- Email: [email protected]
Adaptive filtering for the separation of incoherent scatter and meteor signals for Arecibo observation data
ARTICLE IN PRESS
Journal of Atmospheric and Solar-Terrestrial Physics 67 (2005) 1190–1195 www.elsevier.com/locate/jastp
Adaptive filtering for the separation of incoherent scatter and meteor signals for Arecibo observation data C.-H. Wen, J.F. Doherty, J.D. Mathews Communications and Space Sciences Laboratory, The Pennsylvania State University, University Park, PA 16802-2707, USA Available online 10 August 2005
Abstract The Arecibo 430 MHz incoherent scatter radar (ISR) has been used to observe the ionosphere and meteors over 40 years. The meteor signals have traditionally been treated as the interference in ionosphere observation data, while more recently the incoherent scatter signals have been described as background noise in the meteor observation data. In this paper we present signal processing techniques that separate these two different signals into different geophysically interesting data sets. We investigate two techniques for two different data sets. For meteor observation data an adaptive filter is used to locate and separate meteor returns from the total data so that we improve meteor detection rate. For ionosphere observation data we use a filterbank followed by the short time Fourier transform (STFT) analysis to remove the meteor signals from the incoherent scatter results. By doing this the incoherent scatter data will not be contaminated and we can also analyze meteor characteristics from the separated signal. r 2005 Elsevier Ltd. All rights reserved. Keywords: Radar signal processing; Meteor detection; Interference removal; Incoherent scatter radar
1. Introduction The incoherent scatter radar (ISR) located at Arecibo has been used to observe the vertical ionospheric electron concentration profiles for many years. Earlier studies of E region ion layers at Arecibo include Rowe (1974), Miller and Smith (1975), Mathews and Bekeny (1979), and recent studies include Mathews et al. (1993, 1997a, b), Mathews (2003), Sulzer (2004). The Arecibo Observatory (AO) meteor observations grew from the ISR observations of the ionosphere. It was first introduced by Zhou et al. (1995). Meteor observation techniques and properties of the meteors were reported by Mathews et al. (1997a, b, 2003), Mathews (2003), Zhou et al. (1998), and Janches et al. (2000). Corresponding author. Tel.: +814 865 7226; fax: +814 863 8457. E-mail address: [email protected] (C.-H. Wen).
Currently the ISR ionosphere observations are made using 13-baud Barker or 88-baud pseudo-random coded pulses while separate meteor observations utilize uncoded 45 ms pulses transmitted every millisecond (Mathews et al., 2003). The meteor returns seen in the so-called ISR power-profile results are often spread over twice the code length as the meteor return voltages are incorrectly decoded due to significant Doppler offsets. The rangespread meteor return then contaminates the ISR powerprofile effectively found by squaring and adding—in practice, all processing is done in the transform domain via FFTs. Here we separate ISR and meteor returns using Doppler filters in a manner that preserves maximal information in both signal paths. We describe the design of specific filters to separate the signals based on the inherent differences between the incoherent scatter and meteor signals to separate the signals. For the ISR ionosphere observation data, we use a filterbank followed by the short time Fourier transform (STFT)
1364-6826/$ - see front matter r 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.jastp.2005.06.004
ARTICLE IN PRESS C.-H. Wen et al. / Journal of Atmospheric and Solar-Terrestrial Physics 67 (2005) 1190–1195
analysis to remove the meteor signals. Chau and Woodman (2003) introduced similar technique for meteor head-echo observations using 13 baud Barker code. We also analyze the separated meteor signals thus providing useful information for the meteor head-echo research. For the meteor observation data, we use an adaptive filter to remove the low Doppler velocity incoherent scatter signals, leaving meteor signals for further analysis via techniques similar to those described in Mathews et al. (2003). We introduce the filterbank technique for the ISR observation data in Section 2 and the adaptive filter technique for the meteor observation data in Section 3. Conclusions are given in Section 4.
