NEW Keysight HD3

[ballsystemlord]

@CRTbrain ,
The behavior you have described can be replicated on my Rigol MSO5000 scope. I don't know how you feel about that, but at least you now know that this is not a Keysight scope bug. It does occur across scope makes/models. I'm attaching a gif so you can see what I'm talking about.

Please pardon the glare, I'm uncertain how to eliminate it when filming.
PS: I did try adjusting the brightness and contrast of the display without effect.

[egonotto]

Hello,

in the online event from 18 September 2024 about the HD3, at 16:48 it says that the HD3 has a resolution of 2 uV at 2 mV/div. That corresponds to only 13 bits.

Best regards
egonotto

[ballsystemlord]

Hello,

in the online event from 18 September 2024 about the HD3, at 16:48 it says that the HD3 has a resolution of 2 uV at 2 mV/div. That corresponds to only 13 bits.

Best regards
egonotto

Hmm, maybe the ENOB and actual NOB are different?

[Wolfgang]

Hello,

in the online event from 18 September 2024 about the HD3, at 16:48 it says that the HD3 has a resolution of 2 uV at 2 mV/div. That corresponds to only 13 bits.

Best regards
egonotto

Hmm, maybe the ENOB and actual NOB are different?

... these marketing fuzzies dont get it. Only ENOB bits are useful, all beyond fall victim to noise and nonlinearities.
What I would like to see for the physics and precision measurement guys would be an UXH scope (100MHz, 13-14Bits ENOB) 

[CRTbrain]

I read the user manual for speed vs accuracy tradeoffs.  From the manual, a user will never get anywhere near 14-bit for measurements nor for math.  The manual state that the HD3 using a highly decimated set of data for both measurements and math.  This is truly disappointed.  Hewlett and Packard would be disappointed that marketing overtook engineering truth. 

The Analysis Record and Waveform Math
Keysight InfiniiVision oscilloscopes are designed to provide high waveform update
rates which give you an increased probability of capturing infrequent events. With
high waveform update rates, there is less "dead time" between acquisitions.
To provide high waveform update rates, there are some things the oscilloscope
does with memory:
• When running, acquisition memory is split into two memories so that the
processing of one acquisition can take place while another acquisition is being
captured. For single acquisitions, the full memory is used. Waveform data in
acquisition memory is called the raw acquisition record.
• There is a separate eavesdrop memory used for making measurements and for
calculating math functions. Data in this memory is a reduced-sample
decimation of the data in the raw acquisition record. Waveform data in
eavesdrop memory is called the analysis record.
The default length of the analysis record is 64K points. To improve the precision
of measurements and math functions (including FFT), at the expense of
waveform update rate, you can also select deeper analysis record lengths.
96 Keysight InfiniiVision HD3-Series Oscilloscopes User's Guide
5 Math Waveforms
(If you want to calculate math functions on a longer record, you can save the
raw acquisition record and use programs like Excel or MATLAB to perform the
analysis.)
Waveform data for the acquisition time (that is, the time per division setting
multiplied by 10 divisions across the display) is saved to the memories. When the
oscilloscope analog-to-digital converter's (ADC, or digitizer) sample period yields
more data points than can be stored in the memory, some samples are thrown
away (decimated), and the effective (or actual) sample rate is reduced. This is why,
at greater time/div settings, the displayed sample rate decreases.
For example, suppose an oscilloscope's digitizer has a sample period of 1 ns
(maximum sample rate of 1 GSa/s) and a 1 M memory depth. At that rate, memory
is filled in 1 ms. If the acquisition time is 100 ms (10 ms/div), only 1 of every 100
samples is needed to fill memory. The effective sample rate becomes 10 MSa/s.
The decimator is configured to provide a best-estimate of all the samples that
each point in the record represents. There is no filtering in the decimation.
FFT (Spectral)
Analysis of
Analysis Record
Data
When the FFT operator is turned on, the decimation from the raw acquisition
record to the analysis record works on integer rate down-sampling. For example, a
raw acquisition record of 20M points and an analysis record of 64K points are
already set up with an integer rate decimation of 320 (20M/320 = 64K).
For FFT analysis, the decimated record is then zero-padded to 2(X+1) where 2X >=
record length. For the above example, the power of 2 greater than the record
length is 65536, so the record is zero-padded to 131072 points.
You can look at the FFT math function results on the oscilloscope to work
backward and find the un-zero-padded decimated record length as well as the FFT
length used after the zero-padding:
• First, we know:
• The FFT algorithm translates a time record of N equally spaced samples into
N/2 equally spaced lines in the frequency domain.
• The width of the FFT on the display is the maximum frequency in the FFT.
The maximum FFT frequency is also fS/2, where fS is the effective sample
rate of the decimated data. (Use the center/span or start/stop frequency
controls to display the maximum FFT frequency.)
• The FFT Resolution is displayed in the FFT menu. FFT resolution is also
known as the FFT bin width or a line in the frequency domain.
• Therefore, the un-zero-padded decimated record length is: fS x the acquisition
time (or 2 x max FFT freq x acquisition time).
• Also, the FFT length used after the zero-padding is: 2 x (maximum FFT
frequency)/(FFT Resolution).
All this works the same with analysis records greater than the 64K-point default.

