Non-Uniform Fast Fourier Transformation (NUFFT) and Magnetic Resonace Imaging 2008 년 11 월 27 일. Lee, June-Yub (Ewha Womans University)

Size: px
Start display at page:

Download "Non-Uniform Fast Fourier Transformation (NUFFT) and Magnetic Resonace Imaging 2008 년 11 월 27 일. Lee, June-Yub (Ewha Womans University)"

Transcription

1 Non-Uniform Fast Fourier Transformation (NUFFT) and Magnetic Resonace Imaging 2008 년 11 월 27 일 Lee, June-Yub (Ewha Womans University)

2 1 NUFFT and MRI Nonuniform Fast Fourier Transformation (NUFFT) Basics of MRI Comments on Fourier Inversion 2 The NUFFT of Type-1,2 The Type-1 and Type-2 Nonuniform FFT Fast Gaussian gridding method Numerical Results of Type-1,2 NUFFT 3 The NUFFT of Type-3 The non-uniform FFT of Type-3 Numerical examples of Type-3 NUFFT 4 The Fast Sinc-Transform The Fast Sinc-Transform using NUFFT An example of fast Sinc-transformation 5 Imaging Modality for MRI Basic approaches for Fourier Inversion 6 Conclusions

3 Nonuniform Fast Fourier Transformation (NUFFT) DFT and Type 1-3 NUFFT 0. DFT (Discrete Fourier transforation) : O(N 2 ) FFT : O(NlogN) F (k) = 1 N N 1 j=0 f (j) e 2πijk/N, f (j) = k F (k) e2πijk/n (1) F (k) = 1 2π 2π 0 f (x) e ikx j dx, f (j) = k F (k) e ikx j with x j = 2πj N (2)

4 Nonuniform Fast Fourier Transformation (NUFFT) DFT and Type 1-3 NUFFT 0. DFT (Discrete Fourier transforation) : O(N 2 ) FFT : O(NlogN) F (k) = 1 N F (k) = 1 2π N 1 j=0 f (j) e 2πijk/N, f (j) = k F (k) e2πijk/n (1) 2π 0 f (x) e ikx j dx, f (j) = k F (k) e ikx j 1. Type1-NUFFT in 2D : O(NlogN) instead of O(N 2 ) F (k 1, k 2 ) = 1 N with x j = 2πj N (2) N 1 j=0 f j e i(k1,k2) x j, (3) with f j = f (x j ) w j can be considered as a discretization of F (k 1, k 2 ) = 1 2π 2π (2π) f (x) e i(k1,k2) x dx (4) with the quadrature weight w j corresponding to x j.

5 Nonuniform Fast Fourier Transformation (NUFFT) DFT and Type 1-3 NUFFT II 2. Type2-NUFFT in 2D : O(NlogN) instead of O(N 2 ) f (x j ) = F (k 1, k 2 ) e i(k1,k2) xj, (5) k 1 k 2 is simply the evaluation of a finite Fourier series at an arbitrary set of targets.

6 Nonuniform Fast Fourier Transformation (NUFFT) DFT and Type 1-3 NUFFT II 2. Type2-NUFFT in 2D : O(NlogN) instead of O(N 2 ) f (x j ) = F (k 1, k 2 ) e i(k1,k2) xj, (5) k 1 k 2 is simply the evaluation of a finite Fourier series at an arbitrary set of targets. 3. Type3-NUFFT in R d : O(NlogN) instead of O(N 2 ) F k = N 1 j=0 f j e ±is k x j, (6) is a discretization of the continuous Fourier transform, F (s) = 1 (2π) d... f (x) e ±is x dx (7) at nonuniformly sampled frequencies using nonuniformly sampled points.

7 Basics of MRI Structure of Magnetic Resonance Imaging Machine Without Magnetic Field With Magnetic Field H o Main Magnet Coil Shim Coil Gradient coil (Body Coil) Random orientation S N Aligned Spinning tops Head Coil Signal from everywhere M( r, t) = M 0 ρ( r)e iωt, ω = γ H S(t) = M( r, t)d r

8 Basics of MRI Bloch Equation and Signal localization Signal from everywhere M( r, t) = M 0 ρ( r)e iφ(t, r), φ(t, r) = ωt = γ H( r) t S(t) = M( r, t)d r

9 Basics of MRI Bloch Equation and Signal localization Signal from everywhere M( r, t) = M 0 ρ( r)e iφ(t, r), φ(t, r) = ωt = γ H( r) t S(t) = M( r, t)d r MRI using FT : Echo-Planar Imaging(EPI) by Mansfield in 1977 S(t) = M 0 ρ( r)e iφ(t, r) d r, under H(t) = H 0 + G(t) r, φ(t, r) = H 0 t + t 0 (G x x + G y y)dt Let M 0 = M 0 e iω0t = M 0 ρ( r z0 )e iγ t 0 (Gx x+gy y)dt dxdy, Let k(t) = (G x, G y ) dt 0 ( ) = M 0 ρ( r z0 )e iγ k(t) (x,y) dxdy = M 0 FT [ρ( r z0 )] k(t) t

10 Basics of MRI Image Reconstruction Method Real part Imaginary part Frequency K-space Inverse Fourier Transformation Physical XY-space Magnitude Image Phase Image

11 Basics of MRI Classical sampling on uniform grid Classical Uniform Sampling ky Inverse Fourier Transformation In most clinical system, Fourier data ( ) ρ(kq, 1 kq) 2 := FT [ρ( r z0 )] k(tq ) kx are acquired on a uniform cartesian grid (k 1 m 1, k 2 m 2 ). Then the reconstructed image ˆρ(x 1 n 1, x 2 n 2 ) can be computed using backward Fast Fourier Transformation (FFT), C M 1 M 1 m 1=0 m 2=0 e 2πi m 1 n 1 M e 2πi m 2 n 2 M ρ(k 1 m1, k 2 m 2 )

12 Basics of MRI Fast scanning methods Polar Sampling Spiral Sampling Rosette Sampling ky ky ky kx kx kx Advantages and Disadvantages Pros Fast Scanning Time Cons Computationally. Slow reconstruction (FFT is not applicable) Mathematically. Inversion is ill-conditioned (FT is not unitary)

13 Comments on Fourier Inversion Comments on Fourier integral(fourier Inversion) The Fourier integral H(u 1, u 2 ) = h(v) e i(u1,u2) v dv (8) with h(v) at a scatter of points rather than on a regular Cartesian mesh.

14 Comments on Fourier Inversion Comments on Fourier integral(fourier Inversion) The Fourier integral H(u 1, u 2 ) = h(v) e i(u1,u2) v dv (8) with h(v) at a scatter of points rather than on a regular Cartesian mesh. There are three separate issues involved: 1 aquisition of data h(v j ) in the domain, 2 the choice of a quadrature scheme {v j, w j }, and 3 the availability of a fast algorithm for computing the discrete transform itself.

15 Comments on Fourier Inversion Comments on Fourier integral(fourier Inversion) The Fourier integral H(u 1, u 2 ) = h(v) e i(u1,u2) v dv (8) with h(v) at a scatter of points rather than on a regular Cartesian mesh. There are three separate issues involved: 1 aquisition of data h(v j ) in the domain, 2 the choice of a quadrature scheme {v j, w j }, and 3 the availability of a fast algorithm for computing the discrete transform itself. Remarks: Not all schemes for reconstructing Fourier integrals can be represented as a quadrature H(u 1, u 2 ) = 1 N 1 N j=0 w jh j e i(u1,u2) v j, of the type (3), however, if the decision has been made to use a quadrature approach then the remaining task is entirely computational.

