Académique Documents
Professionnel Documents
Culture Documents
- Trends in Electrical Measurement and Instrumentation - September 13-14, 2001 - Lisbon, Portugal
Abstract - The paper deals with the measurement of to obtain exactly the Fourier spectrum of v(t) except for the
harmonic magnitude and phase angle of a periodic voltage roundoff and truncation errors in digital processing.
v(t) in power systems. The proposed procedure is based on The extraction of the cosine and sine coefficients of
Walsh spectrum measurement followed by digital v(t) from the truncated measure of cal and sal coefficients is
computation to provide the Fourier spectrum of v(t) in achieved by way of a Walsh to Fourier matrix conversion
magnitude and phase. Computer simulation results are process; a subsequent matrix multiplication provides for
presented to validate this method. compensation for the Walsh spectral truncation.
The compensation matrix takes the form of a diagonal
Keywords - Harmonic magnitude and phase angle, Power matrix with fixed elements; each converted Fourier spectral
systems, Walsh spectrum, Fourier spectrum. component requires to be multiplied by a correction factor
which is fixed for a given harmonic.
The Walsh method is attractive for use at power system
1. INTRODUCTION frequency, for then the synchronization of the Walsh waves
with v(t) is not difficult. It is evident that Walsh spectra are
The usual representation of a periodic time-varying superior to the Fourier spectra, in that the multiplication
signal is the Fourier series. The coefficients of the equation with sine and cosine functions, respectively, is obviated by
can be obtained according to two main approaches. In the simple reversal of signs (Walsh waves assume values of
first, the input signal is sampled, converted into digital form plus and minus one). Moreover both odd and even
and the Fourier integral calculation is used in order to harmonics can be measured.
obtain the coefficients. If the period is known, this method
requires only one cycle of the signal: however the main
drawback is the awkward and the time-consuming 2. PRINCIPLE OF OPERATION
multiplications by cosine and sine coefficients of Fourier
processing. Variations have been suggested processing the The Fourier series of a periodic waveform v(t) is written
digitized data using Fast Fourier Transform algorithms:
they allow accurate measurements but the disadvantage is
the large amount of computation involved [1].
In this paper a digital technique of measuring the phase v( t ) = a 0 + [a n cos nt + b n sin nt ] (1)
angle of the harmonics as well their magnitudes is n =1
presented. The method is firstly based on Walsh spectrum where
measurement with successive digital computation to 2 T
provide the Fourier spectrum of the periodic voltage in a n =
T 0 v(t) cos n t dt (2)
magnitude and phase [2]. This measurement method is fast
and accurate and doesnt require expensive instrumentation; 2 T
moreover it is particularly well-suited to power-frequency b n =
T 0 v(t) sin n t dt (3)
waveform measurements [3]-[6].
The Walsh method, in common with Fourier method,
requires that the periodic voltage v(t) be frequency limited, and T=2/ is the period of the fundamental frequency
which implies that its Walsh spectrum is made up of an at 50 Hz.
infinite series of cal and sal terms. However, if the highest Walsh functions, which are shown in Fig. 1 up to sal8(t),
harmonic order of v(t) is H, then measurement of only S can be used as an alternative orthogonal family for the
cal components and S sal components (with S H) permits series representation of a periodic waveform according to
2 T
0 0 0 0
a n, m =
T 0 cal m (t) cos n t dt (10)
-2 -2 -2 -2
0 0.01 0.02 0 0.01 0.02 0 0.01 0.02 0 0.01 0.02
cal2(t) sal3(t) cal3(t) sal4(t)
2 T
(11)
b n, m = 0 sal (t) sin n t dt
2 2 2 2
m
T
0 0 0 0
-2 -2 -2 -2
0 0.01 0.02 0 0.01 0.02 0 0.01 0.02 0 0.01 0.02
2
cal4(t)
2
sal5(t)
2
cal5(t)
2
sal6(t)
where n is the index of row (corresponding to the
harmonic order) and m is the index of column. Moreover
0 0 0 0
the elements of the conversion matrix Fa assume the same
-2 -2 -2 -2
absolute value of the conversion matrix Fb, but their signs
2
0 0.01
cal6(t)
0.02
2
0 0.01
sal7(t)
0.02
2
0 0.01
cal7(t)
0.02
2
0 0.01
sal8(t)
0.02
may differ. The conversion matrices are essentially semi-
infinite; when applied in (8) however they are truncated to
0 0 0 0 SxS element square matrices. Also an additional matrix
multiplication is required which compensated for the Walsh
-2
0 0.01 0.02
-2
0 0.01 0.02
-2
0 0.01 0.02
-2
0 0.01 0.02 spectral truncation
Ka a* = a Kb b* = b (12)
Fig. 1 Walsh functions up to the eighth order.
