You are on page 1of 10

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication.

Phase Locked Loop Based on an Observer for Grid Synchronization


Yongsoon Park

Seung-Ki Sul

Woo-Chull Kim

Hyun-Young Lee

Student member, IEEE


Seoul National University,
Seoul, Korea.

Fellow, IEEE
Seoul National University,
sulsk@plaza.snu.ac.kr

LG Uplus Corp.,
Seoul, Korea.

PLASPO Co., Ltd.


Goyang City, Korea

Abstract A grid synchronization technique is essential for


grid-connected power converters. Phase locked loop (PLL) has
been widely exploited as an implementation technique, but
additional efforts are required to deal with severely distorted
grid voltages. In this paper, an observer is proposed to enhance
the performance of PLL. A state equation is newly formulated
to construct the observer, which extracts positive-sequence
voltage from the distorted grid voltage. Additionally, guidelines
are suggested to set the observer gains by considering its
internal transfer functions. The PLL and proposed observer
have been tested in simulations and experiments in contrast to
dual second order generalized integrator frequency locked loop
(DSOGI-FLL). The results have shown that the proposed
method reveals better phase-tracking performance for grid
synchronization.
Index terms digital signal processing, grid-tied converter,
grid synchronization, observer, phase locked loop, positivesequence voltage.

I. INTRODUCTION
Electricity is delivered from suppliers to consumers via an
electrical grid. The grid voltage is expected to be sinusoidal,
whose frequency is fundamentally 50 or 60 Hz. This normal
voltage is called a positive-sequence voltage. In particular,
electrical power is mainly transferred by modulating the AC
current corresponding to the positive-sequence voltage, and it
is useful to separate this from the rest of apparent power [1].
The grid voltage phase angle should be detected to
elaborately modulate active and reactive powers when a gridconnected converter participates in power delivery. For this
purpose, the method based on phase locked loop in the
synchronous reference frame (SRF-PLL) has been widely
used [2], [3]. However, since the grid voltage is usually
polluted by unexpected distortions such as harmonics,
unbalances, and glitches, the detection of its phase angle may
not be easy with the simple PLL method.
Therefore, many attempts have been made to enhance the
detection performance under polluted grid conditions [4]-[15].
When the recent researches are considered, a better method
for grid synchronization is to fulfill the following
requirements simultaneously. First, the mitigation of
unbalance and harmonics should be achieved under higher
dynamics to track the phase angle variation. Second, the
consistency of grid synchronization has to be guaranteed
even if grid frequency varies.
In general, each required function can be implemented in
several stages with filter-type blocks [4]-[9]. In some cases,
the implementation can be based on parallel structure [12],
[13]. Even if all these methods have already presented

effective performances in functional aspects, the


computational efficiency is not high due to the distributed
structures. Moreover, the integrated system design becomes
more complicated when the number of separate parts
increases. By using a specialized filter, the rejection at every
harmonic has been attempted with the relatively simple
structure [14], [15]. However, this method should save the
previously sampled values, and its sampling frequency must
be set carefully in conjunction with the grid frequency.
Therefore, an appropriate compromise is required between
the implementation effort and the grid synchronization
performance.
In this paper, an observer is proposed to clearly extract the
positive-sequence component of the grid voltage. The
required functions for grid synchronization can be unified
into a single observer by virtue of its state equation, which
has been newly derived in this paper. As a result, the
proposed method has a relatively simple structure while
presenting the phase detection performance comparable to
the dual second order generalized integrator frequency locked
loop (DSOGI-FLL) [10].
In terms of design and analysis, the following points can
be further noted: although the observer gains are generally set
on the basis of the heuristic insight of the designer, this paper
suggests explicit guidelines in setting the gains considering
the internal transfer functions. In addition, the analyzing tools
are newly introduced to discuss the unified effect of the
proposed 4th order observer and to examine the distortions
from digital implementation.
The organization of this paper is as follows. In section II,
how the observer can extract positive-sequence voltage with
less distortion is described. After the observer gains are set,
the observer to PLL attachment is detailed in section III. The
proposed method is then assessed via simulation and
experimental results under grid fault situations in section IV.
Finally, concluding remarks are noted in section V.
II. EXTRACTION OF POSITIVE-SEQUENCE VOLTAGES
A. Definitions of Symmetrical Components
In general, the three-phase voltage can be decomposed
into symmetrical components: positive-, negative-, and zerosequence voltages [18]. Among them, negative-sequence
voltage arises mainly when the grid is under abnormal
operating conditions such as phase-to-ground faults and
unbalanced loading conditions. In SRF-PLL, it is hard to
filter out the negative-sequence voltage in the synchronous
reference frame even though the bandwidth of PLL is quite
reduced. [3], [4]. To effectively remove the negative-

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

sequence volttage, the symm


metrical compponent modeling can
be consideredd.
Initially, thhe positive-seq
quence voltagee can be definned as
v a+ = V+ sin p

v b+ = V+ sin(( p 120)

v c+ = V+ sin(( p + 120)

(1)

where both thhe subscripts + and p rrefer to the ppositivesequence, and p is the phase
p
angle of
o positive-seequence
voltage.
As per the definition off (1), the positive-sequence voltage
drature axis of
o the synchhronous
only exists in the quad
me. This defin
nition has been
n adopted to m
maintain
reference fram
the coherencee between griid and AC-mo
otor applicatioons. By
using the sam
me phase ang
gle in (1), thhe negative-seequence
voltage can thhen be generallized into
v a = V sin( p + n )

v b = V sinn( p + n + 120)

v c = V sin( p + n 120)