16 The Type-1 and Type-2 Nonuniform FFT Type-1 NUFFT: Discrete to Continuum Extension I With x j [0, 2π], the type-1 NUFFT is defined by the calculation of F (k) = 1 N N 1 j=0 f j e ikx j for k = M 2,, M 2 1 (9) is the exact Fourier coefficients of the function f (x) = N 1 j=0 f jδ(x x j ). (10) 1 The convolution of f (x) with g τ (x) = l= e (x 2lπ)2 /4τ, f τ (x) = f g τ (x) = 2π can be well-resolved by a uniform mesh in x. 0 f (y)g τ (x y) dy, (11) N 1 f τ (2πm/M r ) = f j g τ (2πm/M r x j ). (12) j=0

17 The Type-1 and Type-2 Nonuniform FFT Type-1 NUFFT: Discrete to Continuum Extension II 2 The Fourier coefficients of f τ, namely F τ (k) = 1 2π 2π 0 f τ (x)e ikx dx, can be computed using the standard FFT on a over-sampled grid F τ (k) 1 M r 1 M r m=0 f τ (2πm/M r )e ik2πm/mr. (13) 3 Once the values F τ (k) are known, an elementary calculation shows that π F (k) = 2τ τ ek F τ (k). (14) (This is a direct consequence of the convolution theorem and the fact that the Fourier transform of g τ is G τ (k) = 2τ e k2τ ).

18 The Type-1 and Type-2 Nonuniform FFT Discrete Summation using smooth delta functions image xj 0 2π image Figure: In the version of the NUFFT described here, each delta function source at a point such as x j in eq. (10) is replaced by a Gaussian. This smears the source strength to nearby regular grid points. The regular grid must be fine enough to resolve the smeared function f τ in eq. (11). Note that we include 2π-periodic images of the sources in the definition of the heat kernel g τ. They decay sufficiently rapidly that all but the nearest ones can be ignored.

19 The Type-1 and Type-2 Nonuniform FFT The Type-2 Nonuniform FFT The type-2 NUFFT evaluates a regular Fourier series at irregular target points, f (x j ) = M 2 1 F (k) e ikx k= M j. (15) 2 1 We first deconvolve the Fourier coefficients, defining F τ (k) by π F τ (k) = 2τ τ ek F (k) (16) 2 Evaluate f τ (x) on a uniform mesh using the FFT f τ (x) = M r 1 k=0 F τ (k) e ikx. (17) 3 We then compute the desired values f (x k ) from 2π f (x j ) = f τ g τ (x j ) = 1 f τ (x)g τ (x j x) dx 2π 0 1 Mr 1 m=0 M f τ (2πm/M r ) g τ (x j 2πm/M r ). (18) r

20 Fast Gaussian gridding method Fast gridding method using Gaussian Kernel I The dominant task is the calculation of f τ (2πm/M r ) in (12) and f (x j ) in (18). f τ (2πm/M r ) = N 1 j=0 f j l= e (x j 2πm/M r 2lπ) 2 /4τ. An elementary calculation shows that e (x j 2πm/M r ) 2 /4τ = e x 2 j /4τ ( e x j π/m r τ ) m e (πm/m r ) 2 /τ. (19) Fast Gridding Algorithm of Type 1 in two dimension Step I: Initialization 1 Set the oversampling ratio R = M r /M, the spreading parameter M sp, and the Gaussian kernel parameter τ according to the desired precision ɛ. 2 Precompute E 3 (l) = e (πl/mr )2 /τ for 0 l M sp and E 4 (k) = E 4 (M k) = e τk2 for k M 2.

21 Fast Gaussian gridding method Fast gridding method using Gaussian Kernel II Step C: Convolution for each source point (x j, y j ) 1 Find the nearest grid point (ξ 1, ξ 2 ) = 2π M r (m 1, m 2 ) with ξ 1 x j, ξ 2 y j. 2 Compute E 1 = e ((x j ξ 1) 2 +(y j ξ 2) 2 )/4τ, E 2x = e π(x j ξ 1)/M r τ, E 2y = e π(y j ξ 2)/M r τ l and E 2x (l 1 ) = E 1 l 2x, E 2y (l 2 ) = E 2 2y for M sp < l 1, l 2 M sp 3 Convolve the gaussian spreading function with f j as follows: V 0 = f j E 1 for l 2 = M sp +1, M sp V y = V 0 E 2y (l 2 ) for l 1 = M sp +1, M sp Add V y E 2x (l 1 ) to f τ (m 1 + l 1, m 2 + l 2 ). Step D: FFT and deconvolution 1 Compute 2D FFT of f τ (m 1, m 2 ) to obtain F τ (k 1, k 2 ). 2 Set F (k 1, k 2 ) = π τ E 4(k 1 )E 4 (k 2 )F τ (k 1, k 2 ) for M 2 k 1, k 2 < M 2

22 Numerical Results of Type-1,2 NUFFT Example 1. (Verification of Accuracy) We compare the results of the type 1 and 2 transforms with uniformly distributed random data points using direct summation and the NUFFT Type 1: 1024 points in 1D 10 0 Type 1&2: 64*64 grid Type 1&2: points in 2D 0 10 Computed Error Spreading distance Spreading distance Spreading distance Figure: Gridding error in the NUFFT.

23 Numerical Results of Type-1,2 NUFFT Example 2. (Fast gridding compared to gridding) 1.4 1D: Ultra 60(450Mhz) 1 1D: Pentium III(1GHz) 0.7 1D: Pentium IV(2.4Ghz) Time (Sec) Time (Sec) , ,000 Number of Data Points Time (Sec) , ,000 Number of Data Points , ,000 Number of Data Points 10 2D: Ultra 60(450Mhz) 10 2D: Pentium III(1GHz) 6 2D: Pentium IV(2.4Ghz) Time (Sec) 6 4 Time (Sec) 6 4 Time (Sec) , ,000 Number of Data Points , ,000 Number of Data Points , ,000 Number of Data Points Figure: Computing time.

24 Numerical Results of Type-1,2 NUFFT Example 3. (Comparison with direct method) Fast gridding algorithm v.s. direct summation and the FFT D Transformation Cost D Transformation Cost 10 0 Time (Sec) 10 1 Time (Sec) Number of Data Points Number of Data Points Figure: CPU requirements of with tolerance ɛ = 10 3, 10 6, 10 9, D ɛ=10 3 ɛ=10 6 ɛ=10 9 ɛ=10 12 Time (µsec) 3.77 N 4.40 N 4.78 N 5.44 N Breakeven (N) D ɛ=10 3 ɛ=10 6 ɛ=10 9 ɛ=10 12 Time (µsec) 8.45 N N N N Breakeven (N = M 2 ) 10*10 15*15 20*20 26*26

25 Numerical Results of Type-1,2 NUFFT Example 4. (MRI image reconstruction) I The exact reconstruction would obviously be the Fourier integral ˆf (j 1, j 2 ) = 2π 0 0 F (r, θ)e i(j1,j2) (r cos θ,r sin θ) r dr dθ, (20) which can be approximated by a type 1 transformation, ˆf (j 1, j 2 ) = N 1 k=0 F k e i(j1,j2) (sk x,sk y ). (21) Two issues should be considered: (1) the selection of points {(s k x, s k y )} which will certainly affect the image quality, (2) quadrature weights W k so that F k W k F (s k x, s k y ). We simply use (s k x, s k y ) = r j (cos(θ i ), sin(θ i )), r j = πj M, θ i = 2πi 2M (22) and the quadrature weight is r j θ r = (jπ/m) (2π/2M) (π/m).

26 Numerical Results of Type-1,2 NUFFT Example 4. (MRI image reconstruction) II y= y= Figure: Reconstructed image from a radial grid data.

27 The non-uniform FFT of Type-3 The Algorithm of Type-3 NUFFT I Type-3 NUFFT can be interpreted as the continuous Fourier transform of the function f (x) = N 1 2π f j δ(x x j ) (23) evaluated at the point s = s k. j=0 1 We begin by convolving f (x) with g τ (x) = e x 2 /4τ, f τ (x) = f g τ (x) = 1 2π f (y)g τ (x y) dy. (24) Since f τ can be well-resolved by a uniform mesh in x we define the discrete equispaced samples of f τ by f τ (n x ) = N 1 j=0 f jg τ (n x x j ). (25)

28 The non-uniform FFT of Type-3 The Algorithm of Type-3 NUFFT II 2 Using the Fourier transform of g τ (x) by G τ (s) = 2τ e s2τ, we define fτ σ (x) = f τ (x) G σ (x) = 1 e σx 2 f τ (x). (26) 2σ 3 The Fourier transform of fτ σ, namely Fτ σ (s) = 1 2π f σ τ (x)e isx dx, can be computed using the standard FFT on a sufficiently fine grid, Fτ σ (m s ) x 2π n f σ τ (n x )e imn x s. (27)

29 The non-uniform FFT of Type-3 The Algorithm of Type-3 NUFFT III 4 The next step is to recover the values F τ (s k ) by convolving Fτ σ (s) with g σ (s): F τ (s k ) = ( Fτ σ ) g σ (sk ) = 1 2π s 2π m Fτ σ (s)g σ (s k s) ds F σ τ (m s ) g σ (s k m s ). (28) 5 Once the values F τ (s k ) are known, an elementary calculation shows that F (s k ) = 1 2τ e s2 k τ F τ (s k ). (29) This follows again from the convolution theorem.

