The system that requires largest!

Started by santosh nath December 20, 2003
Though it is not very good (rather stupid) to ask the following
questions; I thought this could be a pre-Xmas fun!

1. What is the largest FFT size(point) ever used so far and
where(Application)?
2. What is the largest QAM constellation ever used commercially so far
and where(Application)? ?
3. What is the largest FIR filter taps used so far and
where(Application)?.

 Real time constraint should be applied if needed.


Enjoy your days with happy X-mas!!
Cheers,
santosh
santosh.nath@ntlworld.com (santosh nath) wrote in message
news:<6afd943a.0312201539.555ad3a6@posting.google.com>...
> Though it is not very good (rather stupid) to ask the following > questions; I thought this could be a pre-Xmas fun! > > 1. What is the largest FFT size(point) ever used so far and > where(Application)? > 2. What is the largest QAM constellation ever used commercially so far > and where(Application)? ? > 3. What is the largest FIR filter taps used so far and > where(Application)?. >
Hi, I can only state the number of FFT points for the OFDM (Orthogonal Frequency Division Multiplex) modulation scheme used in digital broadcasting. Terrestrial Digital Video Broadcasting (DVB) uses up to 4096 point FFTs. You can find a high order QAMs also in DVB but this time for cable transmission. From what I know they encode up to eight bits for one symbol resulting in a 256QAM. You require a large number of filter coefficients if you want to alter the sample frequency of a signal. For a direct implementation these interpolation filters can have 100 or even 200 over taps. But sophisticated implementation techniques, (poly phase decomposition for interpolation as example), are used to reduce the number of taps for these filters. Oliver Faust
Hello Santosh,

I think Lake DSP uses extremely long FIR filters. Something like thousands
if not tens of thousands of taps.

Clay


"santosh nath" <santosh.nath@ntlworld.com> wrote in message
news:6afd943a.0312201539.555ad3a6@posting.google.com...
> Though it is not very good (rather stupid) to ask the following > questions; I thought this could be a pre-Xmas fun! > > 1. What is the largest FFT size(point) ever used so far and > where(Application)? > 2. What is the largest QAM constellation ever used commercially so far > and where(Application)? ? > 3. What is the largest FIR filter taps used so far and > where(Application)?. > > Real time constraint should be applied if needed. > > > Enjoy your days with happy X-mas!! > Cheers, > santosh
santosh nath wrote:
> Though it is not very good (rather stupid) to ask the following > questions; I thought this could be a pre-Xmas fun! > > 1. What is the largest FFT size(point) ever used so far and > where(Application)? >
I'm using an ~22050 point fft to look at .5 sec chunks of speech.
> > Real time constraint should be applied if needed. >
Definitely not real time. Data source is CD. Calculations done in Scilab on a PC with a Gig of memory. Therefore "brute force" methods are acceptable. Question: Someone recently pointed out how long time chunks could be built up from overlapping ifft's. Is there a similar method going from time domain -> freq domain? [ Seems it there should by as lots of other things are symmetrical. ] [ My current frequency range of interest is continuous from 20 Hz to 5 kHz. ] Also, later I suspect I will want to compare spectra of data with different window widths while having the same resolution in the frequency domain. Each spectrum will be "normalized against itself", either by saying the largest peak or the value at a certain frequency is 0 dB. How?

santosh nath wrote:
> > Though it is not very good (rather stupid) to ask the following > questions; I thought this could be a pre-Xmas fun! > > 1. What is the largest FFT size(point) ever used so far and > where(Application)?
I have seen, but failed to bookmark, an application that will do a single FFT of an arbitrary length audio file. Wish I could be more helpful as to where to locate it. Bob -- "Things should be described as simply as possible, but no simpler." A. Einstein
santosh nath wrote:

> Though it is not very good (rather stupid) to ask the following > questions; I thought this could be a pre-Xmas fun! > > 1. What is the largest FFT size(point) ever used so far and > where(Application)? > 2. What is the largest QAM constellation ever used commercially so far > and where(Application)? ? > 3. What is the largest FIR filter taps used so far and > where(Application)?. > > Real time constraint should be applied if needed. > > > Enjoy your days with happy X-mas!! > Cheers, > santosh
What's with the preoccupation with biggest and mostest? Financial analysts perform FFTs and transversal FIRs using tens of thousands of bins and taps. I haven't kept track of the actual numbers. Jerry -- Engineering is the art of making what you want from things you can get. &#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;
On 20 Dec 2003 15:39:58 -0800, santosh.nath@ntlworld.com (santosh
nath) wrote:

