Fractionally Spaced Equaliser

HI All

I have some conceptual doubt regarding FSE.

Lets say, I am using rectangular pulse at TX and I have 12 samples per
symbol. Now, at receiver I use rectangular pulse as matched filter. So I
will get triangular shape o/p. And I will take o/p at the peak, i.e.
sampling at symbol rate.

My que is: how I will take o/p of matched filter for FSE equaliser. If I
want to take 2 samples per symbol, where should I sample the MF o/p?

In the proakis comms book, it says that:

matched filter followed by linear equaliser (at symbol rate) is equivalent
to have FSE and it will take care of timing phase error.

I am just not clear: where to take samples for this FSE?

I would be very thankful for any help in this regard.

Your opinion matters a lot.

Thanks

Chintan

cpshah99 wrote:

> HI All
> 
> I have some conceptual doubt regarding FSE.
> 
> Lets say, I am using rectangular pulse at TX and I have 12 samples per
> symbol. Now, at receiver I use rectangular pulse as matched filter. So I
> will get triangular shape o/p. And I will take o/p at the peak, i.e.
> sampling at symbol rate.
> 
> My que is: how I will take o/p of matched filter for FSE equaliser. If I
> want to take 2 samples per symbol, where should I sample the MF o/p?

It doesn't matter. You can sample anywhere.

> 
> In the proakis comms book, it says that:
> 
> matched filter followed by linear equaliser (at symbol rate) is equivalent
> to have FSE and it will take care of timing phase error.
> 
> I am just not clear: where to take samples for this FSE?

Anywhere. No information is gained or lost depending on the sampling 
instant.

> I would be very thankful for any help in this regard.
> Your opinion matters a lot.

Your pompous phrases have no meaning whatsoever.

> Thanks

Go to hell.


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

Hey Vladimir

Thanks a lot.

Just help me understand this last thing.

Take a rectangular pulse. Convolve that pulse with itself. So the duration
of the resulting triangular pulse is *twice* of the rectangular pulse.

eg: if rectangular pulse is 12 samples long then triangular pulse will be
24 samples long.

So it does not matter when I need two samples per symbol:

If I take samples at 1 and 12 or samples at 6 and 17.

And another que: do I really need matched filter or not?

page 633 proakis comms edition 4. 

>> Thanks
>
>Go to hell.

Believe me, I feel like I am already there.

Thanks again.

Chintan

On Apr 15, 3:30&#4294967295;pm, Vladimir Vassilevsky <antispam_bo...@hotmail.com>
wrote:
> cpshah99 wrote:
> > HI All
>
> > I have some conceptual doubt regarding FSE.
>
> > Lets say, I am using rectangular pulse at TX and I have 12 samples per
> > symbol. Now, at receiver I use rectangular pulse as matched filter. So I
> > will get triangular shape o/p. And I will take o/p at the peak, i.e.
> > sampling at symbol rate.
>
> > My que is: how I will take o/p of matched filter for FSE equaliser. If I
> > want to take 2 samples per symbol, where should I sample the MF o/p?
>
> It doesn't matter. You can sample anywhere.
>
>
>
> > In the proakis comms book, it says that:
>
> > matched filter followed by linear equaliser (at symbol rate) is equivalent
> > to have FSE and it will take care of timing phase error.
>
> > I am just not clear: where to take samples for this FSE?
>
> Anywhere. No information is gained or lost depending on the sampling
> instant.
>
> > I would be very thankful for any help in this regard.
> > Your opinion matters a lot.
>
> Your pompous phrases have no meaning whatsoever.
>
> > Thanks
>
> Go to hell.
>

Vlad,

what the fuck is that all about?!?!  what pompous phrase are you
referring to?

are you practicing your Don Rickles routine?

r b-j

cpshah99 wrote:
> Hey Vladimir
> 
> Thanks a lot.
> 
> Just help me understand this last thing.
> 
> Take a rectangular pulse. Convolve that pulse with itself. So the duration
> of the resulting triangular pulse is *twice* of the rectangular pulse.
> 
> eg: if rectangular pulse is 12 samples long then triangular pulse will be
> 24 samples long.