30 The non-uniform FFT of Type-3 Setting the tuning parameters The actual implementation requires a complete specification of all details. We do not repeat the analysis of [2] here, but summarize the relevant results as follows: if we let τ = 2 x m sp, 2π 2( 2 1) x π S 1 2, s π 1 2, (30) X + m sp x σ = 2 s m sp, M r = 2π, 2π 2( 2 1) x s then carrying out the convolutions in (25) and (28) with m sp = 9 yields about six digits of accuracy. Carrying out the convolutions with m sp = 18 yields about twelve digits of accuracy. Efficient implementation of these steps can be carried out using the fast Gaussian gridding algorithm. The higher dimensional versions involve more notation but are obvious extensions of the one-dimensional scheme. Appropriate values of M r, x and s can be chosen for each dimension separately.

31 Numerical examples of Type-3 NUFFT Example 5. (MRI in inhomogeneous magnetic field) In the presence of a field inhomogeneity given by φ(x), s(t) = ρ(x)e i2πk(t) x e iφ(x)t dx. (31) This requires the computation of s(t j ) N n=1 w nρ(x n )e i2πk(t j ) x n e iφ(xn)t j = N n=1 w nρ(x n )e ik j X n where K j = (k 1 (t j ), k 2 (t j ), t j ) and X n = (2πx 1 n, 2πx 2 n, φ(x n )). Thus, by embedding the data points in a higher dimensional space, one can carry out the transformation using a type-3 NUFFT. For a case with a spiral trajectory up to a maximum frequency of 60 and the regular (x 1, x 2 ) mesh on the unit box, direct evaluation required about 140 seconds and the NUFFT required 3.8 seconds to obtain six digits of accuracy when the time interval is [0, 1] and φ(x) varies in the interval [ 5π, 5π].

32 The Fast Sinc-Transform using NUFFT Definition of Fast Sinc-Transformatin I We define the d-dimensional sinc transforms by G l = N j=1 q j sinc(k j v l ). (32) which can be viewed as the evaluation of the function G(k) = sinc(k k )H(k )dk (33) at the points v l, due to the singular source distribution H(k) = N j=1 q jδ(k k j ). This follows from the elementary properties of the δ-function. From the convolution theorem we have that G(k) is given by G(k) = g(x) e 2πix k dx (34) with g(x) = F 1 sinc(k) F 1 H(k). (35)

33 The Fast Sinc-Transform using NUFFT Definition of Fast Sinc-Transformatin II The latter two functions are easily computed. The inverse Fourier transform F 1 sinc(k) in two dimensions is simply Π(x) = { 0 if x1 > 1/2 or x 2 > 1/2 1 if x 1 < 1/2 and x 2 < 1/2, (36) where x = (x 1, x 2 ). Further, it is easy to see that h(x) = F 1 H(k) = N q j e 2πix k j. (37) j=1 Thus, we can compute G(v l ) from (34)-(37): G(v l ) = 1/2 1/2 1/2 1/2 h(x) e 2πix v l dx. (38)

34 The Fast Sinc-Transform using NUFFT Quadrature considerations Equation (38) is an exact relation, and it remains only to discretize the integral with a collection of exponential functions with maximum frequency given by K max = max j k j L 1. Using a tensor-product rule for the double integral, we have G(v l ) = 1/2 1/2 1/2 M j 1=1 j 2=1 1/2 h(x) e 2πix v l dx M h(x j1, x j2 )e 2πi(x j 1,x j2 ) v l q j1 q j2 (39) The error, as in the one-dimensional case, decays at an exponential rate once M exceeds πk max. In summary, the fast sinc transform requires the adjoint NUFFT to compute h(x) via (37) at the tensor product quadrature points. The amount of work is of the order O((N + M 2 ) log(n + M 2 )) = O((N + K 2 max) log(n + K 2 max)). Since the quadrature used is spectrally accurate, the error in the fast sinc transform is dominated by the tolerance requested of the NUFFT.

35 The Fast Sinc-Transform using NUFFT The fast sinc 2 -transform The he inverse Fourier transform of sinc 2 (k) in two dimensions is t(x 1 ) t(x 2 ) where t(x) = We therefore need to compute G(k) = { 0 if x > 1 1 x if 1 < x < 1. h(x 1, x 2 ) t(x 1 ) t(x 2 ) e 2πi(x1,x2) k dx 1 dx 2 (40) where the integrand is smooth, on the four quadrants. In summary, the fast sinc 2 transform requires the adjoint NUFFT to compute h(x) via (37) at the tensor product quadrature points, followed by the forward NUFFT to compute (40) using tensor product Gaussian quadrature. The amount of work is O((N + K 2 max) log(n + K 2 max)).

36 An example of fast Sinc-transformation Computational Speed To test the utility of the fast algorithm, we constructed an Archimedean spiral sampled at N points according to the formula ( ( ) ( )) k j = K max j N cos 3πK max j N, sin 3πK max j N. Sample timings for K max = 64 and two different values for N are given in Table 1. N T FST T FS 2 T T dir Error < , < < , < 10 5 Table: Timing results for FST and FS 2 T on Archimedean spiral with K max = 64. Calculations were carried out on a laptop computer with a 1.2GHz Pentium processor.

37 Basic approaches for Fourier Inversion Inversion of continuous-to-discrete Fourier Transformation A number of imaging modalities require the inversion of the equation s(n) = ρ(r)e 2πık(n) r dr, (41) V where k(n) denotes the location of the nth measurement in the frequency domain ( k-space ) and r denotes position in the image domain. It will be convenient below to write this in operator form as s n = Hρ (k n ), (42) where H is the continuous-to-discrete Fourier operator which maps the image space to the signal space. We are particularly concerned with non-uniform sampling schemes, where the points {k(n)} do not lie on a regular grid. The inversion of (42) is, of course, an inherently ill-posed problem; the space of all possible densities ρ(r) is infinite dimensional space, while the vector of measurements {s(n)} is finite dimensional.

38 Basic approaches for Fourier Inversion Scheme 1 (Optimal quadrature) The first reconstruction scheme relies on the inverse Fourier transform ρ(r) = s(k)e 2πık r dk, (43) via a quadrature formula or, more precisely, its approximation via ρ(r) n s(n)e 2πık(n) r w n. (44) which can be carried out in O(N log N) and an optimal set of weights {w n } is given by the formula 1 w n = m sinc2 (k(m) k(n)). (45) Here, sinc(k) sin(πk) πk and sinc(k) = sinc(k 1 ) sinc(k d ), where k = (k 1, k 2,..., k d ). While the evaluation of these weights appears to require O(N 2 ) operations, the fast sinc 2 -transform, described below, makes use of the NUFFT to reduce the cost to O(N log N).

39 Basic approaches for Fourier Inversion Scheme 2 (Pseudo-inverse) The minimum-norm least-squares solution, ˆρ(x), ˆρ(x) = H + s = H (HH ) + s, (46) can then be computed in two steps: 1 Solve the system for M mn := (HH ) mn = sinc(k m k n ), Ma = s where (47) 2 Compute ˆρ(x) using a ˆρ(x) = H a. The matrix M may be ill-conditioned, so pseudoinverse construction of a = M + s costs O(N 3 ) work using the SVD. The cost of conjugate gradient method is O(J N 2 ) or O(J N log N) with the fast Sinc-Transformation, where J denotes the number of iterations. Remark: Note that ρ(r) n s(n)e 2πık(n) r w n (44) can be written as ρ(r) H W s and W can be viewed as a diagonal approximation of M +.

