22-bit DC voltmeter (in)sane?

[Beowolf]

I'm thinking of building a digital voltmeter using a MCP3550 22-bit sigma-delta ADC interfaced with a PIC and a LCD display.

How complex this could be?

I know that I need a stable reference voltage beside theese chips.

My consideration is about the noise.

Is it possible to build souch a device in DIY conditions ( only 2 sided PCB, limited resolution on routing), and not destroy the initial precision of the ADC by surrounding noise?

I have low expirience on ADC-s beyond 12-bits.

 

[Nigel Goodwin]

I'm thinking of building a digital voltmeter using a MC3550 22-bit sigma-delta ADC interfaced with a PIC and a LCD display.

How complex this could be?

I know that I need a stable reference voltage beside theese chips.

My consideration is about the noise.

Is it possible to build souch a device in DIY conditions ( only 2 sided PCB, limited resolution on routing), and not destroy the initial precision of the ADC by surrounding noise?

I have low expirience on ADC-s beyond 12-bits.

I would suggest it's a waste of time? - why do you think you need 22 bit resolution?, and how are you thinking of calibrating it to that degree of accuracy?. Even with 12 bit, you're likely to lose the lowest bits in noise, and have similar calibration problems (but no where near as bad).

 

[philba]

Let's think about it from the perspective of what you are measuring. Lets say you are using a 5V reference. A 22 bit ADC will have 2^22 steps to it so each step measures 1.2 microVolts. Now, consider what typical noise is going to be. a couple of millivolts? Maybe more. Also, even a linear regulator will have a few millivolts ripple.

Finally. think about what kind of accuracy you care about, 1%? .1%? .01%? Ignoring the above noise issue, a 12 bit ADC will give you .024% accuracy. With noise it's more in the .1 to .5% range. I'd spend more time figuring how to reduce the noise in your system and thus get more use of the bits you have rather than brute force throwing more bits at the problem.

Phil

 

[dknguyen]

Cosmic

Consider the accuracy required in the instruments used for investigating celestial phenomena, the big bang, and the like. Now consider that they use 24-bit ADCs.

I'm also pretty sure that such devices require cryogenic cooling to minimize the noise which would make the resolution irrelevent. So at room temperature, you can forget about it pretty much I think, unless you are working with massive voltages, in which case you then need special high-voltage ADCs and other equipment.

Plus, more bits means more processing power and time to interpret the info. Lower the bits and oversample.

My vote: INsane

 

[JimB]

After some searching and head scatching, I assume you mean the MCP3550.
(MC3550 is a line driver of some sort).

In terms of building a usable voltmeter with 22bit resolution, you must be crazy!

But...
For a compact project, where you will learn about:
PIC programming
Interfacing the (cant remember the type) output from the MCP3550
Problems with voltage references
Problems the voltage dividers (different input ranges)
Noise problems

It makes a lot more sense than some of the wacko ideas brought up on this board, (like trying to perfect the VHF super-regenerative receiver built on a plug in breadboard using random components and running off a worn out battery?).

Yes, give it a go and see how you get on.

JimB

 

[dknguyen]

True

True enough. In an absolute sense it's pretty crazy, but relatively, much crazier things have been floating around

Unlike the crazier ideas, applications do exist for this, but I think there are easier, better, more practical approaches unless you absolutely need 22-bits...

 

[Oznog]

It's not too crazy. I've worked with 24 bit Linear ADCs before. They have their own noise filters on the chip, but there are still noise considerations across the board. Double sided board should be fine. Make a ground plane under the chip itself.

They make nifty shunt meters. You can read a 200amp shunt with an accuracy down to a few milliamps.

Make sure you don't apply voltage >Vdd or <Vss to the ADC inputs. Other concerns- for reasons I still don't get, the accuracy of a SigmaDelta ADC gets wacky if there's a lot of capacitance on the input, particularly when combined with a series resistance between the measured voltage and the capacitor, just like an RC filter would do.

So using using an RC filter in front of the ADC inputs can actually greatly decrease the accuracy. IIRC the tech guy at Linear pointed out that an RC filter followed by a certain amount of resistance would separate the ADC input from the problematic capacitance and fix this.

I know how illogical that seems. It did not agree with my engineering sensibilities that input capacitance would cause inaccuracies, or that series resistance after the cap would restore accuracy. But the guy was quite knowledgeable and I'm taking his word for it because he knew a hell of a lot more about these SigDelts than myself.