[CRTbrain]

Thanks.  I checked out some other scopes in lab to see their behavior.

It doesn't happen on LeCroy, Tek, and R&S MXO scopes.  Nice.  It does happen on InfiniiVision and R&S RTM3000 scopes.  I can't comment on Rigol or Siglent as we don't purchase these.  Users need to be very, very careful how scope settings are made, else they will get inaccurate results.   It's much nicer (and more accurate) to not have to worry about this, and have the scope make measurements on the full waveform value.   

[ballsystemlord]

Thanks.  I checked out some other scopes in lab to see their behavior.

It doesn't happen on LeCroy, Tek, and R&S MXO scopes.  Nice.  It does happen on InfiniiVision and R&S RTM3000 scopes.  I can't comment on Rigol or Siglent as we don't purchase these.  Users need to be very, very careful how scope settings are made, else they will get inaccurate results.   It's much nicer (and more accurate) to not have to worry about this, and have the scope make measurements on the full waveform value.

When I first started looking/asking about scopes I was told that, "Scopes aren't for making accurate measurements." Now you've got me curious as to whether that statement was correct or the gentlemen should have said that, "Most scopes can't make accurate measurements," instead.

[ballsystemlord]

Thanks.  I checked out some other scopes in lab to see their behavior.

It doesn't happen on LeCroy, Tek, and R&S MXO scopes.  Nice.  It does happen on InfiniiVision and R&S RTM3000 scopes.  I can't comment on Rigol or Siglent as we don't purchase these.  Users need to be very, very careful how scope settings are made, else they will get inaccurate results.   It's much nicer (and more accurate) to not have to worry about this, and have the scope make measurements on the full waveform value.

If you have the time, could you do you test against the R&S RTA series?

Thanks!

And if you would be so good, could you explain exactly what you're doing to acquire your data/perform your test? I'll replicate on my MSO5000 (if possible,) and then we can compare scopes head-to-head. In my above GIF, I just guessed what you were doing and tried to replicate. I showed that the maths are based on the displayed data instead of the in-memory data with a stopped scope. That is to say, it was not actively acquiring data, but instead was operating only on the data in-memory.

[electr_peter]

There is nothing new or alarming that Keysight Megazoom scopes make measurements from "eavesdrop memory" (subset of full memory). InfiniiVision scopes (X1/2/3/4/6 series) popularity proves it is not a limitation for informed users, it was discussed many times. Also, HD3 (like X6 series) allows using more memory for measurements if required.

Measurement, for example, of rise time at fast time base is more accurate, but what is arguably even more important, a scope gives visual indication of where and how it measures.

It boils down to knowing your tools before making measurements.

