| Re: time vs longitude FFT [message #56542] |
Mon, 29 October 2007 10:41  |
Kenneth Bowman
Messages: 86
Registered: November 2006
|
Member |
|
|
In article <1193671433.271745.268030@57g2000hsv.googlegroups.com>,
knielsen73@gmail.com wrote:
> Hi,
>
> I have a time vs longitude plot and would like to do a 2-D FFT to get
> a frequency vs wavenumber plot. I used the IDL FFT but I am not sure
> what the output is...I didn't feel the IDL help was very helpful
> there. I found a post from 1998 but there were some non-defined
> functions and I wasn't exactly sure what they did.
>
> Does anybody have a good example on how to use 2D FFT in IDL or some
> good documentation?
>
> Cheers,
> Kim
The main issues with the IDL FFT (or any FFT for that matter) are
sorting out the order in which the coefficients are stored and
understanding how the sine and cosine components are represented in
a complex FFT.
My recommendation is that you create an artificial, low-resolution,
input data set that contains some simple combination of sines and
cosines and that you understand completely. Then look at the FFT.
Or you can buy my book (An Introduction to Programming with IDL),
which has a chapter that does exactly that for the 1-D case.
Ken Bowman
|
|
|
|
| Re: time vs longitude FFT [message #56543 is a reply to message #56542] |
Mon, 29 October 2007 17:29  |
knielsen73
Messages: 9
Registered: October 2007
|
Junior Member |
|
|
On Oct 29, 5:54 pm, "R.G. Stockwell" <noem...@please.com> wrote:
> "R.G. Stockwell" <noem...@please.com> wrote in message
>
> news:47265628$0$497$815e3792@news.qwest.net...
>
>
>
>> Also, do just 1D to ensure that you are calculating the correct
>> period and freq.
>
>> f = 12/len
>> data = amp*cos(2 !pi f t)
>
> oops sorry about that, i've got a 1 yr old climbing on the
> laptop.
>
> Anyways, do the example in 1D first with a simple cos,
> to ensure that you are calculating the correct freqs.
>
> Cheers,
> bob
thanks Bob...got it!
|
|
|
|
| Re: time vs longitude FFT [message #56546 is a reply to message #56542] |
Mon, 29 October 2007 14:54 |
R.G.Stockwell
Messages: 163
Registered: October 2004
|
Senior Member |
|
|
"R.G. Stockwell" <noemail@please.com> wrote in message
news:47265628$0$497$815e3792@news.qwest.net...
>
> Also, do just 1D to ensure that you are calculating the correct
> period and freq.
>
> f = 12/len
> data = amp*cos(2 !pi f t)
oops sorry about that, i've got a 1 yr old climbing on the
laptop.
Anyways, do the example in 1D first with a simple cos,
to ensure that you are calculating the correct freqs.
Cheers,
bob
|
|
|
|
| Re: time vs longitude FFT [message #56547 is a reply to message #56542] |
Mon, 29 October 2007 14:52 |
R.G.Stockwell
Messages: 163
Registered: October 2004
|
Senior Member |
|
|
<knielsen73@gmail.com> wrote in message
news:1193692469.369318.51960@50g2000hsm.googlegroups.com...
>> kxvalues = findgen(nx)/nx - 1./2
>> kyvalues = findgen(ny)/ny - 1./2
Note, in the above, it shows that the sampling interval is 1.
You should divide by your sampling interval there if it is not 1.
> amp = 10.
> wave_number = 2
> wavelength = float(jmax)/wave_number
> period = 6.8333333
>
> for j=0,jmax-1 do begin
> for n=0,nmax-1 do begin
> data_in[j,n] = amp*sin(2.*!pi*j/wavelength - 2.*!pi*n/period) +$
> amp*.5*cos(2.*!pi*j/wavelength - 2.*!pi*n/period)
for one thing, you are adding 2 out of phase waves here. Remove the
first term (and just use the cos).
