| Products > Test Equipment |
| Functional comparison of R&S RTB2000, Siglent SDS2000X and Keysight DSOX1000 |
| << < (22/39) > >> |
| RBBVNL9:
OK, lots of input from various persons, need to find some time to digest all of it (got other work to do today...) --- Quote ---(or mS as you said, aka milli Siemens what is conductance SI derived unit) --- End quote --- You caught me there! It usually upsets my when people make errors in SI units or prefixes, but here I had a slip of the pen and did it myself :( |
| 2N3055:
--- Quote from: Someone on April 05, 2022, 10:37:16 am --- --- Quote from: 2N3055 on April 05, 2022, 08:39:35 am ---They also mention casually memory sizes, but 3000T has only 500k of sample memory when doing 4ch+ digital normal mode (4 buffer-2 per ch-1 with digital -0.5 Mpts for ping pong buffers. 1Mpts for Single mode). Scope with 100Mpts will be 100x slower everything else being equal. Scope with 500MPts will have soo much more work to do. --- End quote --- You can keep pulling up the same "small memory" argument, except it is common for memory to change in different acquisition settings across most scopes. Those characteristics are not avoided in the Keysight data sheets or left as only up-to/maximum/peak values, and made very clear in the manuals (contrasting to these other examples discussed here where it is not clear at all what is expected). But you make the false claim that there is a comparison with memory depth, the waveform capture measurements that competitors use keep the memory depth very short to inflate their numbers. Look at how much slower the competitors are despite choosing less memory! (when at same memory they are far behind). Scopes are not some computer system where the parts are put together in imbalanced ways by the end user, they are a finished product from the manufacturer who has decided on the system performance. Keysight make a real time scope with 100Mpts of memory, the EXR or MXR series, without a significant drop in waveform rate (keeping the system balanced). Where you keep falling over is trying to make out like all scopes work/function the same. The Keysight megazoom method has been to decouple display and waveform memory, they are two different paths that dont interact. Acquisition data is piped to the display plotter/memory (through decimation etc) separately to the waveform memory. As you say before, to make things fast they chose to put most of the emphasis/features on the decimated view (positive example: fast eye diagrams). Most other scopes draw the waveforms from the acquisition memory, and take measurements from the original data (Lecroy being the extreme example of that, positive example: higher resolution measurements). Completely different with advantages and disadvantages, for comparisons they are best listed as characteristics rather than put as a good/bad binary check box against specific/your preference. --- End quote --- I'm not sure what I wrote was false. Fact is that real time processing of 250 Mpts is harder than 4 MPts. Which, I agree with you, is no problem if you're sampling at such sample rate and timebase that all scopes take only 1000 samples.. I wasn't attacking anything and was not commenting OP specific results on this test but in general this old (you surely remember massive talk about this on previous many occasions) topic. Keysight was always big on marketing this Wfms/s advantage (which is real) against other scopes from competition. Always pointing out that this is very important advantage of Megazoom IV to other scopes that don't have it, conveniently forgetting their own higher ends scopes also having 100 Wfms/s if they had same large memory as competition. And thank you for pointing out that Keysight managed to make new gen of scopes that manage to have large memory and fast Wfms/s. Only problem is that those still barely achieve 200000 Wfms/s, same as lowly 1000X. And that despite massive processing power of high end scope with prices north of 20000 USD.. That shows you what price and performance is needed to make a 200 Mpts scope that will achieve high Wfms/s rates. So in a range of prices of several thousands USD, compromises will exist. One extreme is 1 MWfms/s on KS 3000T with small memory. Other side is hundredths of MPts at slower update rate. An it is not about sheer amount of data being processed. It is about architecture that changes with larger memories and high bandwidths needed. Architecture can be more easily optimised if you have small memory. I suspect KS is using dual port memory architecture somewhere in side Megazoom IV to achieve simultaneous sampling in one buffer and rendering from another to achieve ping pong double buffering.. Or maybe hardware bank switching of memory pages onto two separate address/data spaces.. To achieve bandwidths needed, they use wide words and banks, and interleaving. Bottom line small memory (full size small memory) allowed them to optimize architecture to fastest possible for fast acquisition in circular buffer/trigger/switch to other buffer/render in parallel. FPGAs, as powerful as they are, are not as flexible as ASIC, where you can really draw up what you want (*). That means that even with same sizes of data, Megazoom will be faster. It is optimized for that. For instance, datapath on Megazoom IV is 8 bit. On SDS2000X+ is 16 bit, courtesy of 12+bit capable architecture. Etc etc... I'm not trying to pretend that all scopes function the same way. Quite the opposite. General population are thinking of them like they work the same way. They don't, and those types of comparisons end up being looked upon as comparisons which one is better, instead of which one is different and how. And how can you best utilize the tools they provide. They each are best when used and considered in a specific way. These comparisons end up being perceived as some scoring competition, instead of nice resource for people to dig in to make their own decision based on good quality resource. I believe Rudy's original intention was exactly that. And I try to explain that and you say the same, except I'm wrong... I think that sometimes you overestimate my English. Although I'm quite eloquent on occasion, it takes a lot of effort to find right words. It is hard to express what you think in foreign language, despite good command of it in general. False modesty on the side, I know my English is quite OK, but far from efortles or perfect. In a short, I agree with your last paragraph (except first sentence), that is exactly what I try to say. You cannot simplistically compare these platforms based on few numbers taken out of context. Honestly, I miss MSO5000 from Rigol in this. It would have been good to add it to what so far I see as very good and impartial research by Rudy. There have been quite a few discussions where there are no good info on real performance form real life MSO5000 users. I know there are quite a few of them out there, but maybe one or two contribute quality info. (*) "I'm not bad. I'm just drawn that way." Jessica Rabbit |
