Reconstruction of spline spectra-signals from generalized sinc function by finitely many samples

Abstract

Reconstruction of signals by their Fourier (transform) samples is investigated in many mathematical/engineering problems such as the inverse Radon transform and optical diffraction tomography. This paper concerns on the reconstruction of spline-spectra signals in \(V(\hbox {sinc}_{a})\) by finitely many Fourier samples, where \(\hbox {sinc}_{a}\) is the generalized sinc function. There are two main results on this topic. When the spectra knots are known, the exact reconstruction formula conducted by finitely many Fourier samples is established in the first main theorem. When the spectra knots are unknown, in the second main theorem we establish the approximations to the spline-spectra signals also by finitely many Fourier samples. Numerical simulations are conducted to check the efficiency of the approximation.

This is a preview of subscription content, access via your institution.

Fig. 1
Fig. 2
Fig. 3

References

  1. 1.

    Aldroubi, A., Gröchenig, K.: Nonuniform sampling and reconstruction in shift-invariant spaces. SIAM Rev. 43, 585–620 (2001)

    MathSciNet  Article  Google Scholar 

  2. 2.

    Cerejeiras, P., Chen, Q., Kähler, U.: A necessary and sufficient condition for a Bedrosian identity. Math. Methods Appl. Sci. 33(4), 493–500 (2009)

    MathSciNet  Article  Google Scholar 

  3. 3.

    Chen, Q., Wang, Y.B., Wang, Y.: A sampling theorem for non-bandlimited signals using generalized Sinc functions. Comput. Math. Appl. 56(6), 1650–1661 (2008)

    MathSciNet  Article  Google Scholar 

  4. 4.

    Chen, Q., Qian, T.: Sampling theorem and multi-scale spectrum based on non-linear Fourier atoms. Appl. Anal. 88(6), 903–919 (2009)

    MathSciNet  Article  Google Scholar 

  5. 5.

    Chen, Q., Micchelli, C.A., Wang, Y.: Functions with spline spectra and their applications. Int. J. Wavel. Multiresolut. Inf. Process. 8(2), 197–223 (2010)

    MathSciNet  Article  Google Scholar 

  6. 6.

    Chen, Q., Li, L., Malonek, H.R.: The construction of generalized B-spline low-pass filters related to possibility density. Sci. China Math. 55(12), 2481–2491 (2012)

    MathSciNet  Article  Google Scholar 

  7. 7.

    Chen, Q., Li, L., Wang, Y.: Amplitudes of mono-component signals and the generalized sampling functions. Signal Process. 94(1), 255–263 (2014)

    Article  Google Scholar 

  8. 8.

    Chen, Q., Qian, T., Li, Y.: Shannon-type sampling for multivariate non-bandlimited signals. Sci. China Math. 56(9), 1915–1934 (2013)

    MathSciNet  Article  Google Scholar 

  9. 9.

    Fan, S., Smith-Dryden, S., Zhao, J., Gausmann, S., Schhülzgen, A., Li, G., Saleh, B.E.: Optical fiber refractive index profiling by iterative optical diffraction tomography. J. Lightwave Technol. 36(24), 5754–5763 (2018)

    Article  Google Scholar 

  10. 10.

    Han, D., Zayed, A.I.: Sampling expansions for functions having values in a Banach space. Proc. Am. Math. Soc. 133(12), 3597–3607 (2005)

    MathSciNet  Article  Google Scholar 

  11. 11.

    Han, D., Nashed, M., Sun, Q.: Sampling expansions in reproducing kernel Hilbert and Banach spaces. Numer. Funct. Anal. Optim. 30, 971–987 (2009)

    MathSciNet  Article  Google Scholar 

  12. 12.

    Holschneider, M.: Inverse Radon transforms through inverse wavelet transforms. Inverse Probl. 7, 853–861 (1991)

    MathSciNet  Article  Google Scholar 

  13. 13.

    Kak, A.C., Slaney, M.: Principles of Computerized Tomographic Imaging. IEEE Press, New York (1988)

    Google Scholar 

  14. 14.

    Li, Y.: Sampling approximation by framelets in Sobolev space and its application in modifying interpolating error. J. Approx. Theory 175, 43–63 (2013)

    MathSciNet  Article  Google Scholar 

  15. 15.

    Li, Y., Chen, Q., Qian, T., Wang, Y.: Sampling error analysis and some properties of non-bandlimited signals that are reconstructed by generalized sinc functions. Appl. Anal. 93(2), 305–315 (2014)

    MathSciNet  Article  Google Scholar 

  16. 16.

    Li, R., Liu, B., Liu, R., Zhang, Q.: Nonuniform sampling in principal shift-invariant subspaces of mixed Lebesgue spaces \(L^{p, q}({\mathbb{R}}^{d+1})\). J. Math. Anal. Appl. 453, 928–941 (2017)

    MathSciNet  Article  Google Scholar 

  17. 17.

    Liu, Y., Kou, K., Ho, I.: New sampling formulae for non-bandlimited signals associated with linear canonical transform and nonlinear Fourier atoms. Signal Process. 90, 933–945 (2010)

    Article  Google Scholar 

  18. 18.

    Shannon, C.: A mathematical theory of communication. Bell Syst. Tech. J. 27, 379–423 (1948)

    MathSciNet  Article  Google Scholar 

  19. 19.

    Song, Z., Sun, W., Zhou, X., Hou, Z.: An average sampling theorem for bandlimited stochastic processes. IEEE Trans. Inf. Theory 53(12), 4798–4800 (2007)

    MathSciNet  Article  Google Scholar 

  20. 20.

    Song, Z., Liu, B., Pang, Y., Hou, C.: An improved Nyquist-Shannon irregular sampling theorem from local averages. IEEE Trans. Inf. Theory 58(9), 6093–6100 (2012)

    MathSciNet  Article  Google Scholar 

  21. 21.

    Sun, W., Zhou, X.: Reconstruction of band-limited signals from local averages. IEEE Trans. Inf. Theory 48, 2955–2963 (2001)

    MathSciNet  MATH  Google Scholar 

Download references

Acknowledgements

The authors would like to thank the reviewers for their valuable suggestions which improve the presentation of the paper. Youfa Li is supported by Natural Science Foundation of China (Nos: 61961003, 61561006, 11501132), Natural Science Foundation of Guangxi (No: 2019GXNSFAA185035) and the talent project of Education Department of Guangxi Government for one thousand Young-Middle-Aged backbone teachers.

Author information

Affiliations

Authors

Corresponding author

Correspondence to Yanfen Huang.

Additional information

Communicated by Deguang Han.

Appendix

Appendix

Proof of Lemma 2.2

By the definition of \(\{\xi _{k}\}_{k=1}^{\lfloor \frac{2\pi }{\mu _{0}}\rfloor +1}\), we have \(-\pi =\xi _{1}<\xi _{2}<\cdots < \xi _{\lfloor \frac{2\pi }{\mu _{0}}\rfloor +1}=\xi _{1}+\lfloor \frac{2\pi }{\mu _{0}}\rfloor \mu _{0}\le \pi\) and \(\lfloor \frac{2\pi }{\mu _{0}}\rfloor \ge s\). Furthermore, if \(\lfloor \frac{2\pi }{\mu _{0}}\rfloor =\frac{2\pi }{\mu _{0}}\), then

$$\begin{aligned} \bigcup _{k=1}^{\lfloor \frac{2\pi }{\mu _{0}}\rfloor }[\xi _{k},\xi _{k+1})=[-\pi ,\pi ). \end{aligned}$$
(4.1)

And if \(\lfloor \frac{2\pi }{\mu _{0}}\rfloor <\frac{2\pi }{\mu _{0}}\), then

$$\begin{aligned} \bigcup _{k=1}^{\lfloor \frac{2\pi }{\mu _{0}}\rfloor }[\xi _{k},\xi _{k+1})\bigcup \left[ \xi _{\lfloor \frac{2\pi }{\mu _{0}}\rfloor +1},\pi \right) =[-\pi ,\pi ). \end{aligned}$$
(4.2)

We next prove that for any \(i\in \{1,2, \ldots , s\}\), there exists at least a point \(\xi _{k_{i}}\in [t_{i},t_{i+1})\) where \(k_{i}\in \{1, \ldots , \lfloor \frac{2\pi }{\mu _{0}}\rfloor +1\}\). This statement will be proved for the two cases (4.1) and (4.2), respectively.

