Hi,

I've checked the effective number of bits of my PicoScope 5244B and compared it to the specifications with different resolutions.

I get:

Both are close enough, but I'm a little surprised by the 'low' ENOB compared to the physical number of bits. So I've made the same measurements with all possible vertical scale and resolution.

I get:

Except 16bits resolution, there is almost no difference between 8, 12, 14 and 15bits...

Note that I made ENOB measurements with 2 methods (AC rms voltage and FFT noise floor, block mode, PicoScope6 version : 6.10.6.1), both give me exactly the same ENOB value.

Did I made something wrong or could you help me to understand why scope resolution impact is so limited on ENOB? Thank you

Regards

Herve

## Effective number of bits of 5244B

### Re: Effective number of bits of 5244B

Hi Herve,

Would you be able to tell me how you calculated the ENOB?

So that I can test with a device.

Kind Regards,

Would you be able to tell me how you calculated the ENOB?

So that I can test with a device.

Kind Regards,

Karunen

Technical Specialist

Pico Technology

Technical Specialist

Pico Technology

### Re: Effective number of bits of 5244B

Hi Karunen,

I've computed the ENOB according 2 methods (both give almost the same ENOB value)

Method 1: With AC noise

1- I mesure AC Vrms noise with PicoScope6 (using Measure Menu, input loaded with 540ohms). On my scope, I get Noise_ACVrms=118.4µVrms with vertical scale=+/-50mV

2- According to vertical scale, I compute a sinus full scale amplitude. For example, with +/-50mV, SinFS=2*50/(2*SQRT(2)), unit is mVrms

3- Then, SNR is 20*Log10(SinFS/Noise_ACVrms). Here I compute SNR=49.5dB

4- Then, ENOB=(SNR-1.76)/6.02=7.93bits

Method 2: With FFT noise floor

1- With PicoScope6, FFT display, FFT size =65536, unit Arbitrary dB and reference set to SQRT(2) (meaning 1Vp), I get the FFT noise floor.

With +/-50mV scale, I get FFT_NF=-124.2dB (average)

2- Relative to the FFT reference, I compute a 0dBFS correction ratio 0dBFS=20*LOG10(SinFS/1Vp)

For +/-50mV scale, I get 0dBFS=-29dB

3- I compute the FFT processing gain: FFT_PG=10*Log10(FFT_Size/2)=45.15

4- Then, SNR=0dBFS-FFT_NF-FFT_PG = -29+124.2-45.15=50dB

5- Then, ENOB=(SNR-1.76)/6.02=7.98bits

Thank you for keeping me informed of your measurements

Regards

Herve

I've computed the ENOB according 2 methods (both give almost the same ENOB value)

Method 1: With AC noise

1- I mesure AC Vrms noise with PicoScope6 (using Measure Menu, input loaded with 540ohms). On my scope, I get Noise_ACVrms=118.4µVrms with vertical scale=+/-50mV

2- According to vertical scale, I compute a sinus full scale amplitude. For example, with +/-50mV, SinFS=2*50/(2*SQRT(2)), unit is mVrms

3- Then, SNR is 20*Log10(SinFS/Noise_ACVrms). Here I compute SNR=49.5dB

4- Then, ENOB=(SNR-1.76)/6.02=7.93bits

Method 2: With FFT noise floor

1- With PicoScope6, FFT display, FFT size =65536, unit Arbitrary dB and reference set to SQRT(2) (meaning 1Vp), I get the FFT noise floor.

With +/-50mV scale, I get FFT_NF=-124.2dB (average)

2- Relative to the FFT reference, I compute a 0dBFS correction ratio 0dBFS=20*LOG10(SinFS/1Vp)

For +/-50mV scale, I get 0dBFS=-29dB

3- I compute the FFT processing gain: FFT_PG=10*Log10(FFT_Size/2)=45.15

4- Then, SNR=0dBFS-FFT_NF-FFT_PG = -29+124.2-45.15=50dB

5- Then, ENOB=(SNR-1.76)/6.02=7.98bits

Thank you for keeping me informed of your measurements

Regards

Herve

### Re: Effective number of bits of 5244B

Hi Herve,

Sorry for the delayed response. To answer your original question, your calculation sets an upper limit for ENOB but doesn’t give the full picture of dynamic performance. When ADC’s are interleaved in time to create a faster sample rate, there is an inevitable bit of distortion of phase, gain and offset which leads to the generation of spurs. This clearly doesn’t affect the ADC’s when they are run synchronously in parallel, so that their outputs can be summed. Now when you change the resolution from 12 to 14-bits, there is a marginal reduction in range due to the internal configuration, however, the interleaved spurs will disappear, which has a small effect on the calculation of SINAD and ENOB. What is even more significant here is that because SFDR is calculated using only the fundamental frequency, and spurs, the increase in range at the higher resolutions is much more significant, giving another 10dB of signal range. This is extremely useful for applications where you’re trying to discriminate between one frequency that is modulated by others, as in radio work.

Regarding the noise itself, there are essentially 2 components, i.e. the noise that is inherent in the analogue input stages of the scope, and the noise generated by the ADC itself. As the step sizes get smaller and we sum the values of more ADC cores in parallel, so the ADC noise reduces but the input noise is unaffected, and so begins to dominate as you increase the resolution. This masking effect means that you don’t actually see the improvement in ENOB that you would expect as you increase the number of bits used by the ADC. However, there are tools within both the hardware and PicoScope 6 that you can use to counter this masking effect, and lift the signal image from the digitized signal and noise.

