FFT and 60 Notch filter software

[loop123]

Is the .wav file supposed to contain any useful information?
Besides the spurs, everything looks like random noise to me

It is just sine wave generator output of 10uV, 50Hz sine wave. Simple only but why is the output at Audacity has so much noise? Goal is simply to remove the noise and regain the pure 10uV, 50Hz sine wave. pls help do that.

My understanding was that the 50 Hz component (and the other spurs) are interfering signals which you want to suppress. If, OTOH, the 50 Hz signal is the useful signal you want to keep (while all the other spectral components should be suppressed), then this can easily be done with a narrow bandpass filter. However, I think this is not what you really want. A real EEG signal is not a pure 50 Hz sine wave, therefore such a bandpass filter would suppress most of the useful components in a real EEG signal. Is the EEG simulator not able to simulate a more realistic EEG signal?

Above is the specs of the Netech simulator I'm using. It can only output 0.1, 0.5, 2, 50, and 60 Hz.  I chose 50Hz signal. The amplitude choices are 10, 30, 50, 100, 500 μV, 1,2, and 2.5 mV.  I chose 10μV signal.  And I fed the 50Hz, 10uV signal into the BMA-200 bioamplifier. And connect the BNC output of the BMA-200 to the E1DA Cosmos ADC and into Audacity. 

Yes, 50Hz is the signal, not the noise. It's just to test the unit. So if 10uV, 50Hz sine wave is the signal. Why does Audacity show it with so many noises like in following? Are they harmonics? but where is the resultant in 2nd image?  This is only present if 10uV are used and not only higher amplitude signal which has better SN Ratio.






Please remove the noises and present the pure 10uV, 50Hz output from the Netech simulator.

[loop123]

I opened your wav file in EDFbrowser and it doesn't look noisy.

The one you downloaded is the 2.5mV, 60Hz output from Netech Simulator. Instead download the attached one with 10uV, 50Hz signal from the Netech simulator.

I loaded at audacity now and the following is the waveform

[gf]

That's what I get if I notch out 60Hz and 78 Hz and cut-off the bandwidth at 100 Hz. What bandwidth do you need to keep?
Audacity has notch and lowpass filters, too, so I guess you could do the same in Audacity.

[loop123]

That's what I get if I notch out 60Hz and 78 Hz and cut-off the bandwidth at 100 Hz. What bandwidth do you need to keep?
Audacity has notch and lowpass filters, too, so I guess you could do the same in Audacity.

Did you used my audio file above? I want to keep 1000Hz. I just ran EDFbrowser for first time and loaded it up but I can't find how to run the Amplitude FFT done by Karel. where to find it. He loaded  up the 2.5mV file, not the 10uV, 1000hz file which is in my last message. Please remove the noises and retain 1000Hz, 10uV.

[WatchfulEye]

Most of the noise is between 50 Hz and 1000 Hz - it is broad band noise, not at any specific frequency.
Apart from the 60 and 78 Hz spurs, the only way to reduce the noise is reduce the bandwidth. You cannot keep up to 1000 Hz and reduce the noise any further with any simple filter process.

[loop123]

First, please tell me what exactly are these noises? In the 2nd image, they don't look like harmonics:

If they are harmonics. They should be resultant in the waveform. But my audacity waveforms don't show any resultant. The noises are riding in each part of the sine wave. How do you differentiate between EMI, RFI, amplifier noise, and quantization ADC noises? How sure are you they are powerline noises or harmonics? Please tell me what exactly those are first.

You seem to be looking for an individiual frequency as the source of your noise/distortion. It actually appears that this is broadband noise/noise floor that you are seeing. Also, a hint when using the audacity spectrum viewer - you need to amplify the signal first, as the viewer has a fixed minimum on the Y axis. You can then magnify as required in the spectrum viewer.

This is the spectrum of the last waveform you posted. I've also calculated normalised amplitudes for narrow band noise, on the basis that the signal is 10 uV rms.

