NEW Keysight HD3

[tggzzz]

LVC would be my first choice for the fastest commonly available logic since it gets a speed boost from operating at 5 volts, but I wonder if it would be even faster driving the emitter of an NPN cascode RF transistor to produce a clean output.  I mean to try this some day with a microwave layout and an 74LVC125.

At least the 74LVC125 is only 4 channel, compared with the 74LS125's irritating 6 channels! But D/DB/NS/PW packages still having power connections on opposite corners?!

I like the 74LVC1G devices because of the low pin inductance which isn't shared between multiple outputs. Would your circuit need 4 drivers?

[HighVoltage]

It certainly does not seem there is much "special ASIC acceleration" for FFT in HD3. Unlike R&S that really looks like realtime SA at settings shown.

The older Agilent 7000 series scopes have a much faster FFT than newer models.
For many of my FFT needs, I have used a DSO7104A with delight over the years.
The 3000x, 4000x or 6000x Keysight scopes just suck at FFT update rates.

I was hoping that the HD3 would have a similar fast FFT again with the new ASIC.

[David Hess]

... exactly. The noise is why their ENOB values are not really spectacular, as might be hoped from th 14bits ADC resolution.

The noise is one of the reasons I question using a 14-bit ADC, but higher resolution does have advantages for frequency domain measurements.

What I would like to see is a test of settling time.  After all, an oscilloscope is a time domain instrument, and settling time is always limited by things outside of the resolution and bandwidth.

A big improvement over the scope they replaced, which was around 4x noisier across most conditions:
https://www.eevblog.com/forum/testgear/new-keysight-infiniivision-hd3-oscilloscope/msg5609003/#msg5609003
quite a bit of that is noise folding down from insufficient antialiasing! a significant advantage that pure analog has.

At high sensitivity noise is nominated by the high impedance buffer and aliasing of the full bandwidth becomes insignificant.

I do wonder about the low frequency and flicker noise though, but oscilloscopes are all terrible in this regard except for specialty low frequency instruments, and even they are not that good.

At least the 74LVC125 is only 4 channel, compared with the 74LS125's irritating 6 channels! But D/DB/NS/PW packages still having power connections on opposite corners?!

I like the 74LVC1G devices because of the low pin inductance which isn't shared between multiple outputs. Would your circuit need 4 drivers?

No, I mean the 74LVC125 in general and the SOT-23 single packaged 74LVC1G is actually what I would use.

A 74125 type part is preferred because its open drain output will not couple back through the transistor or diode during turn-off, producing a cleaner rising edge in this case.  The 74125 could be used for positive or negative edges, but unfortunately PNP RF transistors no longer exist.

[2N3055]

The noise is specified as 34 microvolts RMS for the 1 megohm input over 100 MHz.  My typical old analog oscilloscopes have a noise of 28 microvolts RMS over 100 MHz for a 200 MHz input.

So I will give them credit for achieving low noise for a 500 MHz or 1 GHz input because I have nothing to compare with, but they did not achieve low noise in comparison to obsolete slower oscilloscopes.  This may be because the design to support 500 MHz or 1 GHz will be inherently higher noise at lower frequencies.

I'm preparing to make some measurements of noise for comparison and will publish when done.
But preliminary, compared to SDS3104xHD, at 50Ω HD3 only has lower noise up to 20 mV/div. At 50mV/div is similar, and then up from there Siglent has less noise..

At 1 MΩ situation is not that clean cut either.  At 1MΩ KS published only data at 500MHz BW.
Where my SDS3104xHD has 1GHz BW, and if I set digital filter (ERES, same type of thing as what HD3 uses) at 500Mhz, I get much better results.
In fact, with full 1GHz BW  I get better noise levels from 50mV/div up, despite having 2x BW.

I also tested shortly with SDS200xHD, a 500MHz scope with 1MΩ. At 500µV/div to 5mV/div I get 75µV RMS (better), at 10mV/div I get 125µ RMS (slightly worse), at 50mV/div I get slightly better, at 100mV/div etc...
So basically, very comparable results...
Meaning, for the 500 MHz version at 1MΩ, it's no better than SDS200xHD, noise vise..

Advantage is only in 50Ω Path.

[EEVblog]

At 1 MΩ situation is not that clean cut either.  At 1MΩ KS published only data at 500MHz BW.

I believe the HD3 is 500MHz limited at 1M

[David Hess]

At 1 MΩ situation is not that clean cut either.  At 1MΩ KS published only data at 500MHz BW.