[tooki]

Please pardon the glare, I'm uncertain how to eliminate it when filming.
PS: I did try adjusting the brightness and contrast of the display without effect.

Insert a USB thumb drive into the scope and save a screenshot.

[tszaboo]

I just played around a little with an HD3 today. I did not make any test with deeper meaning, but the overall UX was rather disappointing. No snappy UI at all.

The nice Keysight guy also mentioned, that a separate SA was initially planned to go to the empty space at the front, but it was decided that there is no real market for it. According to the Keysight guy the FFT became so good, that a separate SA moudle did not make sense anymore or they would be cannibalising their real SA units.

So, don't keep your hopes up for an MDO version that fills that void at the front.

I personally think this is the biggest problem of some manufacturers..

Why would this be a problem? FFT goes to 1GHz, maybe 2 or 4 on the oscilloscopes. Practically DC in the microwave world. SA will do 20GHz+ for the same money.
You cannot set up RBW, dynamic range is much less, you cannot attenuate the signal, 50 Ohm input is so-so. BNC is an outdated connector with really bad connection compared to the N or 3.5mm and other connectors on SAs. You cannot do any of the useful measurement modes of an SA, like occupied bandwidth. You change the time scale knob by accident and all your FFT settings are all messed up.
It's OK for conducted emission testing and such, but it's not an RF instrument, it's not meant to be one.

[nctnico]

I just played around a little with an HD3 today. I did not make any test with deeper meaning, but the overall UX was rather disappointing. No snappy UI at all.

The nice Keysight guy also mentioned, that a separate SA was initially planned to go to the empty space at the front, but it was decided that there is no real market for it. According to the Keysight guy the FFT became so good, that a separate SA moudle did not make sense anymore or they would be cannibalising their real SA units.

So, don't keep your hopes up for an MDO version that fills that void at the front.

I personally think this is the biggest problem of some manufacturers..

Why would this be a problem? FFT goes to 1GHz, maybe 2 or 4 on the oscilloscopes. Practically DC in the microwave world. SA will do 20GHz+ for the same money.
You cannot set up RBW, dynamic range is much less, you cannot attenuate the signal, 50 Ohm input is so-so. BNC is an outdated connector with really bad connection compared to the N or 3.5mm and other connectors on SAs. You cannot do any of the useful measurement modes of an SA, like occupied bandwidth. You change the time scale knob by accident and all your FFT settings are all messed up.
It's OK for conducted emission testing and such, but it's not an RF instrument, it's not meant to be one.

Yes and no. I'm under the impression that EMC checks are a big reason for non-RF companies to buy a spectrum analyser and there are many of such companies. If an oscilloscope can serve that purpose as well (at least with decent spectrum analysis controls like you find on the R&S oscilloscopes for example), I can see having FFT eating into the market of spectrum analysers.

If you look at the current test equipment market, you'll see that low frequency network analysers (going from a few Hz to a couple of hundred MHz) have dissapeared almost completely despite these systems being very useful for (example) doing power distribution impedance measurements. However, with FFT and/or frequency analysis (bode plot) you can do the same measurements with an oscilloscope (and maybe an external signal generator).

[2N3055]

I just played around a little with an HD3 today. I did not make any test with deeper meaning, but the overall UX was rather disappointing. No snappy UI at all.

The nice Keysight guy also mentioned, that a separate SA was initially planned to go to the empty space at the front, but it was decided that there is no real market for it. According to the Keysight guy the FFT became so good, that a separate SA moudle did not make sense anymore or they would be cannibalising their real SA units.

So, don't keep your hopes up for an MDO version that fills that void at the front.

I personally think this is the biggest problem of some manufacturers..

Why would this be a problem? FFT goes to 1GHz, maybe 2 or 4 on the oscilloscopes. Practically DC in the microwave world. SA will do 20GHz+ for the same money.
You cannot set up RBW, dynamic range is much less, you cannot attenuate the signal, 50 Ohm input is so-so. BNC is an outdated connector with really bad connection compared to the N or 3.5mm and other connectors on SAs. You cannot do any of the useful measurement modes of an SA, like occupied bandwidth. You change the time scale knob by accident and all your FFT settings are all messed up.
It's OK for conducted emission testing and such, but it's not an RF instrument, it's not meant to be one.

