-
Bf412 is a typo. The circuit analysis is also wrong at that time. Lf412 buffer stage has gain of 1.09 (just for trimming purposes).
45 nV/Sqrt Hz was miscalculation based on wrong resistor values in schematic and also included additional allowance for 1/f noise at 50Hz and 10kohm source impedance. Based on 2x Lf412 using datasheet spec + 2x 5k input protection resistors, a better estimate would be 37 nV/Sqrt Hz. Still the datasheet is old and it is not impossible that manufacturing processes have improved.
The Lf412 is a JFET input amplofier and as such has very high impedance, very low bias current and very low current noise. This allows it to have consistent performance even with a very high source impedance. This is why it is used as the input stage.
The AMP01 has more bias current and more current noise. These will limit performance and introduce errors and noise with a high impedance signal source. With the problems that come with modifying an existing circuit which has been factory trimmed for precision, I don't think a major modification like bypassing the input stage is going to be a good Idea.
-
Also using the 200Hz flat response portion, the noise density is 1.323mV/Sqrt (Hz) / 50000 = 26.4nV/Sqrt (Hz). Still too low considering AMP01 has 5nV/rtHz and LF412 has 25nv/rtHz noise.
Why? With gain=1 for the LF412 input stage, the combined noise power of 25nV/sqrt(Hz) and 5nV/sqrt(Hz) is still only ~25.5nV/sqrt(Hz). Please remember what I worte:QuoteIf you add two noise sources, then their powers (or squared voltages) sum up, not their voltages
Isn't 26.4 close to this value anyway? IMO it is in the ballpark.
-
Here is the part I need clarification above.
"The dynamic range can limit how much amplification one can have before the ADC. With the modern high resolution ADCs this is usually no longer an issue. E.g. the noise would still be way small than the 250 mV range of the ADC."
Can you provide actual values or example to illustrate what this means "The dynamic range can limit how much amplification one can have before the ADC"?
as for the next statement:
"E.g. the noise would still be way small than the 250 mV range of the ADC."
If your noise is 20uV and your signal is 10uV. What if the 20uV noise is too small than the 250mV range of the g.USBamp, you cant resolve the 10uV. Can you?
Note the BMA uses AMP01 to amplify signal. While the g.USBamp doesnt have amplifier bec you map the signal direct to the ADC.
So my question is. Using purely ADC based amplifier like the USBamp. Can you somehow magically retrieve the signal at 10uV at 1000Hz even if the noises are akin to the BMA? And in connection to this thread. How can I test using the REW RTA on the g.USBamp? maybe there is another RTA software to piggy back on its own software??
I want to justify spending $3000 on the software if the g.USBamp has some magic in it the BMA doesnt. One thing it has magic on is it has DSP that can oversample and remove noises using moving averages and almost brick wall frequency response. I want to test it using REW RTA or similar software. -
Also using the 200Hz flat response portion, the noise density is 1.323mV/Sqrt (Hz) / 50000 = 26.4nV/Sqrt (Hz). Still too low considering AMP01 has 5nV/rtHz and LF412 has 25nv/rtHz noise.
Why? With gain=1 for the LF412 input stage, the combined noise power of 25nV/sqrt(Hz) and 5nV/sqrt(Hz) is still only ~25.5nV/sqrt(Hz). Please remember what I worte:QuoteIf you add two noise sources, then their powers (or squared voltages) sum up, not their voltages
Isn't 26.4 close to this value anyway? IMO it is in the ballpark.
Going back to the BMA. I inputted a 10uV, 50Hz signal to the BMA at 1000Hz, 50000 gain and tested it with REW with the same E1DA and all same settings. The noise is higher, why?
In our last RTA. The noise os 0.746uV rms and 26.4nV/Sqrt(Hz).
Now with input signal, the noise becomes 2.548uV rms and the noise density is 36.8nV/Sqrt (Hz). Why does noise increase when there is actual input signal? The noise happens to be the one that is supported by Audacity showing the 10uV waveform below with 2.548uV noise.
-
I measured the Netech EEG simulator directly set at 2.5mV, 50Hz connected to the E1DA (without the BMA connected). This is the RTA. I was wondering if the Netech introduced additional noise of more than 1.8uV (difference between the 2.548uV (with input) minus 0.746uV (without input)). Are the noises floor supposed to be the same with input and without input?