I believe the HD3 is 500MHz limited at 1M

With a 1 megohm input, the impedance of the probe cable combined with the 24 picofarad input capacitance limits bandwidth, which is why I was also surprised at the high input capacitance.

The standard method of measurement is a terminated 50 ohm source, and 25 ohms into 24 picofarads is only 265 MHz, which explains why past 500 MHz inputs only had like 10 or 12 picofarads of input capacitance.  They must be using a passive probe termination which provides an even lower impedance to drive that 24 picofarads.  This is not impossible, but it is unusual.

Or maybe they are DSPing it?  If so I would expect a bump in the high frequency noise where the equalization has to add gain.

[tggzzz]

At least the 74LVC125 is only 4 channel, compared with the 74LS125's irritating 6 channels! But D/DB/NS/PW packages still having power connections on opposite corners?!

I like the 74LVC1G devices because of the low pin inductance which isn't shared between multiple outputs. Would your circuit need 4 drivers?

No, I mean the 74LVC125 in general and the SOT-23 single packaged 74LVC1G is actually what I would use.

A 74125 type part is preferred because its open drain output will not couple back through the transistor or diode during turn-off, producing a cleaner rising edge in this case.  The 74125 could be used for positive or negative edges, but unfortunately PNP RF transistors no longer exist.

I presume you mean using the 74125 with input connected to GND and on/off implemented via the OE input.

Have you considered the 74LVC1G07, which has a true open drain output?

[Martin72]

Is it 15µV which is passed through an attenuator and then reduced again

Yes.

Thanks, and your FFT settings ?

 

[ftg]

No, I mean the 74LVC125 in general and the SOT-23 single packaged 74LVC1G is actually what I would use.

A 74125 type part is preferred because its open drain output will not couple back through the transistor or diode during turn-off, producing a cleaner rising edge in this case.  The 74125 could be used for positive or negative edges, but unfortunately PNP RF transistors no longer exist.

Apparently they are getting increasingly rare.
Onsemi has MMBTH81 listed as active, but those are only rated at 600MHz fT.
https://www.onsemi.com/products/discrete-power-modules/rf-transistors/mmbth81

Sadly NXP has obsoleted the 5GHz fT BFT93.

But Renesas is still manufacturing the old Harris/Intersil transistor arrays and HFA3096 has 3x 8.5GHz fT NPN + 2x 5.5GHz fT PNP on the chip.
Unfortunately it is only available in SOIC and not QFN or DFN. 
https://www.renesas.com/us/en/products/space-harsh-environment/harsh-environment-products/harsh-environment-transistor-arrays/hfa3096-ultra-high-frequency-transistor-arrays#overview

[tv84]

But preliminary, compared to SDS3104xHD, at 50Ω HD3 only has lower noise up to 20 mV/div. At 50mV/div is similar, and then up from there Siglent has less noise..

Please explain why that happens, if you can.

[tooki]

How do you figure? As best I can tell, all the scope makers support time-limited “demos” of individual option licenses. It’s just that right now, their reps can give those out for free for testing, but usually just once or twice before their bosses get mad. I wish they just sold those time-limited licenses so we could just license certain features as needed.

Sure they do but the problem is the minimum time interval.

For most that is days. The R&S system can potentially generate licenses for a specific date/hour, with minutes granularity. To fully answer your rent periods wish, you need a system like this.

I mean, ok, that could be. But I struggle to see how whole-day granularity would be a problem. I mean, if we look at the software world, typical time-based licenses are for a month or a year. I have not yet encountered software that charged by the hour!

[tooki]

HD3 UI true annoyances for me (from what I saw on Dave's video):

- the screen size and resolution and consequential no-scaling
- the frameless split screens (give it a cheap look , IMO)
- the vertical axes values being on the right
- the top channels/menu bar should be on the bottom

I leave all the engine analysis for the experts, but at the UI level, the MXO4 is much more polished than the HD3. It looks like KS tried to innovate but, IMO, this is an engineering instrument not a smartphone skin project. Much less if we consider that it's something that can cost ~20k€.

I agree. The menus on the HD3 are too deep. They should design the UI to only open one extra window selected from a primary or secundary menu bar. Brands like Micsig and R&S got this right IMHO.

As an example: selecting measurements on the R&S RTM3004 is much easier with pictures that show what the measurements actually do. I have a Lecroy 7300A Wavepro and that has a similar way of selecting measurements (like the KS HD3) which I loath. Scrolling through a long list on a touchscreen just isn't comfortable. Especially when you need to look for something particular which isn't / can't be sorted by the nature of the items.

