Adaptive demodulation!

Hi!

I have to demodulate a signal (detect the envelope)
for which the bandwidth is not known in advance.
By filtering out the "noise" with a bandpass filter, I can get a
substantial improvement in signal to noise ratio of the demodulated signal.

I was thinking about using an adaptive filter of
some kind, but maybe somebody already has
some expirience or ideas on this.

I think that just the RMS (the maximum total power) of
the envelope could be a guide for optimization process
that would be forming the filter.

The learning phase is possible.

Thanks!
Atmapuri.

If you are using an envelope detector to detect AM, and you are
operating near threshold, you will get a much better improvment by
filtering before detection.  Limit the BW as much as possible before
the envelope detector.  Filtering after detection is less effective.
An envelope detector is not a linear process.

If you are operating with a high S/N this is less true, but it is very
true with a low S/N near threshold.

If you use a synchronous detector (which requires carrier recovery) I
belive (I'm not sure) filtering before or after detection is equivalent
theoretically but there are still implementation issues.

In eiter case you will need to make some estimate of the BW of the
signal in order to set the filter BW.  

Mark

Hi!

> If you are using an envelope detector to detect AM, and you are
> operating near threshold, you will get a much better improvment by
> filtering before detection.  Limit the BW as much as possible before
> the envelope detector.  Filtering after detection is less effective.
> An envelope detector is not a linear process.

That is the aim. But the location of the BW is unknown. All that is known
is that:

- it is there, it is modulated.
- the modulate signal has about 1kHz of bandwidth and is actually
pure noise, except for its modulation.
- the signal is fairly strong.
- but the location in the frequency is not know in advance.

From your expirience, is it possible to obtain a better SNR of the resulting
envelope, but using the Hilbert transformer approach?

> If you are operating with a high S/N this is less true, but it is very
> true with a low S/N near threshold.

The envelope is my information.
Any other freqency peaks and all the other stuff in the raw signal, should
be
removed automatically, before the envelope detection to remove
any stuff that does not contribute to the strength of the demodulated
signal.

> In eiter case you will need to make some estimate of the BW of the
> signal in order to set the filter BW.

That was the question . Must be automatic. How?

Thanks!
Atmapuri.

> That is the aim. But the location of the BW is unknown. All that is
known
> is that:
>
> - it is there, it is modulated.
> - the modulate signal has about 1kHz of bandwidth and is actually
> pure noise, except for its modulation.
> - the signal is fairly strong.
> - but the location in the frequency is not know in advance.
>
>

I guess I don't understand the question,

first you said the bandwidth was unknown.

now you are saying the BW is 1 kHz but the center frequency is unknown?
Mark

Hi!

> now you are saying the BW is 1 kHz but the center frequency is unknown?

1kHz is guess. Both is unknown, bandwidth and center frequency. But typical
good enough bandwidth is probably 1kHz.

Thanks!
Atmapuri