40 Basic approaches for Fourier Inversion MR Image Reconstruction Image reconstruction using k-space sampling truncated at K max = 64. Figure: Cartesian (top) and spiral (bottom) Optimal weight reconstruction (left) pseudoinverse approximation using 5 iterations (right)

41 Conclusions We have presented a simple version of the type-3 nonuniform FFT. It can be used to approximate the continuous Fourier transform when neither the spatial nor the Fourier domain spacing is regular. The NUFFT type 3 allows for the evaluation of more general integral operators such as (31) by embedding them in a higher dimensional space. We have constructed a fast algorithm for the (discrete) sinc and sinc 2 transforms which have immediate application in MR image reconstruction. We expect that the algorithms described here will be of fairly broad utility since nufft and sinc-convolution arises naturally in many signal and image processing contexts.

42 References H. Choi and D. C. Munson, Analysis and design of minimax-optimal interpolators, IEEE Trans. Signal Processing 46 (1998), A. Dutt and V. Rokhlin, Fast Fourier tramsforms for nonequispaced data, SIAM J. Sci. Comput. 14 (1993), J. A. Fessler and B. P. Sutton, Nonuniform Fast Fourier transforms using min-max interpolation, IEEE Trans. Signal Proc. 51 (2003), L. Greengard and J.-Y. Lee, Accelerating the Nonuniform Fast Fourier Transform, SIAM Rev., 46 (2004), S. Inati, J.-Y. Lee, L. Fleysher, R. Fleysher, and L. Greengard, Optimal Least Squares Reconstruction of Magnetic Resonance Images from Non-uniform Samples in k-space, in preparation. J.-Y. Lee, and L. Greengard, The type 3 Nonuniform FFT and its Applications, J. Comput. Phys., 206 (2005), 1-5. D. Potts, G. Steidl, M. Tasche, Fast Fourier transforms for nonequispaced data: A tutorial, Modern Sampling Theory, , Birkhauser Boston, Applied and Numerical Harmonic Analysis Series, (2001).

Direct Methods for Reconstruction of Functions and their Edges from Non-Uniform Fourier Data

Direct Methods for Reconstruction of Functions and their Edges from Non-Uniform Fourier Data Direct Methods for Reconstruction of Functions and their Edges from Non-Uniform Fourier Data Aditya Viswanathan aditya@math.msu.edu ICERM Research Cluster Computational Challenges in Sparse and Redundant

More information

Problem descriptions: Non-uniform fast Fourier transform

Problem descriptions: Non-uniform fast Fourier transform Problem descriptions: Non-uniform fast Fourier transform Lukas Exl 1 1 University of Vienna, Institut für Mathematik, Oskar-Morgenstern-Platz 1, A-1090 Wien, Austria December 11, 2015 Contents 0.1 Software

More information

Numerical Approximation Methods for Non-Uniform Fourier Data

Numerical Approximation Methods for Non-Uniform Fourier Data Numerical Approximation Methods for Non-Uniform Fourier Data Aditya Viswanathan aditya@math.msu.edu 2014 Joint Mathematics Meetings January 18 2014 0 / 16 Joint work with Anne Gelb (Arizona State) Guohui

More information

Constructing Approximation Kernels for Non-Harmonic Fourier Data

Constructing Approximation Kernels for Non-Harmonic Fourier Data Constructing Approximation Kernels for Non-Harmonic Fourier Data Aditya Viswanathan aditya.v@caltech.edu California Institute of Technology SIAM Annual Meeting 2013 July 10 2013 0 / 19 Joint work with

More information

Fourier Reconstruction from Non-Uniform Spectral Data

Fourier Reconstruction from Non-Uniform Spectral Data School of Electrical, Computer and Energy Engineering, Arizona State University aditya.v@asu.edu With Profs. Anne Gelb, Doug Cochran and Rosemary Renaut Research supported in part by National Science Foundation

More information

Nuclear Magnetic Resonance Imaging

Nuclear Magnetic Resonance Imaging Nuclear Magnetic Resonance Imaging Jeffrey A. Fessler EECS Department The University of Michigan NSS-MIC: Fundamentals of Medical Imaging Oct. 20, 2003 NMR-0 Background Basic physics 4 magnetic fields

More information

MRI beyond Fourier Encoding: From array detection to higher-order field dynamics

MRI beyond Fourier Encoding: From array detection to higher-order field dynamics MRI beyond Fourier Encoding: From array detection to higher-order field dynamics K. Pruessmann Institute for Biomedical Engineering ETH Zurich and University of Zurich Parallel MRI Signal sample: m γκ,

More information

Biophysics 230: Nuclear Magnetic Resonance Haacke Chapter 11

Biophysics 230: Nuclear Magnetic Resonance Haacke Chapter 11 Biophysics 230: Nuclear Magnetic Resonance Haacke Chapter 11 Daniel B. Rowe, Ph.D. daniel.rowe@marquette.edu Department of Math, Stat, Comp Sci Marquette University dbrowe@mcw.edu Department of Biophysics

More information

Spectral Broadening Mechanisms

Spectral Broadening Mechanisms Spectral Broadening Mechanisms Lorentzian broadening (Homogeneous) Gaussian broadening (Inhomogeneous, Inertial) Doppler broadening (special case for gas phase) The Fourier Transform NC State University

More information

Compensation. June 29, Arizona State University. Edge Detection from Nonuniform Data via. Convolutional Gridding with Density.

Compensation. June 29, Arizona State University. Edge Detection from Nonuniform Data via. Convolutional Gridding with Density. Arizona State University June 29, 2012 Overview gridding with density compensation is a common method for function reconstruction nonuniform Fourier data. It is computationally inexpensive compared with

More information

Name: INSERT YOUR NAME HERE. Due to dropbox by 6pm PDT, Wednesday, December 14, 2011

Name: INSERT YOUR NAME HERE. Due to dropbox by 6pm PDT, Wednesday, December 14, 2011 AMath 584 Name: INSERT YOUR NAME HERE Take-home Final UWNetID: INSERT YOUR NETID Due to dropbox by 6pm PDT, Wednesday, December 14, 2011 The main part of the assignment (Problems 1 3) is worth 80 points.

More information

Spectral Broadening Mechanisms. Broadening mechanisms. Lineshape functions. Spectral lifetime broadening

Spectral Broadening Mechanisms. Broadening mechanisms. Lineshape functions. Spectral lifetime broadening Spectral Broadening echanisms Lorentzian broadening (Homogeneous) Gaussian broadening (Inhomogeneous, Inertial) Doppler broadening (special case for gas phase) The Fourier Transform NC State University

More information

Moving interface problems. for elliptic systems. John Strain Mathematics Department UC Berkeley June 2010

Moving interface problems. for elliptic systems. John Strain Mathematics Department UC Berkeley June 2010 Moving interface problems for elliptic systems John Strain Mathematics Department UC Berkeley June 2010 1 ALGORITHMS Implicit semi-lagrangian contouring moves interfaces with arbitrary topology subject

More information

In this chapter we study elliptical PDEs. That is, PDEs of the form. 2 u = lots,

In this chapter we study elliptical PDEs. That is, PDEs of the form. 2 u = lots, Chapter 8 Elliptic PDEs In this chapter we study elliptical PDEs. That is, PDEs of the form 2 u = lots, where lots means lower-order terms (u x, u y,..., u, f). Here are some ways to think about the physical

More information

A Framework for Discrete Integral Transformations I the Pseudo-polar Fourier transform

A Framework for Discrete Integral Transformations I the Pseudo-polar Fourier transform A Framework for Discrete Integral Transformations I the Pseudo-polar Fourier transform A. Averbuch, R.R. Coifman, D.L. Donoho, M. Israeli, and Y. Shkolnisky June 26, 2007 This paper is dedicated to the

More information

EPI Bildgebung. German Chapter of ISMRM. Doktorantentraining. Freiburg 30 Mai - 1 Juni 2001

