On approximation properties of Kantorovich-type sampling operators Olga Orlova, Gert Tamberg 1 Department of Mathematics, Tallinn University of Technology, E STONIA Modern Time-Frequency Analysis Strobl 2014 AUSTRIA June 5, 2014 1 This is a joint work with Andi Kivinukk (Tallinn University) This research was supported by the Estonian Science Foundation, grant 9383 and by the Estonian Min. of Educ. and Research, project SF0140011s09. O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 1 / 34 Shannon Sampling Series Sampling theory The aim of this talk is to study approximation properties of a generalized Shannon sampling operators (Sw f )(t) := ∞ X k =−∞ f( k )s(wt − k ) w with a bandlimited kernel s, which is a Fourier transform of a certain even window function λ ∈ C[−1,1] , λ(0) = 1, λ(u) = 0 (|u| > 1), i.e. Z1 s(t) := s(λ; t) := λ(u) cos(πtu) du. 0 We give the conditions for the kernel s, which allow us to have the estimate of order of approximation via modulus of smoothness in form: kf − Swr f k 6 Mr ω2r (f ; O. Orlova, G. Tamberg (TUT) 1 ). w Kantorovich-type Sampling Operators Strobl 2014 2 / 34 Shannon Sampling Series Preliminary results Def The Bernstein class Bσp for σ > 0 and 1 6 p 6 ∞ consists of those bounded functions f ∈ Lp (R) which can be extended to an entire function f (z) (z ∈ C)of exponential type σ, i. e., |f (z)| 6 eσ|y | kf kC (z = x + iy ∈ C). The class Bσp is a Banach space if one takes the norm of Lp (R). Lemma We have Bσ1 ⊂ Bσp ⊂ Bσr ⊂ Bσ∞ , Bαp ⊂ Bβp , O. Orlova, G. Tamberg (TUT) 1 6 p 6 r 6 ∞, 06α6β<∞ Kantorovich-type Sampling Operators Strobl 2014 3 / 34 Shannon Sampling Series Whittaker-Kotelnikov-Shannon sampling series Whittaker-Kotelnikov-Shannon sampling theorem Theorem (Whittaker-Kotelnikov-Shannon) p If g ∈ BπW , 1 6 p < ∞, or g ∈ Bσ∞ for some 0 6 σ < πW , then g(t) = ∞ X k =−∞ g( k sinc ) sinc(Wt − k ) =: (SW g)(t), W (1) the series being uniformly convergent on each compact subset of R. p The Bernstein class BπW forms the set of fixed points of the sampling operators (1). Now it is natural to ask what happens in WKS Theorem p if instead of g ∈ BπW we consider uniformly continuous and bounded functions f ∈ C(R)? M. Theis [Theis’19] showed that the equality in (1) does not hold for any g ∈ C(R). O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 4 / 34 Shannon Sampling Series Whittaker-Kotelnikov-Shannon sampling series Whittaker-Kotelnikov-Shannon sampling theorem Theorem (Whittaker-Kotelnikov-Shannon) p If g ∈ BπW , 1 6 p < ∞, or g ∈ Bσ∞ for some 0 6 σ < πW , then g(t) = ∞ X k =−∞ g( k sinc ) sinc(Wt − k ) =: (SW g)(t), W (1) the series being uniformly convergent on each compact subset of R. p The Bernstein class BπW forms the set of fixed points of the sampling operators (1). Now it is natural to ask what happens in WKS Theorem p if instead of g ∈ BπW we consider uniformly continuous and bounded functions f ∈ C(R)? M. Theis [Theis’19] showed that the equality in (1) does not hold for any g ∈ C(R). O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 4 / 34 Shannon Sampling Series Generalized sampling series Generalized Shannon sampling series We get uniform convergence kf − Sw f kC → 0 (w → ∞) for any f ∈ C(R) if, instead of the cardinal sine, P we use different kernel 1 1 functions s ∈ L (R) ∩ C(R) (sinc 6∈ L (R)), k ∈Z s(u − k ) = 1, u ∈ R. This approach gives us