2. Filterbank technique for the ISR ionosphere observation data
rdecoded ½n ¼
M X
rcoded ½n þ k 1c½k;
n ¼ 1; 2; . . . ; N.
k¼1
(1) Notice that the decoded result can be found in transform domain as is done in practice. Because of the Doppler speed of the meteor, the decoding process effectively uses the wrong code thus spreading the meteor energy over a sample twice the interval code length, which deteriorates the result. Fig. 1(a) shows the power profiles of the decoded signals recorded at 22:20:22.990, September 3rd, 2001. A 13 baud Barker code was used. There is one meteor event in this sequence. Fig. 1(b) shows the power profile of the raw undecoded signal of that meteor event; and Fig. 1(c) shows the power profile of the decoding result using (1). In Fig. 1(c) the meteor energies spread in range because it is not decoded correctly, i.e., the decoding assumed zero Doppler offset. To properly decode meteor signals we need to correct the code for the Doppler offset that results in a complex code. The modified decoding equation after multiplication by the appropriate phasor is rmeteor ½n ¼
M X k¼1
rcoded ½n þ k 1c½kejoD kT ,
where oD is the Doppler frequency of the meteor, and T is the sampling period. Fig. 1(d) shows the decoding result using the correct Doppler frequency shifted Barker code. Note that the decoding result for incoherent scatter signals is now erroneous. The figure shows that we can detect the presence of the meteor with proper threshold and precisely identify the location of the meteor. After identifying the meteor, we use the STFT to analyze the corresponding meteor signals. We use the data shown in Fig. 1(b) to illustrate the process. For this data set a 13-baud Barker coded pulse was transmitted. With this prior knowledge and the meteor location identified by the Doppler frequency shifted Barker decoder, we get the Barker coded meteor signals. Let the coded meteor signals be rcoded ½L, rcoded ½Lþ 1; . . . ; rcoded ½L þ 12, where L is the range gate location of the meteor. Each meteor signal sample is modeled by rcoded ½L 1 þ n ¼ AejðoD nTþfÞ c½n;
The ISR ionosphere observations are made using Barker coded carrier pulses with the inter-pulse period (IPP) of 10 ms. The received signals are mixed to baseband (DC center frequency) and then sampled in inphase and quadrature-phase channels. To separate the incoherent scatter and the meteor signals, we first use a filterbank to detect the meteor then use STFT analysis to remove the meteor signals. Let the received signals of one IPP be rcoded ½n, n ¼ 1; 2; . . . ; N, where N is the number of samples and let the code be ccode ¼ ½c½1; c½2; . . . ; c½M, where M is the code length. The decoded result, a voltage, is written as
(2)
1191
n ¼ 1; 2; . . . ; 13, (3)
where A is the amplitude, oD is the Doppler frequency of the meteor signals, T is the sampling period, f is the phase, and cBarker;13 ¼ ½c½1; c½2; . . . ; c½13 ¼ ½1; 1; 1; 1; 1; 1; 1; 1; 1; 1; 1; 1; 1 is the 13 Barker code. The STFT of the meteor signals is MðkÞ ¼ FFT ½rcoded ðLÞ cð1Þ; rcoded ðL þ 1Þ ð4Þ cð2Þ; . . . ; rcoded ðL þ 12Þ cð13Þ . Here we use fast Fourier transform (FFT) to calculate the frequency spectrum of the meteor signals. Note that we can zero pad the signals to get more spectrum samples. Fig. 2 shows the magnitude of the STFT. We can clearly see the meteor energy in the frequency spectrum. To remove the meteor signals we use the spectrum to estimate the frequency spectrum of meteor signals (sinc function) and then subtract it from the original spectrum. Fig. 3(a) shows the power profile of the undecoded IPP with meteor removal. Fig. 3(b) shows the power profile of the decoded result, the process removes 90% of the meteor signal energy. Fig. 3(c) shows RTI plot of the Fig. 1 results with the meteor return removed. Fig. 4 shows the flowchart of the meteor signal detector and removal. This figure summarizes the process we just discussed. The input is one IPP signal at a time—when running a meteor observation program we analyze four or more IPPs at a time (Mathews et al., 2003; Wen et al., 2004). Since we do not know the Doppler frequency of the meteor signals a priori, we construct a filterbank. Each filter of the filterbank is a Barker decoder with different Doppler frequency shift. The filter that produces the maximum peak gives the best estimation of the meteor Doppler frequency. We can increase the resolution of Doppler frequency estimation by adding more filters. A threshold detector
ARTICLE IN PRESS C.-H. Wen et al. / Journal of Atmospheric and Solar-Terrestrial Physics 67 (2005) 1190–1195
1192
x 104
6
600
Meteor
550 5
500
4
400 Power
Altitude (km)
450 350 300 250
3 2
200
Meteor
150
1
100 1
2
(a)
4 5 6 Time (sec)
7
8
9
0
10
0
3.5
6
3 Power
4
7
5 4
2.5 2
3
1.5
2
1
1
0.5
0
500
Meteor
4.5
8
200 300 400 Sample Number
x 106
5
Meteor
9
100
(b)
x 105
10
Power
3
0
100
(c)
200 300 400 Sample Number
0
500
0 (d)
100
200 300 400 Sample Number
500
Fig. 1.