>Though it is not very good (rather stupid) to ask the following >questions; I thought this could be a pre-Xmas fun! > >1. What is the largest FFT size(point) ever used so far and >where(Application)? >2. What is the largest QAM constellation ever used commercially so far >and where(Application)? ? >3. What is the largest FIR filter taps used so far and >where(Application)?. > > Real time constraint should be applied if needed. > > >Enjoy your days with happy X-mas!! >Cheers, >santosh
Hi, years ago I was performing 4 million-point FFTs on a Sun Workstation using MATLAB. I was generating test signals for some others to use in testing their DSP algorithms. I'll bet someone here will probably know of a larger-sized FFT application. Don't the SETI people (looking for little green men) use million-point FFTs? Also, doesn't one of those commercial up/down converter chips (Analog devices, of Harris/Intersil, GreyChip) have a 256-tap FIR filter built in? Merry Xmas, [-Rick-]
On Sun, 21 Dec 2003 02:20:58 -0600, Richard Owlett
<rowlett@atlascomm.net> wrote:

>santosh nath wrote: >> Though it is not very good (rather stupid) to ask the following >> questions; I thought this could be a pre-Xmas fun! >> >> 1. What is the largest FFT size(point) ever used so far and >> where(Application)? >> > >I'm using an ~22050 point fft to look at .5 sec chunks of speech. > >> >> Real time constraint should be applied if needed. >> > >Definitely not real time. >Data source is CD. >Calculations done in Scilab on a PC with a Gig of memory. >Therefore "brute force" methods are acceptable. > >Question: > >Someone recently pointed out how long time chunks could be built up >from overlapping ifft's. Is there a similar method going from time >domain -> freq domain? [ Seems it there should by as lots of other >things are symmetrical. ] [ My current frequency range of interest is >continuous from 20 Hz to 5 kHz. ] > >Also, later I suspect I will want to compare spectra of data with >different window widths while having the same resolution in the >frequency domain. Each spectrum will be "normalized against itself", >either by saying the largest peak or the value at a certain frequency >is 0 dB. How?
Hi Richard, Humm, I didn't understand your last question. Different window widths (i.e., the length of your time sequences) will result in different freq-domain resolutions. That's assuming the sample rate is the same for the various time-domain sequences. You can plot spectra where the largest spec magnitude sample is always at 0 dB by: 1) compute your FFT samples, X(m); 2) compute the magnitudes of your FFT results, |X(m)|; 3) find the largest magnitude sample in the |X(m)| sequence, call that value "|X(m)|max"; 4) divide the |X(m| sequence by the |X(m)|max value to yield a "normalized" sequence, call that sequence "|X(m)|norm"; 5) plot 20*log10 of sequence "|X(m)|norm". The peak |X(m)| value will always be at the 0 dB level, and all other spectral mag samples will be at negative dB levels. This makes it easier to compare two spectra shapes (two spectral envelopes). See Ya' [-Rick-]
Jerry Avins wrote:

> santosh nath wrote: > >> Though it is not very good (rather stupid) to ask the following >> questions; I thought this could be a pre-Xmas fun! >> >> 1. What is the largest FFT size(point) ever used so far and >> where(Application)? >> 2. What is the largest QAM constellation ever used commercially so far >> and where(Application)? ? >> 3. What is the largest FIR filter taps used so far and >> where(Application)?. >> >> Real time constraint should be applied if needed. >> >> >> Enjoy your days with happy X-mas!! >> Cheers, >> santosh > > > What's with the preoccupation with biggest and mostest? Financial > analysts perform FFTs and transversal FIRs using tens of thousands of > bins and taps. I haven't kept track of the actual numbers.
Correction: hundreds of thousands. Jerry -- Engineering is the art of making what you want from things you can get. &#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;
Rick Lyons wrote:

> On Sun, 21 Dec 2003 02:20:58 -0600, Richard Owlett > <rowlett@atlascomm.net> wrote: > > >>santosh nath wrote: >> >>>Though it is not very good (rather stupid) to ask the following >>>questions; I thought this could be a pre-Xmas fun! >>> >>>1. What is the largest FFT size(point) ever used so far and >>>where(Application)? >>> >> >>I'm using an ~22050 point fft to look at .5 sec chunks of speech. >> >> >>> Real time constraint should be applied if needed. >>> >> >>Definitely not real time. >>Data source is CD. >>Calculations done in Scilab on a PC with a Gig of memory. >>Therefore "brute force" methods are acceptable. >> >>Question: >> >>Someone recently pointed out how long time chunks could be built up > >>from overlapping ifft's. Is there a similar method going from time > >>domain -> freq domain? [ Seems it there should by as lots of other >>things are symmetrical. ] [ My current frequency range of interest is >>continuous from 20 Hz to 5 kHz. ] >> >>Also, later I suspect I will want to compare spectra of data with >>different window widths while having the same resolution in the >>frequency domain. Each spectrum will be "normalized against
itself",
>>either by saying the largest peak or the value at a certain frequency >>is 0 dB. How? > > > Hi Richard, > Humm, I didn't understand your last question. > Different window widths (i.e., the length of > your time sequences) will result in different > freq-domain resolutions. That's assuming the sample > rate is the same for the various time-domain sequences. >
Obvious now that you mention it ;{ How do you calculate the nominal center of each point in the spectrum? [ I did a dimensional analysis of the formula I used and came up with trash ;[
> You can plot spectra where the largest spec magnitude > sample is always at 0 dB by: > [snip]
At least I had that much right.
santosh nath <santosh.nath@ntlworld.com> wrote:
> Though it is not very good (rather stupid) to ask the following > questions; I thought this could be a pre-Xmas fun! > > 1. What is the largest FFT size(point) ever used so far and > where(Application)?
NASA has tech reports on doing out of core FFTs on Crays from the 1980's, so assuming they have kept up, they would presently be doing several billion point ffts at the very least.
> > > Enjoy your days with happy X-mas!! > Cheers, > santosh
-- Sander +++ Out of cheese error +++
Ray Andraka wrote:
 > Oops, that should have been 4Giga point.
 >
 > Ray Andraka wrote:
 >>Some time ago someone approached me about doing a 4M point FFT for an
 >>astronomy application.  It wasn't real time, and I suggested it would
 >>be far cheaper doing it with PCs.  Don't know what happened with the 
 >>project since.

We (FFTW authors) have received a number of notes from people doing 
multi-billion point FFTs, and in fact this was the motivation for our 
inclusion of parallel 1d distributed-memory FFTs in FFTW 2.x.  Often, 
it's not even time that makes people want to parallelize, it's memory 
usage.  (The same constraints originally motivated out-of-core FFTs, 
although I don't hear about too many people using those any more.)

Typical applications for this sort of thing seem to be in astronomy, as 
happened in your case.  e.g. analyzing pulsar data, gravitational waves, 
etcetera.

Cordially,
Steven G. Johnson

Oops, that should have been 4Giga point.

Ray Andraka wrote:

> Some time ago someone approached me about doing a 4M point FFT for an > astronomy application. It wasn't real time, and I suggested it would be > far cheaper doing it with PCs. Don't know what happened with the project > since. Large FIR filters are usually required because of a very narrow > or sharp response. These are more efficiently realized using a polyphase > or multi-rate approach, so in practice you'll usually find large filters > broken down into a bank of smaller ones. I've done a number of filters, > that if done in one stage would have numbered in the 100's of thousands of > taps. Even if multi-rate approaches don't help, a filter that large is > more efficiently done with fast convolution: ie in the frequency domain. > > santosh nath wrote: > > > Though it is not very good (rather stupid) to ask the following > > questions; I thought this could be a pre-Xmas fun! > > > > 1. What is the largest FFT size(point) ever used so far and > > where(Application)? > > 2. What is the largest QAM constellation ever used commercially so far > > and where(Application)? ? > > 3. What is the largest FIR filter taps used so far and > > where(Application)?. > > > > Real time constraint should be applied if needed. > > > > Enjoy your days with happy X-mas!! > > Cheers, > > santosh > > -- > --Ray Andraka, P.E. > President, the Andraka Consulting Group, Inc. > 401/884-7930 Fax 401/884-7950 > email ray@andraka.com > http://www.andraka.com > > "They that give up essential liberty to obtain a little > temporary safety deserve neither liberty nor safety." > -Benjamin Franklin, 1759
-- --Ray Andraka, P.E. President, the Andraka Consulting Group, Inc. 401/884-7930 Fax 401/884-7950 email ray@andraka.com http://www.andraka.com "They that give up essential liberty to obtain a little temporary safety deserve neither liberty nor safety." -Benjamin Franklin, 1759
Some time ago someone approached me about doing a 4M point FFT for an
astronomy application.  It wasn't real time, and I suggested it would be
far cheaper doing it with PCs.  Don't know what happened with the project
since.   Large FIR filters are usually required because of a very narrow
or sharp response.  These are more efficiently realized using a polyphase
or multi-rate approach, so in practice you'll usually find large filters
broken down into a bank of smaller ones.  I've done a number of filters,
that if done in one stage would have numbered in the 100's of thousands of
taps.  Even if multi-rate approaches don't help, a filter that large is
more efficiently done with fast convolution: ie in the frequency domain.