(2)

where the subbscript represents the neegative-sequen


nce, and
n is the phaase differencee of negative--sequence volltage to
positive-sequeence voltage.
For exampple, each voltaage in (1) and
d (2) can be depicted
d
in terms of thhe phase anglee as shown in Fig. 1. In thee figure,
V+ in (1) wass set to 179.6 V, and V in (2) was 30 % of V+.
In addition, n in (2) was seet to 30 . The sum of positiive- and
negative-sequuence voltagess presents an unbalance ass shown
in Fig. 1.
mating Positivee-Sequence Vo
oltage
B. An Obserrver for Estim
For the observer design,, the state equ
uation on grid voltage
must first bee established. This state eq
quation explaains the
relationship between
b
statess and their derrivatives. In eessence,
the grid voltaage should bee discussed in
n terms of itts timevariation. In particular,
p
the grid voltage can be considdered in
the synchronoous reference frame
f
by
v
20) cos( p + 1120) a
v d 2 cos p cos( p 12

v =
vb
20) sin( p + 120)
q 3 sinn p sin( p 12
vc

(3)

where the hat ^ hereafter refers


r
to estim
mated value .
For each phase,
p
the sum
m of (1) and (2) can be insertted into
the right-sidee of (3). Th
he zero-sequeence voltage is not
considered, beecause it is naaturally erased
d in the processs of (3).
The d-q voltaages can also be decomposed into positivve- and
negative-sequuence compon
nents in the sy
ynchronous reeference
frame:
s d V sin (2 p + d + n )
v d v d+ vd V+ sin
(4)
v = v + v =
+
os d V ccos (2 p + d + n )
q q+ q V+ co

where d = p p .

(5)

Figg. 1. Positive- andd negative-sequen


nce voltages.

If dd/dt andd dn/dt are negligible


n
undder the assum
mption
of slow variationn in the estimaation error and
d phase difference,
thee derivative of
o (4) can be approximatedd by the chain
n rule
intto
V cos (2 p + d + n )
d vd d vd
d
.
(6)
v v = 2 p
dtt q dt q
dt
V sin (2 p + d + n )

Each negativve-sequence component


c
in (4) is repeateed in
thee other axis inn (6). Equationn (6) is then rearranged into
vq
v q v q+
d vd
d
2 p
= 2 p v v
v

dtt v q
dt
d
d
d+

(7)

whhere is anguular frequency.


The novel deerivation of (7
7) has led to th
he proposal of this
papper. In (7), thhe derivativess of vd and vq can be explaained
wiith respect to themselves
t
annd the positivee-sequence vo
oltage
(w
without the negative-sequ
uence voltagge). Furtherm
more,
because it has been
b
assumedd that dd/dt is
i small enoug
gh to
d
of positive-seque
p
ence voltage in
i (4)
derive (7), the derivative
n, the entire state equation of
o the
cann be regardedd as null. Then
griid voltage cann be derived ass
p
p 0 2

0 2

p 0
p 0 2
2
d
Vs =
0
dtt
0 0
0

0
0 0
0

V = A V
m s
s

vd 1 0 0 0
Vo = =
Vs = Cm Vs
vq 0 1 0 0

Vs = vd

vq

vd+

vq+

(8-a)

(8-b)
(8-c)

whhere the subscripts s andd o refer to state and ou


utput,
resspectively.
A Luenburgeer observer cann be designedd as in (9), beccause
thee system is observable with Am and Cm inn (8) [16]:

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

d
+ L [V C V

Vs = Am V
s
m
o
m s]
dt
p
Lm = 1
p2

= v
V
s
d

p3
p4

v q

q1
q2

v d+

q3
q 4

(9-a)

H t = q 4s3 + (p1q 4 p 2q 3 )s 2

(9-b)

Pt = det[s I m A m + L m C m ]

v q+

(9-c)

where variables in Lm are observer gains.


By virtue of the practical assumption on positive-sequence
voltage, the observer in (9) could be newly proposed, which
can observe the positive-sequence voltage in its internal state.
C. Gain Settings for the Observer
The observer gains of Lm in (9) must be specified. In
general, it is difficult to logically describe why a set of
observer gains have to be selected since certain selections
may have to rely on the insight of the designer. In this paper,
the gain setting of the observer is described with the reason
why they are selected. The internal transfer functions of the
observer are used to explain the gain setting. These functions
can be derived as follows. Initially, the state equation in (9)
can be transformed into (10) when the derivative operator is
replaced with the Laplace operator:

=A V
sV
(10)
s
m s + L m [Vo C m Vs ] .
Equation (10) can be rearranged in terms of Vo:
= [s I A + L C ]1 L V
V
(11)
s
m
m
m m
m o
where Im represents the identity matrix.
Hence, the estimated states in (9-c) can be expressed with
respect to the measurable states in (8-b):
v d = A t /Pt vd + E t /Pt vq

v q = Bt /Pt vd + Ft /Pt vq
(12)

v d+ = C t /Pt v d + G t /Pt v q
v = D /P v + H /P v
t
t
d
t
t
q
q+
where all transfer functions in (12) are detailed with
A t = p1s3 + {2 p (p 3 q 3 ) + p1p 4 p 2 p3}s 2
p q1 )s + 4
p2 (q1q 4 q 2q 3 )
+ 2 p (p1q 2 p 2q1 + p 3q 4 p 4q 3 + 2

(13-b)

2pq 3s
p (q1 p1 )s 2 + 4
Bt = p3s3 + 2
3

Ct = q1s + (p 4q1 p3q 2 )s

(13-a)

2p (q1q 4 q 2q 3 )
p (p1q 2 p 2q1 + 2
p q1 )s + 4
+2

(13-c)

D t = q 3s 3 + (p 4q 3 p3q 4 )s 2
+ 2 p {q 3 (q 2 p 2 ) + q 4 (p1 q1 ) + 2 p q 3}s

(13-d)

2pq 2s
p (p4 q 4 )s 2 + 4
E t = p 2s3 + 2

(13-e)

p (q 2 p 2 ) + p1p 4 p 2 p3}s
Ft = p 4s + {2

p (p1q 2 p 2q1 + p3q 4 p 4q 3 + 2


p q 4 )s + 4
2p (q1q 4 q 2q 3 )
+ 2

G t = q 2s 3 + (p1q 2 p 2q1 )s 2
+ 2 p {q 2 (p3 q 3 ) + q1 (q 4 p 4 ) + 2 p q 2 }s

(13-f)
(13-g)