follows the filterbank. When at least one output of the filters exceeds the threshold we declare a meteor detection. We then remove the meteor signals by STFT analysis, e.g. Fig. 2.
2500 Meteor
Magnitude
2000
1500
3. Adaptive filter technique for the meteor observation data
1000
500
0
0
100
200 Fig. 2.
300
400
500
Meteor observations are made by using 45-ms uncoded carrier pulse with an inter-pulse period of 1 ms (Mathews et al., 2003). The return signal is demodulated in in-phase and quadrature-phase channels and sampled at a 1 ms1 rate. To separate the incoherent scatter signals we apply a least-mean-square (LMS) adaptive filter, Haykin (2002),
ARTICLE IN PRESS C.-H. Wen et al. / Journal of Atmospheric and Solar-Terrestrial Physics 67 (2005) 1190–1195
3
x 104
1193
x 105
7 6
2.5
5 Power
Power
2 1.5 1
(a)
3 2
0.5 0
4
1 0 0
100
200 300 400 Sample Number
500
0
100
200 300 400 Sample Number
(b)
500
600 550 500 Altitude (km)
450 400 350 300 250 200 150 100 1
2
(c)
3
4 5 6 Time (sec)
7
8
9
10
Fig. 3.
Filter 1
I N
STFT Filter 2 YES
Threshold
P
Meteor Signals
Detector
U T NO
Analysis &
O
Meteor
U
Removal
T P U T
Filter N Fig. 4.
ARTICLE IN PRESS C.-H. Wen et al. / Journal of Atmospheric and Solar-Terrestrial Physics 67 (2005) 1190–1195
1194
scatter signals, which is usually the case since the meteor signal is rare, the transversal filter is a low pass filter (the frequency response is adaptively adjusted according to the input signal) which separates the incoherent scatter signal (x[n]) and the white noise(y[n]). In practice we first use the data with incoherent scatter and white noise signals to train this system. After it converges, we then apply the meteor observation data as the input to the system. The meteor signals will appear in y[n] (without incoherent scatter signals) as the result. We use a meteor event recorded at 7:38:42.750 AST 24 February 2001 to demonstrate the incoherent scatter signals removal process. Fig. 6(a) shows the image of the real part signals of the meteor event. There are 160 IPPs in this image; each IPP has 250 complex signal samples. We can see a weak meteor event at IPP #85–#95. Fig. 6(b) shows the real part of IPP #88 signals. Fig. 6(c) and (d) shows the estimated incoherent scatter signals (low Doppler frequency) of the adaptive filter and the
to the received radar signals. Fig. 5 shows the block diagram of the adaptive filter; r½n, x½n, and y½n in the diagram are the received, estimated, and desired signals, respectively. This system is an adaptive line enhancer (ALE) which is used to detect a sinusoidal signal buried in a noise background. The input and output of the transversal filter are the delayed version of input signal and the estimated input signal, respectively. When the input signals only consist of white noise and incoherent
r[n]
Transversal Filter
Delay
Incoherent Scatter Signals
x[n]
LMS
y[n] +
Fig. 5.
250
20 Meteor
Meteor
15
200
150
Voltage
Sample Number
10
100
5 0 -5 -10
50
-15 20
40
(a)
60 80 100 IPP Number
120
140
-20
160
0
50
0
50
(b)
100 150 Sample Number
200
250
200
250
15
15 10
10
5
5 Voltage
Voltage
Meteor
0 -5
-5
-10
-10 -15
-15 0
(c)
0
50
100
150
Sample Number
200
250
(d) Fig. 6.
100
150
Sample Number
ARTICLE IN PRESS C.-H. Wen et al. / Journal of Atmospheric and Solar-Terrestrial Physics 67 (2005) 1190–1195
non-incoherent-scatter part of the signals (y[n] in Fig. 5). The figures clearly show that the adaptive filter removes the incoherent scatter signals and retains the meteor signals almost unchanged.
