ADC limitations for bandpass/IF sampling

"Joerg" <notthisjoergsch@removethispacbell.net> wrote in message
news:qJ_nf.33643$q%.17135@newssvr12.news.prodigy.com...
> Hello Steve,
>
> > This form of polyphase sampling is much loved by those of a masochistic
> > tendency, for the pain and misery it can cause. :-) It is a nightmare to
> > get multiple ADCs like this to track over time and temperatures.
> >
>
> Nah, there are situations where you simply have to do it or else the
> project would die. Done it a few times.
>
> The trick is not to rely on any manual alignment whatsoever but to run a
> little uC alongside and then provide auto-adjust circuitry for all three
> variables for all ADCs minus one. Gain, offset and clock phase. It's not
> really rocket science. On one project they parked a big TI DSP just for
> that purpose. In the end its 'employment rate' hovered around a percent
> or two, almost idle. The cost for the analog parts and the DACs to set
> the trim levels was about $15 or so, peanuts compared to the cost of one
> of the ADCs. This was not a one-off project but it remained in
> production for about seven years until a faster and cheaper ADC type
> came out where we could get it done without multi-phase.

Neat! I know of a few who've done this although I'm not masochistic enough
(sounds like you are :-))
Sometimes there are no other options and these kinds of approaches are
warranged - but in the case of the OP, I'd bet there are several viable
ones.

Cheers
Bhaskar


> Regards, Joerg
>
> http://www.analogconsultants.com

"Joerg" <notthisjoergsch@removethispacbell.net> wrote in message
news:qJ_nf.33643$q%.17135@newssvr12.news.prodigy.com...
> Hello Steve,
>
> > This form of polyphase sampling is much loved by those of a masochistic
> > tendency, for the pain and misery it can cause. :-) It is a nightmare to
> > get multiple ADCs like this to track over time and temperatures.
> >
>
> Nah, there are situations where you simply have to do it or else the
> project would die. Done it a few times.
>
> The trick is not to rely on any manual alignment whatsoever but to run a
> little uC alongside and then provide auto-adjust circuitry for all three
> variables for all ADCs minus one. Gain, offset and clock phase. It's not
> really rocket science. On one project they parked a big TI DSP just for
> that purpose. In the end its 'employment rate' hovered around a percent
> or two, almost idle. The cost for the analog parts and the DACs to set
> the trim levels was about $15 or so, peanuts compared to the cost of one
> of the ADCs. This was not a one-off project but it remained in
> production for about seven years until a faster and cheaper ADC type
> came out where we could get it done without multi-phase.

Neat! I know of a few who've done this although I'm not masochistic enough
(sounds like you are :-))
Sometimes there are no other options and these kinds of approaches are
warranged - but in the case of the OP, I'd bet there are several viable
ones.

Cheers
Bhaskar


> Regards, Joerg
>
> http://www.analogconsultants.com

"Joerg" <notthisjoergsch@removethispacbell.net> wrote in message
news:qJ_nf.33643$q%.17135@newssvr12.news.prodigy.com...
> Hello Steve,
>
> > This form of polyphase sampling is much loved by those of a masochistic
> > tendency, for the pain and misery it can cause. :-) It is a nightmare to
> > get multiple ADCs like this to track over time and temperatures.
> >
>
> Nah, there are situations where you simply have to do it or else the
> project would die. Done it a few times.
>
> The trick is not to rely on any manual alignment whatsoever but to run a
> little uC alongside and then provide auto-adjust circuitry for all three
> variables for all ADCs minus one. Gain, offset and clock phase. It's not
> really rocket science. On one project they parked a big TI DSP just for
> that purpose. In the end its 'employment rate' hovered around a percent
> or two, almost idle. The cost for the analog parts and the DACs to set
> the trim levels was about $15 or so, peanuts compared to the cost of one
> of the ADCs. This was not a one-off project but it remained in
> production for about seven years until a faster and cheaper ADC type
> came out where we could get it done without multi-phase.

Neat! I know of a few who've done this although I'm not masochistic enough
(sounds like you are :-))
Sometimes there are no other options and these kinds of approaches are
warranged - but in the case of the OP, I'd bet there are several viable
ones.