(13-h)

p q 4 )s + 4
2p (q1q 4 q 2q 3 )
p (p3q 4 p 4q 3 + 2
+2
= s 4 + (p1 + p 4 )s3 + 4 p2 (q1q 4 q 2q 3 )
p (p3 p 2 + q 2 q 3 ) + p1p 4 p 2 p3}s 2
+ {4 2p + 2

(14)

p (q1 + q 4 )}s
+ 2 p {p1q 2 p 2q1 + p3q 4 p 4q 3 + 2

Among the transfer functions in (12), the most important


factors are Ct/Pt and Ht/Pt, which explain the influence of
each axis voltage to its estimated positive-sequence voltage.
Therefore, these transfer functions have been crucially
considered in setting the observer gains. Initially, the roll-off
rate of Ct/Pt and Ht/Pt can simply be increased by setting q1
and q4, the highest order coefficients in the numerators, to
zero:
(15)
q1 = q 4 = 0 .
The settings of (15) are intended to enhance the filtering
performance to high-frequency distortions. By (15), Ct and Ht
are simplified into
2pq 2q 3
p p1q 2s 4
(16)
Ct = p3q 2s 2 + 2
2pq 2q 3 .
p p 4q 3s 4
H t = p 2q 3s 2 2

(17)

To prevent non-minimum phase responses, all coefficients


must have the same sign in each of (16) and (17). This
condition causes the roots of each equation to be placed in
the left half plane (LHP) and is presented as
(18)
p1 / p 3 < 0, q 3 / p 3 > 0, p 4 / p 2 > 0, q 2 / p 2 > 0 .
Meanwhile, the observer poles can be determined by using
the concept of the general second order system as follows:
Pset = (s 2 + 21n1s + 2n1 )(s 2 + 2 2n2s + 2n2 )
(19)
where x is damping ratio and nx is natural frequency.
After the damping ratios and natural frequencies are
determined, the observer gains are set by comparing the
coefficients between (14) and (19). Considering that all the
coefficients are positive in (19), the condition in (20) must be
satisfied in the coefficient comparison after (15) is applied:
(20)
p1 + p 4 > 0, q 2q 3 < 0 .
The next design point is to set the observer gains so that
the observer structure is symmetric in the synchronous d-q
reference frame. This can contribute to the simple
implementation of the observer, because similar gains can be
repeated in the d-q reference frame, as shown in Fig. 6. The
symmetric structure can be achieved with (21) when the
conditions of (15), (18), and (20) are assumed:
(21)
p1 = p 4 , p 2 = p 3 , q 2 = q 3 .
By the settings of (15) and (21), the equations of (13) and
(14) are simplified into (22) and (23), respectively. That is,
the internal transfer functions in the d-axis become similar to
those in the q-axis.
p (q 2 p 2 ) + p12 + p 22 }s 2
A t = Ft = p1s3 + {2
p2 q 22
+ 4 p p1q 2s + 4

(22-a)

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

(22-b)

p p1s 2 + 4
2p q 2s)
Bt = E t = (pp 2s3 + 2
2

2
p

p p1q 2s + 4
q
C t = H t = p 2q 2s +2
3

(22-c)

2
2

p )s}
D t = G t = {qq 2s + p1q 2s + 2
2 p q 2 (q 2 p 2 + 2
4

Pt = s + 2p1s + {4 2p + 4 p (q
q 2 p 2 ) + p12 + p 22 }s 2

(22-d)

+ 4 p p1q 2s + 4 2p q 22

(23)

The obserrver gains in (23) can theen be specifieed with


respect to predetermined
p
d damping ratios and natural
frequencies inn (19) by thee coefficient comparison.
c
B
Because
harmonic distortions comm
monly occur at multiples of grid
w
be convvenient to set the
t natural freequency
frequency, it would
to be proportiional to the griid frequency:

n1 = k1 p , n2
(24)
n = k 2p
where p is estimated
e
gridd frequency. p has alreaddy been
used for the design
d
since (77).
For (21) annd (24), the co
oefficient com
mparison betweeen (19)
and (23) resullts in
p
p1 = p 4 = (1k1 + 2 k 2 )
(25)
2p
2p + 4
p (q 2 p 2 ) + p12 + p 22 = (k12 + k 22 + 41 2 k1k 2 )
4

(26)

p p1q 2 = 2k1k 2 (1k 2 + 2 k1 )


3p

(27)

4p .
2p q 22 = k12 k 22
4

(28)

If p1 in (255) is inserted innto (27), q2 is uniquely deteermined.


This resultantt q2 must thenn satisfy (28), which is rearrranged
into (29) by thhe substitutionn of the resultaant q2:
2

1k 2 + 2 k 1

=1.
1k 1 + 2 k 2

(29)

Because each
e
value inn (29) has been
b
assumedd to be
positive, (29) is further sim
mplified into
(30)
( 2 1 )(k1 k 2 ) = 0 .
When the observer polees are placed according
a
to (19), the
o (15) and (21) are used onnly if (30) is saatisfied.
gain settings of
One of two factors
f
in (30)) must be zerro at least forr this to
work. When (26) is considdered in addition to (30), eequation
s
if 1 is equal to 2 while (32) dooes so if
(31) must be satisfied
k1 is equal to k2:
p 2)2 = (k1 k 2 )2 (1 12 )
(p 2 /
(31)
p 2) 2 = k12 (1 2 ) 2 .
(p 2 /

(32)

It is impoortant to remeember that evvery value iss a real


number in (311) and (32). According
A
to (31), k1 and k2 can be
different under (33)s connstraint even if 1 and 2 are the
same. Howevver, 1 must be
b identical to 2 in (32) if k1 is
equal to k2. Therefore, thhe setting forr (31) is seleected to
d
of freed
dom in the obsserver poles.
increase the degree
2
(33)
(1 1 ) 0 1 1 1 .
In a generral second orrder system, ooscillatory ressponses
would occur if the dampping ratio is smaller thann unity.
Therefore, in (33), the dam
mping ratio couuld be set to uunity so
the observer gains
g
in (21) are
a finally speccified as

