How much noise floor and other things matter in oscilloscope usability

[Fungus]

No, that signal was created by combining two signals, either with digital combiner (Siglent AWGs have one) or by simple resistive combiner, that will serve as attenuator at the same time.

Oh, I checked the manual and the MSO5000 AWG can only do AM, FM and FSK. No "add" function. 

(sorry, "combine")

If Rigol releases MSO7000 MarkII with as low noise as Siglent, then I would be willing to reconsider.

I wonder why they don't, you'd think they'd have had time to tweak their ASIC by now. All their expensive devices are based on it so it must be costing them a lot of lost sales.

(or maybe not, Batronix' bestseller list puts the Rigol ahead in sales)

That amount of noise is difficult to defend.

[Fungus]

Anyway, maybe a "Special Informative Instrument/Equipment Section" could be created where only those with actual "hands on" experience could comment, this would be highly beneficial to those seeking information to help decide on a purchase.

Strongly disagree.

Who would be the arbiter of truth? You? Not everybody in the world does the same job as you or has the same needs or the same budget as you.

The Rigol could even be the better choice in many situations.

(I even suspect that hardly anybody in this thread actually sits all day looking at mV signals. All these Siglents probably spend far more time at 1V/div than at 1mV/div).

[mawyatt]

Anyway, maybe a "Special Informative Instrument/Equipment Section" could be created where only those with actual "hands on" experience could comment, this would be highly beneficial to those seeking information to help decide on a purchase.

Strongly disagree.

Who would be the arbiter of truth? You? Not everybody in the world does the same job as you or has the same needs or the same budget as you.

The Rigol could even be the better choice in many situations.

(I even suspect that hardly anybody in this thread actually sits all day looking at mV signals. All these Siglents probably spend far more time at 1V/div than at 1mV/div).

No arbitrator, just only those with actual "hands on" experience with the equipment/instrument under consideration should comment. This way one can get information related to actual experience with such equipment.

In the case of this thread I wouldn't comment because I've never had "hands on" with a Rigol.

Best,

[bdunham7]

(I even suspect that hardly anybody in this thread actually sits all day looking at mV signals. All these Siglents probably spend far more time at 1V/div than at 1mV/div).

Well I do, and I don't mean just responding to this thread.  I may not be looking at 1mV signals, but I often use 100X probes to minimize circuit loading.  So I'm right back in that 1-10mV/div area.  In fact, on occasion even the slightly higher noise of the Siglent is annoying. 

I got a box delivered yesterday and I find that the SDS2000X+ is only very slightly better--if at all--than the SDS1104X-E as far as noise and minimum trigger.  It isn't able to match the Tek 2221A, but it gets pretty close--200uV_rms vs 150uV_rms for a stable display.

[2N3055]

Anyway, maybe a "Special Informative Instrument/Equipment Section" could be created where only those with actual "hands on" experience could comment, this would be highly beneficial to those seeking information to help decide on a purchase.

Strongly disagree.

Who would be the arbiter of truth? You? Not everybody in the world does the same job as you or has the same needs or the same budget as you.

The Rigol could even be the better choice in many situations.

(I even suspect that hardly anybody in this thread actually sits all day looking at mV signals. All these Siglents probably spend far more time at 1V/div than at 1mV/div).

I would be very happy with Mike as arbiter of that.. He is supremely qualified to do so.. Unlike some others..
Not that I want to push that burden on him. Just saying. Not everybody's opinion has same weight, despite size of their egos...Some people simply know more..

In the end your "democratic choice" argument is useless and doesn't hold. Those without knowledge are not capable of making educated decisions, because of lack of knowledge. Nobody is contesting their freedom of choice, they simply don't know enough to make good decision. At one point you have to trust the doctor and not the fear, rumors and your preferences.
Giving patient raw CT, MRI and X-ray images without doctors diagnose for them is not useful. All they can do go to another doctor with them.