You can see the 50 Hz signal, and small 2nd and 3rd harmonics. You can also see a small amount of 60 Hz noise (1 uV), but no harmonics.
However, there are 2 unknown and more important noise sources - 78 Hz (1.7 uV) as well as harmonics of 78 Hz, and broadband noise (approx 100 nV/sqrt Hz).

The 60 Hz is likely coming from power line noise, and could be mitigated with better shielding, or the use of a notch filter (depending on the frequencies of interest in your signal), or differential measurement.
The 78 Hz is a mystery, and may be noise generated by the simulator.
The broadband noise may be the noise floor of your setup, and because of it's 1 kHz bandwidth, it has a total contribution of approx 3 uV.

I suggest repeating the measurements with other waveforms and amplitudes, to see what happens to the unknown noise sources. It may be that some of the unknown noise is from your waveform generator.

Oh. I missed this message. Is the waveform you downloaded the one with lots of noises, like above? That's what I wanted to know, whether it is just powerline harmonics or the noise floor already. But the noise of the AMP01 used in BMA-200 is 5nV/sqrt (Hz). So at 1000Hz bandwidth, the noise is actually 0.158uV. Where did you get the 100nV/Sqrt (Hz)? 

[loop123]

Most of the noise is between 50 Hz and 1000 Hz - it is broad band noise, not at any specific frequency.
Apart from the 60 and 78 Hz spurs, the only way to reduce the noise is reduce the bandwidth. You cannot keep up to 1000 Hz and reduce the noise any further with any simple filter process.

What was what I wanted to know since message 1 of this thread. Whether this is powerline noises or noise floor already. Did you ran them at Audacity? How did you choose the LPF in the last 3 images?   But the AMP01 has 5nV/sqrt (Hz) noise or 0.158uV at 1000Hz. So with 10uV signal, the noise floor is very small fraction.

[WatchfulEye]

Only the 60 Hz noise is powerline.

There is a mystery 78 Hz noise.

The main noise is broadband noise - which looks like noise floor, but by my calculations it is about 100-150 nV/sqrt Hz - much higher than specified noise of the AMP01. It could be noise from the waveform generator.

However, note that BMA-200 specifies 7uV noise - and your recording shows less noise than this.

You could try measuring the noise directly by using a 1 kohm resistor instead of the waveform generator. It would be interesting to have a series of recordings with different waveforms and 1 with just a resistor - all settings left exactly the same.

[loop123]

Only the 60 Hz noise is powerline.

There is a mystery 78 Hz noise. The main noise is broadband noise - which looks like noise floor, but by my calculations it is about 100-150 nV/sqrt Hz - much higher than specified noise of the AMP01.

However, note that BMA-200 specifies 7uV noise - and your recording shows less noise than this.

I don't know how the 0.158uV rms noise at 1000Hz at the AMP01 became so bad at the BMA-200. But is it really possible to make an amplifier that can deal with clean 10uV, 1000Hz bandwidth signal? Is there a known device that can do this already? Using AMP01, why can't one make this?

I'll try the resistor measurement tomorrow.. please let me know what can imitate 10uV too instead of using the Netech simulator.

[gf]

Here is the accumulated spectral power up to 1kHz. The big jump is the wanted signal. Even the spurs result only in small jumps. Everything else is wideband noise as you see. It makes sense to lowpass filter at 1kHz to get rid of the top 5% beyond 1kHz.

EDIT: I should not have labeled the plot with "noise power", but rather spectral power.

[Karel]

I just ran EDFbrowser for first time and loaded it up but I can't find how to run the Amplitude FFT done by Karel. where to find it.

To view the powerspectrum of a signal, leftclick on the signallabel and choose "Spectrum".

https://www.teuniz.net/edfbrowser/EDFbrowser%20manual.html#Powerspectrum

[BrianHG]

LOL, you could just apply a deep 3D statistical FFT noise remover algorithm.  (There is an 20 year old audio editor which can do it, but, I don't think its what you want since it requires example sample noise to work with, then it removes just that noise and leaves everything else automatically.  Though, it only supports 24bit integer sampling, maybe 32bit floats.)