| 2N3055:
--- Quote from: Someone on April 05, 2022, 10:48:18 am --- --- Quote from: rf-loop on April 05, 2022, 10:22:12 am --- --- Quote from: RBBVNL9 on April 05, 2022, 09:43:22 am ---SDS - “Seq. Acq. Switch” set to off. When it is turned on, I see sets of n acquisitions (where n is the number set by the “Seq Segment” parameter” with almost 100mS in between, so a considerable drop in the number of trigger events per second. --- End quote --- This is really weird. Even If I test with this same signal with SDS1104X-E and 200us/div and Sequence. I take oneshot sequence with 50MSa/s 140k length and when I look segments time stamps they have all 3ms (3000us) delta time. So 333.33 segments in second. Then I change it to 14k memory so 5MSa/s and look 1900 segment single sequence. Every single segment in sequence time stamp delta time is 3ms. So 333.33... segment/s Then with 1GSa/s 2.8M current mem lenght (one segment length)... still all 19 segments delta time is 3ms.... 333.33... segment/s --- End quote --- I think the explanation is good, with the sequence mode in run mode: [n sequences without gaps] 100ms processing interval [n sequences without gaps] 100ms processing interval... etc Interesting it was not in a circular mode (does not support it?) where the sequences capture around forever until stop is pressed. --- End quote --- Touchscreen Siglents indeed run Segmented capture in circular (forever as you say :-) mode if you start it in RUN mode. They will get one burst, and again and again with a slight pause in between. In order to get a single segmented burst you must press Single while in segmented mode.. |
| RBBVNL9:
--- Quote ---Honestly, I miss MSO5000 from Rigol in this. It would have been good to add it to what so far I see as very good and impartial research by Rudy. There have been quite a few discussions where there are no good info on real performance form real life MSO5000 users. I know there are quite a few of them out there, but maybe one or two contribute quality info. --- End quote --- Thanks for the nice words. I have been thinking of buying a Rigol MSO5000 for this comparison. Buts it's quite an investment if I do so only for these videos. Also, this comparison plus its videos have proven to be quite time-consuming (surprise, surprise). I have some work trips and a sabbatical leave coming up (was offered a guest professor position in Tokyo for that leave, and I am not going to drag any test equipment along ;-) With less time, adding an extra scope means it would delay the completion of the series. |
| Someone:
--- Quote from: 2N3055 on April 05, 2022, 12:16:37 pm --- --- Quote from: Someone on April 05, 2022, 10:37:16 am --- --- Quote from: 2N3055 on April 05, 2022, 08:39:35 am ---They also mention casually memory sizes, but 3000T has only 500k of sample memory when doing 4ch+ digital normal mode (4 buffer-2 per ch-1 with digital -0.5 Mpts for ping pong buffers. 1Mpts for Single mode). Scope with 100Mpts will be 100x slower everything else being equal. Scope with 500MPts will have soo much more work to do. --- End quote --- You can keep pulling up the same "small memory" argument, except it is common for memory to change in different acquisition settings across most scopes. Those characteristics are not avoided in the Keysight data sheets or left as only up-to/maximum/peak values, and made very clear in the manuals (contrasting to these other examples discussed here where it is not clear at all what is expected). But you make the false claim that there is a comparison with memory depth, the waveform capture measurements that competitors use keep the memory depth very short to inflate their numbers. Look at how much slower the competitors are despite choosing less memory! (when at same memory they are far behind). Scopes are not some computer system where the parts are put together in imbalanced ways by the end user, they are a finished product from the manufacturer who has decided on the system performance. Keysight make a real time scope with 100Mpts of memory, the EXR or MXR series, without a significant drop in waveform rate (keeping the system balanced). Where you keep falling over is trying to make out like all scopes work/function the same. The Keysight megazoom method has been to decouple display and waveform memory, they are two different paths that dont interact. Acquisition data is piped to the display plotter/memory (through decimation etc) separately to the waveform memory. As you say before, to make things fast they chose to put most of the emphasis/features on the decimated view (positive example: fast eye diagrams). Most other scopes draw the waveforms from the acquisition memory, and take measurements from the original data (Lecroy being the extreme example of that, positive example: higher resolution measurements). Completely different with advantages and disadvantages, for comparisons they are best listed as characteristics rather than put as a good/bad binary check box against specific/your preference. --- End quote --- I'm not sure what I wrote was false. Fact is that real time processing of 250 Mpts is harder than 4 MPts. --- End quote --- Right there, the opening statement. You keep bringing it back to memory depth. The main limitation of waveform update rate is the throughput of the plotter, the plotting rate of dots/vectors per second that is sometimes reported. By separating the acquisition memory and waveform plotter/display the Keysight megazoom IV continues fast realtime display with the full ADC rate out to 20GS/s in the higher end uses of it (with a proportionally slower waveform update rate from the higher sample rate). There is no compromising on waveform rate as the display window (and acquisition memory depth) gets larger, if they had more acquisition memory they wouldn't slow down at all, because the plotting keeps up at the highest ADC sample rates already. This is also why the intensity graded display doesn't show aliasing, it has data oversampled enough to produce the real time plotting. No trade off for memory depth vs update rate as they are largely independent. That is a fundamentally different way of operating compared to other scopes, that buffer the acquisition as a continuous (often decimated/filtered) waveform and then turn that data into a waveform for display. Needing the waveform kept in memory while it is being processed for display, blocking further acquisitions, producing longer blind times. Parallel vs Serial. Night vs Day. |
| Navigation |
| Message Index |
| Next page |
| Previous page |