Rigol would be better choice in some situations, which I already pointed out before. SDS2000X+ is better rounded scope overall. Excellent for analog, competent for digital. Rigol OTOH is one show pony, it is quite good for decoding and that is it. You can do analog stuff with it but not well.

MSO8000 OTOH is quite good scope for the price, because it has some more options, and 2 GHz bandwidth. It's noise is good FOR the bandwidth. MSO5000 is just weird. It could have been so much better.

Again, like Nico and others told you, it is not all about uVolts. Rigol shows finger thick trace all the time. It is only exacerbated even more when using 10x probes, where you are really at 100mV div (noise wise) doing digital stuff at 1v/div. It is not about microvolts, it is about details in the picture. Or shall I say the lack of them...

[nctnico]

No arbitrator, just only those with actual "hands on" experience with the equipment/instrument under consideration should comment. This way one can get information related to actual experience with such equipment.

That way you only get mostly information from people who are still on their -what I call- buyer's 'high'. On top of that, not everyone's needs are the same. What is a great tool for one, totally sucks for someone else. It would be a lot more useful to asses equipment using standarised testing but that takes a lot of time. All in all having a mix of owners and people who (based on experience with a wide range of equipment) know what works for a certain use case and what doesn't is a good compromise.

[Fungus]

I would be very happy with Mike as arbiter of that.. He is supremely qualified to do so.. Unlike some others..

Insults are still coming thick and fast I see.

I seem to recall telling Fiorenzo not to use 1x probe without 20Mhz limiter. I even posted a link to Dave's video on that.

I also suggested trying color gradient mode and averaging mode to see what happens.

I even suggested the SDS1104X-E as a way of saving money, because I know the noise is pretty much the same as the 2000 series and it seemed like it would cover his stated needs perfectly well.

At no point was he in any danger of being tricked into buying a Rigol because of fanboyism (imaginary).

Seems like the only people here being "hurt" here are the curmudgeons. 

(...and it's only been one thread, which nobody was forcing you to read anyway. Was the sky really falling?)

[Fungus]

On top of that, not everyone's needs are the same.

Yep. The Rigol MSO5000 is cheaper and it's a better choice for some percentage of users.

[David Hess]

I wonder why they don't, you'd think they'd have had time to tweak their ASIC by now. All their expensive devices are based on it so it must be costing them a lot of lost sales.

The noise is likely a characteristic of the process used to make the ASIC.  It is unlikely any available circuit change would fix it.  Many digital CMOS processes, which can still be used to make fast ADCs, are incredibly noisy, like 100s of nV/SqrtHz and MHz noise corners.

As you pointed out, most applications do not involve inspecting millivolt level signals so the Rigol's high level of noise is irrelevant to most users.

Well I do, and I don't mean just responding to this thread.  I may not be looking at 1mV signals, but I often use 100X probes to minimize circuit loading.  So I'm right back in that 1-10mV/div area.  In fact, on occasion even the slightly higher noise of the Siglent is annoying.

I suspect higher noise is a major reason x100 probes are not very popular despite having better high frequency performance than x10 probes.  My own experience is that the difference in noise is very noticeable.

[nctnico]

I wonder why they don't, you'd think they'd have had time to tweak their ASIC by now. All their expensive devices are based on it so it must be costing them a lot of lost sales.

The noise is likely a characteristic of the process used to make the ASIC.  It is unlikely any available circuit change would fix it.  Many digital CMOS processes, which can still be used to make fast ADCs, are incredibly noisy, like 100s of nV/SqrtHz and MHz noise corners.

As you pointed out, most applications do not involve inspecting millivolt level signals so the Rigol's high level of noise is irrelevant to most users.

No, the problem is not in the milli-volt level signals but the overall high noise which obscures details in the signal and makes the traces much thicker. I have owned a very noisy (high end) DSO myself (from a different brand though) and it was very cumbersome to work with due to the high noise level. Hi-res and averaging can clean up a signal to some extend but you'll also lose the details. And I wasn't even using it to look at particulary special signals; just getting measurements on analog signals in the range of a few Volts already proved difficult.