However, since you have been delivered a simulated sample and not working with true hardware where you can make sure there is no 60hz/50hz noise from the source, you are working with someone who wants you to make your own specific filter for these EEG cases which you will not easily achieve.

NOTE: In real life, the 50hz/60hz over a 24 hour period is typically good, but from minute to minute, I've seen it drift up and down by as much as 0.2hz.  There have been recorded cases where the line frequency has been off by as much as 1hz or more.

[tggzzz]

I just ran EDFbrowser for first time and loaded it up but I can't find how to run the Amplitude FFT done by Karel. where to find it.

To view the powerspectrum of a signal, leftclick on the signallabel and choose "Spectrum".

https://www.teuniz.net/edfbrowser/EDFbrowser%20manual.html#Powerspectrum

You had better tell him how to "leftclick", before he asks us.

[BrianHG]

What you really need is to sample all the EEG channels in parallel.  Then, make an algorithm which subtracts the common noise from all the channels channels together leaving you with the pure EEG signals.  This means you need a 16 channel sampler with 16 channel amp and 16 channel capture software.

Remember, the 60hz/50hz/6.7khz noise will be on all channels, just at slightly different amplitudes and for the high 6.7khz, slightly different phase.  With all that source info, you can negate just the commonalities between channels.  No notch filtering.  This means a filter with no ringing, echo, bounce.  A filter which will also follow power interruptions and surges since the same junk will be present on all channels.  A notch filter may be useful as a selector to zero in on these selective frequencies, but for low frequencies like 60hz, with a 44.1khz sample rate, you will have a huge 65kpoint sample lag to teeth out that narrow band.

[loop123]

Only the 60 Hz noise is powerline.

There is a mystery 78 Hz noise.

The main noise is broadband noise - which looks like noise floor, but by my calculations it is about 100-150 nV/sqrt Hz - much higher than specified noise of the AMP01. It could be noise from the waveform generator.

However, note that BMA-200 specifies 7uV noise - and your recording shows less noise than this.

You could try measuring the noise directly by using a 1 kohm resistor instead of the waveform generator. It would be interesting to have a series of recordings with different waveforms and 1 with just a resistor - all settings left exactly the same.

Edit: this is 30uV in the 2nd BMA instead of 10uV. With both set to 10uV. the noises are identical.



Anyway. Out of curiosity. Have you guys encountered an amplifier that has no obvious damage but the noise just became 20 times worse like from 5nV/sqrt (Hz) to 100 nv/sqrt (Hz)?

What about ADC. Can ADC damaged by ESD etc cause much more noises too and not obvious non-functional state? Because I have another unit, the expensive $16750 g.USBamp used by major R&D centers worldwide. I got it used for $1100. The following are the waveforms. At 10uV, 1000Hz, the waveforms have bad noises which I guess are not powerline noises at all but the noise floor itself? The g.USBamp doesn't have any amplifier, it uses ADC only with +-250mV range. It's noises are Noise level   < 0.4 µV rms 1-30 Hz. Extrapolating it to 1000Hz, I guess the noises would be above 1 uV? Yet the $16750 unit sells so much. I still haven't bought their $4000 software to use it fully so I used their demo software just to test it.

See its specs at

https://www.gtec.at/product/gusbamp-research/

The following are the outputs from USBamp demo (offset is adjusted to 100uV or it can go off the scale). Input all 10uV, 50Hz from the Netech simulator

The following with 100Hz bandwidth in the USBamp app with Notch Filter off:



The following with 100Hz bandwidth in the USBamp app with Notch Filter on:



The following with 250Hz bandwidth in the USBamp app with Notch Filter off:



The following with 250Hz bandwidth in the USBamp app with Notch Filter on:



The following with 500Hz bandwidth in the USBamp app with Notch Filter off:



The following with 500Hz bandwidth in the USBamp app with Notch Filter on:



The following with 1000Hz bandwidth in the USBamp app with Notch Filter off:



The following with 1000Hz bandwidth in the USBamp app with Notch Filter on:

[Andy Chee]

Assuming your equipment is portable and battery powered, how about going into the middle of a nature park reserve, which is a long distance away from any powerlines, TVs, power supplies, and any other electrical equipment.

