Who made the comment about modems and FSK signals?

"Brian Reinhold" <breinhold@comcast.net> wrote in message
news:ziaOb.84961$xy6.144535@attbi_s02...
> [...] method snipped.
> And compare which is largest.

That looks good to me -- providing you remembered to filter out the
out-of-band components first.

> If 'total0' is greater than 'total1' then tone 0 was present.  I think
that
> is what you described below.  It's just in my computations I find that the
> magnitude of the 'totalx' values are consistently larger with one wave
> versus the other, and I beleive this is due to the fact that one tone has
> only 1 and 1/12 of a period while the other has 1 and 3/4 of a period. But
I
> am not sure.

I'm going to guess that your signal has a DC offset or a 60hz hum that you
haven't removed.

"Matt Timmermans" <mt0000@sympatico.nospam-remove.ca> wrote in message
news:o10Pb.8158$U77.601294@news20.bellglobal.com...
>
> "Brian Reinhold" <breinhold@comcast.net> wrote in message
> news:ziaOb.84961$xy6.144535@attbi_s02...
> > [...] method snipped.
> > And compare which is largest.
>
> That looks good to me -- providing you remembered to filter out the
> out-of-band components first.
>
> > If 'total0' is greater than 'total1' then tone 0 was present.  I think
> that
> > is what you described below.  It's just in my computations I find that
the
> > magnitude of the 'totalx' values are consistently larger with one wave
> > versus the other, and I beleive this is due to the fact that one tone
has
> > only 1 and 1/12 of a period while the other has 1 and 3/4 of a period.
But
> I
> > am not sure.
>
> I'm going to guess that your signal has a DC offset or a 60hz hum that you
> haven't removed.

Well, I have not filtered the incoming signal, but I do remove the mean over
the samples which I compute the matched filter. However, I think I am
running a case of a clean signal.  At least the noise that does exist has 32
times smaller amplitude.

I also normalize that above 'sum of products' with the 'sum of products of
the matched filter with itself'.  The normalization value computed with the
two tones are different, but it doesn't help enough.

When I graphically look at this value of 'total' as it is being computed for
a dot pattern (alternating tones of equal baud length), for short bauds the
pattern looks like a sine wave with lots of small scale wiggles on it, and
the wiggles are quite large in amplitude.  As the baud length becomes longer
(so each tone includes more and more periods) the sinusoidal shape
approaches a triangular shape and the wiggles become much smaller in
amplitude and relative scale ( of course the bauds are longer so the
relative scale is expected to shrink).

These wiggles are probably the source of the problems I am having with the
detection algorithm.  They maybe an unavoidable feature of having only about
one period of signal to work with.

Brian
>
>

"Brian Reinhold" <breinhold@comcast.net> wrote in message
news:GZ7Pb.105643$8H.182368@attbi_s03...
> When I graphically look at this value of 'total' as it is being computed
for
> a dot pattern (alternating tones of equal baud length), for short bauds
the
> pattern looks like a sine wave with lots of small scale wiggles on it, and
> the wiggles are quite large in amplitude.  As the baud length becomes
longer
> (so each tone includes more and more periods) the sinusoidal shape
> approaches a triangular shape and the wiggles become much smaller in
> amplitude and relative scale ( of course the bauds are longer so the
> relative scale is expected to shrink).
> These wiggles are probably the source of the problems I am having with the
> detection algorithm.  They maybe an unavoidable feature of having only
about
> one period of signal to work with.

Oh, I see.  That sounds right to me, too.  A baud-length snippet of signal
that isn't aligned to the signal boundaries can look pretty wierd when the
baud length is so small.  If you're not doing any clock recovery, then I
would recommend a detection algorithm with a narrower bandwidth.  You could
try TIs algorithm, or you could pass your total1-total0 through a low-pass
to filter out the ripples.

You can make the system more robust if you recover the clock, and only look
for symbols at the appropriate symbol instants.

"Matt Timmermans" <mt0000@sympatico.nospam-remove.ca> wrote in message
news:R7bPb.5109$rW5.277264@news20.bellglobal.com...
>
> "Brian Reinhold" <breinhold@comcast.net> wrote in message
> news:GZ7Pb.105643$8H.182368@attbi_s03...
> > When I graphically look at this value of 'total' as it is being computed
> for
> > a dot pattern (alternating tones of equal baud length), for short bauds
> the
> > pattern looks like a sine wave with lots of small scale wiggles on it,
and
> > the wiggles are quite large in amplitude.  As the baud length becomes
> longer
> > (so each tone includes more and more periods) the sinusoidal shape
> > approaches a triangular shape and the wiggles become much smaller in
> > amplitude and relative scale ( of course the bauds are longer so the
> > relative scale is expected to shrink).
> > These wiggles are probably the source of the problems I am having with
the
> > detection algorithm.  They maybe an unavoidable feature of having only
> about
> > one period of signal to work with.
>
> Oh, I see.  That sounds right to me, too.  A baud-length snippet of signal
> that isn't aligned to the signal boundaries can look pretty wierd when the
> baud length is so small.  If you're not doing any clock recovery, then I
> would recommend a detection algorithm with a narrower bandwidth.  You
could
> try TIs algorithm, or you could pass your total1-total0 through a low-pass
> to filter out the ripples.
>
> You can make the system more robust if you recover the clock, and only
look
> for symbols at the appropriate symbol instants.

I hear a lot about 'clock recovery' but I am not sure how one would do that.
The timing is supposed to be x baud but there should also be a certain
tolerance to error (there may be a 5% difference between the sender's and
receiver's clocks). So I do not try to 'predict' the length of the next baud
pulse based upon what I have received.  I look for transitions and estimate
the number of baud pulses based on the time between transitions and the
expected length of a baud pulse.  I do know that no more than 7 zero or one
bits can appear sequentially in the message, so that kind of puts an upper
bound on the tolerance.

Right know I am doing a simple computation to see how the wiggles and
overall signal should appear.

Brian
>
>