We develop a theory for understanding and modeling the effects of mid-spatial frequency (MSF) structures on the modulation transfer function (MTF) of an imaging system. We show that the MTF is related through Fourier transformation to what we call the pupil-difference probability density (PDPD) function. This relation is illustrated for several one-dimensional periodic groove patterns. We also consider two-dimensional pupils, particularly those presenting straight and circular concentric periodic surface errors, similar to those resulting from diamond turning and milling processes of freeform optical manufacturing, and develop simple approximations for the MTF based on the PDPD. Our approach provides an analytic way to understand the effects of tool profiles, as well as other freeform manufacturing parameters.
© 2017 Optical Society of America
The ability to more finely control subaperture tools used for the creation and finishing/polishing of optical surfaces has been linked to the possibility of generating surface shapes of greater complexity. Such freeform surfaces provide the opportunity of improvements in system volume and performance. However, subaperture tools are inevitably accompanied by mid-spatial frequency (MSF) error structures. With characteristic spatial frequencies greater than those of the typical low-order Zernike aberrations and lower than those of surface roughness, MSF errors pose a range of problems that include complicating the ability to set effective tolerances for surface quality [1–4].
In this work, we provide a new approach for understanding the effects of MSF structures and other errors on the performance of an optical system, characterized by its optical transfer function (OTF) and its magnitude, the modulation transfer function (MTF). We start by following similar ideas to those presented previously for another performance metric, the Strehl ratio , under the standard perturbation model . The statistics of MSF structures on a given optical surface are gathered in what we refer to as the pupil-difference probability density (PDPD) function, which is the probability density for the aberration difference between any two pupil points with prescribed separation. The link between the PDPD and the OTF, for all wavelengths, is then given by a Fourier relation. This framework is used for analyzing several simple one-dimensional periodic groove patterns. These ideas are then extended to two-dimensional structures, and approximate results are found for the cases of parallel grooves (like those resulting from diamond milling) and concentric circular grooves (such as those resulting from diamond turning). This framework is novel in that it provides mathematical intuition on how surface errors resulting from manufacturing parameters such as tool profile and path affect the degradation of an optical system’s performance. We also comment on the effect of concatenating several types of error.
2. Pupil Difference Probability Density
For a given wavenumber, k = 2π/λ, the optical transfer function (OTF) of a system is defined as the autocorrelation of the system’s field at the exit pupil  due to an object point source:Eq. (1) by adopting the shorthand notation:
We now define the PDPD, denoted by P(η, ρ), as a probability density for the error ΔW with given pupil displacement ρ. That is, for given ρ, P(η, ρ)dη is the fraction of values of q over the overlap region for which the aberration difference ΔW (q, ρ) falls within the infinitesimal range [η, η + dη]. This function can be expressed mathematically as
By inserting unity in the form of ∫ δ[η − ΔW (q, ρ)] dη = 1 into the integrand of Eq. (2) and reversing the order of integration, the OTF can be rewritten as
The corresponding result for the MTF is
Equations (4) and (6) state the first main result of this work: for systems where W can be regarded as independent of wavelength (e.g. reflective systems), the OTF as a function of wavenumber k is simply the Fourier transform of the PDPD times the OTF of the unaberrated system. Note that for systems in which W does depend on wavelength, this formulation is still valid, i.e., Eq. (5) still applies. However, because P(η, ρ) then also depends on k, the inverse process of recovering the PDPD from the OTF by inverse Fourier transformation is not possible.
3. One-dimensional periodic groove structures
To illustrate how the PDPD gives a simple picture of MTF performance, we now consider four different periodic error patterns in one dimension that allow a simple analytic treatment: a binary rectangular groove with fill factor 1/2, a symmetric triangular groove, a sinusoidal groove, and a piecewise parabolic groove. These groove profiles are chosen because their simplicity helps illustrate the concepts introduced here and allow analytic solutions. In particular, the piecewise parabolic groove structure mimics the profile left behind by a diamond tool. For all patterns, the spatial period is denoted by T and the peak-to-valley height by h. Throughout, the dependence on the one-dimensional separation ρ is expressed in terms of the shorthands and . The groove patterns we examine, and the corresponding PDPD and its Fourier transform are presented in Table 1. Illustrations of these results are shown in Fig. 1 and Visualization 1. The plots for the PDPD within the second column are shown on their side so that they can be related to the error differences shown in purple in the first column.
