Enhancing the Filtering Capability and the Dynamic Performance of a Third-Order Phase-Locked Loop under Distorted Grid Conditions

Abstract

This work proposes a structural enhancement and a new technique to design the loop filter (LF) of a third-order phase-locked loop (PLL) to enhance the PLL dynamic performance under abnormal grid conditions. The proposed PLL combines a moving average filter (MAF) and an arbitrarily delayed signal cancelation (ADSC) for structural enhancement to achieve DC-offset rejection and harmonics elimination. The window length of the MAF is selected to be one-sixth of the fundamental grid period to remove non-triple odd harmonics and speed up the PLL dynamic response. The triple harmonics are eliminated, adopting the line-to-line voltage concept, while the ADSC operator rejects the DC offset. The LF design is based on a modified third-order polynomial tuned using stochastic optimization to minimize the settling time of the frequency deviation, offering better dynamic performance over the symmetrical optimum method (SOM) and achieving synchronization within one grid cycle. The PLL mathematical model, small-signal model, and LF design based on the modified polynomial are discussed. Finally, the proposed PLL performance is verified numerically and experimentally with comparisons with other PLLs to demonstrate the effectiveness of the proposed work.

1. Introduction

The feasible control of the grid-connected converters is of prime importance when renewable energy sources are connected to the grid. Therefore, grid synchronization is of interest due to more and more renewable energy sources tied up to the grid using power electronic converters in recent years [1].

The phase-locked loop (PLL) is a technique that can effectively synchronize the phase, frequency, and amplitude of the grid-connected inverters with the grid [2,3]. By maintaining synchronization, PLL helps to ensure that the renewable energy source can be safely and efficiently integrated into the grid. Additionally, PLLs can also be used to optimize the power output of the renewable energy source, further increasing its efficiency [4]. Conventional PLLs work effectively in ideal conditions. However, PLLs face significant challenges in estimating the phase and frequency under abnormal conditions [5,6,7,8,9,10]. PLLs should keep up the synchronization process in the presence of disturbances such as DC offset, phase jump, frequency jump, and the harmonics and return to the steady state within two grid cycles [11]. Therefore, a robust and accurate PLL with high filtering capability to achieve grid synchronization has attracted much attention in the literature [12,13,14,15,16,17]. In addition, the PLL loop filter is responsible for both the system static noise and dynamic performance to be set while considering the constraints imposed by the other system elements [18]. Hence, properly tuning the parameters of the loop filter (LF), including the proportional-integral (PI) controller’s parameters, is one of the most important issues in the PLL design [19]. There are many methods to design the PLL’s LF based on its small-signal model. If the system is third-order, one of the most used methods to design the LF is the symmetrical optimum method (SOM) [20,21,22], whereas if the system is a second-order system, the damping factor and natural frequency of a standard second-order system are used to design the gains [7,23,24]. An adaptive feedforward mechanism can adjust the filter gains according to the estimated grid frequency [25]. In general, the value of the gains is selected targeting a 2% criteria settling time. All the papers above rely on the small-signal model of the PLL to design the LF gains, giving a local nature to the solution. In contrast, the LF gains are designed in this paper based on the actual PLL model without approximations using stochastic optimization.

On the other hand, the filtering capability of DC offset and harmonics are associated with the MAF window length that can be used as a prefilter or as an in-loop filter [24,26]. The DC offset and harmonics are rejected if the window length is the same as the fundamental grid period, but this will add a delay. In [13], the MAF is used as a prefilter in the αβ-reference frame; this speeds up the response because no delay is introduced in the loop. However, under off-nominal frequencies, a phase shift is introduced; hence, a phase error correction must be created, increasing the system’s complexity. Another choice of window length is half of the nominal period. In general, when the window length decreases, the speed of the response increases, affecting the filtering capability such that the DC offset cannot be rejected.

