the PID controller of the converter, unlike FIG. 8a, is tuned using an optimal but non-autotuning method;

FIG. 9a shows the transient response of the output voltage of a DC-DC switching buck converter due to a step in the commanded set-point voltage from 2V to 2.2V, where the converter has an inductance value of 4.8 μH and a capacitance value of 440 μF, and where the PID controller of the converter is tuned with the MRFT autotuning method, according to some embodiments;

FIG. 9b shows a transient response of the output voltage of a DC-DC switching buck having the same inductance and capacitance values as in FIG. 9a and for the same step in set-point like in FIG. 9a, but where the PID controller of the converter, unlike FIG. 9a, is tuned using an optimal but non-autotuning method;

FIG. 10a shows the transient response of the output voltage of a DC-DC switching buck converter due to a step in the load current from 0.27 A to 1.27 A, where the converter has an inductance value of 10 μH and a capacitance value of 660 μF, and where the PID controller of the converter is tuned with the MRFT autotuning method, according to some embodiments;

FIG. 10b shows a transient response of the output voltage of a DC-DC switching buck having the same inductance and capacitance values as in FIG. 10a and for the same step in load current like in FIG. 10a, but where the PID controller of the converter, unlike FIG. 10a, is tuned using an optimal but non-autotuning method;

FIG. 11a shows the transient response of the output voltage of a DC-DC switching buck converter due to a step in the load current from 0.27 A to 1.27 A, where the converter has an inductance value of 10 μH and a capacitance value of 440 μF, and where the PID controller of the converter is tuned with the MRFT autotuning method, according to some embodiments;

FIG. 11b shows a transient response of the output voltage of a DC-DC switching buck having the same inductance and capacitance values as in FIG. 11a and for the same step in load current like in FIG. 11a, but where the PID controller of the converter, unlike FIG. 11a, is tuned using an optimal but non-autotuning method;

FIG. 12a shows the transient response of the output voltage of a DC-DC switching buck converter due to a step in the load current from 0.27 A to 1.27 A, where the converter has an inductance value of 4.8 μH and a capacitance value of 660 μF, and where the PID controller of the converter is tuned with the MRFT autotuning method, according to some embodiments;

FIG. 12b shows a transient response of the output voltage of a DC-DC switching buck having the same inductance and capacitance values as in FIG. 12a and for the same step in load current like in FIG. 12a, but where the PID controller of the converter, unlike FIG. 12a, is tuned using an optimal but non-autotuning method;

FIG. 13a shows the transient response of the output voltage of a DC-DC switching buck converter due to a step in the load current from 0.27 A to 1.27 A, where the converter has an inductance value of 4.8 μH and a capacitance value of 440 μF, and where the PID controller of the converter is tuned with the MRFT autotuning method, according to some embodiments; and

FIG. 13b shows a transient response of the output voltage of a DC-DC switching buck having the same inductance and capacitance values as in FIG. 13a and for the same step in load current like in FIG. 13a, but where the PID controller of the converter, unlike FIG. 13a, is tuned using an optimal but non-autotuning method.

DETAILED DESCRIPTION

A method of autotuning a PID controller 102 of a DC-DC switching buck converter 100 using the modified relay feedback test (MRFT) is provided. The method is comprised of two components, which are:

• • a procedure for implementing the test stage of the MRFT autotuning method on a digitally-controlled DC-DC pulse-width modulated buck converter; and
  • PID controller tuning rules that provide near-optimal performance for virtually any DC-DC buck converter. The tuning rules also provide a desired phase margin.

In the context of the DC-DC converter, a direct implementation of the original MRFT autotuning method would be to disable the DPWM and let the MRFT directly control the converter switches, so that when the MRFT block output is +h, the main switch is turned ON, and when the relay is −h, the main switch is turned OFF. But since the MRFT frequency is typically much lower than the DPWM frequency, switching at the MRFT frequency will result in an increased ripple in both the inductor current and the output voltage. The present invention adopts a more appropriate approach of double-modulation, where the DPWM is still maintained during the test stage, and the MRFT modulation is superposed on top of it. This is done by setting the MRFT block output, u(t), to either D+h or D−h instead of +h or −h, where D is the nominal duty-cycle, i.e., the modulation index that is normally provided by the PID controller to the DPWM to produce an output voltage equal to the set point (V_o-ref). In such approach, the MRFT oscillations appear superposed on top of the nominal output voltage. In terms of gate pulses, the MRFT manifests as a series of slightly wider PWM pulses corresponding to the D+h duty-cycle and lasting for T_u/2, followed by a series of slightly narrower PWM pulses corresponding to the D−h duty-cycle and also lasting for T_u/2, where T_u is the time period of the MRFT oscillations. The advantage of the adopted approach is that the converter maintains the switching frequency it was designed for even during the test stage, thus maintaining an acceptable level of ripple. This minimizes the impact of the MRFT auto-tuning on the converter, making it possible to take place even during normal operation (in most applications).