EPI Bildgebung. German Chapter of ISMRM. Doktorantentraining. Freiburg 30 Mai - 1 Juni 2001 EPI Bildgebung German Chapter of ISMRM Doktorantentraining Freiburg 30 Mai - 1 Juni 2001 Review of k-space and FT The EPI sequence Technical requirements Choice of the readout waveform ramp sampling regridding

More information

Linear second-order differential equations with constant coefficients and nonzero right-hand side

Linear second-order differential equations with constant coefficients and nonzero right-hand side Linear second-order differential equations with constant coefficients and nonzero right-hand side We return to the damped, driven simple harmonic oscillator d 2 y dy + 2b dt2 dt + ω2 0y = F sin ωt We note

More information

Part II: Magnetic Resonance Imaging (MRI)

Part II: Magnetic Resonance Imaging (MRI) Part II: Magnetic Resonance Imaging (MRI) Contents Magnetic Field Gradients Selective Excitation Spatially Resolved Reception k-space Gradient Echo Sequence Spin Echo Sequence Magnetic Resonance Imaging

More information

Lecture 7 January 26, 2016

Lecture 7 January 26, 2016 MATH 262/CME 372: Applied Fourier Analysis and Winter 26 Elements of Modern Signal Processing Lecture 7 January 26, 26 Prof Emmanuel Candes Scribe: Carlos A Sing-Long, Edited by E Bates Outline Agenda:

More information

Numerical stability of nonequispaced fast Fourier transforms

Numerical stability of nonequispaced fast Fourier transforms Numerical stability of nonequispaced fast Fourier transforms Daniel Potts and Manfred Tasche Dedicated to Franz Locher in honor of his 65th birthday This paper presents some new results on numerical stability

More information

Topics in Fourier analysis - Lecture 2.

Topics in Fourier analysis - Lecture 2. Topics in Fourier analysis - Lecture 2. Akos Magyar 1 Infinite Fourier series. In this section we develop the basic theory of Fourier series of periodic functions of one variable, but only to the extent

More information

Discrete Fourier Transform

Discrete Fourier Transform Last lecture I introduced the idea that any function defined on x 0,..., N 1 could be written a sum of sines and cosines. There are two different reasons why this is useful. The first is a general one,

More information

Spin-Warp Pulse Sequence

Spin-Warp Pulse Sequence Bioengineering 28A Principles of Biomedical Imaging Fall Quarter 25 Linear Systems Lecture Spin-Warp Pulse Sequence RF G x (t G y (t k y k x K-space trajectories EPI Spiral k y k y k x k x Credit: Larry

More information

Let R be the line parameterized by x. Let f be a complex function on R that is integrable. The Fourier transform ˆf = F f is. e ikx f(x) dx. (1.

Let R be the line parameterized by x. Let f be a complex function on R that is integrable. The Fourier transform ˆf = F f is. e ikx f(x) dx. (1. Chapter 1 Fourier transforms 1.1 Introduction Let R be the line parameterized by x. Let f be a complex function on R that is integrable. The Fourier transform ˆf = F f is ˆf(k) = e ikx f(x) dx. (1.1) It

More information

Fast Multipole Methods for The Laplace Equation. Outline

Fast Multipole Methods for The Laplace Equation. Outline Fast Multipole Methods for The Laplace Equation Ramani Duraiswami Nail Gumerov Outline 3D Laplace equation and Coulomb potentials Multipole and local expansions Special functions Legendre polynomials Associated

More information

Fourier transform. Stefano Ferrari. Università degli Studi di Milano Methods for Image Processing. academic year

Fourier transform. Stefano Ferrari. Università degli Studi di Milano Methods for Image Processing. academic year Fourier transform Stefano Ferrari Università degli Studi di Milano stefano.ferrari@unimi.it Methods for Image Processing academic year 27 28 Function transforms Sometimes, operating on a class of functions

More information

Fourier transforms, Generalised functions and Greens functions

Fourier transforms, Generalised functions and Greens functions Fourier transforms, Generalised functions and Greens functions T. Johnson 2015-01-23 Electromagnetic Processes In Dispersive Media, Lecture 2 - T. Johnson 1 Motivation A big part of this course concerns

More information

Numerical Methods I Orthogonal Polynomials

Numerical Methods I Orthogonal Polynomials Numerical Methods I Orthogonal Polynomials Aleksandar Donev Courant Institute, NYU 1 donev@courant.nyu.edu 1 Course G63.2010.001 / G22.2420-001, Fall 2010 Nov. 4th and 11th, 2010 A. Donev (Courant Institute)

More information

Convolution Theorem Modulation Transfer Function y(t)=g(t)*h(t)

Convolution Theorem Modulation Transfer Function y(t)=g(t)*h(t) MTF = Fourier Transform of PSF Bioengineering 28A Principles of Biomedical Imaging Fall Quarter 214 CT/Fourier Lecture 4 Bushberg et al 21 Convolution Theorem Modulation Transfer Function y(t=g(t*h(t g(t

More information

MATH 220 solution to homework 4

MATH 220 solution to homework 4 MATH 22 solution to homework 4 Problem. Define v(t, x) = u(t, x + bt), then v t (t, x) a(x + u bt) 2 (t, x) =, t >, x R, x2 v(, x) = f(x). It suffices to show that v(t, x) F = max y R f(y). We consider

More information

Computational Methods in Plasma Physics

Computational Methods in Plasma Physics Computational Methods in Plasma Physics Richard Fitzpatrick Institute for Fusion Studies University of Texas at Austin Purpose of Talk Describe use of numerical methods to solve simple problem in plasma

More information

Introduction to Practical FFT and NFFT

Introduction to Practical FFT and NFFT Introduction to Practical FFT and NFFT Michael Pippig and Daniel Potts Department of Mathematics Chemnitz University of Technology September 14, 211 supported by BMBF grant 1IH81B Table of Contents 1 Serial

More information

First-order overdetermined systems. for elliptic problems. John Strain Mathematics Department UC Berkeley July 2012

First-order overdetermined systems. for elliptic problems. John Strain Mathematics Department UC Berkeley July 2012 First-order overdetermined systems for elliptic problems John Strain Mathematics Department UC Berkeley July 2012 1 OVERVIEW Convert elliptic problems to first-order overdetermined form Control error via

More information

( ) f (k) = FT (R(x)) = R(k)

( ) f (k) = FT (R(x)) = R(k) Solving ODEs using Fourier Transforms The formulas for derivatives are particularly useful because they reduce ODEs to algebraic expressions. Consider the following ODE d 2 dx + p d 2 dx + q f (x) = R(x)

More information

The Application of FROID in MR Image Reconstruction

The Application of FROID in MR Image Reconstruction The Application of FROID in MR Image Reconstruction by Linda Vu A thesis presented to the University of Waterloo in fulfillment of the thesis requirement for the degree of Master of Applied Science in

More information

BME I5000: Biomedical Imaging

BME I5000: Biomedical Imaging BME I5000: Biomedical Imaging Lecture 9 Magnetic Resonance Imaging (imaging) Lucas C. Parra, parra@ccny.cuny.edu Blackboard: http://cityonline.ccny.cuny.edu/ 1 Schedule 1. Introduction, Spatial Resolution,

More information

A fast randomized algorithm for overdetermined linear least-squares regression

A fast randomized algorithm for overdetermined linear least-squares regression A fast randomized algorithm for overdetermined linear least-squares regression Vladimir Rokhlin and Mark Tygert Technical Report YALEU/DCS/TR-1403 April 28, 2008 Abstract We introduce a randomized algorithm

More information

Radiation by a dielectric wedge

Radiation by a dielectric wedge Radiation by a dielectric wedge A D Rawlins Department of Mathematical Sciences, Brunel University, United Kingdom Joe Keller,Cambridge,2-3 March, 2017. We shall consider the problem of determining the

More information

1 Assignment 1: Nonlinear dynamics (due September

1 Assignment 1: Nonlinear dynamics (due September Assignment : Nonlinear dynamics (due September 4, 28). Consider the ordinary differential equation du/dt = cos(u). Sketch the equilibria and indicate by arrows the increase or decrease of the solutions.