Case (4.1): For the interval \([t_{s},t_{s+1})\), it follows from (2.3) and (4.1) that \(\xi _{\lfloor \frac{2\pi }{\mu _{0}}\rfloor +1}= t_{s+1}=\pi\) which together with

$$\begin{aligned} t_{s+1}-t_{s}\ge \mu _{0}=\xi _{\lfloor \frac{2\pi }{\mu _{0}}\rfloor +1}-\xi _{\lfloor \frac{2\pi }{\mu _{0}}\rfloor }\end{aligned}$$
(4.3)

leads to \(\xi _{\lfloor \frac{2\pi }{\mu _{0}}\rfloor }\ge t_{s}\). Then

$$\begin{aligned} \xi _{\lfloor \frac{2\pi }{\mu _{0}}\rfloor }\in [t_{s},t_{s+1}). \end{aligned}$$
(4.4)

For \(i=1,\) it is clear that

$$\begin{aligned} \xi _{1}=-\pi =t_{1}\in [t_{1},t_{2}). \end{aligned}$$
(4.5)

Allowing for (4.4) and (4.5), we just need to prove the statement for the intervals \([t_{i},t_{i+1}),i=2,\ldots ,s-1\) where \(s>2.\) Or else, if there exists \(i_{0}\in \{2, \ldots , s-1\}\) such that \(\xi _{k}\notin [t_{i_{0}},t_{i_{0}+1})\) for any \(k\in \{1, \ldots , \lfloor \frac{2\pi }{\mu _{0}}\rfloor \}\), then by (2.3) and (4.1) we have

$$\begin{aligned} \{\xi _{k}\}_{k=1}^{\lfloor \frac{2\pi }{\mu _{0}}\rfloor }\subset [-\pi ,\pi )\setminus [t_{i_{0}},t_{i_{0}+1})= \bigcup _{i=1}^{s}[t_{i},t_{i+1})\setminus [t_{i_{0}},t_{i_{0}+1}).\end{aligned}$$
(4.6)

It follows from (4.6) that there exists \(k\in \{1, \ldots , \lfloor \frac{2\pi }{\mu _{0}}\rfloor -1\}\) such that \(\xi _{k+1}-\xi _{k}>t_{i_{0}+1}-t_{i_{0}}\ge \mu _{0}\). This contradicts with \(\xi _{k+1}-\xi _{k}=\mu _{0}\). Therefore, the statement holds for the case (4.1).

Case (4.2): For the interval \([t_{s},t_{s+1})\), it follows from \(\xi _{\lfloor \frac{2\pi }{\mu _{0}}\rfloor +1}=\xi _{1}+ \lfloor \frac{2\pi }{\mu _{0}}\rfloor \mu _{0}\) and \(\xi _{1}=-\pi\) that

$$\begin{aligned} \pi -\xi _{\lfloor \frac{2\pi }{\mu _{0}}\rfloor +1}=2\pi - \left\lfloor \frac{2\pi }{\mu _{0}} \right\rfloor \mu _{0}= \left( \frac{2\pi }{\mu _{0}}- \left\lfloor \frac{2\pi }{\mu _{0}} \right\rfloor \right) \mu _{0}< \mu _{0},\end{aligned}$$
(4.7)

which together with \(t_{s+1}=\pi\) and \(t_{s+1}-t_{s}\ge \mu _{0}\) leads to \(\xi _{\lfloor \frac{2\pi }{\mu _{0}}\rfloor +1}>t_{s}\). Then

$$\begin{aligned} \xi _{\lfloor \frac{2\pi }{\mu _{0}}\rfloor +1}\in [t_{s},t_{s+1}).\end{aligned}$$
(4.8)

For \(i=1\), it is easy to prove that (4.5) still holds. By (4.5) and (4.8), we just need to prove the statement for the intervals \([t_{i},t_{i+1}),i=2,\ldots ,s-1\) where \(s>2.\) Or else, if there exists \(i_{0}\in \{2,\ldots ,s-1\}\) such that \(\xi _{k}\notin [t_{i_{0}},t_{i_{0}+1})\) for any \(k\in \{1, \ldots , \lfloor \frac{2\pi }{\mu _{0}}\rfloor +1\}\), then it follows from (2.3) and (4.2) that

$$\begin{aligned} \{\xi _{k}\}_{k=1}^{\lfloor \frac{2\pi }{\mu _{0}}\rfloor +1}\subset [-\pi ,\pi )\setminus [t_{i_{0}},t_{i_{0}+1})= \bigcup _{i=1}^{s}[t_{i},t_{i+1})\setminus [t_{i_{0}},t_{i_{0}+1}). \end{aligned}$$
(4.9)

Now by (4.9), there exists \(k\in \{1, \ldots , \lfloor \frac{2\pi }{\mu _{0}}\rfloor \}\) such that \(\xi _{k+1}-\xi _{k}>t_{i_{0}+1}-t_{i_{0}}\ge \mu _{0}\). This contradicts with \(\xi _{k+1}-\xi _{k}=\mu _{0}\). Summarizing what has addressed above, for any \(i\in \{1,2, \ldots , s\}\) there exists at least a point \(\xi _{k_{i}}\in [t_{i},t_{i+1})\), where \(k_{i}\in \{1, \ldots , \lfloor \frac{2\pi }{\mu _{0}}\rfloor +1\}\).

Next we consider how to determine \(\{d_{i}\}_{i=1}^{s}\). For any \(i\in \{1,\ldots , s\}\), by the above statement there exists \(\xi _{k_{i}}\in [t_{i},t_{i+1})\) where \(k_{i}\in \{1, \ldots , \lfloor \frac{2\pi }{\mu _{0}}\rfloor +1\}\). Then it follows from (2.5) that \(\xi _{k_{i}}+ 2m_{k_{i}}\pi \in {\mathcal {C}}^{m_{k_{i}}}_{i}\) with \(m_{k_{i}}\in {\mathbb {Z}}\), which together with (2.9) leads to

$$\begin{aligned} {\widehat{f}}(\xi _{k_{i}}+ 2m_{k_{i}}\pi )=d_{i}\widehat{\hbox {sinc}_{a}}(\xi _{k_{i}}+ 2m_{k_{i}}\pi ).\end{aligned}$$
(4.10)

Based on the observed samples \(\{{\widehat{f}}(\xi _{k}+2m_{k}\pi )\}^{\lfloor \frac{2\pi }{\mu _{0}}\rfloor +1}_{k=1}\), define \(\{ {\widetilde{d}}_{k}\}^{\lfloor \frac{2\pi }{\mu _{0}}\rfloor +1}_{k=1}\) by

$$\begin{aligned} {\widetilde{d}}_{k}:=\frac{{\widehat{f}}(\xi _{k}+2m_{k}\pi )}{\widehat{\hbox {sinc}_{a}}(\xi _{k}+2m_{k}\pi )}, \end{aligned}$$
(4.11)

where \(m_{k}\in {\mathbb {Z}}\). Obviously, by (4.10) and (4.11) we have

$$\begin{aligned} \{d_{i}\}^{s}_{i=1}\subseteq \{ {\widetilde{d}}_{k}\}^{ \big \lfloor \frac{2\pi }{\mu _{0}}\big \rfloor +1}_{k=1}. \end{aligned}$$
(4.12)

It follows from \(\xi _{1}\in [t_{1},t_{2})\) that \(d_{1}={\widetilde{d}}_{1}\). Note that \(d_{i}\ne d_{i+1}\). Then \(\{d_{i}\}^{s}_{i=2}\) can be obtained iteratively by

$$\begin{aligned} d_{i}={\widetilde{d}}_{\vartheta (i)}, \end{aligned}$$
(4.13)

where \(\vartheta (i)=\min \{\vartheta (i-1)< k\le (\lfloor \frac{2\pi }{\mu _{0}}\rfloor +1) \ \hbox {s.t.} \ {\widetilde{d}}_{k}\ne d_{i-1} \}\) with \(\vartheta (1)=1\).

