Doubt

Dear All

I have some doubts regarding synchronisation. Please tell me if my
understanding is right or not.

Currently I am concerned abt carrier and symbol synchronisation.

Now by symbol synchronisation, I understand that we need to sample the o/p
of matched filer at right instant. NOw due to delay in channel, symbols (or
samples per symbol) can arrive before or after the right sampling
instatnt.

And by carrier phase I mean if my modulated signal 

s=real(data).*cos(2*pi*fc*t)-imag.*sin(2*pi*fc*t)

then I need to estimate the phase of the carrier, in the above case my
phase is 0.

So we correct the timing first, then we give this timing corrected symbols
to equaliser and correct the phase adaptively using costas loop.

Is this correct?

Thanks a lot.

Regards,

Chintan

On Mon, 04 Aug 2008 10:58:27 -0500, "cpshah99"
<cpshah99@rediffmail.com> wrote:

>Dear All
>
>I have some doubts regarding synchronisation. Please tell me if my
>understanding is right or not.
>
>Currently I am concerned abt carrier and symbol synchronisation.
>
>Now by symbol synchronisation, I understand that we need to sample the o/p
>of matched filer at right instant. NOw due to delay in channel, symbols (or
>samples per symbol) can arrive before or after the right sampling
>instatnt.
>
>And by carrier phase I mean if my modulated signal 
>
>s=real(data).*cos(2*pi*fc*t)-imag.*sin(2*pi*fc*t)
>
>then I need to estimate the phase of the carrier, in the above case my
>phase is 0.
>
>So we correct the timing first, then we give this timing corrected symbols
>to equaliser and correct the phase adaptively using costas loop.
>
>Is this correct?
>
>Thanks a lot.
>
>Regards,
>
>Chintan

Generally that works, yes.   As always, there are alternative
arrangements, but what you've described is a common basic
architecture.  I think you could find relevant block diagrams in a
number of communications texts.



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

Blog: http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php

>On Mon, 04 Aug 2008 10:58:27 -0500, "cpshah99"
><cpshah99@rediffmail.com> wrote:
>
>>Dear All
>>
>>I have some doubts regarding synchronisation. Please tell me if my
>>understanding is right or not.
>>
>>Currently I am concerned abt carrier and symbol synchronisation.
>>
>>Now by symbol synchronisation, I understand that we need to sample the
o/p
>>of matched filer at right instant. NOw due to delay in channel, symbols
(or
>>samples per symbol) can arrive before or after the right sampling
>>instatnt.
>>
>>And by carrier phase I mean if my modulated signal 
>>
>>s=real(data).*cos(2*pi*fc*t)-imag.*sin(2*pi*fc*t)
>>
>>then I need to estimate the phase of the carrier, in the above case my
>>phase is 0.
>>
>>So we correct the timing first, then we give this timing corrected
symbols
>>to equaliser and correct the phase adaptively using costas loop.
>>
>>Is this correct?
>>
>>Thanks a lot.
>>
>>Regards,
>>
>>Chintan
>
>Generally that works, yes.   As always, there are alternative
>arrangements, but what you've described is a common basic
>architecture.  I think you could find relevant block diagrams in a
>number of communications texts.
>
>
>
>Eric Jacobsen
>Minister of Algorithms
>Abineau Communications
>http://www.ericjacobsen.org
>
>Blog: http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php
>

%%%%%

Hi Eric,

Thanks a lot. To be honest I am not able to find any good treatment of
this topic. I am following Proakis book , but that book itself is
complicated.

I know certain books by Meyr et al and all, but they are not available.
And I didnt find the book by Mengali also helpful.

The costas loop i m implementing is from my class notes.

Another doubt is that if i implement early late gate synchroniser, so the
o/p will be one sample per symbol right?
 and then i can use the o/p of early late gate synchroniser as i/p of my
adaptive DFE.?

and what do some ppl mean by 1/T where T is symbol period? does that mean
one sample per symbol?

Thanks a lot.