Just an FYI, as someone who has used them both: the UI on the RTM series and MXO4 look superficially similar, and do have some commonalities, but they’re actually quite different. IMHO the RTM (and the RTB series, which shares basically the same software) software is appreciably better thought-out than the MXO software, which has many controls spread around in ways that to me make no sense, and result in you tapping around way more than if they were grouped better, and it lacks many of the little touches that are so nice in the RTM/RTB software.

For example, the MXO software does not have the little pictures on measurements that you like on the RTM. Similarly, on triggers, you don’t see the little picture until after you’ve selected a trigger from the drop-down list. IMHO there’s a lot of low-hanging fruit in terms of usability improvements in the MXO software, much of which they could achieve simply by doing what the RTM/RTM (and the portable RTH, whose software is also superficially similar, but different again) do.

[David Hess]

I presume you mean using the 74125 with input connected to GND and on/off implemented via the OE input.

That is right.

Have you considered the 74LVC1G07, which has a true open drain output?

That would work, but I prefer more universal parts.  When I initially looked into this, and before RF PNP transistors went out of production, I anticipated needing an open drain device of both polarities.

Sadly NXP has obsoleted the 5GHz fT BFT93.

Sadly, and there are no replacements that I have found.

But Renesas is still manufacturing the old Harris/Intersil transistor arrays and HFA3096 has 3x 8.5GHz fT NPN + 2x 5.5GHz fT PNP on the chip.
Unfortunately it is only available in SOIC and not QFN or DFN.

The packaging is very important to keep parasitics down, so the old Harris process transistor arrays are not suitable.

[nctnico]

It certainly does not seem there is much "special ASIC acceleration" for FFT in HD3. Unlike R&S that really looks like realtime SA at settings shown.

The older Agilent 7000 series scopes have a much faster FFT than newer models.
For many of my FFT needs, I have used a DSO7104A with delight over the years.
The 3000x, 4000x or 6000x Keysight scopes just suck at FFT update rates.

Your DSO7104A must be different than the one I had. FFT was very slow on the one I had. Especially with 128kpoints FFT enabled. 

[dkonigs]

Also, I note that none of the Agilent/Keysight scopes will do Dallas "1-Wire" serial protocol decode.. Seems like quite an oversight.. I had to get a Picoscope just to do this..

I wonder if they will ever release a tool to allow us to write our own protocol decoder.. I think the HP/Agilent 16700/800/900 series of logic analyzers had this feature..

This is a huge part of why I picked up a Saleae Logic Analyzer a while back.  While the built-in protocol decoding on these scopes is very useful for most of the common standard protocols, its completely useless once you have some vendor-specific and/or bespoke serial protocol you want to work with.  One thing those USB-attached instruments bring to the table is a plug-in architecture where you can write/modify custom decoders to cover those cases.

[Zucca]

Also, I note that none of the Agilent/Keysight scopes will do Dallas "1-Wire" serial protocol decode.. Seems like quite an oversight.. I had to get a Picoscope just to do this..

I wonder if they will ever release a tool to allow us to write our own protocol decoder.. I think the HP/Agilent 16700/800/900 series of logic analyzers had this feature..

This is a huge part of why I picked up a Saleae Logic Analyzer a while back.  While the built-in protocol decoding on these scopes is very useful for most of the common standard protocols, its completely useless once you have some vendor-specific and/or bespoke serial protocol you want to work with.  One thing those USB-attached instruments bring to the table is a plug-in architecture where you can write/modify custom decoders to cover those cases.

Once you try the Picoscope decoding function, it is very hard to go back to anything else. IMHO.

[egonotto]

But preliminary, compared to SDS3104xHD, at 50Ω HD3 only has lower noise up to 20 mV/div. At 50mV/div is similar, and then up from there Siglent has less noise..

Please explain why that happens, if you can.

Hello,

I'm interested in that too.

From what I have seen in the various data sheets, LeCroy is the best in the 1 V/div range with 4.9 mV RMS at 1 GHz.

The Siglent SDS3104X HD has less than 6 mV RMS in this range. Recently I even had only 5.1 mV RMS (at 1 MOhm) with it.

R&S are pretty weak in this range.
The MXOs are a disappointment regarding noise at 1 V/div and RTA and RTM is even worse at 1 V/div.