[Fungus]

I wonder why they don't, you'd think they'd have had time to tweak their ASIC by now. All their expensive devices are based on it so it must be costing them a lot of lost sales.

The noise is likely a characteristic of the process used to make the ASIC.  It is unlikely any available circuit change would fix it.  Many digital CMOS processes, which can still be used to make fast ADCs, are incredibly noisy, like 100s of nV/SqrtHz and MHz noise corners.

Keysight's ASIC isn't as noisy and that's quite old now. Is that a different process?

Even if the ASIC can't easily be fixed, you'd think there would be more ways to improve the signal in software.

eg. Their HiRes mode seems to be fixed at 32x which is a bit heavy-handed. I'd have preferred a user-selectable setting, eg. 4x, 8x, 16x, 32x... to preserve some detail.

Or even implement a programmable FIR filter on the 8Ghz incoming data.

[bdunham7]

I suspect higher noise is a major reason x100 probes are not very popular despite having better high frequency performance than x10 probes.  My own experience is that the difference in noise is very noticeable.

The main reasons I'll use a 100X probe are HV (not an issue here) and low circuit loading.  For the latter, obviously there are issues with low signal levels and higher frequencies where the capacitive loading takes over.  So usually I'll be looking at some rather mundane medium-level signal, like a crystal oscillator or CMOS logic circuit, and I just want a halfway decent reading with minimal loading.  It never looks really great, but a 100X probe often gets the job done.  More noise?  I'm sure there is, but I don't think it overwhelms scope input noise.

Here are three examples of 100mV/1MHz, 100mV/100Hz and 20mV/100Hz.  This results in a 1mV_rms or 200uV_rms input to the scope.  I find this sort of setup to be very useful in troubleshooting random stuff.

[2N3055]

I wonder why they don't, you'd think they'd have had time to tweak their ASIC by now. All their expensive devices are based on it so it must be costing them a lot of lost sales.

The noise is likely a characteristic of the process used to make the ASIC.  It is unlikely any available circuit change would fix it.  Many digital CMOS processes, which can still be used to make fast ADCs, are incredibly noisy, like 100s of nV/SqrtHz and MHz noise corners.

Keysight's ASIC isn't as noisy and that's quite old now. Is that a different process?

Even if the ASIC can't easily be fixed, you'd think there would be more ways to improve the signal in software.

eg. Their HiRes mode seems to be fixed at 32x which is a bit heavy-handed. I'd have preferred a user-selectable setting, eg. 4x, 8x, 16x, 32x... to preserve some detail.

Or even implement a programmable FIR filter on the 8Ghz incoming data.

Keysight has access to state of the art processes.
Also Keysight has years of experience of making massively parallel ADC.

Your comments about Hi Res are valid. Except, Rigol initially didn't filter so heavily, and problem was that people were saying HiRes is not working. They had to go all the way up to 32x to make it show..

Rigols chipset consists of two components AFE ASIC (Beta Phoenicis) and Signal processing (Ankaa) that contains ADC block. Problems are that both chips were designed for MSO8000 and have up to 4 GHz bandwidth.
AFE ASIC is fully contained, including amplifiers and solid state switches for attenuators. I don't know exact internal architecture. Also SP chip with ADC also has full bandwidth (4GHz) driver buffers to drive ADC cluster. That can be source of noise too. And also noise might be from ADC, quantization noise. If ADC is massively parallel type, there are all kinds of things that can go wrong.. intercalibration, clock distribution, crosstalk etc etc..

One interesting thing would be to disconnect/short input directly into Rigol ADC chip. That would give a better clue where the noise comes from.

[Performa01]

Here's a demonstration, what an SDS2000X Plus can show with an emulated ripple with asynchronous spikes riding on it if optimal probing (without additional noise pickup) is applied. 2.5 mVpp 1 MHz ramp with 300 µV 6.000001 MHz 10 ns wide spikes riding on it.

Nice example (pic worth a thousand words). How did you generate such signal?

