Modern digital scopes: real-time sampling or equivalent-time sampling

[marmad]

For a 100 MHz BW, 250 MS/s is enough to represent the original signal as it complies with the Nyquist theorem.

Digital oscilloscopes can be subject to sampling alias errors. Sampling alias errors occur when the signal has frequency content beyond the Nyquist frequency (which is 125MHz @ 250MSa/s). As mentioned in this Agilent paper:

"...why aren’t oscilloscope sampling rates exactly 2x higher than the oscilloscope’s bandwidth? This is because out-of-band signals may only be attenuated by 10 dB beyond the band-edge of the instrument. Said differently, the frequency response of the oscilloscope does not roll-off infinitely fast and some buffer room is used on the sampling rate to minimize aliasing."

This is the reason the majority of DSOs sample at 4-10x greater than the maximum bandwidth. Unfortunately, the frequency response of the DS1000Z series does not roll-off fast enough to minimize aliasing when sampling at 250MSa/s.

[pascal_sweden]

The analog input path in the frontend attenuates, amplifies, filters, and/or couples the signal to optimize the digitization by the ADC.

How does the analog frontend look like in the Rigol? What sacrifices are made by Rigol in comparison with higher end scopes?

Is this where the high cost comes from in the higher end scopes, or is the majority of the cost driven by the ADC quality as such?

Which types of filters are used in Rigol scopes? Which types of filters are used in higher end scopes?

[edavid]

The analog input path in the frontend attenuates, amplifies, filters, and/or couples the signal to optimize the digitization by the ADC.

How does the analog frontend look like in the Rigol? What sacrifices are made by Rigol in comparison with higher end scopes?

Is this where the high cost comes from in the higher end scopes, or is the majority of the cost driven by the ADC quality as such?

Which types of filters are used in Rigol scopes? Which types of filters are used in higher end scopes?

Rigol doesn't publish any information about their designs, so the little that is known comes from reverse engineering.  It's thought that they use an off the shelf PGA chip with a simple 1 or 2 pole lowpass filter.

Higher end low bandwidth scopes use faster ADCs (or more of the same speed ADCs), and higher end high bandwidth scopes use a combination of custom multipole analog filters and DSP to get closer to the Nyquist limit.

[David Hess]

I am actually disappointed in the current crop of 100 MHz 250 MS/s DSOs because of their lack of ETS.

What "crop" are you referring to? The single DS1000Z model from Rigol, sold as three FW-delineated versions?

The current crop includes the index graded and high waveform aquisition rate DS1000Z and DS2000A series including their MSO cousins which Rigol treats separately.  That is a lot more than three models but I would consider them to be two basic implementations.  Many (all?) of their predecessors support ETS.

The result is shown in the time domain in the above video as what I like to call "wobbulation".  I suspect the same thing on one of the Rigol oscilloscopes with an index graded display makes for waveform thickening which is mistaken for noise.

Yes, we already know you weren't happy with the DS1000E you bought. It still doesn't explain how whatever nonlinearity and sampling clock errors you feel were present in the 5x dual-ADC-chip DS1000E series were magically transferred to the redesigned, single quad-ADC-chip DS1000Z model.

I never bought any Rigol oscilloscope but that had nothing to do with ETS or graded index display support or aliasing in the digitizer.  It was a result of Rigol's misleading documentation and sales representatives as well as some brief evaluations.  This was before the DS1000Z and DS2000A series were available which I might have considered.

Why would you expect the DS1000Z (or DS2000A) digitizer to be free of these errors?  Almost all DSOs display them to one extent or another (*) and even if the DS1000Z (or DS2000A) ADC was free of them, it is still at the mercy of its clock source which is usually a significant source.

I am suspicious that Rigol left ETS out on these models because the sampling performance could not support it but it could have been do to market segmentation or cost.  The graded index display and high waveform acquisition rate would tend to mask the aliasing as noise so it would not be apparent anyway making ETS less useful.

I knew about David's point and was ignoring it for the argument that a scope should have an adequate filter in place.
I have not seen the result for the Rigol, but ok, if it doesn't then the issue becomes relevant.