The HDS is not best in class with 7.26 mV RMS, but it's still not bad. The two times better than others really only applies at high sensitivity and 50 ohms.

I wonder why they specify the noise to 3 digits?

Best regards
egonotto

[2N3055]

But preliminary, compared to SDS3104xHD, at 50Ω HD3 only has lower noise up to 20 mV/div. At 50mV/div is similar, and then up from there Siglent has less noise..

Please explain why that happens, if you can.

Short version:

I think Keysight made good preamplifiers for front end, but ADC is similar in noise to good 12 bit ADC.
Hence in sweet spot where front end stops influencing noise figures ENOB is becoming similar (because ADC is predominant source of noise then).

HD3 at 1GHz and SDS3104xHD both have ENOB of 8.2.

[2N3055]

But preliminary, compared to SDS3104xHD, at 50Ω HD3 only has lower noise up to 20 mV/div. At 50mV/div is similar, and then up from there Siglent has less noise..

Please explain why that happens, if you can.

Hello,

I'm interested in that too.

From what I have seen in the various data sheets, LeCroy is the best in the 1 V/div range with 4.9 mV RMS at 1 GHz.

The Siglent SDS3104X HD has less than 6 mV RMS in this range. Recently I even had only 5.1 mV RMS (at 1 MOhm) with it.

R&S are pretty weak in this range.
The MXOs are a disappointment regarding noise at 1 V/div and RTA and RTM is even worse at 1 V/div.

The HDS is not best in class with 7.26 mV RMS, but it's still not bad. The two times better than others really only applies at high sensitivity and 50 ohms.

Exactly. We must always measure and know our instruments.

I wonder why they specify the noise to 3 digits?

Best regards
egonotto

It looks good? 

[EEVblog]

Thanks, and your FFT settings ?
 

Sorry, don't remember. I'd have to set it up again.

[hpw]

Thanks, and your FFT settings ?
 

Sorry, don't remember. I'd have to set it up again.

And please, to show the maximal possible FFT size as only 64K, while otherwise decimated or not or boxed with related detectors (see given HD3 user manual)

[HighVoltage]

I would like to see the refresh speed of the FFT in comparison to the 3000x at a few different settings.
Thanks!

[2N3055]

I would like to see the refresh speed of the FFT in comparison to the 3000x at a few different settings.
Thanks!

That is going to be hard, because on HD3 (like on R&S scopes) you don't really have all the settings under manual control in a classical FFT sense.
You can get some particular RBW with different windows, sample rate and bin size.

What could be done is to set HD to few different setups, and then trying to provide similar results on SDS3000xHD using it's controls.

For instance, in video, I saw that Dave set HD3 to 3,2GS/s, 128kpts of samples, FFT with 24.4kHz bin size. That is 3.2GS/s divided by 2^17, so 128k of  FFT bins.

So I set similar on Siglent: 4 GS/s, 200kpts, FFT to 128k bins, and got 30,52 kHz bin spacing.
At these settings (and some measurements on), speed was very much the same as HD3. Enabling 4x averaging did not slow things down. With measurements off, it is slightly faster visually, visibly but nothing I would call large improvement.
Taking into account that it does process a bit more data (and internal data path is 16 bit anyways) I would say that at these settings SDS3000xHD processes FFT bit faster.

It could be that HD3 in comparison is faster or slower with smaller or larger number of bins and amount of data to process.  Until thoroughly tested we won't know.
This preliminary simple comparison shows that HD3 and SDS3000xHD are definitely in the same class in this regard.

I would personally be cautious to do some "in depth" testing right now, though.
Despite all the enthusiasm and pomp, HD3 was released in unfinished state..
I would let Keysight feature complete the product and sort it out a bit before testing it in it's actually finished state...

Any testing now could produce overly optimistic (maybe it is fast now but after some debugging they have to implement more safe/slower code) or pessimistic (they find a way to optimize some process yielding faster processing) results.
We have all seen this before, with Keysight, R&S, Siglent....

To say it colloquially, what we have now, what was released, is extended film trailer/teaser not the actual film yet.
Featuring good actors, good filmmakers, from good studio, meaning it should be good.
But I will wait for full premiere to actually pass judgment.

[skander36]

Hi Dave! Thank you for the presentation!
Will be a teardown section?

[ballsystemlord]

Is there some reason that the HD3 was released before the FW was finished?

I really can't imagine why. KS is a big company, not a small startup desperate to make ends meet.