On Fri, 2006-03-24 at 08:56 +0100, OE3ZK Gerhard wrote:
Thank you Bill, Jerry, Sinisa, Lin and others for the debate about NR.
>
> It did help my understanding!
>
> What I read in these responses is that many of us see filtering /bandwidth
> reduction
as the sole means of improving the ability to recover information from noise. I
feel
this misses the point that I am exploring. In his 1948 landmark paper, Claude
Shannon
extends the work of Nyquist and Hartley. Significantly Shannon refers to
statistical
processes and entropy (the degree of randomness) of a signal. In the case of CW
some of
that processing takes place in our heads. This says to me that data processing
ability
ultimately becomes key to signal recovery and as there is a clear difference in
the
entropy of coded data vs analog voice, those processes should be different. I
very much
doubt that deep space communications relies on bandwidth reduction alone
>
>
> What I fail to understand is the claim by users of Linrad and Winrad that DSP
> noise
reduction is improved by processing noise in a much wider band than the signal
itself.
It appears to be born out by examples, I just don't know why...
>
> 73 Gert, OE3ZK
>
>
>
> That's because the noise reduction comes from the signals being correlated
> (e.g. a
collection of sine waves and signals that are the same after some selected time
delay,
at least one cycle) while the noise is truly random and the wider the
bandwidth, the
more random the noise. Listen to noise through a 200 Hz or 100 Hz filter. It
has a
pitch. The narrow filter is suppressing noise spectrum and what is left is
nearly the
same frequency so is partly correlated.
And the spectrum seen ought to fall off about 3KHz, isn't that the roofing
filter at work?
Noise correlates less well at higher audio frequencies than at low with a short
time
correlation. And so without a long correlation time, the higher frequency noise
reduction from lack of correlation is better that contributes to the roll off.
Measuring noise through a narrow bandwidth filter also has lots of amplitude
variation
with time (a time constant probably about the inverse of the filter bandwidth).
The
wider the bandwidth of the measurement, the more constant the meter reading.
Hence
commercial noise figure meters like to use several MHz bandwidth. Measuring NF
with a
noise generator and a 2 KHz filter can be tedious deciding how to visually
average the
meter pointer to that 3 dB rise when its jumping up and down a dB. Going to a
narrower
bandwidth makes it much worse.
--
Gerald J.
Dr. Gerald N. Johnson, electrical engineer
WBCCI #5623, VAC
All content copyright.
_______________________________________________
TenTec mailing list
TenTec@contesting.com
http://lists.contesting.com/mailman/listinfo/tentec
|