Bessel Samp 3

6
5082 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 59, NO. 10, OCTOBER 2011 manipulate such matrices. The good agreement between the theoreti- cally-predicted and the empirically-obtained separation performance was demonstrated in simulation. A remaining key-question is, given the sources’ covariance matrices and our analytic expressions for the performance prediction, whether (and if so, how) the association-matrices can be chosen so as to opti- mize the performance (in some sense). This question will be addressed in future work. REFERENCES [1] A. Yeredo r, “On optimal selection of correlation matrices for matrix- pencil-based separation,” in Proc. 8th Int. Conf. Independent Compo- nent Anal. Source Separation (ICA), 2009, pp. 187–194. [2] L. Tong , V. C. Soon, Y.-F. Huang, and R. Liu, “AMUSE: A new blind identication algorithm, in Proc. ISCAS , 1990, pp. 1784–1787. [3] L. Tong and R. Liu, “Blind estimatio n of correlated source signals, ” in Proc. 24th Asilomar Conf., 1990, pp. 258–262. [4] J. F. Cardo so, “Source separati on usin g high er-ord er moments, in Pro c. IEEE Int. Conf. Acoust., Speech, Signal Process . (ICASSP), 1989, pp. 2109–2112. [5] A. Y eredo r, “Blin d source sepa ratio n via the seco nd chara cteri stic func - tion,” Signal Process., vol. 80, no. 5, pp. 897–902, 2000. [6] A. M. T omé, “The gene ralize d eigen deco mpos ition approa ch to the blind source separation problem,”  Digit. Signal Process., vol. 16, no. 3, pp. 288–302, 2006. [7] L. Parra and P. Sajda, “Blind source separati on via generalized eigen- value decompositio n,” J. Mach. Learn. Res. , vol. 4, pp. 1261–1269, 2003. [8] E. Ollila, H. Oja, and V. Koivune n, “Complex-v alued ICA based on a pair of generalized covariance matrices,” Comput. Statist. Data Anal. , vol. 52, pp. 3789–3805, 2008. [9] J. F. Car doso and A. Souloumiac, “Blind bea mfo rmi ng for non Gaussian signals,”  Proc . Inst. Electr. Eng.—F , vol. 140, no. 6, pp. 362–370, 1993. [10] L. Fêty and J.-P. V. Uffelen, “New methods for signal separatio n,” in Proc. 14th Conf. HF Radio Syst. Tech., 1988, pp. 226–230. [11] L. Parra and C. Spence, “Convolutiv e blind source separation of non- stationary sources,  IEEE Trans. Speech Audio Process., pp. 320–327, 2000. [12] A. Belouchrani, K. Abed-Meraim, J.-F. Cardoso, and E. Moulines, “A blind source separation technique using second-order statistics,”  IEEE Trans. Signal Process., vol. 45, no. 2, pp. 434–444, 1997. [13] D.-T. Phamand J.-F. Card oso, “ Blind separation of instantaneous mix- tures of nonstationary sources,”  IEEE Trans. Signal Process. , vol. 49, no. 9, pp. 1837–1848, 2001. [14] A. Belouch rani and M. G. Amin, “Blind source separ ation based on time-frequency signal representations,  IEEE Trans. Signal Process.. , vol. 46, no. 11, pp. 2888–2897, 1998. [15] K. Abed-Meraim, Y. Xiang, J. H. Manton , and Y. Hua, “Blind source separation using second-order cyclostationary statistics,  IEEE Trans. Signal Process., vol. 49, no. 4, pp. 694–701, 2001. [16] A. Yeredo r, “On using exact joint diagonali zation for non-iterative ap- proximate joint diagonalization ,” IEEE Signal Process. Lett. , vol. 12, no. 9, pp. 645–648, 2005. [17] J.-F. Cardoso and B. Laheld, “Equivariant adaptiv e source separation,”  IEEE Trans. Signal Process. , vol. 44, no. 12, pp. 3017–3030, 1996. A Class of Scaled Bessel Sampling Theorems Luc Knockaert  Abstract—Sampling theorems for a class of scaled Bessel unitary trans- forms are presented. The derivations are based on the properties of the generalized Laguerre functions. This class of scaled Bessel unitary trans- formsincludesthe cla ssi cal sin e andcosin e tra nsf orms, but als o nov el chi rp sine and modied Hankel transforms. The results for the sine and cosine transform can also be utilized to yield a sampling theorem, different from Shannon’s, for the Fourier transform.  Index Ter ms—Besse l funct ions,chirp trans form, Hank el trans form, sam- pling theorems. I. INTRODUCTION The classical Shannon [1] sampling theorem is omnipresent in large areas of signal processing. Numerous generalizations have been de- rived [2]–[5], while extensions of the sampling theorem have arisen in connection with wavelets [6], the fractional Fourier and quasi-Fourier transforms [7]–[10]. In this correspondence, we develop sampling the- orems for a class of scaled Bessel unitary transforms in a constructive way, based on the generalized Laguerre functions. This class includes the sine and cosine transforms, but also novel chirp sine and modied Hankel transforms. The results for the sine and cosine transform can also be utilized to yield a sampling theorem, different from Shannon’s, for the Fourier transform. Novel asymptotic truncation error expres- sions are derived and pertinent examples are presented. II . MAIN RESULTS We begin by stating the following result due to Kramer [2], [4], [5]. Theorem 1:  Let be a rea l un ita ry o per ato r over        with produc t kerne l    such that the set         for ms a comple te or- thogonal basis in      for some countable set of strictly increasing positive constants      . Suppose further that    is -ba ndl imited to  , i.e., the transform             is in        such that        for   . Then,    admits the sampling representation              (1) where the sampling kernel is                     and            . Proof:  See [2], [4], and [5]. Manu scrip t recei ved Febru ary 15,2011; revi sed June14, 2011 ; accep ted June 14, 2011. Date of publication June 27, 2011; date of current version September 14, 2011. The associate editor coordinating the review of this manuscript and approving it for publication was Prof. Xiang-Gen Xia. This work was supported by a grant of the Research Foundation-Flanders (FWO-Vlaanderen). The author is with the INTEC-IBCN-IBBT , Ghent University , B-9050 Gent, Belgium (e-mail: luc.knockaert@in tec.ugent.be). Digital Object Identier 10.1109/TSP.2011.21 60634 1053-587X/$26.00 © 2011 IEEE

