P. 1
00757221 Discrete Fractional Fourier Transform Based on Orthogonal

00757221 Discrete Fractional Fourier Transform Based on Orthogonal

|Views: 26|Likes:
Publicado porOm Prakash Suthar

More info:

Published by: Om Prakash Suthar on Sep 14, 2011
Direitos Autorais:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

09/14/2011

pdf

text

original

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 47, NO.

5, MAY 1999

1335

Discrete Fractional Fourier Transform Based on Orthogonal Projections
Soo-Chang Pei, Senior Member, IEEE, Min-Hung Yeh, and Chien-Cheng Tseng, Member, IEEE
Abstract—The continuous fractional Fourier transform (FRFT) performs a spectrum rotation of signal in the time–frequency plane, and it becomes an important tool for time-varying signal analysis. A discrete fractional Fourier transform has been recently developed by Santhanam and McClellan, but its results do not match those of the corresponding continuous fractional Fourier transforms. In this paper, we propose a new discrete fractional Fourier transform (DFRFT). The new DFRFT has DFT Hermite eigenvectors and retains the eigenvalue-eigenfunction relation as a continous FRFT. To obtain DFT Hermite eigenvectors, two orthogonal projection methods are introduced. Thus, the new DFRFT will provide similar transform and rotational properties as those of continuous fractional Fourier transforms. Moreover, the relationship between FRFT and the proposed DFRFT has been established in the same way as the conventional DFT-tocontinuous-Fourier transform. Index Terms— Discrete Fourier transform, discrete fractional Fourier transform, Fourier transform, fractional Fourier transform.

I. INTRODUCTION HE FOURIER transform (FT) is one of the most frequently used tools in signal analysis [1]. A generalization of the Fourier transform—the fractional Fourier transform (FRFT)—has been proposed in [2] and [3] and has become a powerful tool for time-varying signal analysis. In time-varying signal analysis, it is customary to use the time-frequency plane, with two orthogonal time and frequency axes [4]. Because the successive two forward Fourier transform operations will result in the reflected version of the original signal, the FT can in the be interpreted as a rotation of signal by the angle time–frequency plane and represented as an orthogonal signal representation for sinusoidal signal. The FRFT performs a rotation of signal in the continuous time–frequency plane to any angle and serves as an orthonormal signal representation for the chirp signal. The fractional Fourier transform is also called rotational Fourier transform or angular Fourier transform in some documents. Besides being a generalization of the FT, the
Manuscript received December 2, 1996; revised February 19, 1998. This work was supported by the National Science Council, R.O.C., under Contract NSC 85-2213-E002-025. The associate editor coordinating the review of this paper and approving it for publication was Dr. Akram Aldroubi. S.-C. Pei is with the Department of Electrical Engineering, National Taiwan University, Taipei, Taiwan, R.O.C. (e-mail: pei@cc.ee.ntu.edu.tw). M.-H. Yeh is with the Department of Computer Information Science, Tamsui Oxford University College, Tamsui, Taipei, Taiwan, R.O.C. C.-C. Tseng was with the Department of Electronics Engineering, Hwa Hsia College of Technology and Commerce, Taipei, Taiwan, R.O.C. He is now with the Department of Computer and Communication Engineering, National Kaohsiung First University of Science and Technology, Taipei, Taiwan, R.O.C. Publisher Item Identifier S 1053-587X(99)03244-4.

T

FRFT has been proved to relate to other time-varying signal analysis tools, such as Wigner distribution [4], short-time Fourier transform [4], wavelet transform, and so on. The applications of the FRFT include solving differential equations [2], quantum mechanics [3], optical signal processing [5], timevariant filtering and multiplexing [5]–[8], swept-frequency filters [9], pattern recognition [10], and time–frequency signal analysis [11]–[13]. Several properties of the FRFT in signal analysis have been summarized in [9]. Many methods for realizing the FRFT have been developed, but most of them are to utilize the optical implementation [14], [15] or numerical integration. Because the FRFT is a potentially useful tool for signal processing, the direct computation of the fractional Fourier transform in digital computers has become an important issue. The ideal discrete fractional Fourier transform (DFRFT) will be a generalization of the discrete Fourier transform (DFT) that obeys the rotation rules as the continuous FRFT and provides similar results as the FRFT. In [16], a method for a numerical integration FRFT has been proposed, but the method does not obey the rotation rules, and the signal cannot be recovered from its inverse transform. In [17], Santhanam and McClellan have developed a discrete FRFT, but their method does not provide the same transforms to match those of the continuous case. In this paper, we present a new discrete fractional Fourier transform (DFRFT). This DFRFT is a generalization of the DFT and will provide similar transforms as those of the continuous case. The relationship between the DFRFT and the FRFT can also be established and discussed in detail. Moreover, the proposed DFRFT has important unitary and rotation properties. This paper is organized as follows. In Section II, the previous development of continuous and discrete fractional Fourier transforms are reviewed. The concept for developing the DFRFT to have similar results as the continuous corresponding case are described in Section III. Two acceptable solutions for the DFRFT are considered and proposed in Section III. Then, the relationships between the FRFT and the DFRFT can be established in Section IV. Finally, conclusions and discussions are made in Section V. II. PRELIMINARY A. Continuous FRFT The Fourier transform of a signal can be interpreted as a angle rotation of the signal in the time–frequency plane. The FRFT is then developed and treated as a rotation of signal

