Cable Equalizer

Guy Eschemann wrote:

>>May I suggest an analog RC filter? With some trivial electonics, it
>>could be made adaptive. It is probably not qualified as the solution of
>>the OP's ill-posed homework problem, however it does the job.
>>

> thanks for your suggestion. I actually did my homework before posting
> here, but couldn't find a satisfactory solution by myself so I thought
> I should ask the experts. As I found in this article (http://
> www.planetanalog.com/features/showArticle.jhtml?articleID=188702457),
> metallic cables have a frequency response which is proportional to the
> square root of the frequency,

You are correct. The attenuation exponent is *generally* proportional to 
the root of frequency.

> which seems difficult to mimic with standard analog filters.

The questions are how much of attenuation do you have and how accurate 
the compensation should be. My wild guess (based on something like 
1000ft of something like RG58) is that the basic 1-st order shoulder 
filter would be enough.

> Also, as I have an FPGA, I don't mind about
> implementing a digital equalizer.

Better and better. Do a digital filter then.


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

Guy Eschemann wrote:
>> May I suggest an analog RC filter? With some trivial electonics, it
>> could be made adaptive. It is probably not qualified as the solution of
>> the OP's ill-posed homework problem, however it does the job.
>>
>> Vladimir Vassilevsky
>> DSP and Mixed Signal Design Consultanthttp://www.abvolt.com
> 
> Vladimir,
> thanks for your suggestion. I actually did my homework before posting
> here, but couldn't find a satisfactory solution by myself so I thought
> I should ask the experts. As I found in this article (http://
> www.planetanalog.com/features/showArticle.jhtml?articleID=188702457),
> metallic cables have a frequency response which is proportional to the
> square root of the frequency, which seems difficult to mimic with
> standard analog filters. Also, as I have an FPGA, I don't mind about
> implementing a digital equalizer.

If pulse shape is important, equalizing phase may be as important as 
equalizing amplitude.

Jerry
-- 
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������

On 10/22/2009 11:24 AM, Guy Eschemann wrote:
>> How are the 8 carriers multiplexed onto the cable?  Frequency division?
>>    That's the most common approach, and if that's the case then it
>> shouldn't be too difficult to increase the transmit power on each
>> carrier as frequency increases.
>>
>> I don't know why time-domain equalization wouldn't work, either.   Since
>> the channel response is known, pre-distorting (or pre-equalizing,
>> depending on which term you prefer), shouldn't be difficult at all.
>>
>> Maybe I'm not seeing the difficulty, but it seems straightforward to me.
>> --
>> Eric Jacobsen
>> Minister of Algorithms
>> Abineau Communicationshttp://www.abineau.com
>
> Eric,
> you're right, the carriers are multiplexed onto the cable via
> frequency division, with each channel being approx. 600 kHz wide.
> Unfortunately, I have no control on the transmitter so I have to do
> all of the equalization on the receiver side. There's also no way I
> can transmit a training sequence. I'd also like to avoid having to do
> any demodulation at the receiving end since this is not the final
> receiver but just an interface to an optical transmitter. By the way,
> this goes into a real product so it really should be able to adapt to
> whatever cable type/length it is connected to.
> Regards,
> Guy.

I'm with Vlad, then.  A simple analog filter with tilt opposite the 
cable response would do it, i.e., fixed analog equalizer.

Since the symbol rates are so slow I don't think you'd need to worry too 
much about screwing up the carriers, unless you did a really crappy job 
with the analog filter.

-- 
Eric Jacobsen
Minister of Algorithms
Abineau Communications
http://www.abineau.com

