Products > Test Equipment
Where is the Keysight Megazoom V ASIC?
2N3055:
--- Quote from: nctnico on February 11, 2022, 10:19:38 pm ---
--- Quote from: 2N3055 on February 11, 2022, 09:08:58 pm ---
--- Quote from: nctnico on February 11, 2022, 11:31:18 am ---
--- Quote from: 2N3055 on February 11, 2022, 09:57:05 am ---If you keep analog front end from 3000T/4000, get rid of aMegazoom IV, put a single 16GS/s (a variant of?) converter from EXR (which would still get you 4GS/s in full channel mode, instead of 2,5GS/s now) and pipe that into Zynq, you would get immensely more powerful platform than now, and 10bit too.... You would literally solve all of the shortcomings of current platform. Pretty much what Rigol is doing, but with better performing components.
--- End quote ---
No. In the end the Zync platform is just a crutch to be able to do some functions in software but in the end the performance can't get nowhere near oscilloscopes that do functions like protocol decoding in hardware. Scopes from Rigol and Siglent can't get even close to the performance and comfort you get from Keysight or R&S where it comes to protocol decoding. In the end you pay for the engineering to do the hardware accellerated decoding. If you want to get close with a (mostly) software solution with low NRE then the only way out is a GPU based platform.
--- End quote ---
Zynq is just a sea of logic gates together with a fast cpus on a same chip with a fast interconnect.
You can load same IP into it that they hardcoded into Megazoom IV.
And I don't know what you think hardware decode means in Keysight. Their most advanced Infinium scopes are same software based decoding, like Lecroy, Siglent and the rest. Software based decoding has it's advantages, like for instance where you can enable decoding after you captured signal,
--- End quote ---
The Megazoom ASIC does decoding in hardware. On a PC based scope you have enough processing power to do decoding in software so needing hardware accelleration is less likely. Being able to change the decoding parameters after an acquisition is handy but if it comes at the cost of reduced memory depth, speed and other limitations, the benefit becomes very small quickly.
--- Quote ---And in Zynq you can pipe that to decoders real fast. Not to mention that in Zynq, you can actually use FPGA fabric to have hardware assisted decoding FSMs transparently..
--- End quote ---
But what you forget here is that FPGA development is very time consuming (= extremely expensive) and there are not many that master the skills to do FPGA development effectively so most of the lower end scope (B brand) manufacturers opt for a software solution at the cost of lower performance. It works but it is slow and needs all kinds of workarounds (like only decoding what is on screen or larger decimation steps) to get to an acceptable user experience. The only low-end DSO manufacturer I can think of that does decoding inside the FPGA is MicSig. Also keep in mind that the low cost Zyncs don't have that much gates to begin with so it is easy to run out of space. For example: GW Instek uses partial reconfiguration in their Zync based DSOs to load a different trigger engine to trigger on protocols. I guess they ran out of space to simply include all trigger engines and had to resort to a complicated solution.
--- End quote ---
Read my lips: on 3000T decoding is hardware/software hybrid. Like I explained. It is visible in segmented mode.
Cost of reduced memory depth in software decoding scope is such that they will all still have an order of magnitude more memory than Megazoom does.
FPGA development is expensive? So you're saying that Keysight ASIC development is cheaper and easier than FPGA development? News flash, their scopes are chock full of FPGA already, and EXR/MXR are running lots of internal glue and architecture in them. If anybody would be able to make fantastic scope on Zynq that would be them.
They went with ASICS for their higher end scopes because in a long run it will be cheaper for them to run the heavy lifting in ASICs (and because they have capability to design and make them) and for ADC and such where it provides competitive advantage i.e there is no off the shelf component and they are pushin state of the art.
FPGA that you would use in 3000T replacement would be a bit bigger than one used in Micsig or GW Instek 2000E series.
