narrow band detector for telephones

Hello guys,
I was going through Steves comments about narrow band detector in
dsprelated.com ,i.e.  using auto correlation and finding peaky_ness. The
sources of these tone signal are
1) Echo of dial tone ( which is a amplitude modulated signal , 400 hz
being modulated by 25 hz)
2) a dtmf tone ( may be as high as 0 dbm )
3) may be echo of other tones
4) or a whistle may be from the talker ( I recorded a whistle at around
1225 Hz)

My point is we can perhaps assume the highest frequency tone is around
1633 Hz, now can we devise simplistic auto correlation method so that we
can find a positive to negetive transion ( or vice versa )of this
autocorrelation values . Number of transitions will be highest for 1633 hz
and lowest for 400 hz . Does this number (of transitions) can accurately
tells the presence of narrow band energy ?

Any thought about this?

   
 

>Hello guys,
>I was going through Steves comments about narrow band detector in
>dsprelated.com ,i.e.  using auto correlation and finding peaky_ness. The
>sources of these tone signal are
>1) Echo of dial tone ( which is a amplitude modulated signal , 400 hz
>being modulated by 25 hz)
>2) a dtmf tone ( may be as high as 0 dbm )
>3) may be echo of other tones
>4) or a whistle may be from the talker ( I recorded a whistle at around
>1225 Hz)
>
>My point is we can perhaps assume the highest frequency tone is around
>1633 Hz, now can we devise simplistic auto correlation method so that we
>can find a positive to negetive transion ( or vice versa )of this
>autocorrelation values . Number of transitions will be highest for 1633
hz
>and lowest for 400 hz . Does this number (of transitions) can accurately
>tells the presence of narrow band energy ?
>
>Any thought about this?
>


Tested! Seems to be working ! just that I get a positive to negetive
transition for artificially generated white noise also ! 
>   
> 
>
>
>
>

> Tested! Seems to be working ! just that I get a positive to negetive
> transition for artificially generated white noise also !


Offcourse, unless I misunderstood something, your method just is based
on the zero crossing counting of the autocorrelation function. It does
not seem to check for bandwidth at all...


To check for bandwith I would suggest generating an AR model from the
data and looking at the distance of the poles from the unit circle.
Or, alternatively, generate an SVD decomposition of the
autocorrelation matrix and looking at the condition number of the
singular values (just a thought).