You should be able to at least establish the noise floor performance of your equipment.  My suggestion is at least 1000m distance in all directions from any electrical device, but I don't know whether you have a location that size.

This is for testing your equipment only, I'm not suggesting you perform your actual experiments in the middle of nowhere!

[loop123]

Only the 60 Hz noise is powerline.

There is a mystery 78 Hz noise.

The main noise is broadband noise - which looks like noise floor, but by my calculations it is about 100-150 nV/sqrt Hz - much higher than specified noise of the AMP01. It could be noise from the waveform generator.

However, note that BMA-200 specifies 7uV noise - and your recording shows less noise than this.

You could try measuring the noise directly by using a 1 kohm resistor instead of the waveform generator. It would be interesting to have a series of recordings with different waveforms and 1 with just a resistor - all settings left exactly the same.

Bad news. I made mistake in my last test. I rechecked the amplitude and found out I used 30uV in 2nd BMA instead of 10uV that is why there was minimal noise compared to the 10uV waveform in the first BMA. So ignore my last message except the g.USBamp waveforms.

Anyway. I used 1k ohm resistor as you asked in the original BMA. This is the noise with 1kohm resistor alone connected to BMA with 50000 gain 1000Hz banwidth:



This is 10uV 50Hz from the Netech simulator connected to BMA with 50000 gain 1000Hz bandwidth.



Same setting for both only the resistor was replaced to the Netech as you asked.

computing for the resistor noise at 1k ohm and 1000 Hz

0.13*sqrt(R*f) nV (rms) noise = 0.13 sqrt (1000 x 1000 Hz) nv = 0.13 x 1000 = 130nV rms

the noise of the AMP01 is 5nv/sqrt (Hz) x sqrt (1000Hz bandwidth) = 158.11nV rms

combination is sqrt (130^2 + 158.11^2) = sqrt (16900 + 24999) = sqrt (41899) = 204nV rms

In the Audacity, the noise is not 204nV rms!

with 10uV x 50000 = 0.5V.. the noise is not far from 0.5V.. maybe 100 uV. What's going on? The 1kohm resistor test used the same scale as the 10uV and the noises are closed.

They are the noise floor, isn't it? and not interference.. because for interference, you are supposed to be able to filter it using 60Hz and harmonics. and not wide band noise, right?

With the AMP01 5nV/sqrt (Hz). Why are we getting more than say 1000nV/sqrt (Hz) in the BMA? Even if the BMA is so junk, why is the AMP01 also messed up as well. The 2nd BMA showed the same noise so it is not due to defective AMP01

[loop123]

By using 2.5mV in the Netech Simulator and 2000 gain in the BMA. I get 5 volts in Audacity from 1 to -1. 

Does it mean Audacity 1 to -1 is really 5 Volts? Or does it change depending on your setups? how?

So the 10uV x 50000 gain = 0.5V in last message is right scale at it is 0.1 instead of 1  or 5V/10 =0.5V.

[tggzzz]

Does it mean Audacity 1 to -1 is really 5 Volts? Or does it change depending on your setups? how?

Sigh. You really don't understand.

Audacity processes numbers, either floating point or integer. Those numbers have absolutely no defined relationship to any input voltage.

Sometimes those numbers might originate in an ADC. Often they will have been "artificially generated" inside a computer program.

[gf]

Does it mean Audacity 1 to -1 is really 5 Volts? Or does it change depending on your setups? how?

Sigh. You really don't understand.

Audacity processes numbers, either floating point or integer. Those numbers have absolutely no defined relationship to any input voltage.

Sometimes those numbers might originate in an ADC. Often they will have been "artificially generated" inside a computer program.