Pinkus:
--- Quote from: 2N3055 on February 11, 2022, 10:31:39 pm ---All the tradeoffs Keysight decided to do were done well, and with full understanding of it's target market and audience. It is a masterpiece of engineering: how to achieve most with as little as possible.
It is actual delicate balance of how much user gets while giving as little as possible.
Most users of these scopes are not bothered by low memory. It gets the job done.
Most users of these scopes are not bothered by low screen resolution. It gets the job done.
Most users of these scopes are not bothered by small screen resolution. It makes scope compact and easy to put on the desk. It gets the job done.
etc etc.
All while giving great user experience, instant response to user input. It is closest thing to CRT scope on market.
In its current form it gets the job done. No need to mess with it. There is no need for new design.
These scopes are Fluke 87 of scope industry.. Same rules apply.
--- End quote ---
It is exactly this. Having a DSOX4000a on my bench, I looked at the newer R&S / Rigols / Siglents in the past years. Their UI looks very impressive at first, but if you know a Keysight DSOX3000/4000 from practice, you quickly realize that you can still do your job faster and better with it, faster than with the said 10 years newer scopes with a often sluggish behaviour.
Sure more memory would be nice, but do I really need it? I have segmented memory and otherwise I just trigger on what I want to see. Period. I am convinced, in most cases where people are wanting more memory, they just do not know how to trigger. Very rarely do I need tons of data before and after. Do I need fast boot up time or 12 bit resolution and/or 0.5uV vertically? Yes, it would have been helpful sometimes, but I rather choose the reliability and speed of the Keysight as this is what I need in 99,99% of all time.
2N3055:
--- Quote from: wraper on February 11, 2022, 09:29:01 pm ---
--- Quote from: 2N3055 on February 11, 2022, 09:57:05 am ---If you keep analog front end from 3000T/4000, get rid of aMegazoom IV, put a single 16GS/s (a variant of?) converter from EXR (which would still get you 4GS/s in full channel mode, instead of 2,5GS/s now) and pipe that into Zynq, you would get immensely more powerful platform than now, and 10bit too.... You would literally solve all of the shortcomings of current platform. Pretty much what Rigol is doing, but with better performing components.
--- End quote ---
Somehow I didn't notice a better performance overall. Performance of Zinq based scopes fall flat beyond certain performance friendly settings and do not actually reach the performance of Megazoom to begin with. Megazoom based scopes perform well regardless. If Zinq is so perfect, why Rigol rolled out their own custom ASICs for their Ultravision II scopes (with Zinq also included)?
--- End quote ---
Like I said, "Keysight is fast!" is a U/I responsiveness thing. It is something that "button twiddlers" (people that turn buttons real fast to benchmark how fast scope will respond) notice immediately but rest of people not so much.
It is a credit to how well software in Keysight scopes is written.
People seem to praise 1000X series as being as fast to use as 3000T, glory to the Megazoom IV.
Despite 1000X having less than 1/5th of WFM/s per second.
Clever software writing AND same application processor as in 3000T makes user experience the same. Not Megazoom IV
And, as I know you know damn well, ASIC is short for Application Specific Integrated Circuit. It doesn't even have to have complicated function. You can make 5 transistor IC in SiGe process for your scope front end and it is ASIC.
Rigol has 3 ASIC chips. One is an amplifier for an active probe. Other one is amplifier/attenuator/integrated front end for a scope. Both analog chips. And third one is their ADC. That gets piped into Zynqs where they run all the scope logic...
They could as easily just buy off the shelf ADC and use that with minor adjustments.
And they did it this way because it doesn't lock them in into architecture that will get outdated in few years. And with expectation that FPGA market will keep expanding and FPGA chips they use will get even cheaper with time.
