IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 57, NO.

4, APRIL 2008 813

Shanks’ Method for Dynamic Phasor Estimation

Alejandro Torres Muñoz and José Antonio de la O Serna, Senior Member, IEEE

Abstract—A new algorithm for phasor estimation is proposed. into account the rate of change and acceleration of the dynamic
It is based on a signal model that allows amplitude and phase phasor [3] and [14]. These new terms provide better phasor
dynamic variations. An autoregressive moving average (ARMA) estimates and crucial information about the dynamic change
model is assumed for the oscillating signal. Its autoregressive
part is fixed, and it is defined only by the nominal fundamental of the power system during an oscillation. Thus, they can be
frequency. Its best moving average parameters are estimated with used to improve the monitoring and control of the power system
Shanks’ method. These parameters provide the key information under transient conditions. Previous results that are related to
from which the phasor state vector is estimated through the partial this intention of improving phasor estimation during dynamic
fraction expansion of the ARMA rational polynomial. These esti- system conditions can be found in [4]–[6]. The well-known
mates could be useful, not only for the monitoring and controlling
of the power system, but also for discriminating between a fault Stankovic’s approach can be found in [7] and [8].
and an oscillation state. The representation of the signal by an autoregressive moving
average (ARMA) signal model and the identification of its
Index Terms—Autoregressive moving average (ARMA) signal
model, complex envelope, demodulation, digital filters, dynamic parameters over short time intervals lead to a new method for
phasor, frequency estimation, interpolation, phasor estimation, phasor estimation. The autoregressive (AR) coefficients of the
Shanks’ method. model only depend on the nominal fundamental frequency. On
the other hand, its best moving average (MA) parameters are
I. I NTRODUCTION estimated using Shanks’ method. These parameters carry the
information of the phasor that corresponds to the given power

D UE to the increasing power demand and interarea power

interchanges, power systems often operate closer to the
stability limits. Under these conditions, load switching or fault
signal. The partial fraction expansion of the ARMA model
demodulates the power signal, and the coefficients of each
fraction provide its phasor state vector. The solution of this
tripping produce power oscillations, characterized by slow am- least-squares method is very simple, given that its AR coeffi-
plitude variations at frequencies lower than the fundamental cients are constant; therefore, the inversion of its Gram matrix
one. These variations are good indicators of stability loss or needs to be calculated only once. This can also be regarded
even of system collapse risk. as a bandpass filtering technique around the fundamental fre-
The magnitude of the oscillation depends on the initial con- quency that captures the spectrum of the oscillation and its
ditions of the system when a transient is released. The system derivatives.
reaches nominal values until a new steady state is attained. After the presentation of the basic ideas that concern the
During the oscillation, relays are unable to discriminate a faulty dynamic phasor in [3] and [14], a wide interest was found in
state from an oscillation state. This is mainly because, today, relating the proposed solution to the well-known least-squares
most of the phasor estimates are obtained by assuming constant methods such as Shanks’ method. This paper is the response
amplitude and phase during the observation interval [1], [2], to this significant concern. The main conclusion is that the
which is a severe assumption under transient conditions. ARMA signal model is quite limited, mainly because it only
Thus, monitoring and control of power systems under tran- generates time-asymmetrical signals and, therefore, incomplete
sient conditions need new algorithms for phasor estimation. subspace bases, which are unable to cover symmetric Taylor
The reliable and effective discrimination between a fault and polynomials. However, the ARMA model is able to fit Taylor
an oscillation is a crucial decision to avoid the blackouts of the polynomials at the right-hand side of the center of the interpo-
recent years. lation interval, as we will see in the next sections.
In this paper, we assume amplitude and phase to be quadratic In the following sections, the signal models are developed
time functions over short observation intervals. They are rep- along with Shanks’ method; then, the frequency responses of
resented by second-order Taylor polynomials that, in addition the filters are shown for different time intervals. Then, we
to its constant term, corresponding to the traditional amplitude analyze the noise sensitivity of the filters. Finally, several sim-
and phase assumed nowadays, includes two new terms that take ulation and experimental results are presented. Our conclusion
is that this methodology could be able to discriminate between
oscillation and fault states of the system by the new information
Manuscript received February 8, 2007; revised May 30, 2007. This work about the phasor speed and acceleration.
was supported by the State University of Nuevo León under Project PAICYT
CA-1241-06.
The authors are with the Autonomous University of Nuevo León (UANL),
66451 Monterrey, México. II. O SCILLATION S IGNAL M ODEL
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. Nowadays, phasor estimation techniques are found over a
Digital Object Identifier 10.1109/TIM.2007.913824 steady-state signal model during the whole observation time

