Upload
asr
View
213
Download
1
Embed Size (px)
Citation preview
Doppler Profile Estimation in VHF Radar Signals
using Wavelets 1Leela Lakshmi. S.,
Assistant Professor(Sr.,)
Dept. of ECE, SKIT
Srikalahasti-517640
2Rajani Kanth .V,
Assoc. Professor
Dept. of EEE, SKIT
Srikalahasti-517640
3Varada Rajan. S.,
Professor
Dept. of EEE, SVU
Tirupati-517502
A.S.R.Reddy
Professor
Principal, SKIT
Srikalahasti-517640 [email protected]
Abstract- The wind profile estimations are the fundamental
objective of the Indian MST radar with the received signal
ranging from nearly 3.5 25 Km height. A widely accepted
narrow band model is considered, for the analysis of the
signal received. Though the received signal SNR is stronger
at lower altitudes, as the height progresses the strength
becomes feeble and the noise dominates the received signal.
The classical Time-Frequency analysis is made on the
received signal from different altitudes and observed that the
zero Doppler influence is significant at higher altitudes along
with the noise. A Hilbert transform based DC bias removal is
proposed in contrast with the existing 3 point MA technique.
To handle the noise, in particular at higher altitudes a
wavelet based adaptive denoising is proposed and found
satisfactorily for the spectral cleaning of the Doppler
spectrum. The validation of the proposed algorithm is made
with the GPS data.
Key words: doppler, radial velocity, denoise, threshold
I. INTRODUCTION
Atmospheric research in India has a hoary tradition over
200 year old legacy by the then British India Company at
Colaba Observatory station near Mumbai for the purpose
of meteorological observations. To unravel the mysteries
of the topological of the Indian sub-continent many
programs were instigated, one such program was the
Middle Atmosphere Program (MAP) [1]. The principle
objective of the MAP is to explore the secrets of the
activity of the middle atmosphere on all temporal and
spatial scales and also to untangle the interactions between
the three different height regions viz. Mesosphere,
Stratosphere and Troposphere. The Radar is setup by the
ISRO at NARL (National Atmospheric Research
Laboratory), Gadanki (near Tirupati), Chittoor (dist), A.P.;
@ to study these
regions. This atmospheric radars are also known as clear
air radars and they operate typically in the VHF
( ) band; admirable instrumentation system
for atmospheric probing in the regions of Mesosphere,
Stratosphere and Troposphere (MST) covering up to a
height of about 100 Km generally [1]. It is also used for
coherent backscatter study of the ionosphere irregularities
above 100 km. These radars also used for meteor detection
in the as a tracer of background wind at 70-120 Km in the
astronomical studies. The necessary technology to setup
VHF radar is cheaper and easier to achieve [2]. However,
the VHF region is crowded, band widths are narrow,
external noise can be high, and beamwidths are broad.
Turbulent fluctuations in the refractive index of the
atmosphere serve as a target for these radars. Particularly,
the noise effect becomes more and more predominant as
the height progress beyond 12 Km in case of soft targets.
Therefore, there is a greater need for need to induct the
signal processing techniques for cleaning the Doppler
spectrum, used for wind profile estimation at different
altitudes.
II. 2.0 MST RADAR SIGNAL MODEL
A widely accepted model, without loss of generality, of
the radar signals i.e. narrowband signal model [2] is
considered as the velocity of the target (layers of the
atmosphere in the present case) is much lesser than the
speed of the propagation of the transmitted/received
signal. Accordingly [3], the received signal is given by:
where is the Doppler frequency, - time lag.
The classical Fourier transform (also in turn the
spectrum estimation) of the signal is given
mathematically as:
If, the received signal represented by eq. ( ) consists of
a single frequency component, it is very straightforward to
extract the information from the target like its velocity,
height, etc. In practice, the received signal is much
contaminated with noises that degrade the quality of the
signal received. Hence, it is a cumbersome process to
compute the spectrum of the signal and becomes intricate
to extract the parameters. Thus, the Atmospheric Signal
Processing has been one field of signal processing where
there is an ample scope for development of novel and
efficient tools for cleaning of the spectrum of the signal,
TABLE I: RESOLUTION DETAILS OF MST RADAR
(SOURCES NARL, ISRO, GADANKI, INDIA)
Sampled
Time ( )
Radial
velocity
resolution
(m )
Doppler
resolution
( )
Range of
Radial
Velocity
( )
Range of
Doppler
Shift ( )
256 512 256 512
4 2.8 1.4 1.0 0.5 ±353.75 ±125.0
8 1.4 0.7 0.5 0.25 ±176.87 ±62.5
16 0.7 0.35 0.25 0.12 ±88.44 ±31.25
32 0.35 0.17 0.12 0.06 ±44.23 ±15.63
36 0.17 0.09 0.06 0.03 ±22.10 ±7.81
128 0.09 -- 0.03 -- ±11.10 ±3.91
256 -- -- -- -- ±5.12 ±1.95
512 -- -- -- -- ±2.77 ±0.98
243978-1-4244-8594-9/10/$26.00 c©2010 IEEE
and estimation of the desired parameters. The MST radar
data processing is presented in the following sections.