As a side note, current chip market must worry all those that did that bet. And those (like Keysight) that run their own ASICs must be smiling short term. Until they need to run another batch and then they are also in trouble.
nctnico:
--- Quote from: 2N3055 on February 11, 2022, 10:44:00 pm ---
--- Quote from: nctnico on February 11, 2022, 10:19:38 pm ---
--- Quote from: 2N3055 on February 11, 2022, 09:08:58 pm ---
--- Quote from: nctnico on February 11, 2022, 11:31:18 am ---
--- Quote from: 2N3055 on February 11, 2022, 09:57:05 am ---If you keep analog front end from 3000T/4000, get rid of aMegazoom IV, put a single 16GS/s (a variant of?) converter from EXR (which would still get you 4GS/s in full channel mode, instead of 2,5GS/s now) and pipe that into Zynq, you would get immensely more powerful platform than now, and 10bit too.... You would literally solve all of the shortcomings of current platform. Pretty much what Rigol is doing, but with better performing components.
--- End quote ---
No. In the end the Zync platform is just a crutch to be able to do some functions in software but in the end the performance can't get nowhere near oscilloscopes that do functions like protocol decoding in hardware. Scopes from Rigol and Siglent can't get even close to the performance and comfort you get from Keysight or R&S where it comes to protocol decoding. In the end you pay for the engineering to do the hardware accellerated decoding. If you want to get close with a (mostly) software solution with low NRE then the only way out is a GPU based platform.
--- End quote ---
Zynq is just a sea of logic gates together with a fast cpus on a same chip with a fast interconnect.
You can load same IP into it that they hardcoded into Megazoom IV.
And I don't know what you think hardware decode means in Keysight. Their most advanced Infinium scopes are same software based decoding, like Lecroy, Siglent and the rest. Software based decoding has it's advantages, like for instance where you can enable decoding after you captured signal,
--- End quote ---
The Megazoom ASIC does decoding in hardware. On a PC based scope you have enough processing power to do decoding in software so needing hardware accelleration is less likely. Being able to change the decoding parameters after an acquisition is handy but if it comes at the cost of reduced memory depth, speed and other limitations, the benefit becomes very small quickly.
--- Quote ---And in Zynq you can pipe that to decoders real fast. Not to mention that in Zynq, you can actually use FPGA fabric to have hardware assisted decoding FSMs transparently..
--- End quote ---
But what you forget here is that FPGA development is very time consuming (= extremely expensive) and there are not many that master the skills to do FPGA development effectively so most of the lower end scope (B brand) manufacturers opt for a software solution at the cost of lower performance. It works but it is slow and needs all kinds of workarounds (like only decoding what is on screen or larger decimation steps) to get to an acceptable user experience. The only low-end DSO manufacturer I can think of that does decoding inside the FPGA is MicSig. Also keep in mind that the low cost Zyncs don't have that much gates to begin with so it is easy to run out of space. For example: GW Instek uses partial reconfiguration in their Zync based DSOs to load a different trigger engine to trigger on protocols. I guess they ran out of space to simply include all trigger engines and had to resort to a complicated solution.
--- End quote ---
Read my lips: on 3000T decoding is hardware/software hybrid. Like I explained. It is visible in segmented mode.
Cost of reduced memory depth in software decoding scope is such that they will all still have an order of magnitude more memory than Megazoom does.
FPGA development is expensive? So you're saying that Keysight ASIC development is cheaper and easier than FPGA development? News flash, their scopes are chock full of FPGA already, and EXR/MXR are running lots of internal glue and architecture in them. If anybody would be able to make fantastic scope on Zynq that would be them.
They went with ASICS for their higher end scopes because in a long run it will be cheaper for them to run the heavy lifting in ASICs (and because they have capability to design and make them) and for ADC and such where it provides competitive advantage i.e there is no off the shelf component and they are pushin state of the art.
FPGA that you would use in 3000T replacement would be a bit bigger than one used in Micsig or GW Instek 2000E series.
--- End quote ---
And hence the higher price Keysight charges for their oscilloscopes. They do more engineering work so they have to charge a higher price for their product. B-brands are cutting corners to push engineering and hardware costs down.
