Measure of SNR seems to bad
Measure of SNR seems to bad
Hello,
i watch the video "How to measure the dynamic range of the PicoScope 4262".
See https://www.youtube.com/watch?v=FxluPd1nZ8I
I do the same measurement with my PicoScope 4262.
But my measure give way to bad SNR.
In video the SNR is 97 dBc (sign? and dB).
With PicoScope 6 Version 6.10.16.2 I got 48 dBc (dB?).
With PicoScope 6 Version 6.11.12.1692 I got 76 dBc(dB?)
I believe the SNR should be better than 90 dB as the attached pictures show.
Best Regards
egonotto
i watch the video "How to measure the dynamic range of the PicoScope 4262".
See https://www.youtube.com/watch?v=FxluPd1nZ8I
I do the same measurement with my PicoScope 4262.
But my measure give way to bad SNR.
In video the SNR is 97 dBc (sign? and dB).
With PicoScope 6 Version 6.10.16.2 I got 48 dBc (dB?).
With PicoScope 6 Version 6.11.12.1692 I got 76 dBc(dB?)
I believe the SNR should be better than 90 dB as the attached pictures show.
Best Regards
egonotto
Re: Measure of SNR seems to bad
Hello,
I reflect about the SNR problem and I make new measure of the signal and noise separately.
The noise is about 0.05 mV RMS (in +1V range)
The signal is about 671 mV RMS.
So the SNR is about 82 dB
Thus i think the SNR value in the video is way too good.
97 dB SNR would give a noise about 0.01 RMS in the range +1V
From PicoScope 4262 Specifications:
Noise 8.5 µV RMS (on most sensitive range)
Best Regards
egonotto
I reflect about the SNR problem and I make new measure of the signal and noise separately.
The noise is about 0.05 mV RMS (in +1V range)
The signal is about 671 mV RMS.
So the SNR is about 82 dB
Thus i think the SNR value in the video is way too good.
97 dB SNR would give a noise about 0.01 RMS in the range +1V
From PicoScope 4262 Specifications:
Noise 8.5 µV RMS (on most sensitive range)
Best Regards
egonotto
Re: Measure of SNR seems to bad
Looking at the traces in the first post the difference between the fundamental and the first harmonic / highest harmonic appear to be similar to the video, and similar to readings I am getting here.
How are you connecting the signal generator output to the channel input ?
How are you connecting the signal generator output to the channel input ?
Martyn
Technical Support Manager
Technical Support Manager
Re: Measure of SNR seems to bad
Hello,Martyn wrote:Looking at the traces in the first post the difference between the fundamental and the first harmonic / highest harmonic appear to be similar to the video, and similar to readings I am getting here.
How are you connecting the signal generator output to the channel input ?
yes the trace looks similar as in the video.
Therefore I think it is no problem with the hardware.
I use a 50 ohm BNC cable of length about 1m.
It seems there a no essential disturbance by the cable.
If i set the timebasis to 1ms/div i got about 60 dB SNR (5MHz FFT 16384 bins). With 32768 bins i got again 76 dB SNR.
At timebasis 2ms/div i got about 76 dB SNR (2.5MHz FFT 16384 bins)
Best regards
egonotto
Re: Measure of SNR seems to bad
If you can post a psdata file for the spectrum plot and one for the time plot I will run them against a unit here.
Martyn
Technical Support Manager
Technical Support Manager
Re: Measure of SNR seems to bad
Hello,
here is the psdatafile. It include the time and the spektrum plot.
Best regards
egonotto
here is the psdatafile. It include the time and the spektrum plot.
Best regards
egonotto
 Attachments

 SNR_100ms_100kS.psdata
 Time and spektrum plot
 (176.42 KiB) Downloaded 152 times
Re: Measure of SNR seems to bad
Hi egonotto,
Sorry for the delay in answering your post. I had to first of all make sure that the following discussion was unaffected by a slight issue that we found in PicoScope 6.
