Advise on processing low-level differential signals

[3141592]

I'm a bit lost on what would be the best way in dealing with a low-level differential signal (spanning from 10kHz to 100kHz) that's polluted with both common mode noise (10kHz-100kHz) and a high amplitude, low frequency (50Hz), differential noise. It should be amplified by about 60dB (1000x) for an ADC, but the differential LF noise should be dealt with before this step as it is about 10-100 times larger than the useful signal and would overdrive the amplifier.

I'm thinking of the following setups, but I'm not sure if these are good or what would provide the best SNR possible:

Differential receiver (based on AD797?) -> Sallen-Key high pass filter (AD797?) -> Amplifier

high pass filter on both lines -> Instrumentation Amplifier (AD8429?)

I'm a bit unsure if this belongs here or in the beginner section, sorry if I've made a mistake. I'm thankful for any advice you guys can give me.

[dom0]

Questions not answered:
- Source impedance
- Actual amplitude values

[3141592]

Questions not answered:
- Source impedance
- Actual amplitude values

Thank you for taking your time.

Source impedance is 150 Ohms.
The useful signal is in the 2mVpp range, the LF differential noise can range from 20mVpp to 200mVpp, CM noise is unknown ( depends on the enviroment and cable length [amplification at the source is not possible] ).

[dom0]

Try a differential first order high pass, fc maybe somewhere in the 2..5 kHz, directly at the input, differential amp (you probably want a fully integraded instrumentation amp to get to the required CMRR quickly) with gain of maybe just 10 or 20 or so (don't want to exceed output compliance now, do we?) to get maybe 2-10 Vpp at the output, then add a n-pole high/band pass as you see fit to remove the differential noise.
Since 50 Hz sounds like mains hum I wouldn't expect that all energy is contained within 50 Hz, but rather spread over 50, 100 and 150 Hz. So a notch filter is probably not an option to remove it. Add final gain/buffer stage as needed to comply with ADCs input requirements. Alternatively you can spread the gain over the filter stages by inserting a gain stage after every VCVS.

At 150 ? source impedance you'd want to stick something bipolar in the first stage to get overall noise down since higher current noise is tolerable.

[Kleinstein]

The first stage could be a differential amplification (e.g. 3-10 times) with limited bandwidth, much like the input stage of the classical  instrumentation amplifier, but with a capacitor in series with the gain determining resistor. Later stages are then less critical to noise, as the higher frequency differential signal is amplified much more than LF background and CM parts.

The high frequency somewhat limits the choice of fully integrated instrumentation amplifiers. It could also be a good idea to stay fully differential, as many ADC want a differential signal anyway.

How much residual "noise" can be tolerated also depends on the ADC that is planed. So likely the 50 Hz part only needs to me made smaller than the signal, not to waste too much of the ADC range. So a 100 times damping of the 50 Hz Signal could be sufficient - this does not need a really steep filter - something like 2 nd Order or just two 1.st Order stages may be enough.

[Marco]

Are there any good references which do the math to show the advantages of differential filtering?

Intuitively I have a hard time believing differential filtering -> single ended conversion, will do any better than single ended conversion -> single ended filtering ... there seems so much room for mismatches in the differential filter. My intuition would be to go single ended ASAP.

[dmills]

10KHz  as the lowest frequency of interest?
150 ohm source?
Big common mode noise?
Small signal?

Easy... Use a transformer, 10K is more then high enough to get well away from saturation effects even in something small, and you have to love the very high common mode impedance.

If you want an electronic input stage then have a careful look at the mic preamp chips from THAT corp, as well as their line receiver chips.

Regards, Dan.

[3141592]

there seems so much room for mismatches in the differential filter. My intuition would be to go single ended ASAP.

That's what I'm worried about too, if the components are not matched properly, wouldn't I waste the CMRR of the following InAmp? I guess I'll have to try and do some math as to how much CM noise would get through with certain component tolerances. On the other hand, converting to single ended is another noise contribution before the high gain stage.

The first stage could be a differential amplification (e.g. 3-10 times) with limited bandwidth, much like the input stage of the classical  instrumentation amplifier, but with a capacitor in series with the gain determining resistor. Later stages are then less critical to noise, as the higher frequency differential signal is amplified much more than LF background and CM parts.

The high frequency somewhat limits the choice of fully integrated instrumentation amplifiers. It could also be a good idea to stay fully differential, as many ADC want a differential signal anyway.

Sorry for the possibly very dumb question, but would your idea of a cap in series with the gain setting resistor work with an integrated InAmp? Like the AD8429? I'll try this method in a few days. The ADC in this case needs a single ended input.

Try a differential first order high pass, fc maybe somewhere in the 2..5 kHz, directly at the input, differential amp (you probably want a fully integraded instrumentation amp to get to the required CMRR quickly) with gain of maybe just 10 or 20 or so (don't want to exceed output compliance now, do we?) to get maybe 2-10 Vpp at the output, then add a n-pole high/band pass as you see fit to remove the differential noise.
Since 50 Hz sounds like mains hum I wouldn't expect that all energy is contained within 50 Hz, but rather spread over 50, 100 and 150 Hz. So a notch filter is probably not an option to remove it. Add final gain/buffer stage as needed to comply with ADCs input requirements. Alternatively you can spread the gain over the filter stages by inserting a gain stage after every VCVS.

At 150 ? source impedance you'd want to stick something bipolar in the first stage to get overall noise down since higher current noise is tolerable.

I'm looking at low voltage noise devices disregarding the current noise specs. So you would recommend a multi-stage design with the gain distributed across the stages? Wouldn't that also mean higher noise?

Easy... Use a transformer, 10K is more then high enough to get well away from saturation effects even in something small, and you have to love the very high common mode impedance.

If you want an electronic input stage then have a careful look at the mic preamp chips from THAT corp, as well as their line receiver chips.

Regards, Dan.

I have already checked their preamps out, it seems that they are hard to get in Europe and the SSM2019 is an alternative for the THAT 1510/1512.
Looking at the SSM2019 datasheet however left me with some questions, as it seems a lot less detailed than that of the AD8429. I'll think about using a transformer, I have no experience in using them in these kinda applications.


Thank you all for the answers and ideas! 

[Marco]

AD8429 has pretty poor CMRR at 100 kHz, compare that to the AD8428 ... ain't cheap though.

What is cheap is a 10 mH pulse transformer. They don't specify the usable bandwidth, but the inductance is just about right for the 10 kHz lower cut off ... throw in a 1 uF blocking capacitor, which will also do some preliminary filtering of the LF noise.

PS. just don't put too much signal near 1 kHz through it ...

[3141592]

What is cheap is a 10 mH pulse transformer.

Thanks, I'll definitely try this!