Sei sulla pagina 1di 5

Journal of Systems Engineering and Electronics Vol. 20, No. 3, 2009, pp.

499503

Parametric estimation of ultra wideband radar targets


Fan Ping & Jing Zhanrong
School of Electronics and Information, Northwestern Polytechnical Univ., Xian 710072, P. R. China (Received March 11, 2008)

Abstract: Based on the analysis of impulse response properties, a scattering model of ultra wideband (UWB)
radar targets is developed to estimate the target parameters exactly. With this model, two algorithms of multiple signal classication (MUSIC), and matrix pencil (MP), are introduced to calculate the scattering center parameters of targets and their performances are compared. The simulation experiments show that there are no dierences in the estimation precision of MUSIC and MP methods when the signal-to-noise ratio (SNR) is larger than 13 dB. However, the MP method has a better performance than that of MUSIC method when the SNR is smaller than 13 dB. Besides, the time consuming of MP method is less than that of MUSIC method. Therefore, the MP algorithm is preferred for the parametric estimation of UWB radar targets.

Keywords: ultra wideband radar, parametric estimation, scattering center, modeling.

1. Introduction
In the last few years, ultra wideband (UWB) radar has been widely used in various elds due to its large relative bandwidth and the unique capability for the identication of complex targets. When a target is illuminated by UWB radar, the received signal is not the simple delayed incident signal but a stack of scattering center echoes. Thus, UWB radar targets can be well approximated as a resultant of a series of discrete scattering centers. Many current studies are focused on modeling the scattering characteristics of UWB radar targets[13]. Moreover, various algorithms for parametric estimation of the scattering centers[12] , such as the MUSIC (multiple signal classication) and MP (matrix pencil) methods, are developed because the target scattering centers can describe the target scattering properties. However, there are few reports on the comparison of their performances. Therefore, it is necessary to compare and achieve a preferred method in order to estimate the parameters of target scattering centers exactly. In the present study, a scattering model of UWB radar targets is built to describe its scattering behav-

ior. Then the MUSIC and MP algorithms are employed to calculate the scattering center parameters of the complex target. In addition, the simulation experiments are used to compare the performance of these two methods.

2. Scattering model of UWB radar target


Usually, a complex target with a nite number of scattering centers can be represented as a linear, time-invariant, causal system with a unique impulse response[4]. In principle, the impulse response, h(t), of a scattering target with M point centers for ideal radar reectors can be descried as
M

h(t) =
m=1

(t tm )

(1)

where (t) is the Dirac delta function and tm is the relative round trip propagation time from radar to mth scattering center and back to radar. However, this simple assumption may not be applicable in practice, especially for the targets coated with radar absorbing materials. Due to the losses and electromagnetic led interactions on the surface of targets, each scattering center can be regarded as a dispersive channel where

* This project was supported by the Aerospace Science and Technology Foundation of China (2007CH080004).

500 the incident impulse suers a reduction in amplitude as well as an increase in the nominal duration. Based on the analysis of Gaussian pulse properties and the dispersive channel model, therefore, a more realistic impulse response, h(t), can be represented as
M M

Fan Ping & Jing Zhanrong In practice, the received signal y (t) is sampled with a uniform sampling interval t to obtain a set of discrete samples {yn }. By Eq. (5), the samples can be expressed in terms of poles and their amplitudes, which is
M

h(t) = u(t)
m=1

(t tm ) = u(t)
m=1

2 Am exp(a2 m m )

yn = y (nt) =
m=1

m exp(m nt)

(6)

(2) where u(t) is the unit step function to assure causality. (t tm ) is the Gaussian pulse of the mth scattering center with amplitude of Am , and m = t tm is the relative round trip propagation time delay of the mth scattering center. The parameter am = 2 /Tm is the eective bandwidth, and Tm is the eective duration of (t tm ). Previous experimentation[4] indicates that the time variation of Gaussian pulse, (t), can be employed to represent the incident signals emitted by UWB impulse radar, which is (t) = A exp[a2 (t ti )2 ], a = 2 /T (3)

