Author Topic: Low Voltage Signal Conditioning Theory Question  (Read 2438 times)

0 Members and 1 Guest are viewing this topic.

Offline Evan.CornellTopic starter

  • Regular Contributor
  • *
  • Posts: 189
  • Country: us
Low Voltage Signal Conditioning Theory Question
« on: December 23, 2019, 09:29:00 pm »
If I have an amplifier circuit (say ADA4945-1, with Input Noise Figure of 1.8 nV/√Hz, f = 100 kHz) and an amplifier bandwidth of 200kHz, that implies total noise level of 1.8nV*sqrt(200k)=805nV.

Does that mean that my desired input signal must be greater than that noise level by enough to give me desired ENOB? I have seen posted elsewhere 6dB per ENOB as a rule of thumb.

So for instance, if I need to digitize the input signal to ADA4945-1 at ENOB=5, then my input signal ought to be at least 805nV*10^(5*6/20)=25.46uV.

Furthermore, if the signal I need to digitize does not use that entire bandwidth, say it's only 10kHz wide, somewhere within the 200kHz passband, and I can digitally bandpass filter the digitized signal, then my minimum input signal level would decrease to 1.8nV*sqrt(10k)*10^(5*6/20)=5.69uV.

Are these lines of reasoning correct?
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 22436
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: Low Voltage Signal Conditioning Theory Question
« Reply #1 on: December 23, 2019, 10:00:41 pm »
Yes, as long as the quantization noise is evenly distributed (roughly, that the noise + signal is larger than a few LSB so that dithering occurs naturally) and you have enough extra bits in the DSP filters (so that SNR isn't lost to rounding).

Note also that, if the amplifier's noise bandwidth is more than 200kHz, and the ADC's sampling aperture or analog bandwidth is more than 200kHz, then noise beyond the desired bandwidth can be aliased into the samples.  Design the filter so that the noise at the ADC input is within this limit (or, use an ADC that isn't susceptible to aliasing, such as an integrating or S-D type).

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Online Marco

  • Super Contributor
  • ***
  • Posts: 7044
  • Country: nl
Re: Low Voltage Signal Conditioning Theory Question
« Reply #2 on: December 23, 2019, 10:09:18 pm »
Rule of thumb for the multiplier for peak to peak noise from RMS varies between 6-8.
 

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 15155
  • Country: de
Re: Low Voltage Signal Conditioning Theory Question
« Reply #3 on: December 23, 2019, 10:19:07 pm »
At the relatively low noise level one should also check other noise sources, like the signal source itself and resistors in the feedback part. The 1.8 nV/Sqrt(HZ) about corresponds to a 100 Ohms resistors. 
 

Offline Evan.CornellTopic starter

  • Regular Contributor
  • *
  • Posts: 189
  • Country: us
Re: Low Voltage Signal Conditioning Theory Question
« Reply #4 on: December 26, 2019, 07:41:23 pm »
All very helpful information, thanks!

My application requires high input impedance (think >=100kOhm), therefore fully-differential amplifiers are out of the running, at least as the first stage. Because I would like to maintain a fully-differential signal path, I am looking at implementing Cross-Connected Instrumentation Amplifier using 2x AD8429, see here for hookup: https://www.analog.com/en/analog-dialogue/raqs/raq-issue-161.html

My goal here is to be able to be able to digitize a 40uVpp input signal (~200kHz max bandwidth) with an ENOB>=5. I believe with AD8429 gain=~600, the input referred noise should be somewhere between 1nV/rt(Hz) and 2nV/rt(Hz), thus 871nV total noise at DiffInAmp input (does that work out to be nVpp or nVrms when coming from Noise Density parameter in spec sheet?). 871nV*10^(5*6/20)=27.6uV minimum signal level to get ENOB=5. Take that 27.6uV minimum input signal times gain of 600, that's 16.54mV. ADC I am using is 24-bit, with noise level of 10uVpp. I don't need gain of 600 to get to correct ENOB level, but I do think I need the gain level that low to keep the input noise of the DiffInAmp low enough.

I also want to have some passive high-pass filtering between sensor and AD8429. To get f_3dB=~3.4kHz, I get C=4.7uF, R=10ohm, trying to keep resistor value low to reduce noise contribution. Question: what capacitor characteristics should I look for in terms of low noise differential signal path?
 

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 15155
  • Country: de
Re: Low Voltage Signal Conditioning Theory Question
« Reply #5 on: December 26, 2019, 08:22:17 pm »
A low pass filter at 3.4 kHz would limit the bandwidth. So this does not go together with 200 kHz BW. It may be a good idea to have some low pass filtering, but I don't hink one would have to go all the way to the low kHz range at the input, unless there is interference.
For the RC values it depends on the output impedance of the sensor. 10 Ohms resistors only make sense if the source is really that low in impedance. For the filter I would avoid electrolytic caps and class 2 ceramics.
More filtering could be after the first amplifier stage as this would also filter out out of band amplifier noise.