Transcript of Bessel Samp 3

Page 1: Bessel Samp 3

8/13/2019 Bessel Samp 3

http://slidepdf.com/reader/full/bessel-samp-3 1/5

5082 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 59, NO. 10, OCTOBER 2011

manipulate such matrices. The good agreement between the theoreti-

cally-predicted and the empirically-obtained separation performance

was demonstrated in simulation.

A remaining key-question is, given the sources’ covariance matrices

and our analytic expressions for the performance prediction, whether

(and if so, how) the association-matrices can be chosen so as to opti-

mize the performance (in some sense). This question will be addressed

in future work.

REFERENCES

[1] A. Yeredor, “On optimal selection of correlation matrices for matrix-

pencil-based separation,” in  Proc. 8th Int. Conf. Independent Compo-

nent Anal. Source Separation (ICA), 2009, pp. 187–194.

[2] L. Tong, V. C. Soon, Y.-F. Huang, and R. Liu, “AMUSE: A new blindidentification algorithm,” in Proc. ISCAS , 1990, pp. 1784–1787.

[3] L. Tong and R. Liu, “Blind estimation of correlated source signals,” inProc. 24th Asilomar Conf., 1990, pp. 258–262.

[4] J. F. Cardoso, “Source separation using higher-order moments,” inProc. IEEE Int. Conf. Acoust., Speech, Signal Process. (ICASSP),1989, pp. 2109–2112.

[5] A. Yeredor, “Blind source separation via the second characteristic func-tion,” Signal Process., vol. 80, no. 5, pp. 897–902, 2000.

[6] A. M. Tomé, “The generalized eigendecomposition approach to the

blind source separation problem,”  Digit. Signal Process., vol. 16, no.

3, pp. 288–302, 2006.[7] L. Parra and P. Sajda, “Blind source separation via generalized eigen-

value decomposition,” J. Mach. Learn. Res., vol. 4, pp. 1261–1269,2003.

