14th PSCC, Sevilla, 24-28 June 2002 Session 01, Paper 5, Page 1

FREQUENCY DETECTION METHOD

BASED ON RECURSIVE DFT ALGORITHM
Katsuyasu Nakano*, Yutaka Ota*, Hiroyuki Ukai*,
Koichi Nakamura*, and Hideki Fujita**

*Dept. of Systems Management and Engineering, **Electric Power Research & Development Center
Nagoya Institute of Technology, Chubu Electric Power Co.,Inc
Gokiso-cho, Showa-ku, Nagoya, 466-8555, JAPAN 20-1, Kitasekiyama, Ohdaka-cho,
Phone & Fax: +81-52-734-5428 Midori-ku, Nagoya, 459-8522, JAPAN
E-mail: katsu@franklin.elcom.nitech.ac.jp

Abstract :We have developed the high- tems. The calculation algorithm of the volt-
speed and high precision PMU(Phasor Meas- age/current phasor in the PMU is based on the
urement Unit) by using DSP(Digital Signal recursive DFT(Discrete Fourier Transform). How-
Processor) and the GPS(Global Positioning ever, in the case when frequency of power sys-
System). The developed PMU enables to cal- tems is fluctuating and different from the funda-
culate the voltage and current phasors and the mental frequency of DFT algorithm, the calculated
other state variables in electric power sys- r.m.s. value and phase angle have the detection
tems from the measured instantaneous volt- error due to the deviation of frequency[6][7].
age and current in real-time. In this paper, our The purpose of this paper is to propose a new
purpose is to propose a new detection method detection method of the frequency and the phasor
of frequency in the power systems based on in the power systems by analyzing the error con-
error analysis of the recursive DFT algorithm tained in the voltage phasor due to the deviation of
used in the PMU. This method enable s to frequency.
detect frequency with fine accurate and high- Firstly, assumed the deviation of frequency, the
speed. error equation of voltage phasor on the basis of the
recursive DFT algorithm is derived. Since both the
Keywords : frequency detection, recursive DFT, phase and the magnitude components of this equa-
phasor measurement unit, error analysis tion contain oscillatory error, the running average
filter is applied in order to remove the error. Then
the deviation of frequency is recursively deduced
1 INTRODUCTION from the filtered voltage phasor. Secondly, since
the voltage phasor, which the filter is applied to, has
Frequency is one of the most important state the off-set error caused by deviation of frequency
variables in power systems. It is typically measured and the phase delay due to the filter, this off-set
by detecting the zero-crossing period of the voltage error and phase delay is compensated by applying
wave. However, since power electronics technol- the detected deviation of frequency. On the other
ogy is increasingly applied in the power systems, it hand, it is made clear that the error equation which
is required to detect more precisely and faster in dose not contain oscillatory error is deduced by
the presence of harmonics and noise. using the positive-phase-sequence component
On the other hand, applications of digital signal transformed by the method of symmetrical coordi-
processing technology to the power systems have nates in the case of three phase balanced voltage.
received a great deal of attention in recent years. Finally, the numerical simulation results are shown
There have been many digital algorithm applied to to confirm the effectiveness of the proposed
calculate frequency and phasor[1][2][3][4][5]. In method compared with the zero-cross detection
those, in particular, the PMU is widely applied in method[8]. In particular, the proposed method has
order to detect both r.m.s. value and phase angle of robustness in the case when the voltage wave con-
voltage/current at multi points in the power sys- tains harmonics.
14th PSCC, Sevilla, 24-28 June 2002 Session 01, Paper 5, Page 2

