Why does my Oscilloscope look noisyFollow article
Why does my Digital Scope Look Noisy?
Many oscilloscope users comment that their digital scopes are “noisy”. Often engineers say “they aren’t as clean as their old analogue scopes”. Is this really the case?
This article gives a quick overview of some theory of digital scopes, some techniques to make the scope display look “better” or “worse” – and discuss if really you are looking at noise or the real signal.
Noise cannot be eliminated completely, it’s random and the more bandwidth you have in your system, the more noise you will capture. On an oscilloscope, noise will overlay on the acquired signal, and it may be difficult to see the “real data” behind the noise.
Analogue -v- Digital
Analogue oscilloscope aren’t used that much anymore, but they have a place in the heart of engineers of a certain age. The beam on the screen was deflected from a constant horizontal position directly by the voltage applied to the input. If the beam intensity was kept low, even lots of noise didn’t appear to make the trace look much thicker as the vertical deflection was small, and in many cases the bandwidth of the analogue scope was quite low (typically 20MHz) compared to today’s higher bandwidth digital scopes. The trace on an analogue scope flew across the screen (updated) in relation to the time base setting.
Digital scopes (DSOs) however can make it look like there is lots of noise compared to the same signal on an analogue scope. This is because DSOs digitise the signal and illuminate pixels on the screen, sometimes with equal intensity, and sometimes regardless of their frequency of occurrence. Any small change in the signal amplitude would not appear to change the analogue scope display very much, if at all – but it can result in a different quantisation (digitisation) level, and this can make the trace look thicker and therefore noisier.
Newer digital scopes have overcome the earlier display limitations, and offer intensity grading like an analogue scope, so signals such as noise that don’t occur at the same time and amplitude don’t appear as “bright” as the desired signal.
Be careful with datasheet specifications!
A scope is a signal viewing tool, so it’s no use with no signal applied - you may as well turn it off! Don’t be swayed by really low values or what initially looks like a clean trace with no signal applied.
Oscilloscopes can be optimised for very low signals (little attenuation and high gain in the input path), but with the trade-off poorer performance at higher signal conditions which are typically used much more often. From the chart below - which is taken directly from various manufacturers’ datasheets - we see the declared measured RMS noise with no signal applied/grounded for some 500MHz or 600MHz oscilloscopes. Note how some have lower noise at the 1mV/div setting than others, but have much higher noise at 5mV/div and above – and then not even declaring a measured value at 2V/div input! Think carefully before you buy what you want the scope for, and don’t consider just one optimal condition.
In the case of Supplier A below, the lowest input condition doesn’t even offer the full bandwidth (being limited to 150MHz, not 500MHz), so the user doesn’t get to make a fair comparison of noise performance.
It’s very important to remember that the noise specifications of oscilloscopes are presented as RMS values, whereas the human eye views noise on the display as peak-to-peak.
Many things affect the way the signal is displayed on the oscilloscope.
The signal you see isn’t just the signal, it’s also dependent upon the probing type and technique used, the grounding and various oscilloscope settings, which include: amplification (V/div setting); memory length; waveform update rate; bandwidth, acquisition mode and display mode. We will discuss how changing some of these settings can alter both the actual noise measurement and the way it is displayed.
Assuming that the signal, the probe and ground is the same, if we change the amplification (V/div setting) we can measure different values on the same signal – the smaller the signal displayed (i.e. just 2 of the 8 vertical grids) the less accurate the measurement is as the digitiser is not being used to its the full dynamic range. Also, at lower V/div settings, the more the scope is amplifying - both signal and noise - and so it stands to reason that more signal or system noise will be shown (a lower signal-to-noise ratio). The most effective use of the digitiser is to maximise the signal across the whole display, and using variable gain instead of the fixed 1:2:5:10 steps the V/Div knob offers, exploits this to full effect.
For low signal levels, using a 1:1 probe (i.e. zero signal attenuation) instead of the standard 10:1 (i.e. 10x attenuation) probe would result in both displaying and measuring less noise.