More information

Nov : Lecture 22: Differential Operators, Harmonic Oscillators

Nov : Lecture 22: Differential Operators, Harmonic Oscillators 14 Nov. 3 005: Lecture : Differential Operators, Harmonic Oscillators Reading: Kreyszig Sections:.4 (pp:81 83),.5 (pp:83 89),.8 (pp:101 03) Differential Operators The idea of a function as something that

More information

ECG782: Multidimensional Digital Signal Processing

ECG782: Multidimensional Digital Signal Processing Professor Brendan Morris, SEB 3216, brendan.morris@unlv.edu ECG782: Multidimensional Digital Signal Processing Spring 2014 TTh 14:30-15:45 CBC C313 Lecture 05 Image Processing Basics 13/02/04 http://www.ee.unlv.edu/~b1morris/ecg782/

More information

f (t) K(t, u) d t. f (t) K 1 (t, u) d u. Integral Transform Inverse Fourier Transform

f (t) K(t, u) d t. f (t) K 1 (t, u) d u. Integral Transform Inverse Fourier Transform Integral Transforms Massoud Malek An integral transform maps an equation from its original domain into another domain, where it might be manipulated and solved much more easily than in the original domain.

More information

Direct Sum. McKelvey, and Madie Wilkin Adviser: Dr. Andrew Christlieb Co-advisers: Eric Wolf and Justin Droba. July 23, Michigan St.

Direct Sum. McKelvey, and Madie Wilkin Adviser: Dr. Andrew Christlieb Co-advisers: Eric Wolf and Justin Droba. July 23, Michigan St. and Adviser: Dr. Andrew Christlieb Co-advisers: Eric Wolf and Justin Droba Michigan St. University July 23, 2014 The N-Body Problem Solar systems The N-Body Problem Solar systems Interacting charges, gases

More information

MAGIC058 & MATH64062: Partial Differential Equations 1

MAGIC058 & MATH64062: Partial Differential Equations 1 MAGIC58 & MATH6462: Partial Differential Equations Section 6 Fourier transforms 6. The Fourier integral formula We have seen from section 4 that, if a function f(x) satisfies the Dirichlet conditions,

More information

Convolution/Modulation Theorem. 2D Convolution/Multiplication. Application of Convolution Thm. Bioengineering 280A Principles of Biomedical Imaging

Convolution/Modulation Theorem. 2D Convolution/Multiplication. Application of Convolution Thm. Bioengineering 280A Principles of Biomedical Imaging Bioengineering 8A Principles of Biomedical Imaging Fall Quarter 15 CT/Fourier Lecture 4 Convolution/Modulation Theorem [ ] e j πx F{ g(x h(x } = g(u h(x udu = g(u h(x u e j πx = g(uh( e j πu du = H( dxdu

More information

Velocity k-space analysis of Flow Effects in Echo-Planar, Spiral and Projection Reconstruction Imaging. Sangeetha Somayajula

Velocity k-space analysis of Flow Effects in Echo-Planar, Spiral and Projection Reconstruction Imaging. Sangeetha Somayajula Velocity k-space analysis of Flow Effects in Echo-Planar, Spiral and Projection Reconstruction Imaging Sangeetha Somayajula Introduction Flow and motion in the object causes distortion in the MRI signal

More information

Fast and accurate methods for the discretization of singular integral operators given on surfaces

Fast and accurate methods for the discretization of singular integral operators given on surfaces Fast and accurate methods for the discretization of singular integral operators given on surfaces James Bremer University of California, Davis March 15, 2018 This is joint work with Zydrunas Gimbutas (NIST

More information

Lecture 4: Numerical solution of ordinary differential equations

Lecture 4: Numerical solution of ordinary differential equations Lecture 4: Numerical solution of ordinary differential equations Department of Mathematics, ETH Zürich General explicit one-step method: Consistency; Stability; Convergence. High-order methods: Taylor

More information

Topics. Bioengineering 280A Principles of Biomedical Imaging. Fall Quarter 2004 Lecture 4 2D Fourier Transforms

Topics. Bioengineering 280A Principles of Biomedical Imaging. Fall Quarter 2004 Lecture 4 2D Fourier Transforms Bioengineering 280A Principles of Biomedical Imaging Fall Quarter 2004 Lecture 4 2D Fourier Transforms Topics 1. 2D Signal Representations 2. 2D Fourier Transform 3. Transform Pairs 4. FT Properties 1

More information

L p -boundedness of the Hilbert transform

L p -boundedness of the Hilbert transform L p -boundedness of the Hilbert transform Kunal Narayan Chaudhury Abstract The Hilbert transform is essentially the only singular operator in one dimension. This undoubtedly makes it one of the the most

More information

Hamburger Beiträge zur Angewandten Mathematik

Hamburger Beiträge zur Angewandten Mathematik Hamburger Beiträge zur Angewandten Mathematik Error Estimates for Filtered Back Projection Matthias Beckmann and Armin Iske Nr. 2015-03 January 2015 Error Estimates for Filtered Back Projection Matthias

More information

APPLICATION OF THE SPARSE CARDINAL SINE DECOMPOSITION TO 3D STOKES FLOWS

APPLICATION OF THE SPARSE CARDINAL SINE DECOMPOSITION TO 3D STOKES FLOWS F. Alouges, et al., Int. J. Comp. Meth. and Ep. Meas., Vol. 5, No. 3 (07) 387 394 APPLICATION OF THE SPARSE CARDINAL SINE DECOMPOSITION TO 3D STOKES FLOWS F. ALOUGES, M. AUSSAL, A. LEFEBVRE-LEPOT, F. PIGEONNEAU

More information

The Fourier spectral method (Amath Bretherton)

The Fourier spectral method (Amath Bretherton) The Fourier spectral method (Amath 585 - Bretherton) 1 Introduction The Fourier spectral method (or more precisely, pseudospectral method) is a very accurate way to solve BVPs with smooth solutions on

More information

Fourier Series. (Com S 477/577 Notes) Yan-Bin Jia. Nov 29, 2016

Fourier Series. (Com S 477/577 Notes) Yan-Bin Jia. Nov 29, 2016 Fourier Series (Com S 477/577 otes) Yan-Bin Jia ov 9, 016 1 Introduction Many functions in nature are periodic, that is, f(x+τ) = f(x), for some fixed τ, which is called the period of f. Though function

More information

n 1 f n 1 c 1 n+1 = c 1 n $ c 1 n 1. After taking logs, this becomes

n 1 f n 1 c 1 n+1 = c 1 n $ c 1 n 1. After taking logs, this becomes Root finding: 1 a The points {x n+1, }, {x n, f n }, {x n 1, f n 1 } should be co-linear Say they lie on the line x + y = This gives the relations x n+1 + = x n +f n = x n 1 +f n 1 = Eliminating α and

More information

Introduction to Practical FFT and NFFT

Introduction to Practical FFT and NFFT Introduction to Practical FFT and NFFT Michael Pippig and Daniel Potts Faculty of Mathematics Chemnitz University of Technology 07.09.2010 supported by BMBF grant 01IH08001B Table of Contents 1 Serial

More information

6 The Fourier transform

6 The Fourier transform 6 The Fourier transform In this presentation we assume that the reader is already familiar with the Fourier transform. This means that we will not make a complete overview of its properties and applications.

More information

Fast Multipole BEM for Structural Acoustics Simulation

Fast Multipole BEM for Structural Acoustics Simulation Fast Boundary Element Methods in Industrial Applications Fast Multipole BEM for Structural Acoustics Simulation Matthias Fischer and Lothar Gaul Institut A für Mechanik, Universität Stuttgart, Germany

More information

17 The functional equation

17 The functional equation 18.785 Number theory I Fall 16 Lecture #17 11/8/16 17 The functional equation In the previous lecture we proved that the iemann zeta function ζ(s) has an Euler product and an analytic continuation to the

More information

4.6 Applications of Transforms to Differential Equations Establish the following result by formally inverting the Laplace transform.

