On 07/09/2011 08:15 PM, Randy Yates wrote:
> I may very well be asking this question prematurely, and I may
> actually already know the answer but have forgotten it. In either
> case, please be kind...
>
> In binary FSK demodulation, one normally puts a bandpass filter around
> each of the mark and space frequencies. How small can this filter be?
>
> My thoughts cycle like this:
>
> 1) We want the filter to be as small as possible to remove as much
> noise as possible.
>
> 2) But if the filter is too narrow compared to the baud rate, the
> filter output doesn't reach full output before the transmitter
> switches to the next baud (and thus possibly the other frequency),
> and so we get worse noise performance.
>
> This must be covered somewhere in the basics, but I'm missing and/or
> forgetting it. Memory refreshment would be appreciated!

What kind of FSK?  What's the shift/baud ratio?  In narrow FSK modes, 
like MSK and Bell 103 modems, the spectra of the individual tones bleed 
into each other (this is extreme in the MSK case, which is either PSK or 
FSK depending on how you look at things).  In wideband FSK modes the 
spectra of the individual tones are distinct, and individual matched 
filters followed by a detection scheme that compares the power out of 
each filter at each bit time is the way to go.

-- 

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Do you need to implement control loops in software?
"Applied Control Theory for Embedded Systems" was written for you.
See details at http://www.wescottdesign.com/actfes/actfes.html

On 07/09/2011 08:47 PM, Vladimir Vassilevsky wrote:
>
>
> Randy Yates wrote:
>
>> I may very well be asking this question prematurely, and I may
>> actually already know the answer but have forgotten it. In either
>> case, please be kind...
>>
>> In binary FSK demodulation, one normally puts a bandpass filter around
>> each of the mark and space frequencies. How small can this filter be?
>
> RANDY = STUPIDENT.
>
> In AWGN, the best filter is the filter that matches the transmit pulse
> shape. If noise is not white, the filter should match the SNR.
> This is in the first page of any ABC book.

Whoa -- the two faces of Vladimir.

In _really old_ FSK practice (continuous phase not guaranteed, wide 
shift/baud ratio), you would, indeed, have filtered each frequency 
separately, and you would have probably used envelope detection 
techniques to determine mark and space (and your circuit would have had 
6AU6's, 12AK6's, and the like).

In that case, the optimal filter from an _overall_ standpoint would not 
have been matched -- you could have maybe made it, but then no one would 
have been able to afford your demodulator except the military.

-- 

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Do you need to implement control loops in software?
"Applied Control Theory for Embedded Systems" was written for you.
See details at http://www.wescottdesign.com/actfes/actfes.html

Randy Yates wrote:

> I may very well be asking this question prematurely, and I may
> actually already know the answer but have forgotten it. In either
> case, please be kind...
> 
> In binary FSK demodulation, one normally puts a bandpass filter around
> each of the mark and space frequencies. How small can this filter be?

RANDY = STUPIDENT.

In AWGN, the best filter is the filter that matches the transmit pulse 
shape. If noise is not white, the filter should match the SNR.
This is in the first page of any ABC book.

Randy Yates wrote:
> I may very well be asking this question prematurely, and I may
> actually already know the answer but have forgotten it. In either
> case, please be kind...
> 
> In binary FSK demodulation, one normally puts a bandpass filter around
> each of the mark and space frequencies. How small can this filter be?
> 
> My thoughts cycle like this:
> 
>   1) We want the filter to be as small as possible to remove as much
>   noise as possible.
> 
>   2) But if the filter is too narrow compared to the baud rate, the
>   filter output doesn't reach full output before the transmitter
>   switches to the next baud (and thus possibly the other frequency),
>   and so we get worse noise performance.
> 
> This must be covered somewhere in the basics, but I'm missing and/or
> forgetting it. Memory refreshment would be appreciated!

It depends.

If you cut off a part of the original spectrum of a signal, you will get 
ISI as the result. The power of ISI equals to the power of the part of 
the spectrum which was cut off. The over simplified view would be 
consider ISI as noise; hence the optimum bandwidth is when the power of 
noise in the bandwidth PLUS the power of ISI is at minimum.

VLV

I may very well be asking this question prematurely, and I may
actually already know the answer but have forgotten it. In either
case, please be kind...

In binary FSK demodulation, one normally puts a bandpass filter around
each of the mark and space frequencies. How small can this filter be?

My thoughts cycle like this:

   1) We want the filter to be as small as possible to remove as much
   noise as possible.

   2) But if the filter is too narrow compared to the baud rate, the
   filter output doesn't reach full output before the transmitter
   switches to the next baud (and thus possibly the other frequency),
   and so we get worse noise performance.

This must be covered somewhere in the basics, but I'm missing and/or
forgetting it. Memory refreshment would be appreciated!
-- 
Randy Yates                      % "Watching all the days go by...
Digital Signal Labs              %  Who are you and who am I?"
mailto://yates@ieee.org          % 'Mission (A World Record)',
http://www.digitalsignallabs.com % *A New World Record*, ELO