III. MST RADAR DATA FORMAT
The Indian MST radar data is stored/available in two
formats for the scientists and/or user scientists. They are:
( ) the Time domain data (discrete time series data,
commonly referred as raw data), ( ) the Frequency
domain (spectrum computed using FFT) data, ( )
Moments data, ( ) Power spectrum + moments data and
( ) UVW data types. The data types ( ), ( ), ( ) & ( )
are preprocessed data computed from the discrete time
series data without applying any signal processing
techniques. The typical transformation is a simply the
Fourier transformed data and statistical data only. There is
a little to do by the signal processing engineer with this
data. However, the discrete time series data received from
the atmosphere is associated with enormous amount of
information [4]. The clear air turbulence is essentially a
volume target and the volume reflectivity in the
atmosphere ranges between to
or lower. Thus, the typically the power
aperture values of the order of to are
required for successful MST radar campaign. In most
cases the echo is buried below/under the background
noise/clutter ranges typically to . Therefore,
new advanced signal processing techniques and efficient
algorithms are required to extract the required information.
The spectral estimation of the radar return signal using eq.
(2), is simple and straight, finds useful in extracting the
desired information. The data processing is done in on-line
mode and off-line mode. The data processing operation
consists of applying spectral analysis techniques to the
received radar signal to estimate Doppler spectrum. This is
further useful in estimating the radial velocity and absolute
velocity of the wind. A typical Doppler spectrum of MST
radar is shown in fig 1. The profile of the Doppler
spectrum is clearly traceable at lower ranges of altitude.
As the height progress the spectrum is much contaminated
by the noise in turn the SNR becomes feeble and hard to
trace the wind profile. The table I depicts few technical
details of the Indian MST radar, NARL, Gadanki. To
understand the content of the signal it is required
mathematical tools other than Fourier transform due to the
fact that the classical Fourier transformation is localized in
frequency domain rather time-frequency. This limitation
can be overcome using time-frequency analysis.
IV. TIME-FREQUENCY ANALYSIS
The joint transforms were
used for the purpose of characterizing the time-varying
frequency content of a signal. The best known time
frequency representation of a time signal is short-time
Fourier Transform ( ), also referred as Gabor
transform. The most convenient approach to explore a
signal with content is to split the signal into many
segments in the time domain, and then taking the Fourier
transform. This slant is called as the short time Fourier
transforms ( ) and is given by:
Mathematically, the Gabor window function is given by
(time domain representation):
The fig.2 depicts the time domain representation of the
radar signal at different altitudes. An ample number of
window functions are available in the literature and are
useful to find the frequency content with in temporal
domain, is shown in fig. 3 & 4. The table II depicts figure
of merit made by the observations for different gating
functions when used as the kernel windows in the
analysis of equation ( ). It is observed that the effect of
DC bias in the signal is not much intervening at lower
altitudes particularly nearly below 12 Km and beyond this
Fig. 1: Typical Doppler Spectrum of MST radar (Doppler Frequency in
Hz Vs Height in Km). Fig. 2.a: Time domain representation of MST radar signal (received at nearly 4.25Km).
Fig. 2.b: Time domain representation of MST radar signal (received at
nearly 12.5Km).
Fig. 2.c: Time domain representation of MST radar signal (received at
nearly 22.5Km).
244 2010 International Conference on Signal and Image Processing
range the dominance of the DC component in the signal
progressively increases and causes for erroneous
estimations of wind velocities using the Doppler spectrum.
The wind velocities , using the Doppler frequency is
estimated mathematically as:
and vertical height corresponding to this velocity is
given quantitatively as:
where - velocity of light in free space, - carrier
frequency, - range time delay, - beam tilt angle.