It is a two channel AWG (SDG6052X in this particular case), both channels combined by means of a resistive wideband (DC-12.4 GHz) power combiner/splitter. This will provide 6 dB attenuation for each channel.
Channel 1 generates a 5 mVpp ramp with 5% symmetry at 1 MHz.
Channel 2 generates a 10 ns wide pulse with 3 ns rise/fall times at 6.000001 MHz with 60 mVpp amplitude and this signal is fed through a 1 GHz precision step attenuator, set at 40 dB.
The output of the power combiner is what you see.

First thing, HiRes on MSO5000 has been around for some time now, and it's not big help...
They had to downsample 32 times to get something and then not much.

The manual says that the bandwidth cannot exceed 1/32 the samplerate, so this sounds like 16 times oversampling to me.
This would be roughly equivalent to ERES 2.0 bits on a Siglent, where the resolution enhancement in ENOB can be set from 0.5 to 3.0.

In 10 bit mode together with ERES 2.0, we get a total resolution enhancement to 14 bits (and a theoretical ENOB of 11 bits). The trace gets very thin with this, see attached screenshot.

SDS2354X Plus_Ramp_2.5mV_1M_Pulse_300uV_6000001Hz_ERES2.0

I use two two pass trough terminators  to 1k resistors from siggen, other side connected together, and grounded with 100 Ohm resistor. I have that one because we had some previous discussion about two tone testing, and someone did it with that one so I made it exactly like that so we can compare results...

That might have been me since I was interested in the scopes DR performance. The 1K series R is to isolate the two AWG outputs from each other so they don't "see" the other signal as much with the ~26dB reverse isolation looking back from the shunt 100 ohm resistor. Was concerned about how the AWG output behaves in the presence of another signal and how this might affect the AWG output amp linearity. With this high an isolation the AWG output amplifier effects should be minimal and the resultant two tone IDM representing the scopes performance. BTW using the digital combining isn't as good, since the signal is created and then passed thru the AWG amplifier chain to the output, thus exposing the signal to the chains linearity effects. However may be OK since we are only looking at 65~75dB IMD with these scopes, but with Performa01 Picoscope results of ~100dB some of the AWG amplifier chain effects may contribute if using the AWG digital combining for the two tones.

For the high 3rd order dynamic range on the Picoscope 4262 I did not need to use any additional attenuators apart from the power combiner itself, which only provides 6 dB isolation between its ports. At frequencies as low as 1 MHz, an SDG6000 amplifier output quite obviously cannot be intermodulated that easily. But maybe I should try one more time with additional attenuation – who knows, maybe the Picoscope can do even better…

EDIT: Done - and found no difference. The SDG6052X performs well enough for this test at 1 MHz, even without additional isolation between the outputs.

The integrated digital combiner does indeed generate some IMD products on its own, so it is not suitable for reliable 3rd order dynamic tests. IMD can be as bad as 45 dBc at high output levels.

EDIT: Using the internal digital combiner, the third order intermodulation intercept point is about +21 dBm at an output level of +9 dBm and a frequency of ~50 MHz.
EDIT2: Unsurprisingly (since it's done in the digital domain), the third order intermodulation intercept point is nearly the same (+21.5 dBm) for a -18 dBm output level at 1 MHz.

I got a box delivered yesterday and I find that the SDS2000X+ is only very slightly better--if at all--than the SDS1104X-E as far as noise and minimum trigger.  It isn't able to match the Tek 2221A, but it gets pretty close--200uV_rms vs 150uV_rms for a stable display.

There is almost no difference indeed. If anything, the SDS2000X Plus might show less or lower spurious signals, but the noise level should be about the same. In fact, the little SDS1000X-E even has a slight edge here, because it provides a true full resolution 500 µV/div sensitivity, whereas the SDS2000X Plus is limited to 1 mV/div. Yet the 500 µV/div gain setting is not useless there either, because of the 16 bit display interface. This means you can make use of it in 10 bit mode (which becomes 9 bit at 500 µV/div) and also the results of math functions, especially ERES of course.
 

