Professional Documents
Culture Documents
a r t i c l e in f o a b s t r a c t
Article history: A frequency spectrum refinement and correction algorithm has been put forward to improve the
Received 27 November 2009 accuracy of measurements, and the Cramer–Rao lower bound (CRLB) of Gaussian group Doppler signal
Received in revised form was given based on the introduction of acceleration. Results of simulations and experiments showed
4 January 2010
that the Goertzel refinement algorithm could improve the resolution of the spectrum of the Doppler
Accepted 18 January 2010
Available online 6 February 2010
signal; the ratio correction algorithm made the results closer to the real value. The CRLBs of the
estimated parameters were related to the sampling data, the signal to noise ratio (SNR) and the
Keywords: broadness of Gaussian group, and it can be decreased by increasing the length of the sampling data or
Laser Doppler improving the SNR; the gap between the variances of the measuring results and the CRLBs narrowed
Signal processing
when the SNR of the signal was improved, and was almost eliminated when the SNR was higher than
Cramer–Rao low bound
6 dB.
& 2010 Elsevier Ltd. All rights reserved.
1. Introduction relating to the velocity of the measured point and many kinds of
noise, such as photodetection shot noise, light scattered from
Laser Doppler velocimeter (LDV) has become a very useful tool outside the measuring volume, and so on. To better estimate the
for the measurement of its own velocity in the vehicle inertial Doppler frequency of noisy signals, frequency domain processing
navigation system [1,2]. Precision of measurements, fast response techniques are used. With the fast development of digital signal
time and low invasiveness are among the advantages leading processing, the fast Fourier transform (FFT) is usually used in
to the success of LDV, in which the Doppler shifted and processing Doppler signals. The power spectrum S of a discretized
backscattered light from a target is coherently mixed with the Doppler signal x is given by
source light to produce a beat-frequency intensity modulation.
Fast Fourier transform (FFT) is used to obtain the velocity by X
N 1
Sk ¼ j xn expðj2pkn=NÞj2 ð1Þ
calculating the beat-frequency of the signal spectrum. The n¼0
Cramer–Rao low bound (CRLB) is often used to evaluate the
accuracy degree of the measurement for signal processing. where N is the number of discrete samples and
Firstly, if we regard the frequency of the spectral peak k= N, N+ 1,y,N 1. The Doppler frequency is given by the
as Doppler frequency, the accuracy would not be high [3]; secondly, peak of the spectrum.
ignoring the acceleration, the CRLB of simple-frequency sinusoidal On the other hand as a matter of fact, if we put the frequency
signal was used as an approximate value in previous literature [4]. of spectral peak as the Doppler frequency, the accuracy would not
So, in this paper, a frequency spectrum refinement and correction be high because the frequency resolution of FFT is given by
algorithm was put forward to improve the accuracy of measure-
ments, and the CRLB of Gaussian group Doppler signal was given Df ¼ fs =N ð2Þ
based on the introduction of acceleration.
where fs is the sampling frequency.
It seems that we can improve the frequency resolution by
decreasing the sampling frequency fs or increasing the number
2. Signal processing
of the sampling data N. However, fs cannot be decreased because
of the wide range of the measured velocity, and N cannot be
The primary result of LDV is a current pulse from the increased because of the huge extent of computation. Hence a
photodetector. This current contains the frequency information frequency spectrum refinement and correction algorithm
was used to improve the frequency resolution. To assist in
Corresponding author. Tel.: + 86 13739091383; fax: + 86 073184576314. understanding, a block diagram of the signal processing is shown
E-mail address: wttzhoujian@163.com (J. Zhou). in Fig. 1.