Fig
g. 2. Pole placemeents of the propossed observer acco
ording to k and .

p
p1 = p 4 = (k1 + k 2 )

p
.
p 2 = p 3 = 2

=
q
q
k
k

/
2
3
1 2 p
2

(34)

D. Pole Placem
ment of the Ob
bserver
Considering (12) and (22),
(
the esstimated posiitivequence voltagge is expressed
d by
seq
v d+ = Ct /Pt vd Dt /Pt vq
(35)
.

v q+ = Dt /Pt vd + Ct /Pt vq
The substituution of (36) makes the transfer funcctions
sim
milar to the geeneral second order
o
system:
(36)
k1 = k, k 2 = k .
d
as (37) and
The transfer functions in (35) are then derived
(38
8) after reduciing common factors:
f
(k
( p )2
Ct
= 2
ps + (k
p )2
Pt s + (1 + )kk

(37)

k 2 ps
Dt
1
= 2
.
ps + (k
p )2
Pt
2 s + (1 + )k

(38)

The meaningg of k and can be undersstood from Fiig. 2,


whhere the locattions of obserrver poles are depicted with
h the
speecific values of k and . All poles are
a located on
n the
negative real axxis pertaining to (31) and (3
33). When k is
i set
firrst, two poless are placed at
a a distance ktimes the grid
freequency. Theen, the remaaining two poles
p
are pllaced
acccording to settting , whichh is the relatiive distance to
o the
firrst placement.
o the
By the propoosed setting inn this paper, thhe operation of
4thh order obserrver shown inn (9) can be described
d
with
h the
2nnd order transffer functions in
i (37) and (3
38). This featu
ure is
very helpful in adjusting thee observer reesponses sincee the
m can be understood with simple modeeling.
oriiginal system
Hoowever, it is not clear what
w
the uniffied effect off the
pro
oposed observver is with only (35), (37), and
a (38).
To analyze the unified effect,
e
the folllowing methood is
inttroduced based on the fact that
t
the deriveed system in (8)
( is
a linear
l
system
m. Since the superposition
s
principle hold
ds in
linnear systems, the response to three-phasse voltage at each

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

frequency cann be separately considered. Considering (1) and


[14], the threee-phase voltagge can be geneeralized into
vah (t) = Vh ssin(hp t)

s
v bh (t) = Vh sin(h
p t 120)

vch (t) = Vh ssin(hp t + 120)

(39)

where h cann be any real value.


v
Through Parks
P
transforrmation in (3)), the generalized d-q
voltage can bee derived as
v dh (t) Vh sin((h 1)p t) Vh sin( t t)
=
v (t) =
.
qh Vh cos((h 1) p t) Vh cos( t t))

(40)

If the Laplace transform


m is applied to (40), equationn (41) is
established beetween the d- and
a q-axis volltages:
v dh (s) =

t
v qh (s) .
s

(41)

Fig
g. 3. Frequency reesponse of the observer to three-ph
hase voltage.

Based on (35) and (41


1), the frequeency responsee to the
three-phase voltage at the arbitrary
a
frequuency, denotedd by t,
can be derivedd as
s

/ t D t /Pt ) vdh (s) = Td (s) vdh (s)


v d+ (s) = (C t /P
t
.

t
v (s) = (D /P
/

+
C
/P
)

v
(s)
=
T
(s)

v
(s)
t
t
t
t
qh
q
qh
q+
s

(42)

The Bode plot of Td annd Tq in (42) can be obtaiined by


replacing s wiith jt as show
wn in Fig. 3. Due
D to the sym
mmetric
design of the observer, the Bode plot of Td is the samee as that
h
the propposed observerr works
of Tq. It becoomes evident how
in frequency domain.
d
As shown in Fig. 3, thee proposed obsserver fundam
mentally
w-pass filters in the synchrronous d-q reeference
serves as low
frame while offering
o
the band-stop
b
funcction simultanneously.
In particular, the bandwidthh of low-pass filtering is sett by the
pole placemeent, and the sttop band is loocated at douuble the
estimated freqquency. The low-pass
l
functtion originates in the
common propperty of obserrver while thee band-stop ffunction
results from the proposed
d modeling oof (7) to reject the
negative-sequuence voltage.
Although both filtering
g functions are
a affected by the
estimation errror of the grid
d frequency, thhe movement of stop
band is more critical. Thenn, it has been ddiscussed in teerms of
the grid freqquency error as shown inn Fig. 4. Whhen the
negative-sequuence voltage was considereed in (42) (t = 2p),
the estimatedd frequency in
n (37) and (388) has been cchanged
within the freequency errorr of 20%. As a result, evenn if the
estimated freqquency deviatees from the grrid frequency bby 20%,
the magnitudee of negative--sequence volltage is still exxpected
to be mitigateed by the obserrver by up to oone tenth.
IIII. PLL BASEED ON THE OBSSERVER
The observver proposed in section II is attached tto SRFPLL. By virtuue of this connstruction, filttering harmonnics can
be readily acchieved with less effort inn PLL, which allows
PLL to increaase its bandwiddth and contribbutes to the hiigher

Fig
g. 4. Deterioration
n of band-stop fun
nction by grid freequency error.

Fig
g. 5. Entire PLL structure.
s

phhase-angle traccking dynamiics. The entiree PLL structu


ure is
shoown in Fig. 5 and is referreed to as Synchhronous ObseerverAiided Preprocesssing phase locked loop (SO
OAP-PLL).
The estimateed frequency in
i PLL must be fed back to
t set
thee observer gains accordingg to (34). Usuually, e2 in Fig.
F 5
has been used as the estimateed frequency. However,
H
e22 may
be sensitive to distortions
d
beccause kpp, the proportional gain,
dirrectly deliverrs scaled anggle errors to it. To refinee this
draawback, e1 in Fig. 5 is used as thhe estimated grid
freequency. The effectiveness of this mo
odification caan be
disscussed with the followiing equationss. When (43
3) is
asssumed, (44) caan be derived from Fig. 5:
d d = p p
(43)
p =

k pps + k ip
s 2 + k pps + k iip

(44)

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

where kpp andd kip are propoortional and inntegral (PI) gaains for
PLL, respectivvely.
From Fig. 5 with (43), e1 is derived as
a
k ip
k ip
(45)
d
( p p ) .
s
s
If p in (444) is inserted into
i
(45), it caan be rearranged into

e1 =

e1

k ips
s 2 + k pps + k ip

p =

k ip
s 2 + k pps + k ip

p .