0018-9456/$25.00 © 2008 IEEE

814 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 57, NO. 4, APRIL 2008

interval. The traditional model has constant amplitude and following z-transform:
phase, as in the following equation: 
ρ1 − ρ2 2 ejθ0 z −1
(2) (2)
(2)
(2) 1 ρ0
S (z) = +
s(t) = a0 cos(ω0 t + ϕ0 ), 0 ≤ t ≤ T. (1) 2 1 − ejθ0 z −1 (1 − ejθ0 z −1 )2

ρ2 ejθ0 z −1
(2) (2)∗
ρ0
+ −1
+
This model has no room for the dynamic variations proper (1 − e z )
jθ 0 3 1 − e−jθ0 z −1
∗
to an oscillation and constitutes a severe restriction for the (2)
ρ1 − ρ2
(2)
2 e−jθ0 z −1
monitoring and control of the power systems. In this paper, we +
relax amplitude and phase to a band-limited time function over (1 − e−jθ0 z
−1 )2

ρ2 e−jθ0 z −1
(2)∗
the observation window, as can be seen in
+ . (7)
(1 − e−jθ0 z −1 )3
s(t) = a(t) cos (ω0 t + ϕ(t)) , 0 ≤ t ≤ T. (2) This can be reduced to a rational function of the form
q

By rewriting (2) in terms of complex exponentials, we have bk z −k

S (2) (z) = k=0p (8)

1+ ak z −k
1

s(t) = P (t)ejω0 t + P ∗ (t)e−jω0 t , 0≤t≤T (3) k=1
2
where q = 5 and p = 6 for the quadratic model.
The transfer function in (8) has two triple poles—one triple
where P (t) = a(t)ejϕ(t) is the dynamic phasor of the signal. at ejθ0 and the other at its complex conjugate. Note that they
In telecommunications, this is known as the complex envelope are defined only by the nominal fundamental frequency θ0 . On
of the bandpass [9] signal s(t). This complex time function is the other hand, the coefficients bk , k = 0, . . . , q, of the MA part
crucial for the dynamics of the power system. provide the phasor information that corresponds to the bandpass
It is possible to approximate the dynamic phasor P (t) signal, as we can see in the partial fractions expansion in (7).
over a given interval with ascending-order Taylor polynomi-
als. Without loss of generality, the quadratic approximation III. S HANKS ’ M ETHOD
would be
The best coefficients {bk }qk=0 in (8) are obtained using
Shanks’method [10], [11], which exploits the least-squares
2
(2) (2) t (2) t principle. The estimation error is defined as the difference
P (2) (t) = P0 + P1 + P2 , 0≤t≤T (4)
1! 2! between the given signal segment sd (n) and its model ŝ(n)
in (6) by
where P (2) (t) is the second-order Taylor polynomial in which e(n) = sd (n) − ŝ(n)
the terms P0 = P (0), P1 = P  (0), and P2 = P  (0) ∈ C are  q
the derivatives of the dynamic phasor P (t) at the origin. As the = sd (n) − bk v(n − k), n = 0, . . . , N − 1 (9)
order of the polynomial increases, the estimates will improve, k=0
but the estimates of the quadratic approximation are good −1
where {v(n)}N n=0 is the unit response of the AR filter. The
enough for our purposes. The approximated bandpass signal
best estimates of bk are obtained by applying the orthogonality
will then be
principle (V ⊥ e ∴ VH e = 0) and by solving the normal
equation
1  (2) 
s(2) (t) = P (t)ejω0 t + P (2)∗ (t)e−jω0 t , 0 ≤ t ≤ T. VH Vb̂ = VH sd (10)
2
(5)
where V is a matrix N × (q + 1) whose vectors are delayed
By sampling (5), we will have versions of the unit response v(n), as shown in the following
equation:
 2
   