On Oct 22, 9:54 am, Vladimir Vassilevsky <nos...@nowhere.com> wrote:
> Eric Jacobsen wrote:
> > On 10/22/2009 4:07 AM, Guy Eschemann wrote:
>
> >> I'm looking for some inspiration on the following problem:
>
> >> I want to transmit a signal consisting of 8 QPSK-Modulated carriers
> >> over a coaxial cable, in the frequency range of 1 to 10 MHz. Since the
> >> cable has a frequency-dependent attenuation (e.g. 1dB/100m@1MHz, 3dB/
> >> 100m@10 MHz), I need to somehow compensate for this at the receiver
> >> before it is forwarded to an optical transmitter.
>
> >> Since the signal consists of 8 superimposed modulated carriers, time-
> >> domain equalization is probably out of question. So I thought of a
> >> kind of frequency-domain equalization, which constantly monitors and
> >> adjusts the power of the 8 bands to the same level. Is this the way to
> >> do it? If yes, can you recommend any references to get me started?
>
> > How are the 8 carriers multiplexed onto the cable?  Frequency division?
> >  That's the most common approach, and if that's the case then it
> > shouldn't be too difficult to increase the transmit power on each
> > carrier as frequency increases.
>
> > I don't know why time-domain equalization wouldn't work, either.   Since
> > the channel response is known, pre-distorting (or pre-equalizing,
> > depending on which term you prefer), shouldn't be difficult at all.
>
> > Maybe I'm not seeing the difficulty, but it seems straightforward to me.
>
> May I suggest an analog RC filter? With some trivial electonics, it
> could be made adaptive. It is probably not qualified as the solution of
> the OP's ill-posed homework problem, however it does the job.
>
> Vladimir Vassilevsky
> DSP and Mixed Signal Design Consultanthttp://www.abvolt.com

Examples of the RC portions on pages 5 and 6:

http://www.intersil.com/data/an/an1307.pdf

Dale B. Dalrymple

On Oct 31, 7:38=A0pm, dbd <d...@ieee.org> wrote:
> On Oct 22, 9:54 am, Vladimir Vassilevsky <nos...@nowhere.com> wrote:
>
>
>
> > Eric Jacobsen wrote:
> > > On 10/22/2009 4:07 AM, Guy Eschemann wrote:
>
> > >> I'm looking for some inspiration on the following problem:
>
> > >> I want to transmit a signal consisting of 8 QPSK-Modulated carriers
> > >> over a coaxial cable, in the frequency range of 1 to 10 MHz. Since t=
he
> > >> cable has a frequency-dependent attenuation (e.g. 1dB/100m@1MHz, 3dB=
/
> > >> 100m@10 MHz), I need to somehow compensate for this at the receiver
> > >> before it is forwarded to an optical transmitter.
>
> > >> Since the signal consists of 8 superimposed modulated carriers, time=
-
> > >> domain equalization is probably out of question. So I thought of a
> > >> kind of frequency-domain equalization, which constantly monitors and
> > >> adjusts the power of the 8 bands to the same level. Is this the way =
to
> > >> do it? If yes, can you recommend any references to get me started?
>
> > > How are the 8 carriers multiplexed onto the cable? =A0Frequency divis=
ion?
> > > =A0That's the most common approach, and if that's the case then it
> > > shouldn't be too difficult to increase the transmit power on each
> > > carrier as frequency increases.
>
> > > I don't know why time-domain equalization wouldn't work, either. =A0 =
Since
> > > the channel response is known, pre-distorting (or pre-equalizing,
> > > depending on which term you prefer), shouldn't be difficult at all.
>
> > > Maybe I'm not seeing the difficulty, but it seems straightforward to =
me.
>
> > May I suggest an analog RC filter? With some trivial electonics, it
> > could be made adaptive. It is probably not qualified as the solution of
> > the OP's ill-posed homework problem, however it does the job.
>
> > Vladimir Vassilevsky
> > DSP and Mixed Signal Design Consultanthttp://www.abvolt.com
>
> Examples of the RC portions on pages 5 and 6:
>
> http://www.intersil.com/data/an/an1307.pdf
>
> Dale B. Dalrymple

to the OP

consider the problem at a high level..

if you are able to do this in the digital domain.  your BEST solution,
is to demodulate each of the 8 carriers, and then re modulate them,
this serves not only as an equalizer but as a regenerator, as long as
the demod is able to recover the correct data, then you are starting
with a new perfect "copy" of the signal with ALL the impairments
removed.
That is one of the big advantages of digital over analog.

Now of course, demod/remod is much more complicated.  If you can't
afford that complication then use the DSP to recover each carrier by
band pass filtering and AGC each carrier to the correct amplitude then
recombine them.   This provides overall equalization so all carriers
are the correct amplitude but it doesn't provide equalization across
the channel of each carrer, but at the rate and distance you are
going, you prob don't need that..

You have many more options if you can do this in the digital domain.

Mark