Got a homework question, but not in school. :) FFT windowing functions in LTSpice. None Welch Bartlett Hamming Hann Blackman Lanczos Parzen Kaiser Bessel I no idea when each are applied. Looked at http://en.wikipedia.org/wiki/Window_function Geezz :( Not exactly easy late night reading. Saw the audio slide show tutorial on http://www.ni.com/swf/presentation/us/windowing/ Type of Signal | Window ------------------+------------ Sine waves | Hanning Spectral analysis Hanning (for random excitation) Transients longer than length of window Hanning Unknown content Hanning Two tone with close f Kaiser Bessel Transients duration longer than the length of the window Rectangular The 'none' (no window) choice in LTSpice means rectangular. Who uses Welch? D from BC British Columbia Canada.

# FFT Windowing Functions in LT Spice

Started by ●January 7, 2008

Posted by ●January 7, 2008

"D from BC" <myrealaddress@comic.com> wrote in message news:1go3o35nobptkttckeb5tqf2b8bl1ssl9r@4ax.com...> Got a homework question, but not in school. :) > > FFT windowing functions in LTSpice. > None > Welch > Bartlett > Hamming > Hann > Blackman > Lanczos > Parzen > Kaiser Bessel > > I no idea when each are applied. > > Looked at > http://en.wikipedia.org/wiki/Window_function > Geezz :( Not exactly easy late night reading. > > Saw the audio slide show tutorial on > http://www.ni.com/swf/presentation/us/windowing/ > > Type of Signal | Window > ------------------+------------ > Sine waves | Hanning > Spectral analysis Hanning (for random excitation) > > Transients longer > than length of window Hanning > Unknown content Hanning > > Two tone > with close f Kaiser Bessel > > Transients duration > longer than the length > of the window Rectangular > > The 'none' (no window) choice in LTSpice means rectangular. > > Who uses Welch? > > > D from BC > British Columbia > Canada.I don't use 'em, more trouble than they're worth and all much of a muchness. Perhaps useful on limited data sets but what's the point of doing a sim in the first place?. I FFT the raw data with no window but do it over say 100 cycles. The start and stop discontinuities automatically lose themselves in the noise. More relevant to a worthwhile FFT is getting the number of data points up. If pushed I'd use the Blackman window.

Posted by ●January 8, 2008

On Jan 7, 9:22 am, "john jardine" <john.jard...@idnet.co.uk> wrote:> "D from BC" <myrealaddr...@comic.com> wrote inmessagenews:1go3o35nobptkttckeb5tqf2b8bl1ssl9r@4ax.com...> > > > > Got a homework question, but not in school. :) > > > FFT windowing functions in LTSpice. > > None > > Welch > > Bartlett > > Hamming > > Hann > > Blackman > > Lanczos > > Parzen > > Kaiser Bessel > > > I no idea when each are applied. > > > Looked at > >http://en.wikipedia.org/wiki/Window_function > > Geezz :( Not exactly easy late night reading. > > > Saw the audio slide show tutorial on > >http://www.ni.com/swf/presentation/us/windowing/ > > > Type of Signal | Window > > ------------------+------------ > > Sine waves | Hanning > > Spectral analysis Hanning (for random excitation) > > > Transients longer > > than length of window Hanning > > Unknown content Hanning > > > Two tone > > with close f Kaiser Bessel > > > Transients duration > > longer than the length > > of the window Rectangular > > > The 'none' (no window) choice in LTSpice means rectangular. > > > Who uses Welch? > > > D from BC > > British Columbia > > Canada. > > I don't use 'em, more trouble than they're worth and all much of a muchness. > Perhaps useful on limited data sets but what's the point of doing a sim in > the first place?. > I FFT the raw data with no window but do it over say 100 cycles. The start > and stop discontinuities automatically lose themselves in the noise. > More relevant to a worthwhile FFT is getting the number of data points up. > If pushed I'd use the Blackman window.If your waveform is periodic, and the sampling time is an integer multiple of the period, no windowing is necessary. If the waveform is not periodic, OR if the sampling time is not an integer multiple of the period, then one should use a window. Mark