With the AD8429 keep in mind the 1.5 pA/Sqrt(Hz) of noise current. So 2 amplifiers in parallel would be some 0.7 nV/Sqrt(Hz) and some 2.1 pA/sqr(Hz). So this would be right for a very low impedance source of less than some 500 Ohms, less (e.g. 50 ohms)  if very low frequencies are of interest.  The noise density numbers are RMS.

For the SNR calculation it is not about the BW of the amplifier, but the BW of the signal in question. The amplifier and ADC can be faster and one could than do some digital low pass filtering. In a noise critical system there should be however anti aliasing filtering. If not to fast a SD ADC could make this easier than an SAR type. With an ENOB of only a little over 5, there is no real need for a high resolution ADC. Even 8 or 10 Bits could be enough, especially with some oversampling.

So what is the sensor impedance and what are the actual frequencies of interest.
 

Offline Evan.CornellTopic starter

  • Regular Contributor
  • *
  • Posts: 189
  • Country: us
Re: Low Voltage Signal Conditioning Theory Question
« Reply #6 on: December 26, 2019, 08:46:01 pm »
High-pass, not low-pass is what I said above. Signal range from 10kHz to 200kHz overall. The sensor input has a nominal capacitance (piezo transducer) of 1.7nF, so impedance will range from 9362ohm at 10kHz to 468ohm at 200kHz.
 

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 15155
  • Country: de
Re: Low Voltage Signal Conditioning Theory Question
« Reply #7 on: December 26, 2019, 10:59:54 pm »
Sorry for thinking of a low pass filter - the low R and and large capacitance made sense for a low pass only, if at all.

The sensor is quite high impedance, especially in the lower frequency range. So I am afraid the AD8429 amplifier is not suitable, as the current noise is too high. The best type of amplifier is likely JFET based.  For noise in the low nV/sqrt(Hz) region one may have to use discrete JFETs. Really good BJTs (e.g. super beta to keep current noise low) may still be an option, though not without difficulties due to the current noise. Even for the JFETs one may have to look at the current noise.
Chances are the input would be single sided and the transition to differential would be only later, e.g. from the 2nd stage on.

For the high pass filtering chances are one would not need an extra capacitor but could use just the sensor capacitance and a relatively large resistor in the 100 K range to provide the DC bias. So the input filter cross over would be intentionally considerably lower than the signals of interest. This is to avoid noise from the filter that comes in near the cross over region. Here a larger resistor is better as there is less noise current. In the pass band the sensor capacitance shorts out the voltage noise of the resistor.
Final filtering would be behind the initial amplifier, maybe in software if the ADC has enough dynamic range.
 

Online Marco

  • Super Contributor
  • ***
  • Posts: 7044
  • Country: nl
Re: Low Voltage Signal Conditioning Theory Question
« Reply #8 on: December 26, 2019, 11:49:01 pm »
Just use a single ended JFET opamp like the OPA827 and put it right next to the transducer.
 

Offline Evan.CornellTopic starter

  • Regular Contributor
  • *
  • Posts: 189
  • Country: us
Re: Low Voltage Signal Conditioning Theory Question
« Reply #9 on: December 27, 2019, 01:45:19 pm »
So the Current Noise at 1.5pA/rt(Hz) translates to 14nV/rt(Hz), given sensor impedance of 9362ohm @ 10kHz? I can see now why that's not going to work.

I will continue my search.. thanks for the OPA827 suggestion.

EDIT: Kleinstein, in regards to the 100k resistor, is that in parallel to the sensor?
« Last Edit: December 27, 2019, 01:58:43 pm by Evan.Cornell »
 

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 15155
  • Country: de
Re: Low Voltage Signal Conditioning Theory Question
« Reply #10 on: December 27, 2019, 02:07:16 pm »
Another low noise JFET OP is the ADA4625.

For even lower noise one could use discrete JFETs like SK3557, especially in a single sided amplifier as opposed to the differential input stage in an OP. The frequency range is not that different from AM radio.
I attached an NBS paper on such an discrete amplifier that may be interesting.
 

Offline Evan.CornellTopic starter

  • Regular Contributor
  • *
  • Posts: 189
  • Country: us
Re: Low Voltage Signal Conditioning Theory Question
« Reply #11 on: December 27, 2019, 03:45:35 pm »
So searching for devices with Vn <= 4nV/rt(Hz), and In <= ~100fA/rt(Hz), I find a couple of options:

