Yu et al. Vol. 29, No. 11 / November 2012 / J. Opt. Soc. Am.

A 2415

Band-limited angular spectrum numerical propagation

method with selective scaling of observation
window size and sample number

Xiao Yu, Tang Xiahui,* Qin Yingxiong, Peng Hao, and Wang Wei
Wuhan National Laboratory for Optoelectronics, National Engineering Research Center for Laser Processing, and School
of Optoelectronics Science and Engineering, Huazhong University of Science and Technology,
Wuhan, Hubei 430074, China
*Corresponding author: txh1116@mail.hust.edu.cn

Received July 12, 2012; revised September 24, 2012; accepted September 25, 2012;
posted September 27, 2012 (Doc. ID 172485); published October 22, 2012
Band-limited angular spectrum (BLAS) methods can be used for simulating the diffractional propagation in the
near field, the far field, the tilted system, and the nonparaxial system. However, it does not allow free sample
interval on the output calculation window. In this paper, an improved BLAS method is proposed. This new algo-
rithm permits a selective scaling of observation window size and sample number on the observation plane. The
method is based on the linear convolution, which can be calculated by fast Fourier transform effectively. © 2012
Optical Society of America
OCIS codes: 050.1940, 090.1995, 070.2025.

1. INTRODUCTION function [2,11]. The band-limited angular spectrum (BLAS)

Digital simulation of scalar diffraction is widely used for
studies like laser resonator mode analysis, digital holography,
and imaging. These simulations of diffraction are usually
based on the well-known scalar diffraction theories, such
as the angular spectrum (AS) method, the Rayleigh–
Sommerfeld formula, the Fresnel–Kirchhoff formula, and
the Fresnel and Fraunhofer approximation of the Fresnel–
Kirchhoff diffraction integral.
There are two main kinds of methods for simulating the
diffraction propagation between the parallel planes [1,2].
The first kind is based on the integration in the spatial
domain. The single-transform-based Fresnel method [3], the
convolution-based Fresnel method [3], the convolution-base
Rayleigh–Sommerfeld (CV-RS) method [4], and the scaling
convolution Rayleigh–Sommerfeld (SCV-RS) method [5] be-
long to this kind. The two-step Fresnel propagation method
[6,7] is an extension of the single-transform-based Fresnel
method. For the two-step Fresnel propagation method, the
single-transform-based Fresnel method is used twice and it
allows scaling of the window size but not sample number.
Also, this method is only valid for the propagation in the
Fresnel region. The CV-RS and SCV-RS methods can be eval-
uated by the numerical convolution. SCV-RS allows for inde-
pendent sample intervals but not sample numbers in the input
and out computation windows. In general, these simulation
methods are not suitable for the near-field simulation because
of the sampling problem [2]. Also, these kinds of methods
cannot be used for the propagation between the tilted
planes.
The methods associated with the AS propagation can be
used for the propagation between the tilted planes [8–10]. The
general AS method is valid for the near-field simulation
because of the sampling problem associated with the chirp
method proposed by the authors of [12] can be used for near
and far field propagation simulation due to the truncation of
the unnecessary high-frequency components of the input
spectrum. In the off-axis system, the shifted-AS method would
be effective [13]. The other advantage for the AS is that it is an
exact diffraction method for simulation of the very short pro-
pagation distances like tens of wavelengths. However, one
common drawback for the methods based on the AS propaga-
tion is that they cannot permit the independent sample size
and the calculation window size on the observation plane, be-
cause the method based on fast Fourier transform (FFT)
needs the same sample interval and calculation window size
on the source and observation plane.
In this paper, we propose a scaling BLAS method that fea-
tures validity for the independent sample size and the sample
number on the observation plane. This new method is based
on the linear convolution, which can be evaluated by fast
Fourier transform effectively. The sampling criterion for this
method is also discussed. The numerical verification and an
application example of the scaling BLAS are given.
2. PHYSICAL DIFFRACTION THEORY
The scalar diffraction theory of light is successfully applicable
in the majority of situations in diffractive optics. In the frame
of the scalar theory, the accurate diffraction formula is the
Rayleigh–Sommerfeld diffraction integral
ZZ

z 1
U
x; y; z  U
x1 ; y1 ; 0 exp
ikr − ik dx1 dy1 ; (1)
Σ r r
where r  
x − x1 2 
y − y1 2  z2 1=2 in Fig. 1. k is the wave
number, z is the propagation distance
x1 ; y1 , and U
x1 ; y1 ; 0
are the coordinate and the light field at the input aperture,

1084-7529/12/112415-06$15.00/0 © 2012 Optical Society of America

2416 J. Opt. Soc. Am. A / Vol. 29, No. 11 / November 2012 Yu et al.

Fig. 1. Coordinate systems and the geometry of the model.