A quasi-type1 (QT1) PLL uses the MAF as an internal filter, although utilizing the filter inside the control loop decreases the response speed. The response can be enhanced by utilizing a P-controller with a feedforward term [14]. However, the performance of this PLL degrades under frequency drift with the presence of harmonics and takes more than two grid cycles to settle down. Another approach in [15], similar to [14], adopts the same window length for the MAF to remove the impact of the odd harmonics and uses αβ delayed signal cancellation (DSC) with a fixed time delay to remove the effect of the DC offset and the even harmonics. Under frequency deviation, a phase shift is created; hence, phase error correction is needed. A combination of MAF and DSC in the paralleled filter (PF) is suggested in [16]. The MAF extracts the fundamental frequency negative sequence (FFNS) at 100 Hz and the fundamental frequency positive sequence (FFPS) at 0 Hz, while the modified DSC (MDSC) extracts the FFNS only. Thus, the PF benefit passes the FFNS through MAF and MDSC with a reversed phase. Finally, the FFNS can be removed by an arithmetic operation. The MAF can remove the other harmonics, where the window length equals one-sixth of the nominal period, speeding the response speed more than the previous techniques. However, this PLL suffers from oscillation in the estimated grid information under frequency drift; it also has no dc offset rejection capability.

This paper proposes a new approach to design the loop filter (LF) of a third-order phase-locked loop (PLL), offering contributions in terms of the structural improvement of the three-phase PLL and a method for the optimal loop filter design. Without loss of generality, the proposed PLL combines an MAF with an arbitrarily delayed signal cancellation (ADSC) for structural enhancement to achieve DC-offset rejection and harmonics elimination. The window length of the MAF is selected to be one-sixth of the fundamental grid period to remove non-triple odd harmonics and accelerate the PLL response. The triple harmonics are eliminated, adopting the line-to-line voltage concept, while the ADSC operator rejects the effect of the DC offset. Moreover, different optimization methods are adopted to design the gains of the LF targeting the 2% criterion settling time in grid frequency deviation. The effectiveness of the proposed PLL and the adopted loop filter design method is verified by comparing their performance with other related PLLs and verifying the offered improvements numerically and experimentally.

2. The Proposed Method

2.1. PLL Structure Enhancement

PLL structure improvement includes the phase-to-line voltage transformation to block the triple harmonics. The ADSC, which is not restricted to a specific time delay, blocks the DC offset and the MAF with a window length of one-sixth of the fundamental grid period to block the non-triple odd harmonics.

The schematic diagram of the proposed PLL is shown in Figure 1, where ${\nu }_{a\phi }$, ${\nu }_{b\phi }$, and ${\nu }_{c\phi }$ are the phase voltages; ${\nu }_{ab}$, ${\nu }_{bc}$, and ${\nu }_{ca}$ are the line voltages; ${\omega }_n$ is the nominal angular frequency; $k_{\phi }$ is the phase error correction; $∆{\widehat{\omega }}_g$ is the deviation in the estimated grid frequency; $\widehat{\theta }$ is the estimated phase angle; ${\widehat{\omega }}_g$ is the estimated grid frequency.

2.1.1. Elimination of Odd Triple Harmonics

Typically, the grid-connected converters’ phase at neutral voltage is a source of odd triple harmonics. Thus, a transformation to line voltage is adopted in the proposed PLL to eliminate the odd triple harmonics out of the box without any additional cost or complexity, allowing for the decrease in the MAF window length, hence speeding up the PLL dynamic response. To verify the elimination, assume the input voltages of three-phase has odd triple harmonics as:

$${\nu }_{a\phi }=V\sin \left(\theta \right)+{\sum }_{k=3,9,27,\dots }{V}_k\sin \left(k\theta \right),$$

(1)

$${\nu }_{b\phi }=V\sin (\theta -\frac{2\pi }{3})+{\sum }_{k=3,9,27,\dots }{V}_k\sin \left(k\left(\theta -\frac{2\pi }{3}\right)\right),$$

(2)

$${\nu }_{c\phi }=V\sin (\theta +\frac{2\pi }{3})+{\sum }_{k=3,9,27,\dots }{V}_k\sin \left(k\left(\theta +\frac{2\pi }{3}\right)\right),$$

(3)

where $\theta$ is the phase, $V$ is the amplitude of the grid voltage, and $V_k$ is the amplitude of the odd triple harmonics. After some mathematical simplification, (1) to (3) can be written as shown in (4) to (6).

$${\nu }_{ab}=\sqrt{3}V\sin \left(\theta +\frac{\pi }{6}\right),$$

(4)

$${\nu }_{bc}=\sqrt{3}V\sin \left(\theta -\frac{\pi }{2}\right),$$

(5)

$${\nu }_{ca}=\sqrt{3}V\sin \left(\theta +\frac{5\pi }{6}\right),$$

(6)

It can be noticed from (4) to (6) that the conversion from phase to line voltages eliminates the odd triple harmonics. However, it is worth mentioning that this conversation adds a phase shift of $\pi ⁄6$ rad and scales the magnitude by $\surd 3$.