Proof of Proposition 2.3

We first prove that \(\{\eta _{i}\}^{\lfloor \frac{2\pi }{\mu _{0}}\rfloor \mathbf{n} +1}_{i=1}\) increases as i does. Direct observation on (2.12) (A1-A5) gives us that we just need to prove \(\xi _{k}+2\varepsilon (\mathbf{n} -1)<\xi _{k+1}\) for \(k=1,2,\ldots ,\lfloor \frac{2\pi }{\mu _{0}}\rfloor\). It follows from \(\mathbf{n} =\lceil \frac{\mu _{0}}{2\varepsilon }\rceil\) and \(\xi _{k+1}=\xi _{k}+\mu _{0}\) that \(\xi _{k+1}-\big (\xi _{k}+2\varepsilon (\mathbf{n} -1)\big )=\mu _{0}-2\varepsilon (\lceil \frac{\mu _{0}}{2\varepsilon }\rceil -1) =(\frac{\mu _{0}}{2\varepsilon }-\lceil \frac{\mu _{0}}{2\varepsilon }\rceil )2\varepsilon +2\varepsilon\). Note that \(-1<\frac{\mu _{0}}{2\varepsilon }-\lceil \frac{\mu _{0}}{2\varepsilon }\rceil \le 0\). Then

$$\begin{aligned} 0<\xi _{k+1}-\big (\xi _{k}+2\varepsilon (\mathbf{n} -1)\big )\le 2\varepsilon .\end{aligned}$$
(4.14)

That is, \(\xi _{k}+2\varepsilon (\mathbf{n} -1)<\xi _{k+1}\). And \(\{\eta _{i}\}^{\lfloor \frac{2\pi }{\mu _{0}}\rfloor \mathbf{n} +1}_{i=1}\) increases as i does.

We next prove that \(|t_{k}-{\widetilde{t}}_{k}|\le \varepsilon\) for \(k=2,\ldots ,s\), where \({\widetilde{t}}_{k}\) is defined by (2.14).

Note that \(\{\eta _{i}\}^{\lfloor \frac{2\pi }{\mu _{0}}\rfloor \mathbf{n} +1}_{i=1}\supseteq \{\xi _{k}\}_{k=1}^{\lfloor \frac{2\pi }{\mu _{0}}\rfloor +1}\). Then it is easy to follow from Lemma 2.2, (2.13) and (4.12) that \(\{d_{j}\}^{s}_{j=1}\subseteq \{ {\widehat{d}}_{i}\}^{\lfloor \frac{2\pi }{\mu _{0}}\rfloor \mathbf{n} +1}_{i=1}\).

Specifically, by the similar procedure for Lemma 2.2 (4.13), \(\{d_{j}\}_{j=1}^{s}\) can be determined iteratively by

$$\begin{aligned}d_{1}={\widehat{d}}_{1}, \ d_{k}={\widehat{d}}_{\tau (k)}, k=2,\ldots ,s, \end{aligned}$$

where

$$\begin{aligned} \tau (k)=\min \{\tau (k-1)< i\le \left( \left\lfloor \frac{2\pi }{\mu _{0}} \right\rfloor \mathbf{n} +1 \right) \ \hbox {s.t.} \ {\widehat{d}}_{i}\ne d_{k-1} \} \ \hbox {with} \ \tau (1)=1. \end{aligned}$$
(4.15)

Direct observation on (2.12) (A1–A5) and (4.14) gives us that for any \(k\in \{1, \ldots , \lfloor \frac{2\pi }{\mu _{0}}\rfloor \}\), we have

$$\begin{aligned} \eta _{k\mathbf{n} +1}-\eta _{k\mathbf{n} }=\xi _{k+1}-\big (\xi _{k}+2\varepsilon (\mathbf{n} -1)\big )\le 2\varepsilon .\end{aligned}$$
(4.16)

And for \(i=(k-1)\mathbf{n} +1,\ldots ,k\mathbf{n} -1\), we have

$$\begin{aligned} \eta _{i+1}-\eta _{i}=2\varepsilon .\end{aligned}$$
(4.17)

Now for \(\varepsilon \in (0,\frac{\mu _{0}}{2}]\), (4.16) and (4.17) lead to

$$\begin{aligned} \eta _{i+1}-\eta _{i}\le 2\varepsilon \le \mu _{0}, \ i=1,2,\ldots , \left\lfloor \frac{2\pi }{\mu _{0}}\right\rfloor \mathbf{n} .\end{aligned}$$
(4.18)

It follows from (4.15) that \(d_{1}={\widehat{d}}_{1}=\cdots ={\widehat{d}}_{\tau (2)-1}\), which together with \(\eta _{1}=\xi _{1}\in [t_{1},t_{2})\) and (4.18) leads to \(\eta _{\tau (2)-1}\in [t_{1},t_{2})\).

On the other hand, note that \(\eta _{\tau (2)}-\eta _{\tau (2)-1}\le \mu _{0}\) and \({\widehat{d}}_{\tau (2)}=d_{2}\ne d_{1}\). Then using (4.15) again we have \(\eta _{\tau (2)}\in [t_{2},t_{3})\). For \(k=2,\ldots ,s\), we can also prove that \(\eta _{\tau (k)-1}\in [t_{k-1},t_{k})\) and \(\eta _{\tau (k)}\in [t_{k},t_{k+1})\). Consequently,

$$\begin{aligned} t_{k}\in (\eta _{\tau (k)-1},\eta _{\tau (k)}], \ k=2,\ldots ,s. \end{aligned}$$
(4.19)

For \({\widetilde{t}}_{k}\) defined by (2.14), we next prove that \(|t_{k}-{\widetilde{t}}_{k}|\le \varepsilon\). We first prove \({\widetilde{t}}_{k}\in (-\pi ,\pi )\). By the definition of \(\{\xi _{k}\}_{k=1}^{\lfloor \frac{2\pi }{\mu _{0}}\rfloor +1}\), we have \(\xi _{1}=-\pi\) and \(\xi _{\lfloor \frac{2\pi }{\mu _{0}}\rfloor +1}=\xi _{1}+\lfloor \frac{2\pi }{\mu _{0}}\rfloor \mu _{0}\le \pi\). Recall that \(\{\eta _{i}\}^{\lfloor \frac{2\pi }{\mu _{0}}\rfloor \mathbf{n} +1}_{i=1}\) increases as i does. Then by (2.12) we have

$$\begin{aligned} -\pi \le \eta _{i}\le \pi , i=1,2,\ldots , \left\lfloor \frac{2\pi }{\mu _{0}} \right\rfloor \mathbf{n} +1. \end{aligned}$$
(4.20)

It follows from (2.14) and (4.20) that \({\widetilde{t}}_{k}\in (-\pi ,\pi )\). Now by (4.19), it is easy to prove that \(|t_{k}-{\widetilde{t}}_{k}|\le \frac{\eta _{\tau (k)}-\eta _{\tau (k)-1}}{2}\). Consequently, the proof can be concluded by (4.18).