4. Parallel groove patterns
The one-dimensional cases in Section 3 provide a way of understanding the two-dimensional case corresponding to milled surfaces, namely of surfaces that exhibit periodic groove patterns in one direction and are constant in the other. If the surface in question is located near a plane conjugate to the aperture stop, the error W at the pupil is approximately proportional to the surface error. Consider, for example, a circular pupil of radius R across which the error has the shape of grooves aligned with the y-axis. Under the assumption that R is much larger than the groove width T we can use the approximation P(η, ρ) ≈ P1(η, ρx), where ρx is the x-component of ρ. This means that the MTF is given by
Figure 2 illustrates this result for the case of 5 parabolic groove profiles across the pupil.
5. Circular groove patterns
It is also insightful to see how this framework can be used for estimating the MTF of a system for which the errors are composed of concentric circular grooves with identical cross sections, such as those generated by diamond turning. That is, we assume that the error is only a function of the radial pupil coordinate q according to W (q), with radial period T. The rotational symmetry of the pupil error can be exploited to derive approximate analytic results that build upon the one-dimensional analysis presented in Section 3.
Consider again a circular pupil of radius R. Once again, ρ denotes the relative displacement between the two shifted copies of the pupil, but it is treated as a scalar given the problem’s rotational symmetry. Let a point q within the overlap region be specified by r1 and r2, which are the distances from the centers of the two shifted pupil copies, as shown in Fig. 3. Note that, due to the problem’s symmetry, it is sufficient to consider points above the line joining these centers. In fact, we can just consider points for which r1 > r2. That is, only a quarter of the overlap region is sufficient. Let us define the centroid and difference coordinates:
The change of variables from q to r and has a Jacobian given by
The region of integration O(ρ) is defined by the bounds of r and , given either byFig. 3 for the two distinct regimes ρ < R and ρ > R. In terms of these new variables, Eq. (3) can be written as
It is easy to see that there are two regions where the Jacobian in Eq. (9) becomes large, i.e. where the mapping from the area in the pupil overlap region bunches up in the space of and r. The first of these regions is near the boundary r = ρ/2 and the second is near the boundary . The first corresponds to the line segment between the two centers, and the second to the line segments between the centers and the edges of the overlap region (only when ρ < R). Note from Fig. 3 that these are regions where there is significant alignment of the grooves from both pupil copies (the relative displacement of the grooves of the two copies is at most one period more than at the axis). Over the rest of the area, the grooves of both copies intersect at a considerable angle.
The approximate results that we set out to obtain are expressed in terms of three quantities. The first is precisely the PDPD for the one-dimensional groove structure in question, like those summarized in Table 1. This one-dimensional PDPD can be written as an integral over a single period of in r:
Although its integrand has period 2T in , it can be shown that has period T (as do the PDPDs in Table 1). The second is defined similarly, but the integration is over the difference:
It can be shown that has period T/2 even though the integrand has period T in r. Notice that if the error is symmetric around some pivot point rc, i.e. W (r) = W (rc − r) (as are all groove structures in Table 1), then . The third quantity is defined as5] to describe the Strehl ratio.
