I hope this thread goes on for a while, because I am very interested in
the question "how does DSP noise reduction work?"
I know that some methods of autocorrelation can find correlated signals
that are at, or even significantly below, the noise level. However they
require really long integration times. In our applications, SSB voice or
CW tone detection, there are limitations imposed on the correlation
function, because we can only tolerate a limited amount of latency. In
other words, since our DSP algorithms have to work in near real time, we
cannot necessarily take full advantage of all that is possible with DSP.
As this discussion continues (and I hope it does) I would not be
surprised that someone who knows a lot more about it than I do, explains
that because of these limitations DSP as we use it really does not
accomplish much more than automatic bandpass filtering and leveling.
I do not have an Orion, and the only DSP in Amateur Radio I have
experienced is several years old technology. An Omni VI and also JPS
NIR-12 is the DSP that I have listened to. They sound to me like
bandwidth limiting and AGC and noise gate type of action. So far when
using a really good receiver (Omni VI) , I don't believe that using this
kind of DSP I have ever been able to copy a CW signal with the DSP that
I could not copy without it. When I had the JPS NIR-12 hooked up to my
old Kenwood TS-440, there were some times that the DSP made the
difference between copying and not copying, but then the TS-440 is not
nearly as good a receiver as the Omni VI.
I expect that the Orion has better DSP algorithms than the Omni VI and
the JPS NIR-12, and there may be a large advantage in performing the DSP
digitized IF instead of from audio. Still there may be limitations that
are just unavoidable. DSP is after all NOT magic, and can only do what
is physically and mathematically possible.
DE N6KB
Grant Youngman wrote:
work.) My DSP theory is not fresh in my memory any more, but the
discussions here about DSP only being a narrow filter does
not map with my experience in DSP. Maybe it's old knowledge,
but BW filtering doesn't get the entire job done, it seems to me.
A common DSP algorithm as might be found in many amateur devices for noise
reduction uses autocorrelation to identify the primary spectral components,
and then adapts dynamic filters around those components. That should sound
pretty familiar. It isn't really a one or the other situation. And
autocorrelation isn't the only method for spectral estimation.
There are as many fancy algorithms around as there are graduate students
working on dissertations in the area, speech-recognition companies trying to
get a leg-up on the competition regards recognition accuracy in noisy
places, and on and on.
|