What are some suggestions for digitizing 16 bits @ up to about 2G/s? The resolution and the rate are not hard requirements - I'm just trying to figure out how to do high rate digitizing with non-exotic parts if possible. I had assumed that many A/Ds with serial out could be paralled, and the outputs pipelined, but this is not looking promising. Almost all the fast A/Ds are parallel out, which makes sense, but these parts are also spendy. I've looked at TI and Analog, and the 200 M/s parts tend to be in the $50 range, and of course are quite large. BGA is OK. Do I get 10 or so of these 200 M/s parts and parallel them, or are there other ways? This project is currently in the hobby domain. This will probably start off as a spectrum analyzer. I plan on doing an FPGA-centric design, so interfacing many dozen pins may be an option.
Digitizing advice
Started by ●October 18, 2005
Reply by ●October 18, 20052005-10-18
"Bryan Hackney" <no@body.home> wrote in message news:09%4f.42365$Q53.10742@tornado.texas.rr.com...> > What are some suggestions for digitizing 16 bits @ up to about 2G/s? The > resolution and the rate are not hard requirements - I'm just trying to > figure out how to do high rate digitizing with non-exotic parts ifpossible.> > I had assumed that many A/Ds with serial out could be paralled, and theoutputs> pipelined, but this is not looking promising. Almost all the fast A/Ds are > parallel out, which makes sense, but these parts are also spendy. I'velooked> at TI and Analog, and the 200 M/s parts tend to be in the $50 range, andof> course are quite large. > > BGA is OK. Do I get 10 or so of these 200 M/s parts and parallel them, orare> there other ways? > > This project is currently in the hobby domain. This will probably startoff> as a spectrum analyzer. I plan on doing an FPGA-centric design, sointerfacing> many dozen pins may be an option.If you are making a spectrum analyser you can undersample in bands - rememebr you ened only sample 2B where B is teh bandwidth so you can look at the spectrum in frequency bands which I believe is the method used in some digital scopes.Your A/D must have a good analogue bandwidth though. McC
Reply by ●October 18, 20052005-10-18
Real_McCoy wrote:> "Bryan Hackney" <no@body.home> wrote in message > news:09%4f.42365$Q53.10742@tornado.texas.rr.com... > >>What are some suggestions for digitizing 16 bits @ up to about 2G/s? The >>resolution and the rate are not hard requirements - I'm just trying to >>figure out how to do high rate digitizing with non-exotic parts if possible. > > > If you are making a spectrum analyser you can undersample in bands - > rememebr you ened only sample 2B where B is teh bandwidth so you can look at > the spectrum in frequency bands which I believe is the method used in some > digital scopes.Your A/D must have a good analogue bandwidth though. > > McC > >I am aiming at a radio that can demodulate up to 1GHz without intermediate stages, thus the 2G/s rate. I'm not sure where this will lead, but I'm looking at the fast digitizing part as the real challenge - the application is just there. It could be a scope, spectrum analyzer, radio, etc.
Reply by ●October 18, 20052005-10-18
Bryan Hackney wrote:> Real_McCoy wrote: >> "Bryan Hackney" <no@body.home> wrote in message >> news:09%4f.42365$Q53.10742@tornado.texas.rr.com... >> >>> What are some suggestions for digitizing 16 bits @ up to about 2G/s? The >>> resolution and the rate are not hard requirements - I'm just trying to >>> figure out how to do high rate digitizing with non-exotic parts if possible. >> >> If you are making a spectrum analyser you can undersample in bands - >> rememebr you ened only sample 2B where B is teh bandwidth so you can look at >> the spectrum in frequency bands which I believe is the method used in some >> digital scopes.Your A/D must have a good analogue bandwidth though. >> >> McC >> > I am aiming at a radio that can demodulate up to 1GHz without intermediate > stages, thus the 2G/s rate. I'm not sure where this will lead, but I'm > looking at the fast digitizing part as the real challenge - the application > is just there. It could be a scope, spectrum analyzer, radio, etc.Ok, so I'm guessing this is for some SDR type setup. I'd go with McC's suggestion of using under sampling. You will need to look into the analog bandwidth of the ADCs, to ensure that the sampling aperture is sufficiently small to prevent any unwanted noise from the sampling. This would allow you to use a more modest ADC, perhaps something of around 200MSPS would suffice. Otherwise you'll be hard pressed to get that many bits of resolution for a cheap price. You could do two interleaved 1GSPS ADCs, with parallel LVDS outputs. Two options, personally I'd go with the first. More economical, and easier to interface as well. Just requires a bit more front-end circuitry.