We show in Appendix A that an approximate expression for the PDPD is given by
In these results, A(ρ), B(ρ), and E(ρ) are the areas shown in blue, red, and green in Fig. 3, respectively, such that A(ρ) + B(ρ) + E(ρ) = F(ρ). Their explicit functional forms are given in Appendix A for the case considered here of a circular aperture. Note, however, that were the pupil to have a different shape or not be centered at the center of the grooves (due, for example, to an off-axis beam footprint), the approximate result above would still be valid, as long as the areas A, B, E, and F are calculated with the appropriate boundaries. Also, t = Min(ρ, T) and τ = −Min(2R − ρ, T), and the integral operator applied to a function f(ρ) is defined as
Since this operator has the form of a convolution and is applied to periodic functions with period T, the easiest way to calculate its effect is through a Fourier series expansion of the function it acts on. That is, if the periodic function (assumed here to be even) is written asEq. (17), we have to find the Fourier coefficients am for . These coefficients are summarized in Table 2 for the 1D PDPD Fourier transforms in Table 1. Details of their derivation are given in Appendix B. (It turns out that the coefficients in the third row were found previously in .) Note that, for computational purposes, only a finite number of terms in the Fourier series are used, and the number required to achieve a given level of error depends on kh. The last column on Table 2, labelled as m10%, gives a simple estimate of the terms m ≤ Max(0, m10%) to be used to achieve an error of under 10%. In all cases, a single Fourier coefficient (m = 1) is sufficient to achieve this level of error for peak-to-valley aberration heights up to about half a wavelength. Finally, notice that in Eq. (17) corresponds simply to a0/2, given explicitly in Table 2.
Figure 4 shows the resulting MTF plots of the four groove shapes as calculated by Eq. (17), in comparison to those generated numerically. One can see that, even with the approximations used in the analytic result and despite the fact that there are only ten grooves across the pupil radius, the analytic MTF plots match the numerical ones very well, giving results that are consistent to within about 5%. Figure 5 shows, for the same structures, plots of these MTF estimates, as well as of each of the three individual contributions to the estimate in Eq. (17), for kh = 1.6, and Visualization 3 shows the same data for varying kh. Notice that for kh ± 0 the MTF presents an initial peak for ρ < T, followed by oscillations around a baseline, shown in gray in Fig. 5; this general behavior for the MTF has been described before by employing a different approach . It turns out that the equation for this baseline is very simple, and gives a basic estimate of the MTF for ρ > T. Note from Eq. (20) and the fact that that the first and second terms in Eq. (17) include parts proportional to , and that these, added to the third term of Eq. (17), give simply . Therefore, this baseline is simply the MTF of the unaberrated system scaled by . The number of oscillations around the baseline equals the number of grooves across the diameter, and their amplitude scales not only with kh but also with , since the areas A(ρ) and B(ρ) are roughly proportional to the square root of the period.
6. Concatenation of errors, randomness, and relation to other measures
The use of the PDPD as a probability density provides a simple way of understanding the concatenation of independent errors, and allows the study of not only deterministic error shapes like those described so far, but also of stochastic errors that follow certain statistics. There are often multiple error contributions within an optical system, coming from separate surfaces, from different manufacturing processes for a single surface, or even from fluctuating media such as the atmosphere. For the cases where these contributions are approximately uncorrelated, the total PDPD is constructed following the standard concatenation rule for independent probability densities, as a convolution of the PDPDs Pi from each contribution:
Then, from the Fourier convolution theorem, the MTF of the total system is simply the MTF of the unaberrated system times the product of the Fourier transforms of the individual PDPDs :
Note that, for this relation to hold, the two different error contributions do not need to be uncorrelated due to independent random fluctuations, but simply from the point of view of the PDPD; for example, two deterministic groove structures in different directions or with different periods have PDPDs that are effectively uncorrelated. This is because, for any given ρ, if we were to pick a large series of points q spanning the overlap region, the values of ΔW for the two contributions would be largely uncorrelated.
Let us now discuss the connection between the PDPD and another error measure that has also been used to study effects on optical performance, the structure function [9–11]. The structure function D(ρ) is defined as the integral of ΔW2(q, ρ) over the overlap region . It is easy to see that the structure function is simply the variance of the PDPD, that is,
While the PDPD contains more information than the structure function, it is worth mentioning two cases where D(ρ) gives an approximate full description of both the PDPD and MTF. The first is that where P(η, ρ) is localized within an interval in η smaller than the wavelength. An approximate expression for the MTF can then be found by expanding the exponential in Eq. (5) in a Taylor series up to second order, so that Eq. (6) becomes13]:
Note that both limiting results above have counterparts in the case of the Strehl ratio, as described in , corresponding to the expressions by Maréchal  and Mahajan . The factors multiplying MTFperf in these two simple estimates are shown for the one-dimensional groove structures in the fourth column of Fig. 1 as dashed and dotted green lines, where we see that they only resemble the true results for small kh.