the generalized sampling series for t ∈ R, w >0 ∞ X k (2) (Sw f )(t) := f ( )s(wt − k ). w k =−∞ Whereas a generalized sampling series with a particular kernel were already considered by M. Theis in 1919, a study of those series for arbitrary kernel functions was initiated at RWTH Aachen University (P. L. Butzer et al) and widely studied there since 1977 and in Perugia since 2005. O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 5 / 34 Shannon Sampling Series Generalized sampling series Generalized Shannon sampling series We get uniform convergence kf − Sw f kC → 0 (w → ∞) for any f ∈ C(R) if, instead of the cardinal sine, P we use different kernel 1 1 functions s ∈ L (R) ∩ C(R) (sinc 6∈ L (R)), k ∈Z s(u − k ) = 1, u ∈ R. This approach gives us the generalized sampling series for t ∈ R, w >0 ∞ X k (2) (Sw f )(t) := f ( )s(wt − k ). w k =−∞ Whereas a generalized sampling series with a particular kernel were already considered by M. Theis in 1919, a study of those series for arbitrary kernel functions was initiated at RWTH Aachen University (P. L. Butzer et al) and widely studied there since 1977 and in Perugia since 2005. O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 5 / 34 Shannon Sampling Series Generalized sampling series 0.4 1.0 0.2 0.5 -4 -2 2 4 -4 -2 2 -0.5 -0.2 -1.0 -0.4 4 Take a function f and compute the corresponding sampling series (S2 f )(t) := ∞ X k =−∞ k f s(2t − k ), 2 (W = 2) and the approximation error S2 f − f . O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 6 / 34 Shannon Sampling Series Generalized sampling series 0.4 1.0 0.2 0.5 -4 -2 2 -4 4 -2 2 -0.5 -0.2 -1.0 -0.4 4 Now we compute the sampling series, taking W = 6, (S6 f )(t) := ∞ X k =−∞ k f s(6t − k ) 6 and the approximation error S6 f − f . O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 6 / 34 Shannon Sampling Series Generalized sampling series 0.4 1.0 0.2 0.5 -4 -2 2 -4 4 -2 2 -0.5 -0.2 -1.0 -0.4 4 At last we take W = 16, getting (S16 f )(t) := ∞ X f k =−∞ k 16 s(16t − k ) and the approximation error S16 f − f . O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 6 / 34 Shannon Sampling Series Sampling operators in Lp case Since in practice signals are however often discontinuous, this talk is also concerned with the convergence of Sw f to f in the Lp (R)-norm for 1 6 p < ∞. The sampling series Sw f of an arbitrary Lp -function f may be divergent. Take f ∈ Lp (R), f (k /w) = 1 for a fixed w > 0, then (Sw f )(t) ≡ 1 6∈ Lp (R). Also operators Sw f do not always provide good results in applications, e.g. in Signal Processing. Generalized sampling operators depend on exact values f (k /w), whilst in practice we often have to deal with the so called "jitter errors" i.e. the impossibility to determine the exact values at the nodes. Replacing the exact value f (k /w) with an average of f around k /w might lead to smaller errors and therefore be useful in applications. O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 7 / 34 Shannon Sampling Series Sampling operators in Lp case Z. Burinska, K. Runovski and H. J. Schmeisser introduced in [BRS’06] for 0 < p 6 ∞ additional shifts λ: (Sw;λ f )(t) := ∞ X k =−∞ f( k + λ)s(w(t − λ) − k ) w and estimated the order of approximation via modulus of smoothness. C. Bardaro, P. L. Butzer, R. Stens and G. Vinti in [BBSV’06] introduced a suitable subspace of Lp (R). They