The algorithm of the MRFT block in such approach is expressed as below:

$u\left(t\right)=\{\begin{array}{cc}& \left\{e\left(t\right)\ge -\beta e_{\min }\&u\left(t-\right)=D-h\right\}\\ D+h,\mathrm{if}& \mathrm{or}\\ & \left\{e\left(t\right)\ge -\beta e_{\max }\&u\left(t-\right)=D+h\right\}\\ & \\ & \left\{e\left(t\right)\le -{\mathrm{\beta e}}_{\max }\&u\left(t-\right)=D+h\right\}\\ D-h,\mathrm{if}& \mathrm{or}\\ & \left\{e\left(t\right)\le -{\mathrm{\beta e}}_{\min }\&u\left(t-\right)=D-h\right\}\\ & \end{array}$

The MRFT tuning method can be initiated upon user prompt or programmed by the user in accordance with a certain schedule or event. The MRFT tuning method only lasts for a short duration (typically less than 1 millisecond). The inputs required by the MRFT tuning method routine are:

• • β is a threshold parameter that determines the amount of hysteresis in the modified relay;
  • h is the desired magnitude of the relay (to be added to steady-state duty-cycle);
  • N_discard is the number of initial MRFT oscillation cycles to be neglected in calculating amplitude and frequency of oscillations;
  • N_MRFT is the number of MRFT oscillation cycles to be run after the first N_discard cycles; and
  • N_cycles is the total number of MRFT oscillation cycles per single run of the test. Thus, N_cycles=N_discard+N_MRFT.

Next, the MRFT tuning method data that needs to be recorded for post-processing is as follows:

• • All minimums and maximums of an error signal for duration of the test stage of the

MRFT autotuning method; and

• • T_MRFT is the time duration for the last N_MRFT cycles only, i.e., not including the time duration of the initial N_discard cycles.
  • The complete error signal waveform for duration of the MRFT tuning method is optionally recorded.
  • The complete relay signal waveform for duration of the MRFT tuning method is optionally recorded.

The MRFT tuning method is then executed per the algorithm in the flowchart shown in FIG. 2. After the MRFT tuning method is completed, the following post-processing calculations are carried out:

$T_u=\frac{T_{MRFT}}{N_{MRFT}}$

• • is calculated. This is the time period of the MRFT oscillations (referred to as the ultimate time period). It is calculated as the average time period over the last N_MRFT cycles (i.e., not including the first N_discard cycles).

$a=\frac{{\sum }_{N_{\mathrm{MRFT}}}{e}_{\max }+{\sum }_{N_{\mathrm{MRFT}}}\left|e_{\min }\right|}{2N_{\mathrm{MRFT}}}\to$

• • is calculated. This is the average amplitude of the MRFT oscillations, calculated using the (absolute value of) all minimums and maximums over the last N_MRFT cycles.
  • A validity check on the error waveform may optionally be conducted to check if it contains many noise spikes or highly irregular oscillations. The test may be repeated if deemed necessary.

The above procedure represents the test stage, which is first of two steps of the MRFT tuning method. The second and final step is the tuning stage, which is to update the PID controller parameters using the test results. A PID controller of the following form is used, where K_c is the proportional gain, T_i the integral time, and T_d the derivative time.

$\begin{array}{cc}{W}_c\left(s\right)=K_c\left(1+\frac{1}{T_{i}s}+T_{d}s\right)& \mathrm{Eq}.\left(3\right)\end{array}$

Tuning rules of the following form are used, where coefficients c₁, c₂, and c₃, are positive constants.