What does the RTA in the following say? Also can you tell if it outputs in rms or peak to peak? what voltage does it output?
-
above is 2.5mV, 50Hz set in Netech connected directly to the E1DA without the BMA.
Below is 10uV, 50Hz set in Netech connected directly to the E1DA without the BMA.
Noise is 1.245uV. Is it accurate? does it mean with 10uV Netech output, there is noise of 1.245uV. And if the BMA is connected, it adds to the BMA 0.746uV noise to produce roughly 2uV noise in the 10uV output? Also let me know if the output in both 2.5mV and 10uV is in rms or peak-to-peak. Many thanks guys.
-
I think what the last few measurements from the signal simulator show is the biggest problem of using an audio ADC and generic software for measurement. The software and the hardware are not integrated together and the settings are wrong in some way.
There appears to be a significant gain error, as the signal amplitude measured in the software does not correspond to the signal amplitude expected. We had guessed that the simulator generated a 10 uV p-p signal (3.4 uV rms), which after 50k gain, should have been measured as 170 mV - but instead the measurement is 127 mV. Earlier on we had assumed that the ADC was calibrated for gain, and that the 1.7 V setting corresponded to a 1.7 V rms sine wave (5V p-p). However, this may not be accurate.
Other possibilities are that the signal generator is not calibrated for amplitude, or that there is an error introduced because your amplifier is unable to accurately drive the low input impedance of your ADC. (The gain error is much larger when connecting the signal generator direct to the ADC, likely because as a biophysical simulator it simulates a high source resistance).
I think before proceeding further, it is necessary to verify the amplitude calibration of your measurements, as it is clear that something is wrong, and it is likely that all your measurements so far have been lower than the real value.
In the first instance, you could use the calibration signal in your amplifier, which should be 1 mV p-p, and using an appropriate setting (e.g. gain 1000) use the oscilloscope mode in your REW software to verify the amplitude of the resulting signal. There is theoretically another problem here which is that your amplifier's test signal is 10 Hz, and an audio ADC may not measure this faithfully. If you have a normal laboratory oscilloscope, you should also directly measure the output of your amplifier as a cross check. Similarly, a general purpose laboratory signal generator may also be useful as an additional signal source.
If you are going to be using this setup for scientific purposes, getting something as fundamental as amplitude correct is of great importance. Spending time getting this right strikes me as the most important step now.
-
I think what the last few measurements from the signal simulator show is the biggest problem of using an audio ADC and generic software for measurement. The software and the hardware are not integrated together and the settings are wrong in some way.
There appears to be a significant gain error, as the signal amplitude measured in the software does not correspond to the signal amplitude expected. We had guessed that the simulator generated a 10 uV p-p signal (3.4 uV rms), which after 50k gain, should have been measured as 170 mV - but instead the measurement is 127 mV. Earlier on we had assumed that the ADC was calibrated for gain, and that the 1.7 V setting corresponded to a 1.7 V rms sine wave (5V p-p). However, this may not be accurate.
Other possibilities are that the signal generator is not calibrated for amplitude, or that there is an error introduced because your amplifier is unable to accurately drive the low input impedance of your ADC. (The gain error is much larger when connecting the signal generator direct to the ADC, likely because as a biophysical simulator it simulates a high source resistance).
I think before proceeding further, it is necessary to verify the amplitude calibration of your measurements, as it is clear that something is wrong, and it is likely that all your measurements so far have been lower than the real value.
In the first instance, you could use the calibration signal in your amplifier, which should be 1 mV p-p, and using an appropriate setting (e.g. gain 1000) use the oscilloscope mode in your REW software to verify the amplitude of the resulting signal. There is theoretically another problem here which is that your amplifier's test signal is 10 Hz, and an audio ADC may not measure this faithfully. If you have a normal laboratory oscilloscope, you should also directly measure the output of your amplifier as a cross check. Similarly, a general purpose laboratory signal generator may also be useful as an additional signal source.
If you are going to be using this setup for scientific purposes, getting something as fundamental as amplitude correct is of great importance. Spending time getting this right strikes me as the most important step now.
