Forums

low frequency response of adaptive equalizer?

Started by mk April 21, 2006
Hi,
I have an LMS based adaptive equalizer which mostly works well but it
doesn't tolerate low frequency inputs too well. The channel consists
of two transformers (high pass) and cable (low pass) and when there is
long periods of no-change in data (which is possible unfortunately
even though the data is scrambled) and the SNR goes down during these
long periods of no change. I think this is because the equalizer's
bandwidth is not high enough  ie it is not able to compensate for the
high pass behavior of the transformers. Any idea how to approach this
problem ?
mk wrote:
> Hi, > I have an LMS based adaptive equalizer which mostly works well but it > doesn't tolerate low frequency inputs too well. The channel consists > of two transformers (high pass) and cable (low pass) and when there is > long periods of no-change in data (which is possible unfortunately > even though the data is scrambled) and the SNR goes down during these > long periods of no change. I think this is because the equalizer's > bandwidth is not high enough ie it is not able to compensate for the > high pass behavior of the transformers. Any idea how to approach this > problem ?
You call your transformers high pass; they are in fact band pass. Transformers don't pass DC, and long periods of no change in the signal approximate DC. As a rule of thumb, to limit phase shift to 45 degrees through two transformers, their low-frequency 3 dB points must be (sqrt(2) - 1)/T, where T is the lime of no change. (The low-frequency break points depend on the transformer loading) Does your cable substantially attenuate the high frequencies in your signal? What frequencies do you use? Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
On Fri, 21 Apr 2006 17:05:33 -0400, Jerry Avins <jya@ieee.org> wrote:

>mk wrote: >> Hi, >> I have an LMS based adaptive equalizer which mostly works well but it >> doesn't tolerate low frequency inputs too well. The channel consists >> of two transformers (high pass) and cable (low pass) and when there is >> long periods of no-change in data (which is possible unfortunately >> even though the data is scrambled) and the SNR goes down during these >> long periods of no change. I think this is because the equalizer's >> bandwidth is not high enough ie it is not able to compensate for the >> high pass behavior of the transformers. Any idea how to approach this >> problem ? > >You call your transformers high pass; they are in fact band pass. >Transformers don't pass DC, and long periods of no change in the signal >approximate DC. As a rule of thumb, to limit phase shift to 45 degrees >through two transformers, their low-frequency 3 dB points must be >(sqrt(2) - 1)/T, where T is the lime of no change. (The low-frequency >break points depend on the transformer loading) > >Does your cable substantially attenuate the high frequencies in your >signal? What frequencies do you use? > >Jerry
I understand that in general transformers are band-pass but within my frequency of interest they are high-pass. The ones I am using have an insertion loss of max -1dB 1-100 MHz. My baud rate is 62.5 MHz and the cable has around 16 dB loss at that point.
mk wrote:

   ...

> I understand that in general transformers are band-pass but within my > frequency of interest they are high-pass. The ones I am using have an > insertion loss of max -1dB 1-100 MHz. My baud rate is 62.5 MHz and the > cable has around 16 dB loss at that point.
Thanks for the cable information. How does the longest pulse compare to the transformers' time constant? You can increase the time constant by resistively loading the transformer or, better, by choosing a turns ratio that will make the cable load it more heavily. Jerry -- Engineering is the art of making what you want from things you can get. &#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;
Jerry Avins wrote:
> mk wrote: > > ... > > > I understand that in general transformers are band-pass but within my > > frequency of interest they are high-pass. The ones I am using have an > > insertion loss of max -1dB 1-100 MHz. My baud rate is 62.5 MHz and the > > cable has around 16 dB loss at that point. > >
When there is no change to the data for a while, what pattern is sent over the line? Does this pattern have a DC component? As you note the xformers will not pass DC and the Adaptive EQ cannot correct for this,, you need to use a randomizing pattern that ensures there is no DC component. Mark
On 21 Apr 2006 18:56:57 -0700, "Mark" <makolber@yahoo.com> wrote:

> >Jerry Avins wrote: >> mk wrote: >> >> ... >> >> > I understand that in general transformers are band-pass but within my >> > frequency of interest they are high-pass. The ones I am using have an >> > insertion loss of max -1dB 1-100 MHz. My baud rate is 62.5 MHz and the >> > cable has around 16 dB loss at that point. >> >> >When there is no change to the data for a while, what pattern is sent >over the line? >
I've been talking about the wire data so no change happens on the wire.
>Does this pattern have a DC component? >
You bet :-(
>As you note the xformers will not pass DC and the Adaptive EQ cannot >correct for this,, you need to use a randomizing pattern that ensures >there is no DC component.
Interestingly there is a coder & scrambler in the system but alas it don't work :-( and they are what they, can't be changed. I am applying dc-offset correction at the AFE (to save ADC bits) so my input is not going to droop but equalizer output error still becomes higher during these events.
"mk" <kal*@dspia.*comdelete> wrote in message 
news:m15j42lb2gid729j2ukjoqsghde8oa0t97@4ax.com...
> On 21 Apr 2006 18:56:57 -0700, "Mark" <makolber@yahoo.com> wrote: > > You bet :-( > >>As you note the xformers will not pass DC and the Adaptive EQ cannot >>correct for this,, you need to use a randomizing pattern that ensures >>there is no DC component. > > Interestingly there is a coder & scrambler in the system but alas it > don't work :-( and they are what they, can't be changed. I am applying > dc-offset correction at the AFE (to save ADC bits) so my input is not > going to droop but equalizer output error still becomes higher during > these events.
One technique that may work for this problem is to treat the equalizer output with partial response processing. This will put a zero at DC and eliminate that part of the spectrum from the equalizer's response. This process helps with energy around DC that is hard to equalize, but I confess I don't know what long periods of energy only at DC will do.
mk wrote:
> On 21 Apr 2006 18:56:57 -0700, "Mark" <makolber@yahoo.com> wrote: > > > > >Jerry Avins wrote: > >> mk wrote: > >> > >> ... > >> > >> > I understand that in general transformers are band-pass but within my > >> > frequency of interest they are high-pass. The ones I am using have an > >> > insertion loss of max -1dB 1-100 MHz. My baud rate is 62.5 MHz and the > >> > cable has around 16 dB loss at that point. > >> > >> > >When there is no change to the data for a while, what pattern is sent > >over the line? > > > I've been talking about the wire data so no change happens on the > wire. > > >Does this pattern have a DC component? > > > You bet :-( > > >As you note the xformers will not pass DC and the Adaptive EQ cannot > >correct for this,, you need to use a randomizing pattern that ensures > >there is no DC component. > > Interestingly there is a coder & scrambler in the system but alas it > don't work :-( and they are what they, can't be changed. I am applying > dc-offset correction at the AFE (to save ADC bits) so my input is not > going to droop but equalizer output error still becomes higher during > these events.
If you know you are trying to send data that contains a DC component through transformers, then I think you have identifed the problem... there are various coding solutions to this....try to look it up... Mark
On 22 Apr 2006 19:56:18 -0700, "Mark" <makolber@yahoo.com> wrote:

> >mk wrote: >> On 21 Apr 2006 18:56:57 -0700, "Mark" <makolber@yahoo.com> wrote: >> >> > >> >Jerry Avins wrote: >> >> mk wrote: >> >> >> >> ... >> >> >> >> > I understand that in general transformers are band-pass but within my >> >> > frequency of interest they are high-pass. The ones I am using have an >> >> > insertion loss of max -1dB 1-100 MHz. My baud rate is 62.5 MHz and the >> >> > cable has around 16 dB loss at that point. >> >> >> >> >> >When there is no change to the data for a while, what pattern is sent >> >over the line? >> > >> I've been talking about the wire data so no change happens on the >> wire. >> >> >Does this pattern have a DC component? >> > >> You bet :-( >> >> >As you note the xformers will not pass DC and the Adaptive EQ cannot >> >correct for this,, you need to use a randomizing pattern that ensures >> >there is no DC component. >> >> Interestingly there is a coder & scrambler in the system but alas it >> don't work :-( and they are what they, can't be changed. I am applying >> dc-offset correction at the AFE (to save ADC bits) so my input is not >> going to droop but equalizer output error still becomes higher during >> these events. > >If you know you are trying to send data that contains a DC component >through transformers, then I think you have identifed the problem... > >there are various coding solutions to this....try to look it up...
I know how to generate DC free codes but I don't think you paid enough attention to my post. The system definition is what it's and it's as close to cast in stone as it can get and it can't be changed.
mk wrote:
> On 22 Apr 2006 19:56:18 -0700, "Mark" <makolber@yahoo.com> wrote: > > >>mk wrote: >> >>>On 21 Apr 2006 18:56:57 -0700, "Mark" <makolber@yahoo.com> wrote: >>> >>> >>>>Jerry Avins wrote: >>>> >>>>>mk wrote: >>>>> >>>>> ... >>>>> >>>>> >>>>>>I understand that in general transformers are band-pass but within my >>>>>>frequency of interest they are high-pass. The ones I am using have an >>>>>>insertion loss of max -1dB 1-100 MHz. My baud rate is 62.5 MHz and the >>>>>>cable has around 16 dB loss at that point. >>>>> >>>>> >>>>When there is no change to the data for a while, what pattern is sent >>>>over the line? >>>> >>> >>>I've been talking about the wire data so no change happens on the >>>wire. >>> >>> >>>>Does this pattern have a DC component? >>>> >>> >>>You bet :-( >>> >>> >>>>As you note the xformers will not pass DC and the Adaptive EQ cannot >>>>correct for this,, you need to use a randomizing pattern that ensures >>>>there is no DC component. >>> >>>Interestingly there is a coder & scrambler in the system but alas it >>>don't work :-( and they are what they, can't be changed. I am applying >>>dc-offset correction at the AFE (to save ADC bits) so my input is not >>>going to droop but equalizer output error still becomes higher during >>>these events. >> >>If you know you are trying to send data that contains a DC component >>through transformers, then I think you have identifed the problem... >> >>there are various coding solutions to this....try to look it up... > > > I know how to generate DC free codes but I don't think you paid enough > attention to my post. The system definition is what it's and it's as > close to cast in stone as it can get and it can't be changed.
I don't think you paid enough attention in physics lessons. :-) You can set the design in stone as much as you like. Making it work properly is another matter. Many, many, systems over the years have been anywhere from a little quirky to totally unusable because they required DC response of a channel that doesn't offer it. Regards, Steve