Improving Analog Input Readings into an Arduino - Circuit Design

Question

I am designing a SMPS (see previous question if curious) and need to provide the mcu with an output feedback voltage to dial in the duty cycle for closed loop control.

Using an Arduino (Leonardo in this case), I am always frustrated with the fluctuation of 'counts' returned from AnalogRead(). I have tried slowing the polling (I only want 50-100ms response), running average software filters, and other software band-aids. The purpose of this question is whether it can be solved in hardware instead so I can tinker and learn a bit.

Version 1 of my circuit looked like this:

The purpose of Resistor R8 was to limit current for the D4 5.1V Zener in case the analog input saw more than 5V so as to not damage the mcu pin. But because it was in series with the analog input, it would create a voltage drop and therefore skew results. As the load resistance varied (potentiometer) so did the voltage across R8 and it was hard to calibrate the analog input in software.

I then began learning about using an op-amp as a buffer instead which has "infinite" input resistance, and therefore whatever series resistor value was chosen would be magnitudes smaller and have minimal effect on the outcome.

Version 2 of my circuit looks like this:

The output is much more stable, but my question is whether I am 'over-doing' it with this circuit?. How do you calculate an RC time constant when the op-amp has infinite resistance? As stated above, I want my response time to be 50-100ms with 20-50mV peak to peak range. I feel like this is certainly doable... Any suggestions or tips and tricks would be suggested.

I am guessing the simplest solution would be a low pass filter with R = 100K and C = 150nF but figured this was a good learning opportunity to explore op-amps too.

PS: the 74LVC2G66 device is what I am using to enable and disable 10K pullup and 10K pulldown resistors for other experiments. When both switches are off, they ~should not effect the series resistance of the analog input.

Answer 1

I am always frustrated with the fluctuation of 'counts' returned from AnalogRead()

well, welcome to the world of analog signals and digitization: Noise is ubiquitous, and unless you do proper analog filtering, you'll get aliasing.

So, deal with it, I guess?

I have tried slowing the polling (I only want 50-100ms response), running average software filters, and other software band-aids.

This is very clearly an application for regular time-interval sampling, not polling. Set up your MCU to trigger ADC conversions using a timer. Don't poll – that introduces significant jitter into your process.

The arduino framework is actively hostile towards doing this correct – you need to work around it.

So, my first suggestion is to drop Arduino and simply consider your Arduino Leonardo a slightly strange evaluation board for an ATMega 32u4, and write your controller in bare-metal C
Honestly, the ATMega32u4 is not the microcontroller you'd want for a high-speed controller that might have to do math, but there could be worse choices. I'd personally go for one of the cheap STM32L1 that are optimized for motor- and power control jobs. Cheaper, too.

Once you have sorted out how to do accurate regular sampling, design an analog front-end filter, the anti-aliasing filter that restricts your analog signal's bandwidth to less than half the sampling rate.

You do need something like your U9 opamp to drive your ADC inputs – their input currents during sampling would otherwise degrade the measurement. However, your C16 is highly counter-productive; you need the opamp to drive the capacitive load of your ADC, don't make it any harder!

So typical signal flow would be

analog signal
---> opamp voltage follower/buffer
---> filter (active or passive)
---> opamp ADC buffer
---> ADC input

The opamp buffer before the filter makes sure the filter is driven with sufficient drive strength, the opamp buffer after makes sure the filter isn't loaded by the load of the ADC.

Now you get properly filtered, properly sampled signals; suddenly you can design the digital filters for exactly the purposes you want, and without having to deal with the sample time jitter you get when polling.

How do you calculate an RC time constant when the op-amp has infinite resistance?

This question makes no sense; the time constant of an RC low-pass filter only exists for an R through which a current flows through the C to ground. You're using R21, 1 kΩ, for that.

D6 is cute and all, but the moment D6 starts to conduct significant current, U10A would probably already have broken down!

Answer 2

The purpose of Resistor R8 was to limit current for the D4 5.1V Zener in case the analog input saw more than 5V so as to not damage the mcu pin. But because it was in series with the analog input, it would create a voltage drop and therefore skew results.

Analog inputs draw negligible leakage current when the ADC is not sampling. During sampling, they draw a small current spike to charge the current capacitor. In your case this current comes from the 100nF filter cap, and voltage drop on the 1k resistor will be negligible.

If you measure 0.1V across the 1k resistor, that's 100µA which is orders of magnitude too high.

Maybe you forgot to disable the micro's internal pullup/pulldown on the analog input, or the Zener diode has too much leakage.

I am guessing the simplest solution would be a low pass filter with R = 100K and C = 150nF but figured this was a good learning opportunity to explore op-amps too.

Every time the ADC samples, it draws a spike of current to charge its internal sampling capacitor. The minimum source impedance specified in the datasheet is calculated so the sampling capacitor charges fast enough to settle to <1LSB of the target voltage within the sampling time window. However if you put a filter cap on the input, then the source resistance is near zero, as the sampling capacitor charges directly from the cap. This also discharges your filter cap every time the ADC samples. For accurate measurement, if your source impedance is higher than the maximum specified in the datasheet, then the cap should be 2^N times higher than the ADC sampling cap (with N ADC bits) so when the ADC pulls charge from it, voltage on the sampling cap does not decrease more than 1LSB.

Also if your source impedance is higher than the maximum specified in the datasheet, sampling too frequently will discharge the filter cap before it has time to recharge via the source resistance, so you have to be careful about that.

Some micros have very accurate ADCs, others not so much. It's important to look at the ENOB (Effective Number of Bits) or SNR (Signal Noise Ratio) in the datasheet. If a 12-bit ADC has 9 ENOB (like many STM32) then you get 9 bits of ADC and 3 bits of noise on top for free.

Also the accuracy of your ADC is the accuracy of the voltage reference used by the ADC, because the digital output value is the ratio of input voltage and reference voltage. If the reference is VCC and it has some ripple on it due to varying supply current draw, then your ADC will lose accuracy. Solutions are a separate reference chip which can be a cheap LDO, or at least some decent filtering, or using the internal reference.

Arduino boards are made cheap, they use VCC as reference and use 2 layer board with no ground plane, and there are few ground pins on the connector. This means any current going in and out of the pins will have its return current in the single ground pin, which creates voltage drop. So the actual potential of "ground" as seen by the ADC inside the micro may be different from what you think "0V" is. A proper PCB where you know where "0V" actually is really helps.