[8] E. Ollila, H. Oja, and V. Koivunen, “Complex-valued ICA based on a

pair of generalized covariance matrices,” Comput. Statist. Data Anal.,

vol. 52, pp. 3789–3805, 2008.[9] J. F. Cardoso and A. Souloumiac, “Blind beamforming for non

Gaussian signals,”   Proc. Inst. Electr. Eng.—F , vol. 140, no. 6, pp.

362–370, 1993.

[10] L. Fêty and J.-P. V. Uffelen, “New methods for signal separation,” inProc. 14th Conf. HF Radio Syst. Tech., 1988, pp. 226–230.

[11] L. Parra and C. Spence, “Convolutive blind source separation of non-stationary sources,” IEEE Trans. Speech Audio Process., pp. 320–327,2000.

[12] A. Belouchrani, K. Abed-Meraim, J.-F. Cardoso, and E. Moulines, “A

blind source separation technique using second-order statistics,” IEEE 

Trans. Signal Process., vol. 45, no. 2, pp. 434–444, 1997.

[13] D.-T. Pham and J.-F. Cardoso, “Blind separation of instantaneous mix-tures of nonstationary sources,”  IEEE Trans. Signal Process., vol. 49,

no. 9, pp. 1837–1848, 2001.

[14] A. Belouchrani and M. G. Amin, “Blind source separation based ontime-frequency signal representations,” IEEE Trans. Signal Process..,

vol. 46, no. 11, pp. 2888–2897, 1998.

[15] K. Abed-Meraim, Y. Xiang, J. H. Manton, and Y. Hua, “Blind sourceseparation using second-order cyclostationary statistics,” IEEE Trans.

Signal Process., vol. 49, no. 4, pp. 694–701, 2001.

[16] A. Yeredor, “On using exact joint diagonalization for non-iterative ap-proximate joint diagonalization,”  IEEE Signal Process. Lett., vol. 12,

no. 9, pp. 645–648, 2005.[17] J.-F. Cardoso and B. Laheld, “Equivariant adaptive source separation,”

 IEEE Trans. Signal Process., vol. 44, no. 12, pp. 3017–3030, 1996.

A Class of Scaled Bessel Sampling Theorems

Luc Knockaert

 Abstract—Sampling theorems for a class of scaled Bessel unitary trans-forms are presented. The derivations are based on the properties of the

generalized Laguerre functions. This class of scaled Bessel unitary trans-formsincludesthe classical sine andcosine transforms, but also novel chirpsine and modified Hankel transforms. The results for the sine and cosinetransform can also be utilized to yield a sampling theorem, different from

Shannon’s, for the Fourier transform.

 Index Terms—Bessel functions,chirp transform, Hankel transform, sam-

pling theorems.

I. INTRODUCTION

The classical Shannon [1] sampling theorem is omnipresent in large

areas of signal processing. Numerous generalizations have been de-

rived [2]–[5], while extensions of the sampling theorem have arisen in

connection with wavelets [6], the fractional Fourier and quasi-Fourier

transforms [7]–[10]. In this correspondence, we develop sampling the-orems for a class of scaled Bessel unitary transforms in a constructive

way, based on the generalized Laguerre functions. This class includes

the sine and cosine transforms, but also novel chirp sine and modified

Hankel transforms. The results for the sine and cosine transform can

also be utilized to yield a sampling theorem, different from Shannon’s,

for the Fourier transform. Novel asymptotic truncation error expres-

sions are derived and pertinent examples are presented.

II. MAIN RESULTS

We begin by stating the following result due to Kramer [2], [4], [5].

Theorem 1:   Let be a real unitary operator over  

      with

product kernel     such that the set      

 

   forms a complete or-

thogonal basis in  

    for some countable set of strictly increasingpositive constants

   

 

 

  . Suppose further that 

  is -bandlimited to

  , i.e., the transform

 

     

 

   

is in  

      such that  

       for    . Then,     admits the

sampling representation

   

 

 

 

 

 

 

 

 

(1)

where the sampling kernel is

 

 

   

 

 

 

       

 

 

and 

 

 

   

 

 

  .