2 FREQUENCY DETECTION METHOD where,

BASED ON RECURSIVE DFT V; amplitude of the voltage
ALGORITHM П; initial phase angle of the voltage
f; frequency of power source.
2.1 Recursive DFT Algorithm
r
Voltage phasor Vr is detected by the funda- Now consider that the frequency of power
mental frequency component of DFT(Discrete source fluctuates and is different from the funda-
Fourier Transform) algorithm for the sequential mental frequency of DFT algorithm f0 , that is f ᷎
data of voltage instantaneous value v as follows: (=f-f0). By substituting Eq. (4) into Eq. (1) and (2),
the perturbed voltage phasor is obtained, which
r 2 N consists of two vector components.
− j2π f 0 kTS
Vr = j ∑ v k +r − N e (1)
N k =1
r 2 N − j 2π f0 kTS − j 2π f0 rTS
Xr = j ∑ vk +r − N e e
N k =1
where, V j ( φ + 2π ∆f ( r − N )TS ) N j 2π ∆fkTS
j;unit of complex number, = {e 0 ∑e
2N k= 1
f0 ;fundamental frequency of DFT, j( − φ0 − 2 π ( 2 f 0 + ∆f )( r − N )T S ) N − j 2 π ( 2 f 0 + ∆f )kTS
−e ∑e }
TS;sampling period, k= 1
N(=1/( f0Ts));the number of sampling per one =
V
( Ae
jαr
+ Βe
jβ r
)
cycle of fundamental frequency. 2
(5)
In Eq. (1) the sine component is taken as the refer-
ence, because the phase angle from the zero-cross where,
point of the sine wave should be detected. By in- sin( π∆fNTS )
A=
troducing rotational coordinate system, N sin( π∆fTS )
sin( π ( 2 f 0 + ∆f )NTS ) (6)
r r B=
X r = Vr e − j 2πf0rTS (2)
N sin( π ( 2 f 0 + ∆f )TS )

the following recursive relation is derived. αr =φ − π∆f ( N − 1 )TS + 2π∆frTS

βr = π − φ +π ( 2 f0 + ∆f )( N − 1)TS − 2π( 2 f0 + ∆f )rTS
(7)
r r 2
( vr + 1 − vr− N + 1 )e− j 2πf 0 ( r+ 1 )TS (3)
Xr + 1 = Xr + j
N ᷎f (=f-f0);deviation of frequency of power
source.
Compared with the usual DFT algorithm as Eq.
(1), the calculation number of recursive DFT as Fig.1 shows the relation between A, B and fre-
Eq. (3) is independent from the sampling number. ᷎
quency f (=f0 + f ) by using Eq. (6). In the Fig. 1,
In fact, the calculation number of real part and if the deviation of frequency is less than ±2.0 Hz,
imaginary part of the latter is only six for any N, which is the range of frequency generally consid-
whereas the former is 4N+1. This implies that the ered in the power systems, it is estimated that A≒
recursive DFT is beneficial for calc ulation time. 1.0, B≒ 0.0. Moreover the phasor have high quality
for signal-noise-ratio in high frequency band, in
2.2 Error Analysis of Recursive DFT particular for harmonics. On the other hand, both
Instantaneous value of single phase voltage vr Ћ r and Ќ
r in the Eq. (7) are linear function of ᷎
is expressed as follows: f. The phasor relationship of Eq. (5) is shown in
Fig. 2. It is found that the second term of Eq. (5)
vr = V sin(2πfrTS + φ ) brings about the oscillatory error with the fre-
V jφ j 2πfrTS − jφ − j 2πfrTS quency of about two times of the frequency of
= (e e −e e ) (4)
power source. These error are due to the fre-
᷎
j2
quency deviation f.
14th PSCC, Sevilla, 24-28 June 2002 Session 01, Paper 5, Page 3

1.2 1.2
0.9 1.0
Magnitude

0.6 0.8

Gain
0.3 0.6

0.0 0.4
0.2
-0.3
0 60 120 180 240 300 360 0.0
0 60 120 180 240 300 360
Frequency [Hz]
Frequency [Hz]
A B

Fig.1 Magnitudes of vector A and B Fig. 3 Frequency response of

r running average filter
Xr

2π∆f B

In the Eq. (10) and Eq. (11), both the r.m.s.

A value and the phase angle include off-set error
2π ( 2 f 0 + ∆f ) caused by the deviation of frequency. By substitut-
Fig. 2 Phasor relationship ing the deviation of frequency detected as previous
section into A in Eq. (10), the r.m.s. value is com-
2.3 Frequency Detection pensated. On the other hand, since the exact value
Since the phasor vector is oscillating due to de- of the phase angle consists of the first and the third
viation of frequency, the running average filter is term of Eq. (11), the phase angle is compensated
applied in order to remove the effect of the second by subtracting the detected deviation of frequency
vector in Eq. (5). Fig. 3 shows the frequency char- and Kf from them. As the result, the following
acteristic of the applied filter. The filter is designed precise phasor angle and r.m.s. value are obtained.
in order to remove the band around 60×2k Hz,
r
k=1,2,3,…, which include the oscillatory frequency
˺
r Xr * V
of the error vector. The filtered phasor is ex- Xr ** = (12)
A 2
pressed as follows:
r r
r
Xr * ˺ V
2
Ae
j( α r + K f )
(8)
arg( X r ) * * = arg( X r *) + π∆f ( N − 1 )TS − K f
˺ φ + 2π∆frT S
(13)

where,
Kf ;phase delay due to the filter. 2.5 Balanced Three-Phase Voltage Case
For the case of three-phase voltage, the phasor
The deviation of frequency is deduced from the can be detected by the positive-sequence compo-
difference of the sequential phase angle of Eq. (8) nent voltage. In the case of balanced three-phase
as follows; voltage, the phasor equation is obtained as follows:

