Rigol HDO1000 and HDO4000 12bit oscilloscopes launched in China

[Fungus]

Anybody have contacts at Rigol?

AFAIK they never gave any signs of reading EEVBLOG forums.

[mawyatt]

Yeah, think as tv84 and 2N3055 mentioned it's very early in the lifeline of this new HD DSO.

They did spend significant $ on the custom chip developments, so maybe Rigol decided to release early to save debugging $, and have early adopters help with the debug? This could be their business release plan for these new HD DSOs!!

We were very interested in the Channel & 12 bit ADC Linearity, this would be mostly for use with FFTs. Now raises the question in light of what we know, is linearity a factor for this HD DSO?? 

Anyway, it is what it is and has damped our initial enthusiasm some, being not as well polished as we had hoped.

Best,

[2N3055]

... If you put scope in Auto memory mode, it will have high enough sample rate and you won't see this..

This is one of the problems, my scope was in AUTO memory mode, as i said I use most of the time auto setting, but maybe this is not a good habit. You should pay attention all the time at this parameter as it can change over a lower limit. The automatic memory management is not so reliable and in some cases, like this, it will not allocate the correct memory back to the necessary number.

In my opinion Auto mode is only way not to go mad in everyday work.  You go manual when you need.

[skander36]

In my opinion Auto mode is only way not to go mad in everyday work...

Good point! 

[2N3055]

Yeah, think as tv84 and 2N3055 mentioned it's very early in the lifeline of this new HD DSO.

They did spend significant $ on the custom chip developments, so maybe Rigol decided to release early to save debugging $, and have early adopters help with the debug? This could be their business release plan for these new HD DSOs!!

We were very interested in the Channel & 12 bit ADC Linearity, this would be mostly for use with FFTs. Now raises the question in light of what we know, is linearity a factor for this HD DSO?? 

Anyway, it is what it is and has damped our initial enthusiasm some, being not as well polished as we had hoped.

Best,

I will exercise a bit more restrain before passing judgment and give them a bit more time to start releasing some FW updates.  When R&S released half baked RTB2000 years ago, everybody was full of understanding for almost a year, until R&S made it decent enough..

OTOH I knew it was too early to release it, and never had early adopter fever... But Rigol has a chance now to prove their mettle... or not. It's up to them.

[gf]

The mirror reminds me on the array returned by Matlab's/Octave's fft() function. It contains DC and the positive frequency bins in the first half of the array, followed by Nyquist and the negative frequency bins. For a real-valued signal, plotting¹⁾ the log magnitude of this array shows a mirrored spectrum, as we are seeing it here. However, FFT on scopes usually does not display any negative frequencies, but hides them. I would not a priori overstate the bug -- it could be an easy to fix careless mistake in the software.

¹⁾ E.g. plot(20*log10(abs(fft(samples))))

[Martin72]

.. But Rigol has a chance now to prove their mettle... or not. It's up to them.

I'm toying with the idea of getting the DHO1074, playing with it for a while, and giving it away again.
It costs only 1200€ in the rigolshop and would be here in a few days.
In the time I could again annoy the German Rigol support, as I had once done with the MSO5000.
Seriously, it would be interesting to play a little bit with it.
I like to have things here instead of reading/watching and the 4000 is a little bit too expensive for just playing.
Although...if batronix would lend me one....

[Fungus]

They did spend significant $ on the custom chip developments, so maybe Rigol decided to release early to save debugging $

I think last summer the marketing department planned a November release, because November/December is when most people buy oscilloscopes.

Then... November came along and it wasn't 100% finished because software never is. The rest is history.

Anybody who's ever worked in a company which develops products will have lived through this story ^^^

[2N3055]

I would not a priori overstate the bug -- it could be an easy to fix careless mistake in the software.

Well, that is a definition of a software bug. ...

[2N3055]

They did spend significant $ on the custom chip developments, so maybe Rigol decided to release early to save debugging $

I think last summer the marketing department planned a November release, because November/December is when most people buy oscilloscopes.

Then... November came along and it wasn't 100% finished because software never is. The rest is history.

Anybody who's ever worked in a company which develops products will have lived through this story ^^^

I agree.. That is what happened. Now we wait and see.. Fact that they rushed it out is one thing.
But if people keep buying it like crazy despite bugs, because it's cheap, they have very little incentive to fix things..

[Fungus]

Although...if batronix would lend me one....

Batronix said they wouldn't be selling them because Rigol won't let them so long as they're selling Siglents.

Let's check...

Nope. No HDO series on Batronix. 

[2N3055]

Although...if batronix would lend me one....

Batronix said they wouldn't be selling them because Rigol wouldn't let them so long as they were selling Siglents.

Let's check...

Nope. No HDO series on Batronix. 

No DHO1000... They are selling 4000 tough... So weird....

[Martin72]

Yep, DHO4000 is on sale, DHO1000 not.
And Conrad Electronic offers the DHO1000 for commercial customers only....

[switchabl]

The mirror reminds me on the array returned by Matlab's/Octave's fft() function. It contains DC and the positive frequency bins in the first half of the array, followed by Nyquist and the negative frequency bins. For a real-valued signal, plotting¹⁾ the log magnitude of this array shows a mirrored spectrum, as we are seeing it here. However, FFT on scopes usually does not display any negative frequencies, but hides them. I would not a priori overstate the bug -- it could be an easy to fix careless mistake in the software.