Also, do just 1D to ensure that you are calculating the correct
period and freq.
f = 12/len
data = amp*cos(2 !pi f t)
|
|
|
|
| Re: time vs longitude FFT [message #56548 is a reply to message #56542] |
Mon, 29 October 2007 14:14 |
knielsen73
Messages: 9
Registered: October 2007
|
Junior Member |
|
|
On Oct 29, 2:41 pm, "R.G. Stockwell" <noem...@please.com> wrote:
> <knielse...@gmail.com> wrote in message
>
> news:1193671433.271745.268030@57g2000hsv.googlegroups.com...
>
>> Hi,
>
>> I have a time vs longitude plot and would like to do a 2-D FFT to get
>> a frequency vs wavenumber plot. I used the IDL FFT but I am not sure
>> what the output is...I didn't feel the IDL help was very helpful
>> there. I found a post from 1998 but there were some non-defined
>> functions and I wasn't exactly sure what they did.
>
>> Does anybody have a good example on how to use 2D FFT in IDL or some
>> good documentation?
>
> a little example:nx = 100
> ny = 200
> x = (fltarr(ny)+1) ## findgen(nx)
> y = findgen(ny) ## (fltarr(nx)+1)
> kx = 20.5/nx
> ky = 0./ny
>
> ;image
> a =cos(2*!pi*(kx*x + ky*y))
>
> ;spectrum in 2d
> spe = fft(a)
>
> kxvalues = findgen(nx)/nx - 1./2
> kyvalues = findgen(ny)/ny - 1./2
>
> !p.multi=[0,1,2]
> contour,a,/fill,nlevels=14
>
> ; shift to get the normal representation,
> ; i.e. (neg kx, 0, pos kx)
> shiftspe = shift(spe,nx/2,ny/2)
>
> contour,abs(shiftspe),kxvalues,kyvalues,/fill,nlevels=14
>
> Note: it is symmetric about the origin.
>
> Cheers,
> bob
Thanks Bob,
I works great with your example. However, when I made a wave with a
wavenumber larger than 1, the scale no longer agrees. I can plot it
"raw", but again, not sure how to read the values with respect to the
real wave number and frequency. Here is the wave I gave as input.
;input a wave number two with an amplitude of 10.
;and eastward phase propagation of 1.7 days
amp = 10.
wave_number = 2
wavelength = float(jmax)/wave_number
period = 6.8333333
for j=0,jmax-1 do begin
for n=0,nmax-1 do begin
data_in[j,n] = amp*sin(2.*!pi*j/wavelength - 2.*!pi*n/period) +$
amp*.5*cos(2.*!pi*j/wavelength - 2.*!pi*n/period)
endfor
endfor
Cheers,
Kim
|
|
|
|
| Re: time vs longitude FFT [message #56552 is a reply to message #56542] |
Mon, 29 October 2007 11:41 |
R.G.Stockwell
Messages: 163
Registered: October 2004
|
Senior Member |
|
|
<knielsen73@gmail.com> wrote in message
news:1193671433.271745.268030@57g2000hsv.googlegroups.com...
> Hi,
>
> I have a time vs longitude plot and would like to do a 2-D FFT to get
> a frequency vs wavenumber plot. I used the IDL FFT but I am not sure
> what the output is...I didn't feel the IDL help was very helpful
> there. I found a post from 1998 but there were some non-defined
> functions and I wasn't exactly sure what they did.
>
> Does anybody have a good example on how to use 2D FFT in IDL or some
> good documentation?
a little example:nx = 100
ny = 200
x = (fltarr(ny)+1) ## findgen(nx)
y = findgen(ny) ## (fltarr(nx)+1)
kx = 20.5/nx
ky = 0./ny
;image
a =cos(2*!pi*(kx*x + ky*y))
;spectrum in 2d
spe = fft(a)
kxvalues = findgen(nx)/nx - 1./2
kyvalues = findgen(ny)/ny - 1./2
!p.multi=[0,1,2]
contour,a,/fill,nlevels=14
; shift to get the normal representation,
; i.e. (neg kx, 0, pos kx)
shiftspe = shift(spe,nx/2,ny/2)
contour,abs(shiftspe),kxvalues,kyvalues,/fill,nlevels=14
Note: it is symmetric about the origin.
Cheers,
bob
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
