DSPRelated.com
Forums
Online · Live + on demand
February 9-11, 2027
Price goes up Jan 1: register now to lock in today's rate
Register Now →

implementation of a very fast lookup table

Started by Alireza ●June 10, 2011
Rune Allnor <allnor@tele.ntnu.no> wrote:

(moving to the top, someone wrote)

>> Linearising power amps by lookup in a simplistic way brings >> benefits, although something more clever is needed to get big benefits. >> Linearity isn't really the problem with data converters, though.
(and also)
>> He actually said he's trying to linearise an 800M sample/second data >> converter.
> That's the statement I don't understand: If this is all digital, > then I can't see what the problem is: Exchange one codeword for > another. Possibly throigh a LUT.
Not having followed the state of the art for RF power amps very closely at all, as well as I know it linearity is important.
> If this *is* digital, and there *is* a problem, I can't see where > the problem comes from: If the problem originates in the digital > domain, then use some other LUT. But if you do that, the first > LUT was off, pointing towards a flawed design of that first LUT.
> But then I ask why one would need adaption: If one gets it right > once, in digital domain, there will be no need to introduce > adaptions.
OK, say, for example, that one has the output of a digital modulator, AM, FM, or some other M. Now one wants to send it through a DAC and then to a high-power RF amplifier. The amplifier isn't perfectly linear, and one wants to correct that.
> So why does one need adaptive methods at all? That is usually > associated with stuff that varies in the analog domain. And if > *that* is the case, why do the correction in digital domain?
It is usually easier and cheaper now to do digitally what can be done digitally. As a rough guess, say one knows the transfer function as a function of amplifier temperature, and wants to correct that. Store a bunch of tables, and copy the appropriate one in as the temperature changes. Interesting thought, though, one might not want a partially update table. Reserve one bit as a table select bit, update the appropriate half of the RAM, then switch the bit. It seems that knowing the way that the non-linearities are determined is important to answering the question. -- glen

vbplaya wrote:

> On Jun 16, 6:33 am, Vladimir Vassilevsky <nos...@nowhere.com> wrote: > >>Rune Allnor wrote: >> >>>On Jun 16, 12:21 am, Alireza <dastgh...@gmail.com> wrote: >> >>>>I want to build a predistorter to linearize a data converter >> >>>What on earth does that mean? If you want to linearize >>>something in the ADC/DAC chain, why do it at the digital >>>end? Wouldn't it make a lot more sense after the DAC stage, >>>before or after the nonlinear component in the analog >>>part of the chain? >> >>This is probably has to do with RF power amplifier linearization. This >>is quite popular area of research. The stupident is trying to solve the >>problem by brute force; with a naive hope that the static nonlinearity >>with adaptation could help the case. >> >>Vladimir Vassilevsky >>DSP and Mixed Signal Design Consultanthttp://www.abvolt.com > > > Wow ... do you think some of you (and I don't mean all of you) so > called professionals can step off the belittle train for a moment to > provide some guidance? We all started somewhere and rather than > destroying this person's ego right out the gate ... maybe you could > provide them the tools to approach this implementation in a > constructive manner so that in the future he/she can grow from this > and learn the proper steps to system designing. Any jack___ can > insult someone for lack of experience and knowledge but it takes > someone special to step out and extend a hand to someone who obviously > needs it. Students don't need to be reminded of how "intelligent" you > are ... they get it ... you know what you are doing ... and one > day ... they'll be there one day too. > > I came on this thread because it looked like an interesting read ... > NO I don't have any quality input into this subject because this isn't > my area of research ... and if I know my average forum trolls ... this > is where this entire post will get lost in the noise because the door > has been opened to criticize me on my lack of knowledge in this area. > I'm complentent in this arena but not fluent and if you feel that > gives you reason to cut me down and COMPLETELY miss the point of > helping out a person who needs it ... feel free ... I don't really > care. One of many things I do understand is leading/mentoring ... so > do us all a favor and I apologize for throwing out a classic here but > stop being part of the problem with our youth and start being part of > the solution ... > > And to all those who come on the site wanting someone else to solve > their problems for them ... explain what you are doing and why (ex. > student studies) and ask for direction ... not solutions ... it goes > both ways ... you need to be humble in your approach but want to learn > the correct way to tackle your particular design. > > I'm sorry I went on a bit of a rant there but I get bothered by how > easily people want to tear someone apart rather than just give a bit > of help to someone who usually needs it the most. I mean > seriously ... are we all a bunch of know-it-all Trekkies on here or > what (nerds with attitudes)? We sit here and use 5th grade playground > tactics to cut down a "stupident" so that we can show our superiority > and then expect to be taken seriously as a professional? > > If you want proof that most people who comment on forums are just > wanting to criticize people and that you may be one of them ... think > of this ... how many of you read this and were able to ignore the fact > that I misspelled 'competent' earlier and just read the message for > its content and not critique it? And how many of you were simply > reading the words till the end so that you could quickly reply to it > and point out how much of an idiot I must be since I didn't know how > to spell it? I wonder ... > > S
On Jun 17, 12:48&#4294967295;pm, Vladimir Vassilevsky <nos...@nowhere.com> wrote:
> vbplaya wrote: > > On Jun 16, 6:33 am, Vladimir Vassilevsky <nos...@nowhere.com> wrote: > > >>Rune Allnor wrote: > > >>>On Jun 16, 12:21 am, Alireza <dastgh...@gmail.com> wrote: > > >>>>I want to build a predistorter to linearize a data converter > > >>>What on earth does that mean? If you want to linearize > >>>something in the ADC/DAC chain, why do it at the digital > >>>end? Wouldn't it make a lot more sense after the DAC stage, > >>>before or after the nonlinear component in the analog > >>>part of the chain? > > >>This is probably has to do with RF power amplifier linearization. This > >>is quite popular area of research. The stupident is trying to solve the > >>problem by brute force; with a naive hope that the static nonlinearity > >>with adaptation could help the case. > > >>Vladimir Vassilevsky > >>DSP and Mixed Signal Design Consultanthttp://www.abvolt.com > > > Wow ... do you think some of you (and I don't mean all of you) so > > called professionals can step off the belittle train for a moment to > > provide some guidance? &#4294967295;We all started somewhere and rather than > > destroying this person's ego right out the gate ... maybe you could > > provide them the tools to approach this implementation in a > > constructive manner so that in the future he/she can grow from this > > and learn the proper steps to system designing. &#4294967295;Any jack___ can > > insult someone for lack of experience and knowledge but it takes > > someone special to step out and extend a hand to someone who obviously > > needs it. &#4294967295;Students don't need to be reminded of how "intelligent" you > > are ... they get it ... you know what you are doing ... and one > > day ... they'll be there one day too. > > > I came on this thread because it looked like an interesting read ... > > NO I don't have any quality input into this subject because this isn't > > my area of research ... and if I know my average forum trolls ... this > > is where this entire post will get lost in the noise because the door > > has been opened to criticize me on my lack of knowledge in this area. > > I'm complentent in this arena but not fluent and if you feel that > > gives you reason to cut me down and COMPLETELY miss the point of > > helping out a person who needs it ... feel free ... I don't really > > care. &#4294967295;One of many things I do understand is leading/mentoring ... so > > do us all a favor and I apologize for throwing out a classic here but > > stop being part of the problem with our youth and start being part of > > the solution ... > > > And to all those who come on the site wanting someone else to solve > > their problems for them ... explain what you are doing and why (ex. > > student studies) and ask for direction ... not solutions ... it goes > > both ways ... you need to be humble in your approach but want to learn > > the correct way to tackle your particular design. > > > I'm sorry I went on a bit of a rant there but I get bothered by how > > easily people want to tear someone apart rather than just give a bit > > of help to someone who usually needs it the most. &#4294967295;I mean > > seriously ... are we all a bunch of know-it-all Trekkies on here or > > what (nerds with attitudes)? &#4294967295;We sit here and use 5th grade playground > > tactics to cut down a "stupident" so that we can show our superiority > > and then expect to be taken seriously as a professional? > > > If you want proof that most people who comment on forums are just > > wanting to criticize people and that you may be one of them ... think > > of this ... how many of you read this and were able to ignore the fact > > that I misspelled 'competent' earlier and just read the message for > > its content and not critique it? &#4294967295;And how many of you were simply > > reading the words till the end so that you could quickly reply to it > > and point out how much of an idiot I must be since I didn't know how > > to spell it? &#4294967295;I wonder ... > > > S
Please don't assume anyone else here is like the person you find offensive.  I don't really know why he is like this.  He just is.  He has a good technical background, so some folks here feel like he is entitled to act like a spoiled child and throw tantrums about the newbies.  

Just ignore him and he will go away... or not, either way remember, like the quote from "Chinatown", "Forget it, Jake. It's the Internet".  