(2) (2) (2) (2) n jθ0 n v(0) 0 ··· 0
s (n) = Re ρ0 + ρ1 n + ρ2 e ,
2!  v(1) v(0) ··· 0 
 
0 ≤ n ≤ N − 1 (6)  .. .. . . 
 . . . . .
. 
V=  v(l + 1)
 (11)

 v(l) ··· v(0) 
 .. .. .. 
where, for N0 samples per fundamental period, θ0 = (2π/N0 ),  . . ··· . 
(2) (2) (2) (2)
ρ1 = P1 τ , and ρ2 = P2 τ 2 , in which τ = T0 /N0 . v(N − 1) v(N − 2) · · · v(N − q − 1)
This quadratic model can be considered as an adaptive filter
for each time interval. Its transfer function is given by the and the vector b̂ contains the best estimates of bk .
MUÑOZ AND DE LA O SERNA: SHANKS’ METHOD FOR DYNAMIC PHASOR ESTIMATION 815

Fig. 1. Projection of sd (n) onto the signal model subspace Θ0 .

Fig. 2. Unit responses of the AR signals. The envelopes are given by the
Note that VH V in (10) corresponds to the autocorrelation right-hand side of the constant, linear, and quadratic Taylor polynomials,
matrix Rvv of v(n) and that VH sd corresponds to the cross- respectively.
correlation vector rhd v between the given signal sd (n) and the
unit response of the AR filter v(n). Thus, the solution can be demodulation of the given signal, and to its exact demodulation
written as if the given signal is in the model subspace.
(2)
Remember that P̂0 corresponds to a sample of the dynamic
b̂ = R−1
vv rhd v . (12) (2) (2)
phasor P̂1 to its rate of change and P̂2 to its acceleration.
The relationship between the Taylor coefficients and amplitude
Once the best estimates of {bk }qk=0 are obtained, the transfer and phase are the following:
function of the estimated signal can be written as  
 (2) 
â(0) = 2 P̂0 

q
−k
b̂k z (2)
(2) k=0 ϕ̂(0) = ∠P̂0
Ŝ (z) =
p . (13)  
â (0) = 2 Re P̂1 e−j ϕ̂(0)
(2)
1+ ak z −k
k=1
 
2 Im P̂1 e−j ϕ̂(0)
(2)
By analyzing the former approximation from the linear alge- ϕ̂ (0) =
bra point of view, the unit response of the AR filter v(n) and its â(0)
delays cover the subspaces defined by the component sequences  
â (0) = â(0)ϕ̂2 (0) + 2 Re P̂2 e−j ϕ̂(0)
(2)
of the signal model, namely, the constant component ejθ0 n , the
linear nejθ0 n , and the quadratic n2 ejθ0 n , n ≥ 0. It is precisely  
2 Im P̂2 e−j ϕ̂(0) − 2â (0)ϕ̂ (0)
(2)
onto this subspace that the given signal sd (n) is projected and in
which the ARMA model of the estimated signal ŝ(n) is found, ϕ̂ (0) = . (14)
â(0)
as shown in Fig. 1.
It is important to emphasize that AR signals are causal. Fig. 2 With these relationships, it is also possible to build Taylor
shows the unit responses of the constant, linear, and quadratic polynomials for amplitude and phase as follows:
AR signals. They appear normalized by their maximum value.
â (0)(τ n)2
Note that their envelopes correspond only to the right-hand side â(2) (n) = â(0) + â (0)τ n +
of a zeroth-, first-, and second-order Taylor polynomial, respec- 2!
tively. Thus, the AR signals involved in Shanks’ method are ϕ̂ (0)(τ n)2
unable to generate the left-hand side of the Taylor polynomials, ϕ̂(2) (n) = ϕ̂(0) + ϕ̂ (0)τ n + , 0 ≤ n ≤ N.
2!
and therefore, with right-handed AR signals, the error reduction (15)
of the Taylor approximation will be applied only over the right-
hand side of the Taylor interval, missing the reduction on its left It is important to emphasize that the derivative of the phase
half. This is the main limitation of the vector basis of the signal ϕ̂ (0) corresponds to the frequency deviations around the nom-
subspace used by Shanks’ model. inal fundamental frequency, as can be seen in (2). In addition,
its second derivative ϕ̂ (0) corresponds to the derivative of the
(2)
The best estimates of the Taylor coefficients P̂i , i = 0, 1, 2,
are obtained by expanding (13) in partial fractions, as in (7), frequency that is typical of a disturbance in the power system.
and then solving into (4). This solution corresponds to the best This measurement is very important under power exchanges or
816 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 57, NO. 4, APRIL 2008