It is noted that except Dolph-Chebyshev window [6], all
the windows introduces coherent/DC gain in the signal
(refer to the observation table II). However, the Dolph-
Chebyshev results are deterring at altitudes beyond 16Km
as shown in fig.5b.
To handle this DC bias present in the signal component,
a Hilbert transform based technique is proposed. At
present the NARL uses a moving average method for
reducing the effect of DC component in the Doppler
spectrum based on 3 point averaging technique.
However, it finds its own limitations that this method also
fails at altitudes beyond 12 Km. The proposed Hilbert
transform removes the DC component which is present in
the signal [6] and the results are shown in fig.6, which is
superior to the multi taper technique [11]. Yet, this
technique cannot eliminate the noise present in the signal.
Therefore, further effective signal processing techniques
require handling the noise for the improvement the SNR of
Fig. 3.a: Time Frequency plot using Rectangular windowforMST radar
signal (received at nearly 12 Km).
Fig. 3.b: Time Frequency plot using Gaussian windowforMST radar
signal (received at nearly 12 Km).
Fig.3c: Doppler Spectrum/Frequency domain representation of MST radar signal (received at nearly 12 Km).
Fig. 4.a: Time Frequency plot using Rectangular windowforMST radar signal (received at nearly 22.5Km).
Fig. 4.b: Time Frequency plot using Dolph-Chebyshev window for
MST radar signal (received at nearly 22.5 Km).
Fig. 4.c: Doppler Spectrum/Frequency domain representation of MST
radar signal (received at nearly 22.5 Km).
Fig. 5.a: Time Frequency plot using Rectangular window of MST radar
signal using proposed DC removal (received at nearly 22.5 Km).
Fig. 5.b: Time Frequency plot using Dolph-ChebyshevMST radarsignal using proposed DC removal (received at nearly 25 Km).
Fig. 5.c: Doppler Spectrum /Frequency domain representation of MST
radar signal using proposed DC removal using Hilbert Transform (received at nearly below 22.5 Km).
Fig. 5.d: Time domain representation of MST radar signal using proposed
DC removal using Hilbert Transform (received at nearly 22.5 Km).
2010 International Conference on Signal and Image Processing 245
the signal. The wavelets, a special class of time-frequency
tool, gives the signal information that is localized in both
time and frequency domain, because of their finite energy
and also finite duration of oscillatory in nature rather
harmonically continuous. This oscillatory behavior is the
desired property that a signal processing expert needs. In
the subsequent sections the analysis of the radar signals
using wavelet transform is presented.
V. WAVELET ANALYSIS
finite energy, localized in time. It has an oscillating
wavelike characteristic but also has the ability to
permit/allow simultaneous time and frequency analysis.
This makes wavelet to be a suitable tool for transient, non-
stationary or time-varying phenomena [7,8]. The basis
function in the recursion form is given by:
The coefficients are the inner product of
and . The transformation kernel is an
orthogonal basis. The desirable property seek by the
designers is orthogonalityi.e., the inner product of the
basis must be equal to zero. This means: is
orthogonal to all its dilations/scales and translations. The
wavelet basis, containing these basis functions
form an orthogonal basis. Mathematically:
In the first integral, equation , ,
whereas is positive and then negative over the
support defined. Thus the integral is zero.
Similarly on the second half-interval
where is positive and negative. The second
integral, equation , is therefore zero. The third integral,
equation , vanishes for a different reason that the
functions , do not overlap. One is zero
and where the other is nonzero simultaneously. So the
product , is vanishes everywhere [8] in
the domain as there is no overlap. However, the dyadic
analysis and dyadic synthesissystemobjects are used to
remove noise from a signal (discussed in the section 6).
These properties of wavelets find important applications in
signal denoising.
VI. ADAPTIVE DENOISING USING WAVELETS
The objective of the denoising or estimation function is
to recover the useful signal from the noisy version thereof.
The denoising is the major application of wavelets in
statistics. Let the signal model, equation ( ), has been
contaminated by the noise is given by:
where desired unknown noise free signal,
is an unobservable noise assumed to be
Gaussian in nature.
is the observed signal from the process.
The wavelet expansion of the above function can be
represented in the similar Fourier series analysis form
given mathematically as:
In many applications it is observed that the rapid
variations in the signal possess minimal magnitude. Thus,
a conceivable denoise stratagem consists of:
Fig. 6: Doppler Spectrum of MST radar using Hilbert Transform
(Doppler Frequency in Hz Vs Height in Km).