Fig. 2. One-dimensional geometric parameters.

x; y and U
x; y; z stand for the coordinate and the light field
on the observation plane, respectively. One important
observation about the diffraction integral above, Eq. (1), is
that the integral is in the form of a two-function convolution.
Then the convolution is rewritten using the convolution
theorem as
A
u; v; z  A
u; v; 0G
u; v; z; (2)
PN=2−1
where A1
un ; 0  m−N=2 U
x1 ; 0 exp
−i2πmΔx1 un  · Δx1
is the discrete Fourier spectrum of the input light field and
un  nΔu, for n  −N=2; …; N=2 − 1 and Δu  1=Lx . Then
the field on the sample points of xm on the observation plane
is given as the summation of the contribution for all the input
spectrum components,

X
N=2−1
where the spectrum, A
u; v; 0, and the transfer function, U
xm ; z  A
un ; z exp
i2πun xm  · Δu;
G
u; v; z, are given by n−N=2

ZZ M M
for xm  mΔx and m  − ; …; − 1: (7)
A
u; v; 0  U
x1 ; y1 ; 0 exp−i2π
ux1  vy1 dx1 dy1 ; (3) 2 2

and Note we cannot use the inverse discrete Fourier transfor-

mation to calculate the summation of Eq. (7), because of the
G
u; v; z  expi2πz
λ−2 − u2 − v2 1=2 : (4) difference of the sample number and the sample interval in the
input and output planes (M ≠ N and Δx ≠ Δx1 ).
A
u; v; 0 stands for the spectrum distribution of the input To overcome this problem, a scaling factor α  Δx=Δu is
field. G
u; v; z is the transfer function. Then the light field introduced into Eq. (7). Then
U
x; y; z at the observation plane can be solved as the
 
two-dimensional (2D) inverse Fourier transformation of 2π
A
u; v; z, given as exp
i2πun xm   exp i
−αun xm 
−α
   
ZZ π π
U
x; y; z  A
u; v; z exp
i2π
ux  vydudv:  exp i
αun − xm 2 exp i 
αun 2  x2m  :
(5) −α α
(8)

Substituting Eq. (8) for Eq. (7), then we can get

3. DISCRETE COMPUTATION ALGORITHM
The aim of our algorithm is to use the AS method for the pro-
 N=2−1
X 1  
π π
blems that the sample interval in the source plane is different U
xm ; z  exp i x2m A
un ; z exp i
αun 2
from that on the observation plane. For simplicity, we just dis- α n−N=2
α α
cuss a one-dimensional propagation between the aligned and  
−iπ
parallel plane, as shown in Fig. 2. The extension of this meth- × exp
αun − xm 2 ·
α · Δu
α
od for the diffraction simulation pattern of 2D planar objects is

 X
N=2−1
straightforward. In Fig. 2, W x is the width of the input aper- π
 exp i x2m · Δw B
wn 
ture, Lx is the width of the calculation window, S x denotes the α n−N=2
width of the observation aperture, and z is the propagation  
distance. −iπ
× exp
wn − xm 2
Let N and M be the sample number in the source plane α


and the observation plane, respectively, and the sample inter- π
val in the source region and the observation region is Δx1 and  exp i x2m · ΔwB
wn   f
wn ; (9)
α
Δx, respectively. The discrete form of A
u; v; z in the one-
dimensional system is written as where B
wn  
1=αA
un ; z expi
π=α
αun 2 , f
wn  
exp

−iπ=αw2n , wn  αun , Δw  Δx  αΔu and “” de-
A
un ; z  A1
un ; 0 expi2πz
λ−2 − u2n 1=2 ; (6) notes the linear convolution. In Eq. (9) the summation of
Yu et al. Vol. 29, No. 11 / November 2012 / J. Opt. Soc. Am. A 2417

PN=2−1
n−N=2 B
wn  exp
−iπ=α
wn − xm 2  is a linear convolution
of B
wn ; z and f
wn   exp

−iπ=αw2n , which can be effec-
tively evaluated using FFT. And we can choose an indepen-
dent sample interval (Δx) and calculation window size (MΔx)
for the output plane in the algorithm because, for the linear
convolution, the sample number of M and N has no need to
be the same. When the scaling factor is chosen as α  Δx=Δu,
the ratio between the sample interval on the source plane and
the observation plane is Δx1 =Δx  NΔx21 =α.

