Precursor and postcursor ISI

Hi,

Can anyone please explain what is the meaning of precursor and postcursor
ISI?

May I know how to set the length of feedforward and feedback filters for a
decision feedback equaliser?

Thanks a lot.

Richard_K wrote:
> Hi,
> 
> Can anyone please explain what is the meaning of precursor and postcursor
> ISI?

The current symbol is affected by the preceding and the post going 
symbols accordingly.

> May I know how to set the length of feedforward and feedback filters for a
> decision feedback equaliser?

It depends on how much of equalization do you need and what kind of a 
channel do you have.

Basically, the equalizer inverts the response of the channel. The 
impulse response of the equalizer should be about the same length as 
that of a channel.

Vladimir Vassilevsky

DSP and Mixed Signal Design Consultant

http://www.abvolt.com

As I understood, the ISI is affected by the past (preceding) symbols. I am
still very confused with the concept of "post going" symbols. Can you
please help again?

Many thanks.



>
>
>Richard_K wrote:
>> Hi,
>> 
>> Can anyone please explain what is the meaning of precursor and
postcursor
>> ISI?
>
>The current symbol is affected by the preceding and the post going 
>symbols accordingly.
>
>> May I know how to set the length of feedforward and feedback filters
for a
>> decision feedback equaliser?
>
>It depends on how much of equalization do you need and what kind of a 
>channel do you have.
>
>Basically, the equalizer inverts the response of the channel. The 
>impulse response of the equalizer should be about the same length as 
>that of a channel.
>
>Vladimir Vassilevsky
>
>DSP and Mixed Signal Design Consultant
>
>http://www.abvolt.com
>

Richard_K wrote:

> As I understood, the ISI is affected by the past (preceding) symbols.

The future is in the effect as well as the past.

  I am
> still very confused with the concept of "post going" symbols. Can you
> please help again?


The channel spreads the signal in the time. Because of that, the current 
symbol overlaps with the symbols before and after it, and this is what 
is commonly referred as ISI.


Vladimir Vassilevsky

DSP and Mixed Signal Design Consultant

http://www.abvolt.com

>Vladimir Vassilevsky wrote:
>> May I know how to set the length of feedforward and feedback filters
for a
>> decision feedback equaliser?
>
>It depends on how much of equalization do you need and what kind of a 
>channel do you have.
>
>Basically, the equalizer inverts the response of the channel. The 
>impulse response of the equalizer should be about the same length as 
>that of a channel.


I notice that for a linear adaptive transversal filter using LMS
algorithm, for certain channels, if I set the length of the filter the
same as the channel, it still cannot converge to the optimum value and
much longer taps is needed. May I know what are the reasons? (I thought
that when the length of the adaptive transversal filter is slightly larger
or equal to the length of the channel, then it will be sufficient for the
filter to converge to the optimum value as it can collect all the energy
from all the paths).

Thanks.

Richard_K wrote:
>> Vladimir Vassilevsky wrote:
>>> May I know how to set the length of feedforward and feedback filters
> for a
>>> decision feedback equaliser?
>> It depends on how much of equalization do you need and what kind of a 
>> channel do you have.
>>
>> Basically, the equalizer inverts the response of the channel. The 
>> impulse response of the equalizer should be about the same length as 
>> that of a channel.
> 
> 
> I notice that for a linear adaptive transversal filter using LMS
> algorithm, for certain channels, if I set the length of the filter the
> same as the channel, it still cannot converge to the optimum value and
> much longer taps is needed. May I know what are the reasons? (I thought
> that when the length of the adaptive transversal filter is slightly larger
> or equal to the length of the channel, then it will be sufficient for the
> filter to converge to the optimum value as it can collect all the energy
> from all the paths).

What do your taps look like, when you use a lot and get the equalizer to 
converge? Are the active taps sitting nicely in the centre of the tap 
buffer, or are they near one end? Failing to ensure they centre seems a 
common problem. Its no use having more taps than the finite impulse 
response of your channel simulator (which is what you appear to have, 
from what you said) if your training isn't using them effectively.

Steve

Richard_K wrote:
>>Vladimir Vassilevsky wrote:
>>
>>>May I know how to set the length of feedforward and feedback filters
> 
> for a
> 
>>>decision feedback equaliser?
>>
>>It depends on how much of equalization do you need and what kind of a 
>>channel do you have.
>>
>>Basically, the equalizer inverts the response of the channel. The 
>>impulse response of the equalizer should be about the same length as 
>>that of a channel.
> 
> 
> 
> I notice that for a linear adaptive transversal filter using LMS
> algorithm, for certain channels, if I set the length of the filter the
> same as the channel, it still cannot converge to the optimum value and
> much longer taps is needed. May I know what are the reasons?