Proof:  See [2], [4], and [5].

Manuscript received February 15, 2011; revised June14, 2011; accepted June14, 2011. Date of publication June 27, 2011; date of current version September14, 2011. The associate editor coordinating the review of this manuscript andapproving it for publication was Prof. Xiang-Gen Xia. This work was supportedby a grant of the Research Foundation-Flanders (FWO-Vlaanderen).

The author is with the INTEC-IBCN-IBBT, Ghent University, B-9050 Gent,Belgium (e-mail: [email protected]).

Digital Object Identifier 10.1109/TSP.2011.2160634

1053-587X/$26.00 © 2011 IEEE

Page 2: Bessel Samp 3

8/13/2019 Bessel Samp 3

http://slidepdf.com/reader/full/bessel-samp-3 2/5

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 59, NO. 10, OCTOBER 2011 5083

Note that, by virtue of the unitarity of , the functions  

 

  are

orthogonal in  

 

   .

Next we need to show that there exist product kernels with the re-

quired properties. In fact there are many of them. We have

Theorem 2:  The class of scaled Bessel product kernels

 

 

   

  0 

 

 

 

     

0  

with

 

 

   

 

 

 

 

 

 

 

 

 

 

 

 

     

satisfy the requirements of Theorem 1.

Proof:  Consider the generalized Laguerre functions

 

 

 

   

 

0      

 

 

 

0   

 

 

 

   

0  

where  

 

 

    are the Laguerre polynomials. It is shown in [11] that the

generalized Laguerre functions  

 

 

    form a complete orthonormal

basis for  

 

   . Next consider the operator with kernel

   

 

 

 

 

 

 

   

 

 

Clearly, the operator is real and unitary with eigenfunctions  

 

 

and eigenvalues 0    

   . Moreover,     is a product kernel since

we have

   

 

 

 

 

 

This follows from the well-established formula [12, p. 189]

 

 

 

0      

 

 

 

   

 

 

   

 

 

0  

 

0  

     

 

0  

 

0   

 

 

 

 

 

 

0  

valid for  

     , and letting     tend to 0 

   from inside the unit circle.

From the complete orthonormal basis  

 

 

  , we can construct another

one by non-linear scaling

 

 

 

 

  0   

 

 

 

 

leading to

 

 

 

 

 

 

 

 

 

   

 

 

   

  0   

 

 

 

We areleft with one last, butmajor, requirement :   

 

 

 

 

   should

form a complete orthogonal basis in 

 

  . After some calculations,

we find that

 

 

 

 

   

 

 

   

 

 

 

 

 

 

  0   

 

 

 

 

 

 

 

 

 

 

 

(2)

For orthogonality, expression (2) must vanish when  

    . This is re-

mindful of the Fourier-Bessel [12] relationships, which can be written

as

 

 

 

 

 

 

 

 

 

   

 

 

 

 

and where   

 

   are the positive zeroes of   

 

    . Hence, we must take

 

 

   

 

 

 

 

 

 

 

 

 

0   

 

 

 

 

     

(3)

In other words, we have proved that the functions

 

 

 

 

       

are orthogonal in  

    . Completeness is guaranteed by a theorem

of Szász [13, p. 282].

The sampling kernel  

 

  can be explicitly written as

 

 

 

 

  0 

 

 

 

 

 

 

 

 

 

 

 

0  

(4)

It can be verified that  

0    

 

 

   , implying the characteristic

interpolation property  

 

 

   

  . The sampling representation (1)

can be written as      

 

   

 

  , where  

 

  is the truncated

sampling expression and  

 

  is the truncation error

 

 

 

 

     

 

 

 

 

 

 

The asymptotic behavior of the truncation error for  0  

   is the sub-

 ject of the two following corollaries.

Corollary 1:   Let the -bandlimited function     be  

0   

  for

0  

  with          . Then, for fixed

 

 

   

0        0    for  0  

 

and 

 

 

   

0     

for   0   

Proof:  See Appendix I.The next corollary proves that there are numerous -bandlimited

functions which satisfy the requirements of Corollary 1. We have

Corollary 2:   Let  

      be in  

      such that  

         

 

  for

 

0    with  

  

0        . Then

     

0     

for 0  

 

Proof:  See Appendix II.

