Forums

adaptive equalization benefit

Started by mguo May 8, 2006
How much benefit can adaptive equalization really get you in a packet
transmission system over multipath channel? Since one will need to do
synchronization before training the equalizer, it seems to me that synch
will be the first one to fail most of the time if ISI is really serious. 

So that means you must have a synchronization scheme that can deal with
serious ISI first. Or you will have to do blind equalizer training, then
do synchronization. 

Is there any expert here willing to share some thoughts?

Thanks.
Min

mguo wrote:
> How much benefit can adaptive equalization really get you in a packet > transmission system over multipath channel? Since one will need to do > synchronization before training the equalizer, it seems to me that synch > will be the first one to fail most of the time if ISI is really serious. > > So that means you must have a synchronization scheme that can deal with > serious ISI first. Or you will have to do blind equalizer training, then > do synchronization. > > Is there any expert here willing to share some thoughts? > > Thanks. > Min >
Without adaptive equalisation only the most elementary, low bit density, forms of transmissions will generally work. There are plenty of symbol synchronisation schemes which don't care much about serious ISI. They may pull in faster when there is no ISI, but they will still pull in when there is. Carrier recovery can also be accomplished without decoding the symbols, so ISI is no big deal there, either. If you have a very rounded constellation, like V.34, blindly recovery the carrier might be tough. Most constellations have corners sticking out, even with high ISI, and you can lock to those. Once you have nailed down the symbol timing, and stopped the constellation spinning, using algorithms like CMA to train an equaliser will gradually make the symbols come into focus through the ISI, and the output error rate will fall to something acceptable. Once you are recovering the data accurately, and you can switch to decision feedback adjustment of the parameters if you so desire. Steve
>mguo wrote: >> How much benefit can adaptive equalization really get you in a packet >> transmission system over multipath channel? Since one will need to do >> synchronization before training the equalizer, it seems to me that
synch
>> will be the first one to fail most of the time if ISI is really
serious.
>> >> So that means you must have a synchronization scheme that can deal
with
>> serious ISI first. Or you will have to do blind equalizer training,
then
>> do synchronization. >> >> Is there any expert here willing to share some thoughts? >> >> Thanks. >> Min >> >Without adaptive equalisation only the most elementary, low bit density,
>forms of transmissions will generally work. > >There are plenty of symbol synchronisation schemes which don't care much
>about serious ISI. They may pull in faster when there is no ISI, but >they will still pull in when there is. > >Carrier recovery can also be accomplished without decoding the symbols, >so ISI is no big deal there, either. If you have a very rounded >constellation, like V.34, blindly recovery the carrier might be tough. >Most constellations have corners sticking out, even with high ISI, and >you can lock to those. > >Once you have nailed down the symbol timing, and stopped the >constellation spinning, using algorithms like CMA to train an equaliser >will gradually make the symbols come into focus through the ISI, and the
> output error rate will fall to something acceptable. > >Once you are recovering the data accurately, and you can switch to >decision feedback adjustment of the parameters if you so desire. > >Steve >
Hi Steve, Thanks for your reply. Could you give an example of an symbol timing scheme that does not care about multipath effect? System still need to establish the starting point for the training sequence. -Min
mguo wrote:
>> mguo wrote: >>> How much benefit can adaptive equalization really get you in a packet >>> transmission system over multipath channel? Since one will need to do >>> synchronization before training the equalizer, it seems to me that > synch >>> will be the first one to fail most of the time if ISI is really > serious. >>> So that means you must have a synchronization scheme that can deal > with >>> serious ISI first. Or you will have to do blind equalizer training, > then >>> do synchronization. >>> >>> Is there any expert here willing to share some thoughts? >>> >>> Thanks. >>> Min >>> >> Without adaptive equalisation only the most elementary, low bit density, > >> forms of transmissions will generally work. >> >> There are plenty of symbol synchronisation schemes which don't care much > >> about serious ISI. They may pull in faster when there is no ISI, but >> they will still pull in when there is. >> >> Carrier recovery can also be accomplished without decoding the symbols, >> so ISI is no big deal there, either. If you have a very rounded >> constellation, like V.34, blindly recovery the carrier might be tough. >> Most constellations have corners sticking out, even with high ISI, and >> you can lock to those. >> >> Once you have nailed down the symbol timing, and stopped the >> constellation spinning, using algorithms like CMA to train an equaliser >> will gradually make the symbols come into focus through the ISI, and the > >> output error rate will fall to something acceptable. >> >> Once you are recovering the data accurately, and you can switch to >> decision feedback adjustment of the parameters if you so desire. >> >> Steve >> > > > Hi Steve, > > Thanks for your reply. Could you give an example of an symbol timing > scheme that does not care about multipath effect? System still need to > establish the starting point for the training sequence. > > -Min >
Try http://ieeexplore.ieee.org/iel4/5559/14886/00679062.pdf?arnumber=679062 for a quick and easy to follow introduction to the topic. I don't quite understand the pictures at the end, though. The first one is supposed to be the starting point, but doesn't show the effects of the carrier not being locked - i.e. the constellation should be spinning. Steve
>mguo wrote: >>> mguo wrote: >>>> How much benefit can adaptive equalization really get you in a
packet
>>>> transmission system over multipath channel? Since one will need to
do
>>>> synchronization before training the equalizer, it seems to me that >> synch >>>> will be the first one to fail most of the time if ISI is really >> serious. >>>> So that means you must have a synchronization scheme that can deal >> with >>>> serious ISI first. Or you will have to do blind equalizer training, >> then >>>> do synchronization. >>>> >>>> Is there any expert here willing to share some thoughts? >>>> >>>> Thanks. >>>> Min >>>> >>> Without adaptive equalisation only the most elementary, low bit
density,
>> >>> forms of transmissions will generally work. >>> >>> There are plenty of symbol synchronisation schemes which don't care
much
>> >>> about serious ISI. They may pull in faster when there is no ISI, but >>> they will still pull in when there is. >>> >>> Carrier recovery can also be accomplished without decoding the
symbols,
>>> so ISI is no big deal there, either. If you have a very rounded >>> constellation, like V.34, blindly recovery the carrier might be tough.
>>> Most constellations have corners sticking out, even with high ISI, and
>>> you can lock to those. >>> >>> Once you have nailed down the symbol timing, and stopped the >>> constellation spinning, using algorithms like CMA to train an
equaliser
>>> will gradually make the symbols come into focus through the ISI, and
the
>> >>> output error rate will fall to something acceptable. >>> >>> Once you are recovering the data accurately, and you can switch to >>> decision feedback adjustment of the parameters if you so desire. >>> >>> Steve >>> >> >> >> Hi Steve, >> >> Thanks for your reply. Could you give an example of an symbol timing >> scheme that does not care about multipath effect? System still need to >> establish the starting point for the training sequence. >> >> -Min >> >Try >http://ieeexplore.ieee.org/iel4/5559/14886/00679062.pdf?arnumber=679062 >for a quick and easy to follow introduction to the topic. I don't quite >understand the pictures at the end, though. The first one is supposed to
>be the starting point, but doesn't show the effects of the carrier not >being locked - i.e. the constellation should be spinning. > >Steve >
Steve, Could you post the title of the article? I can't access the link for some reason. This is a very interesting topic. Thank you very much. -Min