So, in order to reduce noise at the higher resolutions you can switch in the 20 MHz bandwidth limiting hardware filter in the advanced input options for the channel. This will reduce the noise from the input stage, and actually give you better performance than your expected values (almost 10-bit resolution). Also, this will not band limit your signal because at a sampling rate of 62.5 MS/s (max rate at 16-bits) you should only be using a signal bandwidth up to 12.5MHz anyway (at 14–bits your maximum signal bandwidth should be 25MHz, so you would only be slightly band limited by 5MHz with the filter switched in).

There are also other techniques that you can use to improve the resolution still further such as adding resolution enhancement (although this may band limit your signal, so you need to be aware of what your bandwidth requirements are). Also, using Maths channel averaging, on a repetitive waveform, can also clean up noise over successive waveforms. These reductions in the input noise, and improvements to the signal imaging in your display, are useful for more general applications that require higher resolution analysis. It goes without saying that you should also observe all the common ways of minimizing noise such as reducing inductive pickup from sources of noise, ensuring good clean grounding, etc.

What you also need to bear in mind is that the High resolution mode, can also be referred to as high precision mode, because the 4 ADC’s in a 5000 device are used in parallel to give higher precision sampling which results in more accurate samples, as well as higher resolution. This increased precision reduces errors associated with accuracy, and also helps with enhancement techniques, in particular, maths averaging.

To put this all in perspective, the PicoScope 5000 series gives you all the features of our standard scopes, but also offers the features of a great high speed scope (which requires very specific circuitry), and the features of an increased resolution and accuracy scope, which requires a different set of optimizations. It’s not possible to satisfy these differing hardware requirements without compromising somewhere on the specification sheet. However, the design of our PicoScope 5000 series has resulted in a minimal specification compromise that our customers can easily further reduce the impact of and is, all in all, an extremely capable, versatile piece of equipment.

I hope this clarifies things for you.

Regards,

Gerry

Sorry for the delayed response. To answer your original question, your calculation sets an upper limit for ENOB but doesn’t give the full picture of dynamic performance. When ADC’s are interleaved in time to create a faster sample rate, there is an inevitable bit of distortion of phase, gain and offset which leads to the generation of spurs. This clearly doesn’t affect the ADC’s when they are run synchronously in parallel, so that their outputs can be summed. Now when you change the resolution from 12 to 14-bits, there is a marginal reduction in range due to the internal configuration, however, the interleaved spurs will disappear, which has a small effect on the calculation of SINAD and ENOB. What is even more significant here is that because SFDR is calculated using only the fundamental frequency, and spurs, the increase in range at the higher resolutions is much more significant, giving another 10dB of signal range. This is extremely useful for applications where you’re trying to discriminate between one frequency that is modulated by others, as in radio work.

Regarding the noise itself, there are essentially 2 components, i.e. the noise that is inherent in the analogue input stages of the scope, and the noise generated by the ADC itself. As the step sizes get smaller and we sum the values of more ADC cores in parallel, so the ADC noise reduces but the input noise is unaffected, and so begins to dominate as you increase the resolution. This masking effect means that you don’t actually see the improvement in ENOB that you would expect as you increase the number of bits used by the ADC. However, there are tools within both the hardware and PicoScope 6 that you can use to counter this masking effect, and lift the signal image from the digitized signal and noise.

So, in order to reduce noise at the higher resolutions you can switch in the 20 MHz bandwidth limiting hardware filter in the advanced input options for the channel. This will reduce the noise from the input stage, and actually give you better performance than your expected values (almost 10-bit resolution). Also, this will not band limit your signal because at a sampling rate of 62.5 MS/s (max rate at 16-bits) you should only be using a signal bandwidth up to 12.5MHz anyway (at 14–bits your maximum signal bandwidth should be 25MHz, so you would only be slightly band limited by 5MHz with the filter switched in).

There are also other techniques that you can use to improve the resolution still further such as adding resolution enhancement (although this may band limit your signal, so you need to be aware of what your bandwidth requirements are). Also, using Maths channel averaging, on a repetitive waveform, can also clean up noise over successive waveforms. These reductions in the input noise, and improvements to the signal imaging in your display, are useful for more general applications that require higher resolution analysis. It goes without saying that you should also observe all the common ways of minimizing noise such as reducing inductive pickup from sources of noise, ensuring good clean grounding, etc.

What you also need to bear in mind is that the High resolution mode, can also be referred to as high precision mode, because the 4 ADC’s in a 5000 device are used in parallel to give higher precision sampling which results in more accurate samples, as well as higher resolution. This increased precision reduces errors associated with accuracy, and also helps with enhancement techniques, in particular, maths averaging.

To put this all in perspective, the PicoScope 5000 series gives you all the features of our standard scopes, but also offers the features of a great high speed scope (which requires very specific circuitry), and the features of an increased resolution and accuracy scope, which requires a different set of optimizations. It’s not possible to satisfy these differing hardware requirements without compromising somewhere on the specification sheet. However, the design of our PicoScope 5000 series has resulted in a minimal specification compromise that our customers can easily further reduce the impact of and is, all in all, an extremely capable, versatile piece of equipment.

I hope this clarifies things for you.

Regards,

Gerry

Gerry

Technical Specialist

Technical Specialist