2 2
( αr + K f ) − ( α r − 1 + K f ) r 1 r j π r −j π r
∆f r = = ∆f (9) X pr = ( Xa r + e 3 Xbr + e 3 X cr )
2πTS 3
 ( Ae jα r + Βe jβ r ) 
 
 2
π −
2
π
2
π

2.4 High Precision Phasor Detection =
1 V 
+ e
j
3
( Ae
jα r
e
j
3
+ Βe
jβ r
e
j
3
) 
From Eq. (8), the r.m.s. value and the phase an- 3 2 
 −j π 2 2
j π −j π 
2

gle is computed as follows:  + e 3 ( Ae r e 3 + Βe r e 3 ) 

jα jβ
 
V
r = Ar e jα r
V
Xr * = A (10) 2
2
(14)
r
arg( X r *) = α r + K f
(11) It is noted that the phasor does not contain the
= φ − π ∆f ( N − 1 )TS + 2π ∆frTS + K f
oscillatory error. Therefore, the deviation of fre-
14th PSCC, Sevilla, 24-28 June 2002 Session 01, Paper 5, Page 4

quency is detected without applying the running case when voltage wave contains harmonics, the
average filter. In this case, the off-set error of the detected frequency contains the significant error.
r.m.s. value and the phase angle is also compen- Therefore, in general, the low-pass filter is applied
sated by Eq. (12) and Eq. (13) in the similar man- in order to remove the harmonics[8].
ner as the single phase case. In this paper, two cases of simulations are car-
The flow chart of the proposed frequency detec- ried out in order to compare four kinds of fre-
tion method and compensation method of the r.m.s. quency detection method as follows:
value and the phase angle is shown in Fig. 4(a) for
single-phase case, and in Fig. 4(b) for three-phase Simulation case
case, respectively. 1. Power source without harmonics
2. Power source with 5% fifth order harmonics
3 SIMULATION RESULTS
Frequency detection method
In this section, the effectiveness of the proposed 1. Single-phase zero-cross method
frequency and phasor detection method is con- 2. Three-phase zero-cross method
firmed with numerical simulations. The proposed 3. Proposed single -phase method
method is compared with the zero-cross detection 4. Proposed three-phase method
method.
The zero-cross method is the frequency detec- At any cases, the balanced three-phase voltage
tion method by measuring the time interval between is used, and the frequency of voltage has step
zero crossing points of the instantaneous voltage. change from 60.00 Hz to 60.10 Hz. The number of
The frequency is detected every 180 deg by using sampling N is 24, in other words, sampling fre-
single-phase voltage, and every 60 deg by using quency is 1440 Hz.
three-phase voltage, respectively. However, in the

r
r Xr r
Xr * r
Xr Compensation X r **
vr r.m.s. value Running
Recursive- by
& average r
DFT Eq. (12) & (13)
phase angle filter arg( X r ) * *
r r
arg( X r ) arg( X r *)

Differentiat - ∆f r
ing ∆f r
By Eq. (9)

(a) Single-phase case

r r
r X p ,r X p ,r * r
r X p ,r r.m.s. value Running Compensation X p,r * *
va ,r X a ,r
Recursive- & average by
DFT Positive- phase angle filter Eq. (12) & (13)
r phase- r r
vb ,r r
X b ,r sequence arg( X p ,r ) arg( X p ,r *) arg( X p,r ) * *
Recursive-
component
DFT
By Eq. (14)
r
vc ,r X c ,r Differentiat - ∆f r
Recursive- ing ∆fr
DFT By Eq. (9)

(b) Three-phase case

Fig. 4 Flow diagram of frequency and phasor detection
14th PSCC, Sevilla, 24-28 June 2002 Session 01, Paper 5, Page 5

