PS I think one way to get the average waveforms I want from the analysis is using cross-correlation, but again the multiple scale problem is present.
On Wed, Oct 21, 2009 at 9:56 AM, William Simpson <william.a.simp...@gmail.com> wrote: > I have three time series, x, y, and z, and I want to analyse the > relations between them. However, they have vastly different > resolutions. I am writing to ask for advice on how to handle this > situation in R. > > x is a stimulus, and y and z are responses. > > x is a rectangular pulse 4 sec long. Its onset and offset are known > with sub-millisecond precision. The onset varies irregularly -- it > doesn't fall on neat 1/2 sec or sec boundaries for example. > > y is a sampled continuous waveform. It is highly noisy, and it is > actually well represented by samples 1/2 sec apart or so. > > z is a very short pulse -- perhaps 5 ms long -- occurring at irregular times. > > I have problems with how to represent these waveforms at the input > stage (during data collection) and at the analysis stage. If I want to > represent x and z with the sort of precision I'd like, I'd have to > sample every ms or so. But: > - that is massive overkill for y, because it is noisy. I will have > maybe 500 times as many pts as I require! That makes for large data > files, for no good reason. > - the representations for x and z are about 99% zeros. Again, wasteful > for storage > - the analysis will be awkward and slow because of the huge number of pts > > If I sample at a rate of every 1/2 sec, z may be not detected at all, > and the edges of x are represented very poorly. > > I could just save y as a separate file, with values sampled every 1/2 > sec, save x as a file containing onset and offset times, and save z as > the times of each impulse. All three files have different lengths. If > I save this way it is awkward keeping track of the three files for > each run, and I'm still left with the problem of analysing them. In my > mind they are actually three continuous waveforms, and my > simple-minded way of analysing would be to create digitised waveform > versions of all three (on some fine grain). If I do that I still have > the problem of zillions of useless points clogging RAM. > > What do I want from the analysis? > I want the average waveform for y that occurs for 30 sec after each x > pulse, and the average waveform for z that occurs for 30 sec after > each x pulse. The average waveform for y surrounding each z (maybe 30 > sec before and 10 sec after) would also be useful. > > Thanks very much for any suggestions about representation for data > collection and analysis, and for data analysis methods (which will > depend on the data representation). > > Bill > ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