$\begin{array}{cc}{K}_c=c_1{K}_u=c_1\frac{4h}{\pi a_0},T_i=c_2{T}_u=c_2\frac{2\pi }{{\Omega }_0},T_d=c_3{T}_u=c_3\frac{2\pi }{{\Omega }_0}& \mathrm{Eq}.\left(4\right)\end{array}$

According to the MRFT autotuning method, selecting c₁, c₂, and c₃, and the threshold parameter β of the MRFT in coordination with each other as per the following constraints would provide a user-specified phase margin, ϕ_m, for any arbitrary system.

$\begin{array}{cc}{c}_1\sqrt{1+{\xi }^2}=1& \mathrm{Eq}.\left(5\right)\end{array}$ $\beta =\sin \left({\varphi }_m-{\tan }^{-1}\xi \right),\mathrm{where}\xi =2\pi c_3-\frac{1}{2\pi c_2}$

Further to this, using a two-stage optimization technique, the set (β, c₁, c₂, c₃) is chosen so that it provides near-optimal transient performance for a wide range of DC-DC buck converters. Details of the optimization procedure used to obtain the optimized tuning rules is provided later. The optimized set (β*, c₁*, c₂*, c₃*), is given below.

β*=−0.2±5%, c₁*=0.69±5%, c₂*=1.14±5%, c₃*=0.19±5%  Eq. (6)

Referring to the example application shown in FIG. 3, a digitally controlled synchronous DC-DC buck converter 100, operating in voltage-control mode, is used for the implementation of the MRFT tuning method. A full schematic of the system is illustrated in FIG. 3. The output voltage (v_o) goes through a sensing circuit 106 that includes a voltage step-down circuit and an anti-aliasing filter.

A digital microcontroller, e.g., Texas Instruments TMS320F28335, is used for sampling and control. The microcontroller includes non-transitory memory containing software instructions to perform the steps of the method described herein. The microcontroller also includes transitory memory containing data generated by the method described herein.

The switching frequency (f_s=1/T_s) of the digital pulse width modulator (DPWM) 108 is set to 200 kHz. The sampling rate is also set to 200 kHz and is synchronized with the PWM. The resistive load (R_o) 110 can be electronically switched between two values to provide load transients for controller testing. The resistor R_s 112 is used to sense the inductor current (I_L) through the inductor L 114 for protection purposes. The output filter capacitance consists of a parallel combination of two larger aluminum polymer low equivalent series resistance (ESR) capacitors (together forming capacitance C₁) 115 and three smaller ceramic capacitors (together forming capacitance C₂) 116. R_L, R_C1, and R_C2 are the parasitic resistances of L, C₁, and C₂, respectively. Nominal input supply voltage (V_s) is 9 V while the set-point output voltage (V_o-ref) is 2 V.

For the first test setup, L=4.8 μH and C₁=440 μF are used (where C₁ is formed using two parallel 220 μF capacitors for lower ESR). The MRFT tuning method is performed using the optimized β in Eq. (6). The experimentally measured oscillations in v_o are shown in FIG. 4. The digital controller is programmed to discard the first few cycles when taking measurements of T_u and a₀, since as seen from FIG. 4 the oscillations stabilize after a short transient of one or two cycles. It is also noted that the amplitude of oscillations is relatively low (only ±0.04 V around the 2 V steady state, i.e., 2%). Such small magnitude of the oscillations, as well the need for only a few oscillation cycles, makes the MRFT method's effect on a converter's operation minimal The same oscillations of FIG. 4 but recorded in the microcontroller memory as contiguous digital conversion samples taken by the Analog to Digital Converter (ADC) 118 (see FIG. 3), are shown in FIG. 5. Also shown superposed in FIG. 5 is a plot of the relay's output, where +1 (relay ON) indicates that h is added to the steady-state controller output, while −1 (relay OFF) indicates that h is subtracted from the steady-state controller output. It is noted that the sampled ADC signal shown in FIG. 5, upon which the control action is based, is much cleaner than that shown in the graph of FIG. 4. This is achieved by sampling between switch transitions to avoid switching noise spikes and is also partially due to the filtering provided by the anti-aliasing low pass filter (LPF) 106.

Immediately following the first stage of the MRFT method, the new PID controller coefficients are calculated. K_u is obtained from Eq. (2) using the measured amplitude a₀, and it is used along with the frequency Ω₀=2π/T_u to calculate the PID controller coefficients using Eq. (4) and (c₁, c₂, c₃) from Eq. (6). T_u for this example application (using a DC-DC buck converter with L=4.8 μH and C₁=440 μF) was 138 μs, while the K_u based on the measured a₀ was 11.4 V⁻¹.