Dening zm = exp(m t), the simplication of Eq. (6) is


M

yn =
m=1

n m zm

n = 0, 1, 2, ..., N

(7)

where N is the number of samples. It indicates that the scattering model of UWB radar targets with a number of scattering centers can be regarded as a linear system. Meanwhile, the scattering echo yn can be used to calculate the complex poles and their corresponding amplitudes, which represent a unique signature of the target and provide useful information for the target identication. Therefore, the scattering model of UWB radar targets given in Eq. (7) is employed to extract the parameters of scattering centers in the present study.

where a is the eective bandwidth, T is the eective duration, and A is the peak amplitude. According to above analyses, the response of a complex target with the impulse response h(t) given in Eq. (2) to the Gaussian pulse in Eq. (3) is y (t) = (t) h(t) = AAm erfc[(a2 ti + a2 m m )/m ] exp[m (t)] 2 m m=1 (4) where the asterisk * denotes convolution, erfc[] is the complementary error function, m = a2 + a2 m , and 2 m = [aam (ti m )/m ] .
M

3. Extraction methods of target scattering center


Based on the proposed model, the MUSIC and MP algorithms are used to calculate the scattering center parameter of {zm }M m=1 . Then their corresponding amplitude {m }M can be achieved by substituting m=1 M {zm }m=1 into Eq. (7) with the least squares tting. Consequently, the M scattering centers with large amplitudes are regarded as the parametric estimation of UWB radar targets. 3.1 MUSIC algorithm

When the parameter of eective bandwidth is in the order of a 1 G (rad/s), the UWB Gaussian pulse approximates an impulse. So the received waveform y (t) of UWB radar targets in Eq. (4) can be replaced by
M

y (t) =
m=1

m exp(m t)

(5)

where m and m are the set of complex poles and corresponding amplitudes, respectively.

The multiple signal classication MUSIC algorithm usually calculates the MUSIC spectral with estimating the noise subspace from the received signal. This process can be accomplished by eigenvalue decomposition of the estimated covariance matrix. If the Gauss noise n with zero mean and identical variance 2 is added, Eq. (7) can be rewritten in vector notation y = Za + n (8)

Parametric estimation of ultra wideband radar targets where y =[y1 , z 1 2 z1 Z= . . .


N z1

501 where v1i is the elements of the rst row of V V H . The roots of this polynomial are {zm }M m=1 . 3.2 MP algorithm

y2 , . . . , yN ]T z2
2 z2 . . . N z2

... ...

zm
2 zm . . .

N . . . zM

MP algorithm is another method to estimate the signal poles by exploiting a matrix pencil pair. Firstly, the matrix pencil {Y0 , Y1 } are dened as Y = [y 0 L1 , yL2 , ..., y0 ] Y1 = [yL , yL1 , ..., y1 ]

a = [1 , 2 , . . . , M ]T and n = [n1 , n2 ,. . . , nN ]T The covariance matrix of the received signal is Ryy = E [yy H ] + 2 I , I is the unit matrix and H denotes the complex conjugate transpose. Because the exact solution of Ryy is dicult to achieve, an estima yy , can be performed tion of the covariance matrix, R by the spatial smoothing preprocessing (SSP) or the modied spatial smoothing preprocessing (MSSP). Since MSSP has a better performance than SSP[5] , MSSP is used in the present study. Therefore, an estimated covariance matrix is yy = 1 R 2L
L

(12)

where the pencil parameter L [M , N M ] and in this study L = N /3 [7] , Y0 and Y1 are (N L) L matrices, respectively. Besides the matrix element vector, yk , is given by yk = [y (k ), y (k + 1), ..., y (N L + k 1)]T (13)