santosh nath wrote:

> Though it is not very good (rather stupid) to ask the following > questions; I thought this could be a pre-Xmas fun! > > 1. What is the largest FFT size(point) ever used so far and > where(Application)? > 2. What is the largest QAM constellation ever used commercially so far > and where(Application)? ? > 3. What is the largest FIR filter taps used so far and > where(Application)?. > > Real time constraint should be applied if needed. > > Enjoy your days with happy X-mas!! > Cheers, > santosh
-- --Ray Andraka, P.E. President, the Andraka Consulting Group, Inc. 401/884-7930 Fax 401/884-7950 email ray@andraka.com http://www.andraka.com "They that give up essential liberty to obtain a little temporary safety deserve neither liberty nor safety." -Benjamin Franklin, 1759
On Mon, 22 Dec 2003 20:01:23 -0800, Bob Cain
<arcane@arcanemethods.com> wrote:

> > >Richard Owlett wrote: >> >> Obviously I asked WRONG question ;) >> >> Given (sample rate) AND (number of points in time domain) >> >> What is "frequency" of each point in FFT? >> >> Next question: >> "Did I ask "right" question ;? > >If this is the answer: the difference in frequency between >any two consecutive points of the transform is rate/#points. > > >Bob
Yep, If the sample rate is, say, 1 kHz and you had eight samples, then the spacing between FFT samples would be 1000/8 = 125 Hz. The cyclic freq associated with the first half of the FFT samples would be FFT sample#: Freq: 0 0x125 = 0 Hz 1 1x125 = 125 Hz 2 2x125 = 250 Hz 3 3x125 = 375 Hz etc. etc. [-Rick-]
On Mon, 22 Dec 2003 15:24:20 GMT, charles@pentek.com (Charles Krug)
wrote:

  (snipped)

> > >Jerry, you have links for that? > >I ran a 4MB fft in LabView once. Started it, went to lunch, went to a >long design review . . . > >Got back just in time (three hours on a 200MHz Pentium)
Hi, Humm, ... three hours sure seems like a long time. My 4,194,304-point (2^22) FFTs sure didn't take that long using MATLAB on a Sun workstation. (I don't now the CPU clock rate. That was in 1999.) However, in the beginning I'd start my FFTs and my machine would just go dead. I finally realized I'd screwed up my indexing and was trying to perform (2^22-1)-point FFTs. Well, what I was doing was performing (2^22-1)-point DFTs !!!!!!!! That's why my machine appeared dead, it was crunching away as fast as it could at an almost impossibly-intensive computation. [-Rick-]

Richard Owlett wrote:
> > Obviously I asked WRONG question ;) > > Given (sample rate) AND (number of points in time domain) > > What is "frequency" of each point in FFT? > > Next question: > "Did I ask "right" question ;?
If this is the answer: the difference in frequency between any two consecutive points of the transform is rate/#points. Bob -- "Things should be described as simply as possible, but no simpler." A. Einstein
On Mon, 22 Dec 2003 09:38:49 -0500, Jerry Avins <jya@ieee.org> wrote:

>Allan Herriman wrote: > > ... > >> I once made a 240 pole *analog* low pass filter. >> >> Regards, >> Allan. > >That's amazing! (I assume it worked.)
Of course! http://groups.google.com/groups?threadm=38e151a7.3033011%40newshost.fujitsu.com.au Allan.
Obviously I asked WRONG question ;)

Given (sample rate) AND (number of points in time domain)

What is "frequency" of each point in FFT?

Next question:
"Did I ask "right" question ;?



"Richard Owlett" <rowlett@atlascomm.net> wrote in message
news:vueinms39rld44@corp.supernews.com...
> > I lurk on comp.lang.research . > A frequent topic of discussion is detecting "silence" [ i.e. pauses > between words/phrases/sentences ] > > Being one who talks a lot ;] > I have a "gut" feel that *each* of those might be of fairly
consistent
> duration for a particular speaker. > > Ignoring for the moment the current push for "real
time"/"instant"
> recognition. > > Given a speech sample of tens of seconds, should/could/would an FFT of > the *ENTIRE* utterance give indications of where and how long pauses were?
I don't think so. The FFT of the entire sequence will only tell you (not sure exactly how) if a certain frequency component existed (or didn't). It will not tell you exactly when during the temporal record this occured. So in order to figure *when* something happened (or didn't), you'd have to do FFTs on shorter segments of the time data. Clearly you will be limited in terms of resolution - you can only get one at a time...good time or frequency resolution. Hopefully some of that made sense. Cheers Bhaskar