Chintan

On Mon, 04 Aug 2008 13:14:20 -0500, "cpshah99"
<cpshah99@rediffmail.com> wrote:

>>On Mon, 04 Aug 2008 10:58:27 -0500, "cpshah99"
>><cpshah99@rediffmail.com> wrote:
>>
>>>Dear All
>>>
>>>I have some doubts regarding synchronisation. Please tell me if my
>>>understanding is right or not.
>>>
>>>Currently I am concerned abt carrier and symbol synchronisation.
>>>
>>>Now by symbol synchronisation, I understand that we need to sample the
>o/p
>>>of matched filer at right instant. NOw due to delay in channel, symbols
>(or
>>>samples per symbol) can arrive before or after the right sampling
>>>instatnt.
>>>
>>>And by carrier phase I mean if my modulated signal 
>>>
>>>s=real(data).*cos(2*pi*fc*t)-imag.*sin(2*pi*fc*t)
>>>
>>>then I need to estimate the phase of the carrier, in the above case my
>>>phase is 0.
>>>
>>>So we correct the timing first, then we give this timing corrected
>symbols
>>>to equaliser and correct the phase adaptively using costas loop.
>>>
>>>Is this correct?
>>>
>>>Thanks a lot.
>>>
>>>Regards,
>>>
>>>Chintan
>>
>>Generally that works, yes.   As always, there are alternative
>>arrangements, but what you've described is a common basic
>>architecture.  I think you could find relevant block diagrams in a
>>number of communications texts.
>>
>>
>>
>>Eric Jacobsen
>>Minister of Algorithms
>>Abineau Communications
>>http://www.ericjacobsen.org
>>
>>Blog: http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php
>>
>
>%%%%%
>
>Hi Eric,
>
>Thanks a lot. To be honest I am not able to find any good treatment of
>this topic. I am following Proakis book , but that book itself is
>complicated.
>
>I know certain books by Meyr et al and all, but they are not available.
>And I didnt find the book by Mengali also helpful.
>
>The costas loop i m implementing is from my class notes.
>
>Another doubt is that if i implement early late gate synchroniser, so the
>o/p will be one sample per symbol right?
> and then i can use the o/p of early late gate synchroniser as i/p of my
>adaptive DFE.?

Yes, you can feed the equalizer with the output of the synchronizer.

>and what do some ppl mean by 1/T where T is symbol period? does that mean
>one sample per symbol?

Yes, generally T is the symbol period.   I usually use Ts to
distinguish it from the sample rate, but I'm unusual in that regard.
;)

A "T-spaced" equalizer is generally understood to be one that uses one
sample per symbol, though.