 

[Beowolf]

My reasonig was that I should build metal casing around the ADC by making a ground plane of the top layer and pass the signals through vias to bottom plane where will be the digital part. Over the chip i thought to place thin copper casing, and solder it on, making it RF-proof (no input noise).
Of course, both signal wires shuuld be in schielded cables, similiar to those used in osciloscope probes..
As a reference, I could use a REF02, or you can think of a better one?

to Oznog;

Thanx .From all above, your reply was the most useful. Could you refer to some tech info on that RCR type filtering of input?
So far, I was sugested to allways use a RC filter on the input of a ADC, so I used 100k, 1µ for the filtering.

to others:
I know that it would be almouset impossible to acheve all the bits, but, how it's don in audio recording equipment??

And one other thing, whattype of protection do you suggest for the input? I Used diodes after serial resistor, drawing the current to VCC and VDD, but I think that their Ir would in this case do a lot of damage on accuracy

 

[Nigel Goodwin]

I know that it would be almouset impossible to acheve all the bits, but, how it's don in audio recording equipment??

CD is only 16 bit anyway, and as you're not measuring anything?, accuracy isn't that important, and as long as noise is below the noise floor of the rest of the system it doesn't matter at all.

Some systems use more bits, and reduce to 16 bit for mastering, but it's probably more posteuring than anything else?.

 

[Beowolf]

That's right.
I never assumed that i will achieve all the 22 bits. I the data sheet of the ADc is said that under best condition it is possible to have 21.9. I will be happy to have 18 bits.

 

[Nigel Goodwin]

That's right.
I never assumed that i will achieve all the 22 bits. I the data sheet of the ADc is said that under best condition it is possible to have 21.9. I will be happy to have 18 bits.

As has been asked in a number of the previous replies, WHAT are you hoping to do?.

 

[Beowolf]

Tu build precise DC DVM to measure voltages in 100, 50 and 5V range

 

[akg]

Tu build precise DC DVM to measure voltages in 100, 50 and 5V range

u may have gone through all the replays above. and u said precise..but how much accuracy u needed.?,

 

[Nigel Goodwin]

Tu build precise DC DVM to measure voltages in 100, 50 and 5V range

So how are you thinking of calibrating it?.

 

[akg]

So how are you thinking of calibrating it?.

may be using a 22bit DAC

 

[Beowolf]

I have acess to 6 1/2 digit DMM by HP.
I intend to use digital calibration ( no analog trimming, just use precision resistors to make voltage divider, and then the given count multiply by a 24-bit coeficient to match the reading of the HP.)

 

[Beowolf]

About the accuracy..
when I build it, I'll se what I'll get.

I postet thi question here to get some info, ideas, so I could make it better.

I you just want to make fun of me, please, skip this thread.

 

[Nigel Goodwin]

I have acess to 6 1/2 digit DMM by HP.
I intend to use digital calibration ( no analog trimming, just use precision resistors to make voltage divider, and then the given count multiply by a 24-bit coeficient to match the reading of the HP.)

Where are you going to find resistors of the required tolerance?, and assuming you could?, how are you going to afford them?.

Next, what are you using for a voltage reference?, is that accurate and stable to the required degree of precision?.

Is the HP recently calibrated by an approved laboratory?, and what is it's specified accuracy anyway?.

I doubt it's anywhere near the 0.00002% you're aiming for?

 

[Optikon]

My reasonig was that I should build metal casing around the ADC by making a ground plane of the top layer and pass the signals through vias to bottom plane where will be the digital part. Over the chip i thought to place thin copper casing, and solder it on, making it RF-proof (no input noise).
Of course, both signal wires shuuld be in schielded cables, similiar to those used in osciloscope probes..
As a reference, I could use a REF02, or you can think of a better one?

to Oznog;

Thanx .From all above, your reply was the most useful. Could you refer to some tech info on that RCR type filtering of input?
So far, I was sugested to allways use a RC filter on the input of a ADC, so I used 100k, 1µ for the filtering.

to others:
I know that it would be almouset impossible to acheve all the bits, but, how it's don in audio recording equipment??

And one other thing, whattype of protection do you suggest for the input? I Used diodes after serial resistor, drawing the current to VCC and VDD, but I think that their Ir would in this case do a lot of damage on accuracy

You'll have a tough time. Don't use Ref02. Too drifty.. too much 1/f noise. At 22 bits (& higher) input noise (RF or otherwise) isnt going to be your major problem. It will be all the components you use in your design.

Your input amplifier should be a low noise Jfet audio type.

 

[Oznog]

Ideally I don't think he wants to make a meter with 0.000000000000000000000000001% accuracy here. It's not even practical to display the digits thus pointless. Rather, just something that can read in the uV range.

For very low voltages, it's also practical to use an op amp. But it's not just noise, it's offset that causes the problems. You need a "chopper" op amp here like the LTC1053. That can measure accurately even far below uV range.

LTC2410 is a good, slow sig-delt ADC. They provide REALLY good support if you try to contact them.