I'm not sure how a low-cost 100MHz DSO could have an adequate enough filter for a 125MHz Nyquist frequency. A Gaussian frequency response would only be around -5db at 125MHz, and even with a flat-response, it would only be approx. -9db at 125MHz. Perhaps the 50MHz model's frequency response is adequate. The DS1000Z series is, for all practical purposes, a 1/2 channel 50/75/100MHz DSO and a 3/4 channel (50?)25MHz DSO.

Even a high-cost DSO is not going to have an adequate enough analog antialiasing filter and doubt such can practically exist at any price except when it is not needed.  The ways to combat this include more linear digitizers which have less clock error and higher sampling rates.

I have never argued that analog antialias filtering in any DSO was inadequate because instead I have argued that analog antialias filtering in any DSO is useless if only because it makes the transient response and bandwidth vary with sample rate.  At low sample rates with an aliased input signal, it produces just as deceptive a display as aliasing would if not more deceptive.

(*) I have seen some high end DSOs which did not but never got to test one specifically for it.  I tried to replicate wobulation on my 2232 which was handy and could not but only because it has no way to disable ETS.  Aliasing produced by sampling error was apparent but not in a particularly revealing way.

[grumpydoc]

I remember wanting a PM3382/PM3384/PM3392/PM3394.  The PM3394 is a 4 channel, 200 MHz, combination analog/DSO with a 200 MS/s real time sample rate which is shared in somewhat between channels.  They support real time and random equivalent time sampling.  Feature wise they seemed better than the Tektronix equivalents.

I have a 3382 which is the 100MHz 2+2 channels model. I rather like it as an analogue 'scope and the digital side is occasionally useful but a bit limited in sample rate and memory depth (8k samples/channel IIRC).

 I've had my eye out for a 3394 but have only been able to spot them in the wild for excessive amounts of money

[nctnico]

I'm not sure how a low-cost 100MHz DSO could have an adequate enough filter for a 125MHz Nyquist frequency. A Gaussian frequency response would only be around -5db at 125MHz, and even with a flat-response, it would only be approx. -9db at 125MHz. Perhaps the 50MHz model's frequency response is adequate. The DS1000Z series is, for all practical purposes, a 1/2 channel 50/75/100MHz DSO and a 3/4 channel (50?)25MHz DSO.

Even a high-cost DSO is not going to have an adequate enough analog antialiasing filter and doubt such can practically exist at any price except when it is not needed.  The ways to combat this include more linear digitizers which have less clock error and higher sampling rates.
[/quote]
The worst case 250Ms/s samplerate for the DS1000Z could be sufficient with a very sharp anti-aliasing filter and lot's of math to reconstruct the signal that close to the Nyquist frequency. With my own (math intensive) algorithms I achieved proper signal reconstruction up to 0.45fs (112MHz at 250Ms/s).
I did some anti-aliasing tests on my Siglent  SDS2024 and it seems that the anti aliasing filter works pretty well. There is only a tiny bit of aliasing.

[marmad]

I am actually disappointed in the current crop of 100 MHz 250 MS/s DSOs because of their lack of ETS.

What "crop" are you referring to? The single DS1000Z model from Rigol, sold as three FW-delineated versions?

The current crop includes the index graded and high waveform aquisition rate DS1000Z and DS2000A series including their MSO cousins which Rigol treats separately.  That is a lot more than three models but I would consider them to be two basic implementations.

The DS2000A is not a "100MHz 250MSa/s" DSO, thus doesn't fit your original description. Not only that, but it's a completely different design involving a different ADC, different processor, different architecture, etc; i.e. extremely dissimilar internally (much closer to the DS4000 then the DS1000Z). Again, your original "100MHz 250MSa/s" description only applies to a single base model: the DS1000Z (and, of course, all it's permutations: -S, MSO, etc).

Many (all?) of their predecessors support ETS.

None of the new generation of high level (>= 64) intensity graded DSOs have ETS, beginning with Agilent's 2000 / 3000 X-Series. The lack of it on Rigol's later-released UltraVision DSOs is as likely to be attributed to copying the Agilent X-Series' list of features as it is to anything else.

Why would you expect the DS1000Z (or DS2000A) digitizer to be free of these errors?  Almost all DSOs display them to one extent or another (*) and even if the DS1000Z (or DS2000A) ADC was free of them, it is still at the mercy of its clock source which is usually a significant source.