ADA4625, previously mentioned
    3.3nV/rt(Hz)
    4.5fA/rt(Hz)
    max Av=90V/V for BW=200kHz
    Iq=5mA
AD8655
    2.7nV/rt(Hz)
    ADI parametric search shows 1fA/rt(Hz), but then that spec is nowhere in the datasheet. Their forums show other people asking, but no official answer besides "10s of fA/rt(Hz) around 10kHz, up to low hundreds of fA/rt(Hz) at lower frequencies" (https://ez.analog.com/amplifiers/w/documents/1686/ad8655-current-noise-density)
    max Av=140V/V
    Iq=5.3mA
OPA607 (pre-release part from TI)
    3.8nV/rt(Hz)
    47 fA/rt(Hz)
    max Av=250V/V
    Iq=1mA
OPA828
    4nV/rt(Hz)
    1.2 fA/rt(Hz)
    max Av=225V/V
    Iq=7.9mA
OPA827
    3.8nV/rt(Hz)
    2.2 fA/rt(Hz)
    max Av=110V/V
    Iq=6mA

It seems like AD8655 is the best of the bunch in terms of voltage noise, but without any sort of characterized spec on current noise density, I'd be hesitant to use it. So then, ADA4625 looks like the best realistic bet. All the other op-amps/instrumentation amps with lower voltage noise seem to all have much higher current noise densities that wipe out the gains, based on the impedance of my sensor.

If I were to use ADA4625-2 (dual version), use the first amp to get all the gain, and then second amp to invert that amplified output, so signal chain is fully differential after that point, the benefit would be in rejecting any common mode noise coupling into signal chain. Is there a realistic benefit to that approach, or is maintaining single-ended until I get to a fully-diff-amp to drive ADC inputs a more realistic topology?
 

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 15155
  • Country: de
Re: Low Voltage Signal Conditioning Theory Question
« Reply #12 on: December 27, 2019, 04:38:16 pm »
The AD8655 is a CMOS OP, so the current noise should not be that large - 100 fA/Sqrt(Hz) would be very high for a non chopper CMOS amplifier. Chances are high the noise is in the 0.1-10 fA range, so nothing to really worry about.
So it is a very real option, especially if 5 V supply a favorable.
For lower noise one could consider 2 amplifiers in parallel.

Doing the single ended to differential step after the first stage can have some benefit, as this tends to avoid ground reference and ground currents from later stages. However the extra inverter is an odd way. The more natural way is to use a second stage with differential output. So the conversion would be part of the 2 nd stage. Internally those amplifiers tend to be differential from the start, not just at the output.

One may not need more than 2 stages anyway, maybe 3 if more anti aliasing filtering is needed.
 
The following users thanked this post: Evan.Cornell

Online Marco

  • Super Contributor
  • ***
  • Posts: 7044
  • Country: nl
Re: Low Voltage Signal Conditioning Theory Question
« Reply #13 on: December 27, 2019, 09:45:42 pm »
How dependable are the noise corners on CMOS amplifiers? They can be low noise, but the noise corner is all the way up to 10K in the datasheet ... don't want it to go any higher.
 

Offline Evan.CornellTopic starter

  • Regular Contributor
  • *
  • Posts: 189
  • Country: us
Re: Low Voltage Signal Conditioning Theory Question
« Reply #14 on: December 28, 2019, 01:16:02 pm »
How dependable are the noise corners on CMOS amplifiers? They can be low noise, but the noise corner is all the way up to 10K in the datasheet ... don't want it to go any higher.

Hence my concern, since the current noise vs. freq. isn't plotted in the spec sheet for AD8655. However, the ADA4625-2 and AD8656 (dual version) both share same package and pinout, so if I can get away with 5V unipolar rail for both, then I can probably use them interchangeably.
 

Offline Evan.CornellTopic starter

  • Regular Contributor
  • *
  • Posts: 189
  • Country: us
Re: Low Voltage Signal Conditioning Theory Question
« Reply #15 on: January 03, 2020, 07:11:18 pm »
So it seems AD8656 is specified only for single rail 5V operation. Since I want input impedance > 100kohm to avoid loading my sensor, that hints at non-inverting. But since AD8656 is unipolar, that means sensor has to be biased around 2.5V. So, that makes me look at ac-coupling the sensor, but that means series C to ac-couple, plus resistor fed from bias supply. Does that resistor (R3 in the snippet) become the input impedance value? So then it's either loading down my sensor or adding a bunch of resistor noise?

Is there a better way to do this? I am trying to achieve ~50V/V, if it makes a difference.

EDIT: Or am I just stuck with bi-polar operation so I can directly DC couple my sensor to non-inverting input?

Ref: https://ocw.mit.edu/courses/media-arts-and-sciences/mas-836-sensor-technologies-for-interactive-environments-spring-2011/readings/MITMAS_836S11_read02_bias.pdf
« Last Edit: January 03, 2020, 07:25:20 pm by Evan.Cornell »
 

Offline Vovk_Z

  • Super Contributor
  • ***
  • Posts: 1478
  • Country: ua
Re: Low Voltage Signal Conditioning Theory Question
« Reply #16 on: January 03, 2020, 07:25:47 pm »
You can feed ad8656 with +- 2.5V.
 
The following users thanked this post: Evan.Cornell


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf