Huygens MTF vs FFT MTF
j4murali
I have been using an optics software to evaluate certain optical setup
and I would like to estimate the MTF of the system. I see that there
are two ways to generate diffractive MTFs: Huygens based and FFT
based.
I understood that these are two different ways to obtain MTF, but
which one is more accurate?
Just to give some more info, when I run FFT based MTF I get very low
values (less than 10%) for pretty much all frequencies (LP/mm), but
when I run Huygens MTF I get decent numbers. I am not sure which one
is correct.
Any thoughts on which is more accurate?
many thanks in advance
-mJ
Ron Gibbs
"j4murali" <j4mu...@gmail.com> wrote in message
news:88a87e8a-8918-47da...@s9g2000yqd.googlegroups.com...
Of course if you had a legal copy you could read about them in the
manual...;-}
Ron
--- news://freenews.netfront.net/ - complaints: ne...@netfront.net ---
Helpful person
You should not be seeing much difference between the two
calculations. However, in general use the FFT calculation because
although it is less accurate it is quicker. For more accurate results
use Huygens.
Do not use either if your optical system is not fairly close to
diffraction limit. In this case use a geometrical calculation. I
suspect that the differences you are seeing is because you are far
from diffraction limited.
Brian
> Hi all
> Any thoughts on which is more accurate?
>
> many thanks in advance
> -mJ
If your system consists of less than 10 surfaces (including the image)
download OSLO EDU (free - and legal) and see what that gives.
Geometric MTF is an approximation, diffraction MTF is exact.
All the best
Brian
Ancient and Modern Optics
Pierre Lanuéjols
"Ron Gibbs" <ron....@physics.org> wrote in message
news:i0d68d$118m$1...@adenine.netfront.net...
Amusing, I just searched "Huygen MTF" w/o Zemax and I got 20 results
in chinese out of 22... :-)
Draw your own conclusions.
I do not own a copy of Zemax, hence simply out of curiosity the following
question.
Why does Zemax apparently call "Huygens MTF" what seems to be simply
the geometric MTF?
I do not get the realtionship.
Ron Gibbs
>> Of course if you had a legal copy you could read about them in the
>> manual...;-}
>>
>> Ron
>
> Amusing, I just searched "Huygen MTF" w/o Zemax and I got 20 results
> in chinese out of 22... :-)
>
> Draw your own conclusions.
>
> I do not own a copy of Zemax, hence simply out of curiosity the following
> question.
>
> Why does Zemax apparently call "Huygens MTF" what seems to be simply
> the geometric MTF?
>
It's the FFT of the Huygens PSF, and NOT geometric MTF, which Zemax also
does. The Zemax manual describes all the MTF calculations in great detail,
together with the circumstances in which each is appropriate ;-)
Ron
--
Gibbs Associates
Optical Design Consultant
www.gibbsassociates.co.uk
AES
"Ron Gibbs" <ron....@physics.org> wrote:
> >
> > Why does Zemax apparently call "Huygens MTF" what seems to be simply
> > the geometric MTF?
> >
>
> It's the FFT of the Huygens PSF, and NOT geometric MTF, which Zemax also
> does. The Zemax manual describes all the MTF calculations in great detail,
> together with the circumstances in which each is appropriate ;-)
>
Right.
Or at least, Huygens integral in the paraxial approximation
(which is a pretty good approximation unless you have really
wide angle aberrations in your beam) _is_ mathematically a Fourier
transform (or can be quickly converted into such).
So, you can, with proper care, evaluate it using an FFT algorithm.
Proper care means paying appropriate attention to windowing, aliasing,
"guard bands", and such details associated with converting a continuous
FT into a discretized DFT.
Assuming you pay proper care, an FFT evaluation will require much less
time, memory, and CPU cycles, and suffer much less round-off error, than
any other numerical method for evaluating a Fourier transform or a
Huygens integral..
On the other hand, modern computers and operating systems have so much
speed, working RAM, and numerical precision that just evaluating a
Huygens integral using any standard numerical library routine is likely
to be, in practice, indistinguishable from doing it with an FFT routine.
Pierre Lanuéjols
"AES" <sie...@stanford.edu> wrote in message
news:siegman-0AB813...@bmedcfsc-srv02.tufts.ad.tufts.edu...