I never said that I expected them to be free of all errors, but you seem to be implying errors which have yet to be proven. Logically, I would think that using four internally-interleaved ADCs would have less interleave distortion (and perhaps a lower noise level) than when using multiple externally-interleaved ADCs, but I haven't seen any evidence either way.

I am suspicious that Rigol left ETS out on these models because the sampling performance could not support it but it could have been do to market segmentation or cost.

Again, none of the new generation of intensity graded DSOs have it (Agilent, Siglent, etc), so I can't see how one could logically deduce anything about Rigol's performance from the fact that the feature is missing on their new DSOs too.

[David Hess]

I remember wanting a PM3382/PM3384/PM3392/PM3394.  The PM3394 is a 4 channel, 200 MHz, combination analog/DSO with a 200 MS/s real time sample rate which is shared in somewhat between channels.  They support real time and random equivalent time sampling.  Feature wise they seemed better than the Tektronix equivalents.

I have a 3382 which is the 100MHz 2+2 channels model. I rather like it as an analogue 'scope and the digital side is occasionally useful but a bit limited in sample rate and memory depth (8k samples/channel IIRC).

I've had my eye out for a 3394 but have only been able to spot them in the wild for excessive amounts of money

I have seen the 3394 at an affordable price a couple of times but was always leery about attempting to repair and maintain one.

The analog operation makes for a very effective sanity check and also usually makes up for the lack of index grading in digital storage mode.  In a very effective way, both modes of operation complemented each other in this type of oscilloscope.  It was many years and a lot of dollars before DSOs included DPO like functionality which could replace or attempt to replace an analog oscilloscope.

I do not know about Philips but Tektronix continued to make this type of oscilloscope until 1994 and it outlived its close analog only cousins by 3 years.  Looking at the production dates make me suspect that they only stopped making them because they stopped making lower bandwidth 100 MHz CRTs at the same time.  Higher bandwidth CRTs and analog oscilloscopes held on for at least two more years but maybe they just had extras in stock.

[Wuerstchenhund]

None of the new generation of high level (>= 64) intensity graded DSOs have ETS.

That's not entirely correct. The new LeCroy Wavesurfer 10 which only came out recently offers 256 intensity grades and does offer ETS (RIS), although only up to 50GSa/s. The other new Wavesurfer (3000) also offers ETS (50GS/s), and I'm pretty sure it offers at least 64 intensity levels (couldn't find it in the spec). Since the WS3000 is a Siglent SDG3000 the same is probably true for the Siglent variant as well.

The WS3000 roughly competes with the Agilent/Keysight DSO-X3k and the WS 10 with the DSO-X-4k Series.

This is in line with the older generation of LeCroy scopes. The predecessor WaveSurfer Xs-B which is still sold also offers 256 intensity grades and ETS (RIS) up to 200GS/s. Same is true for many other LeCroy scopes going back to at least 2005 (including my WaveRunner 64Xi). It might be longer but I couldn't find data about the number of intensity levels on the even older scopes.

[marmad]

That's not entirely correct. The new LeCroy Wavesurfer 10 which only came out recently offers 256 intensity grades and does offer ETS (RIS), although only up to 50GSa/s. The other new Wavesurfer (3000) also offers ETS (50GS/s), and I'm pretty sure it offers at least 64 intensity levels (couldn't find it in the spec). Since the WS3000 is a Siglent SDG3000 the same is probably true for the Siglent variant as well.

Sorry, I meant the new generation of lower cost, intensity-graded DSOs; I would assume the feature would continue to be present on some new intensity-graded DSOs. I shouldn't necessarily even have included the Agilent 3000 X-Series in my above post - but they just happen to share the lack of ETS with their lower-cost 2000X sibling.

[David Hess]

I am actually disappointed in the current crop of 100 MHz 250 MS/s DSOs because of their lack of ETS.

What "crop" are you referring to? The single DS1000Z model from Rigol, sold as three FW-delineated versions?

The current crop includes the index graded and high waveform acquisition rate DS1000Z and DS2000A series including their MSO cousins which Rigol treats separately.  That is a lot more than three models but I would consider them to be two basic implementations.