1053–587X/99$10.00 © 1999 IEEE

Because where the normalized Hermite function with unitary variance. VOL. Imaginary part: dashed line. Thus. However. Fractional Fourier transform of a rectangular window. [9] csc if if if is not a multiple of (1) is a multiple of is a multiple of (2) where indicates the rotation angle of transformed signal for is the th-order normalized Hermite function the FRFT. and x(t) = 0 otherwise. The transform kernel of continuous FRFT is defined as [2]. In order to avoid ambiguity. The discrete results shown in Fig. . and a circular reflected version of its DFT . and the imaginary parts of the FRFT or DFRFT are indicated by dashed lines. The conventional methods to compute the DFRFT in [17] and [20] are mainly on numerical computing the fractional power of the DFT kernel matrix. the FRFT of the signal (4) (5) Equation (5) indicates that the FRFT can be interpreted as a weighting summation of Hermite functions. The fractional power of the DFT kernel matrix and the rotation angles in the DFRFT essentially mean the same thing. [3]. In [19]. the Hermite functions in [19] with a different scaling to those of the FRFT. . 2 shows the results of this DFRFT produced by a discrete rectangular window. Equation where is the linear (8) indicates that the DFRFT of the signal combination of the four major angular parts: the original signal . is the eigenfunction of the FRFT. it will have a different phase term in the weighting coefficients. Furthermore. the Greek subscripts (for example. and . the middle is not the intermediate state of transform result for and . to any angles in the time–frequency plane [9]. The Old DFRFT The kernel matrix of the DFRFT is obtained by computing the fractional power for the DFT kernel matrix. MAY 1999 Fig. a circular reflected version of the signal ( ). the two results . with unit variance. 47. Fig. Fig. 2 are quite different from the results of Fig. otherwise here is defined as [ ]. (2) provides an eigendecomposition representation of the FRFT kernel. The equivalence of (1) and (2) has been proved in [3]. The weighting coefficients are obtained from multiplying the phase term and the inner product of the input signal and the corresponding Hermite function. The fractional power of the DFT kernel matrix [20] is calculated by (6) (6) where (7) Applying the above defined kernel to the signal is computed as DFRFT of the signal .1336 IEEE TRANSACTIONS ON SIGNAL PROCESSING. 1. [6]. elsewhere] for various angles. Using the kernel of the by angle is computed as FRFT. NO. the (8) is the conventional DFT of signal . 1. Real part: solid line. and the rotation output of the signal is also the weighting summation of Hermite functions. its DFT . The th-order normalized Hermite function with variance is defined as (3) is the th-order Hermite polynomial [18]. B. and ) are used to denote the rotation angles in the time–frequency plane. 5. x(t) = 1 when 02  t  2. and the English superscripts are used to denote the fractional power values of the DFT kernel matrices in this paper. 1 shows the FRFT of the rectangular window function [ for . The real parts of the FRFT or DFRFT in this paper are plotted by solid lines. The rectangular window used . a rotation operation using Hermite functions has also been proposed.

[22]. x(n) = 1 when 06  n  6. Proof: See [21]. . . Proof: See [20]. If the Hermite function are sampled in this way. Proposition 3: DFT Hermite eigenvectors should have the . . . and matrix is defined as OF DFRFT The continuous FRFT has a Hermite function with unitary variance as its eigenfunction. 2. . but they correspond to different eigenvalues. .. a novel matrix is introduced to compute the real and complete set of DFT eigenvectors very elegantly. The corresponding eigenfunction property for the DFT would be like (11) is the eigenvector of DFT corresponding to the where th-order discrete Hermite function. In the discussion in the previous subsection. we get (12) is the th-order Hermite polynomial. are .PEI et al. . Eigendecomposition of DFRFT The development of our DFRFT is based on the eigendecomposition of the DFT kernel. (13) will become (14) . —respectively. The eigenvalue multiplicities of DFT matrix indicate the ranks of the DFT eigensubspaces [21]. . and —and each is major eigensubspaces— . . and corresponding to one of the four eigenvalues—1. The eigenvectors of the DFT matrix constitute four . and then. Can the DFT have the eigenvectors with the similar shapes as the Hermite functions? These DFT eigenvectors are known as DFT Hermite eigenvectors in this paper. DFT Hermite Eigenvectors III. it satisfies the commutative property (10) will also be the eigenvectors The eigenvectors of matrix of the DFT kernel matrix .: DISCRETE FRACTIONAL FOURIER TRANSFORM BASED ON ORTHOGONAL PROJECTIONS 1337 Fig. 1. (9) The Fourier transform of (13) can be computed as (15) . DEVELOPMENT A. The parameter is the index for the DFT eigenvalue . The output is quite different TABLE I MULTIPLICITIES OF THE EIGENVALUES OF DFT KERNEL MATRIX where matrix . and many properties of the DFT matrix eigenvalues and eigenvectors have been discussed in [20] and [21]. . In [20]. 1. but it cannot obtain the real-valued DFT eigenvectors. where is the spread variance of the Proof: It is assumed that DFT eigenvectors. we only summarize some results for our further development of DFRFT. . from the continuous FRFT in Fig. Proposition 1: The eigenvalues of and its multiplicities are shown in Table I. a method for computing the DFT eigenvectors has been introduced. Proposition 2: A matrix can be used to compute the real eigenvectors of the DFT matrix . is defined. The continuous approximate form can be written as (13) Sampling by . we have known that the eigendecomposition of the DFT kernel matrix is not unique. . where associated continuous spread variance is the sampling intervals of signal. B. In [22]. This In Table I. Matrix commutes with the DFT kernel . . . Here. a multiplicity function function is used to denote the DFT eigenvalue multiplicity for . Old DFRFT of a rectangular window function. and x (n ) = 0 otherwise. .