Then, the matrices of Y0 and Y1 are decomposed as Y = Z CZ 0 L R Y1 = ZL CZZR where ZL = 1 z1 . . . 1 z2 . . . . . . 1 zM . . . (14)

k J ) (Rk + J R
k=1

(9)

where L is the number of subarray Rk T 0 ... 1 . . . H . . Rk = yk yk , J = . . . . 1 ... 0

nn

k denotes the complex conjugate of Rk , n = N/3[6] is R subarray dimension, and yk = [yk , yk+1 , . . . , yk+n1 ]T . With the eigenvalue decomposition, the above co yy given in Eq. (9) can be expressed variance matrix R as yy = V V H R (10) where V =[v1 , v2 , . . . , vN ]T , and =diag[1 , 2 , . . . , N ]T . The standard MUSIC algorithm has many variations such as the root-MUSIC and the fast rootMUSIC. Due to the pole paring problem of the rootMUSIC, the fast root-MUSIC is used in the present study. The signal poles can be calculated by nding roots of the following polynomial
M +1

N L 1 N L 1 N L 1 z2 zM z1 L 1 L 2 z1 z1 1 L 1 L 2 z2 1 z2 ZR = . . . . . . . . . L 1 L 2 zM 1 zM

C =diag{c1, c2 , . . . , cM }, and Z =diag{z1, z2 , . . . , zM }. Based on Eq. (14), the following relation can be obtained, Y1 Y0 = ZL C (Z I )ZR = 0 (15)

C (z ) =
i=1

v1i z i1

(11)

which indicates that {zm }M m=1 correspond to the generalized singular values associated with the matrix Y1 Y0 .

502 The minimum description length (MDL) is used to estimate the values of parameter M . In order to reduce the noise eect on the result, the singular value decompositions of Y0 and Y1 are adopted, which are S0 0 [V0 , V0 ] Y0 = [U0 , U0 ] 0 S0 (16) S 0 1 [V1 , V1 ] Y1 = [U1 , U1 ] 0 S
1

Fan Ping & Jing Zhanrong Consequently, the signal poles of UWB radar targets 0 . 1 + Y can be achieved by Y

4. Simulation experiments
In order to compare the performance of MUSIC and MP algorithms with the proposed scattering model of UWB radar targets, computer simulation experiments are employed in the present study. The scattering model of an UWB radar target is
6

where S0 , S1 are the diagonal matrices consisted of M main singular values, U0 , U1 , V0 , V1 are the left and right singular matrices consisted of M main singular values respectively. Correspondingly, S0 , S1 , U0 , U1 , V0 , V1 are the left and right singular matrices consisted of L M main singular values, respectively. 0 = U0 S0 V0 and Y1 1 = By dening Y0 = Y = Y U1 S1 V1 , a transformation of Eq. (15) is 0 = 0 1 Y Y1 Y0 =Y
Table 1 SNR 5 dB method MUSIC MP 15 dB MUSIC MP 25 dB MUSIC MP p 1,2 0.019 1 0.198 5i 0.019 0 0.198 6i 0.019 9 0.199 2i 0.019 6 0.199 5i 0.020 0 0.199 7i 0.019 9 0.199 8i

y (n) =
m=1

am epm n ,

n = 0, 1, ...,

(18)

where p1,2 = 0.02 0.2i, p3,4 = 0.03 0.3i, p5,6 = 0.04 0.4i, am =1, m=1, 2, . . . , 6. Firstly, the performance of above two methods is studied when the SNR values of additive Gaussian noises are 5 dB, 15 dB and 25 dB, respectively. The calculated results of the signal poles with MUSIC and MP are given in Table 1.

(17)

Estimated scattering centers p 3,4 0.042 3 0.299 1i 0.037 7 0.300 2i 0.031 3 0.300 2i 0.031 4 0.300 5i 0.030 4 0.300 1i 0.030 4 0.300 2i p 5,6 0.046 5 0.397 5i 0.042 4 0.398 4i 0.038 4 0.399 8i 0.039 1 0.399 7i 0.039 4 0.400 0i 0.039 6 0.399 9i

Then, an error coecient Cerror is introduced to investigate the inuence of noises clearly, which is used to describe the approximation level between the original signal y (n) and the reconstruction signal y (n). Cerror = 10 lg MSE
N 1

of the MP. Therefore, it indicates that the MP algorithm is robust in the SNR range considered in this study.

