EEVblog Electronics Community Forum
Electronics => Projects, Designs, and Technical Stuff => Topic started by: OM222O on November 30, 2018, 09:47:20 pm
-
Hello
I've designed a PCB using the ADS1219 (24 bit ADC) and the ADR440BRZ (2.048V Reference)
which are controlled using the Atmega 328P (Arduino)
I separated the digital and analog grounds as recommended by the data sheet (one side is digital only, one side is analog only, there's just a plated through hole connecting the two together)
I'm also experimenting with input filtering and whatnot which is why you see one side has a ton of RC filters while one side is directly connected to a header.
My main question is should I use star grounding (similar to what I did) or should I just use a ground plane for the analog section?
Pictures of the PCB are attached
-
You can kindof do star grounding with a flood fill, but it really only becomes important when you have multiple devices on the analog side (but do keep the digital ground on its own trace/plane until the board grounding point), so with just an ADC and a reference generator, it may not make much of a difference, but if you had an amplifier chain ahead of the ADC or a DAC onboard too, then it would become more beneficial.
For this sort of board, I always like the idea of an LC filter and some bulk capacitance on the power input. Even better than a simple inductor on the power rail is a common mode choke at the input - keeps out some of the EMI picked up on the power cable. Another easy and nice addition is local regulation, so if your digital and analog sections needed 5V, for example, you'd have a power input of 6V or more, then an LDO for the digital rail and an LDO for the analog rail. Further isolates the two sections from each other in terms of power noise and offers some noise rejection from the supply input as well.
The images are sort of small and I didn't take a close look, but do you have an input buffer for the ADC? A lot of high speed or high resolution ADCs like a close buffer amp to supply enough current to sample, and having one means you can hook up even very low supply current devices or sensors to your board - seems like the default for the ADS1219 includes one too, so it's good to have if it's not already there.
-
I do not have one indeed but I'm planning on using it as a DC voltmeter ... how much does the input op amp matters? I don't want to go to negative supply stuff and even rail to rail op amps have way too much offset for my taste
-
for example here there are no input op amps (just a current sense amplifier) and it seems to be ok?
-
Your right, I mistook that as just part of the standard circuit, from the input current specs, as long as below 5nA is fine for the device you're measuring, there's no need for an amp. If you wanted a very high input impedance, then you would still want one, but that's low enough as to not be an issue for most things you'd want to measure.
You've got that input filtering section and I expect it will be required for good noise performance, 1kS/s sample rate means your Nyquist frequency is 500Hz, so anything above that could cause aliasing, and more importantly, is extra noise in the system. Narrowing the input bandwidth with a low pass filter will improve noise performance - even a filter designed for 10s of kHz will likely make a notable improvement over nothing. The narrower the bandwidth of the part you're measuring, the lower noise.
-
As I mentioned before, it will be used for DC applications, so no worries about sampling frequency ... I will actually drop the output sample rate as well to bump up the ENOB ;D just one last question, what is a reasonable value for the resistor in the RC filter? 10 \$\Omega\$? 100 \$\Omega\$? 1K , 10K? :-//
-
The ADS1219 is an sigma delta ADC with internal buffers. So there is not much to worry about aliasing filtering. The SD ADC only needs aliasing filtering for >100 kHz for the internal sampling, so much less stringent and possibly the buffer is slow enough.
It would be still good to have some filtering at the input to get rid of possible RF interference. In addition the filter gives a little ESD protection. How much resistance is needed kind of depends on the desired protection and noise that is acceptable. I would consider some 10 K maybe 100 K (more noise, but better protection).
Protection of the inputs would be another issue to look at.
-
So you think I'll be fine with a ground plane and no input filtering? I'm afraid that the the resistor noise and temperature coefficient (the best metal foil resistors I could find had 25ppm temp co) throw off my measurements rather than help it ... I'm planning on using it on the high gains which have a resolution of better than 1uV! my desired accuracy is about 5uV so that leaves plenty of room for INL , thermal erros, etc
I'm still not sure which method is best. Also I was wondering if I should include an amplification step to correct for gain and offset errors or if I should take care of that in software?
-
Well given the simple RC filters you can make, and the high impedance of the input, and the want to keep such a filter to smaller SMD parts..... maybe a 100 ohm resistor and a 4.7uF cap? 100 ohms vs. the G Ohm or so worth of input impedance (5nA max input current, ~5V maximum input voltage) is so small as to be in the noise floor (20 bits gives you a million or so divisions, which would be closer to 1k), so you don't have to do any math to adjust the value by the resistor divider in the front, and any thermal drift is going to be small enough to be insignificant (at least for the filtering resistor...). Metal foil is expensive, so the more practical lower current noise choice is thin film (over thick film), or wirewound if you don't mind the inductance (and this application likely wouldn't). Also ends up being a half a millisecond or so RC time constant, so you should be able to read the actual value pretty soon after it happens, even at full sample rate.
The 100 Ohm 4.7uF should give you a -3dB point at 338Hz, and can probably still be done comfortably with 0805 and smaller parts. The cap will likely be microphonic, so you could go with a film cap or something instead, but as long as you can keep it away from impact/vibration/loud noise, the added capacitor noise should be pretty low.
At least, that would be my starting point. Even if you want to use it only near DC, probably worth testing with a sig gen just to see how it reacts at higher frequencies, could help you tweak the input filter to be appropriate for your application, too.