TABLE II: FIGURES OF MERIT OF WINDOWS
Window
Type
DC/Coherent
Gain
Side lobe
fall-off
(dB/octave)
Equivalent
Noise BW
Rectangle 1.00 -6 1.00
Blackman 0.46 -6 1.57
Blackman-
Harris 0.36 -6 2.00
Kaiser 0.42 -6 1.50
Triangular 0.50 -12 1.33
Bohman 0.41 -24 1.79
Taylor 0.57 -6 1.30
Dolph-
Chebyshev 0.53 0 1.39
Gaussian 0.51 -6 1.43
Bartlett-
Hanning 0.43 -18 1.61
Tapered
Cosine 0.64 -12 1.23
Hanning 0.42 -24 1.73
Hamming 0.54 -6 1.36
246 2010 International Conference on Signal and Image Processing
i. Keeping the approximations such that the noise is
absent or at least much attenuated the .
ii. Supplementing this approximation by parts if the
finer details , clearly ascribable to the
useful signal and rejecting the parts which are
regarded as stemming from the noise.
However, two modes of threshold schema are popular
referred as hard and soft. Therefore, by equality the
coefficients carry the noise component of the
equation . In order to reduce noise effect, through
equating to zero of equation and reconstructing
the result yields to , this is expected to
be free from noise.
The application of wavelet decomposition, unfasten the
high frequency sub-bands contain most of the noise
information and little signal information. Many denoising
methods are available in the wavelet literature for cleaning
the signal from contamination. Several denoising schema
like rigrsure, heursure, sqtwolog, minimaxi, etc., are
available for noise removal [9,10], from the contaminated
signal so as to improve the SNR of the signal. In the
current work the signal decomposition is made using non-
decimated wavelet analysis. This kind of redundant,
translation-invariant transform is especially useful for
denoising, which is one of the most important wavelet
applications. This transform however has a serious
limitation: ( ) the signal length must be a power of 2 and
( ) the periodized extension mode must be used for the
underlying DWT. In-spite of these limitations, this
technique eliminates the reconstructing the signal. The
fig.7illustrates the wavelet adaptive denoising
implementation of the radarreturn at various altitudes. The
SNR of the original signal and wavelet adaptive denoised
signal is shown at nearly 22Km. The effect of wavelet
adaptive denoised signal is also observed. The application
of the wavelet adaptive denoising is done in non-
decimated mode so as to preserve the useful content of the
signal through eliminating the convention down sampling.
However, this method is more redundant. The typical
Doppler spectrum of fig.1 after wavelet adaptive denoising
is illustrated in fig.8 and good improvement of the SNR is
keenly observed with the wavelet denoising. The fig. 9 &
10depict the SNR (dB) plot (Range intensity plot before &
after wavelet adaptive denoising respectively. From fig 10
in comparison with fig 9, it is viewed that the proposed
wavelet adaptive denoising method improves the signal
strength, particularly in the range beyond 12 Km. Thus,
the wind profile using doppler spectrum (refer fig 8)
becomes simple.
The proposed Hilbert transform method, the wavelet
based denoising technique is made a comparisonwith the
other techniques developed [12,13], are herewith
presented in the fig.11. It is observed that the wavelet
Fig. 7.a: MST radar signal in time domain using wavelet adaptive denoising (received at nearly 6 Km).
Fig. 7.b: MST radar signal in time domain using wavelet adaptive
denoising (received at nearly 22 Km).
Fig. 8: Typical Doppler Spectrum of MST radar using wavelet adaptive
denoising (Doppler Frequency in Hz Vs Height in Km).
Fig.9: Range Time intensity plot of MST radar.
Fig.10: Range Time intensity SNR plot of MST radar after proposed
wavelet adaptive denoising.
2010 International Conference on Signal and Image Processing 247
based methods are exhibiting superiority over other
techniques. However, the proposed wavelet based adaptive
denoising performance demonstrates less performance
than the simple wavelet based denoising due to the fact
that the proposed method is developed based on adaptive
method in which the denoising is done based on the range
bin signal received and the amount of noise associated
with the signal present.
VII. WIND VELOCITIES ESTIMATION
Estimating the various parameters form the radar return
is the objective of the present work, such that to satisfy the
fundamental requirement of MAP discussed earlier
(section 1 of the present work). The absolute wind velocity
( ) calculated using the radial velocity components of
different beam positions. The computation/estimation of
velocity components viz., requires at least three
non-coplanar beams of the radar data are required. Let
an orthogonal axis defined by , and which are
aligned to east, north and zenith directions.