4. Conclusions In this paper we present signal processing techniques to separate the meteor signals from the incoherent scatter signals of the Arecibo observation data. We use a filterbank followed by the STFT to remove the meteor signals of the incoherent scatter observation data thus rendering the ionosphere results more accurate. We can also analyze the separated meteor signals to get parameters, such as Doppler frequency (speed) and altitude, after the removal, thus providing two geophysically interesting data streams. One thing needs to be noticed, since each IPP for ionosphere observation data is 10 ms, most of the meteor events can be found in only one or two IPPs. The meteor deceleration cannot be determined for most events. For the meteor observation data we apply a LMS adaptive filter to remove the incoherent scatter signals. According to Mathews et al. (2003) the incoherent scatter deteriorates the detection of the meteor signals. Using the adaptive filter technique, we improve the detection probability of the meteor.
Acknowledgments This effort was supported under NSF Grants ATM0113454 and AST-0205848 to The Pennsylvania State University. The Arecibo Observatory is part of the National Astronomy and Ionosphere Center which is operated by Cornell University under cooperative agreement with the National Science Foundation. As always we thank the Arecibo staff for their efforts.
References Chau, J.L., Woodman, R.F., 2003. Observations of meteorhead echoes using the Jicamarca 50 MHz radar in inter-
1195
ferometer mode. Atmospheric Chemistry and Physics 4, 511–521. Haykin, S., 2002. Adaptive Filter Theory, fourth ed. Prentice Hall, Upper Saddle River, New Jersey. Janches, D., Mathews, J.D., Meisel, D.D., Getman, V.S., Zhou, Q.-H., 2000. Doppler studies of near-antapex UHF radar micrometeors. Icarus 143, 347–353. Mathews, J.D., 2003. Radio science issues surrounding HF/ VHF/UHF radar meteor studies. Journal of Atmospheric and Solar Terrestrial Physics 66, 285–299. Mathews, J.D., Bekeny, F.S., 1979. Upper atmosphere tides and the vertical motion of ionospheric sporadic layers at Arecibo. Journal of Geophysical Research 84, 2743–2750. Mathews, J.D., Morton, Y.T., Zhou, Q., 1993. Observations of ion layer motions during the AIDA campaign. Journal of Atmospheric and Terrestrial Physics 55, 447–457. Mathews, J.D., Sulzer, M.P., Perillat, P., 1997a. Aspects of layer electrodynamics inferred from high-resolution ISR observations of the 80–270 km ionosphere. Geophysical Research Letters 24, 1411–1414. Mathews, J.D., Meisel, D.D., Hunter, K.P., Getman, V.S., Zhou, Q., 1997b. Very high resolutioin studies of micrometeors using the Arecibo 430 MHz radar. Icarus 126, 157–169. Mathews, J.D., Doherty, J.F., Wen, C.-H., Briczinski, S.J., Janches, D., Meisel, D.D., 2003. An update of UHF radar meteor observations and associated signal processing techniques at Arecibo Observatory. Journal of Atmospheric and Solar Terrestrial Physics 65, 1139–1149. Miller, K.L., Smith, L.G., 1975. Horizontal structure of midlatitude sporadic E layers observed by incoherent scatter radar. Radio Science 10, 271–276. Rowe, J.F., 1974. Downward transport of nighttime Es-layers into the E-region ar Arecibo. Journal of Atmospheric and Solar Terrestrial Physics 36, 225–235. Sulzer, M.P., 2004. Meteor Science from Regular Incoherent Scatter Radar Ionospheric Observations at Arecibo. Atmospheric Chemistry and Physics Discussions 4, 805–831. Wen, C.-H., Doherty, J.F., Mathews, J.D., 2004. Timefrequency radar processing for meteor detection, geoscience and remote sensing. IEEE Transactions 42 (3), 501–510. Zhou, Q., Tepley, C.A., Sulzer, M.P., 1995. Meteor Observations by the Arecibo 430 MHz ncoherent scatter radar. I. Results from time integrated observations. Journal of Atmospheric and Terrestrial Physics 57, 421–431. Zhou, Q., Perillat, P., Cho, J.Y.N., Mathews, J.D., 1998. Simultaneous meteor echo observations by large aperture VHF and UHF radars. Radio Science 33, 1641–1654.
