Digital control techniques for DC-DC converters have seen significant advancements in recent years, offering improved performance and flexibility compared to traditional analog methods. These developments have been driven by the increasing demand for more efficient and reliable power conversion systems in various applications, including electric vehicles and renewable energy systems.

Several digital control methods have been applied to DC-DC converters, including [1] - [4] .

1) Voltage-mode control [5]

2) Current-mode control [6] [7]

3) PID Control [8] [9]

4) Hysteretic control [10] [11]

5) Sliding mode control [12] [13]

6) Model predictive control [14]

7) Fuzzy logic control [15]

Among these digital control techniques, voltage-mode and current-mode control remain the most prevalent in industry due to their relative simplicity and well-established design procedures.

However, the approach [16] presented in this paper, utilizing Non-Minimal State Space (NMSS) representation combined with Proportional-Integral-Plus (PIP) control and optimized through Linear Quadratic Regulator (LQR) design, offers several advantages over traditional methods. This approach merits consideration for the following reasons [16] .

1). Formulation eliminates the need for state observers, reducing sensitivity to model mismatch and improving overall system robustness.

2) PIP control extends traditional PI control, offering enhanced flexibility and performance, (particularly for higher-order systems or those with significant time delays).

3) The integration of LQR optimization allows for systematic tuning of controller parameters, balancing control effort and system performance.

By combining these advanced techniques, this paper aims to examine a control strategy for DC-DC buck converters that offers improved voltage regulation, transient response, and disturbance rejection compared to conventional methods while maintaining stability under various operating conditions.

The remainder of this paper is organized as follows: Section 2 presents the converter modelling approach. This starts with the state space averaged model and leads to the NMSS model representation. Section 3 undertakes the LQR optimal controller design. An overview of the implementation of the control law using a digital microcontroller is discussed in Section 4. More detailed practical information relating to our use of the TI F28069M microcontroller is presented in Section 5. The effectiveness of the design is shown in Section 6 where the load step response is examined. The conclusion follows in Section 7.

The schematic of a Buck dc-dc converter is shown in Figure 1 . The circuit switches between two topologies depending on the position of the single pole double throw switch. The switch is operated cyclically with a period of $T_s$. The switch is in position 1 for a length of time of $d{T}_s$, where $d$, $0<d<1$, is the duty ratio. During the remainder of the period of length $\left(1-d\right)T_s$, the switch is in position 2. We will denote $d^{\prime }\equiv 1-d$. We use a state space mathematical description of the two topologies.

During the first subinterval we have:

$\frac{\text{d}x}{\text{d}t}=A_{1}x+B_1{v}_g$(1)

$y=C_{1}x+E_1{v}_g$(2)

During the second subinterval we have:

$\frac{\text{d}x}{\text{d}t}=A_{2}x+B_2{v}_g$(3)

$y=C_{2}x+E_2{v}_g$(4)

where $x$ denotes a minimal length state vector, and $v_g$ is the input voltage and $y$ is the output. The state variables are typically the inductor current and capacitor voltage, such that $x={\left[\begin{array}{cc}i& v\end{array}\right]}^{\text{T}}$.

To represent the system as linear and time invariant the State Space Averaging (SSA) [17] model will be used. Variables $d,x,v_g$ and $y$ are assumed to undergo small variations $\hat{d},\hat{x},{\hat{v}}_g$ and $\hat{y}$ around their steady state operating point denoted by the capitalized symbols $D,X,V_g$ and $Y$. The resulting small signal model is given by:

$\frac{\text{d}\hat{x}}{\text{d}t}=A\hat{x}+B{\hat{v}}_g+B_d\hat{d}$(5)

$\hat{y}=C\hat{x}+E{\hat{v}}_g+E_d\hat{d}$(6)

where

$A=D{A}_1+D^{\prime }{A}_2$(7)

$B=D{B}_1+D^{\prime }{B}_2$(8)

$C=D{C}_1+D^{\prime }{C}_2$(9)

$E=D{E}_1+D^{\prime }{E}_2$(10)

$B_d=\left(A_1-A_2\right)X+\left(B_1-B_2\right)V_g$(11)

$E_d=\left(C_1-C_2\right)X+\left(E_1-E_2\right)V_g$(12)

The steady state vector, $X$, is given from the SSA DC model:

$X=-A^{-1}B{V}_g$(13)

The control input duty ratio to output voltage transfer function, $\frac{\hat{y}}{\hat{d}}$ is derived by using the state space quadruple $\left\{A,B_d,C,E_d\right\}$. For the Buck converter of Figure 1 we find:

$A=\left[\begin{array}{cc}0& -\frac{1}{L_f}\\ \frac{1}{C_f}& -\frac{1}{R{C}_f}\end{array}\right],B_d=\left[\begin{array}{c}\frac{V_g}{Lf}\\ 0\end{array}\right],C=\left[\begin{array}{cc}0& 1\end{array}\right],E_d=0$(14)

To apply discrete time control, this continuous time system is discretized using the Matlab c2d function. The resulting discrete time state space model is then transformed into a discrete time transfer function, using the Matlab tf function. The resulting transfer function has the general form:

$\frac{y}{u}=\frac{B\left(z^{-1}\right)}{A\left(z^{-1}\right)}=\frac{b_1{z}^{-1}+\cdots +b_m{z}^{-m}}{1+a_1{z}^{-1}+\cdots +a_n{z}^{-n}}$(15)

This represents the plant (i.e., the converter) as seen in the general block diagram of Figure 2 .

The remaining blocks represent the compensation, which has been termed Proportional Integral Plus (PIP) [16] . In order to reduce steady state errors, integral control has been introduced. The plant, together with integral control, (which adds an extra state variable, $z$), are now represented in a non-minimal state space form. The NMSS state vector is given by:

$\begin{array}{l}x{\left(k\right)}^{\text{T}}=[\begin{array}{cccccc}y\left(k\right)& y\left(k-1\right)& \cdots & y\left(k-n+1\right)& u\left(k-1\right)& \cdots \end{array}\\ \begin{array}{cccc}u\left(k-2\right)& \cdots & u\left(k-m+1\right)& z\left(k\right)\end{array}]\end{array}$(16)

The dimension of the state vector is $n+m$, i.e., the sum of orders of the denominator and numerator. Recall that a minimal state space representation with an added integrator would have state dimension $n+1$, i.e., the order of the denominator plus one.

The resulting NMSS state space representation of the plant augmented with an integrator is given by [16] :

$x\left(k\right)=Fx\left(k-1\right)+gu\left(k-1\right)+d{y}_d\left(k\right)$(17)

$y\left(k\right)=hx\left(k\right)$(18)

where

$F=\left[\begin{array}{ccccccccc}-a_1& \cdots & -a_{n-1}& -a_n& b_2& \cdots & b_{m-1}& b_m& 0\\ 1& \cdots & 0& 0& 0& \cdots & 0& 0& 0\\ ⋮& \ddots & ⋮& ⋮& ⋮& \ddots & ⋮& ⋮& ⋮\\ 0& \cdots & 1& 0& 0& \cdots & 0& 0& 0\\ 0& \cdots & 0& 0& 0& \cdots & 0& 0& 0\\ 0& \cdots & 0& 0& 1& \cdots & 0& 0& 0\\ ⋮& \ddots & ⋮& ⋮& ⋮& \ddots & ⋮& ⋮& ⋮\\ 0& \cdots & 0& 0& 0& \cdots & 1& 0& 0\\ a_1& \cdots & a_{n-1}& a_n& -b_2& \cdots & -b_{m-1}& -b_m& 1\end{array}\right]$

$g={\left[\begin{array}{lllllllll}{b}_1\hfill & 0\hfill & \cdots \hfill & 0\hfill & 1\hfill & 0\hfill & \cdots \hfill & 0\hfill & -b_1\hfill \end{array}\right]}^{\text{T}}$

$d={\left[\begin{array}{lllllllll}0\hfill & 0\hfill & \cdots \hfill & 0\hfill & 0\hfill & 0\hfill & \cdots \hfill & 0\hfill & 1\hfill \end{array}\right]}^{\text{T}}$

$h=\left[\begin{array}{lllllllll}1\hfill & 0\hfill & \cdots \hfill & 0\hfill & 0\hfill & 0\hfill & \cdots \hfill & 0\hfill & 0\hfill \end{array}\right]$

In this section, an optimal proportional-integral-plus (PIP) controller will be designed. The optimal PIP control presented relies on the aforementioned (NMSS) representation. As will be seen, the design process is relatively simple and very effective. The control law takes the form [16] of full state feedback:

$u\left(k\right)=-k^{\text{T}}x\left(k\right)$(19)

where

$k^{\text{T}}=\left[\begin{array}{llllllll}{f}_0\hfill & f_1\hfill & \cdots \hfill & f_{n-1}\hfill & g_1\hfill & \cdots \hfill & g_{m-1}\hfill & -k_I\hfill \end{array}\right]$(20)

and $x\left(k\right)$ is given by (16).