we have the approximation expression (19) is wide This approximation is valid because is for large . Combining (19) and (20) very small when . . we can find that the DFT Hermite eigenvectors should have associate continuous spread variance . Thus. NO. we have . the is. Therefore. VOL. we have (20) (24) for sufficiently large . we only prove the case that odd. and the other is the numerical error in (20). when order becomes large. The variable in (15) can be replaced by equal to to have its discrete form. However.1338 IEEE TRANSACTIONS ON SIGNAL PROCESSING. the larger order is. both errors approach zero. approximated by can be From Proposition 3. MAY 1999 The sampling theorem [23] tells us the continuous scope of . Next. Thus. Proof: Here. For By truncating the integral interval of Fourier transform from to . is obtained by samProposition 4: If the sequence pling the unit variance Hermite function with the sampling . the slower decay rate the Hermite function will have. by replacing the continuous integral with the numerical integral. (16) adjusts the scaling of Hermite funcThe variance value to make the DFT tion. 5. because larger is . and our goal is to adjust the variance have the same eigenvector shape as itself. defined in the range Proposition 5: If the sequence is obtained by shifting Hermite Gauss samples in the following way: If is even for (22) for and if is od. The sampled Hermite functions still can be used to construct the DFRFT kernel because they have the similar shapes and good approximations to the continuous Hermite functions. the decay rate the degree of Hermite polynomial is proportional to for of the Hermite function sufficiently large . for (23) for then it can be shown that the DFT of the . Thus. then it can be shown that interval If is even (17) If is odd (18) for sufficiently large . is even. This implies that the truncation error in (19) is larger for high-order . the better approximation (17) is. but this will not influence the development of the DFRFT. The th-order continuous Hermite function should have zeros [18]. When the value of approaches infinity. The sequence obtained from (12) is also equal to the sampling of the Hermite function with the unit variance by . it is clear that there are two approximation errors in (17). i. frequency spectrum that DFT can cover is the continuous spectrum spacing (or resolution) of the DFT is . we obtain and letting This expression is valid for any both sides. the sampling interval we will prove that even such samplings of Hermite functions can only have approximate Hermite eigenvectors of the DFT. the proof can be easily derived in the same way. In Propositions 4 and 5. The small aliasing will occur while is closer to . One is the truncation error in (19). Next. 47. the approximation in (17) becomes worse. and the decay rate of Gauss function very fast. From the above proof. these functions are not bandlimited. The condition of for the two shapes of (14) and (16) being the same becomes This approximation is also valid because is is very large.. taking at (21) Equation (14) becomes The proof is completed. and the samplings in Proposition 3 cannot guarantee that the numbers of the sign changes in this sampled th-order Hermite function are also .e. Thus. .

we should not expect that a finite vector can express the oscillation behavior of the very high-order Hermite function very well. the choices for the DFT eigenvectors to construct the DFRFT kernel are multiple and not unique. Combining . the terms of the Hermite functions are summed up from order zero to infinity. only DFT Hermite eigenvectors can be added. It is because the approximation error of the low DFT Hermite eigenvectors are small. we can compute in its DFT eigensubspace to obtain a the projections of Hermite-like DFT eigenvector (32) mod . (27) can be The fractional power of matrix can be calculated from its eigendecomposition and the powers of eigenvalues. the DFT is computed from the eigenvectors of Hermite eigenvector matrix in the same DFT eigensubspace. the eigenvectors can be used as bases for individual DFT eigensubspaces. In the case of continuous FRFT. The idea for developing our DFRFT is to find the discrete form for (2). we obtain DFT for for (30) The proof is completed. In order to retain the eigenfunction property in (11). Using the . • Ambiguity in Deciding the Fractional Powers of Eigenvalues: We know that the square root of unity are 1 and 1 from elementary mathematics. Unfortunately. the Because matrix can have complete real orthogonal DFT eigenvectors. we obtain C. The selections of the DFT Hermite eigenvectors are from low to high orders. the proof can also be easily derived. we know that there are several DFRFT kernel matrices that can obey the rotation properties. The While is given by DFT of the sequence DFT TABLE II EIGENVALUES ASSIGNMENT RULE OF DFRFT KERNEL MATRIX (25) Using the equality second term on the right side of (25) becomes . This assignment rule matches the multiplicities of the eigenvalues of the DFT kernel matrix in Table I. (26) Substituting (26) into (25) and using Proposition 4. In Propositions 4 and 5. Table II shows the eigenvalues assignment rules for the DFRFT. • Ambiguity in Deciding the Eigenvectors of the DFT Kernel Matrix: The DFT eigenvectors constitute four major eigensubspaces. it has been proved that the samplings of Hermite functions can have approximate DFT eigenvectors. The normalized vectors for the samplings of Hermite functions are defined as (31) . In addition. is odd.PEI et al. eigenvectors for the However. therefore. Newly Developed DFRFT DFT (27) (28) is limited in the range where equality rewritten as DFT (29) is limited in the range where (28) and (29). Because of the above ambiguity. In (32).: DISCRETE FRACTIONAL FOURIER TRANSFORM BASED ON ORTHOGONAL PROJECTIONS 1339 Proof: Here. This indicates that there exists root ambiguity in deciding the fractional power of eigenvalues. In addition. for the discrete case. . the unit variance Hermite functions are in the following sampled with a period of discussions. there exist two types of ambiguity in deciding the fractional power of the DFT kernel matrix. will be a DFT Hermite eigenvector. and is the eigenvector of matrix where . we also prove the case that is even.