4.6 Applications of Transforms to Differential Equations Establish the following result by formally inverting the Laplace transform. 4.6 Applications of Transforms to Differential Equations 285 19. Establish the following result by formally inverting the Laplace transform ˆF(s) = 1 s sinh(sy) sinh(sl) l> f (x) = y l + 2( 1) n nπ n=1

More information

Ordinary Least Squares and its applications

Ordinary Least Squares and its applications Ordinary Least Squares and its applications Dr. Mauro Zucchelli University Of Verona December 5, 2016 Dr. Mauro Zucchelli Ordinary Least Squares and its applications December 5, 2016 1 / 48 Contents 1

More information

The Chebyshev Fast Gauss and Nonuniform Fast Fourier Transforms and their application to the evaluation of distributed heat potentials

The Chebyshev Fast Gauss and Nonuniform Fast Fourier Transforms and their application to the evaluation of distributed heat potentials The Chebyshev Fast Gauss and Nonuniform Fast Fourier Transforms and their application to the evaluation of distributed heat potentials Shravan K. Veerapaneni George Biros March 14, 27 Abstract We present

More information

Conformal maps. Lent 2019 COMPLEX METHODS G. Taylor. A star means optional and not necessarily harder.

Conformal maps. Lent 2019 COMPLEX METHODS G. Taylor. A star means optional and not necessarily harder. Lent 29 COMPLEX METHODS G. Taylor A star means optional and not necessarily harder. Conformal maps. (i) Let f(z) = az + b, with ad bc. Where in C is f conformal? cz + d (ii) Let f(z) = z +. What are the

More information

Fast Algorithms for the Computation of Oscillatory Integrals

Fast Algorithms for the Computation of Oscillatory Integrals Fast Algorithms for the Computation of Oscillatory Integrals Emmanuel Candès California Institute of Technology EPSRC Symposium Capstone Conference Warwick Mathematics Institute, July 2009 Collaborators

More information

Numerical Analysis Preliminary Exam 10 am to 1 pm, August 20, 2018

Numerical Analysis Preliminary Exam 10 am to 1 pm, August 20, 2018 Numerical Analysis Preliminary Exam 1 am to 1 pm, August 2, 218 Instructions. You have three hours to complete this exam. Submit solutions to four (and no more) of the following six problems. Please start

More information

EE 367 / CS 448I Computational Imaging and Display Notes: Image Deconvolution (lecture 6)

EE 367 / CS 448I Computational Imaging and Display Notes: Image Deconvolution (lecture 6) EE 367 / CS 448I Computational Imaging and Display Notes: Image Deconvolution (lecture 6) Gordon Wetzstein gordon.wetzstein@stanford.edu This document serves as a supplement to the material discussed in

More information

A Fast Algorithm for Spherical Grid Rotations and its Application to Singular Quadrature

A Fast Algorithm for Spherical Grid Rotations and its Application to Singular Quadrature A Fast Algorithm for Spherical Grid Rotations and its Application to Singular Quadrature Zydrunas Gimbutas Shravan Veerapaneni September 11, 2013 Abstract We present a fast and accurate algorithm for evaluating

More information

K-space. Spin-Warp Pulse Sequence. At each point in time, the received signal is the Fourier transform of the object s(t) = M( k x

K-space. Spin-Warp Pulse Sequence. At each point in time, the received signal is the Fourier transform of the object s(t) = M( k x Bioengineering 280A Principles of Biomedical Imaging Fall Quarter 2015 MRI Lecture 4 k (t) = γ 2π k y (t) = γ 2π K-space At each point in time, the received signal is the Fourier transform of the object

More information

FOURIER TRANSFORMS. 1. Fourier series 1.1. The trigonometric system. The sequence of functions

FOURIER TRANSFORMS. 1. Fourier series 1.1. The trigonometric system. The sequence of functions FOURIER TRANSFORMS. Fourier series.. The trigonometric system. The sequence of functions, cos x, sin x,..., cos nx, sin nx,... is called the trigonometric system. These functions have period π. The trigonometric

More information

Contrast Mechanisms in MRI. Michael Jay Schillaci

Contrast Mechanisms in MRI. Michael Jay Schillaci Contrast Mechanisms in MRI Michael Jay Schillaci Overview Image Acquisition Basic Pulse Sequences Unwrapping K-Space Image Optimization Contrast Mechanisms Static and Motion Contrasts T1 & T2 Weighting,

More information

Recent Results for Moving Least Squares Approximation

Recent Results for Moving Least Squares Approximation Recent Results for Moving Least Squares Approximation Gregory E. Fasshauer and Jack G. Zhang Abstract. We describe two experiments recently conducted with the approximate moving least squares (MLS) approximation

More information

The formulas for derivatives are particularly useful because they reduce ODEs to algebraic expressions. Consider the following ODE d 2 dx + p d

The formulas for derivatives are particularly useful because they reduce ODEs to algebraic expressions. Consider the following ODE d 2 dx + p d Solving ODEs using Fourier Transforms The formulas for derivatives are particularly useful because they reduce ODEs to algebraic expressions. Consider the following ODE d 2 dx + p d 2 dx + q f (x) R(x)

More information

Sparse Fourier Transform via Butterfly Algorithm

Sparse Fourier Transform via Butterfly Algorithm Sparse Fourier Transform via Butterfly Algorithm Lexing Ying Department of Mathematics, University of Texas, Austin, TX 78712 December 2007, revised June 2008 Abstract This paper introduces a fast algorithm

More information

1 Distributions (due January 22, 2009)

1 Distributions (due January 22, 2009) Distributions (due January 22, 29). The distribution derivative of the locally integrable function ln( x ) is the principal value distribution /x. We know that, φ = lim φ(x) dx. x ɛ x Show that x, φ =

More information

Review of Fundamental Equations Supplementary notes on Section 1.2 and 1.3

Review of Fundamental Equations Supplementary notes on Section 1.2 and 1.3 Review of Fundamental Equations Supplementary notes on Section. and.3 Introduction of the velocity potential: irrotational motion: ω = u = identity in the vector analysis: ϕ u = ϕ Basic conservation principles:

More information

Lecture 9 February 2, 2016

Lecture 9 February 2, 2016 MATH 262/CME 372: Applied Fourier Analysis and Winter 26 Elements of Modern Signal Processing Lecture 9 February 2, 26 Prof. Emmanuel Candes Scribe: Carlos A. Sing-Long, Edited by E. Bates Outline Agenda:

More information

Nov : Lecture 18: The Fourier Transform and its Interpretations

Nov : Lecture 18: The Fourier Transform and its Interpretations 3 Nov. 04 2005: Lecture 8: The Fourier Transform and its Interpretations Reading: Kreyszig Sections: 0.5 (pp:547 49), 0.8 (pp:557 63), 0.9 (pp:564 68), 0.0 (pp:569 75) Fourier Transforms Expansion of a

More information

Estimates for the resolvent and spectral gaps for non self-adjoint operators. University Bordeaux

Estimates for the resolvent and spectral gaps for non self-adjoint operators. University Bordeaux for the resolvent and spectral gaps for non self-adjoint operators 1 / 29 Estimates for the resolvent and spectral gaps for non self-adjoint operators Vesselin Petkov University Bordeaux Mathematics Days

More information

Fast algorithms for integral formulations of steady-state radiative transfer equation

Fast algorithms for integral formulations of steady-state radiative transfer equation Fast algorithms for integral formulations of steady-state radiative transfer equation Yuwei Fan, Jing An, Lexing Ying February 11, 218 Abstract We investigate integral formulations and fast algorithms

More information

Electrodynamics PHY712. Lecture 4 Electrostatic potentials and fields. Reference: Chap. 1 & 2 in J. D. Jackson s textbook.