Pierre Lanuéjols
"AES" <sie...@stanford.edu> wrote in message
news:siegman-0AB813...@bmedcfsc-srv02.tufts.ad.tufts.edu...
> "Ron Gibbs" <ron....@physics.org> wrote:
>
>> >
>> > Why does Zemax apparently call "Huygens MTF" what seems to be simply
>> > the geometric MTF?
>> >
>>
>> It's the FFT of the Huygens PSF, and NOT geometric MTF, which Zemax also
>> does. The Zemax manual describes all the MTF calculations in great
>> detail,
>> together with the circumstances in which each is appropriate ;-)
>>
>
> Right.
>
> Or at least, Huygens integral in the paraxial approximation
> (which is a pretty good approximation unless you have really
> wide angle aberrations in your beam) _is_ mathematically a Fourier
> transform (or can be quickly converted into such).
As explained by J. Goodman... :-)
> So, you can, with proper care, evaluate it using an FFT algorithm.
> Proper care means paying appropriate attention to windowing, aliasing,
> "guard bands", and such details associated with converting a continuous
> FT into a discretized DFT.
>
> Assuming you pay proper care, an FFT evaluation will require much less
> time, memory, and CPU cycles, and suffer much less round-off error, than
> any other numerical method for evaluating a Fourier transform or a
> Huygens integral..
>
> On the other hand, modern computers and operating systems have so much
> speed, working RAM, and numerical precision that just evaluating a
> Huygens integral using any standard numerical library routine is likely
> to be, in practice, indistinguishable from doing it with an FFT routine.
Thanks to both of you.
OK, I get it, the Huygens MTF turns out to be just my ordinary "diffraction
MTF"
with the usual assumptions, I never thought to call it that way (or forgot
that it
ever was)!
In the days before the FFT (!) I used to compute MTFs according to a
"recipe"
explained by the late Richard Barakat using Gaussian integration (papers in
the
JOSA early or mid sixties), this was *very* effective, the computations were
performed on a mid range IBM 360 of the time and were a very modest task
indeed even for such an early machine .
Barakat was using the pupil auto-correlation method, the PSF was not used.
The Gaussian integration method is one of the wonders of numerical analysis.
Pierre Lanuéjols
news:siegman-0AB813...@bmedcfsc-srv02.tufts.ad.tufts.edu...
> "Ron Gibbs" <ron....@physics.org> wrote:
>
>> >
>> > Why does Zemax apparently call "Huygens MTF" what seems to be simply
>> > the geometric MTF?
>> >
>>
>> It's the FFT of the Huygens PSF, and NOT geometric MTF, which Zemax also
>> does. The Zemax manual describes all the MTF calculations in great
>> detail,
>> together with the circumstances in which each is appropriate ;-)
>>
>
> Right.
>
> Or at least, Huygens integral in the paraxial approximation
> (which is a pretty good approximation unless you have really
> wide angle aberrations in your beam) _is_ mathematically a Fourier
> transform (or can be quickly converted into such).
As explained by J. Goodman... :-)
> So, you can, with proper care, evaluate it using an FFT algorithm.
> Proper care means paying appropriate attention to windowing, aliasing,
> "guard bands", and such details associated with converting a continuous
> FT into a discretized DFT.
>
> Assuming you pay proper care, an FFT evaluation will require much less
> time, memory, and CPU cycles, and suffer much less round-off error, than
> any other numerical method for evaluating a Fourier transform or a
> Huygens integral..
>
> On the other hand, modern computers and operating systems have so much
> speed, working RAM, and numerical precision that just evaluating a
> Huygens integral using any standard numerical library routine is likely
> to be, in practice, indistinguishable from doing it with an FFT routine.
Thanks to both of you.
OK, I get it, the Huygens MTF turns out to be just my ordinary "diffraction
MTF" with the usual assumptions, I never thought to call it that way (or
forgot that it ever was)!
In the days before the FFT (!) I used to compute MTFs according to a
"recipe" explained by the late Richard Barakat using Gaussian integration
(papers in the JOSA early or mid sixties), this was *very* effective, the
computations were performed on a mid range IBM 360 of the time and
were a very modest task indeed even for such an early machine .
Barakat was using the pupil auto-correlation method, the PSF was not used.
The Gaussian integration method is one of the wonders of numerical analysis.
Sorry for the multiple posts... :-(