Hi, due to limited verbal skills in english math, i appoligize for any spelling errors and 'confusing' explanations :) Warning, lengthy skribblings ahead, Scott Dattalo wrote: >1) what is the expected frequency of the noise? Well as we all know it depends on the sampling frequency and the all so famous nyquist frequency :). Let's say that the noice can have higher frequency components than my sampling frequency. This will give me an somewhat visually 'erratic' result as the samples would tend to 'jitter'. Unfortunately I cannot increase the sampling rate. So although the noise is mostly sinusodial (spelling ?) it will not *look* like that :) Further the 'noise' consist of several overlayed frequencies, some which are far below of my sampling frequency. So to summarize, my sampling frequency is 228hz ( or thereabouts :) ) The 'noise' is an mulitude of sinosodial frequenies, all from 1 hz to above my sampling frequency. The 'pulse', which is an ideally an true square pulse, will amplify the 'noise'. It's an fairly 'odd' signal to decribe. The signal has 'mechanical' origin hence the somewhat slow frequencies. 1'st try to explain: Visualize an clean dc level signal, add a number of sinusodial component of magnitude 1 ( well some gaussian to :) ) add an pulse in the middle of magnitude 4, as the pulse will amplify the noise, the start of the pulse will have an *amplified* image of the previous 'noise' level, lets say peak noise magnitude of 3-4. (visually similar to an 'overshot') this will decrease logaritmically until the end of the pulse (ideally), where the pulse level, approximately will be the 'dc'level+pulse amplitude. ( in 'real-life' the pulse amplitude will vary within 16 bits of magnitude ) 2'nd try: Imagine an heavy table mounted ontop of an scale, now add a number running (different freq) motors on this table. Each motor has an weight on one side of the axle thus giving an sinusodial affect on the scale. Add some normal gaussian noise ( albiet fairly small amplitude). Drop an object ( let's say an ball,to simplify ) onto the table the table will be set into motion from the kinetic energy of the ball. The ball rolls over the table and of the edge. Now the 'task' is to measure the weight of the ball. Sounds fun, right :) This is not to far from reality. The affect of the pulse 'amplifying' the noise and decreasing logaritmically is whats pointing me towards wavelet denoising, as it will give an DC residue when run through an normal lowpass filter. Lets say this system gives 19 bits accurate output in 'static' mode (i.e. dc level filtering) given my current filtering. Using 'normal' averaging and median techniqes only gives me about 15 bits of accuracy. I want to achive atleast 1 bit more, to 16 bit accuracy. >2) what is the pulse width? I prefer to 'measure' in samples, lets say about 80-200 samples. Edges about 4-20 samples (unrelated to pulse with). >Is it possible to pass the signal through a differentiator or high pass >filter to extract the edges of the pulse? Well I can apply 'other' means to detect the edges. So yes, the edges can be 'known'. BTW this is used by current median/averaging techniqe to achive about 15 bits of precision (knowing where the edges are). /Tony -- http://www.piclist.com hint: To leave the PICList mailto:piclist-unsubscribe-request@mitvma.mit.edu