Poles are compensated by zeroes and zeroes are compensated by poles. The 
transversal filter is zero only structure. Therefore if the channel has 
zeroes, the filter has to be of infinite length to compensate for it.


  (I thought
> that when the length of the adaptive transversal filter is slightly larger
> or equal to the length of the channel, then it will be sufficient for the
> filter to converge to the optimum value as it can collect all the energy
> from all the paths).
> 
> Thanks.


Vladimir Vassilevsky

DSP and Mixed Signal Design Consultant

http://www.abvolt.com

>>>Vladimir Vassilevsky wrote:
>>>
>>>>May I know how to set the length of feedforward and feedback filters
>> for a decision feedback equaliser?
>>>
>>>It depends on how much of equalization do you need and what kind of a 
>>>channel do you have.
>>>
>>>Basically, the equalizer inverts the response of the channel. The 
>>>impulse response of the equalizer should be about the same length as 
>>>that of a channel.

>> 
>> I notice that for a linear adaptive transversal filter using LMS
>> algorithm, for certain channels, if I set the length of the filter the
>> same as the channel, it still cannot converge to the optimum value and
>> much longer taps is needed. May I know what are the reasons?
>
>Poles are compensated by zeroes and zeroes are compensated by poles. The

>transversal filter is zero only structure. Therefore if the channel has 
>zeroes, the filter has to be of infinite length to compensate for it.

In this case, do you mean that for IIR multipath channel (channel with
both zeroes and poles), the length of the adaptive linear transversal
filter needs to be very much longer than the channel length in order to
achieve optimum performance?

On the hand, if the multipath channel is FIR type (only poles), then
adaptive linear transversal filter with the length of nearly equal with
the channel length will be fine enough to achieve optimum performance?

When we talk about postcursor ISI (or post symbols or future symbols), is
this only happen in the case of IIR multipath channel? For FIR multipath
channel, only the precursor ISI (Past symbols) will cause the ISI?

Thanks for your time and help.

Richard_K wrote:
>>>> Vladimir Vassilevsky wrote:
>>>>
>>>>> May I know how to set the length of feedforward and feedback filters
>>> for a decision feedback equaliser?
>>>> It depends on how much of equalization do you need and what kind of a 
>>>> channel do you have.
>>>>
>>>> Basically, the equalizer inverts the response of the channel. The 
>>>> impulse response of the equalizer should be about the same length as 
>>>> that of a channel.
> 
>>> I notice that for a linear adaptive transversal filter using LMS
>>> algorithm, for certain channels, if I set the length of the filter the
>>> same as the channel, it still cannot converge to the optimum value and
>>> much longer taps is needed. May I know what are the reasons?
>> Poles are compensated by zeroes and zeroes are compensated by poles. The
> 
>> transversal filter is zero only structure. Therefore if the channel has 
>> zeroes, the filter has to be of infinite length to compensate for it.
> 
> In this case, do you mean that for IIR multipath channel (channel with
> both zeroes and poles), the length of the adaptive linear transversal
> filter needs to be very much longer than the channel length in order to
> achieve optimum performance?
> 
> On the hand, if the multipath channel is FIR type (only poles), then
> adaptive linear transversal filter with the length of nearly equal with
> the channel length will be fine enough to achieve optimum performance?
> 
> When we talk about postcursor ISI (or post symbols or future symbols), is
> this only happen in the case of IIR multipath channel? For FIR multipath
> channel, only the precursor ISI (Past symbols) will cause the ISI?

You seem to have a fixation with multipath. In general channels are 
dispersive. When you receive the bulk of the energy from a symbol, you 
most likely get some of the energy from future symbols at that time. 
Multipath can make this more interesting, if the main path is not the 
shortest one. However, ISI from future systems can happen without any 
multipathing.

Steve

Richard_K wrote:

> 
> In this case, do you mean that for IIR multipath channel (channel with
> both zeroes and poles), the length of the adaptive linear transversal
> filter needs to be very much longer than the channel length in order to
> achieve optimum performance?


Richard,

What exactly are you trying to build?


Vladimir Vassilevsky

DSP and Mixed Signal Design Consultant

http://www.abvolt.com