This is a random question on a simple topic. Sorry if this is a little trivial. If you were to filter a signal with a windowed sinc filter, (using a Blackman window), would you get better results using a large 20,000 point window or a smaller 2,000 window? Consider the signal has no step functions or spikes, and it is a 250 sample per second signal. Consider the performance in the time domain and not the frequency domain. Thanks.
Window size for digital filtering.
Started by ●August 13, 2008
Reply by ●August 13, 20082008-08-13
On 13 Aug, 20:33, "tkremund98" <tkremun...@hotmail.com> wrote:> This is a random question on a simple topic. �Sorry if this is a little > trivial. �If you were to filter a signal with a windowed sinc filter, > (using a Blackman window), would you get better results using a large > 20,000 point window or a smaller 2,000 window?That's impossible to answer since you haven't described neither the context of the data or what you attempt to do. Filters are often designed in the context of sampled data. In that case the length of FIR filters is determined by the width of the transition band relative to the sampling frequency. For most window-method FIR filters the transition bandwidth is the only design factor, as ripple and stopband attenuation are implicitly given by the choise of window function. You have stated no design goals, so the only thing that can be said with some certainty is that the 20000 coefficient filter will require about ten times as many FLOPs as the 2000 coefficient filter. Rune
Reply by ●August 13, 20082008-08-13
tkremund98 wrote:> This is a random question on a simple topic. Sorry if this is a little > trivial. If you were to filter a signal with a windowed sinc filter, > (using a Blackman window), would you get better results using a large > 20,000 point window or a smaller 2,000 window? Consider the signal has no > step functions or spikes, and it is a 250 sample per second signal. > Consider the performance in the time domain and not the frequency domain.Define "better" for your needs. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●August 13, 20082008-08-13
On Aug 13, 7:41 pm, Jerry Avins <j...@ieee.org> wrote:> tkremund98 wrote: > > This is a random question on a simple topic. Sorry if this is a little > > trivial. If you were to filter a signal with a windowed sinc filter, > > (using a Blackman window), would you get better results using a large > > 20,000 point window or a smaller 2,000 window? Consider the signal has no > > step functions or spikes, and it is a 250 sample per second signal. > > Consider the performance in the time domain and not the frequency domain. > > Define "better" for your needs. > > Jerry > -- > Engineering is the art of making what you want from things you can get. > �����������������������������������������������������������������������An example of why "better" must be defined based on your needs: the width of the main lobe in the frequency response of the Blackman window is inversely proportional to N, the length of the window. So, if you're looking for a narrow main lobe (i.e. little "smearing" of the frequency response of the sinc filter that you're windowing), a longer window does help. What doesn't improve with a longer window is the sidelobe attenuation; for the classical Blackman/Hamming/etc. windows the sidelobe height is independent of N. So if you're more concerned about the height of the side lobes (i.e. if you're more concerned with suppressing energy outside the main lobe), increasing N won't do anything for you; you'll need to use a different window or use one of the parameterized windows (e.g. Kaiser) and traverse the trade space of mainlobe width versus sidelobe height. Jason
