gardner TED -- loop filter design

Hi~
Here I use gardner TED in 16QAM receiver.
But I got a problem on the design on loop filter.

Is there some guide on the design? I found some references, but not
worked.

Carlin~

On Apr 22, 4:16&#4294967295;am, "Carlin" <myqs...@126.com> wrote:
> Hi~
> Here I use gardner TED in 16QAM receiver.
> But I got a problem on the design on loop filter.
>
> Is there some guide on the design? I found some references, but not
> worked.
>
> Carlin~

Mengali, d'Andrea, "Synchronization Techniques for Digital Receivers."

Hey Julius

Sorry to jump in the discussion.

That book by Meyr is pretty good but I just have one doubt.

My receiver works with the synchronization algorithm I am using. I am
getting the BER results for uncoded QPSK which is very close to that of
perfect synchronization. But when I use channel coding, why the BER is
degraded when compared with BER for coded QPSK with perfect
synchronisation.

I can think of one reason is that Linear Interpolation introduces some
sort of distortion.

Any inputs from your side?

Thanks

Chintan

On Apr 22, 8:49&#4294967295;am, "cpshah99" <cpsha...@rediffmail.com> wrote:
> Hey Julius
>
> Sorry to jump in the discussion.
>
> That book by Meyr is pretty good but I just have one doubt.

Actually, I recommended a book by Mengali and d'Andrea ....

> My receiver works with the synchronization algorithm I am using. I am
> getting the BER results for uncoded QPSK which is very close to that of
> perfect synchronization. But when I use channel coding, why the BER is
> degraded when compared with BER for coded QPSK with perfect
> synchronisation.
>
> I can think of one reason is that Linear Interpolation introduces some
> sort of distortion.

Yes, of course this is the case, even in floating points.

You need to test the SNR degradation from your linear interpolator.
It is always a function of {oversampling rate, interpolator order,
fractional delay, input SNR, bit precision}.  Typically the last two
are ignored, since if you design it properly they should be very
small.

Try plotting the MSE for a sinusoid with different relative frequency
normalized to your sampling rate, averaged over random phase but
same fractional delay.  Now change the fractional delay and try it
again.  Then try different frequencies.

I don't have a reference handy, but I'm sure this is already
tabulated somewhere.

Julius

Hey Julius

Thanks for replying.

I was thinking, initialy I could not understand anything about timing (rt
now also I dont know everything), but after I realised how these book put
this problem in mathematical form, it was easy. Most of the books assume
that after the receive filter they have got raised cosine pulse, which is
not the case in the problem that I am dealing. I have delay, multipath,
time scaling and noise.

But it is more fascinating when the problem is clear.

Thanks again.

Chintan

On Apr 22, 1:50&#4294967295;pm, "cpshah99" <cpsha...@rediffmail.com> wrote:
> Hey Julius
>
> Thanks for replying.
>
> I was thinking, initialy I could not understand anything about timing (rt
> now also I dont know everything), but after I realised how these book put
> this problem in mathematical form, it was easy. Most of the books assume
> that after the receive filter they have got raised cosine pulse, which is
> not the case in the problem that I am dealing. I have delay, multipath,
> time scaling and noise.
>
> But it is more fascinating when the problem is clear.

There you go!  That's the spirit!

On Apr 23, 1:50&#4294967295;am, "cpshah99" <cpsha...@rediffmail.com> wrote:
> Hey Julius
>
> Thanks for replying.
>
> I was thinking, initialy I could not understand anything about timing (rt
> now also I dont know everything), but after I realised how these book put
> this problem in mathematical form, it was easy. Most of the books assume
> that after the receive filter they have got raised cosine pulse, which is
> not the case in the problem that I am dealing. I have delay, multipath,
> time scaling and noise.
>
> But it is more fascinating when the problem is clear.
>
> Thanks again.
>
> Chintan

The raised cosine pulse filtering is largely irrelvant to this
problem. Gardner is normally used before the equalizer, so depending
on the nature of your channel it is trying to sync to a horribly
distorted version of the signal. So, what you have is the normal case.

The problem with comms books is they focus so much on the clean cases,
or taking the possible impairments one by one, that trying to handle
the real world can be a nasty shock after reading them. :-)

Steve

>On Apr 22, 4:16=A0am, "Carlin" <myqs...@126.com> wrote:
>> Hi~
>> Here I use gardner TED in 16QAM receiver.
>> But I got a problem on the design on loop filter.
>>
>> Is there some guide on the design? I found some references, but not
>> worked.
>>
>> Carlin~
>
>Mengali, d'Andrea, "Synchronization Techniques for Digital Receivers."
>

Thank you~ 

>On Apr 22, 4:16=A0am, "Carlin" <myqs...@126.com> wrote:
>> Hi~
>> Here I use gardner TED in 16QAM receiver.
>> But I got a problem on the design on loop filter.
>>
>> Is there some guide on the design? I found some references, but not
>> worked.
>>
>> Carlin~
>
>Mengali, d'Andrea, "Synchronization Techniques for Digital Receivers."
>

Thank you~ 

Hi Steve

>The raised cosine pulse filtering is largely irrelvant to this
>problem. Gardner is normally used before the equalizer, so depending
>on the nature of your channel it is trying to sync to a horribly
>distorted version of the signal. So, what you have is the normal case.
>

I have not worked on Gardner timing recovery yet, but this is conventional
that in receiver first comes synchronization and then equalizer and so on.

But I have had a look on many books and theses that they say received
signal  y=\sum I_n g(t-nT-\tau) where g(t) is raised cosine. So basically
they assume as you have said mostly one impairement.

Now, in my underwater acoustic case, I have additional term, time scaling
on top of all these. So far as I know, it is not normal, because in radio
comms even if you are travelling at 160km/hr, the scalling factor
a=v/c=10^-7 so basically u can ignore the time scaling.

I might be wrong in claiming that the problem that I am dealing is not
normal.

>
trying to handle
>the real world can be a nasty shock after reading them. :-)
>

Tell me about it. I had almost gone mad :-) and I got this clarified when
I was reading one spy novel.

Chintan