4. DISCUSSION
A. Computational Window at the Source Plane
The computational window size in the aperture plane is an
important simulation parameter for a correct AS simulation.
The computational window size influences the calculation ac-
curacy due to the characteristic of the discrete Fourier trans-
form (more precisely, the discrete Fourier series (DTFS) [14]).
The DTFS can only describe the discrete periodic function.
The sample of the input field would result in modulated Fig. 4. (Color online) Aliasing error from the replicas of the physical
window.
and translated replicas of the physical input window at peri-
odic locations at the source plane [15,16]. Therefore, DTFS
based on FFT actually solves the problem of an infinite array replicas of the physical window can influence the field distri-
of identical, iterated cells located side by side with spacing bution on the observation window. When the spatial frequency
equal to the full width of the sample grid, as shown in Fig. 3. is higher than sin
θa =λ, some part of power emitted by the
Out of the aperture, there must be an adequate zero- aliasing window aside the physical one would come into the
padding band in order to avoid the energy diffracted from observation aperture, which would cause the numerical error
the aliasing windows traveling into the physical observation in the observation plane. To avoid the spectrum components
window. The aliasing error caused by the replicas of the com- from the replicas appearing in the observation window, the
putational window is shown in Fig. 4. The width of the sample length of zero padding should be chosen large enough to main-
window is Lx . At the center of the sample window, it is the tain that θa > θn . In Fig. 4, we can see that the length of the zero
input aperture with size W x . The width of the observation area padding (Lx − W x ) should be chosen to be equal or larger than
on the destination plane is S x . The propagation distance is z. that of the observation aperture to make sure the aliasing spec-
In Fig. 4, θn stands for the largest angle that the plane wave trum would not spill over into the observation window. Then
emitted in the source plane can physically influence the field the size computational region in the source plane is given as
of the observation aperture [17]. It represents the propagation
direction of the plane wave of the highest spatial frequency,
Lx  W x  S x : (10)
which is emitted from the point at the lower end of the source
aperture and observed at the upper end of the observation
window. Then, sin
θn =λ denotes the highest frequency which To get discrete Fourier spectrum of the input field, the input
is needed for the diffraction propagation. In Fig. 4, θa is the signals should first be cut off by a window of finite size. Using
smallest angle that the aliasing spectrum emitted by the a window of finite size would cause the spectrum leakage

Fig. 3. (Color online) Replicas of the physical window in the input plane.
2418 J. Opt. Soc. Am. A / Vol. 29, No. 11 / November 2012 Yu et al.

error [18]. The length and the type of the window would in- (i) for S x ≥ W x , umax  uneed ,
fluence the accuracy of the discrete Fourier spectrum. The (ii) for S x < W x , umax > uneed ,
authors of [16] use a cosine window. Here, we use the rectan-
which means Eq. (15) is always satisfied. Then the final
gular window. And the length of the calculation window Lx is
band-limited transfer function is given as
given by [19]

Lx  hW x ; h  2 or 3: (11) G
u; z  expi2πz
λ−2 − u2 1=2 rect
u=
2uneed : (16)

When h  2, it means that the width of the zero-padding area Note that our sample method is different from that of [12],
is the same to that of the source aperture. The larger h we because the authors of [12] do not consider the aliasing infor-
chose, the less spectrum leakage effect (windowing effect) mation from the replica window.
would be introduced into the discrete Fourier spectrum of the
input light field and the higher resolution in frequency domain D. Computational Complexity
would be gotten because the sample interval in frequency do- The computational load of the scaling BLAS method in Eq. (9)
main is 1=Lx . In this way, DTFS can describe the spectrum of comes from a FFT of the input field, calculation of Eq. (16) for
the continuous input light field correctly. From Eqs. (10) and G
u; z, and a linear convolution and two scaling course for the
(11), the calculation window size Lx should be shown as transfer function. The calculation burden associated with the
FFT-based linear convolution is O

M  N 1 log2
M  N 1 . M
Lx  max
W x  S x ; hW x : (12) is the sample number on the input source plane. N 1 is equal to
uneed =Δu, which stands for the effective length of sample in the
B. Needed Bandwidth for AS Propagation
When the AS method is used to simulate the propagation be-
tween the source plane and the finite-size object screen, a lim-
ited bandwidth of the input spectrum is needed. The minimum
bandwidth of the spectrum which is needed for exact numer-
ical propagation is determined by the geometric structure of
the problem [2,17]. As shown in Fig. 4, The highest frequency
needed for the propagation simulation is sin
θn =λ. Then the
effective bandwidth for the input spectrum is 2 sin
θn =λ.
After some calculations, the needed band of spectrum can be
written as follows [12]:

1
uneed  : (13)
λ
2z=
W x  S x 2  11=2

C. Avoiding the Aliasing Error from Sampling the Chirp

Functions
The transform function of Eq. (4) for the AS method is a kind
of chirp function. It contains a complex exponential term with
a phase function whose signal frequency increases as the
square of coordinate value of the position increases. Sampling
a chirp function can be problematic because it is not band-
limited. According to Nyquist sampling theorem, the maxi-
mum bandwidth which can be correctly sampled is given
as [2,12]

1
umax ≤ ; (14)
λ
2zΔu2  11=2

where Δu is equal to 1=Lx , the sample interval in frequency

space. The meaning of Eq. (14) is that when the sample inter-
val in frequency space Δu is chosen, the bandwidth of the
spectrum which is correctly sampled is only 2 umax [12].
Therefore, for a correct sample, we should maintain the rela-
tion that

umax ≥ uneed : (15)

Fig. 5. Setup for (a) numerical simulation of the scaling BLAS meth-
From Eqs. (13) and (14), the relation between umax and uneed od, (b) the amplitude distribution of the input field, and (c) the phase
can be given as distribution of the input field.
Yu et al.
Fig. 6. (Color online) Simulation result for the normal BLAS method and the scaling BLAS method.
frequency space. Then, for a two dimensional problem, the
total computational complexity is
C  O
N 2 log2 N  O

