Hallo Mak,

sounds as if you have a ground problem. It's important that you properly distinguish two separate grounds, the analog ground and the digital one. Otherwise digital ground return currents can totally erode precision of ADC.

I would start with something like that:



On the right you see the digital section with 89C52. According digital ground lines are marked red. These ground lines are contaminated with digital noise. Between two separated digital ground points ground noise of several hundred millivolts can be observed, depending, of course, on speed of used digital logic chips and '51 derivative and depending on accomplished ground routing.

Make no mistake: The ground connections, whether analog or digital, are not meant to be realized by the help of thin wires, as it could be misinterpreted by looking at the simplied schematic. I assume the use of solid ground planes, an analog one and digital one. But they must not be intermixed, arbitrarily. Both ground planes must be separated by at least 3mm. The only point, where both ground planes are connected is the central ground point at VSS pin of MCP3201. The analog ground plane covers the left side of schematic including VREF, +IN, -IN and VSS pins, while digital ground plane covers the right side of schematic including VDD, CLK, DOUT, #CS/SH and VSS pin. Again, the overlap is only made at VSS pin!!
Connect to this point the 0V terminal of +5V supply.

MCP3201 has an advantegous feature: Differential inputs allow to remove common mode noise. So, connect -IN input as shown in the schematic. Then, voltage drop of digital ground noise running through VSS pin to digital ground plane can be removed from input signal to a high degree. Without this feature, digital ground noise through VSS pin would be added to input signal, which would result in a decrease of precision.

Unfortunately, using separate grounds is not enough, if high resolution ADCs are used. Especially if analog circuitry is derived from digital +5V supply, consequent filtering of analog supply voltage is very helpful. This is done not only for the two operational amplifiers, but also for the ADC itself!
The major filtering element is the ferrite choke. Use a type providing at least 100Ohm at 1MHz and showing a maximum of impedance at about 100MHz of about 1kOhm. There are some nice SMD ferrite beads on the market, showing very good performance. But if they are not available, take a 6 hole RF ferrite choke. This gives a very nice dampening.

Unfortunately, big ferrite chokes can introduce heavy ringing, when not used properly. RLC filter do not show a peak in frequency response, means they do not show ringing, if the following relation is valid:

R > SQRT(2 x L / C)

If we assume about L = 20µH for a common 6 hole choke, and choose C rather high, namely C = 10µF, then formula yields:

R > SQRT(2 x 20µH / 10µF) = 2Ohm

So, if we use R > 2Ohm we will not suffer from ringing.

This filter does not only heavily dampen digital noise on supply voltage, but it also helps to suppress noisy ground return currents running over the analog ground plane! Assume a fast digital glitch is present on +5V. Then, without ferrite choke a current would be caused to flow through supply voltage decoupling capacitors at analog side of I = C x dU/dt. If dU/dt is rather high, what must be expected for a digital glitch, then a high spike current will flow through these decoupling capacitors and along analog ground plane back to VSS pin of MCP3201. As consequence also the analog ground plane would be contaminated with digital noise, and the caraful separating of analog ground and digital ground would no longer make sense.
Here the ferrite choke helps. Due to its high impedance at high frequencies noisy digital currents are blocked and will not contaminate analog ground plane.

Of course, the benefit of this ferrite choke is not infinitely. A certain, but very small amount of digital noise will always be present on analog ground plane. That's why it is so important to use a solid ground plane also for the analog ground connections: Only by the ultra low inductivity of a solid ground plane voltage drops of residual parasitic digital currents can be kept at tolerable minimum.

To the both operational amplifiers:
The upper one shall symbolize a precision voltage source with low source impedance. A special precision voltage source, which can directly at output be decoupled by an additional capacitance, is surely the better solution.
The lower operational amplifier shall symbolize the input amplifier, also providing low source impedance. The input filter is a symbol of an antialiasing filter. This does not at all mean, that I recommend you this design as antialiasing filter. Choose an appropriate one, perhaps the one shown in datasheet.
The simplified schematic shown here shall only demonstrate the grounding and filtering philosophy, but not give a recommendation about certain reference voltage source or antialiasing filter!

By the way, also the digital section would benefit from RLC filters like them, shown for the analog section. But in order to minimize costs, they are seldom found, here.
Advantage of such digital filters would be, that the digital +5V supply voltage, outside the circuitry, and especially the ground return currents were no longer be contaminated with noisy digital currents. This can be very helpful, if the whole circuitry must be connected to a metal chassis. This connection should be made near the central ground point, means near the VSS pin of MCP3201. And the filter can help you to move this point a bit into the analog ground plane. By this, common mode noise of analog ground (which becomes the cable screen of analog input cable!) relative to chassis would be minimal and by this possibility of radiation of analog cable is highly minimized.
If on the other hand, the chassis connection is made 'deep' in the digital section, analog section will show excessive common mode noise, with or without digital supply voltage filters. Analog cable would act like an antenna, emitting all this common mode noise...


Good luck,
Kai