2.1.2. Moving Average Filter (MAF)

The MAF is a linear phase filter that is deemed as a low-pass filter (LPF) [27,28]. Its transfer function can be written as in Equation (7):

$$G_{\mathrm{MAF}\left(s\right)}=\frac{1-e^{-T_{w}s}}{T_{w}s},$$

(7)

where $T_w$ is the window length of the MAF; the block diagram of the MAF in the discrete domain is shown in Figure 2 [13].

In Figure 2, $\nu \left(k\right)$ is the input signal, $\overline{\nu }\left(k\right)$ is the output signal, and $N_w$ is the number of samples within the nominal period. The transfer function in the discrete domain can be expressed as:

$$G_{\mathrm{MAF}\left(\mathcal{Z}\right)}=\frac{1}{N_w}\frac{1-{\mathcal{Z}}^{-N_w}}{1-{\mathcal{Z}}^{-1}},$$

(8)

The magnitude and phase can be found by substituting (s = jω) into Equation (7)

$$G_{\mathrm{MAF}}\left(j\omega \right)=\left|\frac{sin\left(\frac{\omega T_w}{2}\right)}{\omega T_w/2}\right|\angle -\frac{\omega T_w}{2},$$

(9)

The harmonic ($k\times f$) positive-sequence harmonic in the αβ-frame is transformed into a $\left(k-1\right)\times f$ harmonics order in the $dq$-frame. In addition, the ($k\times f$) negative-sequence harmonic in the αβ-frame is transformed into a $\left(k+1\right)\times f$ harmonic in the $dq$-frame, where $k$ is the order of harmonics, and $f$ is the nominal grid frequency. Therefore, the most common αβ harmonics that appear in the input voltages of $5^{\mathrm{th}}$, $7^{\mathrm{th}}$, ${11}^{\mathrm{th}}$, ${13}^{\mathrm{th}}$, ${17}^{\mathrm{th}}$, and ${19}^{\mathrm{th}}$ are transferred $6^{\mathrm{th}}$, ${12}^{\mathrm{th}}$, and ${18}^{\mathrm{th}}$ in the $dq$-frame.

Based on Equation (9), the MAF has a unity gain at zero frequency and zero gains at frequencies =$2\pi k/T_w ,$ $\left(k=1,2,3,\dots \right)$. The harmonics rejection in the MAF is associated with its window length, which is adjusted to trade-off between noise reduction and response time, depending on the application’s specific requirements. If the window length equals the nominal period, the DC-offset and the harmonics are rejected. In contrast, only odd harmonics are rejected if the window length is half the nominal period. The non-triple odd harmonics are rejected if the window length is adjusted to one-sixth of the nominal period [27]. This work adopts the window length of the sixth fundamental grid period allowing the PLL to block the non-triple odd harmonics while the line voltage blocks the triple harmonics.

2.1.3. Arbitrary Delayed Signal Cancellation (ADSC)

The ADSC blocks the DC offset in the αβ reference frame [11,23]. It is represented in (10) and shown in Figure 3.

$${\overline{\nu }}_{\alpha \beta }=({\nu }_{\alpha \beta }\left(t\right)-{\nu }_{\alpha \beta }\left(t-d\right))/2,$$

(10)

where $d=T/n$ is the time delay in s, T is the nominal grid period, and n is an arbitrary positive number. The transfer function of the ADSC can be expressed as in (11).

$$\mathrm{ADSC}\left(s\right)=\frac{{\overline{\nu }}_{\alpha \beta }}{{\nu }_{\alpha \beta }}=\frac{1-e^{-\frac{T}{n}s}}{2},$$

(11)

which has the following magnitude and phase in the frequency domain:

$$\mathrm{ADSC}\left(j\omega \right)=\left|\sin \left(\frac{\omega d}{2}\right)\right|\angle \left(\frac{\pi }{2}-\frac{\omega d}{2}\right),$$

(12)