M  N 1 2 log2
M  N 1 
 O

M  N 1 2   O
M 2   O
N 2 :
5. NUMERICAL VERIFICATION AND
APPLICATION
For many optical applications, the different sample interval in
the source plane and the observation plane is needed. For ex-
ample, the field distribution of the focus point is of interest for
the laser material processing and optical signal dealing. In re-
cent years the ring mode laser (optical vortices) beam has re-
cently attracted significant attention for their application,
especially for its field distribution at the focus point of the
lens. Here we give a simulation example about the calculation
of the field of a Gaussian beam focused by a lens. When the
field distribution at the focus point is considered, the different
sample size between the source and the observation plane is
needed, because the size of the focus point is generally very
small. The optical setup for an application example of the scal-
ing BLAS is shown in as shown in Fig. 5(a). We assume the
beam width W x is 30 mm and the computational size is
Lx  3W x . The sample number on the source plane N  1500
is and the sample number on the observation plane is M 
N=2  750. We just change the sample interval Δx in the ob-
servation plane for the different observation window size of
S x  M · Δx. The focal length f is 100 mm. The wavelength
λ is 10.6 μm. The Fresnel number W 2x =
4λf  of the system is
210, which means that the Fresnel diffraction integral is not
suitable for the simulation and the strict AS method is valid
[20]. The input field distribution before the lens is shown in
Figs. 5(b) and 5(c). It is a TEM01 -mode field, which is given
by [21,22]
A1  A0 r exp
−r 2 =2a2  exp
iθ;
Vol. 29, No. 11 / November 2012 / J. Opt. Soc. Am. A 2419
where A0 is the normalized constant and a  5 mm. Then the
field after the lens is given by A2  A1 exp
−i
k=2f 
x21  y21 ,
where
k=2f 
x21  y21  stands for the phase factor of the
lens [23].
The scaling BLAS method is used for the propagation be-
tween the plane after the lens and the focus plane. The simula-
tion result is shown in Fig. 6. In Fig. 6(a), the result for the BLAS
method is shown. It is obvious that, for the BLAS method, it is
hard to describe the detailed field distribution in the focus
point. The field in the focus is just limited in a very narrow re-
gion, which means that the size of focus point is very small.
When Δx1 =Δx  1, the size of focus point is several times
the sample interval in the observation plane, which means
there are only several sample points in the focus point region.
Thus, the details about the field distribution in the focus point
region cannot be told. The simulation results for Δx1 =Δx  10
and Δx1 =Δx  200 are shown in Figs. 6(b) and 6(c), respec-
tively. In Fig. 6(b), some details about the focus point are
shown. And only a small part of sample point is in the region
of the focus point. In Fig. 6(c), when Δx1 =Δx  200, the full
details about the focus point can be told, because all the sample
points on the observation plane are limited in the region of the
focus point.
6. CONCLUSION
A numerical method called a scaling BLAS method is proposed.
It makes use of the convolution theorem to calculate the field
distribution on the observation region. The convolution-
based method can be evaluated by FFT effectively. For the
scaling BLAS method, the window size and the sample number
in the observation can be chosen independently. This new
method makes it possible to calculate the field on the observa-
tion window of different size from the input window.
ACKNOWLEDGMENTS
This work was supported by the National Natural Science
Foundation of China (NSFC) under grant 60808005, the
NSFHB under grant 2008CDB327, and the NKTRDP under
grant 2007BAF11B01.
2420 J. Opt. Soc. Am. A / Vol. 29, No. 11 / November 2012
REFERENCES
1. D. Mendlovic, Z. Zalevsky, and N. Konforti, “Computation con-
siderations and fast algorithms for calculating the diffraction
integral,” J. Mod. Opt. 44, 407–414 (1997).
2. D. G. Voelz and M. C. Roggemann, “Digital simulation of scalar
optical diffraction revisiting chirp function sampling criteria and
consequences,” Appl. Opt. 48, 6132–6142 (2009).
3. T. M. Kreis, M. Adams, and W. P. O. Jüptner, “Methods of digital
holography: a comparison,” Proc. SPIE 3098, 224–233 (1997).
4. F. Shen and A. Wang, “Fast-Fourier-transform based numerical
integration method for the Rayleigh–Sommerfeld diffraction for-
mula,” Appl. Opt. 45, 1102–1110 (2006).
5. V. Nascov and P. C. Logofătu, “Fast computation algorithm for
the Rayleigh–Sommerfeld diffraction formula using a type of
scaled convolution,” Appl. Opt. 48, 4310–4319 (2009).
6. J. D. Schmidt, Numerical Simulation of Optical Wave Propaga-
tion With Examples in MATLAB (SPIE, 2010).
7. C. Rydberg and J. Bengtsson, “Efficient numerical representa-
tion of the optical field for the propagation of partially coherent
radiation with a specified spatial and temporal coherence func-
tion,” J. Opt. Soc. Am. A 23, 1616–1625 (2006).
8. T. Tommasi and B. Bianco, “Frequency analysis of light diffrac-
tion between rotated planes,” Opt. Lett. 17, 556–558 (1992).
9. T. Tommasi and B. Bianco, “Computer-generated holograms of
tilted planes by a spatial frequency approach,” J. Opt. Soc. A 10,
299–305 (1993).
10. K. Matsushima, H. Schimmel, and F. Wyrowski, “Fast calcula-
tion method for optical diffraction on tilted planes by use of
the angular spectrum of plane waves,” J. Opt. Soc. A 20,
1755–1762 (2003).
Yu et al.
11. D. Mendlovic, Z. Zalevsky, and N. Konforti, “Computation con-
siderations and fast algorithms for calculating the diffraction
integral,” J. Mod. Opt. 44, 407–414 (1997).
12. K. Matsushima and T. Shimobaba, “Band-limited angular spec-
trum method for numerical simulation of free-space propagation
in far and near fields,” Opt. Express 17, 19662–19673 (2009).
13. K. Matsushima, “Shifted angular spectrum method for off-axis
numerical propagation,” Opt. Express 18, 18453–18463
(2010).
14. S. Haykin and B. Van Veen, Signals and Systems (Wiley, 2005).
15. L. Onural, “Exact analysis of the effects of sampling of the scalar
diffraction field,” J. Opt. Soc. A 24, 359–367 (2007).
16. E. A. Sziklas and A. E. Siegman, “Mode calculations in unstable
resonators with flowing saturable gain. 2: fast Fourier transform
method,” Appl. Opt. 14, 1874–1889 (1975).
17. D. Wang, J. Zhao, F. Zhang, G. Pedrini, and W. Osten, “High-
fidelity numerical realization of multiple-step Fresnel propaga-
tion for the reconstruction of digital holograms,” Appl. Opt. 47,
D12–D20 (2008).
18. C. L. Phillips, J. M. Parr, and E. Ann Riskin, Signals, Systems,
and Transforms (Pearson, 2003).
19. D. G. Voelz, Computational Fourier Optics: A MATLAB
Tutorial (SPIE, 2011).
20. N. Hodgson and H. Weber, Laser Resonators and Beam Propa-
gation (Springer, 2005).
21. J. Leach, M. J. Padgett, S. M. Barnett, S. Franke-Arnold, and
J. Courtial, “Measuring the orbital angular momentum of a single
photon,” Phys. Rev. Lett. 88, 257901 (2002).
22. J. Curtis and D. G. Grier, “Structure of optical vortices,” Phys.
Rev. Lett. 90, 133901 (2003).
23. J. W. Goodman, Introduction to Fourier Optics (Roberts, 2005).