Posted by ●January 8, 2008

On Mon, 7 Jan 2008 19:52:39 -0800 (PST), redbelly <redbelly98@yahoo.com> wrote:>On Jan 7, 9:22 am, "john jardine" <john.jard...@idnet.co.uk> wrote: >> "D from BC" <myrealaddr...@comic.com> wrote inmessagenews:1go3o35nobptkttckeb5tqf2b8bl1ssl9r@4ax.com...>> >> >> >> > Got a homework question, but not in school. :) >> >> > FFT windowing functions in LTSpice. >> > None >> > Welch >> > Bartlett >> > Hamming >> > Hann >> > Blackman >> > Lanczos >> > Parzen >> > Kaiser Bessel >> >> > I no idea when each are applied. >> >> > Looked at >> >http://en.wikipedia.org/wiki/Window_function >> > Geezz :( Not exactly easy late night reading. >> >> > Saw the audio slide show tutorial on >> >http://www.ni.com/swf/presentation/us/windowing/ >> >> > Type of Signal | Window >> > ------------------+------------ >> > Sine waves | Hanning >> > Spectral analysis Hanning (for random excitation) >> >> > Transients longer >> > than length of window Hanning >> > Unknown content Hanning >> >> > Two tone >> > with close f Kaiser Bessel >> >> > Transients duration >> > longer than the length >> > of the window Rectangular >> >> > The 'none' (no window) choice in LTSpice means rectangular. >> >> > Who uses Welch? >> >> > D from BC >> > British Columbia >> > Canada. >> >> I don't use 'em, more trouble than they're worth and all much of a muchness. >> Perhaps useful on limited data sets but what's the point of doing a sim in >> the first place?. >> I FFT the raw data with no window but do it over say 100 cycles. The start >> and stop discontinuities automatically lose themselves in the noise. >> More relevant to a worthwhile FFT is getting the number of data points up. >> If pushed I'd use the Blackman window. > >If your waveform is periodic, and the sampling time is an integer >multiple of the period, no windowing is necessary. > >If the waveform is not periodic, OR if the sampling time is not an >integer multiple of the period, then one should use a window. > >MarkInteresting... I don't think I can get LT Spice data points that are an integer multiple of the period.. Here's the wave (time domain) that I'm trying out spectrum analysis on. Among the crazy modulation, one can see a main frequency. http://www.members.shaw.ca/chainsaw/SED/Crazywave.wmf 328Kb D from BC British Columbia Canada.

Posted by ●January 8, 2008

On Jan 8, 12:03 am, D from BC <myrealaddr...@comic.com> wrote:> On Mon, 7 Jan 2008 19:52:39 -0800 (PST), redbelly > > > > <redbell...@yahoo.com> wrote: > >On Jan 7, 9:22 am, "john jardine" <john.jard...@idnet.co.uk> wrote: > >> "D from BC" <myrealaddr...@comic.com> wrote inmessagenews:1go3o35nobptkttckeb5tqf2b8bl1ssl9r@4ax.com...> > >> > Got a homework question, but not in school. :) > > >> > FFT windowing functions in LTSpice. > >> > None > >> > Welch > >> > Bartlett > >> > Hamming > >> > Hann > >> > Blackman > >> > Lanczos > >> > Parzen > >> > Kaiser Bessel > > >> > I no idea when each are applied. > > >> > Looked at > >> >http://en.wikipedia.org/wiki/Window_function > >> > Geezz :( Not exactly easy late night reading. > > >> > Saw the audio slide show tutorial on > >> >http://www.ni.com/swf/presentation/us/windowing/ > > >> > Type of Signal | Window > >> > ------------------+------------ > >> > Sine waves | Hanning > >> > Spectral analysis Hanning (for random excitation) > > >> > Transients longer > >> > than length of window Hanning > >> > Unknown content Hanning > > >> > Two tone > >> > with close f Kaiser Bessel > > >> > Transients duration > >> > longer than the length > >> > of the window Rectangular > > >> > The 'none' (no window) choice in LTSpice means rectangular. > > >> > Who uses Welch? > > >> > D from BC > >> > British Columbia > >> > Canada. > > >> I don't use 'em, more trouble than they're worth and all much of a muchness. > >> Perhaps useful on limited data sets but what's the point of doing a sim in > >> the first place?. > >> I FFT the raw data with no window but do it over say 100 cycles. The start > >> and stop discontinuities automatically lose themselves in the noise. > >> More relevant to a worthwhile FFT is getting the number of data points up. > >> If pushed I'd use the Blackman window. > > >If your waveform is periodic, and the sampling time is an integer > >multiple of the period, no windowing is necessary. > > >If the waveform is not periodic, OR if the sampling time is not an > >integer multiple of the period, then one should use a window. > > >Mark > > Interesting... > I don't think I can get LT Spice data points that are an integer > multiple of the period.. > > Here's the wave (time domain) that I'm trying out spectrum analysis > on. > Among the crazy modulation, one can see a mainfrequency.http://www.members.shaw.ca/chainsaw/SED/Crazywave.wmf> 328Kb > > D from BC > British Columbia > Canada.Okay, looks like windowing would be a really good idea for that waveform. The "slow drift" during the scan is from low frequencies that: 1. are impossible to get an accurate spectrum for, given your sample time, and 2. will "leak" into frequencies you would like to have an accurate spectrum for. I'd try a Hamming, as well as either Hann, Gauss, or Blackman. You may or may not see a significant difference between them, but you'll be better off than not using any window. Mark p.s. Many people say "Hanning" to refer to the Hann window. It was named for Julius von Hann.

Posted by ●January 8, 2008

On Jan 8, 5:03=A0am, D from BC <myrealaddr...@comic.com> wrote:> On Mon, 7 Jan 2008 19:52:39 -0800 (PST), redbelly > > <redbell...@yahoo.com> wrote: > >On Jan 7, 9:22 am, "john jardine" <john.jard...@idnet.co.uk> wrote: > >> "D from BC" <myrealaddr...@comic.com> wrote in messagenews:1go3o35nobpt=kttckeb5tqf2b8bl1ssl9r@4ax.com...> > >> > Got a homework question, but not in school. :) > > >> > FFT windowing functions in LTSpice. > >> > None > >> > Welch > >> > Bartlett > >> > Hamming > >> > Hann > >> > Blackman > >> > Lanczos > >> > Parzen > >> > Kaiser Bessel > > >> > I no idea when each are applied. > > >> > Looked at > >> >http://en.wikipedia.org/wiki/Window_function > >> > Geezz :( Not exactly easy late night reading. > > >> > Saw the audio slide show tutorial on > >> >http://www.ni.com/swf/presentation/us/windowing/ > > >> > Type of Signal =A0 =A0| =A0 =A0Window > >> > ------------------+------------ > >> > Sine waves =A0 =A0 =A0 =A0| =A0 =A0Hanning > >> > Spectral analysis =A0 =A0 =A0Hanning (for random excitation) > > >> > Transients longer > >> > than length of window =A0Hanning > >> > Unknown content =A0 =A0 =A0 =A0Hanning > > >> > Two tone > >> > with close f =A0 =A0 =A0 =A0 =A0Kaiser BesselIn your case this might be a good choice. There are clear hints of (at least) two close frequencies (~10% different) beating mostly visible between 273us and 282us. If you can manage it create a data record that is at least piecewise continuous across the start and end boundary before applying any windowing. It is implicit in the FFT that the time series is tiled periodically so if you have a big difference between the first and last samples as in this case you get roughly the FFT of an unwanted sawtooth wave added to your wanted signal.> > >> > The 'none' (no window) choice in LTSpice means rectangular.Crude but gives you the highest resolution as well as the worst artefacts. Each window allows you to trade resolution in frequency against signal to noise in the transform. These days maximum entropy methods can be used to compute the spectrum of a time series without having to trade resolution quite so badly.> >> I don't use 'em, more trouble than they're worth and all much of a much=ness.> >> Perhaps useful on limited data sets but what's the point of doing a sim=in> >> the first place?. > >> I FFT the raw data with no window but do it over say 100 cycles. The st=art> >> and stop discontinuities automatically lose themselves in the noise. > >> More relevant to a worthwhile FFT is getting the number of data points =up.> >> If pushed I'd use the Blackman window. > > >If your waveform is periodic, and the sampling time is an integer > >multiple of the period, no windowing is necessary. > > >If the waveform is not periodic, OR if the sampling time is not an > >integer multiple of the period, then one should use a window. > > >Mark > > Interesting... > I don't think I can get LT Spice data points that are an integer > multiple of the period..But you should try to get the function sampled over a period that makes it continuous. If that low frequency is real then it should return to near the start value after about 2x the time shown. The longer the continuous run of the time series you can feed into the FFT the less effect the edge discontinuity has on your answer.> > Here's the wave (time domain) that I'm trying out spectrum analysis > on. > Among the crazy modulation, one can see a main frequency.http://www.member=s.shaw.ca/chainsaw/SED/Crazywave.wmf> 328KbRegards, Martin Brown