(0) (1)
Fig. 3. Frequency response of the phasor estimate P̂0 . A concave gain is Fig. 4. Magnitude response of the phasor estimate P̂0 . For windows lower
reached at u = 1 and a linear gain at u = −1, indicating the infiltration of the than one cycle, convex gains are achieved at u = ±1, which indicate infiltration
acceleration at u = 1 and the derivative at u = −1. of acceleration at both frequencies. For larger windows, concave and linear
gains are reached, as in the previous model.

for power system state estimation. Even if those applications

are not considered in this paper, this algorithm will certainly
improve them, not only through the reduction of the phasor
error during off-nominal frequency operation, but also by pro-
viding the frequency-offset measurement itself. In Section IV,
the frequency response of the phasor estimates regarded as the
output of complex filters will be analyzed. We will consider the
following three ascending-order Taylor models: 1) the constant
model; 2) the linear model; and 3) the quadratic model.

IV. F REQUENCY R ESPONSE

The frequency response of the phasor estimates are found by
given input signals of the form sd (n) = ejθn , n = 0, . . . , N −1,
over π ≤ θ ≤ π. This analysis is very useful to explain the
behavior of the estimates when the spectral composition of the
input signal sd (n) is known or before fundamental frequency
changes.
(0)
Fig. 3 shows the magnitude response of the phasor P̂0 (2)
Fig. 5. Magnitude response of the phasor estimate P̂0 . Flat gains are only
estimated with the constant model over windows of increasing achieved at u = ±1, with windows lower than one cycle.
lengths: (N/N0 ) = (1/8), (1/4), (1/2), 1 cycle. This model
corresponds to the traditional Fourier phasor estimator, which The linear model preserves the unit gain at the fundamental
assumes a sinusoidal signal with constant amplitude and phase. frequency u = 1 for all lengths, as shown in Fig. 4, but now,
For the first two windows, there is a huge sensitivity to noise. a null gain is achieved around u = −1, indicating the rejection
This sensitivity decreases as the length of the windows in- of the derivative infiltration of the constant model. For windows
creases. However, a concavity around the fundamental fre- lower than one cycle, convex gains are obtained around u = ±1,
quency persists. This concavity indicates infiltration of the indicating infiltration of the acceleration component into the
phasor acceleration into the phasor estimate. This persistence phasor estimates at both frequencies. Note that these filters are
is also found for the linearity around the negative fundamental very sensitive to noise. For windows larger than one cycle,
frequency component, which indicates infiltration of its phasor concave gains at u = +1 and linear responses at u = −1 are
derivative into the phasor estimates. This is a severe limitation achieved as in the previous model.
for the constant model when we want estimates under power Finally, the quadratic model is able to filter the first and
oscillations. These estimates will always distort amplitude and second derivatives out of the phasor estimates, forming band-
phase. This is explained by the fact that the zero subspace pass filters around the fundamental frequency u = 1, as we can
is unable to channel out derivatives and accelerations so they see in Fig. 5. Note that, for windows lower than one cycle,
unavoidably infiltrate into the estimated phasors. unitary and zero flat gains are formed at u = ±1, which allow
MUÑOZ AND DE LA O SERNA: SHANKS’ METHOD FOR DYNAMIC PHASOR ESTIMATION 817

