Theoretical and practical consequences of groupdelay variation in DMT.

Hi all,

I am trying to figure out the consequences of groupdelay variation in a DMT 
system. When some frequencies of the pasband are expossed to a groupdelay 
variation of maybe 50us, will these frequencies get delayed: 50us/Ts 
samples, and thereby exceed the timedomain boundry of the original DMT 
symbol (one N point FFT) and cross over into another symbol?

Are my thoughts on this matter all wrong ?

Thanks in advance

--
Morten M. J�rgensen 

Morten M. J=F8rgensen wrote:
> Hi all,
>
> I am trying to figure out the consequences of groupdelay variation in
a DMT
> system. When some frequencies of the pasband are expossed to a
groupdelay
> variation of maybe 50us, will these frequencies get delayed: 50us/Ts
> samples, and thereby exceed the timedomain boundry of the original
DMT
> symbol (one N point FFT) and cross over into another symbol?
>
> Are my thoughts on this matter all wrong ?

I don't know much about the DMT application, but group delays *may*
have the effect that energy "leaks" from one time frame to the next.

Group delays are no problems if they are equal for all frequencues.
That's just an extra delay of the signal. Problems may occur if
the group delays differ for different frequencies. That's a dispersive
channel, and the net effect could be that the recieved signal is
totally
unintelligable. The symbols may interfer with each other in time, or
the sympol waveforms may have been distorted so they can no longer
be recognized at the reciever.

Channel dispersion is one of the really big problems in low-frequency
acoustic communication under water.

Rune

I dont know the answer to your question, but I have a suggestion that
may lead you to the answer. The channel you are considering has a large
group delay at some frequencies. But what does this do to the impulse
response? You should examine the channel impulse response, using Matlab
or whatever. you will see that  a certain amount of dispersion is
introduced by the group delay on certain frequencies. The key point is:

if the time-dispersion is shorter in duration than the cyclic prefix,
then NO, you will not get interference between successive frames.
However, if the impulse response is significantly longer than the
cyclic prefix, you will get appreciable ISI.

To overcome this, it is common to use a time-domain equaliser (TEQ) to
shorten the channel impulse response. What this equaliser might do is
delay the remaining frequencies to produuce the same group delay on all
frequencies.

> if the time-dispersion is shorter in duration than the cyclic prefix,
> then NO, you will not get interference between successive frames.
> However, if the impulse response is significantly longer than the
> cyclic prefix, you will get appreciable ISI.

Hmm....I'm not quite sure I get it :D Is the ISI caused by the "transient" 
of my impulse respons or is it caused by the groupdelay variation....?

The cyclic prefix is 80 samples with an IFFT size of 1024 and a sampling 
freq of 4.416Mhz. This make the lenght of the cyclic prefix 80*(4.416Mhz)^-1 
=18.1us but still the standard maximum is at 50us ? Why won't a 40us 
groupdelay variation cause ISI ?

A bit confussed ;) 

Hi all,

I am trying to figure out the consequences of groupdelay variation in a DMT
system. When some frequencies of the pasband are expossed to a groupdelay
variation of maybe 50us, will these frequencies get delayed: 50us/Ts
samples, and thereby exceed the timedomain boundry of the original DMT
symbol (one N point FFT) and cross over into another symbol?

Are my thoughts on this matter all wrong ?

Thanks in advance

--
Morten M. J�rgensen

************************************************

Well, if there is a variation in group delay then certainly different 
frequencies will be delayed differently - really by definition of "group 
delay" which is usually supposed to be a single number and not a range. 
But, if in context, it should be understandable.

ISI could be caused by both  group delay variation and impulse reponse 
transient as a general statement.

Generally you design the impulse response so that ISI is zeroed out.  If I'm 
talking about the same thing you are.  Then you suffer system variations on 
top of that.

There is a pretty good discussion at:
http://commsdesign.com/design_corner/showArticle.jhtml?articleID=47902272
and there is a block diagram at:
http://i.cmpnet.com/commsdesign/2004/sept04/dmt-fig2.jpg

also:
http://www.nextep.com.au/upload/DSL_Modulation_Techniques.pdf

and:
http://www.ifip.or.at/con2000/icct2000/icct420.pdf

You can Google on better key words than I.....

Fred

> There is a pretty good discussion at:
> http://commsdesign.com/design_corner/showArticle.jhtml?articleID=47902272
> and there is a block diagram at:
> http://i.cmpnet.com/commsdesign/2004/sept04/dmt-fig2.jpg
>
> also:
> http://www.nextep.com.au/upload/DSL_Modulation_Techniques.pdf
>
> and:
> http://www.ifip.or.at/con2000/icct2000/icct420.pdf

Great links....thanks allot!

--
Morten 

"Morten M. J&#4294967295;rgensen" <grinder213SPAM@HELL_NOhotmail.com> writes:

>> if the time-dispersion is shorter in duration than the cyclic prefix,
>> then NO, you will not get interference between successive frames.
>> However, if the impulse response is significantly longer than the
>> cyclic prefix, you will get appreciable ISI.
>
> Hmm....I'm not quite sure I get it :D Is the ISI caused by the "transient" 
> of my impulse respons or is it caused by the groupdelay variation....?

Thre's a difference?
-- 
%  Randy Yates                  % "Remember the good old 1980's, when 
%% Fuquay-Varina, NC            %  things were so uncomplicated?"
%%% 919-577-9882                % 'Ticket To The Moon' 
%%%% <yates@ieee.org>           % *Time*, Electric Light Orchestra
http://home.earthlink.net/~yatescr

> Thre's a difference?

Saying that the transient Impuls reponse is the same as groupdelay?

--
Morten 

On 19 Apr 2005 04:26:32 +0100, "Morten M. J�rgensen"
<grinder213SPAM@HELL_NOhotmail.com> wrote:

>> Thre's a difference?
>
>Saying that the transient Impuls reponse is the same as groupdelay?

Doesn't one completely define the other?

Chris Hornbeck
6x9=42 April 29

in article 1113808864.056598.59910@g14g2000cwa.googlegroups.com, Rune Allnor
at allnor@tele.ntnu.no wrote on 04/18/2005 03:21:
...
> I don't know much about the DMT application,

can someone define "DMT"?

-- 

r b-j                  rbj@audioimagination.com

"Imagination is more important than knowledge."