(19)

where MSE =
n=0

|y (n) y (n)|2 is the square error.

The range of SNR from 1 dB to 27 dB is sampled at 2 dB steps, and the calculated corresponding Cerror is shown in Fig. 1. It can be seen that only when the additive noise is higher than 13 dB, the MUSIC and MP algorithms show similar results and both of them are reliable. At low SNR, however, the performance of MUSIC is not reliable compared with that

Fig. 1 Eect of additive noises (SNR) on the error coecient Cerror

Parametric estimation of ultra wideband radar targets In addition, because the target identication of the UWB radar is often used in military, it has a high requirement for real time. Table 2 presents the time
Table 2 Scattering center number Time consuming/s MUSIC MP

503 consuming of calculation when dierent numbers of target scattering centers are extracted with these two methods.

Time consuming of the MUSIC and MP methods 2 0.041 1 0.040 9 6 0.043 8 0.043 2 10 0.047 1 0.046 3 14 0.048 9 0.047 8

It is well known that the time consuming is related to the method complication and the computational conguration. The results shown in Table 2 are obtained from Windows XP, Matlab 7, P4, 2.4 GHz, EMS memory 2 G. It can be seen that with the increase in the scattering center number, the time consuming becomes large. However, the time consuming of these two methods are less than one second, and the maximum is only 0.048 9 s. They satisfy the real time requirement. Besides, the time consuming of MP is less than MUSIC.

References
[1] Wang Y, Chen J W, Liu Z. A fuzzy classier for radar target recognition using one-dimensional scattering centers. Journal of Electronics & Information Technology, 2005, 27(5):784788. [2] Naishadham K, Piou J E. A super-resolution method for extraction of modal responses in wideband data. IEEE Trans. on Antennas Propag., 2004: 41684171. [3] Liu D H, Zhang Y S, Chen Z J, et al. Extraction of UWB radar target feature based on ESPRIT matrix pencil algorithm. Systems Engineering and Electronics, 2006, 28(11):

5. Conclusions
According to the scattering properties of UWB radar targets, a target scattering model is developed, which is used to estimate the scattering center parameters of complex targets. Then, the MUSIC and MP algorithms are applied to extracting the scattering centers with this model. Through the simulation experiment, the results show that the similar parameter estimations of complex targets with these two algorithms are obtained in high SNR environment. When the SNR is smaller than 13 dB, the MP has a better performance than the MUSIC. Moreover, the time consuming of MP is less than that of MUSIC. Therefore, it is concluded that based on the proposed scattering model of UWB radar targets, the MP algorithm is more suitable for the scattering center extraction of UWB radar targets.

16551658. [4] Hussain M G M. Principle of high resolution radar based on nonsinusoidal wavesPart III: radar-target reectivity model. IEEE Trans. on Electromagnetic Compat., 1990, 32(5): 144152. [5] Kim K T, Seo D K, Kim H T. Ecient radar target recognition using the MUSIC algorithm and invariant features. IEEE Trans. on Antennas Propag. 2002, 50(3):325337. [6] Keller J B. Geometric theory of diraction. Journal of the Optical Society of America, 1962, 52(2): 116130. [7] Schimidt R O. Multiple emitter location and signal parameter estimation. IEEE Trans. on Antennas Propag., 1986, 34(4): 806811.

Fan Ping was born in 1979. She is a doctoral candidate in School of Electronics and Information, Northwestern Polytechnical Universiey. Her research areas are radar signal processing and target identication. E-mail: fanping 2007@gmail.com

Potrebbero piacerti anche