Fig. 6. Normalized mean-square error in estimates due to noise infiltration,

(0)
using the constant (E0 ) and quadratic (E(2) ) models over two cycles. Fig. 7. Real signal and its spectrum. In the signal, an apparent oscillation
is produced after a released fault. Its spectrum shows the presence of a fifth
harmonic.
the oscillation to pass without amplitude distortion at u = 1
and suppress the component at u = −1. However, the perfect
flat gains are not achieved with this technique for windows
longer than two cycles because Shanks’ signal model is causal
and antisymmetric. Achievement of better flat gains can be
found in [3] and [14].

V. N OISE R EJECTION
The proposed phasor state vector estimate is a least-squares
estimate in the overdetermined case [12] for the following
signal-plus-noise model:

s = Vb + n (16)

assuming V is known, b is deterministic, and n is noise

distributed as N [0, σ 2 I]. Thus, the received signal comes from
the model subspace but with additive noise. In this case,  b
in (12) is an unbiased estimate of b, and s is distributed as
N [Vb, σ 2 I]. Fig. 8. Amplitude and phase reconstructed signal with estimation and inter-
In this section, the noise infiltration into the phasor estimates polation every two cycles. Discontinuities at the extremes of the intervals are
perceptible.
is considered. For a given b, we obtain b̂ from (12) for many
runs, and by partial fraction expansion, the phasor state vectors VI. E XPERIMENTAL R ESULTS
P̂ are obtained from (7). Then, the phasor mean-squared error
is calculated as In this section, several results of phasor estimates are shown.
We will work with a signal obtained with a phasor measurement
P − P̂ = E. (17) unit installed in a substation of the Mexican network. The
signal was sampled at 48 samples per cycle. This signal is
Fig. 6 shows the normalized mean-squared phasor state er- shown in Fig. 7 along with its spectrum. The spectrum of the
rors due to the presence of white noise in the input signal, using oscillation is perceptible around the fundamental frequency, as
the quadratic E(2) and constant E(0) models, over intervals of well as a fifth harmonic (1.51%). The phasor estimates will be
two cycles when the signal-to-noise ratio (SNR) goes from 0 estimated using the quadratic model in (6). The reconstructed
to 80 dB. The phasor error obtained with the quadratic model is amplitude and phase signals are shown in Fig. 8. Such signals
ten times greater than the one obtained with the constant model. are interpolated over intervals of two cycles, using the phasor
This factor is due to the higher sidelobes of the frequency estimated with the quadratic model over those intervals using
response of the quadratic model (see Figs. 3 and 5). Thus, (6). Little discontinuities appear at the interval extremes as it
Shanks’ method must be used with high SNRs (i.e., greater was expected. They can be reduced by reducing the size of
than 40 dB). the interpolation interval or by reconstructing the signal with
818 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 57, NO. 4, APRIL 2008

