You are on page 1of 4

Proceedings of 2007 International Symposium on Intelligent Signal Processing and Communication Systems Nov.28-Dec.

1, 2007 Xiamen, China

NEW METHOD OF ESTIMATION OF PHASE ,AMPLITUDE, AND

FREQUENCY BASED ON ALL PHASE FFT SPECTRUM ANALYSIS


Huang Xiaohong1, Wang Zhaohua2,Hou Guoqiang3
1,3 2
School of Electronic & Information Engineering, School of Electronic & Information Engineering,
Hebei Polytechnic University, Tianjin University,
Tangshan, Hebei, China 063009. Tianjin, China 300072.
Email: tshxh@163.com

ABSTRACT
The traditional input data truncating way of spectrum analysis method[3], average method[4] and so on. A novel algorithm of
is improved and all phase FFT (apFFT) spectrum analysis is put high accuracy parameter estimation of sinusoidal signal based on
forward, we propose the realization diagram and analyze that all phase FFT spectrum analysis is put forward in this paper,
amplitude spectrum of all phase is the square of traditional which is better than all algorithms nowadays not only in the
spectrum analysis. The method of apFFT has perfect precision and the real time property, but also in the calculation
characteristic of phase analysis ,The phase isn’t influenced by complexity. The program of software is simple, so it is an ideal
frequency shift through the method of all phase and no need to measurement method. It is almost the perfect estimation in the
be corrected, the error is extremely small between the phase situation of no noise, especially to the phase estimation, the error
analyzed by apFFT and the signal real phase .We propose a new is only 0.001%.
algorithm of sinusoidal signal’s amplitude frequency and phase
estimation based on apFFT phase difference , simulation result 2. ALL PHASE FFT SPECTRUM
and comparable datas prove that the algorithm is more accurate ANALYSIS
and practicable than algorithms in literatures.
Index Terms- all phase spectrum analysis; parameter estimation; A novel algorithm of spectrum estimation is put forward in the
[5]
phase difference; correction literature , which improves the data truncating way of
traditional DFT spectrum analysis and reduces the leakage
1. INTRODUCTION greatly,the block diagram is shown in the bottom of the figure 1.

We need to estimate the frequency、amplitude and phase of


single-tone signal in many applications. For example, the
measurement of fundamental of the grid voltage and current in
the power system is performed through the traditional DFT
frequency spectrum method. When the signal is
non-inter-period sampled, the leakage error occurs, which makes
the frequency、amplitude and phase deflect from the real value,
especially the phase’s error. There are many improved
algorithms in many literatures: Interpolated FFT method
[1]
,phase difference correcting method[2], tapered-window

1-4244-1447-4/07/$25.00 ©2007 IEEE 284


Proceedings of 2007 International Symposium on Intelligent Signal Processing and Communication Systems Nov.28-Dec.1, 2007 Xiamen, China

X’(
i k)is the discrete Fourier transform of X’i(i=0,1,…N-1) and Xi
(k)is the discrete Fourier transform of Xi(i=0,1,…N-1).

2πki
j
' N
X i ( k ) = X i ( k )e (1)

All phase spectrum is made up of the sum of X’i(k),so:


2πki
1 N −1 ' 1 N −1 j
X AP ( k ) = ∑ X (k ) = ∑ X ( k )e N
N i=0 i N i=0 i

f
Fig. 1. The diagram of traditional FFT spectrum and all phase j 2π 0 ( N − i + n ) 2π 2πki
1 N − 1N − 1 f − j kn j
FFT spectrum analysis = ∑ ∑ e s e N e N
N i = 0n = 0
First, we deduce the amplitude of a signal consisting of a
single frequency f0 .If the signal with single frequency is
f k
2 πf 0 f 0 Sin 2πN ( 0 − )
j n
1 j 2π N fs N
fs
x=e fs
,where f0 is the signal frequency, fs is the sample = e (2)
N 2 f k
frequency.To one sample point x(N) in the time sequence, there Sin π ( 0 − )
fs N
are N vectors of N dimension including this sample point:

T
According to (2),the amplitude of all phase spectrum is as follows
X 0 = [ x ( N ) x ( N + 1) L x (2 N − 1)]
2
T f k
X 1 = [ x ( N − 1) x ( N ) L x (2 N − 2)]
sinπN ( 0 − )
......................... 1 f N
s
, it is the square of traditional DFT
T N f
X N -1 = [ x (1) x (2) L x ( N )] k
sinπ ( 0 − )
f N
Cycle shift every vector, shift the sample x(N) to the first s
position of the sequence and get the other N vectors of N frequency spectrum amplitude, which is benefit to reducing the
dimension: spectrum leakage.