When working in Spectrum mode you need to be aware that, although we are working with digital values and you have what looks like an instantaneous FFT plot, in order to perform the calculation, the bandwidth of measurement needs to be split up into frequency bins in order to produce the FFT values. Processing the time domain signal in frequency bins means that the FFT calculation is acting like a narrowband spectrum analyzer that sweeps over the bandwidth of the signal spectrum. As it is only considering a small spread of noise for each calculation this has the effect of pushing the noise down by an amount known as the process gain, which is calculated as 10 Log ( number of bins / 2 ).
So, for instance, if the number of bins is 32768 (the default) then the process gain is:
10 Log (32768 / 2) = 42.14dB
We can demonstrate how this affects the SNR in an example:
Process gain worked example
In the attached data file you can see that the SNR automatic measurement is 77.43dB. This is the decibel ratio of the noise to the 1kHz sinewave from the internal Signal Generator. When you add the process gain (as previously calculated) to this you get an FFT noise floor of:
77.43 + 42.14 = 120.57dB
You can also see that, with the display mode set to "Average Hold", when you place the bottom ruler in the middle of the noise floor, and the top ruler on the top of the 1KhZ waveform, the rulers measure 118.5dB of signal to noise level. As the measurements are approximations, this is close enough to the software calculated value + process gain.
You can see the process gain in action when you double the number of bins ("Spectrum Options>Spectrum Bins") and the displayed noise floor drops by 3dB, (i.e. the process gain increases from 42.14dB to 10 Log (65536 / 2) = 45.15dB)
We can confirm that the software is doing the right thing by performing a calculation for the signal to noise ratio based upon the signal waveform and noise captured in the above example:
SNR worked example
SNR is calculated as follows:
SNR = 20 Log ( RMS value of signal / RMS value of noise)
= 20 Log ( ( peak to peak value of signal / σ ) / peak to peak value of noise / σ )
(where σ is dependent upon the variance you're using)
In the above screenshots you can see that we have zoomed in to the Time Domain (Scope Mode) waveform to measure the peaktopeak noise, and the peaktopeak signal waveform, with the PicoScope voltage rulers, and the values we get are 1.799V for the pp signal, and 213uV for the pp noise (note that these are clearly estimates that will have large uncertainty, based upon the small sample size, but are used here for a loose reference). So, using these values:
SNR = 20 Log ( (1.799 / σ ) / ( 213 * 10^6 / σ ) )
= 20 Log ( 1.799 / 0.000213)
= 75.41dB
Using this calculated value the SNR + process gain is 118.55dB !
Now, the video you referred to on you tube is using a version of PicoScope that had an error in the calculation for SNR. The noise floor in the spectrum plot of the video is at the same (or similar) level to what we had above, but the SNR is at 97dBc. For a 16bit Scope the maximum possible range is 97.6dB, but in Spectrum mode the minimum number of bins that you can have is 128, so there must be a process gain of at least 18dB to subtract (and that's without considering the quantization noise, input noise and signal noise).
That said what is clear here is that, when using Spectrum Mode in PicoScope 6, you have a very powerful tool available to you for reducing the noise floor, so that you can view extremely low level details in the signal. Note, however, that Process Gain only occurs in a Spectrum plot when the Spectrum Amplitudes represent Voltages (which they do in PicoScope 6 because we apply Coherent Gain, or Process Gain as we are referring to it here, in order to normalize the voltage level before and after the FFT). If the Spectrum plot is of Spectra representing Power (volts squared) then the noise floor will remain unchanged with Changing bin widths. Also, the 3dB gain per halving of the bin width can change with certain less popular windowing functions.
You can do similar things in Scope mode in PicoScope 6, such as Enhanced Resolution Mode, which allows you to trade some signal bandwidth for increased resolution (for instance, this works very well using the PicoScope 4262 for analyzing audio, because the Scope bandwidth is much higher than the Bandwidth of audio).
So, when you add all of this together with Maths Channel Filters, and Maths Channel Averaging which give you further powerful tools for reducing the noise floor in Scope view, you have at your disposal a powerful toolkit for cleaning up signals in either domain.
Regards,
Gerry
Sorry for the delay in answering your post. I had to first of all make sure that the following discussion was unaffected by a slight issue that we found in PicoScope 6.
