Need I normalize signal for symbol synchronization? : DSP

Hello.
I have a problem with dependence filter parameters of feedback symbol
synchronization system on sygnal energy. Should I normalize signal or
timing error detector output? Or I should to do some other actions to
remove dependence mentioned above? I have fixed-point implementation
system
and have no one-cycle division instruction at my disposal for
normaliation.
Can AGC help me?
Thanks.
Pavel Schukin

On Feb 7, 4:20 am, "Pavel" <barra...@mail.ru> wrote:
> Hello.
> I have a problem with dependence filter parameters of feedback symbol
> synchronization system on sygnal energy. Should I normalize signal or
> timing error detector output? Or I should to do some other actions to
> remove dependence mentioned above? I have fixed-point implementation
> system
> and have no one-cycle division instruction at my disposal for
> normaliation.
> Can AGC help me?
> Thanks.
> Pavel Schukin

Normalizing either the signal or the TED output will work, assuming
the
dynamic range is good enough. That is for you to figure out ;-).

If the signal amplitude doesn't change much, you can try a one-time
estimate of it in the beginning of the packet if you have a preamble
sequence or something, by doing correlation. Otherwise you need
some sort of gain control with feedback. So it depends on what your
modulation and packet format is.

For example, in one system I have a preamble sequence at the start
of each packet. Detection of the packet is done by computing the
cross-correlation with this preamble sequence in decent resolution.
Then I can use the magnitude of the peak to estimate the signal
amplitude level quite well, since I'm using the entire preamble
sequence
to do this.

As far as division goes, you can try using a lookup table to speed
up the computation of what scaling factor to use. Then you need one
multiply (with the scaling factor) per output of TED, or per input to
the TED
from the signal. Of course, you have to do your own work in designing
the
range and resolution needed in your lookup table.

Julius

Pavel wrote:

> I have a problem with dependence filter parameters of feedback symbol
> synchronization system on sygnal energy. Should I normalize signal or
> timing error detector output?

Certainly. The timing detector is a kind of phase detector, so its
output should be governed by phase, not the amplitude.

> Or I should to do some other actions to
> remove dependence mentioned above? I have fixed-point implementation
> system
> and have no one-cycle division instruction at my disposal for
> normaliation.

"Premature optimization is the root of many evils" (c) Knuth

> Can AGC help me?

Yes, the AGC could be a solution.


Vladimir Vassilevsky
DSP and Mixed Signal Design Consultant
http://www.abvolt.com

On Thu, 07 Feb 2008 09:17:47 -0600, Vladimir Vassilevsky
<antispam_bogus@hotmail.com> wrote:

>
>
>Pavel wrote:
>
>> I have a problem with dependence filter parameters of feedback symbol
>> synchronization system on sygnal energy. Should I normalize signal or
>> timing error detector output?
>
>Certainly. The timing detector is a kind of phase detector, so its
>output should be governed by phase, not the amplitude.

With a caveat that, depending on the signal and TED, the gain of the
detector can change with the magnitude of the signal. Many TEDs that
I'm familiar with do change gain if the signal amplitude changes. If
the signal magnitude isn't controlled in those cases then the gain
change of the TED will change the characteristics of the symbol
recovery loop if it's the typical second-order (or whatever) feedback
control loop.

It's usually not a catastrophic change, i.e., it's not that tough to
design a loop that'll work acceptably well over a range of signal
amplitudes. In some systems the symbol timing will acquire before
the AGC, since AGC loops often have long time constants to filter out
noise and other perturbations, so it may not be that big of a deal.

>> Or I should to do some other actions to
>> remove dependence mentioned above? I have fixed-point implementation
>> system
>> and have no one-cycle division instruction at my disposal for
>> normaliation.
>
>"Premature optimization is the root of many evils" (c) Knuth
>
>> Can AGC help me?
>
>Yes, the AGC could be a solution.

It should help, at least.

Eric Jacobsen
Minister of Algorithms
Abineau Communications
http://www.ericjacobsen.org

On Feb 8, 2:46 am, "Pavel.Schu...@gmail.com" <Pavel.Schu...@gmail.com>
wrote:
> On 7 ÆÅ×, 23:20, Eric Jacobsen <eric.jacob...@ieee.org> wrote:
>
>
>
> > On Thu, 07 Feb 2008 09:17:47 -0600, Vladimir Vassilevsky
>
> > <antispam_bo...@hotmail.com> wrote:
>
> > >Pavel wrote:
>
> > >> I have ša problem with dependence filter parameters of feedback symbol
> > >> synchronization system on sygnal energy. Should I normalize signal or
> > >> timing error detector output?
>
> > >Certainly. The timing detector is a kind of phase detector, so its
> > >output should be governed by phase, not the amplitude.
>
> > With a caveat that, depending on the signal and TED, the gain of the
> > detector can change with the magnitude of the signal. šMany TEDs that
> > I'm familiar with do change gain if the signal amplitude changes. šIf
> > the signal magnitude isn't controlled in those cases then the gain
> > change of the TED will change the characteristics of the symbol
> > recovery loop if it's the typical second-order (or whatever) feedback
> > control loop.
>
>  What is amplitude control? Should amplitude be -1 ... 1, or i need to
> develop loop filter parameters for certain amplitude (0.7 for example)
> and keep it constant for all signals for wich i use that loop filter?> It's usually not a catastrophic change, i.e., it's not that tough to
> > design a loop that'll work acceptably well over a range of signal
> > amplitudes. š In some systems the symbol timing will acquire before
> > the AGC, since AGC loops often have long time constants to filter out
> > noise and other perturbations, so it may not be that big of a deal.
>
>   Does it mean that AGC is to be performed digitally?
>
>
>

The AGC can be performed digitally within the dynamic range of the A/D
converter, but you'll need to include some ****og attenuation control
to move outside that window.

Incidentally, the Maximum Likelihood timing estimator includes an SNR
scale factor in it that effectively decreases the loop bandwidth for
noisier signals. Most implementations don't include that.

John