0030-3992/$ - see front matter & 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.optlastec.2010.01.027
ARTICLE IN PRESS
J. Zhou, X. Long / Optics & Laser Technology 42 (2010) 1038–1043 1039
2.1. Frequency spectrum refinement using Goertzel algorithm [5–7] x (n) yk (n)
Z ¼ ½ oD rD A j0 Tw T ð21Þ
In Eq. (26), Kw =Tw/T is the width of the Gaussian group and
Based on the theory of mathematical statistics, we need to pffiffiffiffi R x 2
calculate the corresponding Fish matrix J at first to acquire the erf ðxÞ ¼ ð2= pÞ 0 er dr is the error function. Using Eqs. (25) and
CRLBs of the parameters. The relationship is given by [14] (26), the first and second diagonal elements are then given by
From Eqs. (29) and (30), we know that the CRLBs of the
In actual measurements, integral can replace the summation in Doppler circular frequency and its first order rate of change are
Eq. (22) because the sampling frequency fs is often much related to the number of sampling data, the SNR and the width of
larger than the estimated Doppler frequency fD. Hence, Eq. (24) the Gaussian group.
ARTICLE IN PRESS
J. Zhou, X. Long / Optics & Laser Technology 42 (2010) 1038–1043 1041
4. Simulation and experiments Table 2 shows the comparison of results before and after
correction. From the table, we know the error of the frequency is
Simulation of sinusoidal signals with Goertzel refinement 1.56%, 1.56% and 0.39%, respectively, using FFT directly when the
algorithm and ratio correction algorithm as well as comparison frequency of the signal is 100, 200 and 300 kHz; the average error
between the results and their FFT are carried out. The parameters of the frequency is 0.0006%, 0.0026%, and 0.0021%, respectively,
of the simulation and the signal are shown in Table 1. after the frequency correction when the frequency of the signal is
Fig. 3 shows the comparison of results between the spectrum the same as above. Obviously the estimated frequency with
after Goertzel refinement and FFT directly. From the marked correction is much closer to the real value than that without
points in Fig. 3(d)–(f), we know that Df= 7812.5 Hz using FFT correction. The accuracy of the estimated frequency improved
directly when the frequency of the signal is 100, 200 and 400 kHz. considerably.
From Fig. 3(a)–(c), we know that Df =1953.125 Hz after the When the velocity of the moving vehicle was 1.554 m/s
frequency spectrum refinement. The resolution of the spectrum measured by a high-accuracy GPS, the data acquisition system
improved considerably. Fig. 3(a)–(c) also tells us that we can just collected a sequence of the Doppler signal. The corresponding
refine part of spectrum that we interested, so the extent of the Doppler frequency was 1.84 MHz, while the sampling frequency
calculation of the algorithm can be reduced considerably. was 10 MHz and the number of sampling data was 1024 in the
experiment. The Doppler signal and its spectrum after refinement
and correction are shown in Fig. 4. From this figure, we know the
Goertzel refinement algorithm improved the resolution of the
Table 1 Doppler signal spectrum, and the ratio correction algorithm made
Parameter configuration in the simulation. the results closer to the real value.
Fig. 5 shows the relationship among SNR and CRLBs of Doppler
Parameters of the simulation and the signal
circular frequency as well as its first order rate of change and SNR
Sampling frequency 2 MHz with different numbers of sampling data. This figure tells us that
Number of sample data 1024 the higher the SNR, the smaller the CRLBs, and the more the
Refining multiple 4 sampling data, the smaller the CRLBs.
Number of points for FFT 256 The variances of the results were calculated using the
Amplitude of the sinusoidal signal 1
Frequencies of the sinusoidal signal 100, 200 and 400 kHz
frequency spectrum refinement and correction algorithm, and
compared with the CRLBs. Fig. 6 shows the relationship between
amplitude spectrum
amplitude spectrum
X = 99609.375 X = 400390.625
2 2 X = 199218.75 2 Y = 229708.838
Y = 229413.6893
Y = 150781.37
1.5 1.5 1.5
1 1 1
X = 197265.625 X = 398437.5
0.5 X = 97656.25 0.5 0.5 Y = 14264.3571
Y = 6373.5306 Y = 12438.2973
0 0 0
0.5 1 1.5 2 1 2 3 4 1 2 3 4 5
f x 105 f x 105 f x 105
amplitude spectrum
amplitude spectrum
X = 101562.5 X = 398437.5
100 Y = 119.7465 100 Y = 119.4335
60 X = 195312.5
Y = 65.4005
40
50 X = 93750 50
Y = 29.9522 X = 390625
20 Y = 19.6473
0 0 0
0 0.5 1 1.5 2 0 1 2 3 4 0 1 2 3 4 5
f x 105 f x 105 f x 105
Fig. 3. Comparison of results between spectrum after Goertzel refinement and FFT directly: (a) spectrum after Goertzel refinement when the frequency is 100 kHz, (b)
spectrum after Goertzel refinement when the frequency is 200 kHz, (c) spectrum after Goertzel refinement when the frequency is 400 kHz, (d) result of FFT when the
frequency is 100 kHz, (e) result of FFT when the frequency is 200 kHz and (f) result of FFT when the frequency is 400 kHz.
ARTICLE IN PRESS
1042 J. Zhou, X. Long / Optics & Laser Technology 42 (2010) 1038–1043
Table 2
Comparison of results before correction and after correction.
Frequency before Frequency after Frequency before Frequency after Frequency before Frequency after
correction (Hz) correction (Hz) correction (Hz) correction (Hz) correction (Hz) correction (Hz)
0.015 1.5
0.01 X: 1.836e+006
amplitude spectrum
Y: 1.001
0.005 1
voltage
-0.005 X: 1.816e+006
0.5
Y: 0.2968
-0.01
-0.015 0
200 400 600 800 1000 1 1.5 2 2.5
n f x 106
2
refining spectrum
result of correction 1.6
amplitude spectrum
X: 1.836e+006 X: 1.839e+006
amplitude spectrum
0 1
1.5 2 2.5 1.82 1.83 1.84 1.85 1.86
f x 106 f x 106
Fig. 4. Actual measuring Doppler signal and its spectrum after refinement and correction: (a) the actual measuring Doppler signal, (b) result of FFT directly, (c) spectrum
after refinement and correction and (d) local enlarged image of the spectrum after refinement and correction.
the variances and the CRLBs of the Doppler frequency with its first measurement for LDV. The CRLB of Gaussian group Doppler signal
order rate of change when fs = 1 MHz, N= 128 and f=200 kHz. From was given based on the introduction of acceleration. The results of
this figure we know the gap between the variances and the CRLBs simulations and experiments showed that the error of the
narrowed when the SNR of the signal was improved, and was estimated frequency reduced owing to the technology of
almost eliminated when the SNR was higher than 6 dB. frequency spectrum refinement and correction; the CRLBs of the
estimated parameters are related to the sampling data, SNR, and
broadness of Gaussian group, and they can be decreased by
5. Conclusion increasing the length of sampling data or improving the SNR; the
gap between the variances and the CRLBs narrowed when the SNR
We have demonstrated that the frequency spectrum refine- of the signal was improved, and was almost eliminated when the
ment and correction algorithm can improve the accuracy of the SNR was higher than 6 dB.
ARTICLE IN PRESS
J. Zhou, X. Long / Optics & Laser Technology 42 (2010) 1038–1043 1043
70 160
65 150
60
140
55
130
CRLB/dB
CRLB/dB
50
45 120
40 110
35
N = 128 100 N = 128
30 N = 256 N = 256
25 N = 512 90 N = 512
N = 1024 N = 1024
20 80
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
SNR/dB SNR/dB
Fig. 5. Relationship among SNR and CRLBs of Doppler circular frequency (a) as well as its first order rate of change and SNR (b).
x 106 x 1015
8 4
7 3.5
6 3
CRLB
5 CRLB 2.5 variance of spectral
variance
Fig. 6. Relationship between the variance of the Doppler frequency with its first order rate of change and the CRLBs and SNR when N= 128 and f= 200 kHz: (a) Doppler
circular frequency and (b) the first order rate of change of Doppler circular frequency.