Journal of Atmospheric and Solar-Terrestrial Physics 67 (2005) 1190–1195 www.elsevier.com/locate/jastp
Adaptive filtering for the separation of incoherent scatter and meteor signals for Arecibo observation data C.-H. Wen, J.F. Doherty, J.D. Mathews Communications and Space Sciences Laboratory, The Pennsylvania State University, University Park, PA 16802-2707, USA Available online 10 August 2005
Abstract The Arecibo 430 MHz incoherent scatter radar (ISR) has been used to observe the ionosphere and meteors over 40 years. The meteor signals have traditionally been treated as the interference in ionosphere observation data, while more recently the incoherent scatter signals have been described as background noise in the meteor observation data. In this paper we present signal processing techniques that separate these two different signals into different geophysically interesting data sets. We investigate two techniques for two different data sets. For meteor observation data an adaptive filter is used to locate and separate meteor returns from the total data so that we improve meteor detection rate. For ionosphere observation data we use a filterbank followed by the short time Fourier transform (STFT) analysis to remove the meteor signals from the incoherent scatter results. By doing this the incoherent scatter data will not be contaminated and we can also analyze meteor characteristics from the separated signal. r 2005 Elsevier Ltd. All rights reserved. Keywords: Radar signal processing; Meteor detection; Interference removal; Incoherent scatter radar
1. Introduction The incoherent scatter radar (ISR) located at Arecibo has been used to observe the vertical ionospheric electron concentration profiles for many years. Earlier studies of E region ion layers at Arecibo include Rowe (1974), Miller and Smith (1975), Mathews and Bekeny (1979), and recent studies include Mathews et al. (1993, 1997a, b), Mathews (2003), Sulzer (2004). The Arecibo Observatory (AO) meteor observations grew from the ISR observations of the ionosphere. It was first introduced by Zhou et al. (1995). Meteor observation techniques and properties of the meteors were reported by Mathews et al. (1997a, b, 2003), Mathews (2003), Zhou et al. (1998), and Janches et al. (2000). Corresponding author. Tel.: +814 865 7226; fax: +814 863 8457. E-mail address: [email protected] (C.-H. Wen).
Currently the ISR ionosphere observations are made using 13-baud Barker or 88-baud pseudo-random coded pulses while separate meteor observations utilize uncoded 45 ms pulses transmitted every millisecond (Mathews et al., 2003). The meteor returns seen in the so-called ISR power-profile results are often spread over twice the code length as the meteor return voltages are incorrectly decoded due to significant Doppler offsets. The rangespread meteor return then contaminates the ISR powerprofile effectively found by squaring and adding—in practice, all processing is done in the transform domain via FFTs. Here we separate ISR and meteor returns using Doppler filters in a manner that preserves maximal information in both signal paths. We describe the design of specific filters to separate the signals based on the inherent differences between the incoherent scatter and meteor signals to separate the signals. For the ISR ionosphere observation data, we use a filterbank followed by the short time Fourier transform (STFT)
1364-6826/$ - see front matter r 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.jastp.2005.06.004
ARTICLE IN PRESS C.-H. Wen et al. / Journal of Atmospheric and Solar-Terrestrial Physics 67 (2005) 1190–1195
analysis to remove the meteor signals. Chau and Woodman (2003) introduced similar technique for meteor head-echo observations using 13 baud Barker code. We also analyze the separated meteor signals thus providing useful information for the meteor head-echo research. For the meteor observation data, we use an adaptive filter to remove the low Doppler velocity incoherent scatter signals, leaving meteor signals for further analysis via techniques similar to those described in Mathews et al. (2003). We introduce the filterbank technique for the ISR observation data in Section 2 and the adaptive filter technique for the meteor observation data in Section 3. Conclusions are given in Section 4.