Fig. 9. Instantaneous estimates of amplitude and phase obtained sample by

sample with overlapping windows of two cycles.
Fig. 10. Instantaneous estimates of frequency and its derivative. Note that
sudden changes in amplitude in Fig. 9 can be localized by detection of sudden
phasor estimates at every sample, as shown in Fig. 9. It can be changes in frequency.
observed that discontinuities disappear in this figure; however,
the instantaneous estimates still generate waver signals. error is all the more significant as derivatives of the complex
By analyzing the behavior of the derivatives of the phase envelope higher than the adopted order are found at the center of
shown in Fig. 10, abrupt changes of ±4 Hz are perceived during the time interval. In digital transmission systems, the modulator
the fault. After those sudden changes, a frequency offset is generates the transmitted signals from the signal model, so
produced. This offset explains the descending phase during the Taylor errors are null in that application because those signals
oscillation observed in Figs. 8 and 9. In the postfault region, a coincide with the nth-order model.
little variation of ±0.2 Hz around the fundamental frequency is The Shanks’ method is nothing but the least-squares method
observed. with a particular vector basis. We know that the least-squares
The first two derivatives of the phase provide impor- error is orthogonal to the signal subspace, and it is zero when
tant information about the behavior of the frequency during the input signal matches the signal model. The vectors that
the following three characteristic states of a power signal: generate the signal subspace of Shanks’ method are delayed
1) prefault; 2) fault; and 3) oscillation. The second derivative versions of the unit responses of the AR filters, as can be seen
of the phase corresponds to the rate of change of the fre- in (11). We have also shown that the basis vectors of Shanks’
quency, and therefore, it can be used to detect and estimate method are causal AR signals. Thus, time-symmetric signals
the imbalance between generation and load produced by the such as Taylor polynomials are not contained in the subspace
tripping of the generators, load shedding, fault tripping, etc. By of Shanks’ method, and their projection entails an error. In
taking into account the topology and the initial conditions of the addition, the polynomial fitting provided by Shanks’ method
system, it could be possible to establish a threshold for ϕ̂ (n) to will only operate at the right-hand side of the time interval
obtain an indicator of the presence of disturbance in the power required by the Taylor approximation.
system [13]. Finally, the algorithmic errors of the partial fraction expan-
sion are significant because the transfer function of the signal
model has two triple poles. In the case of multiple poles, the
VII. D ISCUSSION
residuez algorithm of MATLAB accepts an error tolerance of
Phasor estimates are obtained from the coefficients of the 10−3 for its residues. To reduce this tolerance, we are working
partial fractions of the ARMA transfer function. This new on finding the partial fraction expansion through a procedure
technique solves the phasor estimation problem in terms of very fitted to our particular ARMA models, instead of using the
well known mathematical procedures in engineering but has general one of MATLAB. The other research area is to rebuild
several sources of error—the Taylor error, the least-squares er- the signal model to obtain symmetric signals. However, even
ror of Shanks’ method, and the algorithmic errors generated by if we have three sources of error in this technique, the phasor
the partial fraction expansion residuez algorithm of MATLAB. estimates could be useful for certain applications.
The Taylor error is due to the truncation of the polynomial This paper also shows how limited are the well-known least-
order, and its assessment requires a priori knowledge of the squares methods based on the ARMA signal model that is
complex envelope. In our application, the input signal already widely used in signal processing. Limitations that are currently
comes modulated, and there is no way of knowing the complex attributed in the literature to Shanks’ method must, in fact, be
envelope in advance. It can be said, however, that the Taylor assigned to the time asymmetry of the AR model. Its drawback
MUÑOZ AND DE LA O SERNA: SHANKS’ METHOD FOR DYNAMIC PHASOR ESTIMATION 819

