Driving a buffered SAR ADC

[SethGI]

Hi all,

I've been working on a 4 channel data acquisition system, and am designing around the analog devices LTC2357-16 ADC. It's a four-channel, SAR, simultaneous sampling (important for me), buffered ADC. I know in general SAR ADCs can be a bit tricky to drive, so my original plan included a buffer (LT6018) which is specifically marketed for driving SAR ADCs. However, since the ADC itself is buffered, can I get around that requirement?

For reference, here's my overall system plan:

piezoelectric hydrophone -> charge preamplifier (recommended by manufacturer to prevent any discrepancies in cable length from affecting system) -> active butterworth bandpass filter -> PGA (LTC6915, fairly important for my application) -> driver? -> ADC

My question is can I swip that last driver step due to the buffered ADC.

Thanks, and sorry if this is a super basic question!

-Seth

[Kleinstein]

The ADC internal  buffer should not need a special driver before the ADC.

The choice of PGA is add: the LTC6915 is a very slow INA with good CMRR - more something for low voltage DC of very low frequency. Its odd to combine it with a LTC2357 and than caring about simultaneous sampling.
I would more expect a faster PGA, with no special need for very good INA. Ideally one may get a fully differential one.
The ADC is also a odd choice here as it's specially made for a high input voltage range. The more normal ADCs are more like +-2 V input range.  Some ADCs also include some PGA function.

A charge amplifier can still get effected by cable capacitance and sound pick-up by the cable. At least a long cable will add to the noise. Ideally the first amplifier would be directly at the sensor, possibly powered through the same cable as the signal.

[SethGI]

Hi! Thanks for the reply. As to the charge amplifier, I know it doesn't totally eliminate the noise, but the manufacturer of the sensor recommended we use it. So, I'll oblige. I totally agree with the other feedback. What if I were to remove the PGA and buffer steps all in one and use ADAR7251 instead. It has a LNA followed by PGA, then samples at the desired rate. I would then switch my charge amplifier and filters to be fully differential for better CMRR.

Any thoughts on that approach? That ADC (ADAR7251) seems a bit harder to work with, but overall better. As I have like six months for this project, I'm fine with that. Plus, I'm much stronger on the digital side of things than analog so complications in digital-land I'm fine with.

Thanks again for your help and patience!

So it would be input -> differential charge amp -> differential 4th-order butterworth filter -> ADAR7251


 

The ADC internal  buffer should not need a special driver before the ADC.

The choice of PGA is add: the LTC6915 is a very slow INA with good CMRR - more something for low voltage DC of very low frequency. Its odd to combine it with a LTC2357 and than caring about simultaneous sampling.
I would more expect a faster PGA, with no special need for very good INA. Ideally one may get a fully differential one.
The ADC is also a odd choice here as it's specially made for a high input voltage range. The more normal ADCs are more like +-2 V input range.  Some ADCs also include some PGA function.

A charge amplifier can still get effected by cable capacitance and sound pick-up by the cable. At least a long cable will add to the noise. Ideally the first amplifier would be directly at the sensor, possibly powered through the same cable as the signal.

[Kleinstein]

The ADAR7251 is a sigma delta ADC and thus does not need much anti aliasing filtering at the input. For the frequency in question here (some 10-50 MHz) the filter could be just simple RC, maybe LC filter, if at all. Just the limited BW of the amplifiers may be enough.

Even with not so high sampling rate (e.g. 300 kSPS) the digital interface is quite fast. This requites a good board layout to keep interference away from the analog side. This also applies to all comparable ADCs - 4 fast 16 ADCs just produce a lot of data and thus need a fast interface.  Chances are the ADC would be coupled to an FPGA or maybe DSP, that can also produce EMI.

An important question is which frequency range is really of interest.

[ogden]

The ADAR7251 is a sigma delta ADC and thus does not need much anti aliasing filtering at the input.

I would put it this way: every ADC needs anti aliasing buffer at the input, not every ADC needs low impedance signal source.

[SethGI]

@kleinstein, before the ADC I'll be using a bandpass with 3db corners at 20 and 45khz. The reason for the filter is to make the digital processing a bit easier; we want to run fast and any steps which can be handled before the a/d conversion is a plus. The reason for the choise of 4th order butterworth (in 2 2nd order multiple feedback stages) is primarily for the good step response characteristics, quick enough rolloff, and no ripple in passband. The freuqency range that I really care about is 25-40khz. As suspected, I will be connecting to an FPGA. I haven't decided which yet, but probably one from Intel as I have the most experience with those.

The ADAR7251 is a sigma delta ADC and thus does not need much anti aliasing filtering at the input. For the frequency in question here (some 10-50 MHz) the filter could be just simple RC, maybe LC filter, if at all. Just the limited BW of the amplifiers may be enough.

Even with not so high sampling rate (e.g. 300 kSPS) the digital interface is quite fast. This requites a good board layout to keep interference away from the analog side. This also applies to all comparable ADCs - 4 fast 16 ADCs just produce a lot of data and thus need a fast interface.  Chances are the ADC would be coupled to an FPGA or maybe DSP, that can also produce EMI.

An important question is which frequency range is really of interest.

[SethGI]

Do I still need that anti-aliasing buffer with an ADC that has the LNA and PGA stages built in? If so, can I just use any old op-amp that's low noise and fast enough, or do I need something in particular? Or will my filter suffice?

I would put it this way: every ADC needs anti aliasing buffer at the input, not every ADC needs low impedance signal source.

[ogden]

Do I still need that anti-aliasing buffer with an ADC that has the LNA and PGA stages built in?

For ADC with LNA you do not need anti-aliasing buffer, you do need anti-aliasing filter. Passive filter is enough, but if you use active filter which will provide low impedance output - it is OK as well.

Anti-aliasing filter removes unneeded unwanted frequency components which can cause aliasing and add noise or even ruin ADC results. Buffer does buffering - for ADC's which does not have built-in buffer.

[Kleinstein]

The active filter for 20-45 kHz also acts as an anti aliasing filter.

For a SD ADC AA filtering is about really high frequencies, not the usual 1/2 the sampling frequency, but more like > 10 MHz.
I have not checked the DS. It is well possible that ADAR7251 with internal amplifier may not need an external AA filter at all, as chances are the internal amplifier is slow enough to also act as AA filter.

If filter response is critical one usually prefers digital filters. Even if the response is not ideal, the response is known and stable.

[SethGI]

OK that makes way more sense to me, I was confused by the previous comments. I didn't know how a buffer could do any anti-aliasing. So we'll call that a typo. Just to clarify, it looks like my approach of:

piezo -> fully differential charge amplifier -> active filter -> ADAR7251

is viable.

As to the doing the filtering digitally, I don't know if that would work. It's a pretty noisy enviornment and I want to make sure I'm isolating the frequencies I care about before amplification. So I'd prefer to keep a bandpass before the amplification step to make sure I'm making best use of the resolution of the ADC.

One final question. I heard someone mention that the output of a piezoelectric transducer is inherently differential. That doesn't make a ton of sense to me intuitively. So then the question becomes: Do I need to convert from single ended to differntial, or can I just amplify the transducer signal as if it is already fully differential. I've found examples of such conversion circuits online, but nothing which treats the sensor as producing a differential output, which makes me doubt even more that it's a real thing.

Here's what I was basing it on:
http://tesi.cab.unipd.it/40176/1/Thesis.pdf (notably pages 2 and 8 )

I attached an image with a sketch of the two ideas.

Again, thanks for your patience with my noobish questions.