Note that when          , Corollary 1 yields  

 

 

   

and also 

 

 

     

  .

III. EXAMPLES

 A. Sine Transform

Here, we take          and      . The product kernel is

 

 

 

 

 

 

 

 

 

and  

 

   

 

     . The sampling kernel is

 

 

 

0   

0   

 

   

The sampling scheme is equispaced with sampling step1    

  , but

unlike the usual Fourier sinc sampling kernel—see the sequel—which

falls off as  

  at infinity, the sine sampling kernel falls off fasterat a  

  rate.

Page 3: Bessel Samp 3

8/13/2019 Bessel Samp 3

http://slidepdf.com/reader/full/bessel-samp-3 3/5

5084 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 59, NO. 10, OCTOBER 2011

 B. Cosine Transform

Here, we take    

0        and      . The product kernel is

 

0   

 

 

 

0   

 

 

 

and  

 

   

 

   

0       . The sampling kernel is

 

 

 

0  

 

0  

 

 

   

 

Note that the sampling kernels for the cosine and sine transforms are

the same, except for a translation 0     

  in the 

 

  coefficients.

C. Fourier Transform

Although the Fourier transform

0   

 

   

  admits the usual

 

sampling theorem

   

 

    0  

    1  

0   1

where 1      , it is possible to utilize the sine and cosine sampling

kernels. If we define the even and odd components of a function as

 

 

 

 

     

0   

 

 

 

 

 

0  

the Fourier transform can be written as

 

 

   

 

0  

 

   

 

 

 

 

 

 

 

 

 

Hence, signal reconstruction with  

  behavior at infinity is pos-

sible by utilizing the sine sampling kernel for the odd component of the

function, and the cosine sampling kernel for the even component of the

function.

We compared the bandlimited function

   

 

0   

 

   

 

 

 

   

   

0     

with its truncated sine plus cosine sampling reconstruction 

 

  for    and      over the range  

 

    . The result is shown in

Fig. 1.

 D. Hankel Transform

We will restrict ourselves to the modified Hankel transform of order

0, in which case      and          The product kernel is

 

 

 

 

 

 

 

and  

 

   

 

  with   

 

   the positive zeroes of   

    . The sampling

kernel is given by

 

 

 

 

 

 

 

 

 

 

 

 

0   

 

Fig. 1.     versus     for the Fourier transform.

Fig. 2.     versus     for the Hankel transform.

Fig. 3.     versus     for the Hankel transform.

We compared the -bandlimited function

   

 

 

 

 

 

   

 

 

 

 

 

with its truncated sampling representation  

 

  for      and    

  over the range  

 

     . The result is shown in Fig. 2. With the

same sampling conditions as above, another example is

   

 

 

 

   

0    

 

 

   

 

 

0     

 

 

 

 

For this example, we have      

0     

  and hence we may ex-

pect a better truncated sampling approximation as compared to the firstexample. The result is shown in Fig. 3.

Page 4: Bessel Samp 3

8/13/2019 Bessel Samp 3

http://slidepdf.com/reader/full/bessel-samp-3 4/5

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 59, NO. 10, OCTOBER 2011 5085

Fig. 4.     versus     for the chirp sine transform.

 E. Chirp Sine Transform

Here, we take          and      . The product kernel is

 

 

 

 

 

and  

 

 

 

   . The sampling kernel is

 

 

 

0   

0   

   

 

 

 

 

We compared the -bandlimited function

   

 

 

   

         

where     is the Fresnel integral      

 

 

      , with its

truncated sampling representation  

 

  for      and      over

the range  

 

     The result is shown in Fig. 4.

IV. CONCLUSION

Sampling theorems for a class of scaled Bessel unitary transforms

have been presented. This class of scaled Bessel unitary transforms in-

cludesthe classical sine andcosine transforms, butalso novel chirp sine

and modified Hankel transforms. Novel asymptotic truncation error ex-

pressions were derived and pertinent examples presented.