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

Blog: http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php

>On Mon, 04 Aug 2008 13:14:20 -0500, "cpshah99"
><cpshah99@rediffmail.com> wrote:
>
>>>On Mon, 04 Aug 2008 10:58:27 -0500, "cpshah99"
>>><cpshah99@rediffmail.com> wrote:
>>>
>>>>Dear All
>>>>
>>>>I have some doubts regarding synchronisation. Please tell me if my
>>>>understanding is right or not.
>>>>
>>>>Currently I am concerned abt carrier and symbol synchronisation.
>>>>
>>>>Now by symbol synchronisation, I understand that we need to sample
the
>>o/p
>>>>of matched filer at right instant. NOw due to delay in channel,
symbols
>>(or
>>>>samples per symbol) can arrive before or after the right sampling
>>>>instatnt.
>>>>
>>>>And by carrier phase I mean if my modulated signal 
>>>>
>>>>s=real(data).*cos(2*pi*fc*t)-imag.*sin(2*pi*fc*t)
>>>>
>>>>then I need to estimate the phase of the carrier, in the above case
my
>>>>phase is 0.
>>>>
>>>>So we correct the timing first, then we give this timing corrected
>>symbols
>>>>to equaliser and correct the phase adaptively using costas loop.
>>>>
>>>>Is this correct?
>>>>
>>>>Thanks a lot.
>>>>
>>>>Regards,
>>>>
>>>>Chintan
>>>
>>>Generally that works, yes.   As always, there are alternative
>>>arrangements, but what you've described is a common basic
>>>architecture.  I think you could find relevant block diagrams in a
>>>number of communications texts.
>>>
>>>
>>>
>>>Eric Jacobsen
>>>Minister of Algorithms
>>>Abineau Communications
>>>http://www.ericjacobsen.org
>>>
>>>Blog: http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php
>>>
>>
>>%%%%%
>>
>>Hi Eric,
>>
>>Thanks a lot. To be honest I am not able to find any good treatment of
>>this topic. I am following Proakis book , but that book itself is
>>complicated.
>>
>>I know certain books by Meyr et al and all, but they are not available.
>>And I didnt find the book by Mengali also helpful.
>>
>>The costas loop i m implementing is from my class notes.
>>
>>Another doubt is that if i implement early late gate synchroniser, so
the
>>o/p will be one sample per symbol right?
>> and then i can use the o/p of early late gate synchroniser as i/p of
my
>>adaptive DFE.?
>
>Yes, you can feed the equalizer with the output of the synchronizer.
>
>>and what do some ppl mean by 1/T where T is symbol period? does that
mean
>>one sample per symbol?
>
>Yes, generally T is the symbol period.   I usually use Ts to
>distinguish it from the sample rate, but I'm unusual in that regard.
>;)
>
>A "T-spaced" equalizer is generally understood to be one that uses one
>sample per symbol, though.
>
>Eric Jacobsen
>Minister of Algorithms
>Abineau Communications
>http://www.ericjacobsen.org
>
>Blog: http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php
>
%%%%%

Hi Eric

Thanks a lot.

Then so far whatever I am doing is correct.

Thanks again.

Chintan

On Aug 4, 1:14&#4294967295;pm, "cpshah99" <cpsha...@rediffmail.com> wrote:

> Thanks a lot. To be honest I am not able to find any good treatment of
> this topic. I am following Proakis book , but that book itself is
> complicated.
>
> I know certain books by Meyr et al and all, but they are not available.
> And I didnt find the book by Mengali also helpful.
>

The Mengali book is a pain to read, but it has all the ingredients
there.
The Proakis book's treatment is just terrible.  It is a pain to read,
and is
incomplete.  The Meyr book is also hard to read, but like the Mengali
book it has all the ingredients there.

There are tutorials here and there, but mostly they cover specific
techniques, such as the Muller-Muller synchronizer, or Gardner.
Only the books above compare all of them all at once.

So in a sense it depends on your ambition.  If you only want to get
something working, then read the tutorials.  Most of them don't even
derive the effect of noise.  But if you want a formal way of deriving
them, then the Meyr books and Mengali's book are the best I have
found so far.

Julius

>On Aug 4, 1:14=A0pm, "cpshah99" <cpsha...@rediffmail.com> wrote:
>
>> Thanks a lot. To be honest I am not able to find any good treatment of
>> this topic. I am following Proakis book , but that book itself is
>> complicated.
>>
>> I know certain books by Meyr et al and all, but they are not
available.
>> And I didnt find the book by Mengali also helpful.
>>
>
>The Mengali book is a pain to read, but it has all the ingredients
>there.
>The Proakis book's treatment is just terrible.  It is a pain to read,
>and is
>incomplete.  The Meyr book is also hard to read, but like the Mengali
>book it has all the ingredients there.
>
>There are tutorials here and there, but mostly they cover specific
>techniques, such as the Muller-Muller synchronizer, or Gardner.
>Only the books above compare all of them all at once.
>
>So in a sense it depends on your ambition.  If you only want to get
>something working, then read the tutorials.  Most of them don't even
>derive the effect of noise.  But if you want a formal way of deriving
>them, then the Meyr books and Mengali's book are the best I have
>found so far.
>
>Julius
>
%%%

Hi Julius

Thanks a lot.

