Archived Forum

That looks quite an interesting set-up.

I appreciate the op. amp's usefulness for scaling and providing the required offset, but I'm surprised you found that the varying input impedance of the multiplexer has a significant effect. The data sheet recommends a source impedance of less than 10 kΩ so that the sample & hold capacitor, which is quite small, charges adequately quickly. And from that, I'd considered that with a source impedance of 10 kΩ or less, the measurement error from that source would be in the order of ½ LSB or less, i.e. it would not be significant. Your source impedance is 4.7 kΩ. Are you able to quantify the effect that lowering the driving impedance to a few tens of ohms had on your readings?

The first 1.5 ADC clocks (out of a total of 13 for a conversion) charge the sample-n-hold cap.  A typical ADC clock rate is 125kHz, so a conversion takes 104 usecs, the first 12 usecs of which is charging the cap.  I just did a spot check on a random ATmega2560 I had handy.  The trace below is the voltage at an ADC input pin being driven to 0V through a 10K R.  Immediately before initiating a conversion on this pin, I was doing a conversion on a different pin being driven to 5V, ensuring the cap was fully charged.  The 0V pin spiked to just over 600mV as it was selected, and then decayed back down to 0V about 3 usecs later, well within the 12 usecs allotted.  The conversion consistently returns 0.

I think that supports the information in the Atmel data sheet.

So it would seem that "the Arduino's impedance varies while sampling, in such a way that the voltage divider's output will be affected. This is because we used high value resistors..." is based on a misconception and there is no discernible effect in practice, and so there is unlikely to be a significant effect until the source impedance is well in excess of the 10 kΩ recommended maximum.

That was the assumption we were operating on, based on comments in other forums about the need for an opamp between a voltage divider and the analog input pin.   We modeled it in the software, and also with the online calculators but without the type of lab equipment you guys have it was hard to verify.  When I'm working in an area of knowledge outside my core competencies, I tend to rely on best practice methods and in researching this came across countless examples of opamps being used to fix the effects that sampling may have on a source.   Didn't to repeat incorrect information, we were just explaining our design process to the best of our abilities.  Thanks for the additional information and insight.

As you pointed out, there are additional benefits to the opamp that still justify its usage.  We are able to scale the values we are working with:  the 23v possible range of our battery state of charge happens in the upper 35% of what would otherwise be a 0-65v measurement.  With an opamp we can focus on the 23 volts that really matter and increase our resolution.   We can also offset these values so they better fit into the range of the Arduino's inputs.

Considering that the opamp only cost $2.65, it seemed like a safe route to go with some additional perks along the way.

>> edited for tone

Are those two benefits worth the inclusion of a $2.65 part?  

Absolutely.  It's just that a lot of the designs used around here do monitor signals with a source impedance of roughly 10K and with no op-amp between them and the ADC input, so if there were a problem with a 4.7K source impedance, people would rightly want to know more.

well in excess of the 10 kΩ recommended maximum

Well in excess might be a bit bold, but I think if you stick with the 10K recommendation you should be fine.  My test above was just performed on a single part, and the ADC internal series resistance can vary anywhere from 1K to 100K.

In my book, "in excess" means a few percent greater than but below about 20% greater, "Well in excess" means greater than that. "Several times" means greater than about 2.5 times.

So while, dBC, your result falls within my definition of "several times", I restrained myself to "well in excess". However, it's quite conceivable that at some combination of parameters (inter alia temperature, clock speed, supply voltage) 10 kΩ is a critical limiting value.

SolarMill: As you were making a PCB and providing a power supply for other reasons, then adding an op-amp was undoubtedly the best way to go. As dBC pointed out, others might have been confused because one of your reasons appears to have gone against established practice. But in many cases, the cost (I'm including battery life as a potential 'cost') or complications of including an op-amp and providing a power supply are unacceptable or unwarranted.

If you are trying to reduce power consumption (and probably even if you're not) remember to disable the digital input buffer on your analog input pins.  Feeding them Vcc/2 causes them to draw significant current as it means both transistors are on (at least partially) providing a path from Vcc to GND within the buffer (the current draw isn't through the input pin).  I've clocked it at about 800uA per pin on a 5V 2560 which potentially dwarfs anything saved by going with a high resistor divider.  See the DIDRn registers for details.

For the same reason, always enable the pull-ups on unused pins.  I've got a 2560 here that draws 62.5mA on the 5V rail when all inputs are configured as inputs (the power-up default) with nothing connected.  That drops to 42.6mA after enabling all pull-ups.   Those digital input buffers are at their best when you feed them 0V or Vcc; leave them floating (or feed them an analog signal) and they start to suck current.  

A few days after sending off the board design for printing I came across some comments in another thread about tying down the unused opamps to reduce noise and power consumption.   We are are using a dual opamp, so there's only one left over.  I'm still not sure about the best way to do this, though.  Do we just connect the unused pins to ground or do we need a resistor in there?  Why does connecting them to either 0v or Vcc have the same effect?  Is Vcc the same as Vcc-?  In our package, we have 3 pins left unconnected:  2out, 2in-, and 2in+.   What should we do with each of them?

The engineering intern who's been working on this design is on vacation and returns next week, but I'm trying to keep things moving along so please excuse all the questions.

Thanks for the help,

-Bert

"A few days after sending off the board design..."
It's always like that!

You connect the -input to the output, (turning into a voltage follower), you tie the +input to a defined in-range voltage (which could be + input of the other half - it looks to be as good a place as any, and a by-product of doing that is you have a buffered version of the divided-down battery voltage, should you want it.) That needs just two wire links on the underside of the PCB.

See http://www.maximintegrated.com/en/app-notes/index.mvp/id/1957

Thanks.  We just received the PCB boards in the mail and and already found a minor issue with them:  the pads for the DCDC supplies are flipped from where they should be (at the opposite edge of the part than where they belong).  Otherwise, they look great.   Since we're gonna have to get them remade anyway, we can incorporate those changes into the opamp while we're at it.  That said, any other thoughts about the design that could be improved or tweaked?  Our ultimate goal is a well designed open source module that anyone can make for monitoring their off-grid systems, complete with power supplies for Arduino and accesories (such as Hall effect current sensors).

The design is somewhat restricted because it does not provide you with galvanic isolation on the voltage, and as a consequence of that it requires the battery negative to be both earthed and the common GND connection for the Arduino. As long as that limitation is understood and is acceptable, and it is being used at low voltages only, that is probably OK. But if for any reason the battery negative is not earthed, then you have a potentially dangerous situation.

The pcb layout also restricts it to low battery voltages, and it will not be possible to use it - even leaving off the dc-dc converters - on higher voltages because of the minimal creepage distance between battery positive and GND (= battery negative).

It is not usual to have components at weird angles (R1, R2) and there's no need for it.

You might find it helpful to put the component values on the silk screen as well as their designations, and outside their profile; you want to show the LED polarity; and the + for C1 wants to go outside its profile too (outside so that you don't need to resort to an empty board to check values and orientation).

I've built a circuit to measure DC Voltage with full galvanic isolation using the HCPL-7520, and an Isolated DC-DC Converter to supply power to the isolated side, only pencil diags thus far, so will post more details later once I have done formal schematics. Also do isolated DC Current measuring with the ACS 758.

One has to measure mV from these devices so precision is required, and I used a Max 6126 for this.

Its taken a while.

I really look forward to seeing how this goes, when I first saw the Open Energy Monitor I just assumed it would have this capability.  Glad to see it will happen!  We have an off grid 48V system here as well.

I'm watching this too!  @SolarMill, thanks for sharing your progress.