the convergence for the eigenvectors obtained from matrix are not so fast for the high-order Hermite functions by the LMS error criterion. It must be mentioned that the last eigenvalue and should be assigned to the is not assigned to to match the eigenvalue multiplicities value of the DFT matrix in Table I.1340 IEEE TRANSACTIONS ON SIGNAL PROCESSING. . It is easy to verify that the angle rotation property of the DFRFT can be preserved only while the DFT eigenvectors are orthogonal. Although the Mathieu functions can converge to a Hermite function [28]. the eigenvalue–eigenfunction relation is retained in defining the DFRFT. In [25]. The symbol notation in developing the two algorithms are as follows: continuous Hermite samples vector. vector orthogonalization is required for the DFT Hermite eigenvectors obtained from (32). 5. whereas is odd.. Two methods for vector orthogonalization are proposed in the following of this paper. nonorthogonalized Hermite eigenvector. It is easy to show that the eigenvectors located in different eigensubspaces will be orthogonal. and for even . . . is the normalized eigenvector corresponding to the th-order Hermite function. VOL. . This is because the eigenvectors located in different eigensubspaces will be always orthogonal. for and the eigenvalues of the DFRFT are . . However. the eigenvectors obtained from matrix are directly considered to be the discrete Hermite functions. Therefore. is the bottom of the page. Unfortunately. The Gram–Schmidt Algorithm (GSA): The Gram–Schmidt method [29] is a well-known orthogonalization approach for vectors. the DFT Hermite eigenvectors obtained from (32) cannot constitute an orthogonal basis for DFT eigenspace. orthogonalized Hermite eigenvector. MAY 1999 The transform kernel of the DFRFT can be defined as (33) for (odd) (34) for (even) where . shown at . Calculate the continuous samples of Hermite functions: Compute the eigenvectors of matrix : Using equation (32) to compute Hermite eigenvectors by projections: for to for to end end . An eigenbased method for computing the DFRFT to have similar continuous results has been proposed by us in [24]–[27]. So the DFT Hermite eigenvectors can be orthogonalized for every eigensubspace individually to obtain orthogonal eigenvectors in the whole eigenspace of DFT. In addition. the transform kernel for the rotation angle can be computed as in (34a). . Equation (32) provides a method for finding DFT Hermite eigenvectors. The role of matrix in (32) is just as a tool to find a complete set of real and orthogonal DFT eigenvectors.. The DFT Hermite eigenvectors in each DFT eigensubspace can be orthogonalized by the Gram–Schmidt method. The DFT eigenvectors are orthogonalized for each eigensubspace. . 47. Therefore. and for is even. NO. Example 1— : The rotation angle is equal to . (34a) . where the eigenvector corresponding to the th continuous Hermite function. the eigenvectors obtained from the matrix are just discrete Mathieu functions [21]. . . Such a method is called the method in this paper. is defined as follows: where For odd . . This means that is replaced by in (34).

) are plotted Transform results of the impulse signal ( in Figs. Because the . for minimize minimize where matrix. GSA. 3. The corresponding samples of the continuous FRFT for the impulse signal are plotted in Fig. In Fig. it can be to found that the transform results change from a chirp signal ) to an impulse-like function ( ). The is equal to 4/13. otherwise. 3. In Fig.PEI et al. We can formulate our problem as the OPA to find the least Frobenius norm between the samples of Hermite functions and orthogonal DFT Hermite eigenvectors. 9 and 10. Frobenius norm of (35) samples of Hermite functions are plotted for . . 8–10. Orthogonal Procrustes Algorithm (OPA): A traditional mathematical problem known as the orthogonal Procrustes Algorithm can be used to find the least Frobenius norm [30. The error vectors between samples of Hermite functions and DFT Hermite eigenvectors are defined as (37) Example 2: In this example. D. 2. 582] for the two given spaces. . 7. [ which is the same as the signal used in Fig. Figs. GSA. E. The norm of the error vectors between the DFRFT and samples of the FRFT are also shown in the titles of Figs. and the number of points sampling interval is equal to 73. and OPA. the solution will also .: DISCRETE FRACTIONAL FOURIER TRANSFORM BASED ON ORTHOGONAL PROJECTIONS 1341 Fig. 4–6 show the DFRFT of the rectangular window function for various angles for method. Implementation of the New DFRFT As in the case of DFT frequency domain. respectively. the last half of the indices in the DFRFT must also be treated as the negative end The GSA minimizes the errors between the samples of Hermite functions and orthogonal DFT Hermite eigenvectors from low to high orders. 11. The continuous FRFT is an orthonormal signal decomposition for chirp signals [9]. p. Therefore. satisfy Algorithm Calculate the continuous samples of Hermite functions: Compute the eigenvectors of : Using equation (32) to compute Hermite eigenvectors by projections: for to Compute the SVD of . where . ( the DFRFT can be used for the chirp signal and chirp rate detection. The chirp signal used here is equal . A more detailed theory and algorithm can be found in [7]. The minimizing calculating the singular value decomposition (SVD) of . Norms of error vectors between DFT Hermite eigenvectors and samples of Hermite functions. Based on the unitary property in the DFRFT and the transform results shown in Figs. 8–10 for the method. Properties of DFRFT The properties of the DFRFT are shown in Table III. and will can be found by be our solution. and OPA. ]. we can find that the proposed DFRFT provides a similar orthonormal signal decomposition for discrete chirplike signals. is the subject to (36) the . The sampled discrete data then becomes . the norms of error vectors between the computed DFT Hermite eigenvectors and . Example 3: The DFRFT by the GSA for a chirp signal is computed in this example. 1. respectively. and the OPA minimizes the total errors between the samples of Hermite functions and orthogonal DFT Hermite eigenvectors. . we use the DFRFT to deal with the rectangular window function shown in Fig.