estimated the order of approximation for the classical (Whittaker-Kotel’nikov-Shannon) operator for 1 < p < ∞ and for sampling operators with time-limited kernels for 1 6 p < ∞ in [Butzer, Stens’08] and [BBSV’10] via averaged modulus of smoothness. We used the same approach and estimated the order of approximation for sampling operators with band-limited kernels for 1 < p < ∞ in [T’13] via the averaged modulus of smoothness and for 1 6 p < ∞ in [Kivinukk,T’14] and [T’14] via the classical modulus of smoothness. O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 8 / 34 Shannon Sampling Series Sampling operators in Lp case Z. Burinska, K. Runovski and H. J. Schmeisser introduced in [BRS’06] for 0 < p 6 ∞ additional shifts λ: (Sw;λ f )(t) := ∞ X k =−∞ f( k + λ)s(w(t − λ) − k ) w and estimated the order of approximation via modulus of smoothness. C. Bardaro, P. L. Butzer, R. Stens and G. Vinti in [BBSV’06] introduced a suitable subspace of Lp (R). They estimated the order of approximation for the classical (Whittaker-Kotel’nikov-Shannon) operator for 1 < p < ∞ and for sampling operators with time-limited kernels for 1 6 p < ∞ in [Butzer, Stens’08] and [BBSV’10] via averaged modulus of smoothness. We used the same approach and estimated the order of approximation for sampling operators with band-limited kernels for 1 < p < ∞ in [T’13] via the averaged modulus of smoothness and for 1 6 p < ∞ in [Kivinukk,T’14] and [T’14] via the classical modulus of smoothness. O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 8 / 34 Shannon Sampling Series Sampling operators in Lp case Z. Burinska, K. Runovski and H. J. Schmeisser introduced in [BRS’06] for 0 < p 6 ∞ additional shifts λ: (Sw;λ f )(t) := ∞ X k =−∞ f( k + λ)s(w(t − λ) − k ) w and estimated the order of approximation via modulus of smoothness. C. Bardaro, P. L. Butzer, R. Stens and G. Vinti in [BBSV’06] introduced a suitable subspace of Lp (R). They estimated the order of approximation for the classical (Whittaker-Kotel’nikov-Shannon) operator for 1 < p < ∞ and for sampling operators with time-limited kernels for 1 6 p < ∞ in [Butzer, Stens’08] and [BBSV’10] via averaged modulus of smoothness. We used the same approach and estimated the order of approximation for sampling operators with band-limited kernels for 1 < p < ∞ in [T’13] via the averaged modulus of smoothness and for 1 6 p < ∞ in [Kivinukk,T’14] and [T’14] via the classical modulus of smoothness. O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 8 / 34 Shannon Sampling Series Sampling operators in Lp case C. Bardaro, P. L. Butzer, R. Stens and G. Vinti in [BBSV’07] introduced Kantorovich-type sampling operators (k +1)/w Z ∞ X f (u) du s(wt − k ) (SwK f )(t) = w k =−∞ k /w and proved the convergence for 1 6 p < ∞ case. We estimated the order of approximation for Kantorovich-type sampling operators (2nk +1)/2nw Z ∞ X I (Sw,n f )(t) = f (u) du s(wt − k ) nw k =−∞ (2nk −1)/2nw with band-limited kernels for 1 6 p 6 ∞ in [Orlova,T’14] via the classical modulus of smoothness. O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 9 / 34 Shannon Sampling Series Sampling operators in Lp case C. Bardaro, P. L. Butzer, R. Stens and G. Vinti in [BBSV’07] introduced Kantorovich-type sampling operators (k +1)/w Z ∞ X f (u) du s(wt − k ) (SwK f )(t) = w k =−∞ k /w and proved the convergence for 1 6 p < ∞ case. We