Exactly.

ffmpeg -i netech\ 10uv\ 50hz\ bma\ 1000hz\ 50000\ gain.wav
...
Input #0, wav, from 'netech 10uv 50hz bma 1000hz 50000 gain.wav':
  Duration: 00:00:29.94, bitrate: 705 kb/s
    Stream #0:0: Audio: pcm_s16le ([1][0][0][0] / 0x0001), 44100 Hz, 1 channels, s16, 705 kb/s

The file contains just a stream of signed 16-bit integer numbers whose full-scale range is mapped by Audacity to a -1...+1 range when it loads the file. Either the creator of the file can tell you the codes per Volt in the file, or you need to find out yourself.

[loop123]

Does it mean Audacity 1 to -1 is really 5 Volts? Or does it change depending on your setups? how?

Sigh. You really don't understand.

Audacity processes numbers, either floating point or integer. Those numbers have absolutely no defined relationship to any input voltage.

Sometimes those numbers might originate in an ADC. Often they will have been "artificially generated" inside a computer program.

Exactly.

ffmpeg -i netech\ 10uv\ 50hz\ bma\ 1000hz\ 50000\ gain.wav
...
Input #0, wav, from 'netech 10uv 50hz bma 1000hz 50000 gain.wav':
  Duration: 00:00:29.94, bitrate: 705 kb/s
    Stream #0:0: Audio: pcm_s16le ([1][0][0][0] / 0x0001), 44100 Hz, 1 channels, s16, 705 kb/s

The file contains just a stream of signed 16-bit integer numbers whose full-scale range is mapped by Audacity to a -1...+1 range when it loads the file. Either the creator of the file can tell you the codes per Volt in the file, or you need to find out yourself.

If I used the same ADC and amplifier like the BMA + E1DA. The output at Audacity is consistent? For example using my 5V full scale 1 to -1 baseline in my last message. Can you tell what is the approximate voltage level of the following?   



I really thought it was 10uV and displaying right at the 2nd BMA. But half day later. I can't get the same waveforms. So I assumed it was 30uV.    Based on my 5V baseline using the same BMA + E1DA. Do you think it's 10uV or 30uV?

With the AMP01 noise of 5nv/sqrt (Hz). I should be getting 0.158uV rms noise. And using 10uV signal. There should be no wide band noise like the ones I shared previously.

Andy suggested I go far away to test it. But then with no specific frequencies like 60Hz and harmonics. It is not interference, right? You can already tell from the analysis that the wideband noise is not interference but the noise floor already. Or can interference mimic for all intent and purposes the noise floor such that you can't differentiate noise floor from interference? Please elaborate. Thanks.

[tggzzz]

Does it mean Audacity 1 to -1 is really 5 Volts? Or does it change depending on your setups? how?

Sigh. You really don't understand.

Audacity processes numbers, either floating point or integer. Those numbers have absolutely no defined relationship to any input voltage.

Sometimes those numbers might originate in an ADC. Often they will have been "artificially generated" inside a computer program.

Exactly.

ffmpeg -i netech\ 10uv\ 50hz\ bma\ 1000hz\ 50000\ gain.wav
...
Input #0, wav, from 'netech 10uv 50hz bma 1000hz 50000 gain.wav':
  Duration: 00:00:29.94, bitrate: 705 kb/s
    Stream #0:0: Audio: pcm_s16le ([1][0][0][0] / 0x0001), 44100 Hz, 1 channels, s16, 705 kb/s

The file contains just a stream of signed 16-bit integer numbers whose full-scale range is mapped by Audacity to a -1...+1 range when it loads the file. Either the creator of the file can tell you the codes per Volt in the file, or you need to find out yourself.

If I used the same ADC and amplifier like the BMA + E1DA. The output at Audacity is consistent? For example using my 5V full scale 1 to -1 baseline in my last message. Can you tell what is the approximate voltage level of the following?   

Read the preceding two messages in your message. Please. Pretty please. Think about what they say.

If you don't come to understand why we can't tell you the "approximate voltage level", then there's little anybody can do to help you.