where a and b are the desired cosine and sine
coefficients, respectively. The simplest form of
The procedure for the Walsh series determination and the compensation occurs when S = 2k-1, i.e., the highest
corresponding Fourier derivation is given below. harmonic order in v(t) doesn't exceed the third, or the
Consider a periodic voltage v(t) whose period is T at the seventh, or the fifteenth,.
fundamental frequency of 50 Hz. Signal v(t) is sampled Then compensation takes the form of a diagonal matrix,
systematically with sampling frequency fs . with fixed elements
The Walsh coefficients An and Bn are obtained
2
according to (5), (6), (7). To yield the Fourier spectrum of n/2 k + 1
v(t), An and Bn are first multiplied by Walsh-Fourier (13)
conversion matrices Fa and Fb, respectively, to give sin( n/2 k + 1 )
approximations a*, b* to the cosine and sine coefficients of
v(t):
where n is the harmonic order.
a* = Fa A b* = Fb B (8)
3. COMPUTER SIMULATION RESULTS
where
The measurement algorithm has been verified by
computer simulation to investigate the validity of this
93
technique. The program generates an input voltage that is 1.273 0 0.527 0 0.105 0 0.253
sampled with sampling frequency fs = 100 kHz; its
0 1.273 0 0 0 0.527 0
waveform is shown in Fig. 2.
0.424 0 1.025 0 0.685 0 0.284
0 0 0 1.273 0 0 0.
0.255 0 0.615 0 0.920 0 0.381
0 0.424 0 0 0 1.025 0
0.182 0 0.075 0 0.379 0 0.914
80
60 30.5046 41.4398
40 6.3403 9.1580
) 20
2.5416 11.9204
V
(
e
g
a
tl
o
v
. 5.0116 = 6.3810 = a *
t 0
u
3.8010 4.0863
p
n
I
-20
2.0578 0.5825
-40
3.6125 0.9971
-60
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
time (s)
-5.00wal0(t)+30.5046cal1(t)6.3403cal2(t)+2.5416cal3(t)-
5.0116cal4(t)-3.8010cal5(t)-2.0578cal6(t)+3.6125cal7(t) 19.5930 26.7350
+19.5930sal1(t)+12.1964sal2(t)-2.0463sal3(t)+3.9259sal4(t)
12.1964 16.6391
-0.7127sal5(t)-2.1051sal6(t)-2.5048sal7(t)
2.0463 5.9965
. 3.9259 = 4.9986 = b *
0.7127 4.0311
2.1051 3.0194
In applying (8) the Fourier spectrum is obtained in
terms of Walsh spectrum 2.5048 1.1577
94
1 . 0130
80
1 . 0530
1 . 1242 60
a= 1 . 2337 . 40
1 . 3941
20
1 . 6260 )
V
(
)t
1 . 9638 (
v 0
-20
-40
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
9 . 1580 9 . 6436 time (s)
11 . 9204 13 . 4004
. 6 . 3810 = 7 . 8722 Fig. 3 Waveform of v(t).
4 . 0863 5 . 6969
0 . 5825 0 . 9472
0 . 9971 1 . 9581
The results obtained of magnitude and phase angle of
the harmonics are illustrated in Table I.
26 . 7350 27 . 0813
16 . 6391 17 . 5215 Fig. 4 shows the instantaneous error of v(t) with respect
5 . 9965 6 . 7410 to the input voltage. As we can see, the instantaneous error
4 . 9986 = 6 . 1668 is within 0.25 V.
4 . 0311 5 . 6199
3 . 0194 4 . 9096
1 . 1577 2 . 2735
0.2
Thus 0.15
0.1
0.05
rr
-0.15
95
Results obtained with the proposed method are then The method is based on Walsh spectrum measurement
compared with those obtained via the fast Fourier transform followed by digital computation for obtaining the Fourier
(FFT) algorithm. spectrum.
Fig. 5 shows the frequency content of the input voltage. This measurement method is fast and accurate and
doesnt require expensive instrumentation; moreover it is
particularly well-suited to power-frequency waveform
6000
Frequency content of the input voltage
measurements, for then the synchronization of the Walsh
waves with the input voltage is not difficult. It is evident
5000 that Walsh spectra are superior to the Fourier spectra, in
that the multiplication with sine and cosine functions,
yt
si
n
4000 respectively, is obviated by simple reversal of signs (Walsh
e
d
l waves assume values 1). Moreover both odd and even
ar 3000
c
e
t harmonics can be measured.
p
s
r 2000
e
w
o
P
1000 REFERENCES
96