DC Scaling Question 2

[Noway2]

This question is related to the question I posted yesterday about the ADC converter application I am working on.

One of the items that I need to meter is the voltage input to the system, which is going to be 6 to 30Vdc. The ADC that I am considering has a much lower input range of 0 to 3Vdc. My initial thought would be to use an opamp circuit to create a 10:1 attenuator to scale the input. However, in this particular case I am measuring the supply voltage, which can vary as it is a dual battery + charger based system, this poses some problems trying to implement this circuit with opamps. Using a DC-DC converter to provide a higher supply voltage than I intend to measure would almost certainly exceed the cost budget.

As an alternative, I am considering using a resistive divider with .01% (1k and 9k) resistors and feeding this signal into an amp pronbably with a 3.3V supply. As the output will be between .6V and 3.0V, with the 3.3V remaing stable, I believe this should be a workable solution. Using .01% resistors, worst case I calcalate a max error of about .2%. While I can live with this error, I was wondering if anyone has a suggestion for a better alternative.

 

[Comcokid]

Do you really need to measure your system voltage so accurately? 0.01% resistors are hard to get. 1% resistors generally are fine for measuring a system supply voltage unless accurate measurement of the voltage is critical in some way.

I have used a resistive voltage divider many times on a extra A/D channel to measure system or battery voltage. You also need to watch A/D input impedance - if it is low just use a OP-amp voltage follower to buffer the divided voltage.

 

[VisiGoth]

Comcokid brings up a very good point. What are your requirments? FOr measuring voltage the basic questions are
1) How much can you load the point you are measuring? (i.e. Ohm)
2) What precision and what accuracy do you need.
3) What is the bandwidth?

 

[Noway2]

Comcokid,

I meant to type .1% resistors. You have a point about the cost, especially about .01% resistors. Since this is an SMT application, .1% resistors shouldn't be too difficult to obtain. I had planned to put a low output impedance amp circuit between the divider and the ADC, as part of the anti-aliasing filter portion.

I am faced with a bit of a problem as far as design accuracy requirements go and an understanding of the requirments is proving to be a real three edged sword. The end product is serviced by reps who tend to complain if the display doesn't match THEIR meter within about 1 unit. Unfortunately, the marketing department can't provide any requirements and if they do, they change with the wind or with the latest complaint.

While I am probably over designing the system a bit, I am doing so out of the fact that I am extremely tired of getting beat up over the poor accuracy of the existing product, that I didn't design. My goal is to design a product that is inherently accurate enough to satisfy the reps 'out of the box' without requiring field set up or calibration.

 

[Noway2]

VisiGoth,

I posted my previous reply before yours showed up... to answer your questions, the input will be in the 0 - 30Vdc range and I would like an output accuracy of .1Vdc for the display.

 

[logbook]

Recognise that a 0.1V error on a 0-30V range is actually a "linearity error" not a gain error. A gain error would be 0.1V at 30V , then 0.03V at 10V for example. Actaully you requirement will have both a linearity part and a gain part. You would not normally get this sort of accuracy cheaply when there are several factors to take into account. Three 0.1% factors would immedaitely use up your error budget.

Suppose you make a passive 10:1 divider using 0.1% resistors. this gives you a 0.2% uncertainty to start with. then your refernce for the ADC is unlikely to be that accurate anyway. The trick is to use relatively poor absolute accuracy parts and then use a gain trim, either manual or automatic.

Make sure you take the zero error into account as well.

 

[Noway2]

Thank you for the suggestion. I hadn't planned on including an adjustment circuit but I think you are correct. In this case, based on my experience with a similar product, I am certain that I would have to make an automatic one.

My first thought would be to switch in a known reference through a low Rds-on FET and then adjust the gain with an e-pot. I could probably use a similar tactic for the zeroing.

Does anybody have a better suggestion?

 

[logbook]

You can adjust the gain by a few percent using an adjustment on the ADC reference voltage. The DAC to adjust the reference should be cheaper than an e-pot.

A low resistance FET may not be a cheap option when you consider the control circuit as well. Ideally use a packaged CMOS switch and feed in to a high impedance input, so the switch resistance is not so critical. Crappy old CMOS switches have been used inside 7 1/2 digit DVM circuits when positioned correctly.