You are right. All the amplitudes dont correspond so let me calibrate first. In the following. I used the ADC to measure the Netech EEG Simulator set at 2.5mV, 50Hz. I selected the 1.7vrms switch at ADC and set also the FS sine Vrms in REW RTA to 1.7. But why do I get only 0.544mV instead of 2.5mV in the RTA voltage display attached? Is 544.5uV the noise or the signal amplitude? how do you make it display the 2.5mV signal input at RTA? is it even possible to measure real voltage in RTA?
-
The E1DA Cosmos ADC likely cannot accurately measure the netech simulator directly. The ADC has an input impedance of 640 Ohms, and is only suitable for measuring low impedance sources (the source impedance must be much lower than 640 Ohms whereas I would expect the netech simulator to simulate a higher source impedance).
The amplifier has a calibration signal source so you may as well use it. However, it is a square wave and the filters will distort it, so you will need to measure it with some oscilloscope software, rather than try to use rms measurements.
-
The E1DA Cosmos ADC likely cannot accurately measure the netech simulator directly. The ADC has an input impedance of 640 Ohms, and is only suitable for measuring low impedance sources (the source impedance must be much lower than 640 Ohms whereas I would expect the netech simulator to simulate a higher source impedance).
The amplifier has a calibration signal source so you may as well use it. However, it is a square wave and the filters will distort it, so you will need to measure it with some oscilloscope software, rather than try to use rms measurements.
What oscilloscope software do you recommend (that you have personally tested)? You mean REW RTA cant measure peak to peak? -
What oscilloscope software do you recommend (that you have personally tested)? You mean REW RTA cant measure peak to peak?
REW scope is fine. It can measure p-p (use the voltage cursor tool) and it uses the calibration setting from the RTA page.
-
Getting back to your original question, ADCs designed for audio often have poor DC specifications.
The bump in noise at low frequencies might be from aliasing of high frequency noise.
-
Have you measured the noise using an identical setup? For example, using the RTA app. In other words directly compare the different op amps with signal inputs shorted to ground and all other settings the same.
-
Have you measured the noise using an identical setup? For example, using the RTA app. In other words directly compare the different op amps with signal inputs shorted to ground and all other settings the same.
By shorting the inputs (NOT using the Netech) and all settings identical, just replacing the OPA2132P and LF412 to compare. The OPA2132P has lower noise! 26mV (541.6uV/Sqrt (Hz) vs 40mV (855.2uV/Sqrt (Hz) to the LF412. See below. But with the Netech connected, the noise is the same. Can the waveforms seen at Audacity be significant for 26mV vs 40mV noise?
This is the for OPA2132P
This is for the LF412
-
By shorting the inputs (NOT using the Netech) and all settings identical, just replacing the OPA2132P and LF412 to compare. The OPA2132P has lower noise! 26mV (541.6uV/Sqrt (Hz) vs 40mV (855.2uV/Sqrt (Hz) to the LF412. See below. But with the Netech connected, the noise is the same. Can the waveforms seen at Audacity be significant for 26mV vs 40mV noise?
As expected, the measurements of the amplifier itself show significantly lower noise with the newer op amps.
However, when making a real measurement: there is also the signal source resistance. Any resistor will generate noise, and that includes the resistance of an electrode or biological specimen.
If your waveform generator is simulating a physiological signal with a high resistance, then there will be a noise contribution from the source resistance. Depending on the resistance, this could be the largest source of noise in your experiment.
Also remember that because noise adds as a root-sum-of-squares (you square the noise amplitude of all sources, add them, then take the square root), it is changes to the largest noise source which make the biggest difference, with changes to minor sources having disproportionaltely small effects.
If you are trying to measure noise density with your simulator connected, there is a problem. Your simulator produces a low fidelity sine wave with huge quantisation noise - the quantisation noise is broadband and not readily distinguishable from other sources of broadband noise. This will prevent any meaningful noise density measurement, even though the quantisation noise is visibly distinct on a waveform.
If you want to get a feel for the amplitude of the noise waveform, you could just record the noise with audacity and inspect the waveform. You could then repeat the recording with a 10k or 20k resistor, to see how much noise the resistance adds. The same experiment could be done with REW to get an rms measurement.
-
By shorting the inputs (NOT using the Netech) and all settings identical, just replacing the OPA2132P and LF412 to compare. The OPA2132P has lower noise! 26mV (541.6uV/Sqrt (Hz) vs 40mV (855.2uV/Sqrt (Hz) to the LF412. See below. But with the Netech connected, the noise is the same. Can the waveforms seen at Audacity be significant for 26mV vs 40mV noise?