The DS2000A is not a "100MHz 250MSa/s" DSO, thus doesn't fit your original description. Not only that, but it's a completely different design involving a different ADC, different processor, different architecture, etc; i.e. extremely dissimilar internally (much closer to the DS4000 then the DS1000Z). Again, your original "100MHz 250MSa/s" description only applies to a single base model: the DS1000Z (and, of course, all it's permutations: -S, MSO, etc).

I was not as precise as I could have been.  I should have been more inclusive and said "the current crop of 100 MHz 250 MS/s and 300 MHz 1 GS/s DSOs".  The DS2000A series is going to suffer from the same aliasing problem to a somewhat lessor degree unless its digitizer performs worse.

Many (all?) of their predecessors support ETS.

None of the new generation of high level (>= 64) intensity graded DSOs have ETS, beginning with Agilent's 2000 / 3000 X-Series. The lack of it on Rigol's later-released UltraVision DSOs is as likely to be attributed to copying the Agilent X-Series' list of features as it is to anything else.

After thinking about it, I have a better explanation; these oscilloscopes which support intensity grading including the Rigol DS1000Z and DS2000A series and Agilent ones you mention do have ETS but not in the way it is generally thought of and the manufacturers are not advertising it as such.

During acquisition, the trigger to sample clock delay is detected in a way similar to how a transition midpoint timing TDC works (or they use an analog TDC but the result is the same unless there is aliasing); this essentially involves a real time reconstruction filter similar if not identical to a sin(x)/x reconstruction filter before the digital trigger.  The trigger to sample clock delay is used to align the waveform acquisition record with the display record.

The difference between this an an older DSO without an index graded display using ETS is that a histogram is generated in real time and later transferred to the display.  An older non-DPO DSO would instead transfer the waveform acquisition record to the display and then optionally generate the histogram and graded index display.

Both methods result in equivalent time sampling whether they call it that or not.  With some difficulty, it should be possible to derive the ETS sampling rate from the specifications but in practice it is high enough not to matter to the user who would not be able to see the effects of ETS in the presence of index grading.

Why would you expect the DS1000Z (or DS2000A) digitizer to be free of these errors?  Almost all DSOs display them to one extent or another (*) and even if the DS1000Z (or DS2000A) ADC was free of them, it is still at the mercy of its clock source which is usually a significant source.

I never said that I expected them to be free of all errors, but you seem to be implying errors which have yet to be proven. Logically, I would think that using four internally-interleaved ADCs would have less interleave distortion (and perhaps a lower noise level) than when using multiple externally-interleaved ADCs, but I haven't seen any evidence either way.

Every DSO with a real digitizer suffers from these errors and they become more serious as the sample rate approaches the Nyquist frequency.  All someone has to do is measure a clean sine wave which is close to but below the Nyquist frequency (Technically it does not have to be below the Nyquist frequency and it is not even desirable for it to be so to make this measurement.) on a DS1000Z or DS2000A in single shot mode and and they will become apparent just like in that video I linked to.

Externally interleaved ADCs do usually perform worse in this respect and the internally interleaved ones should be better if only because they generally implement some form of self calibration to minimize it.  Both are subject to external clock jitter however.  The specifications of the internally interleaved converters that TI makes do show that more distortion occurs when interleaving is used but the difference is small.  Unfortunately they do not include a complete comparison between modes in their datasheets.

I am suspicious that Rigol left ETS out on these models because the sampling performance could not support it but it could have been do to market segmentation or cost.

Again, none of the new generation of intensity graded DSOs have it (Agilent, Siglent, etc), so I can't see how one could logically deduce anything about Rigol's performance from the fact that the feature is missing on their new DSOs too.

See above.  They did not actually leave it out and neither did Agilent or Siglent.  You have convinced me that they still implement ETS but not that the aliasing issue is moot.

Actually, this convinces me that the aliasing issue is more important because it occurs before the reconstruction filter and trigger and this may explain some very odd demonstrations LeCroy made in the past where they showed their very low jitter "digital trigger" which did not look to be very low jitter.

[pascal_sweden]

Interesting thread which I started, which brought up a lot of knowledge up to the horizon.
Although my contribution itself is limited, I initiated the post, so some contribution in that respect

[David Hess]

How does the analog frontend look like in the Rigol? What sacrifices are made by Rigol in comparison with higher end scopes?

