Intrinsic 1/f noise in ADCs and DACs
in [Design]
|
From: Chieftain of the Carpet Crawlers on 8 Jul 2010 13:38 On Thu, 08 Jul 2010 09:33:23 -0700, Joerg <invalid(a)invalid.invalid> wrote: >I don't, but an off-the-cuff question: Can you modulate the input source >so the baseband information rides on a carrier of a few kilohoitzes to >get you out of the 1/f on the receive side? Called chopping?
From: Phil Hobbs on 8 Jul 2010 14:07 Joerg wrote: > Phil Hobbs wrote: >> Having spent my career trying to keep clear of the low baseband, I now >> find myself needing to do very precise measurements of acceleration at >> very low frequencies--like 1 nano g (1 microgal, 10**-8 m/s**2) per root >> hertz at frequencies from 10**-4 Hz to about 100 Hz. This is an >> interesting ride, and will be generating a few discussions here, I hope. >> >> [I bought myself a brass plaque for the wall that says, >> >> DC: The Final Frontier >> > > I know an RF guy who isn't into plaques but his would read "It's all > just jittery DC". > > >> ] >> >> My noise budget is currently dominated by the white noise of a 16-bit >> ADC (AD7699), running at 100 kHz to spread the noise out, and >> subsequently filtered. (We may add some high frequency dither if it >> turns out to be needed.) The DC levels of the signals can be anywhere >> in the ADC range, but any large changes will be very slow. I'm >> therefore looking at a subranging strategy, with a DAC providing an >> offset that gets subtracted off before digitizing, to allow the steps to >> be effectively 32x smaller, say. >> >> That sets up today's question: The low-frequency noise behaviour of >> most ADC and DAC circuits is dominated by the noise of the voltage >> reference, which is almost always really horrible. Using a ratiometric >> measurement I can get rid of this, ideally, so I'm left with the >> intrinsic 1/f noise of the ADC and DAC. >> >> Does anybody have any wisdom about the intrinsic 1/f noise of ADCs and >> DACs? >> > > I don't, but an off-the-cuff question: Can you modulate the input source > so the baseband information rides on a carrier of a few kilohoitzes to > get you out of the 1/f on the receive side? > I'd love to do that, and we'll probably have to eventually. It's a little difficult since it involves changing the sign of the acceleration periodically (i.e. chopping the sample), meaning mechanical motion, which brings in a whole lot of other issues. Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal ElectroOptical Innovations 55 Orchard Rd Briarcliff Manor NY 10510 845-480-2058 hobbs at electrooptical dot net http://electrooptical.net
From: Phil Hobbs on 8 Jul 2010 14:13 John Larkin wrote: > On Thu, 08 Jul 2010 12:22:52 -0400, Phil Hobbs > <pcdhSpamMeSenseless(a)electrooptical.net> wrote: > >> Having spent my career trying to keep clear of the low baseband, I now >> find myself needing to do very precise measurements of acceleration at >> very low frequencies--like 1 nano g (1 microgal, 10**-8 m/s**2) per root >> hertz at frequencies from 10**-4 Hz to about 100 Hz. This is an >> interesting ride, and will be generating a few discussions here, I hope. >> >> [I bought myself a brass plaque for the wall that says, >> >> DC: The Final Frontier >> >> ] >> >> My noise budget is currently dominated by the white noise of a 16-bit >> ADC (AD7699), running at 100 kHz to spread the noise out, and >> subsequently filtered. (We may add some high frequency dither if it >> turns out to be needed.) The DC levels of the signals can be anywhere >> in the ADC range, but any large changes will be very slow. I'm >> therefore looking at a subranging strategy, with a DAC providing an >> offset that gets subtracted off before digitizing, to allow the steps to >> be effectively 32x smaller, say. >> >> That sets up today's question: The low-frequency noise behaviour of >> most ADC and DAC circuits is dominated by the noise of the voltage >> reference, which is almost always really horrible. Using a ratiometric >> measurement I can get rid of this, ideally, so I'm left with the >> intrinsic 1/f noise of the ADC and DAC. >> >> Does anybody have any wisdom about the intrinsic 1/f noise of ADCs and DACs? > > > Actually, no. I do have the name of the guy at ADI who really knows > the AD7699, and he's been willing to go into the lab and and try stuff > when I had problems. > > The 7699 is a charge-balance part so should be pretty good. We're > seeing a couple of LSBs RMS noise out of ours, down to about 0.4 LSB > when digitally filtered at 1 KHz, averaged over a second. > > It's a 500 KHz, mux'd ADC. I wonder if digitizing another, reference > sort of channel, would provide any numbers that would be useful in > processing the real samples. > > Note that ADC power supply noise will be important. Even the > electrical quality of the SPI lines will start to matter at some > point... everything is on one chip. I wouldn't clock it at all while > it's busy digitizing; you can afford that at 100 KHz. > > Thinking about my Audi transmission, you