VOL. 7. computing the DFRFT is . the computation of the DFRFT can be implemented only by a transform kernel matrix multiplication. x (n ) = 1 when 06  n  6. 5. ej 20:1141k . Then. The output is close to Fig. the following method for implementing the DFRFT can be applied: (38) (39) frequency. 4. NO. The complexity of . 5. and x (n ) = 0 otherwise. 47. DFRFT by S method of a rectangular window function. closer output to the continuous FRFT than Fig. k = 032 1 1 1 32. and x(n) = 0 otherwise. and it is the same as in the DFT case. This figure has a Fig. DFRFT by GSA of a chirp signal. x(n) = 1 when 06  n  6. and x (n ) = 0 otherwise. 4. an impulse-like output is obtained. If the rotational angles are adjusting. x (n ) = 1 when 06  n  6. This concept is also applied in the time domain . ( and the rotation angle are When the number of points determined. the continuous FRFT in Fig. When = 3=7. output to the continuous FRFT than Fig. This figure has a closer Fig. identity transform) and any angle transform domains. 4. and the eigenvalues of DFRFT are also determined. DFRFT by OPA method of a rectangular window function. 6. 1. MAY 1999 Fig.1342 IEEE TRANSACTIONS ON SIGNAL PROCESSING. DFRFT by GSA method of a rectangular window function. the DFT Hermite eigenvectors can be computed a priori.

Proof: Let be any DFT eigenvector located in the . and the value is the eigenvalue for this eigenvector . we can obtain is also a DFRFT eigenvector. If the rotation angle is changed. the DFRFT reduces to DFT. The coefficients ’s are in (33).: DISCRETE FRACTIONAL FOURIER TRANSFORM BASED ON ORTHOGONAL PROJECTIONS 1343 Fig. When = =2. x(0) = 1. DFRFT by OPA of an impulse function. Since any eigenvector in Therefore. Here. 6 Fig. It can be seen that chirp-like outputs are obtained while the angle = =2. This is the major reason for the mismatches in [17]. the inner products of signal and eigenvectors. When = =2. 10. the DFRFT reduces to DFT. and they can be computed in advance. the diagonal matrix F. and the output becomes a constant. and x(n) = 0 otherwise. Discussion The method in [17] obeys the rotation properties. and the output becomes a constant. Fig.PEI et al. DFRFT by S method of an impulse function. Proposition 6: The method in (6) assigns all the eigenvectors of the DFRFT matrix to only four different eigenvalues. and x(n) = 0 otherwise. the DFRFT reduces to DFT. . DFRFT by GSA of an impulse function. and x(n) = 0 otherwise. . we stress the major reason for this mismatch. 9. and the output becomes a constant. When = =2. only should be recomputed. It can be seen that chirp-like outputs are obtained while the angle = =2. x(0) = 1. but it cannot have similar results as in the continuous case. It can be seen that chirp-like outputs are obtained while the angle 6= =2. x(0) = 1. A rigorous discussion for the mismatches of [17] has been presented in [16]. Applying the transform eigensubspace kernel defined in (6) to the eigenvector . 8. 6 (40) The definitions of matrix and matrix are the same as those .

we will discuss the transform range and resolution for the DFRFT (see Table IV). (41) is the eigenvector corresponding to the th-order where Hermite function. ( one of the four eigensubspace has the same eigenvalue. 5. The DFRFT developed in this paper is based on the mimicking of the eigenvalue–eigenfunction relationship that the continuous counterpart of this transform has with the normalized unit variance Hermite functions. and more DFT Hermite eigenvectors are used to compute the DFRFT. Here. the periods in the transform results change for different rotation angles. we will review and understand the transform range and resolution for the is the conventional DFT. the signal and transform results in [31] are discrete and periodic. the proposed DFRFT has assigned a different eigenvalue for each eigenvector. there will be four different eigenvalues for all the eigenvectors. b) = F (b. X is odd. Samples of continuous FRFT of an impulse function. It is based on retaining the property of the DFT that a sampled periodic function transforms into a periodic function. X is even. x(0) = 1. The the overall frequency range that the DFT can represent is equal to . it has been proved that the spread variance . Transform Range and Resolution of DFRFT In this subsection. Then. TABLE III PROPERTIES OF DFRFT Unitary Angle additivity Time Inversion Periodicity Symmetry Eigenfunction Impulse Transform Parity F )3 = (F )01 = F0 F F =F F x(0n) = X (0n) F =F F (a. Moreover. sampling interval for the original continuous signal. the DFRFT can be used to give the similar continuous transform results within the accuracy of the discrete finite vector approximation. is the total recorded signal duration and is signal. However. the method assigns the DFRFT eigenvalues to only four values. and x(t) = 0 otherwise. In the DFRFT of [17]. which satisfies the eigenvalue and eigenfunction relationship as a continuous FRFT kernel. 47. the number of DFT Hermite eigenvectors is only finite. NO. MAY 1999 Fig. VOL. In the following discussion. we will of the DFT Hermite eigenvector is compute the FRFT for a Hermite function with any variance . and the frequency resolution is [23]. While errors of the DFT Hermite eigenvectors will be reduced. and transform range in the time domain as well. 11. the approximation eigenvectors. The development of the new DFRFT is based on the same idea as (2). b) is the (a. In Proposition 3. On the other hand. In the conventional DFT analysis. .1344 IEEE TRANSACTIONS ON SIGNAL PROCESSING.b)-element in DFRFT kernel matrix F [vn ] = e0j n vn where vn is Hermite-like function N F [(k)]  2 10j2cot ej cot  If x is even. and there are some approximation errors in the DFT Hermite approaches infinity. we will establish the relationship between the FRFT and the DFRFT. In [31]. If x is odd. Thus. It should be noted that the number of eigenfunctions for the continuous FRFT in (2) is infinite. IV. To begin with. A. the transform range and resolution of the DFT have been well discussed [23]. our DFRFT can be used for any rotation angle and can provide very similar results as continuous cases. it is also is the number of points of the discrete the time resolution. and the rotation angle in [31] is valid for a certain discrete set of rotation angles. RELATIONSHIP BETWEEN FRFT AND DFRFT In this section. a) where F (a. an alternative DFRFT has recently been proposed. Therefore.