The analog front ends have gotten a lot simpler because of integration.  You can get an idea of what is involved by looking at the datasheet for the TI LMH6518 although it does not show the high input impedance buffer or high input impedance attenuator switching which precedes it.  I would be most interested in seeing how that later is done now although I know how I would do it which almost matches some photos from teardowns I have seen.

For a general idea of what is before the first integrated low impedance amplifier, check out chapter 7 of "The Art and Science of Analog Circuit Design" which discusses almost modern oscilloscope vertical input amplifiers.  At the end of the chapter the author, Steve Roach, speculates about the designs we are probably buying now:

http://goo.gl/R44h5l

Based on approximately when various Rigol models were released and their performance, it looks to me like low end DSOs have been following the trend in increasing integration of ADCs, FPGAs, and wideband integrated analog electronics intended for ATE (automated test equipment).  Where older high performance designs would have taken 2 or 4 separate interleaved ADCs and memory channels to meet their sample rate goals, newer models have that all integrated into one ADC and one FPGA.

Is this where the high cost comes from in the higher end scopes, or is the majority of the cost driven by the ADC quality as such?

High bandwidth high sample rate ADCs are not cheap and better ones cost more.  Some designs interleave several ADC for higher sample rates although that is more commonly done now in an integrated manner if possible.

The analog design before the digitizer is not trivial either and becomes increasingly difficult at wide bandwidths for all kinds of arcane reasons.

As far as the distribution of costs, I suspect that is rather closely guarded and largely irrelevant to the price you have to pay when buying one.

Which types of filters are used in Rigol scopes? Which types of filters are used in higher end scopes?

In all of the designs I am familiar with, simple bandwidth limit filters are implemented not for antialiasing but for noise reduction.  The integrated front-end amplifiers which are part of the analog signal chain like the TI LMH6518 include selectable single pole filters.

I have only seen analog antialiasing filters in DSOs that I would consider toys.

[David Hess]

I was hoping someone would bring this up.
I certainly agree that sin(x)/x reconstruction is completely sufficient to reconstruct an unaliased waveform.

Tek on the issue:
http://www.tek.com/dl/55W_17589_2.pdf

I have not read it all, but they basically summarise that there is no practical difference between Sin X/x and ETS.

... LeCroy stuff ...

Bottom line is that all the manufacturers seem to say a similar thing, and I have a more detailed paper mathematically proving 2.4x minimum in some way (can't find it now). So for all but the most critical applications, a 250MSPS 100MHz scope with Sin X/x should be just fine.

Dave, I had a thought about that Tektronix paper (*) in light of my post above about the similarities between traditional ETS and what the current Rigol oscilloscopes and others are doing with digital triggering and this EEVBlog conversation going on here:

https://www.eevblog.com/forum/testgear/new-rigol-ds1054z-oscilloscope/315/

There is a critical difference between traditional ETS using an analog time delay counter and averages of a sin(x)/x reconstructed waveform when digital triggering is used after sin(x)/x reconstruction which is common in DSOs now.

If I use an DSO which supports traditional ETS with a waveform that is above the Nyquist frequency, the TDC (time delay counter) returns the delay between the trigger and sample clock effectively raising the sample period to the resolution of the TDC.  On a Tektronix 2230 for instance, this allows a 20 MS/s digitizer to operate at an equivalent of 2 GS/s but limited of course to repetitive waveforms.

On a modern DSO which relies on reconstruction, usually sin(x)/x but it could be something else, before the digital trigger, a measurement is made which returns the delay between the trigger and sample clock again effectively raising the sample period to the resolution of the TDC but now the TDC uses something like transition midpoint timing which is a complete digital process implemented in logic.  The manufacturers are not calling this capability ETS and maybe the following is a good reason.

If the waveform is above the Nyquist frequency, the reconstruction is going to completely fail causing the trigger to fail as well.  In this case, averaging after sin(x)/x reconstruction is useless because of aliasing before the trigger.

If triggering occurs on a fast edge which is not bandwidth limited and so contains significant frequencies above the Nyquist limit, aliasing during sin(x)/x reconstruction before the trigger is going to corrupt the trigger to sample clock measurement although the results will be less dire.  I think averaging will still work but the histogram of the signal is going to show a lot of excess noise produced by aliasing and corruption of the trigger.

(*) Thanks for pointing this paper out to me.  I know I have read it in the past but lost track of it and have been searching for it in a desultory manner ever since.