I was referring to mentality of delibratelly imposing artificial limits on products so they not compete with their own products from other lines, while ignoring that competition has no such qualms..

[tszaboo]

Yes and no. I'm under the impression that EMC checks are a big reason for non-RF companies to buy a spectrum analyser and there are many of such companies. If an oscilloscope can serve that purpose as well (at least with decent spectrum analysis controls like you find on the R&S oscilloscopes for example), I can see having FFT eating into the market of spectrum analysers.

If you look at the current test equipment market, you'll see that low frequency network analysers (going from a few Hz to a couple of hundred MHz) have dissapeared almost completely despite these systems being very useful for (example) doing power distribution impedance measurements. However, with FFT and/or frequency analysis (bode plot) you can do the same measurements with an oscilloscope (and maybe an external signal generator).

I'm under the same impression, and shops where I was getting quote for an SA immediately wanted to sell me a EMC precompliance test bundle.
But I don't think that network analyzers would disappear. RS ZNLE3 only goes to 3GHz. Which is not a couple hundred MHz, but there is really not too much difference between making a VNA work at 500MHz or 3 GHz. Most of the cost is going to be the fact that you are buying a box from these guys.
Then there is the Bode 100 and 500.
Or probably an Analog discovery can be used for this.

I was referring to mentality of delibratelly imposing artificial limits on products so they not compete with their own products from other lines, while ignoring that competition has no such qualms..

I don't know how easy it would be to implement the analog blocks of a SA into the digital domain of a scope. I mean it can be done, SDRs are kind-of doing that, and we see a digitalization of these things for a while. The question is do they have the processing power dedicated for this?

[2N3055]

I don't know how easy it would be to implement the analog blocks of a SA into the digital domain of a scope. I mean it can be done, SDRs are kind-of doing that, and we see a digitalization of these things for a while. The question is do they have the processing power dedicated for this?

They do not as it is, but they could have been added.

I have realtime SA SignalHound, that uses PC for crunching. For price Keysight asks for this scope, they could have used full blown PC and install full SignalHound SA inside.

This scope can costs as much as a 4GHz full PC scope from Siglent, with giant screen.

While I think Keysight made good upgrade to old Megazooms, apart from deep memory (where they are literally a decade late to the game) and 14 bit /low noise there is nothing else that it brings compared to old 3000T. In fact at his moment there are dozens of things you can do with 3000T that you cannot do with HD3 yet..

FFT speed, screen size and some other parameters puts it into clean modern 3000 device class category.
That means sub 10000 USD for 1 GHz...

I think they are a bit overly optimistic...

[ballsystemlord]

FFT speed, screen size and some other parameters puts it into clean modern 3000 device class category.
That means sub 10000 USD for 1 GHz...

I think they are a bit overly optimistic...

About what exactly?
And which scope of KS's is sub 10000 USD for 1 GHz?

[2N3055]

FFT speed, screen size and some other parameters puts it into clean modern 3000 device class category.
That means sub 10000 USD for 1 GHz...

I think they are a bit overly optimistic...

About what exactly?
And which scope of KS's is sub 10000 USD for 1 GHz?

Competition...

[2N3055]

There is nothing new or alarming that Keysight Megazoom scopes make measurements from "eavesdrop memory" (subset of full memory). InfiniiVision scopes (X1/2/3/4/6 series) popularity proves it is not a limitation for informed users, it was discussed many times. Also, HD3 (like X6 series) allows using more memory for measurements if required.

Measurement, for example, of rise time at fast time base is more accurate, but what is arguably even more important, a scope gives visual indication of where and how it measures.

It boils down to knowing your tools before making measurements.