' T Another important character of all phase FFT spectrum


X 0 = [ x ( N ) x ( N + 1) L x (2 N - 1)]
analysis is that its phase is constant and isn’t influenced by the
' T
X1 = [ x ( N ) x ( N + 1) L x ( N − 1) ] frequency shift, so the phase needn’t to be corrected. That means

LLLLLLL the real phase of signal can get by all phase FFT spectrum

' T analysis when the signal is non-integer truncated . The measured


X N −1 = [ x ( N ) x (1) L x ( N − 1) ]
phase value and the real phase value are almost the same and the
We can get all phase data vector by adding N vectors aiming at error is extremly little. This property does not belong to the
x(N) traditional DFT spectrum analysis and it is the base of parameter
correction of all phase FFT spectrum analysis.
1
X AP = [ Nx ( N ) ( N - 1) x ( N + 1) + Take the signal cos(1.2×2π/6t+100π/180) as an example to
N
search the reason of so little phase error about all phase FFT
T
x (1) L x (2 N - 1) + ( N − 1) x ( N − 1)]
spectrum analysis. We can get 11 samples: -0.1736 -0.9903 -0.4384
According to the shift property of discrete Fourier transform, 0.7193 0.8829 -0.1736 -0.9903 -0.4384 0.7193 0.8829 -0.1736

there has clear relationship between the X’(


i k)and X(
i k), where The input signal of all phase is made up of 6 groups of N=6

285
Proceedings of 2007 International Symposium on Intelligent Signal Processing and Communication Systems Nov.28-Dec.1, 2007 Xiamen, China

samples . The first group consists of the last 6 samples among all the frequency resolution is 1Hz,
11 samples, the second group consists of another 6 samples s1(t)=cos(9*2πt/N+50π/180)+cos(19.1*2πt/N+50π/180)+cos(29.2
which left shift 1 value, but -0.1736 should right shift to the first *2πt/N+50π/180)+cos(39.3*2πt/N+50π/180)+cos(49.4*2πt/N+50
position, other groups could be get as the same way. π/180).The result of all phase hanning(N) double window FFT
-0.1736 -0.9903 -0.4384 0.7193 0.8829 -0.1736 spectrum analysis and traditional DFT spectrum analysis are in
-0.1736 -0.9903 -0.4384 0.7193 0.8829 0.8829 figure 2(a)、(b) .If change the signal’s phase and generate the
-0.1736 -0.9903 -0.4384 0.7193 0.7193 0.8829 signal:s2(t)=cos(9*2πt/N+10π/180)+cos(19.1*2πt/N+30π/180)+c
-0.1736 -0.9903 -0.4384 -0.4384 0.7193 0.8829 os(29.2*2πt/N+50π/180)+cos(39.3*2πt/N+70π/180)+cos(49.4*2π
-0.1736 -0.9903 -0.9903 -0.4384 0.7193 0.8829 t/N+90π/180),Using the same analysis methods, the analysis
-0.1736 -0.1736 -0.9903 -0.4384 0.7193 0.8829 result is shown in figure 2(c)、(d).
The phases of the samples of the 6 group of N=6 signals are :
180.0000 132.5107 -25.7699 0 25.7699 -132.5107

0 112.9053 51.7298 180.0000 308.2702 247.0947

0 112.4978 53.4788 180.0000 306.5212 247.5022

180.0000 88.4598 138.4375 0 -138.4375 -88.4598

180.0000 83.6275 161.4769 0 -161.4769 -83.6275

180.0000 72.5107 214.2301 180.0000 145.7699 287.4893

Because the frequency is 1.2, we should observe the second


phase in every group. Three are bigger and three are smaller than
the real phase (100) during 6 phases. Input data of all phase is
the average of the above 6 groups signals, phases are counteract
each other, which make phase difference zero. So the phase got
by all phase spectrum analysis is the signal real phase. The result Fig. 2. Phase spectrum analyzed by traditional DFT and all phase
of experiment shows that when the signal is inter-period spectrum analysis
sampled ,the phase got by all phase no window FFT spectrum It is obvious in the figure 2 that the measurement phase is
analysis is perfect; when the signal is non-inter-period extremely close to the real signal phase by all phase double
sampled ,the phase got by all phase double window FFT window analysis, while the measurement phase is far away from
spectrum analysis is most close to the real phase, during which the real signal phase by traditional DFT analysis.
the double window is the convolution of kaiser(N,9.5)window
and itself. In this case, the phase of signal got by all phase
3. PARAMETER ESTIMATION OF
kaiser(N,9.5) double window spectrum analysis is as follows:
SINUSOIDAL SIGNAL BASED ON apFFT
180.0000 100.0069 100.0004 180.0000 259.9996 259.9931
PHASE DIFFERENCE
In this case, the real phase is 100, the measured phase is
100.0069,so the error is only 0.69%, we can think them very
The principle of phase difference is given in the literature[2],
similar. According to the above example, we can get the
the acquire of the sequence is improved according to the method
conclusion that all phase has perfect phase analysis property,
of apFFT phase difference. The algorithm is as follows:
especially when the signal is non-inter-period sampled, the phase
(1)Non- inter-period sample a certain length of signal, get a
analyzed by this method is almost the real value, while the phase
sequence u1(t) with the length of 2N-1,which starts from the
analyzed by traditional method is deflect from the real value.
No.i sample , and get another sequence u2(t) with the length of
Now we look at a concrete simulation: generate the following
2N-1,which starts from the No.i+N sample .
signal, the sampling frequency is 128Hz, the FFT point is 128,
(2)Perform N point all phase double window (convolution of

