Some cut & paste info about the technique:
High-resolution mode
The second method of averaging in the digitising oscilloscope that does not require a repetitive signal is high-resolution mode.
These types of oscilloscope provide 8-bit vertical resolution in normal acquisition mode, like most other digitising oscilloscopes. However, high-resolution mode on the oscilloscope offers up to 12 bits of vertical resolution in real-time mode, which reduces noise and increases vertical resolution.
Instead of averaging points from multiple acquisitions in a single time bucket like normal averaging mode, high-resolution mode averages sequential points within the same acquisition together.
In high-resolution mode you cannot directly control the number of averages as you can in averaging mode. Instead, the number of extra bits of vertical resolution is dependent on the time/division setting of the scope.
When operating at slow time base ranges, the oscilloscope serially filters sequential data points and maps the filtered results to the display. Increasing the memory depth of on-screen data increases the number of points averaged together.
High-resolution mode has no effect at fast time/div settings, where the number of on-screen points captured is small. It has a large effect at slow time/div settings, where the number of on-screen points captured is large.
How to achieve 12 bits of resolution
For the oscilloscope in high-resolution mode, vertical resolution varies as the time base changes as below (at 1Gsample/s sample rate).
Time base Bits of resolution
<10ns/div 8
50ns/div 9
200ns/div 10
1µs/div 11
> or =5µs/div 12
To get up to 12 bits of resolution in high-resolution mode, the oscilloscope averages together at least 16 consecutive samples.
Therefore, to achieve 12 bits of resolution, we add 16 consecutive samples together, then divide the total by 16. This process is commonly referred to as decimation. This results in 12 bits of useful data. Notice that these are bits of resolution, not accuracy.
The effectiveness of high-resolution mode depends on the characteristics of the dominant noise sources, which come internally from the oscilloscope or from external circuits measured by the oscilloscope.
In other words, you only get more “resolution” in the presence of white noise. You may not be able to get more resolution for averaging noise-free samples.
Most A/D converters used in digitising oscilloscopes are 8-bit converters, with 8-bit differential non-linearity (DNL) and 8-bit integral non-linearity (INL). DNL error is defined as the difference between an actual step width, and the ideal value of 1LSB (least significant bit).
The INL error is described as the deviation, in LSB or percent of full-scale range (FSR), of an actual transfer function from a straight line. High-resolution acquisition on a noisy signal will tend to improve DNL characteristics, but not INL. The scope probes and pre-amplifier in the front-end of the oscilloscope are calibrated to a few per cent of accuracy, so those are the dominant component needed to improve the integral non-linearity accuracy.