Differential vs. Single-Ended amplification for sensor sampling

Question

I'm designing a load cell amplifier that should be low-cost, and functional in a noisy environment. The load cell is a 1k balanced full-bridge on a single package. The read speed requirement is only 1kHz, and the gain is expected to be ~1k. Here's my current circuit (passive values not final, just an example):

In a typical two-stage instrumentation amplifier, the amplification stage occurs first, as per the picture I've provided, and that differential signal is then delivered to a biased differential amplifier to deliver the final, single-ended signal.

My question -

If the signal AMP_LC- and AMP_LC+ were instead routed directly to a differential SAR ADC input, what potentional losses can occur, as opposed to the differential stage amplifier that converts to single-ended? And why would an instrumentation amplifier be generally preferred to produce single-ended outputs, given the benefits of differential signaling design? Is it likely for my application that cutting the 'differential stage' and 'bias reference' circuits, and running the 'buffer stage' output directly into a differential SAR ADC input would produce a cheaper result that would have the benefit of greater CMRR and noise immunity over distance?

Edit:

Thank you to Chintalagiri Shashank and Scott Seidman for your help. I had wished for a cheap solution, as I'm using 36 load cell sensors in my work, and would like to see at least four separate amplified levels of each signal. 144 inamps doesn't make for a low-cost design :)

Now, understanding that attenuating the common-mode noise is valuable due to the low impedance output of the amplifier, I can compromise and use a separate instrumentation amplifier per sensor at ~32-64x gain and route these signals to a microcontroller's programmable gain amplifier for 1-32x amplification.

Answer 1

The differential configuration is much more immune to common-mode noise -- that is noise that appears on both leads of your sensor. Single-ended amplifiers will amplify that noise with the gain of your circuit.

We use the in-amp as our first stage to effectively attenuate the common mode noise. Also, the input impedance is high and the same for both inputs. In a "normal" diff amp, the input impedance is not the same on both leads, and that may do suboptimal things to your signal.

Once the common mode noise has been attenuated by the in amp by 70 dB or more, it's pretty safe to use single-ended techniques after that stage. I recommend a modest-gain in-amp stage, followed by high-pass filtering to remove any amplification of voltage input offsets, if your system can tolerate it, followed by a single ended amplification stage to reach 1000 gain.

Answer 2

The design you have will be extremely difficult to get right. You've got a large number of resistors that need to be perfectly matched. Even variations in temperature coefficients will result in long term drifts and errors.

I would strongly suggest getting an instrumentation amplifier to begin with. A reasonable Instrumentation amplifier will not be much more expensive than three op amps with specs good enough to handle a load cell's output signal. It will certainly be cheaper than using three opamps + low TC precision resistors used for all the resistors critical to setting gain and maintaining a good CMRR.

Differential signalling is indeed better, but only in some cases. This is not one of those cases. Rather, the things you'd need to do to support differential signalling will intruduce errors far greater than any common mode rejection benefits you might get.

I would recommend the following:

• make sure your inputs from the load cell to the amplifier input are short, and if more than a couple of inches, a twisted pair. This is your differential signal, and where you're most likely to pick up common mode noise.
• before your instrumentation amplifier, whether you make one or buy one, add a differential low pass filter restricting the signal to the frequency you're measuring at (RFI filter). This will prevent the possibility of amplifying any RF noise you've picked up, which otherwise gets rectified at the amplifier inputs and shows up as DC offsets.
• between your amplifier output and the ADC, try to keep your traces short and prevent large loop area from forming. Another stage of low pass filtering here may be added (Anti Aliasing Filter).
• if you really want differential ADC inputs, then you can use a differential inamp or make one using a regular Instrumentation amplifier and an opamp inverter. At this low frequency, I honestly don't see the benefit though.
• since the gain needed for this sort of thing is usually high, you would generally need two stages of amplification before the ADC. So check the GBW of the inamp or opamp you are using. The first stage would generally be an inamp, the second is usually just an opamp.
• make sure your ground is clean. If possible, keep your analog ground separate from any digital circuits.
• make sure your bridge excitation voltage and bias reference are clean.
• try to set your gains so that you make as much use of the ADCs FSR as possible.
• oversample and decimate as much as you can with the ADC you're using and the measurement speed you really need. Usually, with Sigma Delta ADCs which include OSD or similar within, I prefer to use the lowest sampling rate available on the ADC which provides the best ENOB. With SAR ADCs, if there is no digital filter included in the ADC, then I'd use the highest sample rate I can handle and average it down to what is really needed.

Answer 3

There's really two advantages of an instrumentation amplifier over an ADC:

• Noise rejection
• Signal integrity

In common mode, differential noise cancels out with each other. This can be beneficial for signals traveling over a long distance. This would also imply providing better impedance matching and overall gain accuracy, etc.

If you're hesitant about using an instrumentation amplifier, then you'll need to look for an ADC that has a higher CMRR.

Answer 4

Here is a very cheap strain gauge circuit. Cheap components and zero drift. Better linearity than the linearity of the load cell itself.

As for low noise? Well, we were using large numbers of them in a automotive environment and "automotive" tends to be very electrically noisy indeed. We did some noise reduction in s/w but most of the "noise" was mechanical - there was a lot of bumping around.

The microprocessor drives the bridge with the top at VCC and the bottom at 0V. The ADC takes a reading. The microprocessor drives the bridge with the top at 0V and the bottom at VCC. The ADC takes a reading. the difference between the readings is the the "output". Any drift in the (cheap) op amp cancels out.

The gain (and hence the feedback resistor) depends on the resistance of the strain gauge. Ours were 1k per arm I think. They generally have quite good tolerance. We initially calibrated every sensor+amplifier but eventually found it wasn't worth it - the "nominal value" was close enough.

We didn't have problems with changes in the the micro's supply voltage - each board had its own regulator.

It all worked way better than I expected it would.