[Wuerstchenhund]

That's not entirely correct. The new LeCroy Wavesurfer 10 which only came out recently offers 256 intensity grades and does offer ETS (RIS), although only up to 50GSa/s. The other new Wavesurfer (3000) also offers ETS (50GS/s), and I'm pretty sure it offers at least 64 intensity levels (couldn't find it in the spec). Since the WS3000 is a Siglent SDG3000 the same is probably true for the Siglent variant as well.

Sorry, I meant the new generation of lower cost, intensity-graded DSOs; I would assume the feature would continue to be present on some new intensity-graded DSOs. I shouldn't necessarily even have included the Agilent 3000 X-Series in my above post - but they just happen to share the lack of ETS with their lower-cost 2000X sibling.

I guess with the DSO-X3k it's difficult as technically it shares a lot with the DSO-X2k but competes in a higher price bracket.

The interesting thing with these new LeCroy scopes is that the ETS sample rate has dropped to only 50GSa/s while previous models offered 200GSa/s. Either there's a technical reason for the drop (i.e. slower embedded processing on the new scopes vs full PC platform on the older ones), or it's simply an indication that ETS has lost relevance in this day and age.

[David Hess]

The interesting thing with these new LeCroy scopes is that the ETS sample rate has dropped to only 50GSa/s while previous models offered 200GSa/s. Either there's a technical reason for the drop (i.e. slower embedded processing on the new scopes vs full PC platform on the older ones), or it's simply an indication that ETS has lost relevance in this day and age.

My guess is that they decided they did not need 5 picosecond timing resolution in the external TDC and 20 picoseconds was sufficient.  There are lots of possible design reasons this may be the case including noise considerations.  One common one on old DSOs is where the display record only has say a maximum timing resolution of 50 picoseconds but the TDC has a 10 picosecond resolution.

The lack of external ETS in low end DSOs may simply represent a lower cost point and market segmentation.  All digital ETS is *cheaper* to implement once you have some spare FPGA or ASIC logic handling decimation or DPO functions.  Of course it is not needed at all if you have a sampling rate sufficiently high above your analog bandwidth.

[grumpydoc]

I have seen the 3394 at an affordable price a couple of times but was always leery about attempting to repair and maintain one.

The analog operation makes for a very effective sanity check and also usually makes up for the lack of index grading in digital storage mode.  In a very effective way, both modes of operation complemented each other in this type of oscilloscope.  It was many years and a lot of dollars before DSOs included DPO like functionality which could replace or attempt to replace an analog oscilloscope.

Prices in the 'states seem more reasonable, I've seen a few for $500 or less but shipping to the UK is prohibitive. Someone locally had several which he sold through ebay but wanted about £750 for a 3394B (about US $1200). At that price several modern 'scopes can be bought new. I did offer £375 which I thought fair but no dice. It took him a while but in the end it looks as though he sold them for not far off what he wanted - proving him right and me wrong on price.

I have to say that I don't really understand this; it's a nice 'scope and I'd quite like to own one but not for the same price as a Rigol DS1104Z-S or nearly the price of a Siglent SDS1204CFL.

As to repair - yes, well, there is a lot of unobtanium in these 'sopes so if they fail it might be difficult to find spare parts, and Philips designs could be idiosyncratic. It's probably no worse than getting a new U800 for a 2445 or 2465 though. Service manuals are available on the net, at least.

Maintenance is perhaps not so much a problem. If you have the equipment to calibrate a 2465 then a 3394 shouldn't be too much of a challenge - in fact there's an "auto cal" function which serves for keeping in trim on a day-to-day basis - certainly I've found my 3382 very accurate once warmed up.

[David Hess]

I have seen the 3394 at an affordable price a couple of times but was always leery about attempting to repair and maintain one.

The analog operation makes for a very effective sanity check and also usually makes up for the lack of index grading in digital storage mode.  In a very effective way, both modes of operation complemented each other in this type of oscilloscope.  It was many years and a lot of dollars before DSOs included DPO like functionality which could replace or attempt to replace an analog oscilloscope.

...

I have to say that I don't really understand this; it's a nice 'scope and I'd quite like to own one but not for the same price as a Rigol DS1104Z-S or nearly the price of a Siglent SDS1204CFL.