The goal of optimal design is to minimize a quadratic cost function subject to the constraints of the system state equations. The cost function is given by [16] :

$J=\underset{k=0}{\overset{\infty }{{\displaystyle \sum }}}x{\left(k\right)}^{\text{T}}Qx\left(k\right)+r{\left(u\left(k\right)\right)}^2$

where $x\left(k\right)$ is the non-minimum state vector, $u\left(k\right)$ is the control input, $Q$ is the positive definite weighting matrix, and $r$ is the positive weighting scalar. We define the matrix $Q$ as the following diagonal matrix:

$\begin{array}{c}Q=\text{diag}\left(\begin{array}{cccccccc}{q}_1& q_2& \cdots & q_n& q_{n+1}& \cdots & q_{n+m-1}& q_{n+m}\end{array}\right)\\ =\text{diag}\left(\begin{array}{cccccccc}{q}_y& q_y& \cdots & q_y& q_u& \cdots & q_u& q_e\end{array}\right)\end{array}$

In addition, the positive weighting scalar $r$ will be set equal to $q_u$ ( [16] , p. 111), which is defined next. By defining the following equations only three tuning parameters will be left to configure:

$q_y=\frac{W_y}{n};q_u=\frac{W_u}{m};q_e=W_e$

For many practical cases the three tuning variables ( $W_y$, $W_u$, $W_e$) can be simply set to unity. It is necessary to use an Algebraic Riccati Equation to extract the corresponding optimal controller gains. The method proposed by [16] calls for solving the following equations recursively:

$k^{\text{T}}\left(i\right)={\left(r+g^{\text{T}}P\left(i+1\right)g\right)}^{-1}{g}^{\text{T}}P\left(i+1\right)F$

where

$P\left(i\right)=Q+F^{\text{T}}P\left(i+1\right)F-F^{\text{T}}P\left(i+1\right)g{k}^{\text{T}}\left(i\right)$

and

$P\left(N\right)=0,k\left(N\right)=0$

Implementing the recursive equations in MATLAB can be done as shown in Figure 3 .

For the Buck converter, with parameters, $C_f=100$ μF, $L_f=300$ μH, $R=10$ Ω, as used in the constructed prototype discussed in the next section, the optimal controller gains lead to the following results:

$F\left(z^{-1}\right)=22-17.3z^{-1}$

$G\left(z^{-1}\right)=1+0.263z^{-1}$

$k_I=0.736$

A straightforward block diagram reduction of the system shown in Figure 2 leads to that shown in Figure 4 . We see that the equivalent compensator transfer function $G_c\left(z^{-1}\right)$ is given by:

$G_c\left(z^{-1}\right)=\frac{k_{I}A\left(z^{-1}\right)}{\left[G\left(z^{-1}\right)A\left(z^{-1}\right)+F\left(z^{-1}\right)B\left(z^{-1}\right)\right]\left(1-z^{-1}\right)}$(21)

The product of the compensator transfer function with that of the plant, gives the loop gain of the system. The frequency response of the loop gain is shown in Figure 5 , which verifies the good stability margins.

A digital microcontroller, specifically the C2000 development kit for the Texas Instruments F28069M microcontroller, was used to implement the PIP controller in hardware. An overall block diagram of the system is shown in Figure 6 . The important controller modules for the implementation are the pulse width modulator (PWM) and analog to digital converter (ADC). Not explicitly shown is the general purpose input/output (GPIO) which assigns IO pins to the internal modules. The converter requires a PWM signal, which is directly available from the microcontroller.

The PWM outputs a high voltage of 3.3 V or a low voltage of 0 V with a variable duty ratio. The ADC uses a resistor capacitor circuit to sample and hold a voltage ( [18] , pp. 520-525). The output voltage of the plant is scaled down to the ADC’s input range of 0 to 3.3 V with a resistive voltage divider, as seen in the Buck converter (the plant) schematic in Figure 7 . The ADC then converts that value to a digital number that can be read from a register in memory. The controller program then scales that value appropriately to get a precise and accurate reading of the output voltage. If a GPIO pin is in output mode, the voltage can be set to either 3.3 V or 0 V. In input mode, the voltage on the GPIO pin can be routed to an internal module, such as the ADC. The GPIO module can be configured to specify the pins to which the ADC and PWM have access.

The PWM works by using an internal counter, a period register and a counter compare register ( [18] , p. 244). The counter increments at each pulse of the system clock. When the counter is zero, the PWM asserts a digital output pin, i.e., sets the output pin to logical true of 3.3 V. When the counter equals the value in the PWM’s counter compare register, the PWM deasserts the same pin, i.e., sets the output pin to a logical false of 0 V. When the counter equals the value in the PWM’s period register, the counter resets to zero. The PWM also has a synchronization bit that, when written to, will reset the counter to zero. This is the most basic operation of the PWM module. There are more sophisticated counting options to synchronize with external or internal signals and ways to change the behavior of what is done at different points in the count. The basic mode of operation is sufficient for the PIP controller.