estimated the order of approximation for Kantorovich-type sampling operators (2nk +1)/2nw Z ∞ X I (Sw,n f )(t) = f (u) du s(wt − k ) nw k =−∞ (2nk −1)/2nw with band-limited kernels for 1 6 p 6 ∞ in [Orlova,T’14] via the classical modulus of smoothness. O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 9 / 34 Shannon Sampling Series Singular integrals Singular integrals In order to proceed with Kantorovich-type sampling operators we need to define the singular integral. If χ ∈ L1 (R) is such that Z∞ χ(u)du = 1, −∞ then the convolution integral of f with χρ (u) := ρχ(ρu), namely (Iρχ f )(x) Z∞ := (f ∗ χρ )(x) = f (u)ρχ(ρ(x − u))du (x ∈ R) (3) −∞ is called the singular (convolution) integral with kernel χ. O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 10 / 34 Shannon Sampling Series Kantorovich-type sampling operators Kantorovich-type sampling operators As a final step we replace the exact value f (k /w) in (2) with an average χ of f around k /w, using the singular integral Inw f (n ∈ N) with k /w as an argument. Thus for f ∈ Lp (R) (1 6 p 6 ∞) the Kantorovich-type sampling operators are given by (t ∈ R; w > 0; n ∈ N) ∞ Z ∞ X k K (4) (Sw,n f )(t) := f (u)nwχ(nw( − u))du s(wt − k ) w k =−∞ −∞ with kernels s, χ ∈ L1 (R), R∞ −∞ χ(u)du = 1, P k ∈Z s(u − k ) = 1 (u ∈ R). Remark. Bardaro, Butzer, Stens and Vinti considered the case χ = χ[0,1] , n = 1 only. O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 11 / 34 Shannon Sampling Series Bandlimited kernels Bandlimited kernels In this talk we consider an even band-limited kernel s, i.e. s ∈ Bπ1 , defined by an even window function λ ∈ C[−1,1] , λ(0) = 1, λ(u) = 0 (|u| > 1) by the equality Z1 s(t) := s(λ; t) := λ(u) cos(πtu) du. (5) 0 In fact, this kernel is the Fourier transform of λ ∈ L1 (R), r π ∧ λ (πt). s(t) = 2 (6) Lemma (Butzer, Splettstößer, Stens’88) If s ∈ Bπ1 , then for f ∈ C(R) kSws f − f kC kIws f − f kC . O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 12 / 34 Shannon Sampling Series Bandlimited kernels Many kernels can be defined by Z1 s(t) := λ(u) cos(πtu) du, 0 e.g. 1) λ(u) = 1 defines the sinc function; 2) λ(u) = 1 − u defines the Fejér kernel sF (t) = 12 sinc 2 2t (cf. [13]); 3) λj (u) := cos π(j + 1/2)u, j = 0, 1, 2, . . . defines the Rogosinski-type kernel (see [7]) in the form 1 1 1 sinc(t + j + ) + sinc(t − j − ) (7) rj (t) := 2 2 2 4) λH (u) := cos2 πu 2 = 12 (1 + cos πu) defines the Hann kernel (see [8]) sH (t) := O. Orlova, G. Tamberg (TUT) 1 sinc t ; 2 1 − t2 Kantorovich-type Sampling Operators (8) Strobl 2014 13 / 34 Shannon Sampling Series Bandlimited kernels 5) the general cosine window λC,a (u) := m X ak cos k πu (9) k =0 defines the Blackman-Harris kernel (see [9]) m 1X ak sinc(t − k ) + sinc(t + k ) sC,a (t) := 2 (10) k =0 provided (here and following bxc is the largest integer less than or equal to x ∈ R) m m+1 bX bX 2c 2 c 1 a2k = a2k −1 = . (11) 2 k =0 k =1 We get Hann kernel (8) if we take m = 1 in (10). O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 14 / 34 Shannon Sampling Series Bandlimited kernels 6) powers of the Hann window (see [6], formula(25a)) πu λH,m (u) := cosm 2 m m 1 X m cos (k − )πu , = m 2 k 2 (12) (13) k =0 give a general Hann kernel in the form sH,m (t) = 2−m Γ(1 + m 2 Γ(1 + m) − t)Γ(1 + m 2 + t) (14) . Comparing the window function λH,m in (13) and the general cosine window λC,a in (9) we see that the general Hann kernel in case of m = 2n (n ∈ N) is a special case of the Blackman-Harris kernel. Indeed, sH,2n = sC,a∗ , where the parameter vector a∗ ∈ Rn+1 has 2n 1 ∗ components a0∗ = 212n 2n n and ak = 22n−1 n−k for k = 1, 2, . . . , n. O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 15 / 34 Shannon Sampling Series Bandlimited kernels 7) the general Rogosinski-type window λR,a (u) := m X aj λj (u) = m X aj cos π(j + 1/2)u (15) j=0 j=0 defines the general Rogosinski-type kernel m 1X 2j + 1 2j + 1 aj sinc(t − ) + sinc(t + ) ra (t) := 2 2 2 (16) j=0 provided m X aj = 1. (17) j=0 Comparing the window function λH,m in (13) and the general Rogosinski-type window λR,a in (15) we see that the general Hann kernel in case of m = 2n + 1 (n ∈ N0 ) is a special case of the general Rogosinski-type kernel. Indeed, sH,2n+1 = sR,a∗ , where the parameter vector a∗ ∈ Rn+1 has components aj∗ = 212n 2n+1 n−j for j = 0, 1, . . . , n. O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 16 / 34 Shannon Sampling Series Kernels for singular integrals Kernels for singular integrals In this talk we consider for the singular integrals an even kernel χ ∈ L1 (R) with absolute moment X m0 (χ) := sup |χ(u − k )| < ∞, u∈R k ∈Z defined by an even window function µ, µ(0) = 1 by the equality Z∞ χ(t) := χ(µ; t) := µ(u) cos(πtu) du. (18) 0 In fact, this kernel is also a kernel for generalized sampling operators, when we additionally require µ(2k ) = 0, k ∈ N. O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 17 / 34 Shannon Sampling Series Kernels for singular integrals Many kernels can be defined by Z∞ χ(t) := χ(µ; t) := µ(u) cos(πtu) du 0 e.g. 1) µ(u) = sinck u (k ∈ N) defines the B-spline kernel Bk −1 , in 2 particular case k = 1 we get the indicator function χ[−1/2,1/2] ; u m−1 P defines a spline kernel of degree m if 2) µa (u) = ak sinck +1 2 k =0 P ak = 1; u 2 3) µ(u) = e−π( 2 ) defines a Gauss-Weierstraß kernel 2 χGW (t) = e−πt . O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 18 / 34 Operator norms Norms of the Kantorovich-type sampling operators Norms of the Kantorovich-type sampling operators Theorem K = S s (I χ ) we have for For Kantorovich-type sampling operator Sw,n w nw 1 6 p 6 ∞ an estimate of the operator norm (1/p + 1/q = 1) 1 1 p q K kχk1 m0 (s) kSw,n kp→p 6 nksk1 m0 (χ) Remark. We see that in the case of generalized sampling operators Sw (i.e. n → ∞) this estimate is not bounded for 1 6 p < ∞. Remark. Take χ = χ[−1/2,1/2] . We proved that in this case for L1 (R) we K k K have kSw,n 1→1 = nksk1 and for C(R) we have kSw,n kC→C = m0 (s). O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 19 / 34 Operator norms Nikolskii’s inequality Nikolskii’s inequality Theorem (Nikolskii’s inequality) Let 1 6 p 6 ∞. Then, for every s ∈ Bσp , ∞ X ( kskp 6 sup u∈R )1/p |s(u − k )| p 6 (1 + σ)kskp . k =−∞ Our kernels s ∈ Bπ1 . By Nikolskii’s inequality we have ksk1 6 m0 (s) = kSw k, which gives the estimate 1 1 p K kSw,n kp→p 6 kSw k n m0 (χ) kχk1q O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 20 / 34 Order of approximation Theorem If we have for sampling operator Sws 1 )p w kSws f − f kp 6 M1 ωk (f ; and for the singular integral χ kInw f − f kp 6 M2 ω` (f ; 1 )p , w K = S s (I χ ) then for Kantorovich-type sampling operator Sw,n w nw K kSw,n f − f kp 6 M3 ωr (f ; 1 )p , w where r := min{k , `}. O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 21 / 34 Order of approximation Corollary K = S s (I χ ), where s is For Kantorovich-type sampling operator Sw,n w nw Rogosinski-type, Hann, Blackman-Harris or B-spline kernel and χ is a indicator function or Gauss-Weierstraß kernel we have K kSw,n f − f kp 6 Mω2 (f ; 1 )p . w Indeed, for Rogosinski-type, Hann, Blackman-Harris and B-spline kernels we have the estimates of order of approximation for corresponding generalized sampling operators via modulus of smoothness order 2. For singular integrals with indicator function and Gauss-Weierstraß kernels we have the estimates of the order of approximation via modulus of smoothness order 2. O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 22 / 34 Order of approximation The main theorem Theorem K = S s (I χ ) (w > 0, n ∈ N) Let Kantorovich-type sampling operator Sw,n w nw be defined by the kernel s ∈ Bπ1 with window λ and by the kernel χ ∈ L1 with window µ. If for some r ∈ N νn (u) := λ(u)µ u n =1− ∞ X cj u 2j , ∞ X |cj | 6 ∞. (19) j=r j=r Then for f ∈ Lp (R) (1 6 p 6 ∞) K kSw,n f − f kp 6 Mr ω2r (f ; 1 )p . w (20) The constants Mr are independent of f and w. O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 23 / 34 Order of approximation The main theorem Idea of the proof Lemma (cf. Butzer, Stens’85) p 1 (α + β = 2), then If g ∈ Bαπw and s ∈ Bβπ Sws g = Iws g. p p χ K = S s (I χ ) Take g ∈ Bπw/n , then Inw g ∈ Bπw and for Sw,n w nw χ K K kSw,n f − f kp 6 kSw,n kp→p kf − gkp + kIws (Inw g) − gkp + kg − f kp . χ g) = sw ∗ χnw ∗ g = (s ∗ χn )w ∗ g = ϕw ∗ g = Iwϕ g = Swϕ g Iws (Inw where ϕ ∈ Bπ1 is defined Z1 Z1 νn (u) cos πut du = ϕ(t) := 0 O. Orlova, G. Tamberg (TUT) λ(u)µ u n cos πut du. 0 Kantorovich-type Sampling Operators Strobl 2014 24 / 34 Order of approximation The main theorem The space Λp Definition (Bardaro, Butzer, Stens, Vinti’ 2006) Let Σ := (xj )j∈Z ⊂ R be an admissible partition of R, i.e. 0 < inf ∆j 6 sup ∆j < ∞, ∆j := xj − xj−1 and let the discrete j∈Z j∈Z `p (Σ)-seminorm of a sequence of function values fΣ on Σ of a function f : R → C be defined for 1 6 p < ∞ by kf k`p (Σ) := X |f (xj )|p ∆j j∈Z 1/p . The space Λp for 1 6 p < ∞ is defined by Λp := {f ∈ M(R); kf k`p (Σ) < ∞ for each admissible sequence Σ}. O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 25 / 34 Order of approximation The main theorem Denote for 1 6 p < ∞ X p (R) := Λp and X ∞ (R) := C(R). Bσp ⊂ X p (R). Theorem (Kivinukk,T’14) Let sampling operator Swr (w > 0) be defined by the kernel sr ∈ Bπ1 with λ = λr and for some r ∈ N let λr (u) := 1 − ∞ X j=r cj u 2j , ∞ X |cj | 6 ∞. (21) 1 )p . w (22) j=r Then for f ∈ X p (R) (1 6 p 6 ∞) r kSW f − f kp 6 Mr ω2r (f ; The constants Mr are independent of f and w. O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 26 / 34 Order of approximation The main theorem p We have for g ∈ Bπw/n K kSw,n g − gkp = kSwϕ g − gkp 6 Mr ω2r (g; 6 Mr ω2r (f ; 1 )p w 1 1 1 )p + Mr ω2r (g − f ; )p 6 Mr ω2r (f ; )p + Mr 22r kg − f kp . w w w Theorem (Jackson-type theorem) Given f ∈ C(R) or f ∈ Lp (R) (1 6 p < ∞). Then there exists a gσ∗ ∈ Bσp (1 6 p 6 ∞) and a constant Ck > 0 (depending only on k ∈ N) such that 1 kf − gσ∗ kp 6 Ck ωk (f , )p . σ O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 27 / 34 Order of approximation Examples Examples Theorem χ I Let Cw,a = Sws (Inw ) be a Kantorovich-type