An example of changing these first two conditions to optimise performance are shown in the screenshots below. A 20mVpp sinewave is applied: with the standard 10:1 probe we measure 10mV noise – 50% of the signal value. This could cause an engineer to hunt around the circuit to look for problems.
Standard 10:1 probe on a low-amplitude signal results in a visually measurement of 10mVpp noise.
But by changing to a 1:1 passive probe and scaling the display correctly, we dramatically change the measured noise to less than 2mV. Remember that this is an observed value of peak-to-peak noise, not the measured RMS value of noise that is published in oscilloscope datasheets.
The same signal with a 1:1 passive probe and using full scale reduces the visually measured noise to 2mVpp.
The next oscilloscope setting we can change to affect the display is memory length. Some scopes allow you change the number of points of memory used in the scope’s horizontal or acquisition menu. Why does the first trace below (with 1,000 points of memory) look less noisy than the second (with 10 Million points)?
An oscilloscope with 1,000 points of memory, signal shows some small deviations at low and high levels.
The same oscilloscope and the same signal, but now with with maximum memory (10 million points) enabled – this shows a much “thicker” or “noisier” trace.
The answer is two-fold: each point of memory represents a different sample of the signal: with 10M points the sample rate is 5Gs/Sec (5 billion times per second), with 1,000 points of memory, the sample rate drops to just 1Ms/Sec). So with less data to display, there is less data compaction - the process of displaying many sample points on the same one-pixel width (but possibly many pixels high).
With data compaction, if just one of the samples is a different amplitude value then we see a thicker or “noisier” trace. Thus “Deep” or long memory settings will look noisier! Keysight’s MegaZoom technology means you have all the memory all the time, so maximum memory depth is always available and highest sample rate is also always achieved. (Note that it is possible to reduce the sample rate on a waveform by capturing a longer acquisition time, and zooming in to view the part of the signal of interest).
The second reason is the sample rate relates to the digital bandwidth of the scope: with a sample rate of only 1M samples per second, Nyquist’s rule states that the maximum frequency of the signal that can be reconstructed is just 500kHz!
Now we have seen the effects of digital bandwidth on reducing the appearance of noise, we can also see the effect of RF or analogue bandwidth. RF engineers will be familiar with the concept of noise power from a simple load being equal to kTB, where k is Boltzmann's constant, T is the absolute temperature of the load (for example a resistor), and B is the measurement bandwidth – so the more bandwidth being measured, the more noise we will induce or capture.
As we saw in the first table of RMS noise measurements, some manufacturers limit the bandwidth of their oscilloscopes from 1GHz down to just 150MHz at small signal input settings in order to improve the signal-to-noise performance. Perhaps their analogue front-end may not have sufficient gain-bandwidth without suffering from amplification problems such as compression, clipping, non-linearity or poor phase performance. Other manufacturers may specify certain parameters or even demonstrate their products with a 20MHz hardware filter enabled. As ever, be aware that “apples for apples” comparisons are being made.
Oscilloscope users can also implement a hardware filter on their scope, a 20MHz setting is common to pretty much every oscilloscope on the market. This harks back to the 20MHz analogue oscilloscopes engineers grew up with, and is still the measurement bandwidth used to qualify many products such as power supplies. As the previous text has discussed, the lower the bandwidth, the lower the noise. Obviously this can’t be used if the signals the engineer wants to view have any content above 20MHz.
Next we will look at the effect of another oscilloscope parameter that users may not expect to impact the appearance of noise: waveform update rate (the number of times per second a waveform is captured and displayed on the screen). Noise is Gaussian in distribution, so over an infinite period of time, noise will occur at some point from -∞ to +∞. The diagram below shows the dispersion of noise of a single capture, a longer capture will result in the same curve but will extend the outliers.
Dispersion of noise is Gaussian: histogram of occurrence in frequency (bottom), time-domain view (top).