To be honest I am aiming a bit high. Please help to clear another doubt.

Because of doppler:
1. signal expands or compress
2. phase of the signal will be changed
3. it creates problem for synchronisation ( symbol and phase both).
Is that true?

Assuming the way I am modeling the doppler is correct, do these algorithms
presented in Meyr and Mengali books help?

And last question: Can you please suggest a starting point?

Thanks a lot.

Chintan

On Aug 5, 8:11 am, "cpshah99" <cpsha...@rediffmail.com> wrote:
>
> Hi Julius
>
> Thanks a lot.
>
> To be honest I am aiming a bit high. Please help to clear another doubt.
>
> Because of doppler:
> 1. signal expands or compress
> 2. phase of the signal will be changed
> 3. it creates problem for synchronisation ( symbol and phase both).
> Is that true?
>
> Assuming the way I am modeling the doppler is correct, do these algorithms
> presented in Meyr and Mengali books help?
>
> And last question: Can you please suggest a starting point?
>
> Thanks a lot.
>
> Chintan

Chintan, you are correct on all three points.  Unfortunately most
formal
references today were written for "streaming" communication, where
one
deals with long bursts instead of short bursts.

Another thing is that they assume what I call "effective baseband
drift" to be small.  In many cases, such as when the symbol rate is
very low, this is not true.  I recommend in this case to have a
dedicated
subsystem to do re-sampling, or sampling rate correction, at sampling
rate higher than symbol rate, possibly at passband.

Think about it this way also:  if Doppler is the limiting factor, then
try
to estimate it directly or indirectly.  See if estimating based on a
packet
preamble or pilot tone/signal is sufficient.  If not, then you will
have to
continually improve your estimate.  This is done via a loop filter
(voila!).

I'm sorry to hear that you are frustrated, but I commend you on being
able to go this far.  Most others will likely have either pretended
that
the problem does not exist, or fail to realize that it exists at all.

Using the notation of Meyr/Moeneclaey, your inner receiver can look
like this:

ADC -> passband resampling -> mixer -> baseband synch -> demodulate.

Where do you work anyway?  I'm curious.

Good luck,
Julius

>On Aug 5, 8:11 am, "cpshah99" <cpsha...@rediffmail.com> wrote:
>>
>> Hi Julius
>>
>> Thanks a lot.
>>
>> To be honest I am aiming a bit high. Please help to clear another
doubt.
>>
>> Because of doppler:
>> 1. signal expands or compress
>> 2. phase of the signal will be changed
>> 3. it creates problem for synchronisation ( symbol and phase both).
>> Is that true?
>>
>> Assuming the way I am modeling the doppler is correct, do these
algorithms
>> presented in Meyr and Mengali books help?
>>
>> And last question: Can you please suggest a starting point?
>>
>> Thanks a lot.
>>
>> Chintan
>
>Chintan, you are correct on all three points.  Unfortunately most
>formal
>references today were written for "streaming" communication, where
>one
>deals with long bursts instead of short bursts.
>
>Another thing is that they assume what I call "effective baseband
>drift" to be small.  In many cases, such as when the symbol rate is
>very low, this is not true.  I recommend in this case to have a
>dedicated
>subsystem to do re-sampling, or sampling rate correction, at sampling
>rate higher than symbol rate, possibly at passband.
>
>Think about it this way also:  if Doppler is the limiting factor, then
>try
>to estimate it directly or indirectly.  See if estimating based on a
>packet
>preamble or pilot tone/signal is sufficient.  If not, then you will
>have to
>continually improve your estimate.  This is done via a loop filter
>(voila!).
>
>I'm sorry to hear that you are frustrated, but I commend you on being
>able to go this far.  Most others will likely have either pretended
>that
>the problem does not exist, or fail to realize that it exists at all.
>
>Using the notation of Meyr/Moeneclaey, your inner receiver can look
>like this:
>
>ADC -> passband resampling -> mixer -> baseband synch -> demodulate.
>
>Where do you work anyway?  I'm curious.
>
>Good luck,
>Julius
>

%%%%%

Hi Julius

I really appreciate for spending time.

