Large DC bias on single supply non-ground referenced op-amp

[profdc9]

I have a problem with cascaded op-amp amplification stages that are operated by a single supply and are referenced to a voltage in between the two supplies.  Please see the attached schematic for details, in particular the four op-amps on the lower left.

The PCM2900 codec references its audio signals to half the audio supply voltage, which in this case is about 1.66 volts.  I am trying to design a microphone amplifier to amplify the signal of a dynamic microphone (in the millivolt range) to a range for the ADC input of the PCM2900 codec (+/- 1V approximately about the reference voltage).

I have cascaded two inverting amplifiers with the positive input connected to the reference supply through 100 kohms with local capacitor bypass (1 uF).  The first stage inverting op-amp, depending on how it is jumpered, can be either a gain of approximately unity or 22.  The second stage op-amp can be unity gain or a gain of 5, so that an overall gain of 100 (40 dB) is achieved in the op-amps are used together.

Because both stages are referenced to the same voltage, I did not think that a decoupling capacitor is needed between the stages.  However, I have found the first stage produces a very large DC offset from the reference voltage (0.1 to 0.4 volts) which when amplified by the second stage exceeds the common-mode range of the PCM2900 input.

The input bias current of the OPA2170 op-amp is typically 8-15 pA and the offset voltage is 0.25 mV.  These numbers are small enough that I do not think they explain the large output voltage offset.

When operating an inverting amplifier with a single supply with the positive input at an intermediate voltage with the input AC coupled, I would think that ideally the input coupling capacitor would pass no DC current.  If the input bias current at the negative terminal is very small, then the feedback current between the output and negative input is likewise small.  Therefore the negative terminal and the output must be the same voltage.  Since any difference between the negative and positive terminals would result in a change in the output, the equilibirum point would be such that the output would equal the positive terminal.

So if I short the input of the amplifier (jumper JP9), the output is 1.9 V, or about 0.25 V above the 1.66 V reference.  At the same time, the positive terminal is at the reference, and the negative terminal is 20 mV below the reference, so it seems to be correctly amplifying the 20 mV difference to 250 mV output above the reference.  But what could explain the 20 mV difference?  The worst-case input bias current is 3.5 nA, so for 100 kohms feedback, that is 0.35 mV shift between the inputs.  Similarly, the worst-case input offset voltage is 2 mV.  So the 20 mV difference is an order of magnitude more than be explained even by the worst case.

The capacitor C4 is a 10 uF/16 V tantalum capacitor.  Could the DC leakage through this capacitor explain the problem?  I wanted to use this capacitor because I thought it would produce less distortion than a MLCC 10 uF capacitor.  To explain the 20 mV offset, the leakage would be 200 nA.  Is this high or low for a tantalum capacitor?  If so, what kind of capacitor should be used as a decoupling capacitor for audio?  A film capacitor of this capacitance would be very large.

Any insight is appreciated.  Thanks!

[RandallMcRee]

For troubleshooting purposes you can certainly replace the 10uF tantalum with a, say, 1uF film.

[KT88]

Ther are at least two possible problems:
1. The common-mode rage includes V- (GND) but goes only to V+ -2V which gives you an upper limit of 3V.
2. The open loop output resistance is specified with 900 Ohms. No Vs given at what voltage this value was measured - likely at the highest Vs. At 5V this resistance (impedance) would be probably higher. The load impedance would attenuate your signal especially at the upper rail, causing an imbalance/offset...

Cheers

Andreas