2.1.4. Mathematical Model

The grid line voltages can be written as

$$\begin{array}{c}\left[\begin{array}{c}{v}_{ab}\left(t\right)\\ v_{bc}\left(t\right)\\ v_{ca}\left(t\right)\end{array}\right]=\sqrt{3}(\left[\begin{array}{c}{V}^{+}\sin \left({\theta }^{+}+\frac{\pi }{6}\right)\\ V^{+}\sin \left({\theta }^{+}-\frac{\pi }{2}\right)\\ V^{+}\sin \left({\theta }^{+}+\frac{5\pi }{6}\right)\end{array}\right]+\left[\begin{array}{c}{V}^{-}\sin \left({\theta }^{-}+\frac{\pi }{6}\right)\\ V^{-}\sin \left({\theta }^{-}+\frac{5\pi }{6}\right)\\ V^{-}\sin \left({\theta }^{-}-\frac{\pi }{2}\right)\end{array}\right]+\left[\begin{array}{c}{V}_{ab}^0\\ V_{bc}^0\\ V_{ca}^0\end{array}\right]+{\sum }_{k=5,7,11\dots }^{\infty }\left[\begin{array}{c}{V}_k^{+}\sin \left({\theta }_k^{+}+\frac{\pi }{6}\right)\\ V_k^{+}\sin \left({\theta }_k^{+}-\frac{\pi }{2}\right)\\ V_k^{+}\sin \left({\theta }_k^{+}+\frac{5\pi }{6}\right)\end{array}\right]+\\ {\sum }_{k=5,7,11,\dots }^{\infty }\left[\begin{array}{c}{V}_k^{-}\sin \left({\theta }_k^{-}+\frac{\pi }{6}\right)\\ V_k^{-}\sin \left({\theta }_k^{-}+\frac{5\pi }{6}\right)\\ V_k^{-}\sin \left({\theta }_k^{-}-\frac{\pi }{2}\right)\end{array}\right]),\end{array}$$

(13)

where ${\theta }^{+}={\omega }_{g}t+{\phi }^{+}$, ${\theta }^{-}={\omega }_{g}t+{\phi }^{-}$, ${\theta }_k^{+}=k{\omega }_{g}t+{\phi }_k^{+}$, and ${\theta }_k^{-}=k{\omega }_{g}t+{\phi }_k^{-}$. ${\omega }_g$ is the fundamental grid frequency; $\theta$ is the phase angle; ${\phi }^{+}$ and ${\phi }^{-}$ are the initial phase angles of FFPS and FFNS, respectively; ${\phi }_k^{+}$ and ${\phi }_k^{-}$ are the initial phase angles of positive-sequence and negative-sequence harmonics, respectively; $V^{+}$ and $V^{-}$ are the amplitudes of FFPS and FFNS, respectively; $V_k^{+}$ and $V_k^{-}$ are the amplitudes of positive-sequence and negative-sequence harmonics, respectively. $V_{ab}^0$, $V_{bc}^0$, and $V_{ca}^0$ are the amplitudes of the DC offset components. Now, to convert from $abc$-frame to $\alpha \beta$-frame:

$$v_{\alpha \beta }=J_{\alpha \beta }\times v_{abc}\left(t\right),$$

(14)

where

$$J_{\alpha \beta }=\frac{2}{3}\left[\begin{array}{ccc}1& -\frac{1}{2}& -\frac{1}{2}\\ 0& \frac{\sqrt{3}}{2}& -\frac{\sqrt{3}}{2}\end{array}\right],$$

(15)

$$\begin{array}{c}{v}_{\alpha \beta }=\sqrt{3}(\left[\begin{array}{c}{V}^{+}\sin \left(\theta +\frac{\pi }{6}\right)\\ -V^{+}\cos \left(\theta +\frac{\pi }{6}\right)\end{array}\right]+\left[\begin{array}{c}{V}^{-}\sin \left(\theta +\frac{\pi }{6}\right)\\ V^{-}\cos \left(\theta +\frac{\pi }{6}\right)\end{array}\right]+\frac{2}{3}\left[\begin{array}{c}{V}_{ab}^{dc}-\frac{1}{2}{V}_{bc}^{dc}-\frac{1}{2}{V}_{ca}^{dc}\\ 0+\frac{\sqrt{3}}{2}{V}_{bc}^{dc}-\frac{\sqrt{3}}{2}{V}_{ca}^{dc}\end{array}\right]+{\sum }_{k=5,7,11,\dots }^{\infty }\left[\begin{array}{c}{V}_k^{+}\sin \left({\theta }_k+\frac{\pi }{6}\right)\\ -V_k^{+}\cos \left({\theta }_k+\frac{\pi }{6}\right)\end{array}\right]+\\ {\sum }_{k=5,7,11,\dots }^{\infty }\left[\begin{array}{c}{V}_k^{-}\sin \left({\theta }_k+\frac{\pi }{6}\right)\\ V_k^{-}\cos \left({\theta }_k+\frac{\pi }{6}\right)\end{array}\right]),\end{array}$$

