Hi,
  As we know, any non-linearity on matched filter output contians two
spectral lines at +/- symbol rate. I am a little confused about how to use
that for timing compensation. If it is used as a TED for a PLL, will not
the loop filter reject this line? Secondly, coming to the time domain, is
the top of this sinusoid refers to the ideal timing phase?
Thanks.

>Hi,
>  As we know, any non-linearity on matched filter output contians two
>spectral lines at +/- symbol rate. I am a little confused about how to
use
>that for timing compensation. If it is used as a TED for a PLL, will not
>the loop filter reject this line? Secondly, coming to the time domain, is
>the top of this sinusoid refers to the ideal timing phase?
>Thanks.

A classic way to use the band edge effect for symbol timing recovery may be
found in:

Dominique N. Godard, “Passband Timing Recovery in an All-Digital Modem
Receiver,” IEEE Transactions on Communications, Vol. COM-26, No. 5, May
1978

Regards,
Steve

>>Hi,
>>  As we know, any non-linearity on matched filter output contians two
>>spectral lines at +/- symbol rate. I am a little confused about how to
>use
>>that for timing compensation. If it is used as a TED for a PLL, will not
>>the loop filter reject this line? Secondly, coming to the time domain,
is
>>the top of this sinusoid refers to the ideal timing phase?
>>Thanks.
>
>A classic way to use the band edge effect for symbol timing recovery may
be
>found in:
>
>Dominique N. Godard, “Passband Timing Recovery in an All-Digital Modem
>Receiver,” IEEE Transactions on Communications, Vol. COM-26, No. 5, May
>1978
>
>Regards,
>Steve
>


I remember Goddard takes the imaginary part of the conjugate product of the
two bandpass filter outputs and uses that as an error signal for input to
the loop filter. Can there be any feedforward implementation too?

"commsignal" <58672@dsprelated> wrote in message 
news:O...@giganews.com...
> Hi,
>  As we know, any non-linearity on matched filter output contians two
> spectral lines at +/- symbol rate.

"We" know?

> I am a little confused about how to use that for timing compensation

The underlying concept is very simple: at the output of matched filter, 
energy is pulsating with the baud rate.

> If it is used as a TED for a PLL, will not the loop filter reject this 
> line?

No slightest clue, huh ?

> Secondly, coming to the time domain, is
> the top of this sinusoid refers to the ideal timing phase?

Yes. Maximum of energy corresponds to symbol sampling instant.

VLV

"commsignal" <58672@dsprelated> wrote:
 in message news:8...@giganews.com...

> I remember Goddard takes the imaginary part of the conjugate product of 
> the
> two bandpass filter outputs and uses that as an error signal for input to
> the loop filter.

That's minor technical details of implementation.


> Can there be any feedforward implementation too? 

"steveu" <31473@dsprelated> wrote in message 
news:_f...@giganews.com...
> >Hi,
>>  As we know, any non-linearity on matched filter output contians two
>>spectral lines at +/- symbol rate. I am a little confused about how to
> use
>>that for timing compensation. If it is used as a TED for a PLL, will not
>>the loop filter reject this line? Secondly, coming to the time domain, is
>>the top of this sinusoid refers to the ideal timing phase?

>
> A classic way to use the band edge effect for symbol timing recovery may 
> be
> found in:
>
> Dominique N. Godard, “Passband Timing Recovery in an All-Digital Modem
> Receiver,” IEEE Transactions on Communications, Vol. COM-26, No. 5, May
> 1978

There is a lot of smoke and mirrors although underlying concepts are very 
simple: taking the derivative of signal energy either in time or in 
frequency domain. I think OP was inquiring about time domain method.

VLV

>>>Hi,
>>>  As we know, any non-linearity on matched filter output contians two
>>>spectral lines at +/- symbol rate. I am a little confused about how to
>>use
>>>that for timing compensation. If it is used as a TED for a PLL, will
not
>>>the loop filter reject this line? Secondly, coming to the time domain,
>is
>>>the top of this sinusoid refers to the ideal timing phase?
>>>Thanks.
>>
>>A classic way to use the band edge effect for symbol timing recovery may
>be
>>found in:
>>
>>Dominique N. Godard, “Passband Timing Recovery in an All-Digital Modem
>>Receiver,” IEEE Transactions on Communications, Vol. COM-26, No. 5,
May
>>1978
>>
>>Regards,
>>Steve
>>
>
>
>I remember Goddard takes the imaginary part of the conjugate product of
the
>two bandpass filter outputs and uses that as an error signal for input to
>the loop filter. Can there be any feedforward implementation too?

What exactly do you mean by feedforward in this context? Are you looking
for a technique which directly estimates the sampling phase error and just
feeds that to a Farrow filter to retime the signal? Would that have any
advantage over the loop approach? You'd need to heavily damp the timing
error estimate to get good results in poor SNR conditions, so you might as
well let a loop provide your damping.