(46)

e1 is a loow-pass filtered value of actual


a
frequenncy. By
using this freqquency as p , the observerr performancee can be
less sensitive to distortionss. In addition, because the ttransfer
function in (446) takes the form of the ggeneral seconnd order
system, the PI gains can bee set accordinng to the settinng rules
of the second order system::
2
(47)
k pp = 2 pll npll , k ip = npll
where pll annd npll are the dampingg ratio and natural
frequency of P
PLL, respectivvely.
The gain settings
s
of thee observer andd PLL have tto come
together, so all
a the gains arre set where thhe bandwidth of (37)
is larger than that of (44). Empirically,
E
thhe bandwidth of (37)
is set to four times greater than that of ((44). If the doominant
n larger thaan that of PL
LL, the
pole of the observer is not
w
be distuurbed.
dynamics inteended in (44) would
Meanwhilee, based on thhe equation obbtained from ddividing
both sides of (10) by s,, the propossed observer can be
specifically im
mplemented, as
a shown in Fig.
F 6. In the process
of obtaining Fig. 6, thee observer gaains have beeen set
according to (15), (34), annd (36). Even though the oobserver
ms to be raather compliccated, the ppractical
design seem
implementatioon of the prroposed obseerver is simpple. As
discussed earlier, this singlle observer caan offer the loow-pass
multaneously.
and band-stopp functions sim
IV. SIM
MULATION AND
D EXPERIMENT
TAL RESULTS
A Evaluatioon of SOAP-P
A.
PLL
The perforrmance of SO
OAP-PLL can be discussedd in two
aspects: the tracking error on grid parameters and
a
the
mitigating abbility against distortions.
d
Eaach performannce has
been examineed under grid faults
f
where thhe grid voltages were
contaminated by harmoniccs and unballances, and thhe grid
frequency was changed.
AP-PLL is discussed
In particular, the effectivveness of SOA
LL. This methhod has been sselected
as compared to DSOGI-FL
because its im
mplementation
n is relatively simple, and eessential
gain settings are
a known [10
0].

Fig
g. 6. Block diagraam of the proposeed observer.

Fig
g. 7. Grid fault un
nder conditions off (48), (49), and (50).
(

B. Simulation Results
The fault connditions for sim
mulation testss have been qu
uoted
froom the literatuure [12]. Whhen the voltagge amplitude at an
ideeal grid is sellected as the base voltage, the phasor of
o the
noormal voltage is denoted ass 10. Basedd on this conncept,
thee fault conditioons are given as follows:
(48)
V1+ = 0.5 30, V1 = 0.251110
(49)
V5 = 0.20, V7 + = 0.20, V11 = 0.20
(50)
f fauult f nor = 5Hzz
whhere the subsccript number of
o a voltage reepresents harm
monic
ord
der, and the normal
n
conditiion is indicateed by the subsscript
no
or.
2
z, and
When the norrmal grid is asssumed to be 220V
rms-60Hz
alll the conditionns in (48), (49)), and (50) aree applied after fault,
thee fault wavefoorms are shown in Fig. 7.
Initially, the gains of DSO
OGI-FLL werre set accordin
ng to
earrlier work [100]. For a fair comparison
c
off performancess, the
gains of SOAP-P
PLL had to bee set such thatt its dynamics were
sim
milar to those of DSOGI-FL
LL under the unbalance fau
ult of
(48
8). The specifi
fic values of th
hose gains werre
(51)
= 1, k = 1.7, plll = 1, npll = 2 20 .
The gains of and k are diirectly inserted into the observer
of Fig. 6, and the
t gains of pll and npll are
a reflected in
n the
LL of Fig. 5 through the PI gains in (47). With these
PL
speecific values, SOAP-PLL has
h presentedd the same setttling
tim
me under the condition
c
specified in (48), as
a shown in Fiig. 8.
Due to the sttep variations in (48), the estimation erro
ors of
phhase angle andd frequency present
p
oscillaations for a while
w
aftter the fault, as captured in Fig. 8. However,
H
afterr the
osccillations disappear, both methods shhows perfect grid
synnchronizations as if there are no unbaalance distorttions,

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

which confirm
ms that the unnbalance rejecttion of SOAP-PLL is
comparable with
w that of DS
SOGI-FLL.
The perforrmance of SO
OAP-PLL wass different to that of
DSOGI-FLL when each method wass implemented in a
i (49) were inncluded
discrete time domain, or thhe harmonics in
d
implem
mentation of DSOGID
in fault condditions. The digital
FLL was baased on [21] while SOA
AP-PLL was simply
implemented with backw
ward Euler method.
m
Withh these
implementatioons under 10 kHz samplinggs, each methhod was
tested under the
t fault condditions of (48)), (49), and (550). The
simulation ressult is presenteed in Fig. 9.
As shown in Fig. 9, SOAP-PLL obbviously outpeerforms
DSOGI-FLL in harmonic filtering. Thiis is because SOAPPLL consists of two filteriing stages: the observer annd PLL.
t filtering peerformance off DSOGI-FLL
L can be
Even though the
improved by extending to multiple
m
SOG
GI-FLL (MSOG
GI-FLL)
[12], this exttension can onnly be done bby increasing source
code lines forr implementattion. In additiion, the frequeency of
DSOGI-FLL presents the average errorr of 0.79 Hz mainly
due to harmonnics.
The other conspicuous feature
f
is thatt DSOGI-FLL
L shows
the steady-staate error of 2.7 in estimatting the phasee angle.
To discuss thiis phenomeno
on, one of the dual SOGI iss shown
in Fig. 10. To estimatee positive-seqquence voltagge, the
following trannsfer functionss are importannt in Fig. 10:
D(s) =

ps
k

v
=
2p v
s + k ps +

(52-a)