[2N3055]

The manual says that the bandwidth cannot exceed 1/32 the samplerate, so this sounds like 16 times oversampling to me.

Yes, that makes sense.

For the high 3rd order dynamic range on the Picoscope 4262 I did not need to use any additional attenuators apart from the power combiner itself, which only provides 6 dB isolation between its ports. At frequencies as low as 1 MHz, an SDG6000 amplifier output quite obviously cannot be intermodulated that easily. But maybe I should try one more time with additional attenuation – who knows, maybe the Picoscope can do even better…

I tried with 1k/100Ohm combiner and got roughly the same numbers as you for two tone test...
And both are better than spec.. Love that little guy.
And SDG6000 too, it is much better than spec for signal distortion, and digital combiner keeps surprising me...

[normi]

Noise floor, small voltage scales:
 It is the vertical resolution (volts). It is very visible, which is why you came to this forum, so unlike sample rate you will be fully aware of what you can't see.

When most needed:
Looking at very low voltage signals.

Buuzzzzz wrong! As I wrote before: noise floor simply scales along with the V/div settings. In that perspective using low level signals is not representative for regular scope usage. The Rigol MSO5000 also sucks for higher level signals because the noise will still drown details of the signal. Averaging won't help because that also obscures the details that you want to catch.

And having a higher samplerate gives you nothing if it is far beyond the bandwidth of your scope; it only wastes valuable memory. It just becomes a ridiculous number like having an 8000kW engine in a go-kart.

I disagree, the noise on the MSO5000 is about 2mv at the lowest 4mv level, so if that scaled as in your example it would mean that at 10V/div the noise would be 5v. What appears to be the issue is this Phoenix chipset which no one has information on how it works has a noise component which is added in but it does not scale proportionally with V/div, the scope would also have the regular scope noise which scales with the V/div.
It is arguable whether the MSO5000 has more sample rate than it needs but the fact is that it is better than having barely enough. With all channels on it has 2G/S per channel.
 

[mawyatt]

I wonder why they don't, you'd think they'd have had time to tweak their ASIC by now. All their expensive devices are based on it so it must be costing them a lot of lost sales.

The noise is likely a characteristic of the process used to make the ASIC.  It is unlikely any available circuit change would fix it.  Many digital CMOS processes, which can still be used to make fast ADCs, are incredibly noisy, like 100s of nV/SqrtHz and MHz noise corners.

Keysight's ASIC isn't as noisy and that's quite old now. Is that a different process?

Even if the ASIC can't easily be fixed, you'd think there would be more ways to improve the signal in software.

eg. Their HiRes mode seems to be fixed at 32x which is a bit heavy-handed. I'd have preferred a user-selectable setting, eg. 4x, 8x, 16x, 32x... to preserve some detail.

Or even implement a programmable FIR filter on the 8Ghz incoming data.

CMOS comes in many flavors, the modern versions feature very fast and small featured devices. Because MOS devices exhibit more noise than bipolar, we often oped for a SiGe BiCMOS process where we had access to superb bipolars and good CMOS. However as time marched on the BiCMOS couldn't keep up with the SOTA mainly because there's not enough market to justify the enormous foundry costs, so BiCMOS is becoming an orphan. Wise folks realized this long ago and began to move designs to pure CMOS and leverage off the almost unlimited finances of CMOS driven by Smart Phone and Laptop/iPads.

I suspect Rigol's attempt at a fully integrated front end didn't turn out as well as expected, especially noise-wise, and maybe done in pure CMOS. The cost of improving this is likely moving to another CMOS process, and the expense of a complete redesign in a newer CMOS process is very high....maybe too high for Rigol to invest at this time. Hopefully they will produce at better performing ASIC, and I admire them for taking on the ASIC design challenge in the first place.

Keysight has much deeper pockets and experience than just about anyone in the instrument field, and have been developing full custom ASICs for AWGs and DSO for some time. The Griffin AWG ASIC was an absolute masterpiece of engineering (see image), done in a SiGe BiCMOS process back in ~2009, however the next generation is all CMOS. The Stingray ASIC from ~2011 is an ADC and part of the front end of a developmental high performance DSO (see images) and all CMOS. So Keysight has been moving towards CMOS for custom ASICs for a decade now, and likely others too.

Custom ASIC designs in any process are expensive, and may not fit the financial model for all companies. So, at the moment seem relegated to a few A player levels in instrumentation field, except for Rigol.

Best,

[David Hess]

Keysight's ASIC isn't as noisy and that's quite old now. Is that a different process?

Given the age difference, it almost has to be a different process.

Even if the ASIC can't easily be fixed, you'd think there would be more ways to improve the signal in software.

The only way to remove noise is to limit the bandwidth or synchronously demodulate the signal with averaging over multiple triggers.  And limiting the bandwidth does not reduce the noise density.

Their HiRes mode seems to be fixed at 32x which is a bit heavy-handed. I'd have preferred a user-selectable setting, eg. 4x, 8x, 16x, 32x... to preserve some detail.

That is how Tektronix used to do it.  Bandwidth was variable since the adjustment was tied to sweep speed with a fixed maximum sample rate.  It was certainly effective.  I do not know what they do now.

Or even implement a programmable FIR filter on the 8Ghz incoming data.

That is not impossible but it has high cost.  Some high end DSOs manage it if you want to buy a DSO instead of an expensive new car.

Rigols chipset consists of two components AFE ASIC (Beta Phoenicis) and Signal processing (Ankaa) that contains ADC block. Problems are that both chips were designed for MSO8000 and have up to 4 GHz bandwidth.

That explains at least part of the problem.  A signal chain with 4 GHz performance will always have higher noise density than one designed for a much lower frequency.  I gave an example of this earlier.

AFE ASIC is fully contained, including amplifiers and solid state switches for attenuators. I don't know exact internal architecture. Also SP chip with ADC also has full bandwidth (4GHz) driver buffers to drive ADC cluster. That can be source of noise too. And also noise might be from ADC, quantization noise. If ADC is massively parallel type, there are all kinds of things that can go wrong.. intercalibration, clock distribution, crosstalk etc etc..

The attachment I included earlier about signal conditioning in oscilloscopes written by Steve Roach gives some idea of how such a design is possible, although he thought an alternative using MEMS switches was more likely.

One interesting thing would be to disconnect/short input directly into Rigol ADC chip. That would give a better clue where the noise comes from.

Measuring the noise at every volts per division setting would reveal some of what is going on.

[Fungus]

Or even implement a programmable FIR filter on the 8Ghz incoming data.

That is not impossible but it has high cost.  Some high end DSOs manage it if you want to buy a DSO instead of an expensive new car.

It's only 8 bit numbers and trivially parallelizable.

If the FPGA is capable of triggering on incoming serial data then it might be able to do it.

[Fiorenzo]

@Fiorenzo
You face the common issue most new scope buyers face; you want a scope that will do as much as possible for the amount of money you spend, unfortunately that is not the case. The MSO5000 will do things the Siglent can't and visa versa, although the Siglent is 40-50% more expensive. So lets look at the difference, noise floor Vs Sample rate.

Sample Rate:
It is horizontal resolution, it allows you to see finer details horizontally (time). It is an invisible feature, meaning that if your sample rate is low and causing an issue most cases you won't know. Aliasing is a visible issue if you are experienced enough to see it, but other critical flaws will be hidden. (You won't know that you don't know)

When most needed:
If there is a non repetitive and infrequent glitch in a signal that would not be sampled by the scope because the spacing between the samples is too wide (AKA low sample rate).

Impact:
You could go weeks trying to figure out a problem and will not realize that your circuit's failure is caused by the glitch that your slow scope is unable to find, this can be very costly in time.

Work around:
1. Obtain a high sample rate scope, all are very expensive (except the MSO5000).

Noise floor, small voltage scales:
 It is the vertical resolution (volts). It is very visible, which is why you came to this forum, so unlike sample rate you will be fully aware of what you can't see.