The , and are the resolved wind velocities
in , and directions.The wind velocities , and are
computed as explained.
Consider a given set of beams ; (at
present NARL uses for computation of wind
vectors) and , , are the directional cosines
for the beam. The is the radial component of the
wind velocity measured for the beam using equation 5.
Then, the directional velocity components are estimated
using the following relation:
where,
,
, ,
, ,
Thus, the wind vector components are computed and
finds useful for further atmospheric investigations like
study of periodicity of waves, modeling & development of
the standard Indian Atmosphere model, etc.
VIII. RESULTS& APPLICATIONS
All the results shown are extracted from the
experimental data obtained on MST radar instrument
system, NARL, ISRO in the subsequent years 2005, 06,
07.The wind velocities and directional components are
computed using the equations (5) & (11) are computed.
The results are compared with the ADP software at
NARL, Gadanki, India. The results obtained using
proposed algorithm wavelet based adaptive denoising is
also validated with the GPS experiment, SHAR data and
the results are shown in fig. 12 [12], which illustrates GPS
data Wavelet adaptive denoised Data (ALG). The
results obtained are concurrent with GPS data with a
deviation of ±2%.
The computation of wind velocity components and thus
the wind profile obtained from the MST radar, NARL,
Gadanki is used during launching of the space vehicles at
Sateesh Dhawan Space Centre, ISRO, Sriharikota, A.P.,
India. This information is critical during take-off the space
vehicles because the stress on the launch vehicle due to
wind gusts have greater influence below 15Km. Therefore
the dynamics of the wind information is essentially needed
for the launching program to steer the launch of space-
crafts and to assess mechanical stress during ascent.
ACKNOWLEDGMENT
The authors sincerely express their deep sense of
gratitude to Mr. B.L.Prakash, Associate System Engineer,
IBM, Bangalore, India Ltd for his support in building the
robust and testing algorithm, the package developed for
the current work. The authors also wish to acknowledge to
Mr. P.Srinivasulu, Scientist-F, NARL, ISRO, India who
helped the authors for making use of the instrument
facility to validate the results.
REFERENCES
[1] G Viswanathan, MST Radar System An Overview, Second Winter School on Indian MST Radar, pp.1-21, Feb 1995, UGC-SVU
Centre for MST Radar Application, S V University, Tirupati.
[2] M.I Skolnik, Boston, Radar Hand Book, McGraw-Hill, 1990, pp. 1.16.
[3] Lora G. Win 1994 IEEE Signal Processing Magazine, pp. 13-31.
[4] Radar Observations of Winds and
Turbulence in the Stratosphere and Mesosphere J.Atmos.Sci., , pp 493-505, 1985.
[5] Kusuma G Rao and S C Chakravarty,MST RADAR Theory,
Techniques and Applications, ISRO, 2000.
Fig.12: Typical Wind Profile validation / comparison of the proposed
wavelet adaptive denoised algorithm (ALG) Vs GPS experiment data.
Fig.11: Typical Wind Profile validation / comparison of the proposed
wavelet adaptive denoised algorithm (ALG) Vs GPS experiment data.
248 2010 International Conference on Signal and Image Processing
[6] Vijay K.Madisetti and Douglas B.Williams, Digital Signal
Processing Hand Book,Chapter 6, CRC Press, 1999. [7] C Sidney Burrus, R A Gopinath, and HaitaoGuo, Introduction to
Wavelets and Wavelet Transforms: A Primer, Prentice Hall, NJ,
1998. [8] StephaneMallat, A Wavelet Tour of Signal Processing, 2/ed,
Academic Press, 1999.
[9] Donoho, D.L.; I.M. Johnstone, Ideal de-noising in an orthonormal basis, C.R.A.S. Paris, t. 319, Ser. I, pp. 1317-1322.
[10] Donoho, D.L. (1995), "De-noising by soft-thresholding," IEEE
Trans. on Inf. Theory, 41, 3, pp. 613-627. [11]
Atmospheric Letters, pp 1672-1688
[12] S.Varada Rajani, et.al.Radar Signals using wavelets and Harmonic decomposition
techniques, Atmos. Sci. Letters, Royal Meteorological Society, UK,
pp 1221-1230[13]
Techniques
Development Cell, JNTU, pp 58, Anatapur, 2010.
2010 International Conference on Signal and Image Processing 249