When working in Spectrum mode you need to be aware that, although we are working with digital values and you have what looks like an instantaneous FFT plot, in order to perform the calculation, the bandwidth of measurement needs to be split up into frequency bins in order to produce the FFT values. Processing the time domain signal in frequency bins means that the FFT calculation is acting like a narrowband spectrum analyzer that sweeps over the bandwidth of the signal spectrum. As it is only considering a small spread of noise for each calculation this has the effect of pushing the noise down by an amount known as the process gain, which is calculated as 10 Log ( number of bins / 2 ).
So, for instance, if the number of bins is 32768 (the default) then the process gain is:
10 Log (32768 / 2) = 42.14dB
We can demonstrate how this affects the SNR in an example:
Process gain worked example
In the attached data file you can see that the SNR automatic measurement is 77.43dB. This is the decibel ratio of the noise to the 1kHz sinewave from the internal Signal Generator. When you add the process gain (as previously calculated) to this you get an FFT noise floor of:
77.43 + 42.14 = 120.57dB
You can also see that, with the display mode set to "Average Hold", when you place the bottom ruler in the middle of the noise floor, and the top ruler on the top of the 1KhZ waveform, the rulers measure 118.5dB of signal to noise level. As the measurements are approximations, this is close enough to the software calculated value + process gain.
You can see the process gain in action when you double the number of bins ("Spectrum Options>Spectrum Bins") and the displayed noise floor drops by 3dB, (i.e. the process gain increases from 42.14dB to 10 Log (65536 / 2) = 45.15dB)
We can confirm that the software is doing the right thing by performing a calculation for the signal to noise ratio based upon the signal waveform and noise captured in the above example:
SNR worked example
SNR is calculated as follows:
SNR = 20 Log ( RMS value of signal / RMS value of noise)
= 20 Log ( ( peak to peak value of signal / σ ) / peak to peak value of noise / σ )
(where σ is dependent upon the variance you're using)
In the above screenshots you can see that we have zoomed in to the Time Domain (Scope Mode) waveform to measure the peaktopeak noise, and the peaktopeak signal waveform, with the PicoScope voltage rulers, and the values we get are 1.799V for the pp signal, and 213uV for the pp noise (note that these are clearly estimates that will have large uncertainty, based upon the small sample size, but are used here for a loose reference). So, using these values:
SNR = 20 Log ( (1.799 / σ ) / ( 213 * 10^6 / σ ) )
= 20 Log ( 1.799 / 0.000213)
= 75.41dB
Using this calculated value the SNR + process gain is 118.55dB !
Now, the video you referred to on you tube is using a version of PicoScope that had an error in the calculation for SNR. The noise floor in the spectrum plot of the video is at the same (or similar) level to what we had above, but the SNR is at 97dBc. For a 16bit Scope the maximum possible range is 97.6dB, but in Spectrum mode the minimum number of bins that you can have is 128, so there must be a process gain of at least 18dB to subtract (and that's without considering the quantization noise, input noise and signal noise).
That said what is clear here is that, when using Spectrum Mode in PicoScope 6, you have a very powerful tool available to you for reducing the noise floor, so that you can view extremely low level details in the signal. Note, however, that Process Gain only occurs in a Spectrum plot when the Spectrum Amplitudes represent Voltages (which they do in PicoScope 6 because we apply Coherent Gain, or Process Gain as we are referring to it here, in order to normalize the voltage level before and after the FFT). If the Spectrum plot is of Spectra representing Power (volts squared) then the noise floor will remain unchanged with Changing bin widths. Also, the 3dB gain per halving of the bin width can change with certain less popular windowing functions.
You can do similar things in Scope mode in PicoScope 6, such as Enhanced Resolution Mode, which allows you to trade some signal bandwidth for increased resolution (for instance, this works very well using the PicoScope 4262 for analyzing audio, because the Scope bandwidth is much higher than the Bandwidth of audio).
So, when you add all of this together with Maths Channel Filters, and Maths Channel Averaging which give you further powerful tools for reducing the noise floor in Scope view, you have at your disposal a powerful toolkit for cleaning up signals in either domain.
Regards,
Gerry
Gerry
Technical Specialist
Technical Specialist
Re: Measure of SNR seems to bad
Hi Gerry,
many thanks for your extensive answer.
Best regards
egonotto
many thanks for your extensive answer.
Best regards
egonotto