2. Filterbank technique for the ISR ionosphere observation data
rdecoded ½n ¼
M X
rcoded ½n þ k 1c½k;
n ¼ 1; 2; . . . ; N.
k¼1
(1) Notice that the decoded result can be found in transform domain as is done in practice. Because of the Doppler speed of the meteor, the decoding process effectively uses the wrong code thus spreading the meteor energy over a sample twice the interval code length, which deteriorates the result. Fig. 1(a) shows the power profiles of the decoded signals recorded at 22:20:22.990, September 3rd, 2001. A 13 baud Barker code was used. There is one meteor event in this sequence. Fig. 1(b) shows the power profile of the raw undecoded signal of that meteor event; and Fig. 1(c) shows the power profile of the decoding result using (1). In Fig. 1(c) the meteor energies spread in range because it is not decoded correctly, i.e., the decoding assumed zero Doppler offset. To properly decode meteor signals we need to correct the code for the Doppler offset that results in a complex code. The modified decoding equation after multiplication by the appropriate phasor is rmeteor ½n ¼
M X k¼1
rcoded ½n þ k 1c½kejoD kT ,
where oD is the Doppler frequency of the meteor, and T is the sampling period. Fig. 1(d) shows the decoding result using the correct Doppler frequency shifted Barker code. Note that the decoding result for incoherent scatter signals is now erroneous. The figure shows that we can detect the presence of the meteor with proper threshold and precisely identify the location of the meteor. After identifying the meteor, we use the STFT to analyze the corresponding meteor signals. We use the data shown in Fig. 1(b) to illustrate the process. For this data set a 13-baud Barker coded pulse was transmitted. With this prior knowledge and the meteor location identified by the Doppler frequency shifted Barker decoder, we get the Barker coded meteor signals. Let the coded meteor signals be rcoded ½L, rcoded ½Lþ 1; . . . ; rcoded ½L þ 12, where L is the range gate location of the meteor. Each meteor signal sample is modeled by rcoded ½L 1 þ n ¼ AejðoD nTþfÞ c½n;
The ISR ionosphere observations are made using Barker coded carrier pulses with the inter-pulse period (IPP) of 10 ms. The received signals are mixed to baseband (DC center frequency) and then sampled in inphase and quadrature-phase channels. To separate the incoherent scatter and the meteor signals, we first use a filterbank to detect the meteor then use STFT analysis to remove the meteor signals. Let the received signals of one IPP be rcoded ½n, n ¼ 1; 2; . . . ; N, where N is the number of samples and let the code be ccode ¼ ½c½1; c½2; . . . ; c½M, where M is the code length. The decoded result, a voltage, is written as
(2)
1191
n ¼ 1; 2; . . . ; 13, (3)
where A is the amplitude, oD is the Doppler frequency of the meteor signals, T is the sampling period, f is the phase, and cBarker;13 ¼ ½c½1; c½2; . . . ; c½13 ¼ ½1; 1; 1; 1; 1; 1; 1; 1; 1; 1; 1; 1; 1 is the 13 Barker code. The STFT of the meteor signals is MðkÞ ¼ FFT ½rcoded ðLÞ cð1Þ; rcoded ðL þ 1Þ ð4Þ cð2Þ; . . . ; rcoded ðL þ 12Þ cð13Þ . Here we use fast Fourier transform (FFT) to calculate the frequency spectrum of the meteor signals. Note that we can zero pad the signals to get more spectrum samples. Fig. 2 shows the magnitude of the STFT. We can clearly see the meteor energy in the frequency spectrum. To remove the meteor signals we use the spectrum to estimate the frequency spectrum of meteor signals (sinc function) and then subtract it from the original spectrum. Fig. 3(a) shows the power profile of the undecoded IPP with meteor removal. Fig. 3(b) shows the power profile of the decoded result, the process removes 90% of the meteor signal energy. Fig. 3(c) shows RTI plot of the Fig. 1 results with the meteor return removed. Fig. 4 shows the flowchart of the meteor signal detector and removal. This figure summarizes the process we just discussed. The input is one IPP signal at a time—when running a meteor observation program we analyze four or more IPPs at a time (Mathews et al., 2003; Wen et al., 2004). Since we do not know the Doppler frequency of the meteor signals a priori, we construct a filterbank. Each filter of the filterbank is a Barker decoder with different Doppler frequency shift. The filter that produces the maximum peak gives the best estimation of the meteor Doppler frequency. We can increase the resolution of Doppler frequency estimation by adding more filters. A threshold detector
ARTICLE IN PRESS C.-H. Wen et al. / Journal of Atmospheric and Solar-Terrestrial Physics 67 (2005) 1190–1195
1192
x 104
6
600
Meteor
550 5
500
4
400 Power
Altitude (km)
450 350 300 250
3 2
200
Meteor
150
1
100 1
2
(a)
4 5 6 Time (sec)
7
8
9
0
10
0
3.5
6
3 Power
4
7
5 4
2.5 2
3
1.5
2
1
1
0.5
0
500
Meteor
4.5
8
200 300 400 Sample Number
x 106
5
Meteor
9
100
(b)
x 105
10
Power
3
0
100
(c)
200 300 400 Sample Number
0
500
0 (d)
100
200 300 400 Sample Number
500
Fig. 1.
follows the filterbank. When at least one output of the filters exceeds the threshold we declare a meteor detection. We then remove the meteor signals by STFT analysis, e.g. Fig. 2.