Yes that requires hard work on your part. We've had to go through that process, and so will you.

Alternatively, get out your meter and simply measure the bloody input voltage!

[loop123]

I have thought and tested a lot what you mentioned. I really can estimate approximate voltage level in my files. Here is why.

In my BMA + E1DA, 5V corresponds to 1 to -1. If I adjusted the amplitudes or the gain, the amplitudes can decrease proportionally. This is also why your audio or music can sound as it is because Audacity can consistently track the amplitudes that is why your wave files don't suddenly have jumping amplitudes that don't corresponds to original recording.

Not only that. But if I save the Audacity file into WAV. And later retrieve. It still shows the same amplitude. Therefore I can estimate approximate voltage level because I know the baseline in my particular equipment. Although you can't estimate the approximate level because you don't know my baseline. Yes. I have from start read gf punchline "The file contains just a stream of signed 16-bit integer numbers whose full-scale range is mapped by Audacity to a -1...+1 range when it loads the file.". I know my full range to be 5V so I can estimate my voltage level. But you can't estimate my voltage level becauase you don't know what voltage corresponds to my full range during the mapping of Audacity of the full range in the stream of signed 16-bit integers as gf put it to 1 to -1. Do you guys agree here? Or do you emphasize even I can't estimate my own voltage level in my own file where I know the full range corresponds to 5V??

[nali]

The point being made is that a WAV file has no concept whatever of what a voltage is. It's just a table of numbers. Those numbers are 0 through 2^n-1 e.g. 0-65535; the +/-1 is just Audacity normalising.

The ONLY part of your system that can correlate the information in your WAV file to actual voltages in the real world is your ADC. Fine, if you're 100% confident that your ADC maps to a peak of 5V (is that 5V pk-pk or 5V RMS?) then yes you can use it to work out voltages.

I assume tha ADC is intended for audio. Does it have any form of automatic gain control or clipping protection? If it does you can forget using it for measurements - sell it and buy a scope instead.

[loop123]

The point being made is that a WAV file has no concept whatever of what a voltage is. It's just a table of numbers. Those numbers are 0 through 2^n-1 e.g. 0-65535; the +/-1 is just Audacity normalising.

The ONLY part of your system that can correlate the information in your WAV file to actual voltages in the real world is your ADC. Fine, if you're 100% confident that your ADC maps to a peak of 5V (is that 5V pk-pk or 5V RMS?) then yes you can use it to work out voltages.

I assume tha ADC is intended for audio. Does it have any form of automatic gain control or clipping protection? If it does you can forget using it for measurements - sell it and buy a scope instead.

I'll ask Ivan about the E1DA. So when I told "gf" my full range is 5V, and asked him to estimate the voltage level of a lower amplitude waveform then he should be able to do it since I gave the 5V baseline for the 1 to -1. The reason I need to estimate is because when I used the 2nd backup BMA initially. I got what I thought was a 10uV with BMA 1000Hz switch with no noise. but I couldn't repeat it. Instead I always got the same noise as the one I shared early which some of you made frequency analysis with unknown 78Hz peak. The following is the 5V reference waveform to 1 to -1, the 10uV with 100Hz selected (here I can't duplicate it so gf, Karel, WatchfulEye, please run the Wav file attached and let me know how to duplicate the dancing waveforms).. In the third image is the one I thought was 10uV with 1000Hz bandwidth with no noise, but the amplitude is much larger so I was wondering if it is really 30uV instead of 10uV. Hence asking gf to verify the voltage level with the 1 to -1 referenced to 5 Volts (I am sure) given.







gf. please check if the last image above has the same amplitude as the middle image with confirmed 10uV or is the last image not 10uV? If so, then it is 30uV and the noises in the earlier image I shown which you made frequency analysis is correct. If so, my next goal is figuring out how a AMP01 with excellent 5nV/sqrt (Hz) noise figure turn to 1000nV/sqrt (Hz) noise figure in the BMA and build a virgin AMP01 that can do 10uV with 0.158uV noise.