Reply by ●October 18, 20052005-10-18
If you undersample, then in fact you do need an analog hold circuit that is fast enough for the total carrier frequency (2G) - but your ADC need only be fast enough for the band you are in fact measuring. This is because when undersampling, once the signal is held (the expensive part) you have all the time in the world (relatively..) to actually convert that held voltage to digital, so you can use a cheaper ADC. There is a short (and old) paper on our web site that refers to undersampling, though it does not discuss this hardware aspect: it is the 'zoom FFT' paper lited on the home page (bottom left link) www.bores.com Chris ===================== Chris Bore www.bores.com
Reply by ●October 18, 20052005-10-18
Real_McCoy wrote: ...> If you are making a spectrum analyser you can undersample in bands - > rememebr you ened only sample 2B where B is teh bandwidth so you can look at > the spectrum in frequency bands which I believe is the method used in some > digital scopes.Your A/D must have a good analogue bandwidth though.Clock jitter and aperture uncertainty need to be superb too. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●October 19, 20052005-10-19
"Bryan Hackney" <no@body.home> wrote in message news:Fr%4f.50622$zw5.37793@tornado.texas.rr.com...> Real_McCoy wrote: > > "Bryan Hackney" <no@body.home> wrote in message > > news:09%4f.42365$Q53.10742@tornado.texas.rr.com... > > > >>What are some suggestions for digitizing 16 bits @ up to about 2G/s? The > >>resolution and the rate are not hard requirements - I'm just trying to > >>figure out how to do high rate digitizing with non-exotic parts ifpossible.> > > > > > If you are making a spectrum analyser you can undersample in bands - > > rememebr you ened only sample 2B where B is teh bandwidth so you canlook at> > the spectrum in frequency bands which I believe is the method used insome> > digital scopes.Your A/D must have a good analogue bandwidth though. > > > > McC > > > > > > I am aiming at a radio that can demodulate up to 1GHz without intermediate > stages, thus the 2G/s rate. I'm not sure where this will lead, but I'm > looking at the fast digitizing part as the real challenge - theapplication> is just there. It could be a scope, spectrum analyzer, radio, etc.Unless your frequency response runs from DC to 1GHz then you need not sample so high! Divide your signal into bands.Mobile phones work at around a 1GHz but the bandwidth is in kHz. McC
Reply by ●October 19, 20052005-10-19
Real_McCoy wrote:> "Bryan Hackney" <no@body.home> wrote in message > news:Fr%4f.50622$zw5.37793@tornado.texas.rr.com... > >>Real_McCoy wrote: >> >>>"Bryan Hackney" <no@body.home> wrote in message >>>news:09%4f.42365$Q53.10742@tornado.texas.rr.com... >>> >>> >>>>What are some suggestions for digitizing 16 bits @ up to about 2G/s? The >>>>resolution and the rate are not hard requirements - I'm just trying to >>>>figure out how to do high rate digitizing with non-exotic parts if > > possible. > >>> >>>If you are making a spectrum analyser you can undersample in bands - >>>rememebr you ened only sample 2B where B is teh bandwidth so you can > > look at > >>>the spectrum in frequency bands which I believe is the method used in > > some > >>>digital scopes.Your A/D must have a good analogue bandwidth though. >>> >>>McC >>> >>> >> >>I am aiming at a radio that can demodulate up to 1GHz without intermediate >>stages, thus the 2G/s rate. I'm not sure where this will lead, but I'm >>looking at the fast digitizing part as the real challenge - the > > application > >>is just there. It could be a scope, spectrum analyzer, radio, etc. > > > Unless your frequency response runs from DC to 1GHz then you need not sample > so high! Divide your signal into bands.Mobile phones work at around a 1GHz > but the bandwidth is in kHz. > > McC > >Say baseband is audio. Carrier is fc. Sampling rate can be fc/2. The carrier is aliased to fc/4, and baseband can be extracted by removing the aliased carrier from the digitized stream. What can prevent signals modulated at fc/2 and fc/4 showing up? If this requires an analog bandpass before digitizing, why not just downconvert? It seems that a hybrid downconverter is more flexible. BTW, thanks to all. I'm taking a crash course in undersampling.
Reply by ●October 19, 20052005-10-19
Analog Devices had a recent "Analog Dialogue" article related to this topic. See http://www.analog.com/library/analogDialogue/archives/39-05/multichannel.html These also might be useful to you: http://www.analog.com/library/analogDialogue/archives/39-04/transformer.html http://www.analog.com/library/analogDialogue/archives/39-06/architecture.html Best wishes! -- Jon Harris SPAM blocker in place: Remove 99 (but leave 7) to reply "Bryan Hackney" <no@body.home> wrote in message news:09%4f.42365$Q53.10742@tornado.texas.rr.com...> > What are some suggestions for digitizing 16 bits @ up to about 2G/s? The > resolution and the rate are not hard requirements - I'm just trying to > figure out how to do high rate digitizing with non-exotic parts if possible. > > I had assumed that many A/Ds with serial out could be paralled, and the > outputs > pipelined, but this is not looking promising. Almost all the fast A/Ds are > parallel out, which makes sense, but these parts are also spendy. I've looked > at TI and Analog, and the 200 M/s parts tend to be in the $50 range, and of > course are quite large. > > BGA is OK. Do I get 10 or so of these 200 M/s parts and parallel them, or are > there other ways? > > This project is currently in the hobby domain. This will probably start off > as a spectrum analyzer. I plan on doing an FPGA-centric design, so interfacing > many dozen pins may be an option.
Reply by ●October 19, 20052005-10-19
> Say baseband is audio. Carrier is fc. Sampling rate can be fc/2. The carrier is > aliased to fc/4, and baseband can be extracted by removing the aliased carrier > from the digitized stream. > > What can prevent signals modulated at fc/2 and fc/4 showing up? If this requires > an analog bandpass before digitizing, why not just downconvert? It seems that > a hybrid downconverter is more flexible. > > BTW, thanks to all. I'm taking a crash course in undersampling.In this situation would the carrier not be aliased down to baseband (ie fc-(n*fc/2) = 0Hz, for n = 2) which would cause some problems with the spectrum folding back. You'd need to use quadrature sampling (in the above example), or alternatively sample slightly slower such that the carrier + modulation bandwidth has its lowest frequency component at 2*fs. This would mean that the entire modulated spectrum would occupy from 0Hz up to B Hz (where B is the modulation bandwidth). fs would still have to be greater than 2*B of course.