3.1 Case-1 Fig. 6 shows the phase angle compensated by de-

Fig. 5 shows the simulation results of the case-1. tected frequency. In Fig. 6(a), it is found that when
Fig. 5(a) shows the instantaneous voltage wave. frequency of power source is different from the
Fig. 5(b) shows the frequency detected by zero- fundamental frequency, the phase angle detected
cross methods. In Fig. 5(b), both single - and three- from single-phase component contains oscillatory
phase zero-cross methods have dead time to detect errors. The off-set and oscillatory error are com-
frequency in step wise. However, the proposed pensated by proposed method within 1.5 cycles. On
methods in Fig.5(c) and (d) detect in lump wise, and the other hand, in Fig. 6(b), it is found that phase
the changes of frequency can be detected instanta- angle detected from positive-sequence component
neously. In particular, the proposed three-phase contains only off-set errors. The error is also
method shows good transient performance without compensated by proposed method within 1.5 cycles.
running average filter. If the fluctuation of fre- Therefore, the precise detection of phase angle is
quency is within ±2.0 Hz, the proposed methods realized by the proposed compensation method. The
enable to detect frequency in the same way. r.m.s. value is compensated in the similar way.

60.00Hz 60.10Hz
1.5 Frequency Deviation [Hz] 0.12
1.0
0.08
Voltage [v]

0.5
0.0 0.04
-0.5
0.00
-1.0
-1.5 -0.04
0 17 33 50 67 83 100 117 133 150 0 17 33 50 67 83 100 117 133 150
Time [ms] Time [ms]
va vb vc
input zero1 zero3
(a) Instantaneous voltage (b) Zero-cross method

0.24 0.12
Frequency Deviation [Hz]
Frequency Deviation [Hz]

0.20
0.16 0.08

0.12
0.04
0.08
0.04 0.00
0.00
-0.04 -0.04
0 17 33 50 67 83 100 117 133 150 0 17 33 50 67 83 100 117 133 150
Time [ms] Time [ms]
input without RAF with RAF input without RAF with RAF

(c) Proposed single-phase method (d) Proposed three-phase method

Fig. 5 Step response of case-1
Error of Phase Angle [deg]

Error of Phase Angle [deg]

0.10 0.10

0.00 0.00

-0.10 -0.10

-0.20 -0.20

-0.30 -0.30

-0.40 -0.40
0 17 33 50 67 83 100 117 133 150 0 17 33 50 67 83 100 117 133 150
Time [ms] Time [ms]
before after before after

(a) Single-phase method (b) Three-phase method

Fig. 6 Compensation results of phase angle
14th PSCC, Sevilla, 24-28 June 2002 Session 01, Paper 5, Page 6

3.2 Case-2 cross methods enable to detect frequency precisely

Fig. 7 shows the simulation results in the case by applying the low-pass filter to the voltage. How-
when the voltage wave contains harmonics. Fig. ever, in this way, the delay time to detect frequency
7(a) shows the instantaneous voltage wave. Fig. becomes longer than the case without low-pass
7(b) and (c) shows frequency detected by zero- filter. Finally, dead time become to be about 2.0
cross methods. It should be noted that the both sin- cycles.
gle and three phase zero-cross methods are very On the other hand, in Fig. 7(d) and (e), the pro-
sensitive to harmonics. This is due to that the zero- posed methods are less affected by harmonics and
cross point of voltage wave is shifted from the zero- the delay time is within 1.5 cycles. There are two
cross point of the fundamental frequency wave by reasons why the proposed methods have essentially
the harmonics. robustness against harmonics. Firstly, proposed
Therefore, in the case when the zero-cross meth- methods pick out fundamental frequency as Fig. 1.
ods are used, the low-pass filter is applied to remove Secondly, the running average filter applied to r e -
harmonics. The frequency response of the low-pass move the oscillatory error is also useful to remove
filter is shown in Fig. 8. The phase delay of this the affect of harmonics. Comparing Fig. 9(b)(c) and
filter is about 1.0 cycle. Fig. 9 shows the simulation Fig. 7(d)(e), it is found that the proposed methods
results by using zero-cross methods with low-pass enable to detect frequency more accurately and
filter. Fig. 9(a) shows the instantaneous voltage more instantaneously than zero-cross methods when
wave after filtering. In Fig.9 (b) and (c), the zero- the fluctuation of frequency occurs.

