Choosing between entry-level 12-bit DSOs

[gf]

please note what you keep talking is a combination of sample rate and Sinc reconstruction. turn off Sinc and your point will become moot, granted signal can be a little weird when spectral content contains element more than half the sampling rate, using "Line" or vector plot. but not as severe when your turn on Sinc.

Umm... How would you describe, say, a 170 MHz sine signal sampled at 312 MSa/s, when shown in dot mode or with linear interpolation? It would look severely misleading in my book. Sinc interpolation is not the issue here.

170 MHz folds back to 142 MHz in the first Nyquist zone. So it looks like you had sampled a 142 MHz sine wave.

[Mechatrommer]

Umm... How would you describe, say, a 170 MHz sine signal sampled at 312 MSa/s, when shown in dot mode or with linear interpolation? It would look severely misleading in my book. Sinc interpolation is not the issue here.

and you keep showing the worst case scenario to prove your point when user really fucked up trying to probe >170MHz using 4CH cluelessly. in reality we will avoid such test condition. we will turn on only 2 or 1 channel in operation. even if we have to use all 4CH, when in doubt, we can switch to 1CH to probe the suspicious high frequency signal. granted its not foolproof, thats why you have to gain knowledge and theories as you go, otherwise if you keep doing your setup example you mentioned again and again and made mistake again and again, this job is actually not for you, ymmv.

170 MHz folds back to 142 MHz in the first Nyquist zone. So it looks like you had sampled a 142 MHz sine wave.

if you really try to probe 170MHz fundamentals at 312 MSa/s really you have something wrong within you, not the scope, scope is just a tool. usually spectral content in excess of Nyquist limit are harmonics whose magnitudes are much smaller hence aliasing will not affect much the fundamentals shape, unless your circuit is oscillating unexpectedly. ymmv.

analogy is manual transmission car vs automatic transmission car. manual is not foolproof, automatic transmission is, but for some people like us, we still prefer manual and know what we are doing, and also know what we should not do. because we need performance not available in automatic cars. ymmv.

[ebastler]

and you keep showing the worst case scenario to prove your point when user really fucked up trying to probe >170MHz using 4CH cluelessly. in reality we will avoid such test condition. we will turn on only 2 or 1 channel in operation. even if we have to use all 4CH, when in doubt, we can switch to 1CH to probe the suspicious high frequency signal. granted its not foolproof, thats why you have to gain knowledge and theories as you go, otherwise if you keep doing your setup example you mentioned again and again and made mistake again and again, this job is actually not for you, ymmv.

Not sure where the agitation comes from. Feel free to choose frequencies you consider more realistic. My point was just that aliasing remains aliasing, also in dot mode or with linear interpolation.

[Mechatrommer]

Not sure where the agitation comes from. Feel free to choose frequencies you consider more realistic. My point was just that aliasing remains aliasing, also in dot mode or with linear interpolation.

sorry sometime i got mixed up with 2n**55 posts, but i read his argument all too often. so its somewhat similar to your post... Sinc reconstruction CAN exaggerate aliasing more than Line plot, that was my point. we discussed Gibbs effect that will not be seen on Line plot, esp on fast rising edge square signal. but my reply still stands, if a user wants to probe fundamentals in violating of nyquist limit, then the user got problem, or he/she may not know enough, not the scope... ymmv. otoh talking about properly designed BW limited scope, it will not be able to see ill behaved oscillation in circuit, sometime aliasing can give enough suspicion something happened, ymmv.

[shapirus]

But facts are facts and DHO900 is hardware that is a shame of Rigol and should not have left prototype phase. It is a 250MHz scope (claimed) that in worst case samples at 156MSps/s which is FINRSI type of specmanship.
And since many people hack their DHO800 to 900 that defect is applicable to 800 too.

When does it switch to 156Msa/s?
(other than zooming out horizontally to >= 500 us/div)

[2N3055]

p

But facts are facts and DHO900 is hardware that is a shame of Rigol and should not have left prototype phase. It is a 250MHz scope (claimed) that in worst case samples at 156MSps/s which is FINRSI type of specmanship.
And since many people hack their DHO800 to 900 that defect is applicable to 800 too.

When does it switch to 156Msa/s?
(other than zooming out horizontally to >= 500 us/div)

900 switches when all 4 ch on and MSO

[Mechatrommer]

But facts are facts and DHO900 is hardware that is a shame of Rigol and should not have left prototype phase. It is a 250MHz scope (claimed) that in worst case samples at 156MSps/s which is FINRSI type of specmanship.
And since many people hack their DHO800 to 900 that defect is applicable to 800 too.

When does it switch to 156Msa/s?
(other than zooming out horizontally to >= 500 us/div)

when you turn on all 4CH and 16bit LA and expect you can probe 250MHz with the setup

[BillyO]

Question for those that know.

With the DHO900 (non 'S' model) can you still do automated Bode plots with an external Sig. Gen.?

[Fungus]

I've mentioned before that I think Rigol should drop to 125Mhz when more than 2 channels are active.

(Probably a very difficult hack...)

[Mechatrommer]

With the DHO900 (non 'S' model) can you still do automated Bode plots with an external Sig. Gen.?

offically? no. but unofficially, other than hacking it to S model in vendor.bin, you can make PC SW to control both and plot in PC, its possible.

[shapirus]

900 switches when all 4 ch on and MSO

I see. Yeah that's kinda expected.
A hacked 800 doesn't do it, it stays on 312.5.

It however starts to fail well below Nyquist (which is 156.25 MHz at this sampling rate): the displayed signal begins to have lower frequency noise at about 100 Mhz and it becomes next to useless at ~130 MHz.

Frequency counter works only up to 142 MHz.

Here's a 30s frequency sweep from 70 MHz to 170 MHz. Nevermind the intermittent glitches and some jitter, they may be a result of the signal gen that I'm using (TinySA Ultra). A 50 Ohm feed-through terminator was used on the scope's BNC input.

Memory length was deliberately set to a maximum to get a minimal waveform update rate to get rid of the display persistence effects, which are not desirable in this case.

(the 156.25 MHz point is at ~00:25)

[Fungus]

It however starts to fail well below Nyquist (which is 156.25 MHz at this sampling rate): the displayed signal begins to have lower frequency noise at about 100 Mhz and it becomes next to useless at ~130 MHz.

sin(x)/x will start to go wrong at sample rate / 2.5

That video shows exactly what the math says it should.

[shapirus]

sin(x)/x will start to go wrong at sample rate / 2.5

That video shows exactly what the math says it should.

Yeah, and it actually works fine up to 500 MHz at 1.25 GSa/s (save for the greatly reduced amplitude because of the input filter).

[ebastler]

900 switches when all 4 ch on and MSO

I see. Yeah that's kinda expected.
A hacked 800 doesn't do it, it stays on 312.5.

No, it's not really expected. The DHO900 is the only scope I'm aware of which behaves this way.

In typical DSOs, the analog sampling rate is limited by the ADC. (Which is shared across two or four channels, hence sampling rate per channel drops as you activate more analog channels.) But the digital channels are read separately -- they do not need the ADC, and hence don't eat into the ADC bandwidth available for analog sampling.

In contrast, the DHO900 series seems to be limited by the bandwidth of the FPGA/DRAM interface. It uses the same ADC as the larger DHO1000 series, yet runs it at only 1.25 instead of 2 GHz. And when you want to read digital data in parallel, that eats into the available sampling rate.

A hacked DHO800 will behave the same -- as soon as you actually connect and enable a digital probe. Mechatrommer has done that and can comment.

[KungFuJosh]

I'm using (TinySA Ultra).

The TinySA Ultra (as stated) is not the issue. I had clean signals up to 841MHz on my SDS2504X+.

[shapirus]

The TinySA Ultra (as stated) is not the issue. I had clean signals up to 841MHz on my SDS2504X+.

With frequency sweeps?

[KungFuJosh]

The TinySA Ultra (as stated) is not the issue. I had clean signals up to 841MHz on my SDS2504X+.

With frequency sweeps?

Nothing automated, I stepped up 1MHz or 10 MHz at a time until the signal went wonky. Up to 841MHz was fine, it got goofier after that.

[shapirus]

Nothing automated, I stepped up 1MHz or 10 MHz at a time until the signal went wonky. Up to 841MHz was fine, it got goofier after that.

Try it with a sweep. It seems that it occasionally (maybe between steps) resets to the base 30 MHz frequency.

[BillyO]

Try it with a sweep. It seems that it occasionally (maybe between steps) resets to the base 30 MHz frequency.

Yes, I found the glitching too.