(16)

The DC components are canceled after using arbitrary delay signal cancellation (ADSC).

$${\overline{\nu }}_{\alpha \beta }=\left(v_{\alpha \beta }^o\left(t\right)-v_{\alpha \beta }^o\left(t-d\right)\right)-\left(v_{\alpha \beta }^{dc}-v_{\alpha \beta }^{dc}\right)=v_{\alpha \beta }^o\left(t\right)-Y\left({\omega }_g\tau \right)v_{\alpha \beta }^o\left(t\right),$$

(17)

where $v_{\alpha \beta }^{dc}$ is the DC-offset component, and $v_{\alpha \beta }^o$ are the other components of $v_{\alpha \beta }$.

Now, to convert from $\alpha \beta$-frame to $dq$-frame:

$$\left[\begin{array}{c}{v}_d\\ v_q\end{array}\right]=Y\left(\theta +{\theta }_0\right)\left[\begin{array}{c}{v}_{\alpha }^o\\ v_{\beta }^o\end{array}\right]-Y\left(\theta +{\theta }_0+{\omega }_{g}d\right)\left[\begin{array}{c}{v}_{\alpha }^o\\ v_{\beta }^o\end{array}\right],$$

(18)

where

$$Y\left(\theta \right)=\left[\begin{array}{cc}\cos \left(\theta \right)& \sin \left(\theta \right)\\ -\sin \left(\theta \right)& \cos \left(\theta \right)\end{array}\right],$$

(19)

$$\left[\begin{array}{c}{v}_d\\ v_q\end{array}\right]=2\sqrt{3}\sin \left(\frac{{\omega }_{g}d}{2}\right)\left(V^{+}\left[\begin{array}{c}1\\ 0\end{array}\right]+V^{-}\left[\begin{array}{c}-\cos \left(2\theta \right)\\ \sin \left(2\theta \right)\end{array}\right]+{\sum }_{5,7,11\dots }^{\infty }{V}_k^{+}\left[\begin{array}{c}\cos \left(\theta -{\theta }_k\right)\\ -\sin \left(\theta -{\theta }_k\right)\end{array}\right]+{\sum }_{5,7,11\dots }^{\infty }{V}_k^{-}\left[\begin{array}{c}-\cos \left(\theta +{\theta }_k\right)\\ \sin \left(\theta +{\theta }_k\right)\end{array}\right]\right),$$

(20)

$Y\left(\theta \right)$ is the transformation matrix, and ${\theta }_0=-\frac{{\omega }_{g}d}{2}-\frac{\pi }{6}$. Based on Equation (20), $v_q$ is zero and the amplitude is scaled by $2\sqrt{3}\sin \left(\frac{{\omega }_{g}d}{2}\right)$. Moreover, the ADSC introduces a phase shift that can be corrected through feedback. In addition to that, the line-to-line transformation introduces a phase of $\frac{\pi }{6}$ that can be simply subtracted from the estimated phase to compensate for it, as shown in Figure 1. The $v_d$ is multiplied by $\frac{1}{2\sqrt{3}\sin \left(\frac{{\widehat{\omega }}_{g}d}{2}\right)}$ to compensate for the amplitude.

