AAC | ADPCM | Convolution | DAFx | FFT | IIR | Mixer | MP3 | MPEG | MPEG-4

Hi,

I have been using a recursive Butterworth filter with lag compensation implemented in Excel. It
works very well in my data application in a non-real time application. I want to use the same
filter in realtime while capturing data, however, since the indices look ahead, that's not
possible. With a non-recursive filter, the filter indices can be made to operate only on the
current data and previous samples thus making it causal. In this application, I don't not see
how to do this. Can anybody in this group offer a suggestion?

The Excel file I started working from as well as information about the file can be found at the
websites below.

Thanks,

Art

http://users.wfu.edu/keadjr3/Biomechanics/Project/Menu.htm
 
http://users.wfu.edu/keadjr3/Gait_DATA.xls
	

(You need to be a member of audiodsp -- send a blank email to audiodsp-subscribe@yahoogroups.com )

From: <djtrader10@djtr...>

> I have been using a recursive Butterworth filter with lag
> compensation implemented in Excel. It works very well in
> my data application in a non-real time application. I want to
> use the same filter in realtime while capturing data, however,
> since the indices look ahead, that's not possible. With a non-
> recursive filter, the filter indices can be made to operate
only
> on the current data and previous samples thus making it
> causal. In this application, I don't not see how to do this.

There are two issues here. First, the webpage says that you run
the filter once in each time direction*. This means that the
effective impulse response is infinite in both directions; hence
you need to window the effective response to get an implementable
FIR filter. I once found Harris' classic window survey paper
(citation in [1]) online but can't relocate it. I'll send you the
PDF if you want. If you use it, note the papers correcting typos
in that article also cited in [1]. In any case, search the
comp.dsp newsgroup for discussion of FIR windowing.

Second, once you have a filter of finite extent into the future,
you use it online by simply shifting it to be causal and
relabeling time in the output sequence. Mind the startup
transient, though.

* I guess that's what "lag compensation" means to you. People in
other fields, especially audio, might take that to be just
realigning signals from parallel processing paths with different
delay.
[1] http://www.faqs.org/faqs/dsp-faq/part1/ , section 1.1.4
	Martin
	

(You need to be a member of audiodsp -- send a blank email to audiodsp-subscribe@yahoogroups.com )

From: "Art Zikorus" <djtrader10@djtr...>

>   Thanks for your reply and comments. I must be missing it,
> but I don't see how to rearrange the indices in this filter for
> causal operation. I experimented with indices and could not
> get back to the zero lag ( zero phase shift between input and
> second pass output) after the second pass in the non-causal
> case. Seems that both the first and second pass indices
> should simply be offset, however, it doesn't seem to work
> correctly. If you can help out here, I would greatly appreciate
> it and if possible, please make changes directly to the Excel
> file cited in my earlier post.

I don't have Excel, so I can only say that the same filter
applied in both directions should give a zero-phase result
whatever the offset, causal or not. Maybe you can show what
you've done in pseudocode or equation form?
	Martin
	

(You need to be a member of audiodsp -- send a blank email to audiodsp-subscribe@yahoogroups.com )