60.00Hz 60.10Hz
1.5
1.0
Voltage [v]

0.5
0.0
-0.5
-1.0
-1.5
0 17 33 50 67 83 100 117 133 150
Time [ms]
va vb vc
(a) Instantaneous voltage
0.12 3.00
Frequency Deviation [Hz]

Frequency Deviation [Hz]

2.00
0.08 1.00
0.00
0.04
-1.00
0.00 -2.00
-3.00
-0.04 -4.00
0 17 33 50 67 83 100 117 133 150 0 17 33 50 67 83 100 117 133 150
Time [ms] Time [ms]
input output input output

(b) Single-phase zero-cross method (c)Three-phase zero-cross method

0.24 0.16
Frequency Deviation [Hz]

Frequency Deviation [Hz]

0.20
0.12
0.16
0.12 0.08
0.08 0.04
0.04
0.00
0.00
-0.04 -0.04
0 17 33 50 67 83 100 117 133 150 0 17 33 50 67 83 100 117 133 150
Time [ms] Time [ms]
input without RAF with RAF input without RAF with RAF

(d) Single-phase proposed method (e) Three-phase proposed method

Fig. 7 Step response of case-2
14th PSCC, Sevilla, 24-28 June 2002 Session 01, Paper 5, Page 7

the case when the voltage wave contains harmon-

1.2
ics. Moreover, this method enables to detect fre-
1.0
quency instantaneously when the fluctuation of fre-
0.8
quency occurs. In addition, the r.m.s. value and the
Gain

0.6
phase angle is compensated by using the detected
0.4
frequency.
0.2
0.0
REFERENCES
0 60 120 180 240 300 360
Frequency [Hz]
[1] A.G.Phadke, J.S.Thorp, ”Improved Control and Pro-
Fig. 8 Frequency response tection of Power Systems Through Synchronized
of low-pass filter Phasor Measurements”, Control and Dynamic Sys-
tems, Advances in Theory and Applications, Vol.43,
60.00Hz 60.10Hz Academic Press, 1991, pp.335-376.
1.5
[2] P.J.Moore, R.D.Carranza, and A.T.Johns, ”Model
1.0
System Tests on a New Newmeric Method of Power
Voltage [v]

0.5
System Frequency Measurement”, IEEE Trans. on
0.0
Power Delivery, Vol.11, No.2, April.1996, pp.696-701.
-0.5
-1.0
[3] J.Z.Yang and C.W.Liu, “A Precise Calculation of
-1.5
Power System Frequency and Phasor”, IEEE Trans.
0 17 33 50 67 83 100 117 133 150 on Power Delivery, Vol.15, No.2, April.2000, pp.494-
Time [ms] 499.
va vb vc
[4] Yutaka Ota, Masatake Kawada, Hiroyuki Ukai, Koich
(a) Instantaneous voltage after filtering Nakamura, Nobuyuki Matsui, Mototaka Sone,
0.12
Frequency Deviation [Hz]

S.C.verma, Hideki Fujita, ”Synchronized Phasor

0.08 Measurement of State Parameters at Multi-points in
Electric Power System by using DSP and GPS”, Pro-
0.04
ceedings of International Conference on Signal Proc-
0.00 essing Applications and Tecnology(ICSPAT), No-
vember 1999.
-0.04
0 17 33 50 67 83 100 117 133 150 [5] Yutaka Ota, Hiroyuki Ukai, Koich Nakamura, Hideki
Time [ms] Fujita, ”Evaluation of Stability and Electric Power
input output Quality in Power System by using Phasor Measure-
(b) Single-phase zero-cross method ments”, in Proceedings 2000 International Conference
0.12
Power System Tecnology(PowerCon2000), vol.3, De-
Frequency Deviation [Hz]

cember 2000, pp.1335-1340.

0.08
[6] Tsuyoshi Funaki, Kenji Matsuura, Shunsuke Tanaka,
0.04 “Error Estimation and Correction of Detectied Phase
by Real Time DFT”, Trans. IEE of Japan, Vol.120-B,
0.00
No.12, December 2000, pp.1682-1690. (in Japanese)
-0.04 [7] Katsuyasu Nakano, Yutaka Ota, Hiroyuki Ukai, Koich
0 17 33 50 67 83 100 117 133 150
Nakamura, Hideki Fujita, ”Application of Phasor
Time [ms]
input output
Measurement Method to Frequency Detection of
Power System”, Trans. IEE of Japan, Vol.121-B, No.12,
(c) Three-phase zero-cross method December 2001, pp.1856-1857. (in Japanese)
Fig. 9 Step response of case-2 with LPF
[8] Tomio Chiba, Mitsuyasu Kido, Junzou Kawakami,
Katsuhiko Yoneda, Tadao Kawai, Kotaro Hirasawa,
“High-Precision Voltage Measuring Method using
4 CONCLUSIONS Fast Sampled Data for Electric Power Systems ”,
Trans. IEE of Japan, Vol.113-B, No.5, May 1993,
This paper proposed a new detection method of pp.553-561. (in Japanese)
frequency in electric power systems. The proposed
method based on recursive DFT algorithm realizes
the accurate detection of frequency, in particular, in