286
Proceedings of 2007 International Symposium on Intelligent Signal Processing and Communication Systems Nov.28-Dec.1, 2007 Xiamen, China

Hann and hanning ) FFT spectrum analysis to the sequence difference method and phase difference method in the literature
u1(t),the frequency can be got through the following formula: under the situation of no noise
fs fs amplitude frequency phase
f = k1 ,the frequency resolution F0 = ,where k1 is
N N Ideal value 1.0000000 4.3900000 40.0000000

the serial number of the main spectrum line,d is the coefficient apFFT phase difference 0.9999999 4.3900000 39.9999999
[6]
Method in the literature 0.9988161 4.3868350 40.5486832
of leakage error,the frequency correction is ∆f = dF0 。
5. CONCLUSION
(3)Frequency after signal leakage error correction is:
^ This paper improves the way of input data truncation based on
f = f + ∆f = f + dF0 = (k1 + d ) F0 (3)
the traditional spectrum analysis, and brings forward all phase
spectrum analysis, which has less leakage and high precision
at the same time,we can get phase ϕ1 .
compared to the traditional spectrum analysis. There is no need
(4) Perform N point all phase double window (convolution of to inter-period sample signal strictly .This algorithm can
Hann and hanning ) FFT spectrum analysis to the sequence u2(t), restrain leakage and picket effects of discrete Fourier
transform(DFT) effectively, and no need to do phase correction.
we can get phase ϕ2 .according to the phase difference
The algorithm is simple, and it can be used in many engineering
and has great practical value, such as in the vector measurement
∆ϕ = ϕ2 − ϕ1 = 2π d , we can get the leakage error
of power system automation and the harmonic analysis in the
coefficient d。 sinusoidal alternating current signal, and signal processing in the
(5)Correct frequency、amplitude、phase。Corrected frequency can electronic reconnaissance.
be obtained through (3) when d is known. Let the phase

difference of the main spectrum line is ∆ϕk 1 ,the approximate 6. REFERENCES

[1]Vijay K Jain,William L Collins and David C David,


amplitude of the main spectrum line is Ak1 ,then the amplitude
“High-Accuracy Analog Measurements via Interpolated FFT,”
correction formula is as follows: IEEE Trans on IM,vol.im-28,pp. 113-122, 1979.
[2]Cao Guojian,HuangChun,“Research on quasi-synchronization
2π∆ϕ k1 (1 − ∆ϕ k21 ) Ak1 2
A=( ) (4) algorithm based on time-shifting and phase difference
2 sin(π∆ϕ k1 )
correcting ,”RELAY,vol. 32(13),pp.1-5, 2004.
The initial signal phase is as follows: [3]H Rensers,J Schoukens,G Vilain,“High-Accuracy Spectrum
Analysis of Sampled Discrete Frequency Signals,by Analytical
ϕ 0 = 2ϕ k1 − ϕ k 2 (5)
Leakage Compensation, ” IEEE Trans on IM,
4. COMPUTER SIMULATION RESULT vol.33(4),pp.287-292, 1984.
[4]Wang Zhaohua,Hou Zhengxin,Sufei,“All phase FFT spectrum
The sampling frequency is fs=1024 and the spectrum analysis,”Jounal of comunications,vol.24,pp.16-19,2003.
analysis points are N=1024, the sampled cosine signal with [5]Dingkang, Zhu Xiaoyong,“ Synthesized correcting method of
single frequency is s(t)=cos(4.39*2πt/fs+40π/180). Parameters phase difference on discrete spectrum, ” Journal of vibration
are estimated through apFFT phase difference method and phase engineering , vol.15(1),pp.114-117, 2002.
difference method in the literature[6] separately, the result is [6] Zhu Xiaoyong, Dingkang,“ The synthetical comparision of
shown in table 1. correcting methods on discrete spectrum,”Signal processing,
Table 1 correction results obtained through apFFT phase vol.17(1),pp.91-97,2001.

287

You might also like