Regards,
Steve

>
>"steveu" <31473@dsprelated> wrote in message 
>news:_f...@giganews.com...
>> >Hi,
>>>  As we know, any non-linearity on matched filter output contians two
>>>spectral lines at +/- symbol rate. I am a little confused about how to
>> use
>>>that for timing compensation. If it is used as a TED for a PLL, will
not
>>>the loop filter reject this line? Secondly, coming to the time domain,
is
>>>the top of this sinusoid refers to the ideal timing phase?
>
>>
>> A classic way to use the band edge effect for symbol timing recovery may

>> be
>> found in:
>>
>> Dominique N. Godard, “Passband Timing Recovery in an All-Digital
Modem
>> Receiver,” IEEE Transactions on Communications, Vol. COM-26, No. 5,
May
>> 1978
>
>There is a lot of smoke and mirrors although underlying concepts are very

>simple: taking the derivative of signal energy either in time or in 
>frequency domain. I think OP was inquiring about time domain method.
>
>VLV
>
>
>
>

I guess differentiating that in frequency domain leads to a frequency
matched filter, and the power difference between the two bandedge filters
can be used as a frequency error detector. However, I am still confused
about that twice symbol rate spectral line. Essentially it's a sinusoid in
noise. How can that be used as a timing error detector?

>>>>Hi,
>>>>  As we know, any non-linearity on matched filter output contians two
>>>>spectral lines at +/- symbol rate. I am a little confused about how to
>>>use
>>>>that for timing compensation. If it is used as a TED for a PLL, will
>not
>>>>the loop filter reject this line? Secondly, coming to the time domain,
>>is
>>>>the top of this sinusoid refers to the ideal timing phase?
>>>>Thanks.
>>>
>>>A classic way to use the band edge effect for symbol timing recovery
may
>>be
>>>found in:
>>>
>>>Dominique N. Godard, “Passband Timing Recovery in an All-Digital
Modem
>>>Receiver,” IEEE Transactions on Communications, Vol. COM-26, No. 5,
>May
>>>1978
>>>
>>>Regards,
>>>Steve
>>>
>>
>>
>>I remember Goddard takes the imaginary part of the conjugate product of
>the
>>two bandpass filter outputs and uses that as an error signal for input
to
>>the loop filter. Can there be any feedforward implementation too?
>
>What exactly do you mean by feedforward in this context? Are you looking
>for a technique which directly estimates the sampling phase error and
just
>feeds that to a Farrow filter to retime the signal? Would that have any
>advantage over the loop approach? You'd need to heavily damp the timing
>error estimate to get good results in poor SNR conditions, so you might
as
>well let a loop provide your damping.
>
>Regards,
>Steve
>
>

You are right. But the confusion still remains. How can that spectral line
be used in the timing error detector of the loop?

On Tuesday, October 16, 2012 1:40:03 AM UTC-4, commsignal wrote:
> >
> 
> >"steveu" <31473@dsprelated> wrote in message 
> 
> >news:_f...@giganews.com...
> 
> >> >Hi,
> 
> >>>  As we know, any non-linearity on matched filter output contians two
> 
> >>>spectral lines at +/- symbol rate. I am a little confused about how to
> 
> >> use
> 
> >>>that for timing compensation. If it is used as a TED for a PLL, will
> 
> not
> 
> >>>the loop filter reject this line? Secondly, coming to the time domain,
> 
> is
> 
> >>>the top of this sinusoid refers to the ideal timing phase?
> 
> >
> 
> >>
> 
> >> A classic way to use the band edge effect for symbol timing recovery may
> 
> 
> 
> >> be
> 
> >> found in:
> 
> >>
> 
> >> Dominique N. Godard, “Passband Timing Recovery in an All-Digital
> 
> Modem
> 
> >> Receiver,” IEEE Transactions on Communications, Vol. COM-26, No. 5,
> 
> May
> 
> >> 1978
> 
> >
> 
> >There is a lot of smoke and mirrors although underlying concepts are very
> 
> 
> 
> >simple: taking the derivative of signal energy either in time or in 
> 
> >frequency domain. I think OP was inquiring about time domain method.
> 
> >
> 
> >VLV
> 
> >
> 
> >
> 
> >
> 
> >
> 
> 
> 
> I guess differentiating that in frequency domain leads to a frequency
> 
> matched filter, and the power difference between the two bandedge filters
> 
> can be used as a frequency error detector. However, I am still confused
> 
> about that twice symbol rate spectral line. Essentially it's a sinusoid in
> 
> noise. How can that be used as a timing error detector?

Feed the nonlinearity result through a narrow bandpass filter, then interpolate the matched
filter outputs at alternate zero crossings of the BPF output using a cubic interpolator. 

John