comes, in fact, from the flaw of its time-unbalanced vector [7] A. M. Stankovic and T. Aydin, “Analysis of asymmetrical faults in power
basis. It is also known that this method becomes more accurate systems using dynamic phasors,” IEEE Trans. Power Syst., vol. 15, no. 3,
pp. 1062–1068, Aug. 2000.
when the order of the MA part is increased. This is because [8] A. M. Stankovic, H. Lev-Ari, and M. Perisic, “Analysis and implementa-
high-order MA models are able to fit time-symmetric signals tion of model-based linear estimation of dynamic phasors,” IEEE Trans.
such as Taylor polynomials. Finally, we need to mention that Power Syst., vol. 19, no. 4, pp. 1903–1910, Nov. 2004.
[9] L. W. Couch, II, Digital Analog Communication Systems, 5th ed.
our main goal was to theoretically relate the solution proposed Englewood Cliffs, NJ: Prentice-Hall, 1997, ch. 4, p. 228.
in [3] and [14] to Shanks’ method, not to research the best [10] J. G. Proakis and G. Monolakis, Digital Signal Processing, 3rd ed.
hardware platforms or algorithmic execution times for critical New York: Prentice-Hall, 1996, ch. 8, pp. 709–710.
[11] J. G. Proakis and C. M. Rader, Algorithms for Statistical Signal
applications. Processing, 1st ed. Englewood Cliffs, NJ: Prentice-Hall, 2001, ch. 4,
pp. 177–217.
[12] A. J. Thorpe and L. L. Scharf, “Data adaptive rank-shaping methods for
VIII. C ONCLUSION solving least squares problems,” IEEE Trans. Signal Process., vol. 43,
no. 7, pp. 1591–1601, Jul. 1995.
An ARMA signal model has been proposed for estimat- [13] J. Zuo, Z. Zhong, R. M. Gardner, H. Zhang, and Y. Liu, “Off-line event
ing the dynamic phasor of a power signal through its partial filter for the wide area frequency measurements,” in Proc. IEEE PES
General Meeting, Montreal, QC, Canada, Jun. 2006, pp. 1–6.
fraction expansion. ARMA parameters were estimated with [14] J. A. de la O Serna, “Dynamic phasor estimates for power systems os-
Shanks’ method, and phasor estimates were obtained from cillations,” IEEE Trans. Instrum. Meas., vol. 56, no. 5, pp. 1648–1657,
the partial fraction expansion of its transfer function. Shanks’ Oct. 2007.
phasor estimates could be useful for certain applications such
as discriminating between fault and oscillation states in power
signals and controlling applications that require a phasor state
vector. Alejandro Torres Muñoz was born in San Nicolás
The main limitation of the ARMA model is its time asym- de los Garza, Nuevo León, México, in 1982. He
received the B.S. degree in mechanical and elec-
metry, which allows approximation of the signal only over the trical engineering from the Autonomous University
right part of the Taylor interval. Errors are also generated with of Nuevo León (UANL), Monterrey, México, where
the partial fraction expansion algorithm. Thus, further research he is currently working toward the M.S. degree in
electrical engineering.
is needed to resolve these shortcomings, although a well-known
algorithm has already been applied to estimate phasors under
power system oscillations.

R EFERENCES
[1] L. Wang, “Frequency responses of phasor-based microprocessor relay-
ing algorithms,” IEEE Trans. Power Del., vol. 14, no. 1, pp. 98–109,
Jan. 1999. José Antonio de la O Serna (SM’03) was born in
[2] A. G. Phadke and J. S. Thorp, Computer Relaying for Power Systems. San Pedro, Coahuila, Mexico, in 1953. He received
Hoboken, NJ: Wiley, 1988. the Ph.D. degree from Telecom Paris, Paris, France,
[3] J. A. de la O Serna, “Dynamic phasor estimates for power systems oscilla- in 1982.
tions and transient detection,” IEEE Trans. Instrum. Meas., vol. 56, no. 5, From 1982 to 1986, he was a Professor with
pp. 1648–1657, Oct. 2007. Monterrey Institute of Technology, Monterrey,
[4] J. A. de la O Serna, “On the use of amplitude shaping pulses as win- México. In 1987, he joined the Ph.D. program in
dows for harmonic analysis,” IEEE Trans. Instrum. Meas., vol. 50, no. 6, electrical engineering at the Autonomous University
pp. 1556–1562, Dec. 2001. of Nuevo León (UANL), Monterrey, where he was
[5] J. A. de la O Serna, “Phasor estimation from phasorlets,” IEEE Trans. a member of the Doctoral Committee. From 1988
Instrum. Meas., vol. 54, no. 1, pp. 134–143, Feb. 2005. to 1993, he was with the Department of Electrical
[6] J. A. de la O Serna and K. Martin, “Improving phasor measurement Engineering, National Polytechnic School, Yaoundé, Cameroon. He is currently
under power system oscillations,” IEEE Trans. Power Syst., vol. 18, no. 1, a Research Professor with UANL.
pp. 160–166, Feb. 2003. Dr. de la O Serna is a member of the Mexican Research System.