APPENDIX I

PROOF OF COROLLARY  1

Corollary 1:   Let the -bandlimited function 

  be 

0   

  for 0  

  with          . Then for fixed

 

 

   

0        0   

for  0  

 

and 

 

 

   

0     

for   0   

Proof:  The asymptotic McMahon expansion [14] of the zeros of 

Bessel functions guarantees that 

 

     

  . This implies

   

 

     

0   

 

   

0   

From formula (4) we deduce that

 

 

   

  0   

 

 

 

 

 

 

 

   

    0   

We do not have to worry about  

 

 

 

  or for that case  

 

 

 

  since

these values are asymptotically located on the 6    

 

  envelope of the

Bessel function. Hence

 

 

 

 

     

 

0        0   

   

0        0   

For the squared truncation error, we utilize the orthogonality of the

functions 

 

  . We have

 

 

 

 

 

     

   

 

 

 

 

 

 

 

     

   

 

 

 

 

From (3), we deduce that

   

 

 

 

 

   

0      0 

leading to

 

 

 

   

0     

APPENDIX  II

PROOF OF COROLLARY  2

Corollary 2:   Let  

      be in  

      such that  

         

 

for   0      with  

  

0        . Then

     

0     

for 0   

Proof:  Without loss of generality, we can take      . Consider

the integral [15, p. 684]

 

 

 

 

   

0    0 

     

0   

 

   

  0    0 

0   

  0 

 

 

   

Here,  

    

0      is a constant and  

 

  is the Lommel

function. Since  

 

   

  0 

  for

0  

   , we readily obtain that

 

 

 

 

     

0    0   

 

  

  

   

0   

 

   (5)

Now let   

 

   be a strictly increasing sequence of real numbers greater

than 0        and take  

   

  

0        . Then Müntz’s theorem [16]

states that the system of powers   

 

   is complete in  

    if and

only if 

 

 

 

 

     

 

 

   

 

  (6)

Condition (6) is not difficult to satisfy, i.e., taking  

 

     for  

  

is just fine. Completeness implies that  

      can be written as

 

     

 

 

 

 

 

Page 5: Bessel Samp 3

8/13/2019 Bessel Samp 3

http://slidepdf.com/reader/full/bessel-samp-3 5/5

5086 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 59, NO. 10, OCTOBER 2011

and this leads to

   

 

 

 

 

 

 

 

 

For any of the exponents      

 

  , we can write

 

 

 

 

   

  0   

 

     

 

 

 

By construction  

 

            and hence (5) implies that

 

 

 

 

   

  0   

0   

 

0     

for

0  

 

Since     consists of an infinite sum of elements with identical be-

haviour at infinity,     itself must exhibit the same behavior at infinity

and the proof is complete.

REFERENCES

[1] C. E. Shannon, “Communication in the presence of noise,” in  Proc.

 IRE , 1949, vol. 37, pp. 10–21.[2] H. P. Kramer, “A generalized sampling theorem,” J. Math. Phys., vol.

38, pp. 68–72, 1959.[3] A. J. Jerri, “Some applications for Kramer’s generalized sampling the-

orem,” J. Eng. Math., vol. 3, no. 2, pp. 103–105, 1969.[4] A. J. Jerri, “The Shannon sampling theorem—its various exten-

sions and applications: A tutorial review,”  Proc. IEEE , vol. 11, pp.1565–1596, 1977.

[5] A. I. Zayed , Advances in Shannon’s Sampling Theory. Boca Raton,

FL: CRC Press, 1993.[6] I. W. Selesnick, “Interpolating multiwavelet bases and the sampling

theorem,” IEEE Trans. Signal Process., vol. 47, no. 6, pp. 1615–1621,1999.

[7] X. G. Xia, “On bandlimited signals with fractional Fourier transform,” IEEE Signa l Process. Lett., vol. 3, no. 3, pp. 72–74, 1996.

[8] T. Erseghe, P. Kraniauskas, and G. Cariolaro, “Unified fractionalFourier transform and sampling theorem,”   IEEE Trans. Signal

Process., vol. 47, no. 12, pp. 3419–3423, 1999.[9] Q. Wang and L. Wu, “A sampling theorem associated with

quasi-Fourier transform,”   IEEE Trans. Signal Process., vol. 48,no. 3, pp. 895–895, 2000.

[10] D. Y. Wei, Q. W. Ran, and Y. M. Li, “Generalized sampling expansionfor bandlimited signals associated with the fractional Fourier trans-form,” IEEE Signal Process. Lett., vol. 17, no. 6, pp. 595–598, 2010.

[11] G. Szego , Orthogonal Polynomials. New York: Amer. Math. Soc.Coll., 1939.

[12] A. Erdélyi, W. Magnus, F. Oberhettinger, and F. G. Tricomi , Higher Transcendental Functions. New York: McGraw-Hill, 1953, vol. 2.

[13] P. J. Davis , Interpolation and Approximation. New York: Dover,1975.

[14] M. Abramowitz and I. Stegun , Handbook of Mathematical Func-

tions. New York: Dover, 1965.[15] I. S. Gradshteyn and I. Ryzhik  , Table of Integrals, Series and Prod-

ucts. New York: Academic, 1965.[16] V. Operstein, “Full Müntztheoremin     ,” J. Approx. Theory, vol.

85, pp. 233–235, 1996.

Sparse Approximation Property and Stable Recovery of 

Sparse Signals From Noisy Measurements

Qiyu Sun

 Abstract—In this correspondence, we introduce a sparse approximationproperty of order for a measurement matrix

 

  :  

for all     where     is the best -sparse approximation of the vector

 

  in ,

  is the -sparse approximation error of the vector

 

in , and and are positive constants. The sparse approximationproperty for a measurement matrix can be thought of as a weaker versionof its restricted isometry property and a stronger version of its null spaceproperty. In this correspondence, we show that the sparse approximationproperty is an appropriate condition on a measurement matrix to considerstable recovery of any compressible signal from its noisy measurements.In particular, we show that any compressible signal can be stably recov-ered from its noisy measurements via solving an -minimization problem

if the measurement matrix has the sparse approximation property with , and conversely the measurement matrix has the sparse ap-

proximation property with

  if any compressible signal can be

stably recovered from its noisy measurements via solving an -minimiza-tion problem.

 Index Terms—Additive noise, approximation methods, compressedsensing, signal reconstruction.

I. INTRODUCTION

Given positive integers     and     with        and a measurement

matrix     of size   2     , we consider the problem of compressive sam-

pling in recovering a compressible signal    

   from its noisy mea-

surements            via solving the following  

   -minimization

problem:

       

 

 

      0     

 

   

(I.1)

where          ,           ,        , and the measurement noise  

satisfies      

 

     [1]–[8]. Here, 1  

 

  ,        , stands for the

“  

   -norm” on the Euclidean space.

Given a subset            and a vector    

   , denote by  

 

the vector whose components on     are the same as those of the vector

 

and vanish on the complement 

  . A vector   

   is said to be 

  -

sparse if       

 

  for some subset            with its cardinality

    less than or equal to    , where        . Denote by 6 

  the set of 

all    -sparse vectors. Given a vector     , its best    -sparse approximation

vector 

  in 

   is an 

  -sparse vector which has minimal distance to 

in  

   ; i.e.,     0   

 

 

   

 

   

    6 

    0     

 

  . For      ,

we use  

    instead of   

    for brevity.

In this correspondence, we introduce a new property of a measure-

ment matrix     : there exist positive constants     and     such that

   

 

 

 

         

 

 

   

  0 

 

 

 

 

for all    

 

  (I.2)

Manuscript received June 06, 2010; revised October 26, 2010 and April 13,2011; accepted June 28, 2011. Dateof publication July 12, 2011; date of currentversion September 14, 2011. The associate editor coordinating the review of thismanuscript and approving it for publication was Prof. Namrata Vaswani. Part of this work was done when the author visited Ecole Polytechnique Federale deLausanne.

The author is with the Department of Mathematics, University of CentralFlorida, Orlando, FL 32816 USA (e-mail: [email protected]).

Digital Object Identifier 10.1109/TSP.2011.2161470

1053-587X/$26.00 © 2011 IEEE