Sub: Rigol's DHO800 Oscilloscope (Gibbs Effect & Aliasing Misunderstanding)

[Mechatrommer]

right! playing with sim tonight to design 3rd order elliptic filter (analog/physical) filter (about 60MHz cutoff)... no gibbs effect on sim. so there must be a special one to build a physical filter that can cause "gibbs like" effect. fwiw..

[wasedadoc]

right! playing with sim tonight to design 3rd order elliptic filter (analog/physical) filter (about 60MHz cutoff)... no gibbs effect on sim. so there must be a special one to build a physical filter that can cause "gibbs like" effect. fwiw..

Can you plot the phase versus frequency?

[gf]

right! playing with sim tonight to design 3rd order elliptic filter (analog/physical) filter (about 60MHz cutoff)... no gibbs effect on sim. so there must be a special one to build a physical filter that can cause "gibbs like" effect. fwiw..

I do see overshoot/ringing caused by attenuation of higher harmonics. Why do you classify it as "no Gibbs"? Symmetry is not the criterion. Gibbs ringing introduced by a filter when applied to an ideal square wave can be either symmetric or asymmetric, depending on the filter's phase response.

The only difference between the filters that produce the step responses in figure1 and figure2 is their phase response. Both filters have the same magnitude response (figure3), i.e. attenuation of harmonics is the same. Figure1 is minimum phase and figure2 is linear phase. A causal, stable IIR cannot have linear phase, though.

[nctnico]

I still don't agree with attributing ringing due to low pass filtering a non-perfect square (IOW: a square wave as it exists in the real world) wave to Gibbs effect. The way I read the mathematical description is that Gibbs ears occur because you can't create a perfect square wave from adding sine waves (a Fourier series) which in turn is because in the mathematical world a square wave is not a continuous function. You could call Gibbs ears a 'residu'. However, a discontinuous square wave can exist in the digital world after sampling a square wave or other waveforms with  steep edges in relation to the samplerate. When the sampled signal is brought back to the real world using sin x/x reconstruction (for example), you'll get Gibbs ears as a displayed artefact.

[Mechatrommer]

right! playing with sim tonight to design 3rd order elliptic filter (analog/physical) filter (about 60MHz cutoff)... no gibbs effect on sim. so there must be a special one to build a physical filter that can cause "gibbs like" effect. fwiw..

I do see overshoot/ringing caused by attenuation of higher harmonics. Why do you classify it as "no Gibbs"? Symmetry is not the criterion. Gibbs ringing introduced by a filter when applied to an ideal square wave can be either symmetric or asymmetric, depending on the filter's phase response.

The only difference between the filters that produce the step responses in figure1 and figure2 is their phase response. Both filters have the same magnitude response (figure3), i.e. attenuation of harmonics is the same. Figure1 is minimum phase and figure2 is linear phase. A causal, stable IIR cannot have linear phase, though.

i only focus on pre ringing part, the unrealistic one. the post ringing overshoots are normal, so i will not question that part. Not just in electrical signal, but also normal in physical control system n mechanical. Your car doesnt get bumpy before hitting the bump, otherwise thats paranormal.

[bdunham7]

Your car doesnt get bumpy before hitting the bump, otherwise thats paranormal.

As I pointed out earlier, the causality argument is a trap, a red herring, a.. well, whatever.  There is no theoretical causality issue if the display on the screen doesn't appear until long (relatively speaking) until the event is over.  In the case of  just the analog filter, this means there could be 'pre-ringing' to the extent that the portion of the signal that represents the bulk of the step is delayed in the filter while other components get through faster.  This can be demonstrated with with real components, although I don't have an all-analog example that would give you nice pronounced and symmetrical Gibbs ears.

[gf]

i only focus on pre ringing part, the unrealistic one. the post ringing overshoots are normal, so i will not question that part. Not just in electrical signal, but also normal in physical control system n mechanical. Your car doesnt get bumpy before hitting the bump, otherwise thats paranormal.

Minimum phase filters don't pre-shoot. The traditional all-pole filters like Bessel, Butterworth, Chebycheff type1 are minimum phase, IIRC. But not every IIR (or analog) filter is mininimum phase. If you linearize the phase response of a (say) Butterworth filter with additional allpass filters you can even approximate a linear phase response in the analog domain.