Cheers
Bhaskar


> Regards, Joerg
>
> http://www.analogconsultants.com

Richard Owlett wrote:
> Jerry Avins wrote:
> 
>> Steve Underwood wrote:
>>
>>   ...
>>
>>> Everyone misses something, even if its only their lost youth. :-)
>>
>>
>>
>> Ooh! I have to remember that one!
>>
>> Jerry
> 
> 
> I was wonder to what Jerry was responding so I did a Google of comp.dsp 
> for Author=Underwood and found the relevant message.
> 
> My ISP provides a Supernews account, that message does not appear.
> 
> *HOWEVER* his 12/10 message "Re: Happy Hmas!" appears on Supernews, but 
> not on Google search ;}
> 
> Aside to Steve re his post
> Never fought battles you implied but reminded me of two aspects of "good 
> old days"
> -- chopper stabilized op amps
> -- multi ampere ground loops in audio racks of a radio station

Steve Underwood has no weirdness, but newsgroups generally do. :-)

Google groups usually has everything. However, other news servers 
typically seem to miss stuff. Sometimes a little, and sometimes a lot. I 
often go to Google groups to figure out what a thread was really about.

Steve

Hello Bhaskar,


> Neat! I know of a few who've done this although I'm not masochistic enough
> (sounds like you are :-))


I am not a masochist, at least I hope not. We just didn't have any other 
  choice back then to get to 12 bits. But I have to confess that I 
really like those challenges. And when some folks say that it cannot be 
done it's even more fun to get it done :-)

Regards, Joerg

http://www.analogconsultants.com

Joerg wrote:
> Hello Bhaskar,
> 
> 
>> Neat! I know of a few who've done this although I'm not masochistic 
>> enough
>> (sounds like you are :-))
> 
> 
> 
> I am not a masochist, at least I hope not. We just didn't have any other 
>  choice back then to get to 12 bits. But I have to confess that I really 
> like those challenges. And when some folks say that it cannot be done 
> it's even more fun to get it done :-)
> 
> Regards, Joerg
> 
> http://www.analogconsultants.com

Back in 1962 interviewed for a position at RCA Labs. One question was 
about making 10 watts of square wave at 100 MHz. All I could do was tick 
off all the tube types that couldn't do it and profess ignorance of any 
semiconductors that might. The interviewer agreed that those avenues 
were closed, but added, "I need it, so what should I do?" A said that 
the only way I could think of was to synthesize is by generating the 
components separately. I said it would be a dog to tweak the phases, but 
that it could be made stable enough in the lab. He told me "That's what 
we did" and I got the job. Sometimes the hard way is the only way.

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

Jerry Avins wrote:

> Joerg wrote:
>
>> Hello Bhaskar,
>>
>>
>>> Neat! I know of a few who've done this although I'm not masochistic 
>>> enough
>>> (sounds like you are :-))
>>
>>
>>
>>
>> I am not a masochist, at least I hope not. We just didn't have any 
>> other  choice back then to get to 12 bits. But I have to confess that 
>> I really like those challenges. And when some folks say that it 
>> cannot be done it's even more fun to get it done :-)
>>
>> Regards, Joerg
>>
>> http://www.analogconsultants.com
>
>
> Back in 1962 interviewed for a position at RCA Labs. One question was 
> about making 10 watts of square wave at 100 MHz. All I could do was 
> tick off all the tube types that couldn't do it and profess ignorance 
> of any semiconductors that might. The interviewer agreed that those 
> avenues were closed, but added, "I need it, so what should I do?" A 
> said that the only way I could think of was to synthesize is by 
> generating the components separately. I said it would be a dog to 
> tweak the phases, but that it could be made stable enough in the lab. 
> He told me "That's what we did" and I got the job. Sometimes the hard 
> way is the only way.
>
> Jerry

I remember getting to 12 bits at 3M samples/second in the 1970s, using 
four 750k sample/second Philbrick ADCs polyphased. In those days using a 
processor to manage them wasn't on (though there were dead times in 
their operation, every few seconds, where we could have done some kind 
of calibration). Being large bricks, with several chips inside, making 
them thermally track wasn't easy. Yeah, sometimes the hard way is the 
only way. However, it sucks up so much of your time trying to deal with 
this stuff, it seriously detracts from doing the interesting stuff with 
the digitised data. I've never been big on brute force solutions.