When I was looking, the DS1000Z and DS2000A series of Rigol oscilloscopes did not exist.  Having decided that their other models were unsuitable, I rebuilt a pair of Tektronix 2230s for about $80 each and a 2232 for $120 instead.

As to repair - yes, well, there is a lot of unobtanium in these 'sopes so if they fail it might be difficult to find spare parts, and Philips designs could be idiosyncratic. It's probably no worse than getting a new U800 for a 2445 or 2465 though. Service manuals are available on the net, at least.

Maintenance is perhaps not so much a problem. If you have the equipment to calibrate a 2465 then a 3394 shouldn't be too much of a challenge - in fact there's an "auto cal" function which serves for keeping in trim on a day-to-day basis - certainly I've found my 3382 very accurate once warmed up.

Calibration is definitely a problem although I would have needed the equipment to do transient response calibration anyway whether I had a new DSO or not and that is the hardest one to do.

I never really considered the 24xx series analog oscilloscopes because I already had a 2247A and a fast 7000 series mainframe.  The Philips oscilloscopes were not common enough for me to risk having to find a parts donor.

[pascal_sweden]

The PM3340 was a 2GHz scope with 250MSa/s real-time sampling rate and 14bit 10bit vertical resolution which also offered 2GSa/s in ETS mode. It was a good scope at it's time (1989; I had a PM3343 PM3320A back then which was the 200MHz version of the PM3340) but by today's standards it's a boat anchor. It's also difficult to fix with lots of unobtainium parts. The high vertical resolution made (and still makes) especially the PM3343 PM3320A sought after for audio work, though.

Are you sure that it was not 12-bit? It contained the ADC601JG.


The datasheet can be found here.
http://pdf.datasheetcatalog.com/datasheet/BurrBrown/mXyrvyx.pdf

The ADC601JG can do a fast conversion in 900ns.

Was the maximum sample rate limited by the processing power of the Motorola 68000 processor?

This scope featured already a 12-bit AD converter in 1989.

Most of the mainstream scopes in 2016 still have a 8-bit AD converter.

We are talking more than 25 years ago! What happened?

Do the manufacturers want to keep us stupid?

Are we facing the same evolution as with the Maya civilization?

[wraper]

Are you sure that it was not 12-bit? It contained the ADC601JG.


The datasheet can be found here.
http://pdf.datasheetcatalog.com/datasheet/BurrBrown/mXyrvyx.pdf

The ADC601JG can do a fast conversion in 900ns.

Was the maximum sample rate limited by the processing power of the Motorola 68000 processor?

This scope featured already a 12-bit AD converter in 1989.

Most of the mainstream scopes in 2016 still have a 8-bit AD converter.

We are talking more than 25 years ago! What happened?

Do the manufacturers want to keep us stupid?

Are we facing the same evolution as with the Maya civilization?

You talk about 1.1 MSPS ADC and then bash modern scopes for no improvement.

[pascal_sweden]

You have a point about the sample rate, but it would have been nice if 12-bit would have become mainstream as well, next to faster sample rates.

[borjam]

This is the reason the majority of DSOs sample at 4-10x greater than the maximum bandwidth. Unfortunately, the frequency response of the DS1000Z series does not roll-off fast enough to minimize aliasing when sampling at 250MSa/s.

I have observed this on my "enhanced" DS1074Z actually.

I was looking at some Wordclock and S/PDIF signals. When using a single channel I saw a nice square signal. When enabling another channel, I noticed some "ringing" on the right half of the high part of the signal. I think it was aliasing causing a real misbehavior of the sin(x)/x interpolation. 

At first I thought that I had some odd problem with the converter, but I think it was aliasing. It didn't help that I was looking at a clock signal expecting a 75 ohms termination using a 10x oscilloscope probe (LeCroy PP005) with the BNC adaptor and no proper 75 ohm termination.

I can try to reproduce it and show some captures.

[rf-loop]

....causing a real misbehavior of the sin(x)/x interpolation. 

If not reason in just this special case but...

Rigol DS1000Z Sin(x)/x is just bad joke.


About DS1000Z it is explained here  from start  to message #7
https://www.eevblog.com/forum/testgear/rigol-ds1074z-weird-signal-level-problem/msg649723/#msg649723

[borjam]

....causing a real misbehavior of the sin(x)/x interpolation. 