frequency ( • The length of the time range is shorter than that of the ). 12. it has been shown that there are approximation errors in the DFT Hermite eigenvectors. Angle actual angle for the elliptical rotation. it can be found that the signal rotation of the DFRFT is an elliptical rotation in . it has been known that the . Proposition 8: The resolution of the DFRFT with angular is equal to parameter (44) is the sampling interval of the signal. is used to denote the resolution of the FRFT. the spread variance of the DFRFT for angle is equal to (47) The overall transform range the DFRFT can cover is equal to (50) In (34). In this subsection. The new variance is com- (43) Proof: This proposition can be easily proved by computing the FRFT of the normalized Hermite functions. 12. as drawn in Fig. therefore. but it actually performs elliptical rotation in the continuous time–frequency plane. the DFRFT performs independent circular rotation in the discrete notation. • The lengths of time and frequency ranges are equal ). Elliptical Rotation versus Circular Rotation The FRFT performs circular rotation in the continuous time–frequency plane. will be equal to 1 for any value of . the approximation errors of the DFT Hermite eigenvectors are ignored in the following discussion. ( • The length of the time range is longer than that of the ). It is easy to check some special cases for (43): and . The sampled spread variance of the DFRFT is vector in (12) can also be an approximate eigenvector in the fractional Fourier domain (angle ). (46) Thus. whereas This means that the DFRFT is implemented in a circular rotation for the discrete case. the resolution of the FRFT can be obtained as (49) (42) where puted as . the continuous time–frequency plane. For simplification of analysis. but the DFRFT performs elliptical rotation in the continuous time–frequency plane while .PEI et al. we can substitute the variance in (43) by to get the spread variance of the DFRFT in the fractional Fourier domain (angle ) (48) Both (47) and (48) indicate the corresponding variance of DFT Hermite eigenvectors in the fractional Fourier domain. only the last term can affect the envelope . In (42). From (45) and Fig. and is the where number of points for the discrete signal. B. If . The overall transform range of the DFRFT can cover is equal to (45) Proof: From Proposition 3. • Circular Rotation: (51) . The symbol and are the two special cases. frequency ( The rotation concept of the DFRFT in the continuous is the time–frequency plane is plotted in Fig.: DISCRETE FRACTIONAL FOURIER TRANSFORM BASED ON ORTHOGONAL PROJECTIONS 1345 TABLE IV TRANSFORM RANGE AND RESOLUTION OF DFRFT Proposition 7: The continuous FRFT of a Hermite function with variance is equal to Moreover. Three cases can be realized for the signal rotation of the DFRFT in the continuous time–frequency plane. 12. of the FRFT of From Propositions 4 and 5. we will define circular and elliptical rotations clearly and establish the relationship between circular and elliptical rotations.

Proposition 9: The circular rotation of signal can be implemented by an angle modification of the elliptical rotation and post-multiplying a phase compensation factor. and the post-phase compensa- is the transform kernel of elliptical rotation. NO. Three different cases of rotation concept for DFRFT in continuous time-frequency plane. and where is defined as in (43). The critical point for which we care is the (58) can be written as the weighted Proof: The signal sum of the normalized Hermite functions. 13. is the circular rotation of signal . the elliptical rotation will of signal become the circular rotation. (57) where tion factor is equal to . is the elliptical rotation . 47. Results of Example 4 showing that the error norms between DFRFT and continuous FRFT samples are greatly reduced by angle modification and post-phase compensation. • Elliptical Rotation: where (54) (55) (56) spread variance changes in these two schemes. The transform kernels of the circular and elliptical rotations defined above are infinite numbers of sums for Hermite eigenvectors. the circular rotation is just the FRFT of the signal.1346 IEEE TRANSACTIONS ON SIGNAL PROCESSING. 12. (59) . (a) (b) (c) Fig. 5. While . VOL. If the Hermite eigenvectors in the circular rotation are from order 0 to infinity. MAY 1999 Fig. (52) (53) is the transform kernel of circular rotation.