A fast update rate, such as Keysight’s 1,000,000 waveforms per second, means more signals are captured and displayed. As each individual waveform has unique and random additive noise pattern, when these signals are overlaid, a larger number of waveforms would look “nosier”.
Full bandwidth and 1,000,000 waveforms per second, the noise appears to be 2mVpk-pk.
By using the hold-off function to reduce the waveform update rate of the scope (in fact to bring it down to the update rate of other scopes on the market with as few as 2,000 waveforms per second), we can reduce the appearance of noise. This is because fewer waveforms are being captured, and thus less data (regardless of being signal or noise) - and as noise is random and Gaussian, there are fewer “outliers” which appear to increase the peak-to-peak content.
Full bandwidth, but reduced waveform update rate: the peak-to-peak appearance of noise is cut by 50%.
But how important is noise anyway? If the scope has a slow update rate it will not capture unusual signal condition such as an error as often or even at all as a scope with a fast waveform update rate. So while a slower scope may appear to be less noisy, it is not helping the user as much as a fast scope which is showing more signal data but appears noisier!
Now we can look at some of the oscilloscope settings that can smooth the signal. An acquisition mode common in many oscilloscopes is Hi-Res mode (some call it ‘enhanced resolution’ or ‘ERES’). This uses over-sampling (also known as hyper-sampling) to create new quantisation levels (or “more bits”) by averaging adjacent samples. Note that this also reduces the digital bandwidth. In some scopes, the level of enhanced resolution can be controlled on a bit or half-bit resolution scale, on others, it is automatically processed depending on the time-base. Typically this enables a traditional 8-bit digitiser in a scope to produce results akin to a 12-bit scope.
Let’s say you have a 100MHz sinewave and your scope samples at 5Gs/Sec. The scope is oversampling 50x when you only need 4x or 5x to reconstruct the waveform accurately. Enhanced resolution takes the 4 adjacent sample points and averages them to create a new value – so instead of the actual acquisition being quantised and displayed as discrete levels of 2-2-3-3, it will show a value of 2.5 (2+2+3+3/4), effectively creating a value you would get if the oscilloscope’s digitiser had another bit (another quantisation level). The great thing about High-resolution mode is that it can be used on any signal type, either single-shot, or repetitive.
Oscilloscope set in normal acquisition mode (8 bit) on the left. On the right, the scope is in High-Resolution mode which creates a trace with a cleaner signal.
There is also the averaging acquisition mode: while this is good for repetitive signals, it can’t be used on signals such as serial data. Averaging takes the entire acquisition data (or display data) and averages this with other acquisitions, the effect of this is the first few acquisitions look noisy, then gradually fade to a single “clean” signal. This is good for seeing just a repetitive signal, but does not permit you to view any rare events, such as a signal spike or change in timing as they would be averaged out.
The first generation of digital scopes illuminated every pixel where the signal was, with no intensity differentiation. This gave them a very “blocky” display, as shown below.
Modern digital oscilloscopes offer intensity grading where the scope automatically increases or decreases the brightness of the trace depending on the frequency of its occurrence, to try and replicate the display characteristic users saw in analogue scopes. But because digital scopes can capture much more signal detail with more rare and unusual events (which are often the things engineers want to see!) we can turn adjust waveform intensity to 100%: this illuminates each pixel with equal intensity, regardless of frequency of its occurrence. This creates a “fat signal” like the first generation of digital scopes.
A noisy signal with 100% signal intensity – all pixels are illuminated with equal intensity.
By turning down the intensity, we get a Gaussian-spread of illumination, depending on the frequency of the occurrence of the signal like an analogue scope.
The random noise elements of the signal are no longer as intense, the main signal content stands out.
Users have a lot of control over how the view, interpret and measure signals depending on the probing technique, the acquisition mode, the display mode and other scope settings. Optimising these reduces or eliminates the impact of the oscilloscope on the signal and can assist in resolving signal integrity issues within the user’s system.
For more information, Keysight have a dedicated application note.