Electrodynamics PHY712. Lecture 4 Electrostatic potentials and fields. Reference: Chap. 1 & 2 in J. D. Jackson s textbook. Electrodynamics PHY712 Lecture 4 Electrostatic potentials and fields Reference: Chap. 1 & 2 in J. D. Jackson s textbook. 1. Complete proof of Green s Theorem 2. Proof of mean value theorem for electrostatic

More information

Chapter 8 Integral Operators

Chapter 8 Integral Operators Chapter 8 Integral Operators In our development of metrics, norms, inner products, and operator theory in Chapters 1 7 we only tangentially considered topics that involved the use of Lebesgue measure,

More information

e iωt dt and explained why δ(ω) = 0 for ω 0 but δ(0) =. A crucial property of the delta function, however, is that

e iωt dt and explained why δ(ω) = 0 for ω 0 but δ(0) =. A crucial property of the delta function, however, is that Phys 531 Fourier Transforms In this handout, I will go through the derivations of some of the results I gave in class (Lecture 14, 1/11). I won t reintroduce the concepts though, so you ll want to refer

More information

Numerical Solutions to Partial Differential Equations

Numerical Solutions to Partial Differential Equations Numerical Solutions to Partial Differential Equations Zhiping Li LMAM and School of Mathematical Sciences Peking University Numerical Methods for Partial Differential Equations Finite Difference Methods

More information

The Gaussians Distribution

The Gaussians Distribution CSE 206A: Lattice Algorithms and Applications Winter 2016 The Gaussians Distribution Instructor: Daniele Micciancio UCSD CSE 1 The real fourier transform Gaussian distributions and harmonic analysis play

More information

Analysis IV : Assignment 3 Solutions John Toth, Winter ,...). In particular for every fixed m N the sequence (u (n)

Analysis IV : Assignment 3 Solutions John Toth, Winter ,...). In particular for every fixed m N the sequence (u (n) Analysis IV : Assignment 3 Solutions John Toth, Winter 203 Exercise (l 2 (Z), 2 ) is a complete and separable Hilbert space. Proof Let {u (n) } n N be a Cauchy sequence. Say u (n) = (..., u 2, (n) u (n),

More information

A Randomized Algorithm for the Approximation of Matrices

A Randomized Algorithm for the Approximation of Matrices A Randomized Algorithm for the Approximation of Matrices Per-Gunnar Martinsson, Vladimir Rokhlin, and Mark Tygert Technical Report YALEU/DCS/TR-36 June 29, 2006 Abstract Given an m n matrix A and a positive

More information

ON INTERPOLATION AND INTEGRATION IN FINITE-DIMENSIONAL SPACES OF BOUNDED FUNCTIONS

ON INTERPOLATION AND INTEGRATION IN FINITE-DIMENSIONAL SPACES OF BOUNDED FUNCTIONS COMM. APP. MATH. AND COMP. SCI. Vol., No., ON INTERPOLATION AND INTEGRATION IN FINITE-DIMENSIONAL SPACES OF BOUNDED FUNCTIONS PER-GUNNAR MARTINSSON, VLADIMIR ROKHLIN AND MARK TYGERT We observe that, under

More information

Topics. Example. Modulation. [ ] = G(k x. ] = 1 2 G ( k % k x 0) G ( k + k x 0) ] = 1 2 j G ( k x

Topics. Example. Modulation. [ ] = G(k x. ] = 1 2 G ( k % k x 0) G ( k + k x 0) ] = 1 2 j G ( k x Topics Bioengineering 280A Principles of Biomedical Imaging Fall Quarter 2008 CT/Fourier Lecture 3 Modulation Modulation Transfer Function Convolution/Multiplication Revisit Projection-Slice Theorem Filtered

More information

Introduction to MRI. Spin & Magnetic Moments. Relaxation (T1, T2) Spin Echoes. 2DFT Imaging. K-space & Spatial Resolution.

Introduction to MRI. Spin & Magnetic Moments. Relaxation (T1, T2) Spin Echoes. 2DFT Imaging. K-space & Spatial Resolution. Introduction to MRI Spin & Magnetic Moments Relaxation (T1, T2) Spin Echoes 2DFT Imaging Selective excitation, phase & frequency encoding K-space & Spatial Resolution Contrast (T1, T2) Acknowledgement:

More information

APPROXIMATING THE INVERSE FRAME OPERATOR FROM LOCALIZED FRAMES

APPROXIMATING THE INVERSE FRAME OPERATOR FROM LOCALIZED FRAMES APPROXIMATING THE INVERSE FRAME OPERATOR FROM LOCALIZED FRAMES GUOHUI SONG AND ANNE GELB Abstract. This investigation seeks to establish the practicality of numerical frame approximations. Specifically,

More information

Books. B. Borden, Radar Imaging of Airborne Targets, Institute of Physics, 1999.

Books. B. Borden, Radar Imaging of Airborne Targets, Institute of Physics, 1999. Books B. Borden, Radar Imaging of Airborne Targets, Institute of Physics, 1999. C. Elachi, Spaceborne Radar Remote Sensing: Applications and Techniques, IEEE Press, New York, 1987. W. C. Carrara, R. G.

More information

I. Introduction. A New Method for Inverting Integrals Attenuated Radon Transform (SPECT) D to N map for Moving Boundary Value Problems

I. Introduction. A New Method for Inverting Integrals Attenuated Radon Transform (SPECT) D to N map for Moving Boundary Value Problems I. Introduction A New Method for Inverting Integrals Attenuated Radon Transform (SPECT) D to N map for Moving Boundary Value Problems F(k) = T 0 e k2 t+ikl(t) f (t)dt, k C. Integrable Nonlinear PDEs in

More information

Backprojection. Projections. Projections " $ & cosθ & Bioengineering 280A Principles of Biomedical Imaging. Fall Quarter 2014 CT/Fourier Lecture 2

Backprojection. Projections. Projections  $ & cosθ & Bioengineering 280A Principles of Biomedical Imaging. Fall Quarter 2014 CT/Fourier Lecture 2 Backprojection Bioengineering 280A Principles of Biomedical Imaging 3 0 0 0 0 3 0 0 0 0 0 3 0 0 3 0 3 Fall Quarter 2014 CT/Fourier Lecture 2 0 0 0 1 1 1 0 0 0 1 0 0 1 2 1 0 0 1 1 1 0 1 3 1 0 1 1 1 1 1

More information

2 Voltage Potential Due to an Arbitrary Charge Distribution

2 Voltage Potential Due to an Arbitrary Charge Distribution Solution to the Static Charge Distribution on a Thin Wire Using the Method of Moments James R Nagel Department of Electrical and Computer Engineering University of Utah, Salt Lake City, Utah April 2, 202

More information

ECG782: Multidimensional Digital Signal Processing

ECG782: Multidimensional Digital Signal Processing Professor Brendan Morris, SEB 3216, brendan.morris@unlv.edu ECG782: Multidimensional Digital Signal Processing Filtering in the Frequency Domain http://www.ee.unlv.edu/~b1morris/ecg782/ 2 Outline Background

More information

2 Fourier Transforms and Sampling

2 Fourier Transforms and Sampling 2 Fourier ransforms and Sampling 2.1 he Fourier ransform he Fourier ransform is an integral operator that transforms a continuous function into a continuous function H(ω) =F t ω [h(t)] := h(t)e iωt dt

More information

Discrete Fourier Transform

Discrete Fourier Transform Discrete Fourier Transform Virtually all practical signals have finite length (e.g., sensor data, audio records, digital images, stock values, etc). Rather than considering such signals to be zero-padded

More information

Wave packet decompositions adapted to (non-self-adjoint) operators

Wave packet decompositions adapted to (non-self-adjoint) operators 1 / 42 Wave packet decompositions adapted to (non-self-adjoint) operators Joe Viola Laboratoire de Mathématiques Jean Leray Université de Nantes Joseph.Viola@univ-nantes.fr 26 March 2018 2 / 42 Outline

More information

Ordinary Differential Equation Theory

Ordinary Differential Equation Theory Part I Ordinary Differential Equation Theory 1 Introductory Theory An n th order ODE for y = y(t) has the form Usually it can be written F (t, y, y,.., y (n) ) = y (n) = f(t, y, y,.., y (n 1) ) (Implicit

More information

TOOLS FROM HARMONIC ANALYSIS

TOOLS FROM HARMONIC ANALYSIS TOOLS FROM HARMONIC ANALYSIS BRADLY STADIE Abstract. The Fourier transform can be thought of as a map that decomposes a function into oscillatory functions. In this paper, we will apply this decomposition

More information