As expected, the measurements of the amplifier itself show significantly lower noise with the newer op amps.
However, when making a real measurement: there is also the signal source resistance. Any resistor will generate noise, and that includes the resistance of an electrode or biological specimen.
If your waveform generator is simulating a physiological signal with a high resistance, then there will be a noise contribution from the source resistance. Depending on the resistance, this could be the largest source of noise in your experiment.
Also remember that because noise adds as a root-sum-of-squares (you square the noise amplitude of all sources, add them, then take the square root), it is changes to the largest noise source which make the biggest difference, with changes to minor sources having disproportionaltely small effects.
If you are trying to measure noise density with your simulator connected, there is a problem. Your simulator produces a low fidelity sine wave with huge quantisation noise - the quantisation noise is broadband and not readily distinguishable from other sources of broadband noise. This will prevent any meaningful noise density measurement, even though the quantisation noise is visibly distinct on a waveform.
If you want to get a feel for the amplitude of the noise waveform, you could just record the noise with audacity and inspect the waveform. You could then repeat the recording with a 10k or 20k resistor, to see how much noise the resistance adds. The same experiment could be done with REW to get an rms measurement.
What frequency in the Netech output should you use to compare with the resistor noise? The one at 0.5Hz, 10uV setting at Netech simulator was close to the 10k Ohm resistance shorted value. Note it's difficult to compare noises by looking at the amplitudes at Audacity. They have somewhat same amplitudes, but it is only when the REW RTA was used that the noises could be distinguished.
The following was when 1k Ohm resistor was shorted in the input of the BMA. It's closed to the value when the input is totally shorted. But looking at Audacity, how can you compare the display at 1k ohm vs 10 k ohm? All results were from the OPA2132P chip.
The following is when 10k ohm shorted at input
The following is when the Netech simulator was used with input frequency of 0.1Hz and 10uV. I chose 0.1 Hz because it was the only way for the sine wave to be almost straight at the horizontal, to compare with the resistors shorted displays. It is at this 0.1Hz, 10uV Netech output setting where the RTA noise is similar to the 10k ohm resistor shorted value above.
The following is when the Frequency was increased to 2 Hz. There are 0.1Hz, 2Hz, 5Hz, 50Hz, 60Hz frequency options. I displayed them to ask you what frequency must I use to compare it to the 1k and 10k resistors shorted. As the frequency increases, the noises increased. The display at Audacity just show more sine wave as the amplitude is increased.
The following is when the frequency was increased to 5Hz.
The following is when the frequency was increased to 50Hz.
Since the noises at RTA matched at Netech 0.1Hz 10uV output and the 10k resistor shorted. Does it mean the noise of the Netech is like 10k resistor? If it is, then it's bad because how can you decrease the resistance of the skin to mere 100ohms. Even then, the shorted input noise is already high at 26.4mV as shared in last message.
-
By shorting the inputs (NOT using the Netech) and all settings identical, just replacing the OPA2132P and LF412 to compare. The OPA2132P has lower noise! 26mV (541.6uV/Sqrt (Hz) vs 40mV (855.2uV/Sqrt (Hz) to the LF412. See below. But with the Netech connected, the noise is the same. Can the waveforms seen at Audacity be significant for 26mV vs 40mV noise?
As expected, the measurements of the amplifier itself show significantly lower noise with the newer op amps.
However, when making a real measurement: there is also the signal source resistance. Any resistor will generate noise, and that includes the resistance of an electrode or biological specimen.
If your waveform generator is simulating a physiological signal with a high resistance, then there will be a noise contribution from the source resistance. Depending on the resistance, this could be the largest source of noise in your experiment.
Also remember that because noise adds as a root-sum-of-squares (you square the noise amplitude of all sources, add them, then take the square root), it is changes to the largest noise source which make the biggest difference, with changes to minor sources having disproportionaltely small effects.
If you are trying to measure noise density with your simulator connected, there is a problem. Your simulator produces a low fidelity sine wave with huge quantisation noise - the quantisation noise is broadband and not readily distinguishable from other sources of broadband noise. This will prevent any meaningful noise density measurement, even though the quantisation noise is visibly distinct on a waveform.