The next system used a form of pipelined ADC, with each chip handling 2 
bits of the final result. These were made by Quantel (the TV effects 
people). They were dramatically better (though still somewhat 
problematic) without really using devices that were inherently faster. 
Maybe the smart way was better than the hard way? :-)

Regards,
Steve

Jerry Avins wrote:
> Richard Owlett wrote:
> 
>> Bhaskar Thiagarajan wrote:
> 
> 
>>> You need to be sure that the analog bandwidth of the ADC is higher than
>>> 1MHz. You should find this as a parameter in the data sheet.
>>> Also, I'm assuming your BW of the RF signal is not much more than about
>>> 5kHz.
>>>
>>> Cheers
>>> Bhaskar
>>>
>>>
>>
>> I assume Bhaskar means at least the sample acquisition time is less 
>> than .5 microsecond [ That's how I would apply Nyquist to the 
>> situation. ]
> 
> 
> No. He means that the electronics between the analog input and the 
> sampler will pass 1 MHz without significant attenuation or phase shift. 
> (It's also important that the aperture uncertainty be as good as would 
> be needed to sample at the full (2 MHz) rate.)
> 
>> Given that assumption and assuming the 20kHz sampling rate is not 
>> locked to the 1 MHz carrier, I can not see what information could be 
>> obtained.
>>
>> Obviously I'm missing something.
> 
> 
> Yes. Read the chapter on sub-band sampling in Rick's book.

I assume you mean "Sampling Bandpass Signals" in Chapter 2.
I "understood" each of his math steps but I still miss big picture.
Math seems to clash with my concept of physical reality.
[part of reason I never got my EE degree? ;]


> (You located it on the shelf behind you.)

It moved next to recliner.

> 
> Jerry

Hello Steve,

> I remember getting to 12 bits at 3M samples/second in the 1970s, using 
> four 750k sample/second Philbrick ADCs polyphased. In those days using a 
> processor to manage them wasn't on (though there were dead times in 
> their operation, every few seconds, where we could have done some kind 
> of calibration). Being large bricks, with several chips inside, making 
> them thermally track wasn't easy. Yeah, sometimes the hard way is the 
> only way. However, it sucks up so much of your time trying to deal with 
> this stuff, it seriously detracts from doing the interesting stuff with 
> the digitised data. I've never been big on brute force solutions.
> 

We could have done it without a micro because all it did was regulate 
out the deviations while applying a test signal, one function at a time. 
So a three bit shift register and some analog "glue" would have worked. 
But we used a DSP because they had already parked one on the board, it 
was there when I came and they didn't want to take it back out. IIRC it 
already had one rather mundane job like a slow RS232 comm. Yawn. It was 
like using a Ferrari to go grocery shopping.

Regards, Joerg

http://www.analogconsultants.com

Richard Owlett wrote:

   ...

> I assume you mean "Sampling Bandpass Signals" in Chapter 2.
> I "understood" each of his math steps but I still miss big picture.
> Math seems to clash with my concept of physical reality.
> [part of reason I never got my EE degree? ;]

I'm with you there: math is useful to me for quantifying, but not very 
much for conceptualizing. I always have a niggling doubt about my 
understanding of things I can explain only with math. I think I 
understand sampling subband signals.

Shannon tells you that you need at a bit more than two samples per cycle 
to nail down a frequency component, and that the smaller that bit is, 
the more samples it takes for it to come clear.* Believe him. (He didn't 
need to tell me because I learned how to extract Fourier components by 
hand.) The penalty for not sampling fast enough is aliasing.

Aliasing isn't necessarily bad; there are times when it doesn't bother 
us at all. Consider a sampler running at 10 KHz. Frequencies in the 
sampled signal from 0 to 5 KHz will be sampled just fine. Those from 5 
to 10 KHz will alias and end up looking like they ran from 5 to 0 kHz -- 
an inversion. Frequencies from 10 to 15 KHz will look like (be aliases 
of) the frequencies from 0 to 5 KHz. A mess? Only maybe.

Normally, before the signal is sampled. it will be run through a lowpass 
filter to cut off everything above, say, 4.5 KHz, and the reconstruction 
filter will be similar. Suppose instead that we use bandpass filters 
that pass 5.5 to 9.5 KHz. Can you see that the signal /in that band/ 
will be faithfully reproduced? If "yes", I don't have to write more. If 
"no", tell me what eludes you.

   ...

Jerry
__________________________________________
* We're talking obtainable samples here. I'll give up on you if you 
argue from infinite precision just to be ornery.
-- 
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������