2500 Meteor
Magnitude
2000
1500
3. Adaptive filter technique for the meteor observation data
1000
500
0
0
100
200 Fig. 2.
300
400
500
Meteor observations are made by using 45-ms uncoded carrier pulse with an inter-pulse period of 1 ms (Mathews et al., 2003). The return signal is demodulated in in-phase and quadrature-phase channels and sampled at a 1 ms1 rate. To separate the incoherent scatter signals we apply a least-mean-square (LMS) adaptive filter, Haykin (2002),
ARTICLE IN PRESS C.-H. Wen et al. / Journal of Atmospheric and Solar-Terrestrial Physics 67 (2005) 1190–1195
3
x 104
1193
x 105
7 6
2.5
5 Power
Power
2 1.5 1
(a)
3 2
0.5 0
4
1 0 0
100
200 300 400 Sample Number
500
0
100
200 300 400 Sample Number
(b)
500
600 550 500 Altitude (km)
450 400 350 300 250 200 150 100 1
2
(c)
3
4 5 6 Time (sec)
7
8
9
10
Fig. 3.
Filter 1
I N
STFT Filter 2 YES
Threshold
P
Meteor Signals
Detector
U T NO
Analysis &
O
Meteor
U
Removal
T P U T
Filter N Fig. 4.
ARTICLE IN PRESS C.-H. Wen et al. / Journal of Atmospheric and Solar-Terrestrial Physics 67 (2005) 1190–1195
1194
scatter signals, which is usually the case since the meteor signal is rare, the transversal filter is a low pass filter (the frequency response is adaptively adjusted according to the input signal) which separates the incoherent scatter signal (x[n]) and the white noise(y[n]). In practice we first use the data with incoherent scatter and white noise signals to train this system. After it converges, we then apply the meteor observation data as the input to the system. The meteor signals will appear in y[n] (without incoherent scatter signals) as the result. We use a meteor event recorded at 7:38:42.750 AST 24 February 2001 to demonstrate the incoherent scatter signals removal process. Fig. 6(a) shows the image of the real part signals of the meteor event. There are 160 IPPs in this image; each IPP has 250 complex signal samples. We can see a weak meteor event at IPP #85–#95. Fig. 6(b) shows the real part of IPP #88 signals. Fig. 6(c) and (d) shows the estimated incoherent scatter signals (low Doppler frequency) of the adaptive filter and the
to the received radar signals. Fig. 5 shows the block diagram of the adaptive filter; r½n, x½n, and y½n in the diagram are the received, estimated, and desired signals, respectively. This system is an adaptive line enhancer (ALE) which is used to detect a sinusoidal signal buried in a noise background. The input and output of the transversal filter are the delayed version of input signal and the estimated input signal, respectively. When the input signals only consist of white noise and incoherent
r[n]
Transversal Filter
Delay
Incoherent Scatter Signals
x[n]
LMS
y[n] +
Fig. 5.
250
20 Meteor
Meteor
15
200
150
Voltage
Sample Number
10
100
5 0 -5 -10
50
-15 20
40
(a)
60 80 100 IPP Number
120
140
-20
160
0
50
0
50
(b)
100 150 Sample Number
200
250
200
250
15
15 10
10
5
5 Voltage
Voltage
Meteor
0 -5
-5
-10
-10 -15
-15 0
(c)
0
50
100
150
Sample Number
200
250
(d) Fig. 6.
100
150
Sample Number
ARTICLE IN PRESS C.-H. Wen et al. / Journal of Atmospheric and Solar-Terrestrial Physics 67 (2005) 1190–1195
non-incoherent-scatter part of the signals (y[n] in Fig. 5). The figures clearly show that the adaptive filter removes the incoherent scatter signals and retains the meteor signals almost unchanged.
