This action might not be possible to undo. Are you sure you want to continue?

Welcome to Scribd! Start your free trial and access books, documents and more.

Find out more5, 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

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

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

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

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

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

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

and x (n ) = 0 otherwise. x(n) = 1 when 06 n 6. x (n ) = 1 when 06 n 6. DFRFT by GSA method of a rectangular window function. the DFT Hermite eigenvectors can be computed a priori. k = 032 1 1 1 32. NO. When = 3=7. ( and the rotation angle are When the number of points determined. 5. 4. identity transform) and any angle transform domains. the computation of the DFRFT can be implemented only by a transform kernel matrix multiplication. the following method for implementing the DFRFT can be applied: (38) (39) frequency. This ﬁgure has a Fig. DFRFT by GSA of a chirp signal. MAY 1999 Fig. DFRFT by OPA method of a rectangular window function. Then. an impulse-like output is obtained. and x(n) = 0 otherwise. computing the DFRFT is . DFRFT by S method of a rectangular window function. 6. closer output to the continuous FRFT than Fig. ej 20:1141k . and it is the same as in the DFT case. If the rotational angles are adjusting. 4. 4. and the eigenvalues of DFRFT are also determined. output to the continuous FRFT than Fig. x (n ) = 1 when 06 n 6. the continuous FRFT in Fig. 1. 7.1342 IEEE TRANSACTIONS ON SIGNAL PROCESSING. This concept is also applied in the time domain . 47. VOL. and x (n ) = 0 otherwise. The complexity of . The output is close to Fig. This ﬁgure has a closer Fig. 5.

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

. VOL. In Proposition 3. While errors of the DFT Hermite eigenvectors will be reduced. the periods in the transform results change for different rotation angles. On the other hand. and the rotation angle in [31] is valid for a certain discrete set of rotation angles. In the following discussion. It is based on retaining the property of the DFT that a sampled periodic function transforms into a periodic function. The the overall frequency range that the DFT can represent is equal to . X is odd. the transform range and resolution of the DFT have been well discussed [23]. it is also is the number of points of the discrete the time resolution. Therefore. we will discuss the transform range and resolution for the DFRFT (see Table IV). the approximation eigenvectors. we will of the DFT Hermite eigenvector is compute the FRFT for a Hermite function with any variance . A. an alternative DFRFT has recently been proposed. b) is the (a. which satisﬁes the eigenvalue and eigenfunction relationship as a continuous FRFT kernel. we will review and understand the transform range and resolution for the is the conventional DFT. 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. Thus.1344 IEEE TRANSACTIONS ON SIGNAL PROCESSING. Samples of continuous FRFT of an impulse function. RELATIONSHIP BETWEEN FRFT AND DFRFT In this section. there will be four different eigenvalues for all the eigenvectors. our DFRFT can be used for any rotation angle and can provide very similar results as continuous cases. 11. In [31]. and there are some approximation errors in the DFT Hermite approaches inﬁnity. To begin with. 47. x(0) = 1. and x(t) = 0 otherwise. b) = F (b. the number of DFT Hermite eigenvectors is only ﬁnite. (41) is the eigenvector corresponding to the th-order where Hermite function. In the DFRFT of [17]. Here. NO. Transform Range and Resolution of DFRFT In this subsection. the method assigns the DFRFT eigenvalues to only four values. the proposed DFRFT has assigned a different eigenvalue for each eigenvector. and the frequency resolution is [23]. and more DFT Hermite eigenvectors are used to compute the DFRFT. MAY 1999 Fig. The development of the new DFRFT is based on the same idea as (2). we will establish the relationship between the FRFT and the DFRFT. it has been proved that the spread variance . Moreover. In the conventional DFT analysis. is the total recorded signal duration and is signal. and transform range in the time domain as well. IV.b)-element in DFRFT kernel matrix F [vn ] = e0jn vn where vn is Hermite-like function N F [(k)] 2 10j2cot ej cot If x is even. a) where F (a. ( one of the four eigensubspace has the same eigenvalue. However. the signal and transform results in [31] are discrete and periodic. sampling interval for the original continuous signal. X is even. the DFRFT can be used to give the similar continuous transform results within the accuracy of the discrete ﬁnite vector approximation. Then. If x is odd. It should be noted that the number of eigenfunctions for the continuous FRFT in (2) is inﬁnite. 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.

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

is the elliptical rotation . MAY 1999 Fig. The transform kernels of the circular and elliptical rotations deﬁned above are inﬁnite numbers of sums for Hermite eigenvectors. 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. (57) where tion factor is equal to . the circular rotation is just the FRFT of the signal. • Elliptical Rotation: where (54) (55) (56) spread variance changes in these two schemes. 12. Three different cases of rotation concept for DFRFT in continuous time-frequency plane. VOL. (59) . (52) (53) is the transform kernel of circular rotation. 13. While .1346 IEEE TRANSACTIONS ON SIGNAL PROCESSING. the elliptical rotation will of signal become the circular rotation. (a) (b) (c) Fig. 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. NO. and where is deﬁned as in (43). is the circular rotation of signal . 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. and the post-phase compensa- is the transform kernel of elliptical rotation. 5. 47. If the Hermite eigenvectors in the circular rotation are from order 0 to inﬁnity.

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

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

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue reading from where you left off, or restart the preview.

scribd