PicoScope 7 Software
Available on Windows, Mac and Linux
Error due to low resolution
-
Franz Ferdinand
- Newbie
- Posts: 0
- Joined: Tue May 14, 2019 9:45 am
Error due to low resolution
Hello,
i write, because i didn't find anything suitable in the forum.
I use:
PicoScope 3406DMSO
PicoScope Software Version: 6.13.11.3952
PicoLog - 6.1.8
I'm logging Data with PicoScope (PicoLog has the same result) over a relatively long time (1 hour) (see Images). One Image is created with standard 8-Bit-mode and the other image shows the same signal, but with the 12 Bit resolution increase.
- +-10V
- 200 s/div
- 300 kS
- 8 Bit/ 12 Bit
- (Zoom-Funktion)
In 12 Bit-Mode in my expectation the graph should be like the red line in the image - but it's not. The graph I get has waves with plateaus and I don't know if these waves/plateaus are real or if it's just a measurement error.
Is it possible that the measurement resolution (8 Bit) is too low and the waves are caused by a quantisation error, because the measuring range is very large.
Thank you for your help.
Best regards
i write, because i didn't find anything suitable in the forum.
I use:
PicoScope 3406DMSO
PicoScope Software Version: 6.13.11.3952
PicoLog - 6.1.8
I'm logging Data with PicoScope (PicoLog has the same result) over a relatively long time (1 hour) (see Images). One Image is created with standard 8-Bit-mode and the other image shows the same signal, but with the 12 Bit resolution increase.
- +-10V
- 200 s/div
- 300 kS
- 8 Bit/ 12 Bit
- (Zoom-Funktion)
In 12 Bit-Mode in my expectation the graph should be like the red line in the image - but it's not. The graph I get has waves with plateaus and I don't know if these waves/plateaus are real or if it's just a measurement error.
Is it possible that the measurement resolution (8 Bit) is too low and the waves are caused by a quantisation error, because the measuring range is very large.
Thank you for your help.
Best regards
- Attachments
-
-
Re: Error due to low resolution
Hi Franz,
You are correct about what it is that you are seeing. On a +/-10V input range the minimum step size for a sample is 20/2^8 = 78mV and if you measure the distance between the plateaus you will find that it is very close to this. Your capture doesn't use all of the input range, so you do have to zoom in to see it which means that for your signal the quantisation steps are a more significant part of the signal. So you are partly right in that a signal that is too small for the input range is one factor that affects how much quatisation you have, but there is a more significant reason why you are seeing the quantisation levels, which is based upon how Resolution Enhancement works.
Resolution enhancement needs some noise in the signal to work because noise in the signal provides more statistical data that the software can get an average value from. Resolution Enhancement involves passing a moving average filter over the data, and the moving average filter that you select (i.e. 1/2 a bit to 4 bits) has a width defined by the number of samples. So, the width of the smoothing effect seen on your screen is dependent upon how many samples you are using for your capture. If you use too many samples then the same width filter is spread over samples concentrated into a smaller area, so you get what you are seeing, i.e. the smoothing effect working over a range of values that is less than the width of the 8 bit steps, so it doesn't smooth away the quantisation steps. If you use too few samples then the smoothing effect is over a much larger part of the signal so it then starts to affect features of the signal itself, e.g. reducing the peaks of the signal. So the aim (if you can do it) with resolution enhancement is to use just enough samples so that you are removing the unwanted parts of the signal (e.g. noise and quantisation) without reducing its features.
The examples below show a sinewave on an Picocope 5444D, set to 8-bits, with enhanced resolution to 12-bit. Example (1) shows the selected number of samples (19,231 for 20k requested) that generate the best filtering effect (negligible visible quantisation steps, minimal reduction of the signal peak), example (2) shows that decreasing the number of samples (4,902 for 5k requested) makes the negative signal peak start to reduce in size, and example (3) shows that increasing the number of samples (50,000 for 60k requested) makes the quantisation steps start to become visible.
1 - just enough samples
2 - too few samples
3 - too many samples
In practice you rarely capture pure sinewaves so, when using Resolution Enhancement with other more rapidly changing waveforms, such as squarewaves, it's not possible to adjust the number of samples to remove what you want without removing faster changing parts of the signal. So you have to request a number of samples based upon how much unwanted data you want to leave unextracted, versus how much signal you are prepared to lose.