This same test stage of the MRFT autotuning method was repeated using the same example converter referred above but with different values of L and C₁. For each design, two tests are conducted to test the MRFT auto-tuned controller's performance In the first test, a step in V_o-ref from 2 V to 2.2 V is applied, while in the second test a step in the output current (I_o) from 0.27 A to 1.27 A is applied.

TABLE 1
Controller Parameters for MRFT Auto-
Tuned and Optimized Controllers
	Controller		Ti	Td
Design #	Type	Kc	(μs)	(μs)
1	L = 10 μH	MRFT	10.5	150	25
	C1 = 660 μF	(Adaptive)
		Optimized	7.3	193	36
		(Non-adaptive)
2	L = 10 μH	MRFT	7.9	157	26
	C1 = 440 μF	(Adaptive)
		Optimized	5.6	145	35
		(Non-adaptive)
3	L = 4.8 μH	MRFT	4.6	100	17
	C1 = 660 μF	(Adaptive)
		Optimized	4.8	132	16
		(Non-adaptive)
4	L = 4.8 μH	MRFT	3.7	96	16
	C1 = 440 μF	(Adaptive)
		Optimized	4.3	105	16
		(Non-adaptive)

For comparison, another PID controller that is tuned optimally for each of the specific converter designs is used. This optimized controller is obtained using the following procedure. Experimentally measured frequency response data of the buck converter system is obtained with the aid of a software frequency response analyzer (SFRA) tool, such as that produced by Texas Instruments Incorporated, and a high order transfer function (TF) is fitted to the measured frequency response data. Using this fitted model, a controller that gives the lowest integral of time-weighted absolute error (ITAE) (in simulation) for a step in V_o-ref for this TF is obtained through optimization using a processes of finding a minimum of an unconstrained multivariable function using derivative-free method, e.g., the fminsearch function in MATLAB® available from MathWorks, Inc. Note that this optimized controller differs from the MRFT block in that it 1) is optimized only for a given converter design and cannot be re-tuned to any other converter, and 2) does not guarantee a specific gain or phase margin.

Table 1 above lists the PID controller parameters for the MRFT blocks and the optimized controllers of each of the four tested designs. FIGS. 6a-9b show the experimental voltage output waveforms for the V_o-ref step tests, while FIGS. 10a-13b show the experimental voltage output waveforms for a step change in the load. The overall performance of the MRFT block is quite close to that of the optimized controller. Furthermore, in many of the cases where the overshoot/undershoot of the optimized controller is (expectedly) less than that of the MRFT-autotuned controller, the settling time of the latter is the same or even better, as seen in FIG. 8. Finally, the cases of high overshoot/undershoot (such as FIGS. 8a, 9a, 12a, and 13a) are associated with low damping due to the choice of a smaller inductor L.

In the remainder of this section, the optimization framework used to arrive at the optimal set of (β, c₁, c₂, c₃) is described. The model below describes a general buck converter. The term e^−τs represents the total delay due to sampling/control and the PWM operation, given by τ=1.5T_s.

$G_{buck}\left(s\right)=\frac{e^{-\tau s}}{L{C}_o{s}^2+\frac{L}{R_o}s+1}$

Let T₁=√{square root over (LC_o)} and

$T_2=\frac{L}{R_o}.$

G_buck is said to have three following situational parameters, namely T₁, T₂, and τ. The situational parameters may be reduced to two through use of the scaled Laplace variable s′=T₁s, which results in the TF below having only

$\frac{\tau }{T_1}\mathrm{and}\frac{T_2}{T_1}$

as situational parameters.

$G_{buck}\left(s\right)=\frac{e^{\frac{\tau }{T_1}{s}^{\prime }}}{s^{\prime 2}+\frac{T_2}{T_1}{s}^{\prime }+1}$

In order to obtain a realistic range of TFs covering practical buck converter designs, basic component-sizing equations are utilized. The following equation may be used to determine the inductive value L in a DC-DC buck converter:

$L=\frac{V_o\left(1-D\right)}{K_L{I}_o{f}_s}$

where V_o and I_o are the nominal output voltage and nominal output current, respectively. K_L is a coefficient related to the inductor current ripple and is typically set between 0.2 and 0.3 for a buck converter. K_L=0.3 is chosen to provide the minimum L, since L will be scaled up later. The nominal duty-cycle (D) is taken as 0.5. Noting that the ratio of V_o to I_o is equal to the load resistance (R_o), minimum inductance (L_min) is given by

$L_{\min }=\frac{0.5R_o}{0.3f_s}=\frac{5}{3}\frac{R_o}{f_s}.$

The design criteria for C_o, in the case of a low ESR, is given below where K_c is the desired maximum percent ripple in the capacitor voltage.