I am doing research in underwater acoustics. So far I just worked on AWGN
channel and then turbo equalisation, where I tried to replicate paper by
Tuchler.

And then I got the project on doppler correction. And I struggled to model
that. With the help of my supervisor I modelled it using linear
interpolation i.e. resampling and changing the time scale. I have also
posted solution in this forum.

And from a sea trial experiment that we did, I can say that even a doppler
of 0.2 m/s can creat a lot of pain!!!!

And the doppler correction also worked well. The following paper is very
good.

''A Computationally Efficient Doppler Compensation System for Underwater
Acoustic Communications''. 

But the technique in the above paper can not correct differential doppler
and I would like to design a separate 'symbol synchronise' and 'phase
correction' keeping in mind turbo decoding.

I also found two papers by T H Eggen: Communication over doppler spread
channels: Part 1 and 2

Also his PhD thesis. 

And this topic looks very interesting because it is challenging. So as I
have designed a transmitter and also I have modelled multipath channel with
doppler I would like to design a good receiver.

But I want to go step by step. 

I am working on baseband signals i.e. after down converting them to
baseband. I find it easy to work on baseband.

But the heart of the system should be Turbo decoding.

As u have said, I am going to read Mengali again.

Also there is a good book: 

Simulation of Communication Systems, 2nd ed, by  Michel C Jeruchim et al.

Thanks a lot. It would be great if you could give some basic tutorials.

Regards,

Chintan

On Aug 5, 3:13 pm, "cpshah99" <cpsha...@rediffmail.com> wrote:
> Hi Julius
>
> I really appreciate for spending time.
>
> I am doing research in underwater acoustics. So far I just worked on AWGN
> channel and then turbo equalisation, where I tried to replicate paper by
> Tuchler.
>
> And then I got the project on doppler correction. And I struggled to model
> that. With the help of my supervisor I modelled it using linear
> interpolation i.e. resampling and changing the time scale. I have also
> posted solution in this forum.
>
> And from a sea trial experiment that we did, I can say that even a doppler
> of 0.2 m/s can creat a lot of pain!!!!
>
> And the doppler correction also worked well. The following paper is very
> good.
>
> ''A Computationally Efficient Doppler Compensation System for Underwater
> Acoustic Communications''.
>
> But the technique in the above paper can not correct differential doppler
> and I would like to design a separate 'symbol synchronise' and 'phase
> correction' keeping in mind turbo decoding.
>
> I also found two papers by T H Eggen: Communication over doppler spread
> channels: Part 1 and 2
>
> Also his PhD thesis.
>
> And this topic looks very interesting because it is challenging. So as I
> have designed a transmitter and also I have modelled multipath channel with
> doppler I would like to design a good receiver.
>
> But I want to go step by step.
>
> I am working on baseband signals i.e. after down converting them to
> baseband. I find it easy to work on baseband.
>
> But the heart of the system should be Turbo decoding.
>
> As u have said, I am going to read Mengali again.
>
> Also there is a good book:
>
> Simulation of Communication Systems, 2nd ed, by  Michel C Jeruchim et al.
>
> Thanks a lot. It would be great if you could give some basic tutorials.
>
> Regards,
>
> Chintan

Chintan, let me give you a homework ;-).

Suppose that you have a simple QPSK modem at carrier freq fc,
sampled at sampling rate fs > 2*(fc+R), symbol rate R, and doppler
is \delta.

Suppose that you sample directly at passband.

Now look at your current receiver and answer the following:
1.  At the output of the ADC, what is the dopper rate normalized
    per sample?  What is the normalized frequency offset?
2.  Same thing, but now after your mixer.
3.  Same thing, but now after you decimate to close to R.

I hope that you can at least convince yourself if working in baseband
is a good idea or not.

Good luck!
Julius