[Dspforum] demodulator performance/frame synchronization

[Joseph Haber]

You could also just calculate the FFT output for the two bins (from the DFT
formula), you could design a FIR filter, or you could design an IIR. Lot's
of options. With all the solutions, make sure that the time response of the
filter is short enough so that you can distinguish each of the symbols
(i.e., a very long FIR filter would "smear" a lot of symbols together).

Joseph

On Sun, Nov 16, 2008 at 5:23 PM, Ryan Merton <lucidsmog at gmail.com> wrote:

> Michael,
>
> I did not use EDMA for this lab.  I thought about it, but as Joseph
> said, the added hassle of dealing with baud boundaries across EDMA
> frames would have been annoying.  As it is I have enough processing
> time to run all of the code for each sample using polling (without any
> optimization yet).
>
> You just need to come up with a good way to make a decision on where a
> latch/release occurred and where the baud boundary is and then make a
> decision every 8 samples after that for however many bits there are.
>
> Also, I found that the DSPLIB functions weren't very useful as the FFT
> stuff requires at least 32 points and complex samples.  I just wrote
> the literal implementation for the two bins that I cared about from
> the description in 'help fft' in Matlab.  I'm certain it's painfully
> slow to calculate even just those two with unoptimized C, but like I
> said... it's fitting in the time constraints.
>
> -Ryan
>
> On Sun, Nov 16, 2008 at 4:46 PM, Joseph Haber <josephhaber at gmail.com>
> wrote:
> > Mike,
> >
> > Don't make any presumptions that a start falls on an "8" boundry -- in
> fact,
> > that concept doesn't exist until a start notification is found.
> >
> > My first instinct is that EDMA makes this more complicated than you want
> --
> > you would have to be careful about processing across two sequential
> > "blocks", which seems like a pain to me.
> >
> > Joseph
> >
> > On Sun, Nov 16, 2008 at 4:39 PM, michael norris <mnorris at gmail.com>
> wrote:
> >>
> >> Instructors/students,
> >>
> >> Can we assume that a frame begins on an 8 (f_sample/f_symbol) sample
> >> boundary? In the general case, I don't think we can. If we can't, then
> it
> >> seems we need to compute an 8-point FFT per sample in order to synch the
> >> frame.
> >>
> >> Once we've synched on the frame, we could then compute the FFT once per
> 8
> >> samples. Then we would benefit from interleaving the sampling and
> >> computation with EDMA. But if our computation time must already be under
> one
> >> sampling time, then what benefit is there to implementing EDMA with this
> >> lab?
> >>
> >> (Maybe it prevent us from missing data during the printf() call. But do
> we
> >> care about this?)
> >>
> >> Could someone add a little more about detecting the frame boundary?
> >> Assuming we are sampling once-at-a-time to do the synchronization, where
> in
> >> the start bit's 8 samples do we fall? Are we one past the half-way
> sample,
> >> i.e. sample 5/8?
> >>
> >> We should have plenty of time to do our 8-point FFT with this board. I'm
> >> just curious where EDMA could benefit us here.
> >>
> >> Thanks,
> >> -Michael
> >>
> >> _______________________________________________
> >> Dspforum mailing list
> >> Dspforum at echelonembedded.com
> >>
> http://echelonembedded.com/mailman/listinfo/dspforum_echelonembedded.com
> >>
> >
> >
> > _______________________________________________
> > Dspforum mailing list
> > Dspforum at echelonembedded.com
> > http://echelonembedded.com/mailman/listinfo/dspforum_echelonembedded.com
> >
> >
>
> _______________________________________________
> Dspforum mailing list
> Dspforum at echelonembedded.com
> http://echelonembedded.com/mailman/listinfo/dspforum_echelonembedded.com
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://echelonembedded.com/pipermail/dspforum_echelonembedded.com/attachments/20081116/88557328/attachment.html>