sampling operator whith s = sC,a (a ∈ Rm+1 ) and χ = χ[−1/2,1/2] , then for f ∈ Lp (R) (1 6 p 6 ∞) 1 I kCw,a,n f − f kp 6 Ma,2 ω2 (f ; )p . w Moreover, if there holds m X k =1 ak k 2 = − 1 , (22 − 1)22 n2 then I kCw,a,n f − f kp 6 Ma,4 ω4 (f ; O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators 1 )p . w Strobl 2014 28 / 34 Order of approximation Examples The Bibliographic References I Bardaro, C., Butzer, P.L., Stens, R.L., Vinti, G.: Approximation error of the Whittaker cardinal series in terms of an averaged modulus of smoothness covering discontinuous signals. J. Math. Anal. Appl. 316, 269–306 (2006) Bardaro, C., Butzer, P.L., Stens, R.L., Vinti, G.: Prediction by samples from the past with error estimates covering discontinious signals. IEEE Transactions on Information Theory 56, 614–633 (2010) Burinska, Z., Runovski, K., Schmeisser, H.J.: On the approximation by generalized sampling series in Lp -metrics. Sampling Theory in Signal and Image Processing 5, 59–87 (2006) O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 29 / 34 Order of approximation Examples The Bibliographic References II Butzer, P.L., Splettstößer, W., Stens, R.L.: The sampling theorems and linear prediction in signal analysis. Jahresber. Deutsch. Math-Verein 90, 1–70 (1988) Butzer, P.L., Stens, R.L.: Reconstruction of signals in Lp (R)-space by generalized sampling series based on linear combinations of B-splines. Integral Transforms Spec. Funct. 19, 35–58 (2008) Harris, F.J.: On the use of windows for harmonic analysis. Proc. of the IEEE 66, 51–83 (1978) Kivinukk, A., Tamberg, G.: On sampling series based on some combinations of sinc functions. Proc. of the Estonian Academy of Sciences. Physics Mathematics 51, 203–220 (2002) O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 30 / 34 Order of approximation Examples The Bibliographic References III Kivinukk, A., Tamberg, G.: On sampling operators defined by the Hann window and some of their extensions. Sampling Theory in Signal and Image Processing 2, 235–258 (2003) Kivinukk, A., Tamberg, G.: On Blackman-Harris windows for Shannon sampling series. Sampling Theory in Signal and Image Processing 6, 87–108 (2007) Kivinukk, A., Tamberg, G.: Interpolating generalized Shannon sampling operators, their norms and approximation properties. Sampling Theory in Signal and Image Processing 8, 77–95 (2009) O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 31 / 34 Order of approximation Examples The Bibliographic References IV Kivinukk, A., Tamberg, G.: On window methods in generalized Shannon sampling operators. In: New Perspectives on Approximation and Sampling TheoryFestschrift in honor of Paul Butzer’s 85th birthday. Birkhauser Verlag (2014 (to appear)). In press Tamberg, G.: Approximation error of generalized shannon sampling operators with bandlimited kernels in terms of an averaged modulus of smoothness. Dolomites Research Notes on Approximation 6, 74–82 (2013) Theis, M.: Über eine interpolationsformel von de la Vallee-Poussin. Math. Z. 3, 93–113 (1919) O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 32 / 34 Order of approximation Examples Thank you! O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 33 / 34 Order of approximation Examples [4],[13],[1],[2],[11], [5], [12],[10], [3] O. Orlova, G. Tamberg (TUT) Kantorovich-type Sampling Operators Strobl 2014 34 / 34