$C_o=\frac{1}{8f_s}\frac{K_L{I}_o}{K_C{V}_o}$

Choosing K_C as 1%, the following expression is obtained for the minimum inductance, C_o-min:

$C_{o-\min }=\frac{1}{8f_s}\frac{0.3I_o}{0.01V_o}=\frac{15}{4}\frac{1}{f_s{R}_o}=\frac{3.75}{f_s{R}_o}$

Defining scaling factors a_L for L and a_C for C_o, the situational parameters and thus the expression for G_buck may be rewritten as follows:

$T_1=\sqrt{L{C}_o}=\sqrt{{\alpha }_L{L}_{\min }{\alpha }_C{C}_{o-\min }}=\frac{5}{2}\frac{\sqrt{{\alpha }_L{\alpha }_C}}{f_s}$ $T_2=\frac{L}{R_o}=\frac{{\alpha }_L{L}_{\min }}{R_o}=\frac{5}{3}\frac{{\alpha }_L}{f_s}$ $\to \frac{\tau }{T_1}=\frac{0.6}{\sqrt{{\alpha }_L{\alpha }_C}}\mathrm{and}\frac{T_2}{T_1}=\frac{2}{3}\sqrt{\frac{{\alpha }_L}{{\alpha }_C}}$ $G_{buck}\left(s\right)=\frac{e^{-\left(\frac{0.6}{{\alpha }_L{\alpha }_C}\right)s^{\prime }}}{s^{\mathrm{\prime 2}}+\left(\frac{2}{3}\sqrt{\frac{{\alpha }_L}{{\alpha }_C}}\right)s^{\prime }+1}$

Comparing the characteristic polynomial in the denominator to the standard second order system format of (s²+2ζω_n+ω_n²), the damping factor ζ is found to be

$\frac{1}{3}\sqrt{\frac{{\alpha }_L}{{\alpha }_C}},$

while the natural frequency ω_n is 1. By considering all combinations of a_L and a_C from 1 to 10, a whole range of converters are represented. An illustration of this in grid format is provided in Table 2.

TABLE 2
Grid of TFs representing the range of buck converter designs
	αC →
αL ↓	1	2	3	4	5	6	7	8	9	10
1	1
2	2	3
3	4	5	6
4	7	8	9	10
5	11	12	13	14	15
6	16	17	18	19	20	21
7	22	23	24	25	26	27	28
8	29	30	31	32	33	34	35	36
9	37	38	39	40	41	42	43	44	45
10	46	47	48	49	50	51	52	53	54	55

Finally, to justify the approximation of neglecting the zero that appears in the transfer function when the output capacitor has a non-negligible ESR, the ESR effect is represented to some extent by increasing the damping (ζ). This is done by making sure a_L≥a_C, which results in ζ being ≥⅓. A total of 55 TFs (shown as shaded boxes in the figure below) are obtained after applying the criterion a_L≥a_C. The TFs are numbered serially from 1 to 55, with the TF number appearing in the respective box in Table 2.

The optimization framework consists of two stages. In the first stage, the optimized value (β, c₁, c₂, c₃) for every TF of the grid in Table 2 is obtained through optimization of the transient response in simulation, using the derived model G_buck. The transient performance is evaluated using the “integral of time weighted absolute error” (ITAE), calculated using the following equation, where X is the vector [β, c₁, c₂, c₃]. t_init is the time instant at which the transient occurs, and t_s is the settling time.

$\mathrm{ITAE}\left(X\right)={\int }_{t_{\mathrm{init}}}^{t_s}\left(t-t_{\mathrm{init}}\right)❘e\left(t-t_{\mathrm{init}},X\right)❘\mathrm{dt}$

A combination of computer-based simulation programs, such as MATLAB® and SIMULINK®, also available from MathWorks, Inc., may be used for this purpose. A processes of finding a minimum of an unconstrained multivariable function using derivative-free method, e.g., the fminsearch function in MATLAB® which is based on the Nelder-Mead simplex direct search algorithm, is used for the optimization. Each iteration of the optimization consists of the following 3 steps.

