Weighing is an impressive process. PPM accuracy with primitive gear.
A good load cell is amazing. A platform can be bolted to a cell off-center, but it measures the same no matter where you put the load on the platform. That alone is mind boggling.
Put weight A on the pan. Measure the weight. Put B, ditto. Put both, ditto. Do the math. The sum will be right to PPMs.
--
John Larkin Highland Technology, Inc
lunatic fringe electronics
jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com
No so much on this project, it's similar to seismic monitoring where you need all the dynamic range you can get. With this one there is a tight budget interms of cost, power and size.
Awful doesn't matter as long as it works well and is cheap. During some design reviews I am thinking about handing out sick bag just in case.
Most actually use more than twice that amount. That defeats the purpose because the whole idea behind the S&H is to use it for major power savings. Sample the sensor and then give the slow ADC all the time it needs to crunch the sample.
OK, imagine applying pulsed excitation to the xducer and *integrating* into a gated s/h. Do that for a while at low duty cycle, to average out vibration noise, and then let the ADC have the result.
I told you that it would be awful.
Or use a fast SAR ADC, pulse the system, average.
Are you ratiometric on the excitation supply?
--
John Larkin Highland Technology, Inc
lunatic fringe electronics
jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com
It would also still eat too much power. We need to run at about 60 samples/sec and if we'd pulse-average over a whole sample frame there wouldn't be much in power savings.
A single shot S&H could track for a millisecond or so, then hold. Requires very high input impedance opamps and that's a little challenge because the good ones still use around 100uA per amp (got two channels).
Some uC have SARs in them but not more than 16 bits and that's only for the more expensive class of uC. Fast SARs also tend to use a lot of power. We can't expend more than around 100uA while converting and next to nothing ( Are you ratiometric on the excitation supply?
What's tragic here is that there are some delta-sigma ADCs that have differential inputs and super PGAs, which means that they could connect directly to the bridge output with no diffamp or any of that nonsense. Adding the s/h suggests adding a diffamp too.
What's tempting is two series CMOS switches and two biggish caps to ground, between the bridge and the ADC. Cheap differential s/h. Charge injections cancel. [1] [2]
If you tare the thing often to find zero, it might work.
I'm grateful that this isn't my problem.
[1] As if!
[2] Pulsing the strain gage could wreck the ratiometric-ness of the system.
--
John Larkin Highland Technology, Inc
lunatic fringe electronics
jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com
If you could drive the sensor in bridge, you could sample + and - peaks separately for + and - A/D inputs. A dual MOSFET gate driver might do it. The S/H would take little power, and you'd get twice the signal voltage, helping with noise.
Fast, low-power, and cheap.
Do you need an op amp? You might be able to sample directly into a cap. The ADCs have pretty hi-Z inputs--dropping an op amp from the signal chain would save power, error, and cost.
Hmm. If we take 3mS as "low single digit" milliseconds, settling to 20 bits takes twenty time-constants, so that gives a max. time constant of 150uS. With a 1k source, that means a max. cap value of 150nF.
Doesn't seem horrible. 1pC injection is a 1-in-150,000 error on that. I can imagine a cheap buffer scheme to speed that up too, like ye olde PNP-NPN emitter-follower, with a gated power supply and a straight- through resistor that takes over after the buffer gets the sampling cap close.
The two-phase sampling scheme above partially removes the sampling error, too. If the samplers match, it converts their charge-injection offsets into a common mode signal, which the A/D rejects.
We have a diff delta-sigma. But according to the datasheet it spits out spikes and the main issue is the low input impedance, only 2.4M. That causes too much droop in the capacitor for a 22-bit system. So I'll be looking for a CMOS dual opamp (to keep it differential) that can live on less than the 100uA per amp that we use now in the opamps.
And a low charge-injection dual switch for sampling. Can be a few hundred ohms on resistance but that's a small market nowadays. Most are low ohms bus switches with lots of charge injection.
Well, almost :-)
The good thing is we do not need any longterm DC stability.
I love those kinds of problems. Keeps the old gray matter up there exercised or whatever is left after some local brewskys. Yesterday night we were here:
formatting link
Kate Gaffney was playing and singing. Good old Americana. Fresh beer made right in the community. Sometimes with country bands we can even dance there. Our dogs can attend as well, plenty of other dogs loose or on leashes just hanging out together. Patrons are mostly kayakers, horse riders and mountain bikers. Dress code none whatsoever, folks can even come in swim trunks. Life is good.
That's why I am looking for low charge injection dual switches. What makes the situation less convenient is that due to space constraints it has to be in a tiny DFN, SON or similar package.
The input impedance is actually charge injection as a function of diff input voltage. All it will do is change the gain a little, which you can cal out.
You can get analog switches with less than 1 pC injection. Again, you can cal that out. Mostly.
I was thinking about the power supply voltage (aka ADC reference) shifting as you power the xdcr on and off. That could be a lot worse than charge injection.
--
John Larkin Highland Technology, Inc
lunatic fringe electronics
jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com
It'll cause substantial droop. My main worries are that it possibly isn't constant over time and that there may be spikes coming out of the converter that also get integrated up. I'll buffer it.
Only found the 74HC2G66 as a low cost dual switch that works at 3.3V and has next to no power consumption. Maybe there are others but this one should do fine. Size also matters so we can't have something like a SD5400.
[...]
That will be heavily lowpassed to tide it over but the regulator is already fairly stiff. Also, that effect can be calibrated out because it'll be always the same step. The good thing is that DC almost doesn't matter in this application.
I don't see a charge injection number in the datasheet. Did you measure it?
Cheers
Phil Hobbs
--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics
160 North State Road #203
Briarcliff Manor NY 10510
hobbs at electrooptical dot net
http://electrooptical.net
Of course, Luther and Henry VIII weren't that hot on monasteries. ;)
Cheers
Phil Hobbs
--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics
160 North State Road #203
Briarcliff Manor NY 10510
hobbs at electrooptical dot net
http://electrooptical.net
Yes, but very expensive and more in the low Rdson range where the charge injection is higher again. Also, recent not so stellar experiences with their ADSP-CM40x series have made me much less enthusiastic about that company.
Yeah, but on that ARM processors they seem to have slapped together the usual Blackfin Peripherals and an ARM core, without testing much. Once you get into it then it turns out that much of the advertised stuff doesn't work well or requires some expensive RTOS to work at all. Probably some Outsourcia effort. That is not the old AD I know. Therefore, next time I will be more reserved.
I bet their DSP operation is a different group from the analog things, almost a different company. The semi processes will be completely different, too.
They are ANALOG Devices!
ADI makes very good analog parts, amps and references and converters. They keep stuff in stock and in production and provide pretty good support... or at least they try.
I think a lot of uP sort of stuff uses outsourced IP these days, things like ADCs, Ethenet, caches, DRAM controllers, PCIe, all that. So the seller sometimes can't support parts of their chips, because they didn't design it. The peripherials in the NXP ARMs are a good example. And the PCIe blocks in FPGAs.
The really bad thing about Analog Devices is their hideous new web site.
--
John Larkin Highland Technology, Inc
lunatic fringe electronics
jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.