Q(s) =

2p
k

qv
=
ps +
2p
s + k

(52-b)

where the subbscript reffers to the d-aaxis in the staationary


reference fram
me. Q(s) in particular
p
geneerates the quaadrature
signal of inpuut while both
h D(s) and Q(s) serve as ffiltering
harmonics.
When a syystem is transfformed from thhe s-domain to the zdomain, its orriginal properrty can be disttorted. Moreovver, the
computationaal delay of z-1
, which is indicated
i
by dashed
circles in Figgs. 6 and 10, may
m be includded for feedbaack. To
specify thesee types of distortions, the
t
z-transforrmation
distortion (ZT
TD) is introducced:
H (e js )
ZTD H ( ) = z
H s (j)

(53)

where H sym
mbolizes the system of inteerest, Hs referrs to the
original systeem in the s-domain, andd Hz represents the
transformed system
s
in thee z-domain. In addition, is the
frequency of concern
c
and Ts is the samplling period forr digital
implementatioon.
The meanning of ZTD is the distortted degree off the zdomain system
m from the s--domain systeem. The Bode plot of
ZTD correspoonds to the diifferences betw
ween the Bodde plots
of each system
m:
20 log10 | ZTD H () |= 20 log10 | H z (e j ) | 20
0 log10 | H s (j) | (54)
s

Fig
g. 8. Simulation reesults under unbaalance fault of (48
8) with analog
imp
plementations.

Fig
g. 9. Simulation reesults under faultt conditions of (48
8), (49), and (50)) with
dig
gital implementatiions.

Fig
g. 10. Second order generalized inttegrator.

ZTD
Z H ( ) = H z (e js ) H s (j) .

(55)
The Bode plots
p
of critical ZTDs inn SOAP-PLL and
DS
SOGI-FLL arre depicted in
n Fig. 11 whhere the sam
mpling
freequency is seet at 10 kHzz. In the Bo
ode plot of ZTD,
Z
maagnitude and phase must be
b 0 dB and 0 , respectiively,
whhich corresponnd to unity, whhen the z-trannsformation is ideal.
Ass shown in Figg. 11, the ZTD
Ds deviate froom the unity when
w
thee frequency of concern approaches to the Ny
yquist
freequency.
n for the steeady-state erroor in
From Fig. 111, the reason
DS
SOGI-FLL and not in SOAP
P-PLL can be explained. SO
OAPPL
LL and DSOG
GI-FLL are diff
fferent theoretiically since SO
OAPPL
LL can sepaarate the positive- and negative-sequ
uence
vooltages via Paarks transform
mation. By thhis separation
n, the

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

negative-sequuence voltagee appears aat double thhe grid


frequency whhile the positive-sequence voltage
v
is at the DC
band. In conttrast, DSOGI--FLL deals with
w the positivve- and
negative-sequuence voltagess in the stationnary referencee frame,
where both vooltages appearr at the grid freequency.
As shownn in Fig. 11, thhe ZTDs of D
D(s) and Q(s) in (52)
exhibit the phhase distortionns of 2.2 and 3.2 degrees at tthe grid
frequency, where
w
the posiitive-sequencee voltage apppears in
DSOGI-FLL. For AC signnals, these noonzero phases mean
parallel displaacements in the
t time-domaain. Then, thee phase
distortion of ppositive-sequeence voltage can
c cause the steadystate error in DSOGI-FLL,, and can be iintensified at smaller
sampling frequencies as shown in Fig. 11 (denooted by
c
the ZTDs of SOAP
P-PLL presentt almost
@5kHz). In contrast,
unity at the DC
D band, whhere the positiive-sequence voltage
appears in SO
OAP-PLL. In other
o
words, the
t positive-seequence
voltage wouldd not be distoorted in SOAP
P-PLL. Althouugh the
observer funcction is slighhtly distorted at double thhe grid
frequency as shown in Fig.. 11, the effecctiveness of thhe bandstop functionn in SOAP-PL
LL can be coorroborated w
with the
result of Fig.. 9. Unlike SOAP-PLL,
S
caareful consideerations
are required for
f digital impplementation of
o DSOGI-FL
LL [20][22].
a been inclluded in
In additionn, the conditionn of (50) has also
the fault test of Fig. 9 to
o examine thee frequency-aadaptive
OAP-PLL. Beecause the seccond order disttortions
property of SO
are not observved often in Fig.
F 9, the SOA
AP-PLL succeessfully
rejects negattive-sequence voltage evven after freequency
variation.
C. Experimeental Results
To demonstrate the feassibility of SOA
AP-PLL, a grrid fault
was emulatedd using an AC programmablle source, MX
X30. All
algorithms were
w
implemennted in a diggital signal prrocessor
(DSP), TMS320F28335, and
a
grid volttages were ssampled
through voltaage sensors per 100 s. All the gains for SOAPPLL were set according to (51).
(
f
fault is the single phase-to-groun
p
nd fault
The most frequent
[17], which iss commonly recognized
r
as phase-to-phaase fault
by grid-conneected converteers due to traansformers [19]. The
phase-to-phasse fault betw
ween b- and c-phases haas been
considered:
Va = 1

Vb = 1 / 2 jVsag 3 / 2

Vc = 1 / 2 + jVsag 3 / 2

(56)

where Vsag iss a complex number indiccating the degree of


voltage sag. Inn addition, all magnitudes aare in per unit.
The magnnitude of Vsagg is determined according to the
phase of Vsag [19]. As an exxperiment, Vsaag was set to 0
0.38
40 . The magnitudes
m
of harmonics att 5th, 7th, annd 11th
orders were set to about 0.08 per unitt, and the freequency
variation in (550) was includded as well. Thhe grid voltagees

Fig
g. 11. Bode plots of ZTDs in SOAP
P-PLL and DSOG
GI-FLL under 10
0 kHz
sam
mplings.