2.1.5. Small-Signal Model

The small-signal model can be written using $\widehat{\theta }$ and substituting ${\widehat{\theta }}_0=-\frac{{\widehat{\omega }}_{g}d}{2}-\frac{\pi }{6}$ into (20).

$$\left[\begin{array}{c}{v}_d\\ v_q\end{array}\right]=Y\left(\widehat{\theta }-\frac{{\widehat{\omega }}_{g}d}{2}\right)\left[\begin{array}{c}{v}_{\alpha }^o\\ v_{\beta }^o\end{array}\right]-Y\left(\widehat{\theta }-\frac{{\widehat{\omega }}_{g}d}{2}+{\omega }_{g}d\right)\left[\begin{array}{c}{v}_{\alpha }^o\\ v_{\beta }^o\end{array}\right],$$

(21)

After some approximations, with some trigonometric identities and using the MAF transfer function, ${\nu }_q\left(t\right)$ can be expressed as in Equation (22).

$${\widehat{\nu }}_q=2V^{+}\sin \left(\frac{{\omega }_{n}d}{2}\right)\left[\left(-\frac{∆{\omega }_{g}d}{2}\right)+\left(\frac{∆{\widehat{\omega }}_{g}d}{2}\right)+\left(∆\theta -∆\widehat{\theta }\right)\right]\times G_{\mathrm{MAF}}\left(s\right),$$

(22)

$${\widehat{\nu }}_q=V^{+}\left(2\sin \left(\frac{{\omega }_{n}d}{2}\right)\right)\left[\underset{\mathrm{ADSC}\left(s\right)}{\underbrace{\frac{\left(1+e^{-ds}\right)}{2}}}∆\theta \left(s\right)-∆\widehat{\theta }\left(s\right)+\frac{∆{\widehat{\omega }}_{g}d}{2}\right]\underset{\mathrm{MAF}\left(s\right)}{\underbrace{\frac{1-e^{-T_{w}s}}{T_{w}s}}},$$

(23)

$${\widehat{\nu }}_q=\left[\underset{\mathrm{ADSC}\left(s\right)}{\underbrace{\frac{\left(1+e^{-ds}\right)}{2}}}∆\theta \left(s\right)-∆\widehat{\theta }\left(s\right)+\frac{∆{\widehat{\omega }}_{g}d}{2}\right]\underset{\mathrm{MAF}\left(s\right)}{\underbrace{\frac{1}{\frac{T_w}{2}s+1}}},$$

(24)

The small-signal model of the proposed method is shown in Figure 4.

2.2. The Loop Filter Design

The LF is designed based on a modified third-order polynomial (25), in which the polynomial coefficients are tuned using stochastic optimization with the objective function and constraints to minimize the settling time of the frequency deviation.

$$s^3+a_2{\omega }_0{s}^2+a_1{\omega }_0^{2}s+{\omega }_0^3=0$$

(25)

The stability of the modified polynomial is determined using the Routh–Hurwitz criterion, from which the coefficients of the modified polynomial must obey the following conditions:

$${\omega }_0>0,$$

(26)

$$a_2>0,$$

(27)

$$a_2{a}_1>1,$$

(28)

Based on the small-signal model, the closed-loop transfer function is shown in (29):

$$\frac{∆\widehat{\theta }}{∆{\theta }_i}=\frac{1-e^{-Ts/n}}{2}\frac{k_{p}s+\frac{2}{T_w}{k}_i}{s^3+\frac{2}{T_w}{s}^2+\frac{2}{T_w}\left(k_p-k_i\frac{d}{2}\right)s+\frac{2}{T_w}{k}_i},$$

(29)

Comparing (25) with the characteristic equation of (29), $k_p$ and $k_i$ can be rewritten in terms of the coefficients of the modified polynomial as:

$$k_i=\frac{4}{T_w^2{a}_2^3},$$

(30)

$$k_p=\frac{2}{T_w{a}_2^2}\left(\frac{d}{T_w{a}_2}+a_1\right),$$

(31)

where $T_w=T/6$ is the window length, $d$ is the phase delay, and $a_1$ and $a_2$ determine the optimum parameter of the LF proportional-integral (PI)-gains. The optimization formulation is shown in (32). The flowchart that summarizes the optimization process is shown in Figure 5.

$$\begin{array}{c}\mathrm{Min} J\left(a_1, a_2\right)=t_{ss}\\ \mathrm{subject} \mathrm{to}\left\{\begin{array}{c}\begin{array}{c}OS<2\%\\ a_1{a}_2>1\\ a_2>0\end{array}\\ \begin{array}{c}{\left(a_1\right)}_{min}\le a_1\le {\left(a_1\right)}_{max}\\ {\left(a_2\right)}_{min}\le a_2\le {\left(a_2\right)}_{max}\end{array}\end{array}\right.\end{array},$$

(32)

where the objective function ($J$) is the settling time ($t_{ss}$) of the frequency deviation, OS is the overshoot, and ${\left(a\right)}_{min}$ and ${\left(a\right)}_{max}$ are the coefficients’ boundaries that are selected to be $1\le a_1\le 20$ and $1\le a_2\le 20$.