Proof of Theorem 2.5

Some lemmas for proving Theorem 2.5

Lemma 4.1

Let \(\{\eta _{i}\}^{\lfloor \frac{2\pi }{\mu _{0}}\rfloor \mathbf{n} +1}_{i=1}\) be defined as Proposition 2.3 (2.12), where \(\mu _{0}=\min \{t_{j+1}-t_{j}: j=1,\ldots ,s\}\) with \(\{t_{j}\}_{j=1}^{s+1}\) satisfying (2.3) and \(\mathbf{n} =\lceil \frac{\mu _{0}}{2\varepsilon }\rceil\) with \(\varepsilon \in (0,\frac{\mu _{0}}{2}]\). If \(\{{\widetilde{t}}_{k}\}_{k=2}^{s}\) are constructed via (2.14) such that \(|t_{k}-{\widetilde{t}}_{k}|\le \varepsilon\), then

$$\begin{aligned} {\widetilde{t}}_{k+1}-{\widetilde{t}}_{k}\ge 2\varepsilon \ \hbox {for} \ k=2,\ldots ,s-1.\end{aligned}$$
(4.21)

Proof

For any \(k\in \{2,\ldots ,s-1\}\), it follows from (2.14) that

$$\begin{aligned} {\widetilde{t}}_{k+1}-{\widetilde{t}}_{k}=\frac{\eta _{\tau (k+1)}+\eta _{\tau (k+1)-1}-\eta _{\tau (k)}-\eta _{\tau (k)-1}}{2}. \end{aligned}$$
(4.22)

Now by Proposition 2.3, \(\{\eta _{i}\}^{\lfloor \frac{2\pi }{\mu _{0}}\rfloor \mathbf{n} +1}_{i=1}\) increases as i does which together with (4.15) leads to that

\(\eta _{\tau (k+1)-1}\ge \eta _{\tau (k)}\). Invoking (4.22), we have

$$\begin{aligned} {\widetilde{t}}_{k+1}-{\widetilde{t}}_{k}\ge \frac{\eta _{\tau (k+1)}-\eta _{\tau (k)-1}}{2}.\end{aligned}$$
(4.23)

We next prove that \({\widetilde{t}}_{k+1}-{\widetilde{t}}_{k}\ge 2\varepsilon\) for the three cases: \(\varepsilon =\frac{\mu _{0}}{2}\), \(\varepsilon \in (\frac{\mu _{0}}{4},\frac{\mu _{0}}{2})\) and \(\varepsilon \in (0,\frac{\mu _{0}}{4}]\).

If \(\varepsilon =\frac{\mu _{0}}{2}\) then \(\mathbf{n} =\lceil \frac{\mu _{0}}{2\varepsilon }\rceil =1\), which together with (2.12) leads to \(\{\eta _{i}\}^{\lfloor \frac{2\pi }{\mu _{0}}\rfloor \mathbf{n} +1}_{i=1}=\{\xi _{i}\}_{i=1}^{\lfloor \frac{2\pi }{\mu _{0}}\rfloor +1}\).

Note that \(\xi _{i+1}-\xi _{i}=\mu _{0}\) for \(i=1,\ldots ,\lfloor \frac{2\pi }{\mu _{0}}\rfloor\). Then by (4.15) we have

$$\begin{aligned} \eta _{\tau (k+1)}-\eta _{\tau (k)-1}\ge 2\mu _{0}=4\varepsilon . \end{aligned}$$
(4.24)

It follows from (4.23) and (4.24) that \({\widetilde{t}}_{k+1}-{\widetilde{t}}_{k}\ge 2\varepsilon .\)

If \(\varepsilon \in (\frac{\mu _{0}}{4},\frac{\mu _{0}}{2})\) then \(\mathbf{n} =\lceil \frac{\mu _{0}}{2\varepsilon }\rceil =2\). Allowing for \(\mathbf{n} =2\) and (2.12), we have

$$\begin{aligned} \{\eta _{i}\}^{\lfloor \frac{2\pi }{\mu _{0}}\rfloor \mathbf{n} +1}_{i=1}= \left\{ \xi _{1},\xi _{1}+2\varepsilon ,\xi _{2},\xi _{2}+2\varepsilon ,\ldots , \xi _{\big \lfloor \frac{2\pi }{\mu _{0}}\big \rfloor }+2\varepsilon ,\xi _{\big \lfloor \frac{2\pi }{\mu _{0}} \big \rfloor +1} \right\} . \end{aligned}$$
(4.25)

And by (4.19), we have

$$\begin{aligned} \eta _{\tau (k+1)}-\eta _{\tau (k)-1}>t_{k+1}-t_{k}\ge \mu _{0}. \end{aligned}$$
(4.26)

That is, \(\eta _{\tau (k+1)}-\eta _{\tau (k)-1}>\mu _{0}\).

Consequently, it follows from (4.25), (4.26) and \(\xi _{i+1}-\xi _{i}=\mu _{0}\) for \(i=1,\ldots ,\lfloor \frac{2\pi }{\mu _{0}}\rfloor\) that

$$\begin{aligned} \eta _{\tau (k+1)}-\eta _{\tau (k)-1}\ge \mu _{0}+2\varepsilon >4\varepsilon .\end{aligned}$$
(4.27)

Then (4.23) and (4.27) lead to \({\widetilde{t}}_{k+1}-{\widetilde{t}}_{k}>2\varepsilon .\)

If \(\varepsilon \in (0,\frac{\mu _{0}}{4}]\) then it follows from \(|t_{k}-{\widetilde{t}}_{k}|\le \varepsilon\), \(|t_{k+1}-{\widetilde{t}}_{k+1}|\le \varepsilon\) and \(t_{k+1}-t_{k}\ge \mu _{0}\) that

$$\begin{aligned}{\widetilde{t}}_{k+1}-{\widetilde{t}}_{k}\ge t_{k+1}-t_{k}-2\varepsilon \ge 2\varepsilon . \end{aligned}$$

The proof is concluded. \(\square\)

Lemma 4.2

Suppose that \(f\in V(\hbox {sinc}_{a})\) has the spline spectra such that (1.15) holds with \(d_{k}\ne d_{k+1}\) and the spectra knots \(t_{i}, i=1,\ldots ,s+1\) satisfying (2.3). Denote \({\widetilde{t}}_{1}:=-\pi\) and \({\widetilde{t}}_{s+1}:=\pi\). Let \(\varepsilon\) (being sufficiently small) and \(\{{\widetilde{t}}_{k}\}^{s}_{k=2}\) (being the approximation to \(\{t_{k}\}^{s}_{k=2}\)) be as in Proposition 2.3. Define \({\widetilde{f}}\) via (2.6) with \(\{t_{i}\}^{s+1}_{i=1}\) being replaced by \(\{{\widetilde{t}}_{i}\}^{s+1}_{i=1}\). Then for any fixed \(k\in \{2,\ldots ,s\}\), we have

$$\begin{aligned}&\displaystyle \sum _{m\in {\mathbb {Z}}\setminus \{0\}}\Vert {\widehat{f}}-\widehat{{\widetilde{f}}}\Vert _{L^{2}([{\widetilde{t}}_{k}-\varepsilon +2m\pi ,{\widetilde{t}}_{k}+\varepsilon +2m\pi ))}^{2}\\&\quad \le 2\varepsilon (1+a)^{2}\sum _{m\in {\mathbb {Z}}\setminus \{0\}}a^{4|m|-2}(|d_{k-1}|+|d_{k}|)^{2}. \end{aligned}$$