Posted by ●January 8, 2008

On Tue, 8 Jan 2008 05:15:11 -0800 (PST), redbelly <redbelly98@yahoo.com> wrote:>On Jan 8, 12:03 am, D from BC <myrealaddr...@comic.com> wrote: >> On Mon, 7 Jan 2008 19:52:39 -0800 (PST), redbelly >> >> >> >> <redbell...@yahoo.com> wrote: >> >On Jan 7, 9:22 am, "john jardine" <john.jard...@idnet.co.uk> wrote: >> >> "D from BC" <myrealaddr...@comic.com> wrote inmessagenews:1go3o35nobptkttckeb5tqf2b8bl1ssl9r@4ax.com...>> >> >> > Got a homework question, but not in school. :) >> >> >> > FFT windowing functions in LTSpice. >> >> > None >> >> > Welch >> >> > Bartlett >> >> > Hamming >> >> > Hann >> >> > Blackman >> >> > Lanczos >> >> > Parzen >> >> > Kaiser Bessel >> >> >> > I no idea when each are applied. >> >> >> > Looked at >> >> >http://en.wikipedia.org/wiki/Window_function >> >> > Geezz :( Not exactly easy late night reading. >> >> >> > Saw the audio slide show tutorial on >> >> >http://www.ni.com/swf/presentation/us/windowing/ >> >> >> > Type of Signal | Window >> >> > ------------------+------------ >> >> > Sine waves | Hanning >> >> > Spectral analysis Hanning (for random excitation) >> >> >> > Transients longer >> >> > than length of window Hanning >> >> > Unknown content Hanning >> >> >> > Two tone >> >> > with close f Kaiser Bessel >> >> >> > Transients duration >> >> > longer than the length >> >> > of the window Rectangular >> >> >> > The 'none' (no window) choice in LTSpice means rectangular. >> >> >> > Who uses Welch? >> >> >> > D from BC >> >> > British Columbia >> >> > Canada. >> >> >> I don't use 'em, more trouble than they're worth and all much of a muchness. >> >> Perhaps useful on limited data sets but what's the point of doing a sim in >> >> the first place?. >> >> I FFT the raw data with no window but do it over say 100 cycles. The start >> >> and stop discontinuities automatically lose themselves in the noise. >> >> More relevant to a worthwhile FFT is getting the number of data points up. >> >> If pushed I'd use the Blackman window. >> >> >If your waveform is periodic, and the sampling time is an integer >> >multiple of the period, no windowing is necessary. >> >> >If the waveform is not periodic, OR if the sampling time is not an >> >integer multiple of the period, then one should use a window. >> >> >Mark >> >> Interesting... >> I don't think I can get LT Spice data points that are an integer >> multiple of the period.. >> >> Here's the wave (time domain) that I'm trying out spectrum analysis >> on. >> Among the crazy modulation, one can see a mainfrequency.http://www.members.shaw.ca/chainsaw/SED/Crazywave.wmf>> 328Kb >> >> D from BC >> British Columbia >> Canada. > >Okay, looks like windowing would be a really good idea for that >waveform. The "slow drift" during the scan is from low frequencies >that: >1. are impossible to get an accurate spectrum for, given your sample >time, and >2. will "leak" into frequencies you would like to have an accurate >spectrum for. > >I'd try a Hamming, as well as either Hann, Gauss, or Blackman. You >may or may not see a significant difference between them, but you'll >be better off than not using any window. > >Mark > >p.s. Many people say "Hanning" to refer to the Hann window. It was >named for Julius von Hann.Yup.. That pic is <100uS .. The >1mS oscillograph had hard to see HF detail. I zoomed in to show the HF detail. I'll use more time data for better spectral LF accuracy. I'm just learning about spectral leakage. (Do I need a diaper? :P ) However... I'm learning EMI sim'ing with spice. Perhaps I should high pass filter at ~1Mhz prior to FFT ? Note: These two could be mistaken: Hanning Hamming I suspect that's why it's listed 'Hann' and not Hanning in LTSpice. D from BC British Columbia Canada.

