Uld be obtained very easily inside the frequency domain. JPH203 custom synthesis within this case, the Fourier transforms on the ^ dual-frame vectors, n (k ), where k denotes the frequency variable, are provided by [10] n (k ) = n (k ) nn (k )(five)3. Signal Reconstruction from Redundant and Non-Uniformly Spaced Samples three.1. Redundant and Non-Uniformly Spaced Samples of Bandlimited Functions as Frame Coefficients As outlined by the Nyquist hannon sampling theorem, non-redundant and uniformly spaced samples, obtained at, tn = nT, of bandlimited signals with bandwidths, [-/T, /T ], could possibly be viewed as the coefficients representing these signals in an orthonormal basis,basis n (t) n Z[10]. The basis vectors are given bybasis n (t) = T -1/sin( (t – nT )/T ) (t – nT )/T(6)exactly where T is definitely the Nyquist sampling 5-Methyltetrahydrofolic acid web interval in the provided space of bandlimited functions. Similarly, redundant and non-uniformly spaced samples obtained at tn of bandlimited signals whose bandwidths are included in [-/T, /T ], may be viewed as the coefficients resulting from representing these signals inside a frame, n (t – tn )nZ . The frame vectors are offered by sin( (t – tn )/T ) (7) n (t) = n T -1/2 (t – tn )/T where tn may be the arbitrary place on the nth sample, n = [(tn+1 – tn-1 )/2T ]1/2 , and T is once once again the Nyquist sampling interval inside the given space of bandlimited functions. If the maximum sampling distance satisfies = max|tn+1 – tn | Tn Z(8)then the frame is bounded by A (1 – /T )two and B (1 + /T )two . We note that T/ represents the oversampling ratio, relative for the Nyquist sampling rate, and also the scale element, n , in Equation (7) accounts for this oversampling and location irregularities of your sampled points. 3.two. Frame-Based Reconstruction of an OCT A-Scan The reconstruction of an OCT A-scan, u(z), exactly where z will be the axial spatial domain variable, from its redundant and non-uniformly spaced frequency domain (k-space) samples, isSensors 2021, 21,4 of^ ^ equivalent to reconstructing its Fourier transform, u(k), from its frame coefficients, u(k n ) = ^ u, n (k – k n ) , followed by an inverse Fourier transform. Consequently, applying Equation (four) ^ u(k) = ^ u(kn ) n (k – kn ) (9)n ZFor ease of implementation, as an alternative to applying the inverse Fourier transform, we get a flipped version with the A-scan by applying the forward transform to Equation (9). By utilizing Equation (5) we get u(-z) =n Zu(k n )^ n (z) ^ n n (z)(10)exactly where z may be the axial spatial domain variable. The Fourier transforms on the frame vectors corresponding to Equation (7), but inside the frequency domain, are provided by ^ ^ n (z) = F n (k – k n ) = n 0 (z)e- jkn z (11)^ where 0 (z) could be the Fourier transform from the unshifted frame vector 0 (k) = K -1/2 sin(k/K )/ (k/K ), exactly where K is definitely the Nyquist sampling interval within the frequency domain. Substituting Equation (11) into Equation (10), we’ve got u(-z) = ^ 0 (z) ^ n n (z)n Zn^ u(k n )e- jkn z(12)Because the Fourier transform of sin(/( is usually a rectangular window, and assuming a finite number of frame expansion coefficients, N f , then replacing the continuous spatial variable, z, in Equation (12) by its corresponding uniformly sampled spatial variable, zs = 0, 1, . . . , NT – 1, we could approximate u(-zs ) as 1 u(-zs ) = NTN f -1 n =^ [n u(k n )]e- jkn zs(13)exactly where NT would be the number of Nyquist samples, NT = N f /T. Equation(13) is often a scaled version from the non-uniform discrete Fourier transform (NDFT) at arbitrary points k n , as opposed to the common NDFT employed for OCT reconstruction in [9,11,12]. We note that,.