In case you are unaware of what the "number of samples for the samples requested" refers to, in PicoScope 6 you can control the number of samples used for capturing data by requesting a specific number of samples (typing or selecting a value in the 'Number of samples' text box just to the right of the Timebase selection drop down list). So, when you request a number of samples, PicoScope 6 will use a number of samples that is as close to the number you entered as possible, but that ensures the display will have a complete trace, for the current Timebase value, (with the few sample rates that it can use for that Timebase). The actual number used will always be less (as close as possible but not larger) than the number you request. (The requested number of samples, and the Timebase are what you use to select the sample rate that you want from the fixed values that are available.)
Regards,
Gerry
You are correct about what it is that you are seeing. On a +/-10V input range the minimum step size for a sample is 20/2^8 = 78mV and if you measure the distance between the plateaus you will find that it is very close to this. Your capture doesn't use all of the input range, so you do have to zoom in to see it which means that for your signal the quantisation steps are a more significant part of the signal. So you are partly right in that a signal that is too small for the input range is one factor that affects how much quatisation you have, but there is a more significant reason why you are seeing the quantisation levels, which is based upon how Resolution Enhancement works.
Resolution enhancement needs some noise in the signal to work because noise in the signal provides more statistical data that the software can get an average value from. Resolution Enhancement involves passing a moving average filter over the data, and the moving average filter that you select (i.e. 1/2 a bit to 4 bits) has a width defined by the number of samples. So, the width of the smoothing effect seen on your screen is dependent upon how many samples you are using for your capture. If you use too many samples then the same width filter is spread over samples concentrated into a smaller area, so you get what you are seeing, i.e. the smoothing effect working over a range of values that is less than the width of the 8 bit steps, so it doesn't smooth away the quantisation steps. If you use too few samples then the smoothing effect is over a much larger part of the signal so it then starts to affect features of the signal itself, e.g. reducing the peaks of the signal. So the aim (if you can do it) with resolution enhancement is to use just enough samples so that you are removing the unwanted parts of the signal (e.g. noise and quantisation) without reducing its features.
The examples below show a sinewave on an Picocope 5444D, set to 8-bits, with enhanced resolution to 12-bit. Example (1) shows the selected number of samples (19,231 for 20k requested) that generate the best filtering effect (negligible visible quantisation steps, minimal reduction of the signal peak), example (2) shows that decreasing the number of samples (4,902 for 5k requested) makes the negative signal peak start to reduce in size, and example (3) shows that increasing the number of samples (50,000 for 60k requested) makes the quantisation steps start to become visible.
In practice you rarely capture pure sinewaves so, when using Resolution Enhancement with other more rapidly changing waveforms, such as squarewaves, it's not possible to adjust the number of samples to remove what you want without removing faster changing parts of the signal. So you have to request a number of samples based upon how much unwanted data you want to leave unextracted, versus how much signal you are prepared to lose.
In case you are unaware of what the "number of samples for the samples requested" refers to, in PicoScope 6 you can control the number of samples used for capturing data by requesting a specific number of samples (typing or selecting a value in the 'Number of samples' text box just to the right of the Timebase selection drop down list). So, when you request a number of samples, PicoScope 6 will use a number of samples that is as close to the number you entered as possible, but that ensures the display will have a complete trace, for the current Timebase value, (with the few sample rates that it can use for that Timebase). The actual number used will always be less (as close as possible but not larger) than the number you request. (The requested number of samples, and the Timebase are what you use to select the sample rate that you want from the fixed values that are available.)
Regards,
Gerry
Gerry
Technical Specialist
Technical Specialist
Re: Error due to low resolution
One addition to answer of Gerry, the resolution enhancement is actually a moving average.
So if you have a short anomaly (short spike or dip) you will barely see it in the results. Or even worse you wont see it if it is a short enough spike or dip.
Be aware of this when you use resolution enhancement and are on the hunt for anomaly's.
Benno
So if you have a short anomaly (short spike or dip) you will barely see it in the results. Or even worse you wont see it if it is a short enough spike or dip.
Be aware of this when you use resolution enhancement and are on the hunt for anomaly's.
Benno
-
Franz Ferdinand
- Newbie
- Posts: 0
- Joined: Tue May 14, 2019 9:45 am
Re: Error due to low resolution
Hello,
thank you Gerry for your detailed explanation. bennog also thank you very much. This makes sense. I'll test your advice on future measurements.
Meanwhile, I had the opportunity to measure my signal with a 16 bit data logger and I get the signal pattern that I was expecting (similar to the red line in my first image).
So I think my question can be marked as answered.
Thank you.
thank you Gerry for your detailed explanation. bennog also thank you very much. This makes sense. I'll test your advice on future measurements.
Meanwhile, I had the opportunity to measure my signal with a 16 bit data logger and I get the signal pattern that I was expecting (similar to the red line in my first image).
So I think my question can be marked as answered.
Thank you.