Conventional matched filtering should really be understood as a truncated version of the signal recovery method in the Excel example at http://www.bretcahill.com.
It sounds counter intuitive but by omitting the signal recovery steps traditional match filtering has the often desirable additional effect of low pass filtering. Since noise is often at a higher frequency than signals few were willing to question a situation where you get 2 filters just by being too lazy to complete what should be considered the first filtering operation.
Even when signal detection is all that is desired, however, it would be better to separate what should really be treated two distinct processes. Recover the original signal by taking the deconvolution of the match filter output, and then, if any additional frequency filtering in necessary, tailor it to the situation and to your needs.
And, of course, when it comes to signal recovery this should be the optimal filter.
> I just assumed "matched filtering" included the deconvolution and > other steps to recover the original signal's shape. > > In this case it was actually easier to invent a new filter -- I > assumed it already existed -- than to be scholarly and do some > research. > > To avoid confusion with the terminology from now on the filter that > takes the convolution of a noisy signal like a conventional matched > filter and then takes the deconvolution to recover the original wave > form should be called "signal recovery matched filtering." > > This will distinguish it from the signal detection matched filter > _even if the new filter is used for signal detection_. > > I've updatedhttp://www.bretcahill.comto reflect the new terminology. > > If anyone can think of a better name, i.e., the "Cahill Filter" please > post.