¹⁾ E.g. plot(20*log10(abs(fft(samples))))

I am not sure that the part to the right of f_s/2 is actually a perfect mirror image. If you look at the second screenshot in https://www.eevblog.com/forum/testgear/rigol-hdo1000-and-hdo4000-12bit-oscilloscopes-launched-in-china/msg4668886/#msg4668886 the image seems to be a few dB down. Also in this case sampling rate is 200 kSa/s (acquistion) -> 256.41 kSa/S (FFT). That suggests an interpolation/resampling filter. In the next screenshot there is downsampling 2 GSa/s -> 25.64 MSa/s, which actually makes some sense to improve resolution at low frequencies.

Unlike the 2x upsampling case, that actually has potential for introducing less-obvious images < f_s / 2. So here's hoping they got that right.

[Fungus]

Remember that China is still in all sorts of lockdowns. It's easy to sit here saying they ought not to have released it and that they should be fixing it, but... we have no idea what's going on over there.

[tv84]

But if people keep buying it like crazy despite bugs, because it's cheap, they have very little incentive to fix things..

True words.

I think this forum is a niche... Most people only look at the wallet.   

But, I'm still betting on Rigol to fix all this stuff. Looking at the odds, I'm might win a load of $$$.   

[Njk]

Shouldn't the FFT stop at 100kHz if you have the 200kSa/s ? That would solve the picture at 200 kSa/s and save Rigol's honor. 

But then, why does the 400kSa/s show the same mirroring?   

These are the moments where you need an engineer next to the GUI programmers or in the betatesters group...

I wonder how such a bug passed the FFT QC tests?!? 

It seems Rigol uses some code that is shared. So the bugs are shared as well. Early I've noticed the same inconsistent Sa/s values for FFT sampling rate vs. scope's sampling rate in the 1054Z model. No wonder the HDO series are also affected.

https://www.eevblog.com/forum/testgear/rigol-ds1000z-series-buglist-continued-(from-fw-00-04-04-03-02)/msg4536227/#msg4536227

[Martin72]

It's easy to sit here saying they ought not to have released it and that they should be fixing it, but... we have no idea what's going on over there.

At least they could build the scopes and ship them to the world, despite lockdown.

[Martin72]

I think this forum is a niche...

You underestimate this forum my friend...

[gf]

I am not sure that the part to the right of f_s/2 is actually a perfect mirror image. If you look at the second screenshot in https://www.eevblog.com/forum/testgear/rigol-hdo1000-and-hdo4000-12bit-oscilloscopes-launched-in-china/msg4668886/#msg4668886 the image seems to be a few dB down. Also in this case sampling rate is 200 kSa/s (acquistion) -> 256.41 kSa/S (FFT). That suggests an interpolation/resampling filter. In the next screenshot there is downsampling 2 GSa/s -> 25.64 MSa/s, which actually makes some sense to improve resolution at low frequencies.

Unlike the 2x upsampling case, that actually has potential for introducing less-obvious images < f_s / 2. So here's hoping they got that right.

You are right, if the FFT sample rate is not 200kSa/s, but 256.41 kSa/s, then at least this image can't be a mirror.
Actually only the 3rd image with 2 GSa/s -> 25.64 MSa/s decimation looks OK.
The 4th image with 500 MSa/s -> 5.26 MSa/s decimation has too many spurs, IMO.
Decimation can make sense to improve resolution, by why such weird sample rates like 25.64 MSa/s (decimation by factor ~78 which is neither decimal nor power of two)?

[switchabl]

Maybe they are trying to line up the bins of a power-of-two FFT with the center/span setting? I kind of hate that approach though, CZT is a much cleaner way to get arbitrary spans (even at the cost of somewhat longer FFTs).

I hadn't looked at the fourth image properly. As far as I can tell the only difference between third and fourth is sample rate. So this is genuinely worrying.

It might make sense to download the raw samples at those settings and compare against an FFT in Matlab or Python.

[Martin72]

Although...if batronix would lend me one....

Request made...

[aurel]

I tried to reproduce this FFT issue on my HDO1K and I think I nailed down the bug.

First of, in Auto memory mode I don't see any issue, so I did all the tests in 10k manual memory mode and centered the FFT at 100 kHz.

- Setting the time base to 2 ms/div results in a 500 kSa/s for both the main sampling rate and the FFT window => no visible issue
- Setting the time base to 5 ms/div results in a 200 kSa/s for both the main sampling rate and the FFT window => FFT display is cut after the 100 kHz point (above Nyquist) => no visible issue
- Setting the time base anywhere between 2 and 5 ms/div is the interesting part ! For example setting to 4.5 ms/div results in a 200 kSa/s main sampling rate but a 222.22 kSa/s in the FFT window => FFT display is cut after the 111 kHz point (above Nyquist of the wrong sampling frequency in the FFT window) => there's your issue !

So the bug is that the sampling rate used in the FFT window is computed straight from the timebase, rather than computer from a "rounded up" timebase like it is done for the main sampling frequency.
This should be quite an easy fix for Rigol (let see how much time it takes them !)

For now, to use the FFT in a safe way, you should just avoid manually setting any non "round" timebase, just stay on a 1-2-5 timebase (which is what most users will do anyway).

[Fungus]

So you have to manually set the time base to something weird for it to appear?

[Fungus]

- Setting the time base to 5 ms/div results in a 200 kSa/s for both the main sampling rate and the FFT window => FFT display is cut after the 100 kHz point (above Nyquist)

You mean 'cut' like in this screenshot, nothing displayed?