Fact that previous, more than decade old generation did the same is not really a good reason to keep doing something.
Other scopes also have measurement cursors that show where exactly measurement is taken from.
But also other scopes have deep memory measurements where it keeps accuracy at very deep memory.
And also takes multiple measurements at the same time, so you can get,say, 10000 measurements from single capture. And have full statistical analysis with histogram.

I have MSOX3104T for some years. I know how to "squeeze out" measurements I need from it.
Fact that I can get some good numbers out of it is more of "despite the limitations" thing than "it is perfect the way it is".

I kind of feel that it was lost opportunity to make things better if they only can do it in limited way as the old ones.
But nobody really demonstrated yet how it actually works. I hold hope that it is not exactly as it was.
I also hope I won't have to wait for Electronica to test it myself to find out. Maybe some of the people that have it for test can dig in deeper into what it really can do.


Attachments: Single shot at 1ms/div, and risetime measurement, then changed to 10ns/div on same buffer, and then back at 10ms/div with histogram and Cycle to Cycle Jitter.

[Someone]

From the manual, a user will never get anywhere near 14-bit for measurements nor for math.

Why do think the vertical resolution is reduced? Yes, the horizontal resolution is reduced by decimation, but generally the bit depth is maintained. Which would make your claim false.

[pdenisowski]

You change the time scale knob by accident and all your FFT settings are all messed up.

Unless you have a scope where the time domain and frequency domain settings are fully independent - like the R&S MXO series oscilloscopes

[David Hess]

When I first started looking/asking about scopes I was told that, "Scopes aren't for making accurate measurements." Now you've got me curious as to whether that statement was correct or the gentlemen should have said that, "Most scopes can't make accurate measurements," instead.

Most oscilloscopes are not for making accurate measurements likely includes every oscilloscope you are aware of, at least in the voltage domain.  With an ENOB in the 6 to 8 bit range, that is at about 2-1/2 digits at most, and at low frequencies, almost all oscilloscopes have a huge amount of flicker noise.

The exception includes oscilloscopes which can perform slideback measurements with a differential comparator input.  In this case, the oscilloscope is used to measure a null, so linearity and ENOB become irrelevant; all that matters is determining zero.  My ancient Tektronix 7000 series analog oscilloscopes can make accurate slideback voltage measurements to 1 part in 10,000.  If you want this capability, look for a DSO which has offset voltage adjustment as well as vertical position adjustment.  But if you need this capability, then there is likely a better way.

Time measurements are another matter and digital storage oscilloscopes can be very good here.

From the manual, a user will never get anywhere near 14-bit for measurements nor for math.

Why do think the vertical resolution is reduced? Yes, the horizontal resolution is reduced by decimation, but generally the bit depth is maintained. Which would make your claim false.

Linearity and other errors reduce the ENOB (effective number of bits) limiting accuracy.  At low frequencies where the ENOB is higher, low frequency noise in the input buffer dominates.  There are some other things going on as well like setting time; most instruments will never settle to their digitizer resolution, and high resolution instruments will not even come close.

[pdenisowski]

I'm under the impression that EMC checks are a big reason for non-RF companies to buy a spectrum analyser and there are many of such companies. If an oscilloscope can serve that purpose as well (at least with decent spectrum analysis controls like you find on the R&S oscilloscopes for example), I can see having FFT eating into the market of spectrum analysers.

Yes, if you're designing a product, you'll almost certainly need to pass some (maybe many) EMC compliance tests.  And unless you're both (a) really good and (b) really lucky, you're very likely to fail those compliance tests the first (few, dozen, etc.) times unless you do pre-compliance testing.  I did a webinar / whitepaper on this last year:

https://cdn.rohde-schwarz.com.cn/pws/dl_downloads/dl_common_library/dl_brochures_and_datasheets/pdf_1/Educational_Note_-_Understanding_EMI_Precompliance_Testing.pdf 

What often happens is that people use an oscilloscope with a (sufficiently good) FFT and a set of near field probe to do a lot of EMI debugging during the precompliance process.  Since someone developing an electrical / electronic product would ... I hope ... have a decent oscilloscope or two, it makes sense to use a scope for EMI debugging / "sniffing" with probes.  You can also easily correlate time and frequency domain with a scope, which is really helpful for debugging.

