Tchamov Nikolay nikolay at cs.tut.fi
Mon Oct 19 07:26:46 EST 1992

                                   October 19th, 1992, Tampere

Dear Coleagues,

I am working with real data from analog sensor - a common problem
for the newsgroups where I am posting this Request. I promice to post 
summary of your responses - I do believe this concern quite many 
of us. Thanks is advance for your Attention and time for Reading and
Responding to this. 

  Description of the problem:
1. I have pulses obtained by Analog to Digital converter (A/D) with 
correct sampling rate. The pulses are only positive and for each of 
them it is assumed that there are at the output of the A/D several
samples (2, 3, 4 or more). 

2. From those samples I need to find out very quickly (the speed is
quite important) the following parameters of the pulse:
     - Time for reaching the maximum of the voltage
     - The Value of the maximum of the voltage
     - Energy related parameter = the surface covered by the pulse,
				  (meaning the Integral of its shape)

   The General Solution (is this the most common on ???):
The approaches like Aproximation of the shape and then calculating
the parameters needed are suitable of course, but they are not enough
fast and actually we don't need to know all shape, but only some of
its parameters = 3 numbers. Of course polynomial(quadratic, cubuc or
spline) approximations will be most safety approach giving us the 
coefficients of the polynom, easy applicable in certain formulas for 
calculation of parameters needed. Any thing else recognised to be
common general solution ?

   Perhaps it is possible (please give your comments):
1. It is possible (perhaps?) to derive (using symbolical calculations 
by packages like "Mathematica") straight fromulas for calculating the 
parameters needed having as an input only the coordinates of the samples. 
Those formulas can be simplified up to some stage for obtaining sufficient
precision or approximated by FIR-alike sums to facilitate application 
by high-speed signal processors having fast multiplicaion & summation
operations build in. Is it worth trying this ???

2. The shape of the pulses can be assumed to converge to some well known
shapes of statistics destributions. From the comming in samples it will be
(perhaps?) possible to calcualte the parameters needed meaning: using
the formulas derived for destribution shape and assuming that this is 
the shape of the pulse as well. Is this sounds too strange as an approach?

3. In the literature reviewed I have met some (lets say) modern approaches 
dealing with simulaire processing of incoming from A/D data, like using 
deconvolution, spectral approaches, wavelets and so on but nothing very 
closed to my problem. Do you think that there are some really very effective 
ways to treat this problem ? I have a strong hope that very interesting and 
even crazy approaches have been applyied to this so common problem, but I 
wasn't happy to meet them in the literature reviewed. Let me know the your 
opinions, please.

Any comments, titles of papers, books, reports, software etc. will be great 
help I hope not only for me. I guess, many of us are looking not only to 
solve somehow the problem but to make something better than the standard 

Waiting for your reply,

Best regrds,


 Nikolay Tchamov, Signal Processing Laboratory    Tel: 358-31+16-1885
 Tampere University of Technology                 FAX: 358-31+16-1857       
 P.O.Box 553, SF-33101 Tampere, Finland     E-mail: nikolay at cs.tut.fi

More information about the Comp-bio mailing list