Fig
g. 12. Grid fault for
fo experiment.

aroound the faultt occurrence are


a shown in Fig.
F 12, which are
thee voltages recoognized by thee DSP board.
Initially, the c-phase
c
voltagge, whose neggative peak waas the
staarting point off the fault, wass selected as th
he reference signal
s
to capture severral waveform
ms from the liimited channeels of
osccilloscope. Itt was assumeed that the AC source could
c
gennerate almostt time-invariannt faults accoording to the same
setttings.
f
situationn, SOAP-PLL
L has the feewest
Under the fault
harmonics in thee frequency estimation show
wn in Fig. 13. The
ver can be moore obvious viia the
efffect of the prooposed observ
comparison withh SRF-PLL, where
w
only thee observer parrt has
been eliminatedd from SOAP
P-PLL. As shown in Fig. 13,
negative-sequennce distortionns were eviddently rejectedd by
OAP-PLL wheen compared with SRF-PL
LL. After trannsient
SO
situuations, the RMS of freqquency ripplee was 0.1 Hzz for
SO
OAP-PLL, 1.552 Hz for SRF
F-PLL, and 0.4
45 Hz for DSO
OGIFL
LL. Furthermoore, the frequency of DSO
OGI-FLL preseented
thee average deviation of 0.102 Hz when coompared to SO
OAPPL
LL.
a
Because the actual phase angle was noot accessed, angle
errrors were calcuulated with reespect to the ph
hase angle of

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

[2]

[3]

[4]

[5]
Fig. 13. Frequenccy estimations in experiment: timee [10 ms/div].

[6]

[7]

[8]

[9]
Fig. 14. Angle esstimations in expeeriment: time [10
0 ms/div].

SOAP-PLL, SOAP, when considering thhe simulationn result.


After transiennt situations, the phase anngle of DSOG
GI-FLL,
DSOGI, deviatted from that of SOAP-PLL
L by 2.4 , ass shown
in Fig. 14.
V. CONCLUSION
In this papper, an obserrver has been proposed to clearly
extract positiive-sequence voltage from
m the polluteed grid
voltage. This extraction is intended to reeinforce the ffunction
of PLL by using the obseerver as a preeprocessor. Thhe state
b
newly deerived on gridd voltage to coonstruct
equation has been
the observer, and the gain settings of thee observer havve been
suggested whhen considering the mitigatioon of grid disttortions
and their com
mbination with PLL.
In a functioonal aspect, thhe proposed PLL
P method hhas been
specifically scrutinized in terms of filteering harmonnics and
rejecting unbbalances undder polluted grid conditioons. In
particular, thee performancees have been ddiscussed in ddigitally
implemented structures unnder the consiideration of ppractical
implementatioon. As a resultt, the effectiveeness of the prroposed
method has been
b
confirmeed by simulatiion and experrimental
results. The proposed method
m
exceelled in mittigating
distortions of the estimaated frequenccy and phasee angle
without steadyy-state errors.
REFERENCES
E
[1]

A. E. Eman
nuel, Summary of
o IEEE Standard
d 1459: Definitionns for the
Measuremeent of Electric Power Quanttities Under Siinusoidal,
Nonsinusoiidal, Balanced, or
o Unbalanced Conditions,
C
IEE
EE Trans.
Ind. Appl., vol. 40, no. 3, pp
p. 869-876, May/JJune 2004.

0]
[10

[11
1]

[12
2]

[13
3]

[14
4]

[15
5]

[16
6]
[17
7]

[18
8]

[19
9]

[20
0]

V. Kaura and V. Blasko, Opeeration of a Phasee Locked Loop System


S
Under Distorteed Utility Condittions, IEEE Tra
ans. Ind. Appl., vo
ol. 33,
no. 1, Jan./Feb
b. 1997.
S.-K. Chung, A Phase Traccking System fo
or Three Phase Utility
U
Interface Inverters, IEEE Trans. Power Electrron., vol. 15, no. 3, pp.
431-438, May
y 2000.
S.-J. Lee, H. Kim
K and S.-K. Su
ul, A New Phasee Detecting Meth
hod for
Power Conveersion Systems Considering Distorted Conditio
ons in
Power System
m, in Conf. Rec. IEEE
I
34th IAS An
nnu. Meeting, Occt. 3-7,
1999, pp. 21677-2172.
R. K. Sinha, aand P. Sensarma, A pre-filter baseed PLL for three--phase
grid connected
d applications, Electric
E
Power Systems
S
Research
h, vol.
81, Issue 1, pp
p. 129-137, Jan. 2011.
2
M. P. Kazmiierkowski, M. Jasinski,
J
and G. Wrona, DSP--Based
Control of Grrid-Connected Po
ower Converters Operating Under Grid
Distortions, IEEE
I
Trans. Ind. Informatics, vol. 7, no. 2, pp. 204
4-211,
May 2011.
F. Liccardo, P.
P Marino, and G.
G Raimondo, R
Robust and Fast ThreeT
Phase PLL Traacking System, IEEE Trans. Ind
d. Electron., vol. 58,
5 no.
1, pp. 221-231
1, Jan 2011.
F. Gonzalez-E
Espin, E. Figueeres, and G. Gaarcera, An Ad
daptive
Synchronous-R
Reference-Framee Phase-Locked Loop
L
for Power Quality
Q
Improvement in a Polluted Utiility Grid, IEEE
E Trans. Ind. Elecctron.,
vol. 59, no. 6, pp. 2718-2731, June
J
2012.
Y. F. Wang, and Y. W. Li, Grid Synchroniization PLL Bassed on
Cascaded Delaayed Signal Canccellation, IEEE Trans.
T
Power Eleectron.,
vol. 26, no. 7, pp. 1987-1997, July
J
2011.
P. Rodriguez, A. Luna, R. S. Munoz-Aguilar,
M
I. Etxeberria-Otad
dui, R.
Teodorescu an
nd F. Blaabjerg, A Stationary Reference
R
Framee Grid
Power
Synchronizatio
on System for Three-Phase Grid-Connected
G
Converters U
Under Adverse Grid
G
Conditions, IEEE Trans. Power
P
Electron., vol.. 27, no. 1, Jan. 20
012.
P. Rodriguez, J. Pou, J. Bergaas, J. I. Candela, R. P. Burgos and
a D.
Boroyevich, Decoupled

