I don't think I'm really looking at something a lot different --
probably I need to do some more building and measuring and analyzing.
I was thinking that 2kHz to 3kHz would contain both a good amount of
noise, be below the pass-band cut-off , and have enough voice
components in it.... That was a wild guess based on the Spetra-TAC
voter working above 2kHz with the theory of operation section
describing, essentially, a fair amount of noise and mostly vocal
harmonics with less energy, though it still looks for valleys.
My thinking on the peak-valley thing worked like this and I'M probably
the one not thinking clearly: I know voice isn't a "tone" but I'll use
the tone concept for simplicity -- I'm essentially freezing a couple
of moments in time, so that's cool? If I have that ubiquitous 1kHz
audio signal modulated at +/- 3kHz fed to two receivers, and dead
carrier -- my two "test cases" if you will. If one receiver is
considerably more quieted than the other then:
When there is dead air, the valley would be detected. The noisy
receiver would show a "higher" level -- more energy in the audio
spectrum than the less noisy. Low level is best signal.
When the 1kHz +/-3kHz audio signal is present, the noisy receiver
would show a "higher" level -- more energy in the audio spectrum due
to the signal + noise component. Lower level is the best signal.
Here's where I'm probably making a mistake: The "energy" in the audio
spectrum is the intelligence (signal) energy + noise energy at any
time t. Am I missing something where the intelligence energy is
reduced on a noisy signal or something?
Seriously, I am doing building and measuring, but don't have the
experience of the group, and also may not be using some terms correctly.
On Jun 20, 2008, at 5:52 PM, Jeff DePolo wrote:
> Not quite enough of a programmer to take on the DSP, but will
> likely look at the peaks and valleys with the ADC. I'm not
> sure why I'd need more than one noise circuit though. I don't
> want to do a sample and hold, the ADC and software can do
> that. I was thinking build one analog circuit and look for
> valleys and peaks when the ADC reads the analog circuit
> output -- that is to say, keep track of the highest level and
> lowest level over a certain very short time period
I'm not sure I follow. I would think that for peak and valley
detection to
work right, you need to look at the voice spectrum, not the noise
spectrum,
and use the ratio of the peaks to valleys to compute a value
indicitive of
the S/N, and then compare S/N values among the active channels to
determine
which gets voted. I think that this kind of peak to valley ratioed
comparison would help "even out" differences in audio levels between
receivers (since you're comparing ratios, not absolute levels). I
would
also think that by looking at the audio passband alone, it would also
minimize the detrimental effect of frequency response differences
between
sources, particulary with regard to the typical high-end rolloff
above the
audio passband for sources backhauled across links as compared to
the local
receiver, which is often the most challenging obstacle to overcome as
mentioned previously.
If you look at only the noise spectrum, then peaks and valleys don't
have
much relevance; you would only be able to compare the average noise
levels
to determine which channel gets voted. This is way the LDG is
designed. If
you look at the noise spectrum only, audio that is full-quieting
would have
a peak to valley ratio approaching unity, but so would a hissy, but
stable,
signal. But if you looked at the voice spectrum, the full-quieting
signal
would have a greater ratio between peaks and valleys than would one
with a
steady hiss behind the audio.
Maybe you're suggesting something completely different and it went
over my
head?
--- Jeff
--
Cort Buffington
H: +1-785-838-3034
M: +1-785-865-7206