 

[Noway2]

Thank you for the suggestion. I will have to look into CMOS switches. I was looking at some analog muxes and finding that the input voltage range was limited to the supply voltage at the Vdd pin, which poses some problems trying to switch the potential 30Vdc input. It turns out I need some way to disable the analog inputs on powerup to make sure that the core and IO voltages are up and stabilized first, so some form of switch is going to be required.

Ultimately, the goal is to allow the system to self calibrate to an onboard reference(s) because I can't rely on the idea of having only calibrated boards going to the customer (been there, done that disaster).

My thinking, at the moment is to combine the calibration circuit with the 10:1 attenuator. I found a 2.5K epot(about $1) that I could use in a divider with a 9.1K. On powerup it is at midscale which is close to 10:1 factor. I could then switch in a known reference (another FET or switch) and ajust the epot until I hit the proper value on the ADC. I could then re-run this cal periodically to account for the wide range of temperatures that the system will be exposed to.

I am not certain if I would need to calibrate to multiple references and a way to compensate for the curve of the ADC or not, though.

 

[Madcow]

I've had good luck doing dc measurements with the ads1110 from TI.

http://focus.ti.com/lit/ds/symlink/ads1110.pdf

upto 8 on an i2c bus, nice and small, built in PGA (for those small signalls) , built in ref and selfcal. About 2$ each.

 

[logbook]

I’m confused. You want to measure 30V and can’t get a CMOS mux to switch this much signal. Why don’t you pot the signal down to 3V using a pair of resistors then MUX it? Are you worried about the load on the 30V signal?

Your idea of putting the epot as the bottom end of the 10:1 divider is horrible form a technical point of view. The resistance of the epot is probably horrible, so even if it is stable as a potentiometer, your circuit won’t be stable as a divider. If you are not too bothered about the loading effect of the epot on the signal being divided use a pair of resistors as the main divider. Put the epot in parallel with the real divider but stick a larger resistor in series with the epot "wiper". The epot has a very small effect on the overall output (as required) and it is much more stable with time and temperature.

 

[Noway2]

Madcow,

Thank You. I must admit from a quick glance at the datasheet, it looks like a nice part. Especially for the price! Right now, the application is still pretty much developing in my head and I am working with some prototype experiments, so I will keep this in mind going forward as the requirements and final architecture take shape.

Logbook,

The MUXes and CMOS switches I pulled up (digikey search engine) indicate that they have an analog input maximum of Vdd. I am not planning on using the 30V as a supply as I can't rely on it, but I need to measure the signal that could possibly be as high as 30V to determine if the equipment (that generates the signal is present and functional). At the same time, I can't guarentee that the 30V signal won't be applied without power being available at my circuit's Vdd inputs (which is one of the data sheet requirements) as the signals are external to my controller in the system. Consequently, I need to put some form of switch in line that I can use to ensure that the analog channels are off, prior to and during power up of my controller.

As far as the other portion goes, about putting the epot in parallel with a divider and putting a large resistor on the wiper... I almost get what you are suggesting. I am interpreting your description to mean that the wiper connects to the 'center' tap of the resistor divider via a large resistor to prevent loading the divider. From this description, though, I am having trouble "seeing" where the other ends of the epot go and where I would pick off the attenuated signal. I want to do a 10:1 divider, would I still do 10:1 with two resistors or would I divide this partially and then further divide it with the epot? I am not concerned about loading the 30V signal as it is a combination lead acid battery + charger output. Could you please explain further? I think what you are describing may be what I am after, once I fully grasp the description.

 

[logbook]

Hope these pictures help.

http://www.logbook.freeserve.co.uk/image/poorpot.bmp

http://www.logbook.freeserve.co.uk/image/goodpot.bmp

 

[logbook]

So you take the 30V signal and pot it down to 3V with a pair of resistors. The maximum voltage the MUX sees is therefore 3V. If the MUX has no power it may get upset so you make the divider resistors big enough such that the current into the protection diodes of the MUX is not more than say 1mA and the MUX is then protected.

 

[Noway2]

Logbook,

I really appreciate the pictures. All morning, I kept trying to think of a better way to implement the circuit than I had planned, which was the poorpot method, but just couldn't come up with anything.

The circuit makes sense now that I see it. You create a 10:1 divider with regular resistor and put a large, but adjustable, resistance in parallel with one half of the divider, allowing you to vary the attenuation factor. The output of this circuit, however, has a significantly lower senstivity to variations in the pot.

Thank you for your help!