Rigol DS1000Z Sin(x)/x is just bad joke.


About DS1000Z it is explained here  from start  to message #7
https://www.eevblog.com/forum/testgear/rigol-ds1074z-weird-signal-level-problem/msg649723/#msg649723

Not exactly, mine was certainly not aliasing on the fundamental frequency, as I was displaying several periods on screen, which means that necessarily my sampling frequency was higher than double the fundamental frequency. I think that some harmonics were causing aliasing due to not good enough filtering. The interpolation (if I remember well) was just going nuts on it

Anyway I will try to do some captures this week. At first it was quite puzzling.

[macboy]

Philips PM3340: This scope offered 2GS/s back at that time. But it was equivalent-time sampling. Does anybody know if it also offered real-time sampling, and what the actual bandwidth was?

The PM3340 manual shows that it has a 10 bit ADC, not 14 bit.

I had a look at some old documents. The scope I had was a PM3320A, not PM3343 (not sure the latter even exists).

I couldn't find the specs on a quick search but you're probably right that it was 10bit only.

Are you sure it was 250MSPS?  It's hard to believe they could build even 10bit  250MSPS ADCs back then.

Yes, 200MHz bandwidth and 250MSa/s sample rate.

As someone already pointed out, the PM3340 was 2 GHz, not 2 GSps. I have the PM3320A, little cousin to 3340. The PM3320A is only 250 MHz but did up to 10 GSps effective-time sampling. At 5 ns/div the register is 512 samples deep, which affords 50 samples per division (slightly more than 10 divisions are shown on screen). Then 50 samples / 5 ns is 10 samples per ns, or 10 GHz.  The PM3320A service manual goes into much more detail about the sampling system than the user manual. Do you have the PM3340 service manual? The following is for the PM3320A but the PM3340 is so similar this will be useful to understand the PM3340 too. The PM3320A has three sampling modes: Real time, effective time, and random.

The PM3320A had two different real-time sampling modes. First, using the ADC directly. This was good up to 200 kSps, so 2 ms/div (4096 samples/ ~10 div screen or exactly 400 samples per div. At 2 ms/div, that works out to 5 microseconds/sample or 200 kSps). The other way is using the CCD, a charge-coupled device, a.k.a. bucket-brigade device. Analog samples are clocked into the CCD at a high rate, then clocked out later at a lower rate (50 kHz or so I think) for digitization. The CCD was only 512 samples deep (actually there are two 512 sample CCDs, but every other sample on each was the ground reference, so still 512 effective samples), so for true real time, the resolution is limited to 512 samples/screen (1/8 of the maximum resolution), with interpolation between those for the display. If you use the "Max Res" function, then it combines 8 captures of 512 samples each, shifted slightly in time, to give a 4096 sample deep capture, but that's not real time any longer. The maximum real-time sample rate is 250 MSps at 200 ns/div (2 Gsps effective time at 200 ns/div). If using "Max Res" at the same time/div setting, the effective time sample rate is then 2 GSps (8x as high).

Beyond the 200 ns/div or 250 MSps limit, the PM3320A did random sampling. The CCD would still be filled at 250 MSps, but when reading/digitizing the samples, only 1 in N were kept and stored in the register (sample memory). At 5 ns/div, only 1 in 40 are kept, meaning that only 6 samples are captured and stored for each trigger. The displayed waveform is build up over many triggers, so obviously the input waveform must be repetitive and stable. It could take a few seconds to get a complete waveform on the display, and if you were using averaging to reduce noise, then you are in for a good wait while enough of each sample slot are captured. Luckily you can see the waveform building on the display. It's called random sampling because the phase/time difference between the 250 MHz sampling clock and the trigger is effectively random, so the first captured sample is randomly positioned on the incoming waveform. This contributes to the rather long capture time, luck determines when you get the full set of 512 samples from those 6 randomly placed samples per trigger event.

The antiquated sampling system of the PM3320/PM3340 is offset by a couple of features that set this apart from modern competition. First, true 10 bit resolution. I don't need to say more. Second, the screen is 4k. Yes that's right, 4k. The displayed waveform on the CRT is 4000 pixel wide by 1024 tall. What's the screen resolution of your Rigol? or your >$30k Keysight?