If you want to get a feel for the amplitude of the noise waveform, you could just record the noise with audacity and inspect the waveform. You could then repeat the recording with a 10k or 20k resistor, to see how much noise the resistance adds. The same experiment could be done with REW to get an rms measurement.
The above tests used the latest OPA2132P. When I used the old LF412. I can see that even the output of the Netech Simulator has more noise (something I couldn't easily detect just looking at Audacity). I'd just share the results below of the Netech output of 0.5Hz, 10uV vs 10k Ohm resistor connected to the BMA at 50k gain and 1000Hz.
This is Netech 0.1Hz, 10uV RTA using the LF412. This has more noise compared to the 36.73mV result using the OPA2132P chip.
This is 10k resistor shorted at the BMA RTA using LF412 too. This has more noise compared to the 33.97mV result using the OPA2132P chip.
So the Netech at 10uV is like simulating 10k Ohm. But you said something about "If you are trying to measure noise density with your simulator connected, there is a problem. Your simulator produces a low fidelity sine wave with huge quantisation noise - the quantisation noise is broadband and not readily distinguishable from other sources of broadband noise. This will prevent any meaningful noise density measurement, even though the quantisation noise is visibly distinct on a waveform."
You mean the Netech at 10uV can't be likened to 10k ohm noise? But why is its quantization noise like 10k ohm? but in last paragraph you just said to try comparing them.
Whatever. Since changing the buffer amp can lower the noise in both Netech and resistor tests. Then putting the 1nV/Sqrt (Hz) INA849 can produce even lesser noise at the Netech and 10k resistor tests??
What are the formulas again to compute what the noise in nV contributed by resistors? I want to see the total noise when the INA849 will be put (is this compatible by putting a second socket on top of the existing socket and rewiring the top socket of the AMP01??)
-
I think you need to take a step back and consider what you are trying to do.
This thread was about suitability of an audio ADC for instrumentation use. The main problems mentioned are calibrating the gain and the frequency response. At present, the gain has been guessed but not checked, and frequency response remains unknown. It is also not clear what software you plan to use for analysis of your recordings and what it needs for calibration of gain.
There has been a lot of discussion about noise, but you don't seem to know what signal-to-noise ratio you need for your analysis. You have been trying to track down noise which appears in measurements of a simulator, but the amount and type of noise generated by the simulator is unknown, meaning that it may or may not be representative of real data. The noise hunt has been made more difficult because the noise measurements may be inaccurate as the gain of your measurement system is not calibrated.
I would suggest taking some time to calculate where your major noise sources are, and how they impact your signal-to-noise ratio, and in turn how they affect your analysis.
The formula for thermal noise in a resistor is easily found with google, and a writeup is found at https://en.wikipedia.org/wiki/Johnson%E2%80%93Nyquist_noise#Noise_voltage_and_power.
If you want to reanalyse your amplifier to see what changing various components makes then, I think main noise sources in the signal path are:
Source resistance (e.g. 10 kOhm)
2x 5k protection resistors
2x op amps (20 nV/sqrt Hz for the original LF412)
Instrumentation amp (5 nV/sqrt Hz for the AMP01).
You are interested in a bandwidth of 1 kHz:
So the calculation becomes:
10k Source resistance: 0.13 * Sqrt (10000) * Sqrt (BW) = 411 nV rms
2x 5k Protection resistors: Sqrt(2) * 0.13 * Sqrt (5000) * Sqrt (BW) = 411 nV rms
2x OP amps: Sqrt(2) * 20 * Sqrt (BW) = 894 nV rms
I amp: 5 * Sqrt (BW) = 158 nV rms
The noise powers sum:
Total = Sqrt (411 ^2 + 411 ^2 + 894 ^2 + 158 ^2) = 1078 nV rms
-
This is a red herring! The poster is using the instrumentation to detect bioelectrical signals like EEG, EMG and ECG. They are definitely NOT performing a particle physics experiment!I think you need to take a step back and consider what you are trying to do.
I'm doing a critical CERN-like experiment so need the most sensitive instrument like they do at the Large Hadron Collider. Remember the movie Interstellar where they need to collect data to finish the equations of quantum gravity and save the future. My experiment is like it.
That's not to say their experiment is any more or less in value to the scientific world. Just that from an engineering technical perspective, the poster is dealing with human biosignals.