Proof

We first prove that

$$\begin{aligned}{}[{\widetilde{t}}_{k}-\varepsilon ,{\widetilde{t}}_{k}+\varepsilon )\subseteq [t_{k-1},t_{k+1})\end{aligned}$$
(4.28)

and

$$\begin{aligned}{}[{\widetilde{t}}_{k}-\varepsilon ,{\widetilde{t}}_{k}+\varepsilon )\subseteq [{\widetilde{t}}_{k-1},{\widetilde{t}}_{k+1})\end{aligned}$$
(4.29)

for \(k=2,\ldots ,s.\) Recall in Proposition 2.3 that \(\varepsilon \in (0,\frac{\mu _{0}}{2}]\) with

$$\begin{aligned} \mu _{0}=\min \{t_{i+1}-t_{i}: i=1,\ldots ,s\},\end{aligned}$$
(4.30)

and \(\{{\widetilde{t}}_{k}\}_{k=2}^{s}\) are constructed via (2.14) such that \(|t_{k}-{\widetilde{t}}_{k}|\le \varepsilon \ \hbox {for}\ k=2,\ldots ,s.\) Clearly,

$$\begin{aligned} {\widetilde{t}}_{k}-\varepsilon \le t_{k}\le {\widetilde{t}}_{k}+\varepsilon , \ \hbox {for}\ k=2,\ldots ,s.\end{aligned}$$
(4.31)

It follows from (4.30) that

$$\begin{aligned} t_{i+1}-t_{i}\ge \mu _{0}\ge 2\varepsilon \ \hbox {for}\ i=1,\ldots ,s.\end{aligned}$$
(4.32)

Now by (4.31) and (4.32), it is easy to check that \(t_{k-1}\le {\widetilde{t}}_{k}-\varepsilon\) and \({\widetilde{t}}_{k}+\varepsilon \le t_{k+1}\). Then (4.28) holds. We next prove that (4.29) holds. Recall that \(\{{\widetilde{t}}_{k}\}_{k=2}^{s}\) are constructed via (2.14) such that \(|t_{k}-{\widetilde{t}}_{k}|\le \varepsilon\). Then by Lemma 4.1 (4.21), we have

$$\begin{aligned} {\widetilde{t}}_{i}<{\widetilde{t}}_{i}+\varepsilon \le {\widetilde{t}}_{i+1}-\varepsilon <{\widetilde{t}}_{i+1},\ i=2,\ldots ,s-1, \end{aligned}$$
(4.33)

where \(s>2\). For \(k=2\), by \({\widetilde{t}}_{1}=-\pi =t_{1}\), \(|t_{2}-{\widetilde{t}}_{2}|\le \varepsilon\) and \(t_{2}-t_{1}\ge 2\varepsilon\) we have

$$\begin{aligned} {\widetilde{t}}_{2}+\varepsilon -{\widetilde{t}}_{1}\ge t_{2}-t_{1}\ge 2\varepsilon .\end{aligned}$$
(4.34)

Then \({\widetilde{t}}_{1}\le {\widetilde{t}}_{2}-\varepsilon\). Now (4.33) and (4.34) with \(i=2\) lead to \([{\widetilde{t}}_{2}-\varepsilon ,{\widetilde{t}}_{2}+\varepsilon )\subseteq [{\widetilde{t}}_{1},{\widetilde{t}}_{3})\). For \(k=s\), by \({\widetilde{t}}_{s+1}=\pi =t_{s+1}\), \(|t_{s}-{\widetilde{t}}_{s}|\le \varepsilon\) and \(t_{s+1}-t_{s}\ge 2\varepsilon\) we have

$$\begin{aligned} {\widetilde{t}}_{s+1}-({\widetilde{t}}_{s}-\varepsilon ) \ge t_{s+1}-t_{s}\ge 2\varepsilon .\end{aligned}$$
(4.35)

Namely, \({\widetilde{t}}_{s}+\varepsilon \le {\widetilde{t}}_{s+1}\). Then (4.33) and (4.35) with \(i=s-1\) lead to \([{\widetilde{t}}_{s}-\varepsilon ,{\widetilde{t}}_{s}+\varepsilon )\subseteq [{\widetilde{t}}_{s-1},{\widetilde{t}}_{s+1})\). On the other hand, the direct observation on (4.33) with \(i=2,\ldots ,s-1\) gives us that (4.29) holds for \(k=3,\ldots ,s-1\). Summarizing what has been addressed above, (4.29) holds.

We next estimate \(\sum _{m\in {\mathbb {Z}}\setminus \{0\}}\Vert {\widehat{f}}-\widehat{{\widetilde{f}}}\Vert _{L^{2}([{\widetilde{t}}_{k}-\varepsilon +2m\pi ,{\widetilde{t}}_{k}+\varepsilon +2m\pi ))}^{2}\) for the two cases: \({\widetilde{t}}_{k}\le t_{k}\) and \({\widetilde{t}}_{k}> t_{k}\). By (1.15) and the definition of \({\widetilde{f}}\), for any \(\xi \in {\mathbb {R}}\) we have

$$\begin{aligned} \left\{ \begin{array}{llllll} {\widehat{f}}(\xi )=\left( \sum _{l\in {\mathbb {Z}}}\sum _{k=1}^{s}d_{k}\chi _{[t_{k}+2l\pi ,t_{k+1}+2l\pi )}(\xi )\right) \widehat{\hbox {sinc}_{a}}(\xi ),\\ \widehat{{\widetilde{f}}}(\xi )=\left( \sum _{l\in {\mathbb {Z}}}\sum _{k=1}^{s}d_{k}\chi _{[{\widetilde{t}}_{k}+2l\pi ,{\widetilde{t}}_{k+1}+2l\pi )}(\xi )\right) \widehat{\hbox {sinc}_{a}}(\xi ). \end{array}\right. \end{aligned}$$
(4.36)

If \({\widetilde{t}}_{k}\le t_{k}\), then it follows from (4.28) and the first equation in (4.36) that

$$\begin{aligned} \begin{array}{lll} {\widehat{f}}\chi _{[{\widetilde{t}}_{k}-\varepsilon +2m\pi ,{\widetilde{t}}_{k}+\varepsilon +2m\pi )}(\xi ) &{}=&{}{\widehat{f}}\chi _{[{\widetilde{t}}_{k}-\varepsilon +2m\pi ,{\widetilde{t}}_{k}+2m\pi )}(\xi ) +{\widehat{f}}\chi _{[{\widetilde{t}}_{k}+2m\pi ,t_{k}+2m\pi )}(\xi )\\ &{} \ &{}+{\widehat{f}}\chi _{[t_{k}+2m\pi ,{\widetilde{t}}_{k}+\varepsilon +2m\pi )}(\xi )\\ &{}=&{}d_{k-1}\widehat{\hbox {sinc}_{a}}(\xi )\chi _{[{\widetilde{t}}_{k}-\varepsilon +2m\pi ,{\widetilde{t}}_{k}+2m\pi )}(\xi )\\ &{}\ &{}+d_{k-1}\widehat{\hbox {sinc}_{a}}(\xi )\chi _{[{\widetilde{t}}_{k}+2m\pi ,t_{k}+2m\pi )}(\xi )\\ &{} \ &{}+d_{k}\widehat{\hbox {sinc}_{a}}(\xi )\chi _{[t_{k}+2m\pi ,{\widetilde{t}}_{k}+\varepsilon +2m\pi )}(\xi ). \end{array}\end{aligned}$$
(4.37)

And by (4.29) and the second equation in (4.36), we have

$$\begin{aligned} \begin{array}{lll} \widehat{{\widetilde{f}}}\chi _{[{\widetilde{t}}_{k}-\varepsilon +2m\pi ,{\widetilde{t}}_{k}+\varepsilon +2m\pi )}(\xi ) &{}=&{} d_{k-1}\widehat{\hbox {sinc}_{a}}(\xi )\chi _{[{\widetilde{t}}_{k}-\varepsilon +2m\pi ,{\widetilde{t}}_{k}+2m\pi )}(\xi )\\ &{}\ &{}+d_{k}\widehat{\hbox {sinc}_{a}}(\xi )\chi _{[{\widetilde{t}}_{k}+2m\pi ,t_{k}+2m\pi )}(\xi )\\ &{}\ &{}+d_{k}\widehat{\hbox {sinc}_{a}}(\xi )\chi _{[t_{k}+2m\pi ,{\widetilde{t}}_{k}+\varepsilon +2m\pi )}(\xi ). \end{array}\end{aligned}$$
(4.38)

Correspondingly, (4.37) and (4.38) lead to

$$\begin{aligned} \begin{array}{lll} ({\widehat{f}}-\widehat{{\widetilde{f}}})\chi _{[{\widetilde{t}}_{k}-\varepsilon +2m\pi ,{\widetilde{t}}_{k}+\varepsilon +2m\pi )}(\xi ) =(d_{k-1}-d_{k})\widehat{\hbox {sinc}_{a}}(\xi )\chi _{[{\widetilde{t}}_{k}+2m\pi ,t_{k}+2m\pi )}(\xi ). \end{array}\end{aligned}$$
(4.39)

Direct observation on (4.39) gives us that we need to estimate

$$\begin{aligned}\widehat{\hbox {sinc}_{a}}(\xi )\chi _{[{\widetilde{t}}_{k}+2m\pi ,t_{k}+2m\pi )}(\xi ).\end{aligned}$$

For any \(m\in {\mathbb {Z}}\setminus \{0\}\), define

$$\begin{aligned} T^{1}_{m}:=[(2m-1)\pi ,2m\pi ), \ T^{2}_{m}:=[2m\pi ,(2m+1)\pi ). \end{aligned}$$
(4.40)

From (1.1) and (4.40) we arrive at

$$\begin{aligned} \big (T^{1}_{m}\cup T_{m}^{2}\big )\subset \big ( I_{2|m|-1}\cup I_{2|m|}\big ).\end{aligned}$$
(4.41)

For any fixed \(k\in \{2,\ldots ,s\}\), it follows from (2.14) that \({\widetilde{t}}_{k}\in (-\pi ,\pi )\), which together with (2.3) and \({\widetilde{t}}_{k}\le t_{k}\) leads to \(-\pi<{\widetilde{t}}_{k}\le t_{k}<\pi\). Then by (4.40) we have

$$\begin{aligned}{}[{\widetilde{t}}_{k}+2m\pi ,t_{k}+2m\pi )\subset [-\pi +2m\pi ,\pi +2m\pi )=T_{m}^{1}\cup T_{m}^{2}.\end{aligned}$$
(4.42)

Now combining (4.41) and (4.42) we have

$$\begin{aligned}{}[{\widetilde{t}}_{k}+2m\pi ,t_{k}+2m\pi )\subset \big (I_{2|m|-1}\cup I_{2|m|}\big ).\end{aligned}$$
(4.43)

Note that by (1.2) we have

$$\begin{aligned} \widehat{\hbox {sinc}_{a}}(\xi )=\left\{ \begin{array}{lllll}(1+a)a^{2|m|-1}, &{} \quad \xi \in I_{2|m|-1},\\ (1+a)a^{2|m|}, &{} \quad \xi \in I_{2|m|}, \end{array}\right. \end{aligned}$$
(4.44)

where \(a\in (-1,1)\).

Then it follows from (4.43) and (4.44) that

$$\begin{aligned} \widehat{\hbox {sinc}_{a}}(\xi )\chi _{[{\widetilde{t}}_{k}+2m\pi ,t_{k}+2m\pi )}(\xi )\le (1+a)a^{2|m|-1}\chi _{[{\widetilde{t}}_{k}+2m\pi ,t_{k}+2m\pi )}(\xi ).\end{aligned}$$
(4.45)

Consequently, by (4.39) and (4.45) we have

$$\begin{aligned} \begin{array}{lll} &{}\displaystyle \sum _{m\in {\mathbb {Z}}\setminus \{0\}}\Vert {\widehat{f}}-\widehat{{\widetilde{f}}}\Vert _{L^{2}([{\widetilde{t}}_{k}-\varepsilon +2m\pi ,{\widetilde{t}}_{k}+\varepsilon +2m\pi ))}^{2}\\ &{} \quad \displaystyle \le \sum _{m\in {\mathbb {Z}}\setminus \{0\}}|t_{k}-{\widetilde{t}}_{k}|(1+a)^{2}a^{4|m|-2}|d_{k-1}-d_{k}|^{2}\\ &{} \quad \displaystyle \le 2\varepsilon (1+a)^{2}\sum _{m\in {\mathbb {Z}}\setminus \{0\}} a^{4|m|-2}(|d_{k-1}|+|d_{k}|)^{2}. \end{array}\end{aligned}$$
(4.46)

If \({\widetilde{t}}_{k}> t_{k}\), then by the similar procedures for \({\widetilde{t}}_{k}\le t_{k}\), we can prove that (4.46) still holds for any fixed \(k\in \{2,\ldots ,s\}\). The proof is concluded. \(\square\)

Proof of Theorem 2.5

Note that \({\widetilde{t}}_{1}=-\pi\) and \({\widetilde{t}}_{s+1}=\pi\). Then by Lemma 4.2 (4.29), we have \(-\pi ={\widetilde{t}}_{1}<{\widetilde{t}}_{2}<\ldots<{\widetilde{t}}_{s}<{\widetilde{t}}_{s+1}=\pi\). Consequently,

$$\begin{aligned}{}[-\pi ,\pi )=\bigcup _{i=1}^{s}[{\widetilde{t}}_{i},{\widetilde{t}}_{i+1}).\end{aligned}$$
(4.47)

Define \(Q_{1}:=\cup _{k=2}^{s}[{\widetilde{t}}_{k}-\varepsilon ,{\widetilde{t}}_{k}+\varepsilon )\) and

$$\begin{aligned} Q_{2}:=\left\{ \begin{array}{rcllc} \displaystyle [{\widetilde{t}}_{1},{\widetilde{t}}_{2}-\varepsilon )\cup [{\widetilde{t}}_{2}+\varepsilon ,{\widetilde{t}}_{3}), &{}&{}s=2\\ \displaystyle [{\widetilde{t}}_{1},{\widetilde{t}}_{2}-\varepsilon ) \cup [{\widetilde{t}}_{s}+\varepsilon ,{\widetilde{t}}_{s+1})\cup \{\cup _{k=2}^{s-1}[{\widetilde{t}}_{k}+\varepsilon ,{\widetilde{t}}_{k+1}-\varepsilon )\}, &{}&{} s>2, \end{array}\right. \end{aligned}$$
(4.48)

where \(\varepsilon \in (0,\frac{\mu _{0}}{2}]\). Incidentally, for (4.33), (4.34) and (4.35), \(Q_{2}\) is well defined. By (4.47), it is easy to check that

$$\begin{aligned} \begin{array}{lll} [-\pi ,\pi ) = Q_{1}\cup Q_{2}. \end{array} \end{aligned}$$
(4.49)

Next we prove that

$$\begin{aligned} {\widehat{f}}(\xi )=\widehat{{\widetilde{f}}}(\xi ) \end{aligned}$$
(4.50)

for \(\xi \in Q_{2}+2l\pi\) with any \(l\in {\mathbb {Z}}\). For \(s\ge 2\), recall that \(\{{\widetilde{t}}_{k}\}_{k=2}^{s}\) is constructed via (2.14) such that \(|t_{k}-{\widetilde{t}}_{k}|\le \varepsilon\). Then by \(|t_{2}-{\widetilde{t}}_{2}|\le \varepsilon\) and \(|t_{s}-{\widetilde{t}}_{s}|\le \varepsilon\) we have \({\widetilde{t}}_{2}-\varepsilon \le t_{2}\) and \(t_{s}\le {\widetilde{t}}_{s}+\varepsilon\), which together with \({\widetilde{t}}_{1}=t_{1}\) and \({\widetilde{t}}_{s+1}=t_{s+1}\) lead to

$$\begin{aligned}{}[{\widetilde{t}}_{1},{\widetilde{t}}_{2}-\varepsilon )\subseteq [t_{1},t_{2}),\ [{\widetilde{t}}_{s}+\varepsilon ,{\widetilde{t}}_{s+1})\subseteq [t_{s},t_{s+1}).\end{aligned}$$
(4.51)

Now combining (4.51) and Lemma 4.2 (4.36) we have

$$\begin{aligned} \left\{ \begin{array}{lll} \widehat{{\widetilde{f}}}\chi _{[{\widetilde{t}}_{1}+2l\pi ,{\widetilde{t}}_{2}-\varepsilon +2l\pi )}(\xi ) &{}={\widehat{f}}\chi _{[{\widetilde{t}}_{1}+2l\pi ,{\widetilde{t}}_{2}-\varepsilon +2l\pi )}(\xi )\\ &{}=d_{1}\widehat{\hbox {sinc}_{a}}(\xi )\chi _{[{\widetilde{t}}_{1}+2l\pi ,{\widetilde{t}}_{2}-\varepsilon +2l\pi )}(\xi ),\\ \widehat{{\widetilde{f}}}\chi _{[{\widetilde{t}}_{s}+\varepsilon +2l\pi ,{\widetilde{t}}_{s+1}+2l\pi )}(\xi ) &{}={\widehat{f}}\chi _{[{\widetilde{t}}_{s}+\varepsilon +2l\pi ,{\widetilde{t}}_{s+1}+2l\pi )}(\xi )\\ &{}=d_{s}\widehat{\hbox {sinc}_{a}}(\xi )\chi _{[{\widetilde{t}}_{s}+\varepsilon +2l\pi ,{\widetilde{t}}_{s+1}+2l\pi )}(\xi ). \end{array}\right. \end{aligned}$$
(4.52)

That is, (4.50) holds for \(s=2\), or for \(s>2\) and \(\xi \in \Big ([{\widetilde{t}}_{1},{\widetilde{t}}_{2}-\varepsilon ) \cup [{\widetilde{t}}_{s}+\varepsilon ,{\widetilde{t}}_{s+1})\Big )+2l\pi\). By (4.48), to complete the proof of (4.50) we just need to prove it for \(s>2\) and

$$\begin{aligned}\xi \in \bigcup _{k=2}^{s-1}[{\widetilde{t}}_{k}+\varepsilon +2l\pi ,{\widetilde{t}}_{k+1}-\varepsilon +2l\pi ). \end{aligned}$$

For \(k\in \{2,\ldots ,s-1\}\), it is easy to derive from \(|t_{k}-{\widetilde{t}}_{k}|\le \varepsilon\) and \(|t_{k+1}-{\widetilde{t}}_{k+1}|\le \varepsilon\) that

$$\begin{aligned}{}[{\widetilde{t}}_{k}+\varepsilon ,{\widetilde{t}}_{k+1}-\varepsilon )\subseteq [t_k, t_{k+1}). \end{aligned}$$
(4.53)

By (4.36), the restriction of \({\widehat{f}}\) on the interval \([t_k, t_{k+1})+2l\pi\) is

$$\begin{aligned}{\widehat{f}}\chi _{[t_{k}+2l\pi ,t_{k+1}+2l\pi )}(\xi ) =d_{k}\widehat{\hbox {sinc}_{a}}(\xi )\chi _{[t_{k}+2l\pi ,t_{k+1}+2l\pi )}(\xi ),\end{aligned}$$

which together with (4.53) leads to

$$\begin{aligned} {\widehat{f}}\chi _{[{\widetilde{t}}_{k}+\varepsilon +2l\pi ,{\widetilde{t}}_{k+1}-\varepsilon +2l\pi )}(\xi ) =d_{k}\widehat{\hbox {sinc}_{a}}(\xi )\chi _{[{\widetilde{t}}_{k}+\varepsilon +2l\pi ,{\widetilde{t}}_{k+1}-\varepsilon +2l\pi )}(\xi ). \end{aligned}$$
(4.54)

Using (4.36) again, the restriction of \(\widehat{{\widetilde{f}}}\) on \([{\widetilde{t}}_k, {\widetilde{t}}_{k+1})+2l\pi\) is

$$\begin{aligned}\widehat{{\widetilde{f}}}\chi _{[{\widetilde{t}}_{k}+2l\pi ,{\widetilde{t}}_{k+1}+2l\pi )}(\xi ) =d_{k}\widehat{\hbox {sinc}_{a}}(\xi )\chi _{[{\widetilde{t}}_{k}+2l\pi ,{\widetilde{t}}_{k+1}+2l\pi )}(\xi ),\end{aligned}$$

which together with \([{\widetilde{t}}_{k}+\varepsilon ,{\widetilde{t}}_{k+1}-\varepsilon )\subset [{\widetilde{t}}_k, {\widetilde{t}}_{k+1})\) leads to that

$$\begin{aligned} \widehat{{\widetilde{f}}}\chi _{[{\widetilde{t}}_{k}+\varepsilon +2l\pi ,{\widetilde{t}}_{k+1}-\varepsilon +2l\pi )}(\xi ) =d_{k}\widehat{\hbox {sinc}_{a}}(\xi )\chi _{[{\widetilde{t}}_{k}+\varepsilon +2l\pi ,{\widetilde{t}}_{k+1}-\varepsilon +2l\pi )}(\xi ). \end{aligned}$$
(4.55)

Then it follows from (4.54) and (4.55) that

$$\begin{aligned} \begin{array}{lll} \widehat{{\widetilde{f}}}\chi _{[{\widetilde{t}}_{k}+\varepsilon +2l\pi ,{\widetilde{t}}_{k+1}-\varepsilon +2l\pi )}(\xi ) ={\widehat{f}}\chi _{[{\widetilde{t}}_{k}+\varepsilon +2l\pi ,{\widetilde{t}}_{k+1}-\varepsilon +2l\pi )}(\xi ). \end{array}\end{aligned}$$
(4.56)

Summarizing what has been addressed above, (4.50) holds.