With regards to EMC, people do also use spectrum analyzer or receivers (similar, but different), often when their design is stable enough to attempt to pass conducted and/or radiated testing.  This is also more similar to full compliance testing, which is basically always done with a receiver these days.

I'm not sure that scopes with an FFT (even a really, really nice FFT, like on our MXO series scopes) are eating into our spec an sales.  I would even say there are more areas where scopes and VNAs overlap for things like signal integrity applications. 

In my personal opinion, most scope FFT implementations weren't that useful until about 10 years ago, and it's only in the last few years that scope FFTs have entered the "wow" region of spectrum analysis.  So things might change in the future, but for now I think the biggest change has been that a LOT more scope users are taking advantage of FFT functionality and that's good for everybody

[pdenisowski]

I'm under the same impression, and shops where I was getting quote for an SA immediately wanted to sell me a EMC precompliance test bundle.

Probably because it's often a really cost-effective combination   Seriously.  The cost of failing one full compliance test is usually significantly more than the cost of a spec an and a precompliance bundle.  We have lots and lots of very happy EMC precompliance bundle customers.

But I don't think that network analyzers would disappear.

I really, really hope not   I've probably invested more time learning about VNAs than any other T&M instrument category.

RS ZNLE3 only goes to 3GHz. Which is not a couple hundred MHz, but there is really not too much difference between making a VNA work at 500MHz or 3 GHz.

That's true.  But not all 1 GHz VNAs are the same   And there is a substantial difference between a 500 MHz and a 20 GHz VNA (which is also reflected in the price). 

[Someone]

From the manual, a user will never get anywhere near 14-bit for measurements nor for math.

Why do think the vertical resolution is reduced? Yes, the horizontal resolution is reduced by decimation, but generally the bit depth is maintained. Which would make your claim false.

Linearity and other errors reduce the ENOB (effective number of bits) limiting accuracy.  At low frequencies where the ENOB is higher, low frequency noise in the input buffer dominates.  There are some other things going on as well like setting time; most instruments will never settle to their digitizer resolution, and high resolution instruments will not even come close.

Yes and what's that got to do with the price of encoder knobs?

CRTbrain claims the horizontal decimation will affect vertical resolution, but provides no evidence that is so. Do you have some examples or evidence of horizontal decimation changing the vertical resolution or accuracy?

[ballsystemlord]

There is nothing new or alarming that Keysight Megazoom scopes make measurements from "eavesdrop memory" (subset of full memory). InfiniiVision scopes (X1/2/3/4/6 series) popularity proves it is not a limitation for informed users, it was discussed many times. Also, HD3 (like X6 series) allows using more memory for measurements if required.

Measurement, for example, of rise time at fast time base is more accurate, but what is arguably even more important, a scope gives visual indication of where and how it measures.

It boils down to knowing your tools before making measurements.

Just to be clear, I'm *not* saying that there's anything "new or alarming" about the new KS scope. Rather, I find it to be a curious detail in the sense that some scopes are limited by this and others are not. For example, Dave's review video showed how much slower the R&S MXO 4 scope was when doing measurements. I'm wondering if this is, in part, due to the need to perform said measurements on the whole of the capture vs. the KS doing it on just 64K.

Additionally, if CRTBrain's research is all correct (and I'm assuming it is), then potentially you could use one of the (higher bit) oscilloscopes that he's listed in place of an RF power meter, frequency counter, and probably a few more things. Because those scopes can do measurements over a sizable amount of time and with reasonable accuracy.
You see, up until now I've treated my (8-bit MSO5000) oscilloscope, at the advice of others, as a "look but don't measure" type of device. Being able to do more with a scope was something I didn't think was possible. They were just not designed for making accurate measurements, or so I was told.