4. Conclusions In this paper we present signal processing techniques to separate the meteor signals from the incoherent scatter signals of the Arecibo observation data. We use a filterbank followed by the STFT to remove the meteor signals of the incoherent scatter observation data thus rendering the ionosphere results more accurate. We can also analyze the separated meteor signals to get parameters, such as Doppler frequency (speed) and altitude, after the removal, thus providing two geophysically interesting data streams. One thing needs to be noticed, since each IPP for ionosphere observation data is 10 ms, most of the meteor events can be found in only one or two IPPs. The meteor deceleration cannot be determined for most events. For the meteor observation data we apply a LMS adaptive filter to remove the incoherent scatter signals. According to Mathews et al. (2003) the incoherent scatter deteriorates the detection of the meteor signals. Using the adaptive filter technique, we improve the detection probability of the meteor.
Acknowledgments This effort was supported under NSF Grants ATM0113454 and AST-0205848 to The Pennsylvania State University. The Arecibo Observatory is part of the National Astronomy and Ionosphere Center which is operated by Cornell University under cooperative agreement with the National Science Foundation. As always we thank the Arecibo staff for their efforts.
References Chau, J.L., Woodman, R.F., 2003. Observations of meteorhead echoes using the Jicamarca 50 MHz radar in inter-
1195
ferometer mode. Atmospheric Chemistry and Physics 4, 511–521. Haykin, S., 2002. Adaptive Filter Theory, fourth ed. Prentice Hall, Upper Saddle River, New Jersey. Janches, D., Mathews, J.D., Meisel, D.D., Getman, V.S., Zhou, Q.-H., 2000. Doppler studies of near-antapex UHF radar micrometeors. Icarus 143, 347–353. Mathews, J.D., 2003. Radio science issues surrounding HF/ VHF/UHF radar meteor studies. Journal of Atmospheric and Solar Terrestrial Physics 66, 285–299. Mathews, J.D., Bekeny, F.S., 1979. Upper atmosphere tides and the vertical motion of ionospheric sporadic layers at Arecibo. Journal of Geophysical Research 84, 2743–2750. Mathews, J.D., Morton, Y.T., Zhou, Q., 1993. Observations of ion layer motions during the AIDA campaign. Journal of Atmospheric and Terrestrial Physics 55, 447–457. Mathews, J.D., Sulzer, M.P., Perillat, P., 1997a. Aspects of layer electrodynamics inferred from high-resolution ISR observations of the 80–270 km ionosphere. Geophysical Research Letters 24, 1411–1414. Mathews, J.D., Meisel, D.D., Hunter, K.P., Getman, V.S., Zhou, Q., 1997b. Very high resolutioin studies of micrometeors using the Arecibo 430 MHz radar. Icarus 126, 157–169. Mathews, J.D., Doherty, J.F., Wen, C.-H., Briczinski, S.J., Janches, D., Meisel, D.D., 2003. An update of UHF radar meteor observations and associated signal processing techniques at Arecibo Observatory. Journal of Atmospheric and Solar Terrestrial Physics 65, 1139–1149. Miller, K.L., Smith, L.G., 1975. Horizontal structure of midlatitude sporadic E layers observed by incoherent scatter radar. Radio Science 10, 271–276. Rowe, J.F., 1974. Downward transport of nighttime Es-layers into the E-region ar Arecibo. Journal of Atmospheric and Solar Terrestrial Physics 36, 225–235. Sulzer, M.P., 2004. Meteor Science from Regular Incoherent Scatter Radar Ionospheric Observations at Arecibo. Atmospheric Chemistry and Physics Discussions 4, 805–831. Wen, C.-H., Doherty, J.F., Mathews, J.D., 2004. Timefrequency radar processing for meteor detection, geoscience and remote sensing. IEEE Transactions 42 (3), 501–510. Zhou, Q., Tepley, C.A., Sulzer, M.P., 1995. Meteor Observations by the Arecibo 430 MHz ncoherent scatter radar. I. Results from time integrated observations. Journal of Atmospheric and Terrestrial Physics 57, 421–431. Zhou, Q., Perillat, P., Cho, J.Y.N., Mathews, J.D., 1998. Simultaneous meteor echo observations by large aperture VHF and UHF radars. Radio Science 33, 1641–1654.