Rigol DHO804 Test and Compare Thread

[gf]

Once with 50ohm termination, once blank 1Mohm, full bandwidth (200Mhz), 20Mhz limited.
Then once at 1µs/div, 1ms/div.

Hmmm... Why is there so much difference between 1.25GSa/s and 625MSa/s?

[Randy222]

On the DHO804, why does the scope lose signal sync if the vertical scale is increased?
Example;
Ch1 on the internal 1kHz 3v peak signal, set trigger to 1v, but then dial up the scale to 10v/div. The trigger does not change, so why the scope lose sync on the square wave (dancing signal)?

[2N3055]

Once with 50ohm termination, once blank 1Mohm, full bandwidth (200Mhz), 20Mhz limited.
Then once at 1µs/div, 1ms/div.

Hmmm... Why is there so much difference between 1.25GSa/s and 625MSa/s?

1µs/div as opposed to 1ms/div i would presume, not the sampling rate itself?

[Martin72]

Hmmm... Why is there so much difference between 1.25GSa/s and 625MSa/s?

If I reduce it from 1.25GSa/s  to 312MSa/s(by adding channels) at 1µs/div, nothing happens.
As the memory depth increases, so does the noise, see pictures (Auto, 10M, 50Mpts, 1Mohm termination, full bandwidth).
I'll take a closer look tomorrow or later.

[gf]

As the memory depth increases, so does the noise, see pictures

Very strange. The standard deviation of noise samples should not depend significantly on the memory depth (except when the number of samples is very small).

Could this be related to the mapping of the samples to the screen?
Are the measurements taken directly from the captured samples or from the screen data?

[Martin72]

I can't tell you how Rigol handles it.
I can test tomorrow whether the measured values are only taken from the screen.

[Martin72]

On the DHO804, why does the scope lose signal sync if the vertical scale is increased?
Example;
Ch1 on the internal 1kHz 3v peak signal, set trigger to 1v, but then dial up the scale to 10v/div. The trigger does not change, so why the scope lose sync on the square wave (dancing signal)?

Have a look at the data sheet under "Trigger Sensivity".
It says 0.5div, at 10V/div and a 3V signal it is no wonder why it no longer triggers.

[Martin72]

The standard deviation of noise samples should not depend significantly on the memory depth (except when the number of samples is very small).

Here is another picture at 10ms/div, automemory (50Ohm).


I'll get back to it tomorrow and "drive through" everything.

[Performa01]

As the memory depth increases, so does the noise, see pictures

Very strange. The standard deviation of noise samples should not depend significantly on the memory depth (except when the number of samples is very small).

Could this be related to the mapping of the samples to the screen?
Are the measurements taken directly from the captured samples or from the screen data?

The differences come from the very pronounced 1/f noise characteristic of scope frontends. Faster time base setting rises the lower bandwidth limit, thus cutting off a significant portion of the LF noise.

For example, a DSO with 1 GSa/s and 10 horizontal divisions (and automatic memory management) will have a record length of just 10 kpts and a lower bandwidth limit of 100 kHz at a time base setting of 1 µs/div.
The same DSO will have a record length of 10 Mpts and a lower bandwidth limit of 100 Hz at a timebase of 1 ms/div.

[gf]

The differences come from the very pronounced 1/f noise characteristic of scope frontends.

Oh sh*t, I didn't expect it to be that much
Your explanation makes sense, of course.

[Martin72]

That sounds sensible, the longer the time base, the lower the lower cut-off frequency, the higher the noise.
You know what, I'm trying it out now, from the living room (scope is on the router in the study).

[mawyatt]

The differences come from the very pronounced 1/f noise characteristic of scope frontends.

Oh sh*t, I didn't expect it to be that much
Your explanation makes sense, of course.

If the Rigol front end chip is implemented in pure CMOS then likely a relatively high 1/f corner. Scope front ends don't/can't take advantage of signal processing like CAZ and Chopping which help reduce 1/f noise as well as offsets.

Some utilize bipolar front ends, often SiGe or SiGe BiCMOS, which have good BW and low 1/f corners, but tend to be more expensive, with fewer chip foundry sources available.

Best,

[Martin72]

So...
The scope is now the only thing that was on in the room, even the light was off, so the noise values are correspondingly lower compared to before.
Input open, 1Mohm, full bandwidth, auto-memory.
Starting at 1µs/div.
The values increase up to 1ms/div, after that no more.

[pakakezu]

Maybe a bit unrelated. But i could not find the file format for the ARB waveform for the DHO924S. Looked in the programming and user manual. Can select csv from the menu but shows load fail if i load a saved trace.

I expected the arb file to be selectable and even uploadable from SCPI but only the arb mode can be seleceted without specifics.

If someone managed to output an arb trace can please post how the csv should look like?

[Performa01]

If the Rigol front end chip is implemented in pure CMOS then likely a relatively high 1/f corner. Scope front ends don't/can't take advantage of signal processing like CAZ and Chopping which help reduce 1/f noise as well as offsets.

This is only the harmless part of the story.

The major contribution to the extremely pronounced 1/f noise is the LF path of the split path input buffer.

The HF path is not bad at all and its noise can be well below 3 nV/sqrt(Hz) at e.g. 10 MHz.

The LF path on the other hand has a noise figure of up to 20 dB by design - in addition to the 1/f noise of the FET amplifier utilized here. Bipolar wouldn't be an option since high input impedance is required, and it is almost irrelevant because of the high NF by design mentioned before.

The crossover frequency between LF and HF path is usually only a few kHz, yet the LF path affects the noise of the HF path up to at least 100 kHz.

In practice we get the expected 1/f noise characteristics down to about 100 kHz and  see an excessive increase in noise below 10 kHz. The reason why even lower bandwidth limits than 100 Hz would not make much of a difference anymore might be the small portion of the total bandwidth.

[core]

I have started to "play through" the Batronix demo board again with the rigol.
I will probably upload the results to googledrive.
I already have a few pictures in the box today.
Note to self: In the pass/fail menu do not activate the screenshot at fail, but save it yourself - otherwise the save message will be included in the picture.
DS17: Demo Acquisitionmode Normal/Average (16x)
DS18/19: Demo Acquisitionmode Normal/Peak
DS20: Demo cursors
DS21-DS24: Demo measurements, several modes
DS25: Demo persistence 1sec
DS26/DS27: Demo Pass/Fail

Please go through all the tests on this demo board, as much as you have time to do so. Very interesting.

[Martin72]

Promised.

I still have some time, all scopes except the 1104X-E are back to work.

[Randy222]

On the DHO804, why does the scope lose signal sync if the vertical scale is increased?
Example;
Ch1 on the internal 1kHz 3v peak signal, set trigger to 1v, but then dial up the scale to 10v/div. The trigger does not change, so why the scope lose sync on the square wave (dancing signal)?

Have a look at the data sheet under "Trigger Sensivity".
It says 0.5div, at 10V/div and a 3V signal it is no wonder why it no longer triggers.

Trigger is based on the display? If the signal is 3v and the trigger is 1v, why should it matter what the V/div is set to?

[ebastler]

Trigger is based on the display? If the signal is 3v and the trigger is 1v, why should it matter what the V/div is set to?

Trigger hysteresis is based on the vertical sensitivity, it seems. You set the trigger threshold to 1V, and that's where it will remain. But there is no independent control over what is considered a significant signal excursion across the threshold (which leads to a trigger event), rather than noise. That is derived from your vertical sensitivity setting (V/div).

Not sure whether that's how other scopes commonly do it? Are there scopes which give you a separate hysteresis control?

[Martin72]

I have just tried this on the 1104X-E and it is exactly the same.
But it makes no sense, you can hardly see anything of the 3V signal at 10V/div.

[2N3055]

On the DHO804, why does the scope lose signal sync if the vertical scale is increased?
Example;
Ch1 on the internal 1kHz 3v peak signal, set trigger to 1v, but then dial up the scale to 10v/div. The trigger does not change, so why the scope lose sync on the square wave (dancing signal)?

Have a look at the data sheet under "Trigger Sensivity".
It says 0.5div, at 10V/div and a 3V signal it is no wonder why it no longer triggers.

Trigger is based on the display? If the signal is 3v and the trigger is 1v, why should it matter what the V/div is set to?

Trigger is not based on display.

Even on analog scope minimum signal amplitude for reliable triggering will be based on certain percentage of dynamic range, i.e. input sensitivity.

Ths scope has digital trigger. It works on signal after ADC has sampled it... So if you have signal that is less than step of ADC you cannot trigger on it, right?  And of course you don't need only few steps of ADC, because you need hysteresis, and enough sample points to interpolate curve through trigger point with enough accuracy...

[Randy222]

I have just tried this on the 1104X-E and it is exactly the same.
But it makes no sense, you can hardly see anything of the 3V signal at 10V/div.

Try 5V/div 1V trigger, which fails to sync on the Rigol.
The signal is the signal in the front-end, trigger setting is relative to signal, so why can't it display it, even if squashed in the display, a 10V/div makes a 3v signal look almost flat on a small 7" screen. What if the actual screen size was 14" with same number of div boxes making a div box physically bigger on the screen, that squashed line on 7" screen suddenly is not so squashed on a 14" screen.

Is it just that it did actually trigger but the device can't paint it to the screen?

[ebastler]

The signal is the signal in the front-end, trigger setting is relative to signal, so why can't it display it, even if squashed in the display, a 10V/div makes a 3v signal look almost flat on a small 7" screen. What if the actual screen size was 14" with same number of div boxes making a div box physically bigger on the screen, that squashed line on 7" screen suddenly is not so squashed on a 14" screen.

Is it just that it did actually trigger but the device can't paint it to the screen?

I'm afraid you are on the wrong track there. Yes, "the signal is in the front-end". But it is then amplified or attenuated according to your V/div setting, A/D-converted, and the trigger engine works based on those digital data.

If you change the V/div setting, your "trigger setting" does indeed not change: As mentioned before, the trigger threshold you have set will remain unchanged. (I.e. the scope will adjust its ADC count value for the trigger threshold, such that the trigger threshold's voltage remains unchanged.) But the trigger hysteresis, i.e, the signal swing required to detect a "significant" crossing of the trigger threshold, will change.

The hysteresis gets larger as you go to higher V/div settings. That makes sense from a technical perspective: The scope cannot determine an accurate trigger time from just a few ADC counts of signal swing.

[Martin72]

Finally...
After excactly 3 month of testing my conclusion, only from my point of view, of course.
Summarized in pros and cons:

+Very compact housing with external power supply unit.
+Sharp 7" display with high resolution
+Very low noise, native 12 bit vertical resolution
+Very clear, easy-to-use user interface.
+Unique window function in this price range and far beyond.
+Easy to hack
+Inexpensive

-Touch function in the screen size can no longer be used only with fingers
-No hires mode (14, 16 bit)
-Sample rate drops to 312.5 MSa/s from 3 active channels, must be kept in mind when displaying high-frequency signals
-No bodeplot
-Memory drops to a meager 1Mpt when all channels are active(original state)
-Few decoding functions
-No upgradeable options(original state)
-Relatively useless "ultra acquisition" mode
-Poorly equipped FFT function, also error-prone
-Various inconsistencies in the software in general
-Only one usb port
-Somewhat high-frequency sounding fan
-Limited window functions

As I said, just my impressions, I won't compare it with other scopes.
Some, if not many negative points could easily be corrected by rigol via the software, but unfortunately we all know that probably won't happen.
Instead, the users (Turbotom, FFT) have to fix their bugs - very embarrassing.
But that doesn't mean that you can't work with these scopes, of course you can.
After the first firmware update, I didn't experience any more crashes.
And if you don't need a bodeplot and only rarely use FFT, then it's a good scope with some innovative new features compared to older models in this price range.
But if rigol does nothing more to the scope and if the SDS800X HD becomes available on the world market, at a similar price, then it will be very, very difficult to find reasons to buy it, except perhaps that you wouldn't buy anything from the competition.
But it's not currently available, so the DHO800 is the choice if you don't want an older entry-level scope.

That´s all folks, thanks for reading.

Martin

[Randy222]

The FFT cal file is fixed using the info from the DHO1000 thread.