whereas the OPA minimizes the total errors between the samples of the Hermite functions and the orthogonal DFT Hermite eigenvectors. Using Proposition 9. 13(a) shows the DFRFT with angular parameter for the original signal. and Fig.PEI et al. such DFT Hermite eigenvectors cannot form an orthogonal basis for the DFT eigenspaces. the elliptical rotation can be implemented by circular rotation. (62) .: DISCRETE FRACTIONAL FOURIER TRANSFORM BASED ON ORTHOGONAL PROJECTIONS 1347 where . 13(b) is the DFRFT with angular parameter modification and postphase compensation. In Example 2. The rotation angle . The new transform retains the eigenvalue–eigenfunction relationship using the sampled version of the normalized unit-variance Hermite functions that the continuous FRFT has with the unit variance Hermite functions. The rotation will almost be of the frequency range ( a circular rotation. V. 13(b) match the corresponding continuous FRFT cases shown in Fig.0488 in Example sponding variance 2. Therefore. If variance of the continuous-time counterpart of the eigenvector will approximate to unity. In this case. 13(c) due to the elliptical rotation and not in the circular rotation in the continuous time–frequency plane. Compute the FRFT of (circular rotation) for the both sides of (59). the sampling . the modification of the angle and postphase compensation discussed above are very critical. and the effects of angle modification and postphase compensation will be very small. If the values of (57) are . we only compute the results of by the GSA. the the unitary property in the DFRFT. we directly compute the DFRFT without angle modification and postphase compensation. Fig. and the transform . The in (58) is replaced by in (62) for preserving variable . CONCLUSIONS The development of this DFRFT is based on the eigendecomposition of the DFT kernel matrix . Fig. 13(c) shows the sample values of the continuous FRFT . The results shown in Fig. Two vector orthogonalization processes for the DFT Hermite eigenvectors are accordingly proposed in this paper: One is GSA. 13(c) very well. the following steps must be used to make the results of discrete cases match those of the continuous case. The discrete post-phase compensation resolution is still factor for the DFRFT is the length of time range ( ) will almost equal the length ). The length of the time range is almost equal to the length of the frequency range. Step 3) Multiply the result obtained from Step 2 by the post-phase compensation factor shown in (62). the complete real and orthonormal eigenvectors of the DFT kernel matrix can be computed. The GSA minimizes the errors between the samples of the Hermite functions and the orthogonal DFT Hermite eigenvectors from low to high orders. 13(a) is quite different from Fig. Example 4: In this example. where . and the other is OPA. the DFRFT in Example 2 is almost a circular rotation. However. Here. therefore. However. The DFT Hermite eigenvectors can be calculated by the projection of samples of the unit variance Hermite functions in the DFT eigensubspaces through the help of the eigenvectors of the matrix . . While the variance is far from unity. Fig. for the indices . which are equal to continuous counterpart of the variance in this example is . Step 1) Compute the modification angle . It must be noted that the postphase where compensation factors in (58) and (62) are different. With the help of the commutative matrix . we again deal with the rectangular window shown in Example 2. The correis equal to 1. and the number of points in the interval is still signal becomes 37. The operation in the DFRFT performs elliptical rotation in the continuous time–frequency plane for the case . it will become evaluated at the points (60) (61) can be Equation (61) indicates that the FRFT with angle implemented by a DFRFT with angle . Step 2) Calculate the -point DFRFT of the signal with the angular parameter .