-
I would not make the filter so low impedance, as this would not provide protection and the time constant would depend on the source impedance, possibly way too long with a higher impedance source. Since it is an SD ADC there is no need to have the filter frequency so low - something like 10 K and 10 nF and thus 100 µs time constant. At 10 nF there are already G0C MLCCs available.
Depending on the use some extra protection could be a good idea - so not just the simple RC.
For the filter, there is nothing to worry about the TC of the resistor : it would only change the time constant. So for the filter it is more about not using a single 0805, but more like 2-4 in series to better protect against ESD. So resistors in the filter could be normal 100ppm/K thick film. Even carbon thin film would work - but they get rare.
Resistor TC will get important if there is an amplifier or divider to get other ranges than just some +- 2 to 5 V.
The concept with digital on one side and analog the opposite side is not good - one usually wants some distance and not just the 1.5 mm FR4. For the digital part a ground plane is probably OK, especially if it is really one side ground (and not much else) and the other side the main circuit. Just a ground fill an hope to get the ground connected is about the worst solution - it may still work digital, but it could still fail and produce EMI problems.
The analog part should be more like in a different corner / half of the board and better using a star ground than a ground plane. However with only a simple RC filter there is not much circuit anyway.
-
I'd use a ground plane. Partial grounds can have currents running where you can't predict them. If the digital supply is filtered properly then it won't end up in the analog section (remember current flows in circles so if it can't flow in the supply, then it can't flow in the ground). A good trick is to use series resistors in the digital lines towards the ADC. Usually high resolutions ADCs have differential inputs. Connect the - from the input as close to the 0V of the voltage you are measuring so you basically do end up with a differential input.
-
Interesting, especially the Figures 10,11,12..
https://www.maximintegrated.com/en/app-notes/index.mvp/id/5450 (https://www.maximintegrated.com/en/app-notes/index.mvp/id/5450)
-
input protection doesn't matter here as I'm careful with using the device. I have SMD resistors from 100 ohms to 10K as well as 4.7u tantalums so I can experiment with the filter value (or if one is needed at all). the analog section is also fairly small so I'm guessing a ground plane would be good enough in this case?
Someone mentioned it's not a good idea to keep analog and digital sections on different sides of the PCB ... The manufacturers suggest that you keep them separate ... besides it acts as a natural star ground and I don't need to worry about digital return paths anymore. Also the final version would be a 4 layer board so there will plenty of decoupling provided between the signals. Currently I'm using a 47u tantalum right on the input section + lots of 100n local decoupling caps so I'm not really worried about supply noise either. I'm not sure why there is such a huge disagreement on different methods here but this thread has confused me even more XD I think I'll just test things for myself and find out.
-
Well typically with analog/digital planes, you split the pour with a small gap and then run the two under the chip, just keeping the input side on the analog plane and the output side on the digital - so not really two separate full planes, but one plane area split into analog and digital sections.
You may say you're not worried about power supply noise, but if you want 20 significant bits, you definitely will be ;D For a quantification, a million or so divisions (20 bits worth) at 0-5V full scale is just 5uV per division... and a decent linear lab power supply like the Rigol DP832 is specified for 350uV RMS and 2mVpp... meaning the peak to peak noise if coming from the power supply could lower your performance to about 13 significant bits. Of course there's going to be some power supply rejection inherent to the ADC and some benefit gained from the reference voltage and low sample rate and using averaging.... but having an extra 30+dB worth of power supply noise rejection onboard could go a long way to actually getting your usable resolution close to the spec of the ADC.
My advice would be even if you don't want a common mode choke on the input, extra bulk capacitance, and local regulation for the analog section (if not both), add in the footprints, have a couple jumpers to bypass them, and build the boards like that. Then if your first design without the extra power filtering stuff doesn't hit your performance target, you can add in the LDO or other filtering parts to see if they offer better performance. I suspect they'll be needed, but doing the experiment yourself has its own value too.
-
For the signal filtering I would not use tantalum caps, as they have quite some possible leakage and dielectric absorption. So settling might take long. The better choice are film caps or NP0/COG ceramic capacitors. Unless the signal source is known low impedance, there is not much use in a large cap at the filter. It would be more about maybe having a two stage filter, to better suppress the really high frequencies from the RF range (e.g. GSM, WLAN).
A tantalum cap might be OK for reference filtering / buffering.
For a 4 layer board, one can have the analog and digital part directly on different sides, just have one closed layer of ground in between.
For the ground it is also important where the ground is connected to the outside world.
Power supply noise can become important, as the PSRR may no be that good. A common mode choke might be a good idea - this could be for the input (and corresponding ground), but also for the supply and control.
At the low frequency 20 bit resolution is not as difficult as it sounds. If possible chose an aperture time (possibly manual averaging) that is a multiple of a power line cycle to suppress mains hum an related.
-
I meant the noise between the digital and analog sections. For the power input I will be using a huge capacitance multiplier circuit (2 stages, each using a 47u tantalum and a 10k resistor) as well as two different LDOs (LM2940CT-5.0/LF01 for the analog section and a much cheaper one for the digital section ... I can't remember the exact part number but I don't think it matters tbh).There are more 47u tantalums on the outputs as well. this is used to measure extremely small voltage drops across shunt resistors (10uV accuracy is required for my application which is why I said 5uV error would be acceptable). I know the first thing that comes to mind is a current sense amplifier, but I want to use it over a wide range of resistor values using the PGA and a switchable current source (1mA/10mA/100mA/1A) so a current sense amplifier is not a good choice in my case.
Why do you say it's bad to have them on entirely different planes and not just a slot in the middle? it should be more effective if anything.