On Tue, May 11, 2010 at 4:46 PM, Erik de Castro Lopo <mle+la@xxxxxxxxxxxxx> wrote: > Fons Adriaensen wrote: > >> 2. the sum of the filter outputs equals the original input. >> >> This is a *hard* problem. Basically none of the classical filter >> types have these properties. There are filter sets that can do >> this but they are quite esoteric. > > Linkwitz-Riley filters: > > http://en.wikipedia.org/wiki/Linkwitz-Riley_filter > > I'm pretty sure they can just be converted to the digital domain > using the bilinear transform. > >> I'd have a look at wavelet filters for example. > > Do look at the discrete time version of the continuous wavelet > transform, but from my experience, the discrete dyadic wavelet > transform is a complete waste of time for anything other that > sub-band coding applications. > > Also, is this a bit more appropriate for the LAD list? > > Erik > -- Thanks Erik. I do think this is more appropriate for LAD. I'm not subscribed there but will get subscribed. This topic is, for me, mainly about learning something about the topic of DSP and then applying it to something a bit different. It has application, I think, to stock market trading, but at the same time I've run across a few papers here and there talking about 'listening' to the stock market - converting what's going on into audio to then viewing the market from another perspective. Our ears are far more sensitive to changes than our eyes so why not use them? If I develop something interesting then giving back to the Linux audio community that has given me so much over the years would be cool. Keeping in mind that I'm not a programmer I think that is unlikely, but who knows and why not try? Addressing your comment about the non-DC bias, I agree so I detrend all the data first with longer term trend info to remove that component of the data and give it something approaching zero offset over time. While not necessarily appropriate for audio, one difference with stock data is there is no absolute high or low - no real voltage range - and time is fungible (there aren't necessarily specific time periods like we have with audio) so I'm looking for things like 30-bar oscillations, and not necessarily 333 Hz signals. (I hope that's clear...) Still, I'm free to do anything I want with the data to make it easier to process which might be completely inappropriate for real audio data. Again, thanks for the ideas. Moving this conversation to LAD in the next day or two but will certainly correspond with this list if there's interest or anyone off list if they want to. Cheers, Mark _______________________________________________ Linux-audio-user mailing list Linux-audio-user@xxxxxxxxxxxxxxxxxxxx http://lists.linuxaudio.org/listinfo/linux-audio-user