• • 1. The MRFT block described by the diagram of FIG. 1a, with the modified relay as described in Eq. (2) and with W(s)=G_buck(s), is simulated by the computer-based simulation programs using the initial/current value of β. Measurements of T_u and a₀ from the simulation are obtained.
  • 2. PID controller coefficients are calculated from (4) using T_u and a₀ from step 1 and the initial/current (c₁, c₂, c₃).
  • 3. Using the PID controller coefficients from step 2 and the current G_buck, a closed-loop simulation of a step in V_o-ref is performed and the ITAE is calculated.

After every iteration (of steps 1-3), the set (β, c₁, c₂, c₃) is updated by the fminsearch function to result in a lower ITAE—while still ensuring that the updated (β, c₁, c₂, c₃) still satisfy the constraints in Eq. (5). The iterations (of steps 1-3) are repeated until the improvement in ITAE is below the specified tolerance. Let the 55 TFs in the grid of FIG. 4 be referred to as G_buck-n for n=1 to 55. The final output of the first stage of optimization is therefore a group of locally optimized sets (βⁿ, c₁ⁿ, c₂ⁿ, c₃ⁿ) for n=1 to 55, where each (βⁿ, c₁ⁿ, c₂ⁿ, c₃ⁿ) results in the lowest ITAE for the corresponding G_buck-n.

The second stage of the optimization applies the principle of “least performance degradation” to obtain a globally optimized set of (β, c₁, c₂, c₃). This is done as follows:

• • 1. Using β¹ (the optimized β found for G_buck-1), MRFT is simulated for G_buck-1 to G_buck-55. Respective MRFT measurements, T_u1ⁿ and a₀₁ⁿ (for n=1 to 55), are recorded.
  • 2. Now using c₁¹, c₂¹, and c₃¹ (which are locally optimized for G_buck-1), along with T_u1ⁿ and a₀₁ⁿ (for n=1 to 55) from the previous step, a PID controller C₁ⁿ(s) is designed for each of G_buck-1 to G_buck-55 using (10). Thus, C₁ⁿ(s) is based on G_buck-n and the set (β¹, c₁¹, c₂¹, c₃¹) that is locally optimized to G_buck-1.
  • 3. Next, each G_buck-n is simulated in closed-loop using its corresponding PID controller from the previous step, C₁ⁿ(s). For example: G_buck-1 with C₁¹(s), G_buck-2 with C₁²(s), and so on. Each time the corresponding ITAE, denoted ITAE₁ⁿ(for n=1 to 55) is recorded. The single maximum (worst-case) ITAE

$\underset{n=1,2,\dots 55}{\max }{\mathrm{ITAE}}_1^n,$

is then found.

• • 4. Steps 1, 2, and 3 are repeated using (β², c₁², c₂², c₃²) to obtain ITAE₂ⁿ for n=1 to 55, and

$\underset{n=1,2,\dots 55}{\max }{\mathrm{ITAE}}_2^n$

is recorded. This is repeated until all worst-case ITAEs,

$\underset{n=1,2,\dots 55}{\max }{\mathrm{ITAE}}_k^n\left(\mathrm{for}k=1\mathrm{to}55\right),$

are recorded.

The globally optimized set, (β*, c₁*, c₂*, c₃*), is the set with the lowest

$\underset{n=1,2,\dots 55}{\max }{\mathrm{ITAE}}_k^n\mathrm{for}k=1\mathrm{to}55.$

In other words, it is the set that results in the least performance degradation when applied to all converters in the considered range. This completes the second stage of the optimization. Constraints used in the optimization process include restricting ϕ_m in Eq. (5) to [20°, 60°], c₂ to [0.1, 100] to limit the integral action, and restricting c₁ and c₃ to being positive. A final constraint was to set the minimum β to −0.2, since large negative values of β may be difficult to apply in practice due to the oscillations associated with having a smaller amplitude thus making their accurate measurement a challenging task. Performing the optimization process above for the system represented by G_buck and for the range of parameters defined by the grid of Table 2, the globally optimized set was found to be (β^{3, c}₁³, c₂³, c₃³), which resulted in ϕ_m=35°. The numerical values of the parameters in this set were:

β*=−0.2, c₁*=0.69, c₂*=1.14, c₃*=0.19

In other words, using (β*, c₁*, c₂*, c₃*)=(β³, c₁³, c₂³, c₃³) to tune a controller for every other G_buck-n model in the considered range resulted in the least performance degradation from the locally optimized performance, ITAE₃³. This worst-case degradation,

$\underset{n=1,2,\dots 55}{\max }{\mathrm{ITAE}}_3^n,$

only 13% more than the locally optimized case ITAE₃³. It is noted that the (β*, c₁*, c₂*, c₃*) provided earlier in the manuscript included a ±5% tolerance for each parameter in the set. This is because as long as the four parameters satisfy the constraints in Eq. (5), the specified phase margin is guaranteed, and further if they are in the vicinity of the original values of (β³, c₁³, c₂³, c₃³) , then very similar dynamic performance is obtained.

While the invention has been described with reference to an exemplary embodiment(s), it will be understood by those skilled in the art that various changes may be made, and equivalents may be substituted for elements thereof without departing from the scope of the invention. In addition, many modifications may be made to adapt a particular situation or material to the teachings of the invention without departing from the essential scope thereof. Therefore, it is intended that the invention is not limited to the disclosed embodiment(s), but that the invention will include all embodiments falling within the scope of the appended claims.

Claims

1. A method of automatically tuning a proportional-integral-derivative (PID) controller controlling a DC-DC buck switching power converter, comprising: a) substituting the PID controller with a modified relay feedback test (MRFT) block having an MRFT magnitude (h) and a threshold parameter (β);b) determining a current error signal (e[k])by measuring an output voltage (Vo[k]) of the DC-DC buck switching power converter and comparing the output voltage (vo) to a desired reference output voltage (Vo-ref);c) inputting the error signal (e[k]) to the MRFT block;d) producing a MRFT block output signal u[k] in accordance with the MRFT block equations, which can only be D+h or D−h;e) operating the DC-DC buck switching power converter using the updated nominal duty cycle (D);f) repeating steps from b) through e), which results in generation of self-excited oscillations due to the switching of the MRFT block output between the values D+h and D−h, until a predefined number of cycles (Ncycles) is observed;g) measuring a frequency (Ω0) and amplitude (a0) of oscillations of the output voltage (Vo[k]) of the DC-DC buck switching power converter;h) updating PID controller parameters based on the frequency (Ω0) and amplitude (a0) of the oscillations of the voltage output (Vo[k]) of the DC-DC buck switching power converter; andi) substituting the MRFT block with the PID controller.

2. The method according to claim 1, further comprising: j) comparing the error signal (e[k]) against previous maximum and minimum values of the error signal (emax, emin);k) determining a new maximum value (emax) if the current error signal value (e[k]) is greater than the previous maximum value;l) determining a new minimum value (emin) if the current error signal value (e[k]) is less than the previous minimum value; andm) determining the current value of the MRFT block output signal (u[k]) based on the new maximum or minimum value (emax, emin), the previous value of the MRFT block output signal (u[k−1]), the current error signal value (e[k]), and the MRFT block threshold parameter (β) and magnitude (h) in such a way that the current value of the MRFT block output signal (u[k]) is set to be D+h if (e[k]≥−β×emin and u[k−1]=D−h) or if (e[k]≥−β×emax and u[k−1]=D+h); and the current value of the MRFT block output signal (u[k]) is set to be D−h if (e[k]≤−β×emax and u[k−1]=D+h) or if (e[k]≤−β×emin and u[k−1]=D−h)

3. The method according to claim 2, wherein an initial number of cycles (Ndiscard) are disregarded when measuring the frequency (Ω0) and amplitude (a0) of the oscillations of the output voltage (Vo[k]) of the DC-DC buck switching power converter (step g).

