The Designer's Guide Community
Forum
Welcome, Guest. Please Login or Register. Please follow the Forum guidelines.
Aug 1st, 2021, 11:59pm
Pages: 1
Send Topic Print
DFE (Read 13995 times)
raja.cedt
Senior Fellow
******
Offline



Posts: 1516
Germany
DFE
Jul 11th, 2010, 10:09am
 
hi, can any one please provide some reference  implementation details for decision feedback equalization?

thanks.
Rajasekhar.
Back to top
 
 
View Profile WWW raja.sekhar86   IP Logged
neoflash
Community Fellow
*****
Offline

Mixed-Signal
Designer

Posts: 389

Re: DFE
Reply #1 - Jul 13th, 2010, 1:33pm
 
Go IEEE and do the search.
Back to top
 
 
View Profile   IP Logged
love_analog
Senior Member
****
Offline



Posts: 101

Re: DFE
Reply #2 - Aug 7th, 2010, 1:38pm
 
Back to top
 
 

loveanalog.blogspot.com
The Power of Analog
View Profile   IP Logged
loose-electron
Senior Fellow
******
Offline

Best Design Tool =
Capable Designers

Posts: 1638
San Diego California
Re: DFE
Reply #3 - Aug 31st, 2010, 4:27pm
 
DFE - ah yes -- for what application?

The concept works well in some places and fails in a big way in others...



Back to top
 
 

Jerry Twomey
www.effectiveelectrons.com
Read My Electronic Design Column Here
Contract IC-PCB-System Design - Analog, Mixed Signal, RF & Medical
View Profile WWW   IP Logged
neoflash
Community Fellow
*****
Offline

Mixed-Signal
Designer

Posts: 389

Re: DFE
Reply #4 - Aug 31st, 2010, 5:30pm
 
loose-electron, would you like to list more details?

It still helps to give general ideas, but the results are far more better if more deails/examples/data are given.


loose-electron wrote on Aug 31st, 2010, 4:27pm:
DFE - ah yes -- for what application?

The concept works well in some places and fails in a big way in others...




Back to top
 
 
View Profile   IP Logged
loose-electron
Senior Fellow
******
Offline

Best Design Tool =
Capable Designers

Posts: 1638
San Diego California
Re: DFE
Reply #5 - Sep 2nd, 2010, 8:23pm
 
DFE can work pretty well if the system has a constant predictable amplitude signal.

Compensating high speed data paths is a good example. The system adapts to the amplitude and can work reasonably ok.

Where DFE falls apart is when the signal amplitude changes from transition to transition, or even slowly over time.

With that, the compensation system does not provide the
appropriate FB amplitude, and the system starts to generate errors.

I have successfully used DFE in high speed data paths, and have proven that it would not work for use in Disk Drive read channels.

You might want to consider a Viterbi detector. Just my opinion...

--- Jerry
Back to top
 
 

Jerry Twomey
www.effectiveelectrons.com
Read My Electronic Design Column Here
Contract IC-PCB-System Design - Analog, Mixed Signal, RF & Medical
View Profile WWW   IP Logged
bharat
Community Member
***
Offline



Posts: 57

Re: DFE
Reply #6 - Feb 2nd, 2011, 8:05pm
 
loose-electron,
"and have proven that it would not work for use in Disk Drive read channels"

should the amplitude be not the same for Disk Drive read channels?
OR , it might be due to noise but not that we want?
Back to top
 
 
View Profile   IP Logged
loose-electron
Senior Fellow
******
Offline

Best Design Tool =
Capable Designers

Posts: 1638
San Diego California
Re: DFE
Reply #7 - Feb 2nd, 2011, 9:19pm
 
amplitude on a pulse by pulse basis for magnetic channels varies quite a bit - head fly height, media variance, media speed as a function of head position (it changes ID to OD)
are just a few of the things that lead to signal variance.

DFE is not a good option for magnetic recording. Been there, done that as a research effort.
Back to top
 
 

Jerry Twomey
www.effectiveelectrons.com
Read My Electronic Design Column Here
Contract IC-PCB-System Design - Analog, Mixed Signal, RF & Medical
View Profile WWW   IP Logged
love_analog
Senior Member
****
Offline



Posts: 101

Re: DFE
Reply #8 - Mar 3rd, 2011, 12:01pm
 
Thats interesting that DFE doesn't work in magnetic channels.

I haven't followed disk drive channels for some time but I thought Viterbi was used because the ISI was so severe that a symbol-by-symbol detector would not work. You need a sequence detector.

I wasn't aware that Viterbi was being used because DFE doesn't work. Are there no commerical products out there which have DFE and Viterbi. I did do DFE on disk drive channel but that was a university project so curious about industry experience.
Back to top
 
 

loveanalog.blogspot.com
The Power of Analog
View Profile   IP Logged
carlgrace
Senior Member
****
Offline



Posts: 231
Berkeley, CA
Re: DFE
Reply #9 - Mar 4th, 2011, 8:29am
 
loose-electron wrote on Feb 2nd, 2011, 9:19pm:
amplitude on a pulse by pulse basis for magnetic channels varies quite a bit - head fly height, media variance, media speed as a function of head position (it changes ID to OD)
are just a few of the things that lead to signal variance.

DFE is not a good option for magnetic recording. Been there, done that as a research effort.


Loose-Electron,

I disagree that DFE can't work in read channels.  In fact, Lucent and Silicon Systems (later TI) were using DFEs in their read channel products in the mid-90s to late-90s.  The emergence of Marvel put an end to all that (and to those companies), for the most part.  So you have an existence proof right there.  There was a time in history where the power dissipation of Viterbi decoders was high enough in contemporary process technology that it made sense to make read channel products with DFEs.  They were full DFEs with feed-forward and feedback equalizers.

In your research project where you determined that DFE couldn't work, did you include a dc tap in the feedback equalizer?  It can be tricky to make it converge but it is really helpful.  I worked on an equalizer for a really nasty channel (not magnetic, but bad) and the dc tap was a life-saver.  Also, generally, magnetic recording systems have a decision-directed AGC loop that does the lion's share of the work regarding the amplitude variations across the disk you were experiencing.  
Back to top
 
 
View Profile   IP Logged
loose-electron
Senior Fellow
******
Offline

Best Design Tool =
Capable Designers

Posts: 1638
San Diego California
Re: DFE
Reply #10 - Mar 14th, 2011, 2:33pm
 
I have been involved with DFE (used to be known as QFB or quantized feedback) several times. Once as part of Maxtor, in their advanced channel development group (circa 1989-1992) then as a read channel IC designer doing designs for Seagate and Conner (circa  1994) and then as  part of Quantum (around 1995)

I have also used DFE in SerDes design for Cray, back in 2004. (My dates may be off a touch.)

The limited use of DFE in disk drives was an exercise in duct tape being placed on top of duct tape to make the concept work.

The issues of ISI were compensated by using pattern dependent feedback, making for a rather messy system architecture, and multiple signal processing paths. Not easily done and not reliable.

A few disk drives went out with DFE systems in them, but the compensation to deal with amplitude variance head to head, media to meda, and ID to OD on the drive was again, a messy case in tuning the system to work. Huge adaptive data got developed to tune the devices (a unique lookup table for each readback head/media, with a cylinder dependency set of factors was needed, to adjust the feedback, plus a calibration routine to get things set up was needed)

Magnetic tape drives have used DFE with success, because the amplitude changes are not as nasty as with a disk, and compensation for a track can be made readily.

SerDes has used DFE successfully because the channel characteristics remain consistent, and can be compensated for. Pulse to pulse variance is minimal, and the overall compensation feedback can be adjusted for.

What happens is that people fall in love with the simple concept of DFE, and then to make it work reliably, it becomes a very complex system, with lots of adjustments needed.

A few years back a VC talked to me about funding a DFE startup, my advice at the time, was not to do it. About 2 years and $3 (5?) million dollars later, they had something that would work under very limited and controlled situations, and never sold a single one.
Back to top
 
 

Jerry Twomey
www.effectiveelectrons.com
Read My Electronic Design Column Here
Contract IC-PCB-System Design - Analog, Mixed Signal, RF & Medical
View Profile WWW   IP Logged
loose-electron
Senior Fellow
******
Offline

Best Design Tool =
Capable Designers

Posts: 1638
San Diego California
Re: DFE
Reply #11 - Mar 14th, 2011, 2:47pm
 
[quote author=carlgrace link=1278868143/0#9 date=1299256179]loose-electron wrote on Feb 2nd, 2011, 9:19pm:
There was a time in history where the power dissipation of Viterbi decoders was high enough in contemporary process technology that it made sense to make read channel products with DFEs.  They were full DFEs with feed-forward and feedback equalizers.
 


Ah yes, with all those messy compensation paths. Adaptive equalizers and all the reliability issues associated therewith. Time in history? Well, I was in the middle of it at the time.

A little memory refresh for you - When PRML and Viterbi detection was a new and shiny thing (circa 1993) there was a few generation of PRML channels that were done in an analog manner, not digitally! It was sample and holds, in a pipeline sampler with differencing done with amplifiers.

At the time they could not get to a fast enough 6 bit converter to do the job with DSP methods. I seem to remember SSI doing that for Seagate, IIRC. Another semiconductor company totally dependent on Seagate, that then got destroyed by them.

Back to top
 
 

Jerry Twomey
www.effectiveelectrons.com
Read My Electronic Design Column Here
Contract IC-PCB-System Design - Analog, Mixed Signal, RF & Medical
View Profile WWW   IP Logged
loose-electron
Senior Fellow
******
Offline

Best Design Tool =
Capable Designers

Posts: 1638
San Diego California
Re: DFE
Reply #12 - Mar 14th, 2011, 2:57pm
 
Ah yeas - one addendum, What remained of SSI got bought by TI - thats correct, there was some good design talent there at the time, no idea what has happened to that group since then.
Back to top
 
 

Jerry Twomey
www.effectiveelectrons.com
Read My Electronic Design Column Here
Contract IC-PCB-System Design - Analog, Mixed Signal, RF & Medical
View Profile WWW   IP Logged
carlgrace
Senior Member
****
Offline



Posts: 231
Berkeley, CA
Re: DFE
Reply #13 - Apr 1st, 2011, 9:11am
 
Loose-electron,

Actually only the HDD business of SSI got bought by TI.  They were located in Tustin.  The comms business of SSI (Tustin, Mountain View, and Nevada City) was bought by TDK Semiconductor, it was then spun off to become Teridian Semiconductor (Nevada City was closed).  They focused on smart grid ICs and low bit-rate modems.  Teridian was recently bought by Maxim, so there you go.

The HDD activities in Tustin were shut down years ago.  TI just laid all those guys off.  I worked for a couple of companies down in OC and we interviewed a few of those guys.

Regarding your comments about DFE, I'm not surprised.  I was the first IC design engineer at ClariPhy Communications and initially we were looking to use DFE to equalize multi-mode and polarization dispersion (the marketing jargon was "Electronic Dispersion Compensation) on fiber optic channels.  It was really, really hard.  I designed an FFE in 0.13um that, quite frankly, sucked.  Turns out optical channels are pretty tough to equalize.  We ended up going with a Viterbi decoder using a monsterous 10 GHz, 6b ADC on the front end.  That was in 90 nm, and I think they are at 45nm now and equalizing 100 Gb channels, so the concept was a big success (at least technically...).

Was the DFE startup in electronic dispersion compensation?
Back to top
 
 
View Profile   IP Logged
loose-electron
Senior Fellow
******
Offline

Best Design Tool =
Capable Designers

Posts: 1638
San Diego California
Re: DFE
Reply #14 - Apr 6th, 2011, 1:46pm
 
carlgrace wrote on Apr 1st, 2011, 9:11am:
Regarding your comments about DFE, I'm not surprised.  I was the first IC design engineer at ClariPhy Communications and initially we were looking to use DFE to equalize multi-mode and polarization dispersion (the marketing jargon was "Electronic Dispersion Compensation) on fiber optic channels.  It was really, really hard.  I designed an FFE in 0.13um that, quite frankly, sucked.  Turns out optical channels are pretty tough to equalize.  We ended up going with a Viterbi decoder using a monsterous 10 GHz, 6b ADC on the front end.  That was in 90 nm, and I think they are at 45nm now and equalizing 100 Gb channels, so the concept was a big success (at least technically...).

Was the DFE startup in electronic dispersion compensation?


The DFE question has come up many times, several times while doing read channels for disk drives, SerDes receivers for digital I/O, and optical channels.

Anyplace where amplitude variance is dynamic, I have tried my best to stop it, because of the problems associated therewith. If the powers that be are in love with it, I suggest building a matlab or other behavioral model and then use that to show how the idea has problems.

Your decision to go with a Viterbi detector for optical data was a good one. Fujitsu does an ADC thats one purpose is for this

http://www.lightwaveonline.com/equipment-design/fujitsu-launches-adc-technology-...

I was briefly involved with a fiber company doing something similar to the above. When Fujisu announced, I handed the place the press announcement and told them to go buy it off the shelf instead.
Back to top
 
 

Jerry Twomey
www.effectiveelectrons.com
Read My Electronic Design Column Here
Contract IC-PCB-System Design - Analog, Mixed Signal, RF & Medical
View Profile WWW   IP Logged
Pages: 1
Send Topic Print
Copyright 2002-2021 Designer’s Guide Consulting, Inc. Designer’s Guide® is a registered trademark of Designer’s Guide Consulting, Inc. All rights reserved. Send comments or questions to editor@designers-guide.org. Consider submitting a paper or model.