Matched filter is defined in the continuous time. Think about t, not 
samples.

> So it does not matter when I need two samples per symbol:
> If I take samples at 1 and 12 or samples at 6 and 17.

You can recover the original signal regardless of the sampling instants 
after the matched filter.

> And another que: do I really need matched filter or not?

Do I really need a dog or not?


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

robert bristow-johnson wrote:


>>>I would be very thankful for any help in this regard.
>>>Your opinion matters a lot.
>>
>>Your pompous phrases have no meaning whatsoever.
>>
>>
> 
> Vlad,
> 
> what the fuck is that all about?!?!  what pompous phrase are you
> referring to?

I don't like the meaningless turgid phrases like "matters a lot", 
"highly appreciated", "very thankful" and such. Your majesty's gratitude 
is too much for my unpretentious person; just $100 would be enough.

> 
> are you practicing your Don Rickles routine?

What Don Rickles?

VLV

Hey

Thanks again for spending time on my problem.

but the last few lines on page 633 proakis comms edition 4 are nagging
me.

I m going mad.

according to that what it means is, just receive the signal, take 2
samples per symbol and pass it to FSE. do not do matched filter. I am still
talking about baseband. If we work at passband, we will need lowpass filter
to remove image freq, which MF can remove.

And what abt continous time abt MF? What I said before is not correct? We
work with discrete time isnt it?

I am sorry but I am going mad. I badly need some help.

Thanks again

Chintan

cpshah99 wrote:

> according to that what it means is, just receive the signal, take 2
> samples per symbol and pass it to FSE.

This describes the ideal noiseless case.
If the signal is sampled, the noise is aliased.
Hence you need a matched filter before.

> do not do matched filter. I am still
> talking about baseband. If we work at passband, we will need lowpass filter
> to remove image freq, which MF can remove.

You got all things mixed up. Stick with the basics.

> And what abt continous time abt MF? What I said before is not correct? We
> work with discrete time isnt it?
> 
> I am sorry but I am going mad. I badly need some help.

Sure. I do homework for the usual consulting rate.


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

Ok. So I stic to basic.

Just last thing. Please tell me if my basic r correct or not. 

Receive the signal(multipath,noise), downconvert, pass it thru matched
filter, take 2 samples per symbol(anywhere within that symbol duration) and
give them to FSE.

Thanks again. Really appreciated. (I know you hate these words)

Chintan

On Apr 15, 5:55&#4294967295;pm, Vladimir Vassilevsky <antispam_bo...@hotmail.com>
wrote:
> robert bristow-johnson wrote:
> >>>I would be very thankful for any help in this regard.
> >>>Your opinion matters a lot.
>
> >>Your pompous phrases have no meaning whatsoever.
>
> > Vlad,
>
> > what the fuck is that all about?!?! &#4294967295;what pompous phrase are you
> > referring to?
>
> I don't like the meaningless turgid phrases like "matters a lot",
> "highly appreciated", "very thankful" and such. Your majesty's gratitude
> is too much for my unpretentious person; just $100 would be enough.
>
>
>
> > are you practicing your Don Rickles routine?
>
> What Don Rickles?
>

an American comedian from the 60s, 70s, and 80s.  i think the guy is
still alive.  he appeared on the Tonight Show before the Jay Leno era
(the Johnny Carson era).  his schtick was insulting people and groups
of people (sometimes ethnic groups), sometimes directly to their
faces.  sometimes it was outrageously funny.  particularly when it was
"well deserved" but only in a stereotypical sense.

other folks have picked up on it.  most recently might be South Park
(in which i, myself, am stereotypically ridiculed as "Aging Hippie
Liberal Douche").

r b-j