Rick
On Jun 17, 8:51&#4294967295;pm, rickman <gnu...@gmail.com> wrote:
> Please don't assume anyone else here is like the person you find offensive. &#4294967295;I don't really know why he is like this. &#4294967295;He just is. &#4294967295;He has a good technical background, so some folks here feel like he is entitled to act like a spoiled child and throw tantrums about the newbies. &#4294967295;
i don't think that. but i am happy to chat with Vlad about technical issues when we both find something interesting in common. but maybe it's good to ignore the STUPENDOUSIENT posts. r b-j

robert bristow-johnson wrote:

> On Jun 17, 8:51 pm, rickman <gnu...@gmail.com> wrote: > >>Please don't assume anyone else here is like the person you find offensive. I don't really know why he is like this. He just is. He has a good technical background, so some folks here feel like he is entitled to act like a spoiled child and throw tantrums about the newbies. > > > i don't think that. but i am happy to chat with Vlad about technical > issues when we both find something interesting in common. > > but maybe it's good to ignore the STUPENDOUSIENT posts.
I don't have anything personal about pesky flies, which meaniglessly buzz and bustle in the window pane. Sometimes it is fun to drive them by a flyswatter.
On the other hand I don't like it when they land on me... I know what they eat...

Rick
On Jun 17, 9:09=A0am, glen herrmannsfeldt <g...@ugcs.caltech.edu> wrote:
> Rune Allnor <all...@tele.ntnu.no> wrote: > > (moving to the top, someone wrote) > > >> Linearising power amps by lookup in a simplistic way brings > >> benefits, although something more clever is needed to get big benefits=
.
> >> Linearity isn't really the problem with data converters, though. > > (and also) > > >> He actually said he's trying to linearise an 800M sample/second data > >> converter. > > That's the statement I don't understand: If this is all digital, > > then I can't see what the problem is: Exchange one codeword for > > another. Possibly throigh a LUT. > > Not having followed the state of the art for RF power amps very > closely at all, as well as I know it linearity is important.
..
> It seems that knowing the way that the non-linearities > are determined is important to answering the question.
Exactly. I can't see how that can be done on the fly: If you are good enough to *estimate* the change in nonlinearities, then you are good enough to *compensate* for those changes, by far better and more efefctive means than a pre-determined LUT. To me, the problem statement seems a bit off. Rune
Rune Allnor <allnor@tele.ntnu.no> wrote:

(snip, I wrote)

>> Not having followed the state of the art for RF power amps very >> closely at all, as well as I know it linearity is important.
>> It seems that knowing the way that the non-linearities >> are determined is important to answering the question.
> Exactly. I can't see how that can be done on the fly: If you > are good enough to *estimate* the change in nonlinearities, > then you are good enough to *compensate* for those changes, > by far better and more efefctive means than a pre-determined LUT.
OK, but I was assuming that the OP knew, but didn't say. Some possibilities: (Assume that non-linearities change slowly.) 1) Run an ADC on the amplifier output, compute the difference between that and the input, adjust as needed. 2) As fast ADC are harder to make than fast DAC, run a DAC on the uncorrected value, compare using analog circuits to the amplifier output, then (slowly) generate the compensation data. 3) Assume non-linearities change in a known way with amplifier parameters, such as temperatures. Measure and correct using previously computed functions. 4) Add an extra signal to the amplifier input that can be later filtered out. Use the resulting components in the output to measure the non-linearity. A small AM component on an FM signal might be possible, for example. OK, so four possibilities that may or may not work. Somehow this reminds me of the person asking about a self-tuning piano that ran current through the wires to adjust the tuning. That, as well as I remember, was open loop, such that it tuned on power up and kept the computed values. I thought it was more interesting to do closed loop, such as running an inaudible signal to track the changes, though not so fast as to track the desired vibrations of the wire. -- glen
Thanks for the comments everyone.

The main idea of the project is that we can build nonlinear analog circuits=
 and correct for nonlinearity with DSP. Relaxing the linearity requirements=
 in analog saves power and on the other hand DSP is cheap and efficient tha=
nks to scaling. Here, instead of a fast and accurate DAC we build a fast an=
d nonlinear DAC plus a slow and accurate calibration ADC plus a digital cal=
ibration circuit and a lookup table. Not any nonlinear DAC can easily be ca=
librated, it has to be carefully designed such that the correction is feasi=
ble. A proof-of-concept is already built and what I am trying to do is to s=
implify the digital calibration algorithm. Here is the title of a previous =
publication (with a different calibration method) if you are interested:=20
"A 12-bit 800-MS/s switched-capacitor DAC with open-loop output driver and =
digital predistortion".

I didn't intend to discuss the main idea here so I didn't provide enough in=
formation. My question was merely about an implementation detail: how to bu=
ild a fast refreshable lookup table. I learned that even though the interna=
l clocking rates of FPGAs are lower than my speed requirements they can sti=
ll handle much faster data streams through SerDes. I can use the block RAM =
in the FPGA as a lookup table and I can write to it through e.g. the PCIe i=
nterface. All is left to do is just quite a bit of work to set it all up an=
d running...

Alireza