4. The method according to claim 3, wherein the PID controller parameters are updated based on a number of recorded cycles (NMRFT) which equals a total number of cycles (Ncycles) minus the initial number of cycles (Ndiscard), an elapsed time period (TMRFT) of the number of recorded cycles (NMRFT), and the frequency (Ω0), where

5. The method according to claim 4, wherein the PID controller parameters are updated based on the amplitude (a0) where

6. The method according to claim 5, further comprising the step of: n) calculating a proportional gain (Kc), an integral time (Ti), and a derivative time (Td) for the PID controller, where:

7. The method according to claim 6, where c1=0.69±5%, c2=1.14±5%, and c3=0.19±5%.

8. The method according to claim 6, in which the coefficients c1, c2, and c3 are selected to satisfy the equations: c1√{square root over (1+ζ2)}=1 and β=sin(ϕm−tan−1ζ), where

9. The method according to claim 1, where a threshold parameter β=−0.2±5%.

10. A non-transitory computer readable storage medium having data stored therein representing software executable by a computer, the software including instructions to: switch a controller circuit of a DC-DC buck switching power converter from a proportional-integral-derivative (PID) controller enforcing a nominal duty cycle (D) of the converter to a modified relay feedback test (MRFT) block having an MRFT magnitude (h) and a threshold parameter (β);determine a current error signal (e[k]) by measuring an output voltage (Vo[k]) of the DC-DC buck switching power converter and comparing the output voltage (Vo[k]) to a desired reference output voltage (Vo-ref);input the error signal (e[k]) to the MRFT block;determine a controller output signal (u[k]) of the MRFT block;update the nominal duty cycle (D) to the value given by the controller output signal (u[k]);operate the DC-DC buck switching power converter using the updated nominal duty cycle (D);observe appearance of self-excited oscillations and repeat the preceding instructions for a predetermined number of cycles (Ncycles) of these oscillations;measure a frequency (Ω0) and amplitude (a0) of oscillations of the output voltage (Vo[k]) of the DC-DC buck switching power converter;update PID controller parameters based on the frequency (Ω0) and amplitude (ao) of the oscillations of the voltage output (Vo[k]) of the DC-DC buck switching power converter; andswitch the controller circuit of the DC-DC buck switching power converter from a MRFT block to the PID controller.

11. The non-transitory computer readable storage medium according to claim 10, wherein the software further includes instructions to: compare the error signal (e[k]) against previous maximum and minimum values of the error signal (emax, emin);store a new maximum value (emax) in transitory memory if the current error signal value (e[k]) is greater than the previous maximum value;store a new minimum value (emin) in transitory memory if the current error signal value (e[k]) is less than the previous minimum value; andcalculate current value of the MRFT block output signal (u[k]) based on the new maximum or minimum value (emax, emin), the previous value of the MRFT block output signal (u[k−1]), the current error signal value (e[k]), and the MRFT block threshold parameter (β) and magnitude (h) in such a way that the current value of the MRFT block output signal (u[k]) is set to be D+h if (e[k]≥−β×emin and u[k−1]=D−h) or if (e[k]≥−β×emaxu[k−1]=D+h); and the current value of the MRFT block output signal (u[k]) is set to be D−h if (e[k]≤−β×emax and u[k−1]=D−h) or if (e[k]≤−β×emin and u[k−1]=D−h), and store it in transitory memory.

12. The non-transitory computer readable storage medium according to claim 10, wherein the software further includes instructions to discard the frequency (Ω0) and amplitude (a0) of the oscillations of the output voltage (Vo[k]) of the DC-DC buck switching power converter data collected during an initial number of cycles (Ndiscard).

13. The non-transitory computer readable storage medium according to claim 12, wherein the software further includes instructions to update the PID controller parameters based on a number of recorded cycles (NMRFT) which equals a total number of cycles (Ncycles) minus the initial number of cycles (Ndiscard), an elapsed time period (TMRFT) of the number of recorded cycles (NMRFT), and the frequency (Ω0), where

14. The non-transitory computer readable storage medium according to claim 13, wherein the software further includes instructions to update the PID controller parameters based on the amplitude (a0) where

15. The non-transitory computer readable storage medium according to claim 14, wherein the software further includes instructions to calculate a proportional gain (Kc), Ti an integral time (Ti), and a derivative time (Td) for the PID controller, where:

16. The non-transitory computer readable storage medium according to claim 15 having data stored therein representing a value of c1 as 0.69±5%, a value of c2 as 1.14±5%, and a value of c3 is 0.19±5%.

17. The non-transitory computer readable storage medium according to claim 15, wherein the software further includes instructions to calculate the coefficients c1, c2, and c3, which are selected to satisfy the equations: c1√{square root over (1+ζ2)}=1 and β=sin(ϕm−tan−1ζ), where

18. The non-transitory computer readable storage medium according to claim 11 having data stored therein representing a value of threshold parameter β as −0.2±5%.