Next we estimate \(\Vert {\widehat{f}}-\widehat{{\widetilde{f}}}\Vert _{L^{2}({\mathbb {R}})}^{2}\). By (4.49) and (4.50), it is easy to check that

$$\begin{aligned} \begin{array}{lll} \Vert {\widehat{f}}-\widehat{{\widetilde{f}}}\Vert _{L^{2}({\mathbb {R}})}^{2} =P_{1}+P_{2}, \end{array}\end{aligned}$$
(4.57)

where

$$\begin{aligned} \left\{ \begin{array}{lll} P_{1}=\sum _{k=2}^{s}\Vert {\widehat{f}}-\widehat{{\widetilde{f}}}\Vert _{L^{2}([{\widetilde{t}}_{k}-\varepsilon ,{\widetilde{t}}_{k}+\varepsilon ))}^{2},\\ P_{2}=\sum _{m\in {\mathbb {Z}}\setminus \{0\}}\sum _{k=2}^{s}\Vert {\widehat{f}}-\widehat{{\widetilde{f}}}\Vert _{L^{2}([{\widetilde{t}}_{k}-\varepsilon +2m\pi ,{\widetilde{t}}_{k}+\varepsilon +2m\pi ))}^{2}. \end{array}\right. \end{aligned}$$
(4.58)

Note that \(\{t_{k},{\widetilde{t}}_{k}\}\subset [{\widetilde{t}}_{k}-\varepsilon ,{\widetilde{t}}_{k}+\varepsilon )\) for \(k\in \{2,\ldots ,s\}\). Motivated by this, \(P_{1}\) is estimated for the two cases: \({\widetilde{t}}_{k}\le t_{k}\) and \({\widetilde{t}}_{k}> t_{k}\). If \({\widetilde{t}}_{k}\le t_{k}\), then it follows from (4.28) and the first equation in (4.36) that

$$\begin{aligned} \begin{array}{lll} {\widehat{f}}\chi _{[{\widetilde{t}}_{k}-\varepsilon ,{\widetilde{t}}_{k}+\varepsilon )}(\xi ) &{}=&{}{\widehat{f}}\chi _{[{\widetilde{t}}_{k}-\varepsilon ,{\widetilde{t}}_{k})}(\xi )+ {\widehat{f}}\chi _{[{\widetilde{t}}_{k},t_{k})}(\xi ) +{\widehat{f}}\chi _{[t_{k},{\widetilde{t}}_{k}+\varepsilon )}(\xi )\\ &{}=&{}d_{k-1}\widehat{\hbox {sinc}_{a}}(\xi )\chi _{[{\widetilde{t}}_{k}-\varepsilon ,{\widetilde{t}}_{k})}(\xi ) +d_{k-1}\widehat{\hbox {sinc}_{a}}(\xi )\chi _{[{\widetilde{t}}_{k},t_{k})}(\xi )\\ &{}\ &{}+d_{k}\widehat{\hbox {sinc}_{a}}(\xi )\chi _{[t_{k},{\widetilde{t}}_{k}+\varepsilon )}(\xi ). \end{array}\end{aligned}$$
(4.59)

On the other hand, by (4.29) and the second equation in (4.36) we have

$$\begin{aligned} \begin{array}{lll} \widehat{{\widetilde{f}}}\chi _{[{\widetilde{t}}_{k}-\varepsilon ,{\widetilde{t}}_{k}+\varepsilon )}(\xi )&{}=d_{k-1}\widehat{\hbox {sinc}_{a}}(\xi )\chi _{[{\widetilde{t}}_{k}-\varepsilon ,{\widetilde{t}}_{k})}(\xi ) +d_{k}\widehat{\hbox {sinc}_{a}}(\xi )\chi _{[{\widetilde{t}}_{k},t_{k})}(\xi )\\ &{} \quad +d_{k}\widehat{\hbox {sinc}_{a}}(\xi )\chi _{[t_{k},{\widetilde{t}}_{k}+\varepsilon )}(\xi ), \end{array}\end{aligned}$$

which together with (4.59) leads to

$$\begin{aligned} ({\widehat{f}}-\widehat{{\widetilde{f}}})\chi _{[{\widetilde{t}}_{k}-\varepsilon ,{\widetilde{t}}_{k}+\varepsilon )}(\xi )= (d_{k-1}-d_{k})\widehat{\hbox {sinc}_{a}}(\xi )\chi _{[{\widetilde{t}}_{k},t_{k})}(\xi ). \end{aligned}$$
(4.60)

Direct observation on (4.48) and (4.49) gives us that \([{\widetilde{t}}_{k}-\varepsilon ,{\widetilde{t}}_{k}+\varepsilon )\subset [-\pi ,\pi )\). Recall that \([{\widetilde{t}}_{k},t_{k})\subset [{\widetilde{t}}_{k}-\varepsilon ,{\widetilde{t}}_{k}+\varepsilon )\). Then \([{\widetilde{t}}_{k},t_{k})\subset [-\pi ,\pi )\). Clearly, by (1.2) with \(n=0\) we have

$$\begin{aligned} \widehat{\hbox {sinc}_{a}}(\xi )\chi _{[{\widetilde{t}}_{k},t_{k})}(\xi )=(1+a)\chi _{[{\widetilde{t}}_{k},t_{k})}(\xi ).\end{aligned}$$
(4.61)

Now (4.60) and (4.61) lead to

$$\begin{aligned} \begin{array}{lll} \Vert {\widehat{f}}-\widehat{{\widetilde{f}}}\Vert _{L^{2}([{\widetilde{t}}_{k}-\varepsilon ,{\widetilde{t}}_{k}+\varepsilon ))}^{2} &{}=|t_{k}-{\widetilde{t}}_{k}|(1+a)^{2}|d_{k-1}-d_{k}|^{2}\\ &{}\le 2\varepsilon (1+a)^{2}(|d_{k-1}|+|d_{k}|)^{2}. \end{array}\end{aligned}$$

Consequently,

$$\begin{aligned} P_{1}\le 2\varepsilon (1+a)^{2}\sum _{k=2}^{s}(|d_{k-1}|+|d_{k}|)^{2}.\end{aligned}$$
(4.62)

If \({\widetilde{t}}_{k}> t_{k}\), then by the similar procedures for \({\widetilde{t}}_{k}\le t_{k}\), (4.62) still holds. For the second term \(P_{2}\), it follows from Lemma 4.2 that

$$\begin{aligned} \begin{array}{lll} P_{2}\le 2\varepsilon (1+a)^{2}\sum _{m\in {\mathbb {Z}}\setminus \{0\}}a^{4|m|-2}\sum _{k=2}^{s}(|d_{k-1}|+|d_{k}|)^{2}. \end{array}\end{aligned}$$
(4.63)

Now from (4.57), (4.62) and (4.63) we arrive at

$$\begin{aligned} \begin{array}{llllll} \Vert f-{\widetilde{f}}\Vert _{L^{2}({\mathbb {R}})}^{2}&{}=&{}\frac{1}{2\pi }\Vert {\widehat{f}}-\widehat{{\widetilde{f}}}\Vert _{L^{2}({\mathbb {R}})}^{2}\\ &{}\le &{}\displaystyle \frac{\varepsilon }{\pi }(1+a)^{2}\left( 1+\sum _{m\in {\mathbb {Z}}\setminus \{0\}}a^{4|m|-2}\right) \sum _{k=2}^{s}(|d_{k-1}|+|d_{k}|)^{2}\\ &{}=&{}\displaystyle \frac{\varepsilon }{\pi }\left\{ \frac{1+a}{1-a}+ \frac{a^{2}(1+a)^{2}}{1+a^{2}}\right\} \sum _{k=2}^{s}(|d_{k-1}|+|d_{k}|)^{2}. \end{array}\end{aligned}$$
(4.64)

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Li, Y., Huang, Y. & Zhou, C. Reconstruction of spline spectra-signals from generalized sinc function by finitely many samples. Banach J. Math. Anal. 15, 33 (2021). https://doi.org/10.1007/s43037-020-00116-4

Download citation

Keywords

  • Generalized sinc function
  • Shift-invariant space
  • Spline-spectra signals
  • Fourier samples

Mathematics Subject Classification

  • 42C40
  • 41A30
  • 42A16