[KungFuJosh]

Latest firmware update was 5 days ago on the TinySA Ultra. I wonder if they fixed that bug. There's been significant updates apparently since the version I had. I was using 1.4-120, now they're at 1.4-159.

[Mechatrommer]

900 switches when all 4 ch on and MSO

I see. Yeah that's kinda expected.
A hacked 800 doesn't do it, it stays on 312.5.

you havent activated the LA in your video. if that D (with number 1-16) button on the bottom screen becomes green and you have many horizontal green lines on the screen, then your sample rate will become halved again iirc (156.25MSps)

It however starts to fail well below Nyquist (which is 156.25 MHz at this sampling rate): the displayed signal begins to have lower frequency noise at about 100 Mhz and it becomes next to useless at ~130 MHz.

sin(x)/x will start to go wrong at sample rate / 2.5

1) to be precise, anything more or equal to Sample rate / 2 will not be reconstructed correctly, i havent found literature saying exactly 2.5, i guess 2.5 figure is just a safer bet from scope manufacturers for Sinc interpolation to be correct, or the right BW limit roll off for anything Sr / 2 and higher to be suppressed to an acceptable level. iirc even signal with Sr / 2.1 can be reconstructed correctly, given the condition... "thou shalt not have any harmonics (higher frequency elements) equals to or more than the sample rate / 2, not even the slightest" which comes to the second statements.

2) we usually try to prove Sinc failure with fundamentals or pure sine signal very close to Sr / 2 but infact if front end BW is not brickwalled at sample rate / X ... (X > 2, eq 2.1, 2.2, 2.5 etc) Sinc will still produce "not exactly the original" signal on screen due to presence of non zero harmonics above nyquist limit, even if fundamentals is way below the limit. its just its not so obvious when harmonics magnitudes are small. a good example is low frequency square wave signal but very high rise/fall time, on larger time scale it seems ok, until we zoom closely into the rise/fall time portion (also read as "artefacts").

so please bear in mind Sinc inaccuracy doesnt only happens at fundamentals near nyquist limit, what matters is the highest frequency content presence, the bigger the magnitude the worst the reconstruction will be. thats why some people swear by the "properly BW limited" scope.

unfortunately they sometime wreak havoc from reality, if your circuit is already BW limited (no oscillation) then it should be fine... but if you expect your circuit to contain higher harmonics at significantly high magnitudes and you choose to probe with <= Sr / 2 anyway, then imho its you that have problem, not the scope. otoh built in properly BW limited scope will not be able to sense circuit misbehavior above Sr / 2, everything will just look fine, until the problem manifests/propagates itself on another portion of the circuit. fwiw.

[Fungus]

1) to be precise, anything more or equal to Sample rate / 2 will not be reconstructed correctly, i havent found literature saying exactly 2.5

Anything equal to sample rate/2 can NOT be reconstructed.

Think: You might sample on the zero crossing points.

Now work from there to something a tiny bit below that ... you'll go between zero crossings and peaks over time, and get an AM effect as seen in that video.



2.5 isn't a mathematically derived number, it's what works in practice.

[gf]

Now work from there to something a tiny bit below that ... you'll go between zero crossings and peaks over time, and get an AM effect as seen in that video.

It looks like most interpolation filters are realized as Nyquist Filters, having a symmetric N*SR...(1-N)*SR transition band, where 0<N<0.5, and SR is the sample rate. This is fine for frequencies below N*SR, but leads to the observed "AM effect" for frequencies in the N*SR...0.5*SR range.

An interpolation filter with a N*SR...0.5*SR transition band would avoid that, and only reduce the amplitude of frequencies in the transition band. However, the drawback is that it requires about 4x as many non-zero filter taps as a Nyquist filter with the same N.

And in general, the closer you want to get to 0.5 with N, the more filter taps you need. There is no fixed N set in stone, but there are practical considerations/limits.

[Fungus]

Yes, but with sin(x)/x filter (a very practical one) the 2.5x rule is about right.

[gf]

Yes, but with sin(x)/x filter (a very practical one) the 2.5x rule is about right.

2.5 (i.e. a transition band of 0.4...0.6 * sample_rate) seems to be a "typical" value for DSOs.
But it is of course possible to design interpolation filters with a narrower or wider transition band.

In the video, a noticeable wobble already started to appear at 100 MHz (or approx. sample_rate/3).
So it seems that the DHO800 uses a softer interpolation filter with a wider transition band (at least at 312 MSa/s -- this may be different at higher sample rates).