We showed how the MTF of an optical system for all wavelengths can be understood in terms of the PDPD, a function that characterizes the MSF phase errors (whether these are periodic or not). Once the PDPD is known for a given wavefront error in an optical system, it can be Fourier transformed to obtain the MTF. The implementation of this framework is illustrated for several simple 1D periodic groove structures. Further, it is shown that for 2D errors consisting of parallel or concentric periodic grooves over the pupil, the MTF can be easily calculated from the PDPD of the corresponding one-dimensional pattern. These surfaces are important because they are characteristic of the MSF errors left behind on optical surfaces under the diamond milling and turning processes of freeform manufacturing, and when these surfaces are nearly conjugate to the aperture stop, the wavefront error they produce on the pupil mimics their surface error .
Several simple generalizations can be made to this framework. First, the groove patterns could differ from the linear and concentric ones examined explicitly in this paper. Changing the groove pattern would change the effective areas A(ρ) and B(ρ) over which the grooves from the pupil replicas align to within a period. A similar modification can be made to account for fluctuations in the periodicity, which blur the oscillations in the MTF. In fact, the case of non-periodic surface errors can be thought of as the limit when A(ρ) and B(ρ) tend to zero. Therefore, aside from the initial peak of the MTF for |ρ| smaller than the typical feature size, we can see from Eq. (17) that the MTF is in this case dominated by . That is, except for the central peak, the MTF for non-periodic errors is well approximated by the baseline .
Finally, while we focused on the MTF, these results can be extended to another well-known metric of optical performance, the encircled energy, which can be relatede to the OTF as :Eq. (17), one can find an expression for the encircled energy explicitly in terms of the PDPD, as we have done with the MTF. Note that for a point source and a point detector, Eq. (26) reduces to the expression for the Strehl ratio .
A. Derivation of PDPD for circular grooves
Note that for I1 and I2 the limits of integration were specified according to Eqs. (10) and (11), respectively. In particular, let us look at the integral in r within I1, and break it as a sum of many integrals over a period each. Due to the periodicity of we can writeEquation (29) is an approximation because the integration range is generally not an integer multiple of T. If T is small compared to ρ, we can approximate the summation as another integral:
We can use an analogous procedure for I2 in Eq. (28) to find
To simplify these expressions, we discretize again the outermost integrals. Let us this time start with I2 (and then apply a similar procedure to I1). The integral in r in Eq. (32) is broken into a sum of integrals, each over an interval of T/2 (the period of ). By writing rm = ρ/2 + mT/2 + u with M = ⌊R − ρ/2⌋, and using , we find
Notice that for ρ > 0, β is a smooth, nonsingular function of its second argument. However, we cannot approximate the sum in Eq. (33) as an integral, given the strong dependence of its first term (m = 0) on u. It can be shown, however, that if this term is separated the remaining sum can be approximated as an integral:Eq. (34) is surprisingly small, even for small M.
An analogous result can be found for I1 by breaking the integral in into a sum of integrals over a period each. However, it is important to note that the singularity of integrand is contained in the range of integration only when ρ ≤ R. Therefore, only in this case must we separate one term of the sum (the last one) before approximating the remaining sum as an integral. The result is then
To finish the derivation, note thatFig. 3. Similarly, for ρ ≤ R we get Fig. 3. Finally, the sum of the coefficients of P0 in Eqs. (34) and (35) is approximately equal to E(ρ) = F(ρ) − A(ρ) − B(ρ), the remaining area shown in green in Fig. 3.
The approximations made in this appendix depend on the fact that T is small compared to ρ. This allowed the approximations in Eqs. (30) and (34). However, when considering ρ < T, adjustments are necessary in the expression for α(ρ, ρ). In particular, the “average value” over the integrals would not be over an entire period. That is, the first integral in Eq. (35) would go from 0 to ρ, rather than T. As a result,
Similar adjustments are necessary for β(ρ, ρ/2) when ρ > 2R − T. That is, in the region where B(ρ) is the dominant share of the overlap area F(ρ). In this case, the first integral in the last line of Eq. (34) would go from 0 to R − ρ/2, rather than T/2. As a result,
These changes are taken into account in Eq. (16).