Doub
ble Synchronous Reference Frame PLL
for Power Con
nverters Control, IEEE Trans. Po
ower Electron., vol. 22,
no. 2, Mar. 2007.
P. Rodriguez, A. Luna, I. Can
ndela, R. Mujal, R. Teodorescu and
a F.
Blaabjerg, Multiresonant Freuqency-Lockeed Loop for Grid
Synchronizatio
on of Power Converters Un
nder Distorted Grid
Conditions, IEEE
I
Trans. Ind. Electron., vol. 58
8, no. 1, Jan. 2011.
X. Guo, W. Wu,
W and Z. Chen,, Multiple-Comp
plex Coefficient-FilterBased Phase-L
Locked Loop and
d Synchronization
n Technique for ThreeT
Phase Grid-In
nterfaced Converrters in Distributted Utility Netw
works,
IEEE Trans. In
Ind. Electron., voll. 58, no. 4, pp. 1194-1204, April 2011.
2
E. Robles, S. Ceballos, J. Pou
u, J. L. Martin, J.
J Zaragoza, P. Ib
banez,
Variable-Freq
quency Grid-Seq
quence Detector Based
B
on a Quasii-Ideal
Low-Pass Filteer Stage and a Ph
hase-Locked Loop
p, IEEE Trans. Power
P
Electron., vol.. 25, no. 10, pp. 2552-2563,Oct.
2
20
010.
I. Carugati, S. Maestri, P. G. Donato, D. Carrrica, and M. Ben
nedetti,
Variable Sam
mpling Period Filter
F
PLL for Distorted
D
Three--Phase
Systems, IEE
EE Trans. Powerr Electron., vol. 27,
2 no. 1, pp. 321-330,
Jan. 2012.
D. G. Luenbberger, An intro
oduction to obsservers, IEEE Trans.
Automatic Conntrol, vol. 16, no.. 6, pp. 596-602, Dec. 1971.
J. Keller and B. Kroposki, U
Understanding Fault
F
Characteristtics of
Inverter-Based
d Distributed En
nergy Resources, NREL, Golden
n, CO,
Tech. Rep. NR
REL/TP-550-46698, Jan. 2010.
A. R. Bergen
n and V. Vittal, Symmetrical Components
C
in Power
P
Systems Analyysis 2nd edition, Korea:
K
Pearson Educ.
E
Korea Ltd, 2009,
ch. 12-1, pp. 446-450.
4
A. Sannino, M.
M H. J. Bollen, and J. Svensson, Voltage Toleerance
Testing of Thhree-Phase Voltaage Source Con
nverters, IEEE Trans.
Power Deliverry, vol. 20, no. 2, pp. 1633-1639, April
A
2005.
A. G. Yepes, F. D. Freijedo, J. Doval-Gandoy
y, O. Lopez, J. Malvar
M
and P. Fernan
ndez-Comesana, Effects of Disccretization Metho
ods on
the Performan
nce of Resonan
nt Controllers, IEEE Trans. Power
P
Electron., vol.. 25, no. 7, July 2010.

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

[21] F. J. Rodrig
guez, E. Bueno, M.
M Aredes, L. G. B. Rolim, F. A. S. Neves
and M. C. Cavalcanti, Disccrete-time implem
mentation of seco
ond order
generalized
d integrators forr grid converterrs, in Conf. Reec. IEEE
IECON200
08, Nov. 10-13, 20
008, pp. 176-181.
[22] A. G. Yepees, F. D. FreiJedo
o, O. Lopez and J. Doval-Gandoyy, HighPerformancce Digital Reson
nant Controllers Implemented W
With Two
Integrators,, IEEE Trans. Po
ower Electron., vol.
v 26, no. 2, Febb. 2011.

Yongsoon Park
k (S12) received
d the B.S.
and M.S. degrees in electrical enggineering
from Seoul Naational Universityy, Seoul,
Korea in 2008 and 2010, resppectively,
t Ph.D.
where he is currrently pursuing the
degree. His currrent research inteerests are
sensorless drives of electrical m
machines,
and power conveersion circuits.

Hy
yun-Young Lee received the M.S
S.E.E.
deegree from Ulsan University, Ulsaan,
Ko
orea, in 2004. Hee is currently a Seenior
Ellectrical Engineerr at PALSPO Co..,Ltd,
Ko
orea. His research
h interests includ
de
po
ower electronics, motor controls, softs
sw
witching, photovo
oltaic system and
co
ommunication tecchniques.

Seung-Ki Sul (S78-M80-SM98-F00)


received the B.S
S, M.S, and Ph.D
D. degrees
in electrical engineering
e
from
m Seoul
National Univeersity, Seoul, K
Korea, in
1980, 1983, and 1986, respectiveely. From
1986 to 1988, he was an Associate
A
Researcher in thee Department of E
Electrical
and Computer Engineering,
E
Univ
versity of
Wisconsin, Mad
dison. From 1988 to 1990,
he was a Princcipal Research Engineer
with Gold-Staar Industrial Systems
Company. Sincee 1991, he hass been a
member of the faculty of the S
School of
Electrical Engin
neering, Seoul National
y a Professor. Froom 2005 to 20077, he was
University, wherre he is currently
the Vice Dean oof Electrical Engiineering, Seoul National
N
Universiity. From
2008 to 2011, he
h was the President of Korea Electrical Engineering &
Science Researcch Institute (KE
ESRI). His curreent research inteerests are
power electronicc control of electrrical machines, electric/hybrid
e
veehicle and
ship drives, and ppower-converter circuits.

Woo-Chull Kim
m was born in Korea in
1976. He receiived the B.S. ddegree in
Information
and
comm
munication
engineering froom Suwon Unniversity,
Suwon, Korea, in 2003. Since 20011, he is
a research engin
neer with LG Upplus, Inc.,
Seoul, Korea. His current research
interests includee power electroniic control
of renewable eneergy.

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

You might also like