When most needed:
Looking at very low voltage signals.

Impact:
If your system is susceptible to very small noise signals or you work with tiny signals then you want as low noise as possible. Since you can clearly observe your scopes noise level you won't spend as much time searching as you would be aware that you can't see signals cleanly below a certain level. So not likely to have a dramatic impact as the unseen glitch, because you can quickly proceed to a work around if you want to investigate low noise.

Workaround:
1. Use averaging to clean up noisy signal
2. use an amplifier - some a very cheap and can be modified for Oscilloscopes or Spectrum analyzers, or can be built. Professional ones are very expensive but are usually used for differential measurements and are needed by even the scopes with low noise floor to see much smaller signals.
3. Obtain a scope with lower noise, in some cases these lower noise scope can be cheaper than the MSO5000

Outside of the fact that the RIGOL MSO5000 has a much lower price than the Siglent SDS - Plus, you could still buy a cheaper Siglent scope which would have the same noise performance as the SDS -Plus, however it is impossible to find another non Rigol 8G/S scope for under a $1000, or under $2000. This is the reason why the scope is attractive to many buyers.

Thank you normi, very interesting and usefull explanation.

And also thank you again to everybody, you are super kind to support this thread

[Fiorenzo]

The Rigol doesn't do HiRes mode though.

I don't remember that I had seen a HiRes screenshot from the MSO5000 in this thread.

But the MSO5000 User Guide claims that it does:

High Resolution

This mode uses an over-sample technique to average the neighboring points of the sample waveform. This reduces the random noise on the input signal, generates a much smoother waveform on the screen and improves the vertical resolution. This is generally used when the sample rate of the digitalconverter is greater than the storage rate of the acquisition memory.

Note:
* The"Average"and "High Res"modes use different averaging methods. The former uses "Multi-sampleAverage"and the latter uses "Single-sampleAverage".
* In "High Res"mode,the signal bandwidth does not exceed 1/32 of the sampling rate.
* In "High Res"mode,the highest waveform refresh rate mode is not supported.

Edit: It does not tell the actual decimation factor or the number of averaged neighbor samples, though. A boxcar filter with 16 taps had a -3dB cut-off of ~fs/35, with 8 taps it were ~fs/17, and in order to get the documented fs/32, the closest number of required taps were 15. But this is pure speculation now and I think the "truth of the actual implementation" can only be determined experimentally by an owner.

Btw: One disadvantage of a boxcar averaging filter (and thus disadvantage of HiRes, if based on boxcar averaging) is that its sinc frequency response starts rolling off already beyond DC, i.e. the passband has no pronounced "flat top". If this matters for a particular use case, the filter cut-off should be rather chosen several times higher than the highest frequency of interest.

The Rigol has hi-res mode but many times It was ineffective.... I don't know why

I have some photos but i should search for those.

[Fiorenzo]

The Rigol doesn't do HiRes mode though.

I don't remember that I had seen a HiRes screenshot from the MSO5000 in this thread.

But the MSO5000 User Guide claims that it does:

High ResolutionThis mode uses an over-sample technique to average the neighboring points of the sample waveform. This reduces the random noise on the input signal, generates a much smoother waveform on the screen and improves the vertical resolution. This is generally used when the sample rate of the digitalconverter is greater than the storage rate of the acquisition memory.

Note:
* The"Average"and "High Res"modes use different averaging methods. The former uses "Multi-sampleAverage"and the latter uses "Single-sampleAverage".
* In "High Res"mode,the signal bandwidth does not exceed 1/32 of the sampling rate.
* In "High Res"mode,the highest waveform refresh rate mode is not supported.

STOP THE THREAD!

I just downloaded the latest manual from Rigol and it says they've now added "High Res" mode.

HiRes will make a huge difference to the noise level by leveraging that massive 8Ghz sample rate.



Fiorino, we need another test with "HiRes" mode enabled.

My name Is Fiorenzo ahahah....

In regard of the photos I have some, give me some time and i will post them.
Also, I cannot do more of them because i sent back the Rigol.

I experimented a week with the hi-res mode.
My thought was: well if It has 8GSa/s It should give good results with the hi-res mode but It was not many times.... unfortunately.... Most times It gave only a reduction in the visibile noise of about 10-20%.... some time nothing, and free times not more than 50%....
So I came to the conclusion that hi-res was not the solution for the noise problem...

[Fiorenzo]

Hi-Res mode was implemented by the second firmware update, AFAIK.

Yep, you're right.

So...

And, yes, Fiorenzo did post a screenshot of his test signal in Hires mode, my bad:

The manual says "In "High Res" mode, the signal bandwidth does not exceed 1/32 of the sampling rate" so it sounds like they do 32x oversampling with no user control. Not ideal.

You'd probably see the spikes on that signal if you zoom in a bit. They'd be much more visible in "peak" mode, too. I wish I had one here to fiddle with.

No you wouldn't see anything because it was filtered out.
And in Peak Detect mode you would see 10 mm thick solid wall of noise.. You don't seem to understand exactly how these acquisition modes work.

Yes that Is right. I have seen that many times on the Rigol. Also hi-res mode seemed to me more of an avarage than a bit resolution improve...

[Fungus]

I experimented a week with the hi-res mode.
My thought was: well if It has 8GSa/s It should give good results with the hi-res mode but It was not many times.... unfortunately.... Most times It gave only a reduction in the visibile noise of about 10-20%.... some time nothing, and free times not more than 50%....
So I came to the conclusion that hi-res was not the solution for the noise problem...

My impression from the manual is that it would depend a lot on the horizontal zoom.

It's not a "solution for the noise" but it might make the spikes more visible when you zoom in a bit.

[gf]

Yes that Is right. I have seen that many times on the Rigol. Also hi-res mode seemed to me more of an avarage than a bit resolution improve...

When I look at your 20mV/div screenshot showing 815µV RMS of noise (w/o HiRes and w/o 20MHz BW limit), this is a SNR of only ~36.8dB_FS, or ~5.8 ENOB.
Assuming white noise, the theoretical noise voltage reduction with 16x HiRes is factor 4 = sqrt(16), or ~12dB, leading to a SNR of ~48.8dB_FS then, or ~7.8 ENOB.
So the result is not at all an enhancement beyond 8 ENOB, but HiRes just helps to bring 8 ENOB almost back, at the cost of limiting -3dB bandwidth to ~230MHz, and a sampling rate reduction to 500MSa/s. That's only 1/2 the sampling rate of a 1GSa/s scope then.

[ I don't know whether the same applies to other V/div settings, too. To get evidence, each one would need to be evaluated indiviually. ]

originating
My thought was: well if It has 8GSa/s It should give good results with the hi-res mode but It was not many times.... unfortunately.... Most times It gave only a reduction in the visibile noise of about 10-20%.... some time nothing, and free times not more than 50%....

Two points that come into my mind:

1) If the initial noise (w/o HiRes) happens to be low, then the iprovement can possibly only be seen on the screen when zooming in vertically.

2) The amount of noise reduction achievable by HiRes depends on the spectral power distribution of the noise. The theoretical factor of sqrt(N), where N is the number of adjacent samples being averaged, only applies to white noise. If the noise power is already concentrated at low frequencies, then the HiRes low-pass filtering can't help much. To assess this, it were necessary to do a FFT of the captured signal (w/o HiRes processing applied). E.g. if you already apply a 20MHz BW limit in the AFE, then an additional low-pass with a cut-off at ~230MHz won't make too much difference, but it can only reduce any additional noise originating after the AFE's 20MHz filter (e.g. noise from the ADC).

Edit: Changed some phrasing in order to be hopefully more clear.

Edit: Attached the frequency response of a 16-tap boxcar moving avarage lowpass filter, when running at a sample rate of 8GSa/s. Maybe it helps to understand what kind of filtering is happening when HiRes is enabled.