The PIP program controls the output by setting the counter compare register to a value that will produce the desired duty cycle. The new counter compare value is written to a shadow register which serves as a temporary holding place for the active register ( [18] , p. 244). Then, a software PWM resync is performed to load the new counter compare value from the shadow register and reset the counter to zero. There may be risks that this will result in inconsistent PWM periods. However, the processing should be precisely scheduled to ensure that this operation is done at a 100 kHz frequency, which is the switching frequency of the Buck converter. This requirement for the constant switching frequency was confirmed by measurement with the oscilloscope that showed that the PWM period was always 10 μs.

As previously mentioned, the ADC uses a resistor capacitor circuit to sample and hold a voltage it receives on a designated input pin. The ADC hardware then uses a set of comparators to convert that value to a digital number that can be read from a register in memory.

There are different modes of operation to control the timing of the ADC operation. For the purpose of this PIP program the ADC operates in non-continuous mode. This means a start of conversion (SOC) signal must be sent to the ADC to initiate a read of the voltage on the input pin. An adequate sample and hold time must be set such that the ADC charges its capacitor sufficiently ( [18] , p. 521). Once the timer indicates the sample and hold period is over, the ADC produces a digital read of the voltage. This value is in terms of “ticks.” The F28069M controller has a 12-bit ADC register, therefore 0 ticks correspond to 0 V and 4095 ( $=2^{12}-1$) ticks corresponds to 3.3 V. After the ADC finishes the read, it will update its register with a new value and produce an end of conversion (EOC) signal. This EOC signal is useful for minimizing delay between receiving a new digital input and processing it.

The PIP program, seen as the control law in Figure 6 , collects input from the ADC and produces output with the PWM. A common software element, the Interrupt Service Routine (ISR), is introduced to ensure that the control law is executed periodically at a frequency of 100 kHz.

The architecture of the PIP program relies upon the ISR. This is for two reasons. The first is that the ISR can trigger the ADC’s EOC signal. This will help ensure minimum delay between sampling the input and executing the difference equation, and ultimately producing a PWM output. The second is that the SOC is routed to a timer such that the ISR runs at the sampling frequency. This will ensure that the coefficients in the PIP’s difference equation will be valid for the sample period for which they were derived.

Therefore, the program can be split into two main components, the main function and the ISR function. These two main components are discussed in the next two subsections.

To complete the implementation and form the hardware prototype, the microcontroller is attached to the Buck converter as pictured in Figure 6 . The divided output voltage of the plant, $v_{div}$, as shown in Figure 7 , is connected to the ADCINA4 pin, which is the input for the ADC used by the controller implementation. The controller’s EPWM 2 output pin is connected to point $d$ in Figure 7 . The signal $d$ drives the power mosfet of the Buck converter. The converter takes a nominal 10 V input and steps it down to 5 V. The quiescent duty ratio is thus 50%.

To test the efficacy of the controller, a step load disturbance signal is instigated by periodically switching an extra load across the output. This is done by turning mosfet Q3 on and off which adds and removes an additional shunt 20 Ω resistor. The results of this test are shown in Figure 10 . The top trace shows the open loop response and the bottom trace is the closed loop response of the output voltage $v$. The open loop response shows a large transient of 727 mV peak to peak and a non-zero steady state difference of 100 mV between switched states. The closed loop implementation greatly reduces the transient to 95 mV peak to peak and eliminates steady state difference completely. This zero steady state error is achieved by having integral control in the loop as implemented by the PIP controller. Also evident is that the transient settling time is greatly reduced in the closed loop configuration.

The use of a non-minimal state space representation permits the use of output and input signals as the states thus obviating the need for a state estimator. Absent an estimator there is no need for procedures such as loop transfer recovery (LTR) needed to optimize system response.

With NMSS, optimal control may more straightforwardly be implemented. In this paper we have shown its application in the design of a Buck converter controller. The resulting discrete control law can be quickly implemented. We illustrated this with a Texas Instruments F28069M microcontroller. As demonstrated by the implementation process, a microcontroller is suitable for quickly prototyping a discrete controller since it is elementary to store the discrete set of inputs and outputs as required by NMSS control. To aid in future design, some fundamental issues concerning microcontroller implementation have been discussed at length. The hardware prototype demonstrates the efficacy of this design approach through excellent disturbance rejection performance.

Samuel Jacobs, a former student in the Electrical and Computer Engineering Department at Portland State University is acknowledged for his assistance in microcontroller implementation and providing the oscillogram and a number of the figures appearing in this paper.