Olin Lathrop <PICLIST@mitvma.mit.edu> wrote:

> It might be better to detect and reject spikes up front, then use
> a simple filter afterwards.

    That's what Fr. Tom seems to have suggested, as well.

    That approach, however, won't give the same sort of results as a
    median filter would.  Median filters are WONDERFUL for pulling
    square-wave pulses out of noise; because a median filter always
    SELECTS from the sampled data rather than creating new data,
    pulse edges are REALLY sharp.

    Additionally, the performance of the filter is relatively
    insensitive to the width of the moving filter window; the
    filtered output doesn't degrade until the window gets VERY large
    relative to the width of the pulses you're trying to find.

    If I were Andy Shaw, I'd probably sample data into a circular
    buffer, then NOT sort it.

    Instead of sorting, I'd find the median value by just comparing
    each value in the buffer with the others, until I found a value
    that was greater-than or equal to "x" other values AND less-than
    or equal to "x" other values, where "x" is (number of samples in
    the buffer - 1)/2.  I'm assuming an odd number of samples in the
    buffer, and that the buffer is relatively small... Like 5 to 15
    samples.

    That process can be really fast, especially if you think a
    little before implementing the algorithm, and it will let you do
    the FIFO thing with your circular buffer without requiring any
    extra RAM for a sorted buffer or array of timestamps.

    -Andy

=== Andrew Warren -- aiw@cypress.com
=== Principal Design Engineer
=== Cypress Semiconductor Corporation
===
=== Opinions expressed above do not
=== necessarily represent those of
=== Cypress Semiconductor Corporation

--
http://www.piclist.com hint: The PICList is archived three different
ways.  See http://www.piclist.com/#archives for details.