Posted by ●January 8, 2008

On Tue, 8 Jan 2008 05:57:12 -0800 (PST), Martin Brown <|||newspam|||@nezumi.demon.co.uk> wrote:>On Jan 8, 5:03�am, D from BC <myrealaddr...@comic.com> wrote: >> On Mon, 7 Jan 2008 19:52:39 -0800 (PST), redbelly >> >> <redbell...@yahoo.com> wrote: >> >On Jan 7, 9:22 am, "john jardine" <john.jard...@idnet.co.uk> wrote: >> >> "D from BC" <myrealaddr...@comic.com> wrote inmessagenews:1go3o35nobptkttckeb5tqf2b8bl1ssl9r@4ax.com...>> >> >> > Got a homework question, but not in school. :) >> >> >> > FFT windowing functions in LTSpice. >> >> > None >> >> > Welch >> >> > Bartlett >> >> > Hamming >> >> > Hann >> >> > Blackman >> >> > Lanczos >> >> > Parzen >> >> > Kaiser Bessel >> >> >> > I no idea when each are applied. >> >> >> > Looked at >> >> >http://en.wikipedia.org/wiki/Window_function >> >> > Geezz :( Not exactly easy late night reading. >> >> >> > Saw the audio slide show tutorial on >> >> >http://www.ni.com/swf/presentation/us/windowing/ >> >> >> > Type of Signal � �| ��Window>> >> > ------------------+------------ >> >> > Sine waves � � � �|� �Hanning>> >> > Spectral analysis � � �Hanning (forrandom excitation)>> >> >> > Transients longer >> >> > than length of window �Hanning >> >> > Unknown content � � ��Hanning>> >> >> > Two tone >> >> > with close f � � � ��Kaiser Bessel> >In your case this might be a good choice. There are clear hints of (at >least) two close frequencies (~10% different) beating mostly visible >between 273us and 282us. If you can manage it create a data record >that is at least piecewise continuous across the start and end >boundary before applying any windowing.Neato :) I'll try that. . That 'clicks in' from what I recall from that slide show tutorial link.. (D)> >It is implicit in the FFT that the time series is tiled periodically >so if you have a big difference between the first and last samples as >in this case you get roughly the FFT of an unwanted sawtooth wave >added to your wanted signal. >> >> >> > The 'none' (no window) choice in LTSpice means rectangular. > >Crude but gives you the highest resolution as well as the worst >artefacts. Each window allows you to trade resolution in frequency >against signal to noise in the transform. These days maximum entropy >methods can be used to compute the spectrum of a time series without >having to trade resolution quite so badly.Maximum entropy methods?? Huhh... that doesn't sound like light reading.. :) Googled and found http://cmm.cit.nih.gov/maxent/cos.html ??? Wow looks like MEM is better than FFT. Now I'm wondering why it's not included in LTSpice... (D)> >> >> I don't use 'em, more trouble than they're worth and all much of a muchness. >> >> Perhaps useful on limited data sets but what's the point of doing a sim in >> >> the first place?. >> >> I FFT the raw data with no window but do it over say 100 cycles. The start >> >> and stop discontinuities automatically lose themselves in the noise. >> >> More relevant to a worthwhile FFT is getting the number of data points up. >> >> If pushed I'd use the Blackman window. >> >> >If your waveform is periodic, and the sampling time is an integer >> >multiple of the period, no windowing is necessary. >> >> >If the waveform is not periodic, OR if the sampling time is not an >> >integer multiple of the period, then one should use a window. >> >> >Mark >> >> Interesting... >> I don't think I can get LT Spice data points that are an integer >> multiple of the period.. > >But you should try to get the function sampled over a period that >makes it continuous. If that low frequency is real then it should >return to near the start value after about 2x the time shown. The >longer the continuous run of the time series you can feed into the FFT >the less effect the edge discontinuity has on your answer.Ahhhh... :) Thanks..>> >> Here's the wave (time domain) that I'm trying out spectrum analysis >> on. >> Among the crazy modulation, one can see a mainfrequency.http://www.members.shaw.ca/chainsaw/SED/Crazywave.wmf>> 328Kb > >Regards, >Martin BrownD from BC British Columbia Canada.