Different optimization methods were investigated for the selection of the coefficients $a_1$ and $a_2$: Particle swarm optimization (PSO), genetic algorithm (GA), bee algorithm (BA), and cuckoo optimization algorithm (COA) [29,30,31]. The results of the aforementioned optimization methods are shown in

Table 1. Optimization method comparison.

Method	$a_1$	$a_2$	$J$	$OS$
GA	2.2776	2.0026	0.0165	0.0625
PSO	2.2531	1.9532	0.0159	0.0675
BA	2.24531	1.9668	0.0156	0.0660
COA	2.27480	2.0444	0.0164	0.0586

.

The results of the COA optimum parameters are adopted without loss of generality. The corresponding PI-controller gains for all $d$ values are listed in

Table 2. The optimum PI-controller.

$d$	$k_p$	$k_i$
$T/2$	537.22	42,131
$T/4$	431.89	42,131
$T/8$	379.22	42,131
$T/10$	368.69	42,131
$T/12$	361.67	42,131
$T/16$	352.89	42,131
$T/32$	339.73	42,131

. It can be noticed that $k_i$ is constant for any $d$. This is because (30) depends on the window length and the coefficient $a_2$ only, which is a constant independent of $d$.

Figure 6 shows the response of the actual model along with its small-signal model under a $40°$ phase jump and a 6 Hz frequency jump from a 50 Hz grid frequency, validating the small-signal model in predicting the response of the actual PLL at a distinct phase delay.

3. Simulation Results

The performance of the utilized PLL adopting the proposed third-order optimized polynomial with DC-offset and harmonics rejection is verified using numerical simulation. It is compared with $\alpha \beta$MAF-PLL [13] and hybrid-PLL (HPLL) [15]. The nominal grid frequency is $f_n=50 \mathrm{Hz}$, and the parameters of the compared PLLs are summarized in

Table 3. Control parameter values.

Method	${\mathit{k}}_{\mathit{p}}$	${\mathit{k}}_{\mathit{i}}$
HPLL	94	-
$\alpha \beta$MAF-PLL	439.6	48,312

, while the grid harmonics are listed in

Table 4. Parameters of distortion in input voltage.

Component	−1	−3	+5	+7	+9	+11	+13	+27
Amplitude (p.u.)	0.01	0.05	0.01	0.01	0.05	0.01	0.01	0.05
$\mathrm{THD}\approx 9\%$

. Several case studies were considered for a fair comparison, as shown below:

• Case 1: A phase jump of $40°$ at 0.02 s and $d=T/4, T/16$, and $T/32$. The results are shown in Figure 7, Figure 8 and Figure 9, respectively, and summarized in

  Table 5. The results of case 1 (Phase Jump).

  	The Proposed PLL
  	n
  	32	16	4	$\alpha \beta$MAF-PLL	HPLL
  40 deg phase jump
  Phase settling time (ms)	16.8	17.3	21.3	44.4	36.7
  Overshoot (%)	47.62	49.95	52.23	50.13	51.24
  Peak frequency error (Hz)	13.36	13.1	10.25	5.23	5.78

  .
• Case 2: A frequency jump in the grid from 50 to 56 Hz at 0.02 s and $d=T/4, T/16$, and $T/32$. The results are shown in Figure 10, Figure 11 and Figure 12, respectively, and summarized in

  Table 6. The results of case 2 (Frequency Jump).

  	The Proposed PLL
  	n
  	32	16	4	$\alpha \beta$MAF-PLL	HPLL
  6 Hz frequency jump
  2% frequency settling time (ms)	15.6	16	19.3	39.3	40.9
  Overshoot (%)	0	0	0	0	2.61
  Peak phase error (°)	6.96	7.43	8.82	15.16	13.02

  .
• Case 3: Only DC-offset is added to the grid voltage by 0.1 pu of phase a, −0.1 of phase b, and 0.05 of phase c at 0.02 s, and $d=T/4, T/16$, and $T/32$. The results are shown in Figure 13, Figure 14 and Figure 15, respectively, and summarized in

  Table 7. The results of case 3 (DC-Offset).

