On Tue, Mar 30, 2010 at 9:57 PM, Olin Lathrop <olin_piclist@embedinc.com> w=
rote:
> The recent discussion on DTMF decoding made me think about a project I had
> been considering for a while. =A0I did some simulations to verify my math=
 was
> right and look at filter time constants and the like.
>
> You can see the result at http://www.embedinc.com/temp/dtmf.gif. =A0The i=
nput
> is bursts of the DTMF tones 697, 770, and 853 Hz. =A0Each burst and the g=
aps
> between the bursts are 50mS long, which is the time you are supposed to be
> able to detect a DTMF tone within. =A0The blue trace is the square of the
> magnitude of the detected tone. =A0The algorithm was set up to detect 770=
Hz,
> which is the frequency of the center burst.
>
> As you can see, this worked very well. =A0The blue trace would have event=
ually
> reached 1.0 if the center burst persisted. =A0But clearly the center burs=
t was
> well detected and the other two were not, even though the frequencies
> between adjacent tones differ by less than 11%.
>
> Below is the core code of the simulation. =A0The complete code contains t=
oo
> many distractions to show here, like CSV file writing and other logistics.
>
> =A0for sampn :=3D 0 to nsamp do begin =A0 =A0 {once for each input sample}
> =A0 =A0t :=3D sampn * sampdt; =A0 =A0 =A0 =A0 =A0 =A0 =A0 {make time of t=
his sample}
> =A0 =A0samp :=3D getsamp (t); =A0 =A0 =A0 =A0 =A0 =A0 =A0 {get input samp=
le}
> =A0 =A0r :=3D t * freq; =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 {make ref=
erence frequency phase}
> =A0 =A0ii :=3D trunc(r);
> =A0 =A0r :=3D r - ii;
> =A0 =A0r :=3D r * pi2;
> =A0 =A0prods :=3D samp * sin(r); =A0 =A0 =A0 =A0 =A0 =A0{mix by ref freq =
sine and cosine}
> =A0 =A0prodc :=3D samp * cos(r);
> =A0 =A0filter (filts, prods); =A0 =A0 =A0 =A0 =A0 =A0 {low pass filter mi=
xer results}
> =A0 =A0filter (filtc, prodc);
> =A0 =A0magsq :=3D sqr(filts.val) + sqr(filtc.val); {make square of magnit=
ude}
> =A0 =A0magsq :=3D magsq * 4.0; =A0 =A0 =A0 =A0 =A0 =A0 =A0{normalize}
>
> Obviously you wouldn't normalize the result in the PIC, you'd adjust the
> detection threshold instead. =A0The FILTER subroutine does a two pole low=
 pass
> filter. =A0Each pole has a filter fraction of 1/128, which means it can be
> realized inside a PIC with a right shift of 7 bits. =A0The input was samp=
led
> and processed every 100uS.
>
>
> ********************************************************************
> Embed Inc, Littleton Massachusetts, http://www.embedinc.com/products
> (978) 742-9014. =A0Gold level PIC consultants since 2000.
> --
> http://www.piclist.com PIC/SX FAQ & list archive
> View/change your membership options at
> http://mailman.mit.edu/mailman/listinfo/piclist
>


Very nice! Some questions for the details:
- filter() looks something like 1/128 * getsamp(t) + 1/128 * getsamp(t
- sampdt) ? Or is it an IIR filter?
- What is the trunc() bit? A peculiarity of the arcane language (I'm
guessing fortran) you're using?
Thanks,
- Marcel

-- =

http://www.piclist.com PIC/SX FAQ & list archive
View/change your membership options at
http://mailman.mit.edu/mailman/listinfo/piclist