could use three DACs for the > zoom offset: The current one, the next one up, and the next one down. > Lowpass the hell out of all three and switch between them as needed. > The next logical step would be to make your own multiple-output, say, > 16-step string dac, with just low-value resistors and a few of mux's, > to make the basic steps per above, and fine-tune them with low noise > dacs. > > Hey, make all three signals available (current zoomed signal, next one > up, next one down) and let the ADC mux switch between them all the > time. Then you can splice the crossovers perfectly. > > Analog Devices has a new 20-bit DAC that's designed for MRI gradient > driver apps. It probably has low 1/f noise, because that matters for > imaging. > > I've seen ghastly popcorn noise, tens of PPM, from supposedly > audiophool-quality CMOS dacs. That can be a nightmare. Some parts will > sit for hours and just occasionally make a few pops. > > > John > Good idea about the homemade DAC--we can use a MF resistor string and switch between them with a mux. With zero volts across the mux, there should be zero 1/f noise. An old thin-film-resistor DAC like a DAC08 might do more or less the same. This gizmo has 8 channels in a miniature package--it's on both sides of a 20mm diameter circular board, with a compound TIA for each channel. Not a lot of room for manoeuvre. Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal ElectroOptical Innovations 55 Orchard Rd Briarcliff Manor NY 10510 845-480-2058 hobbs at electrooptical dot net http://electrooptical.net
From: Spehro Pefhany on 8 Jul 2010 14:34 On Thu, 08 Jul 2010 09:44:18 -0700, John Larkin <jjlarkin(a)highNOTlandTHIStechnologyPART.com> wrote: >Analog Devices has a new 20-bit DAC that's designed for MRI gradient >driver apps. It probably has low 1/f noise, because that matters for >imaging. We'll see.. getting some in soon. Nice power supply range.
From: Phil Hobbs on 8 Jul 2010 14:34
Vladimir Vassilevsky wrote: > > > Phil Hobbs wrote: > >> Having spent my career trying to keep clear of the low baseband, I now >> find myself needing to do very precise measurements of acceleration at >> very low frequencies--like 1 nano g (1 microgal, 10**-8 m/s**2) per >> root hertz at frequencies from 10**-4 Hz to about 100 Hz. This is an >> interesting ride, and will be generating a few discussions here, I hope. > > I design seismic data acqusition equipment. The frequency range is about > the same as yours. > > http://www.abvolt.com/projects_and_solutions/data_acquisition_board.htm > > >> My noise budget is currently dominated by the white noise of a 16-bit >> ADC (AD7699), > > Junk. > >> running at 100 kHz to spread the noise out, and subsequently >> filtered. (We may add some high frequency dither if it >> turns out to be needed.) The DC levels of the signals can be anywhere >> in the ADC range, but any large changes will be very slow. I'm >> therefore looking at a subranging strategy, with a DAC providing an >> offset that gets subtracted off before digitizing, to allow the steps >> to be effectively 32x smaller, say. > > This is equvalent to highpass filter. There could be better ways for > doing that. With good ADC, there should be enough of linearity and > dynamic range to make the offset step unnecessary. > >> That sets up today's question: The low-frequency noise behaviour of >> most ADC and DAC circuits is dominated by the noise of the voltage >> reference, which is almost always really horrible. > > It depends. If the reference is external, you can fix the noise. > >> Using a ratiometric measurement I can get rid of this, ideally, so I'm >> left with the intrinsic 1/f noise of the ADC and DAC. > > There is no 1/F in the chopper stabilized ADCs. > >> Does anybody have any wisdom about the intrinsic 1/f noise of ADCs and >> DACs? > > Check ADS1282 from TI. This state of the art ADC has the noise floor of > ~5nv/root(Hz) all the way down to DC. Total dynamic range ~ 133dB > (yes, thus is true number). > > > > Vladimir Vassilevsky > DSP and Mixed Signal Design Consultant > http://www.abvolt.com Thanks, that's great info. I agree that the ADC isn't the greatest for the application. It does have to fit 8 channels' worth on a 20-mm diameter board, so a delta-sigma is a bit problematical. I've seen the ADS1282, which is a very nice part, but (a) it's too slow, and (b) we can't afford 8 of them per sensor anyway, either from a cost or board space point of view. We may be better off in the end using more than one board--we have a bit of space to play with in the axial direction. This sensor uses an interferometer rather than coils or piezos, so we start out with a sensitivity advantage--even with the AD7699, our sensitivity is quite adequate in the white noise region, so if the 1/f can be controlled, we should be in good shape. Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal ElectroOptical Innovations 55 Orchard Rd Briarcliff Manor NY 10510 845-480-2058 hobbs at electrooptical dot net http://electrooptical.net |