  	The Proposed PLL
  	n
  	32	16	4	$\alpha \beta$MAF-PLL	HPLL
  DC-offset
  Phase settling time (ms)	19	19.4	22.4	44.6	41.4
  Peak frequency error (Hz)	0.91	0.91	0.87	0.3	0.34
  Peak phase error (°)	6.08	6.13	6.18	4.16	4.06

  .
• Case 4: A frequency jump of 5 Hz with DC-offsets similar to case three with harmonics, listed in Table 4, is added at 0.02 s for $d=T/4, T/16$, and $T/32$. The results are shown in Figure 16, Figure 17 and Figure 18, respectively, and summarized in

  Table 8. The results of case 4 (Frequency Jump with Harmonics and DC-Offset).

  	The Proposed PLL
  	n
  	32	16	4	$\alpha \beta$MAF-PLL	HPLL
  6 Hz frequency jump with harmonics and DC-offset
  2% frequency settling time(ms)	14.8	15.7	18.1	37.1	41.1
  Peak phase error (°)	11.53	12.8	14.57	19.57	17.35
  Peak frequency error (Hz)	0.15	0.15	0.13	0	0.26

  .

4. Discussion

Table 5, Table 6, Table 7 and Table 8 summarize the performance comparisons between the proposed PLL, the αβMAF-PLL, and HPLL. The comparison is made considering the phase settling time, the frequency settling time, the overshoot, the estimated frequency peak, and the phase error peak. In the first case, it can be noticed that the proposed PLL has the fastest dynamic response reaching the steady state in less than 0.02 s for any delay factor except $\tau =T/4$ for which it needs about 1.1 grid cycles, while the other PLLs need about two grid cycles to settle down. The phase percent overshoot is slightly better than those of the other methods. In the case of frequency jump, the proposed PLL has the fastest dynamic response and reaches the steady state in less than one grid cycle without overshooting for any delay factor. Concerning DC offset, the proposed method rejects the DC offset two times faster than the other methods. The peak frequency and phase errors are almost the same as the other methods. In the last case—the frequency jump with DC-offset and harmonics—the proposed PLL achieves the fastest response with less peak phase error with respect to other methods, synchronizing with the grid in a fraction of a grid cycle.

5. Experimental Verification

To demonstrate the performance enhancement of the proposed PLL and the LF design with comparisons to HPLL and αβMAF-PLL, digital implementation of all the PLLs is made utilizing the DE2-115 development and education board from Altera. An AC power supply is used to generate the voltage signal. The experiment maintains the grid voltage amplitude (V) and sampling frequency ($f_S$) at 1 pu and 10 kHz, respectively. The nominal grid frequency is 50 Hz. The results are captured using RIGOL MSO5354 mixed-signal oscilloscopes. Two cases are considered adopting $d=T/4$ without loss of generality.

• Case A: From the ideal grid condition, three disturbances occur simultaneously, including the dominant harmonics listed in Table 4 with DC-offset (0.1 pu to phase a, −0.1 pu to phase b, and 0.05 pu to phase c) and a frequency jump by 6 Hz. The results of the phase error and the estimated grid frequency of all the PLLs are shown in Figure 19.
• Case B: The frequency jump in case A is replaced by a phase jump of 45° while keeping the harmonic and DC-offset. The results of the phase error and the estimated grid frequency of all the compared PLLs are shown in Figure 20.

The experimental results agree with the simulation, which validates the proposed method.

6. Conclusions

A new approach for designing the loop filter of a third-order phase-locked loop is proposed in this paper. The proposed PLL consists of a moving average filter and an arbitrarily delayed signal cancelation. The arbitrary delay signal cancelation blocks the DC offset. In contrast, the moving average filter, with a window length of one-sixth of the fundamental grid period, blocks non-triple odd harmonics. The remaining triple harmonics are blocked utilizing the line voltage. The loop filter design is based on a modified third-order polynomial derived from stochastic optimization with the settling time of the frequency deviation as the objective function. The effectiveness of the proposed PLL and the adopted loop filter design method is verified by comparing their performance with other related PLLs, demonstrating the offered improvements. The simulation and experimental results show that the proposed PLL achieves synchronization within one grid cycle two times faster than the other PLLs. Therefore, the proposed PLL can be used to ensure that the renewable energy source can be safely and efficiently integrated into the grid.