He then received the M. and D. National Kaohsiung First University of Science and Technology. Bitran. Gori.” IEEE Trans. pattern recognition. and O. He received the B.” IEEE Trans. His research interests include digital signal processing. 1995. V. “Digital computation of the fractional Fourier transform.” Appl. filtering. C. F. The DFRFT proposed in this paper not only supplies the similar transforms to match with those of the continuous case but also preserves the rotation properties. pp. 21–67.” J.. Oppenheim. Math. From 1997 to 1998. [19] B. 1986. 5.” in Proc. 25. A. “Eigenvalue and eigenvector decomposition of the discrete Fourier transform. Taipei. Opt. The Fourier Transform and Its Applications. and wavelets. Nov. W. time–frequency analysis.” IEEE Signal Processing Mag. H. Amer. Tamsui Oxford University College. 1798–1801. 1998. Inst. 547–559. in 1970 and the M. Mar. Min-Hung Yeh was born in Taipei. Aademir. Taipei. Circuits Syst. Circuits Syst. 40–41. Lohmann and B. July 15.. Amer. in 1965... Rev. [12] D. G.S. vol. [8] A. Ozaktas. Dickinson and K.. A. “Relationships between the Radon–Wigner and fractional Fourier transforms. 1994. NTU. 205–207. 1996. J.” Signal Process. pp. N.E. 1972. pp. W. He is currently an Assistant Professor with the Department of Computer Information Science. Ozaktas. and M.” IEEE Trans. M. with honors. Kutay. 7599–7602. REFERENCES [1] R. in 1964. Mendlovic. W. [5] H. A. [22] G. 303–309. [7] R. and laser holography. 1995. both from the National Taiwan University. Ltd. Amer. Taipei. NJ: Prentice-Hall. M. and Ph. Barshan. Yeh. E. 10. T. he was an Assistant Professor of the Department of Electronics Engineering. G.. Opt. 25–31. Kerr. vol. K. Discrete-Time Signal Processing. Kutay. Namias. MAY 1999 Furthermore. 33. 1980. pp. 42. Taipei. pp. R. 2522–2531. . Erkaya. A.S. vol. A. 1191–1194. Chunghwa Telecom Company.” Signal [26] Process. [15] .-C. Speech. Ozaktas. M. 10. Ozaktas. McClellan. pp. Chien-Cheng Tseng (S’90–M’95) was born in Taipei. Mendlovic. IEEE Int. vol.. “The fractional Fourier transform and time-frequency representation. Sansone. degree in computer engineering from the National Chiao-Tung University. Tamsui. Math. Soc. W. NJ: Prentice-Hall.E. degree. McClellan and T.” IEEE Trans. “Improved discrete fractional Fourier transform. M. Tatung Institute of Technology. Apr. pp. New York: McGraw-Hill. Dr.-H. pp. Almeida. 1993. H. vol. He is currently an Assistant Professor with the Department of Computer and Communication Engineering. vol. Arikan. “The discrete rotational Fourier transform. “Linear and quadratic timefrequency signal representations. respectively. vol.” J. Linear Algebra. 1997. “The discrete fractional Fourier transform. Soc.. June 1994. Amer. 3084–3091. 67. 536–539. F. Soc. Golub and C. pp. “Fractional Fourier transformations and their optical implementation—II. Hsinchu. Signal Processing. Spence. A. degrees from the University of California. [20] B. Onural. vol.O. Apr. Taipei. From 1995 to 1997. “Convolution. Methods of Theoretical Physics. Phys.C. 2141–2150. [31] O. W. 1994. A.” J.. N. [16] H. C. [21] J. He received the B. [6] H.. R. 3. 99–108. M. pp.. [23] A. Friedberg. Mendlovic.D. Symp. Soo-Chang Pei (SM’89) was born in Soo-Auo. F. O.” IMA J. 1989. [14] D. W. 6th IEEE DSP Workshop. 1994. 1987. in 1972 and 1975. and it is an important tool for signal processing. Dragonman. Soffer. in 1988. 66–74.. pp. [29] S. [24] S. 15–16. “Discrete fractional Hartley and Fourier transforms. Opt. 2nd ed. “Two dimensional discrete fractional Fourier transform. Santarsiero. Lohmann. the relationship between the FRFT and the . An angle modification and a postphase compensation in the DFRFT for elliptical rotation are required to match the results that are similar for the continuous FRFT. pp. and L. 13. Taiwan. pp. ASSP-30. 1998. His main research interests are in fractional Fourier transforms. 1047–1049. Feb. Taipei.. “Fractional correlation. A. June 1996. Santhanam and J. Lohmann. vol. Steiglitz. Shyu. He was an Engineering Officer in the Chinese Navy Shipyard from 1970 to 1971. This DFRFT provides a method for implementing the discrete FRFT. Sept. New York: McGraw-Hill. vol. degree in electrical engineering in 1997.” IEEE Trans. 45. he was a Research Assistant at UCSB. 1875–1881. vol. H. Press.. Audio Electroacoust.E. 25. Kutay.D. Pei and M.” in Proc.D. 1994. Jan. [10] D. C. 39. 1989. 994–998. pp. pp. Oct. Matrix Computations. Symp. [17] B. from the Tatung Institute of Technology. R. Opt. vol. VOL. Parks. Appl. Bourdeaux-Bartels. NO. pp. Taoyuan. Feschbach.. in 1990 and 1995. vol. pp. “Fractional Fourier transformations and their optical implementation—I. Opt. Signal Processing. 241–265. 9. Feb. p. 1982. [9] L. Y.” IEEE Trans. DFRFT can be established as follows. [18] G. Dorsch. 42. M. he was an Associate Research Engineer at Telecommunication Laboratories. from 1981 to 1983.” in Proc. Time–Freq. His research interests include digital signal processing. Taiwan. he is Professor and Chairman of the Electrical Engineering Department.S.” Opt. which is the same as that implementing the DFRFT is of the DFT. 11. Ozaktas.” Appl. Morse and H. 1994. 1993.” J. A.. [2] A. H. 119–124. vol.O. 1416. pp. [25] 22. 665–675. . Englewood Cliffs. 34. and R. and L. Opt. and D. time–frequency plane. [11] H. pp. Englewood Cliffs. Ozaktas. Acoust. optical information processing. M. H. [4] F. 11. MD: Johns Hopkins Univ. Parks.S. New York. He was a Professor and Chairman with the Electrical Engineering Department. 44. “Chirp filtering in the fractional Fourier domain. 1996. “Fractional Wigner distribution function. Time-Scale Anal. [30] G. J. Ozaktas. “Convolution and filtering in fractional Fourier domain.. Arikan. 1996. Orthogonal Functions. “Effect of fractional Fourier transformation on time-frequency distributions belonging to the Cohen class. H. AU-20. Weisburn. Taipei. If the DFRFT performs a circular rotation of the signal in the . “On Namias’ fractional Fourier transforms. degree from the National Taiwan University (NTU). . vol. M.” Opt.1348 IEEE TRANSACTIONS ON SIGNAL PROCESSING. A. Baltimore. “Generalized self-Fourier functions. [27] S. Mendlovic.” IEEE Signal Processing Lett. He received the B. 199–203. “The fractional order Fourier transform and its application to quantum mechanics. Van Loan. From 1971 to 1975. C. pp. May 1996. Pei is a Member of Eta Kappa Nu and the Optical Society of America. [3] V. 159–175. [13] H. 1996.C. IEEE Int. 47. H. Soc. vol. Cincotti. degrees from the National Taiwan University. and the M. 1. pp. Shenoy. 1992. Santa Barbara (UCSB). “Separation of transient signals. if the DFRFT becomes an elliptical rotation in the continuous time–frequency plane. pp. Ozaktas. A. Applicat. Hwa Hsia College of Technology and Commerce. H. all in electrical engineering. M. H. Tseng. vol. respectively. “Eigenvectors and functions of the discrete Fourier transform. Soc. “Discrete fractional Fourier transform. and G. Insel. vol. 1992. and J. D. Mar. vol. 46. pp. and electronic commerce. 1959. vol. Taiwan. The complexity for .S. Feb. pp.” J. and Ph. 1953. Interscience. and multiplexing in fractional Fourier domains and their relationship to chirp and wavelet transforms. Taiwan. vol. Taipei. On the other hand.. [28] P. vol. image processing. Bozdagi. and A. Yeh.O. Pei.S. Signal Processing. B. M.” J.. Barshan. M. Amer. Lett. Signal Processing. B. in 1987. in 1949.. 474–478.C. degree in computer science and information engineering in 1992 and the Ph. pp. McBride and F. and M. 1989. Hlawatsch and G. Bracewell.” J. Taiwan. Mendlovic and H.E. B. Opt. M. II. “Fractional Fourier domains. Presently.

You're Reading a Free Preview

Descarregar
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->