To finish this appendix, we give the explicit expressions for the areas A, B, E and F for a circular pupil of radius R. For ρ < R we have
Finally, for all ρ ∈ [0, 2R], we have
B. Fourier coefficients for
Here we calculate the Fourier series coefficients of the functions given in the third column of Table 1, according to17]. We therefore focus on the case within the fourth row, corresponding to piecewise parabolic groove structures. The substitution of in Eq. (47) gives
Then we have
Notice that we can make the change of variables 1 − u → u in the second integral without changing the limits of integration, which gives
However, cos(2πmu) = cos[2πm(1 − u)], so the two integrals are identical. We can then write
Note that the ratio within the integrand can now be written as another integral in the form
We can now change variables of integration according to u = (u1 + u2)/2, u′ = (u1 − u2)/2, so that du du′ = du1 du2/2, giving
Using the cosine angle addition formula, we have
National Science Foundation (NSF) I/UCRC Center for Freeform Optics (IIP-1338877)
We thank the members of CeFO for their support and insights, and acknowledge in particular useful discussions with Thomas Suleski, Reza Aryan, Christoph Menke, Thomas Köhler, Flemming Tinker, James Fienup, Angela Davies, Chris Evans, Jannick Rolland, and Greg Forbes.
References and links
1. R. J. Noll, “Effect of Mid- and High-Spatial Frequencies on Optical Performance,” Opt. Eng. 18(2), 182137 (1979). [CrossRef]
2. D. Aikens, J. E. DeGroote, and R. N. Youngworth, “Specification and Control of Mid-Spatial Frequency Wavefront Errors in Optical Systems,” in Frontiers in Optics 2008/Laser Science XXIV/Plasmonics and Metamaterials/Optical Fabrication and Testing, OSA Technical Digest (CD) (Optical Society of America, 2008), paper OTuA1. [CrossRef]
3. J. M. Tamkin, T. D. Milster, and W. Dallas, “Theory of modulation transfer function artifacts due to mid-spatial-frequency errors and its application to optical tolerancing,” Appl. Opt. 49, 4895 (2010).
4. G. W. Forbes, “Never-ending struggles with mid-spatial frequencies,” Optical Measurement Systems for Industrial Inspection IX, Proc. of SPIE9525, 95251B (2015).
6. R. N. Youngsworth and B. D. Stone, “Simple estimates for the effects of mid-spatial-frequency surface errors on image quality,” Appl. Opt. 39, 2198–2209 (2000). [CrossRef]
7. J. W. Goodman, Introduction to Fourier Optics (Roberts & Company Publishers, 2005), Chap. 6.
8. G. D. Boremann, Modulation Transfer Function in Optical and Electro-Optical Systems (SPIE, 2001), pp. 85–88. [CrossRef]
10. L. He, C. J. Evans, and A. Davies, “Optical surface characterization with the area structure function,” CIRP Annals -Manufacturing Technology , 62(1), 539–542 (2013). [CrossRef]
11. R. E. Parks and M. T. Tuell, “The structure function as a metric for roughness and figure,” Proc. of SPIE 995199510 (2016).
12. J. W. Goodman, Statistical Optics (John Wiley & Sons, 2015), Chap. 8.
13. R. E. Hufnagel and N. R. Stanley, “Modulation Transfer Function associated with Image Transmission through Turbulent Media,” J. Opt. Soc. Am. 54, 52–61 (1964). [CrossRef]
14. A. Maréchal, “Etude des effets combinés de la diffraction et des aberrations géométriques sur l’image d’un point lumineux,” Rev. Opt. 2, 257–277 (1947).
15. V. N. Mahajan, “Strehl ratio for primary aberrations in terms of their aberration variance,” J. Opt. Soc. Am. 73, 860–861 (1983). [CrossRef]
16. J. V. Baliga and B. D. Cohn, “Simplified method for calculating encircled energy,” Proc. SPIE 892, 152 (1988). [CrossRef]
17. I. S. Gradshteyn and I. M. Ryxhik, Table of Integrals, Series, and Products (Academic, 1980), p. 739, ET II 360(14).