See also https://cdn.teledynelecroy.com/files/whitepapers/group_delay-designcon2006.pdf

[iMo]

As it has been discussed earlier, the pre-ringing has nothing to do with hardware.
The scope samples a couple of samples, and in order to draw a "continuous solid yellow signal" on the LCD display it uses sinc interpolation in math (it creates additional LCD points based on the "sinc" function). You may see it on previous picture in the replay #220 (the Bodnar 312.5MSa/s) where there were only 20 samples taken (3ns from sample to sample, 60ns wide screen, therefore 20 points to interpolate) and the graphical function based on the "sinc" interpolated those 20points into a "continuous picture" you see on the screen. Therefore you see the "pre-ringing". With more points (samples) taken you do not see the pre-ringing.
You cannot get "pre-ringing" when the electrical signal has not entered the filter.

Thus the "sinc" here has nothing to do with a ringing signal, "filtering in hw" or "dsp". It is a graphical interpolation method/tool they have chosen to create continuous visual lines on the LCD screen. You may interpolate number of bananas exported each month within a year and you will get the same pre-ringing and post-ringing with the sinc interpolator when having 12 numbers..

PS: below the picture, the display even says "16points" taken in total, 3ns apart over the entire screen width (info on top)..
Look at the nodes where the wave changes its phase - aprox six times (20ns/3), 3ns apart, mind the time scale below - there are the "6 real physical points sampled", all yellow between them has been "created artificially by the graphical Sinc interpolation".
The ringing before and after the rising edge is made by the sinc graphical interpolation such you get a solid line instead of only 16 points placed on the LCD screen..

[Mechatrommer]

right! playing with sim tonight to design 3rd order elliptic filter (analog/physical) filter (about 60MHz cutoff)... no gibbs effect on sim. so there must be a special one to build a physical filter that can cause "gibbs like" effect. fwiw..

Can you plot the phase versus frequency?

i just got home, see attached. not sure what it tells. but anything larger than 40MHz is more than 180deg phase?

[bdunham7]

Here is a rather lame attempt to throw something together with assorted junk within reach on my messy bench.  First screenshot is a square wave from an FY6900 through a Mini Circuits BLP70+.  Second is the same except a small modification was made.  Obviously I'd have to tweak my modification to get nicer looking symmetric results.

Edit:  I did tweak the modification a bit to look 'nicer', but I obviously still don't have a symmetrical Gibbs ears appearance.  Note that even 'real' Gibbs ears are not always perfectly symmetric.

https://www.minicircuits.com/pdfs/BLP-70+.pdf

[wasedadoc]

right! playing with sim tonight to design 3rd order elliptic filter (analog/physical) filter (about 60MHz cutoff)... no gibbs effect on sim. so there must be a special one to build a physical filter that can cause "gibbs like" effect. fwiw..

Can you plot the phase versus frequency?

i just got home, see attached. not sure what it tells. but anything larger than 40MHz is more than 180deg phase?

To see if linear phase or not, plot both axes as linear, not log frequency.

Large phase angles per se are not necessarily a problem.  Think about it.  To preserve wave shape all frequencies must be delayed by the same amount of time.   That means phase angle increases linearly with frequency.  Going past 180 or even past 360 degrees is not an automatic no-no.

[Mechatrommer]

Can you plot the phase versus frequency?

i just got home, see attached. not sure what it tells. but anything larger than 40MHz is more than 180deg phase?

To see if linear phase or not, plot both axes as linear, not log frequency.

Large phase angles per se are not necessarily a problem.  Think about it.  To preserve wave shape all frequencies must be delayed by the same amount of time.   That means phase angle increases linearly with frequency.  Going past 180 or even past 360 degrees is not an automatic no-no.

unfortunately tina-ti doesnt have linear scale frequency plot. but eyeballing it, the curvy shape could be linear phase when plotted on linear scale x-axis? here's the group delay plot, so i guess its linear phase up to 3MHz? or with some compromise up to 40MHz? see attached. https://audiojudgement.com/group-delay-explanation-measurement-audio/

[ebastler]

unfortunately tina-ti doesnt have linear scale frequency plot.

Click the "Log" button in the lower right, next to the "Resolution" entry field.

[Mechatrommer]

unfortunately tina-ti doesnt have linear scale frequency plot.

Click the "Log" button in the lower right, next to the "Resolution" entry field.

[mawyatt]

Here's something that folks might find interesting. Gibbs Ears created on DHO824 with SDG2042X AWG utilizing Sinewave and Harmonics feature.

Best,