And sure, at a high volume an ASIC is cheaper compared to bringing money to Xilinx. I never claimed that designing an ASIC is cheaper BTW. Putting functionality into an ASIC or FPGA is more expensive to develop compared to software.
2N3055:
--- Quote from: Pinkus on February 11, 2022, 10:56:54 pm ---
--- Quote from: 2N3055 on February 11, 2022, 10:31:39 pm ---All the tradeoffs Keysight decided to do were done well, and with full understanding of it's target market and audience. It is a masterpiece of engineering: how to achieve most with as little as possible.
It is actual delicate balance of how much user gets while giving as little as possible.
Most users of these scopes are not bothered by low memory. It gets the job done.
Most users of these scopes are not bothered by low screen resolution. It gets the job done.
Most users of these scopes are not bothered by small screen resolution. It makes scope compact and easy to put on the desk. It gets the job done.
etc etc.
All while giving great user experience, instant response to user input. It is closest thing to CRT scope on market.
In its current form it gets the job done. No need to mess with it. There is no need for new design.
These scopes are Fluke 87 of scope industry.. Same rules apply.
--- End quote ---
It is exactly this. Having a DSOX4000a on my bench, I looked at the newer R&S / Rigols / Siglents in the past years. Their UI looks very impressive at first, but if you know a Keysight DSOX3000/4000 from practice, you quickly realize that you can still do your job faster and better with it, faster than with the said 10 years newer scopes with a often sluggish behaviour.
Sure more memory would be nice, but do I really need it? I have segmented memory and otherwise I just trigger on what I want to see. Period. I am convinced, in most cases where people are wanting more memory, they just do not know how to trigger. Very rarely do I need tons of data before and after. Do I need fast boot up time or 12 bit resolution and/or 0.5uV vertically? Yes, it would have been helpful sometimes, but I rather choose the reliability and speed of the Keysight as this is what I need in 99,99% of all time.
--- End quote ---
I have on my desk both 3104T with full options and new 6000A Siglent. And Siglent is not slow. U/I is not as fast as Keysight at moments but slow it is not, certainly not sluggish. Certainly not to the point that it would slow me down.
These comparisons start to sound like benchmarking the PCs: even if you can reliably say some PC is 28% faster running some benchmark, what does it mean to me in real life when editing a document in Word?
But as a sidenote: I did try R&S 2000/3000/4000 and Rigols 5000/7000 and I agree with you. Actually, early versions of R&S 3000 series felt slower than DS7000 from Rigol in general. Both felt slow compared to 3000T. I don't know if they made it faster in newer versions of firmware.
But I could have worked with R&S 3000 (it is noticeable, but not that slow that it slows you down, mind you) but it lacked features that 3000T had.
More memory is very important. My biggest complaint on 3000T is how fast sampling rate drops as soon as I start going to slower timebases.
Also there are two school of though (let's say you're looking at rise time of some signal):
- You put shortest possible time on screen, and let scope gather statistics over thousands of triggers. Here Keysight fast retrigger comes in..
- Or (if your scope has enough memory) you capture very long sequence at full sample rate. And you run analysis on ALL the edges inside that buffer. Thousands of those. You get low triggers/seconds, but at every trigger you get statistics worth of 10000 of triggers on other scope.. In the end you get same numbers per second. But there is one benefit: inside that capture, there was NO blind time. So if you have one glitchy edge happening every 10 ms, in one 50 ms long capture you will have POI of 100%. And not only that, you will be able to see and measure if it is random or periodic.
Therefore : long memory -> useful.
As for reliabilty, that is superb on all mentioned scopes.
As for 12 bit, that is something that is very, very noticeable. If you do analog, you wouldn't want to go back to 8 bit. Trust me. It is addictive as hell.. That thin line that looks like it came fro best analog Tek that let's you see things you didn't see before...
I do agree wholeheartedly that triggering is the key to using the scope. But sometimes it is hard to find what to trigger on. It is easy when you're going for something specific. But, when examining signal for unknowns you might spend some time until you find something of interest and then you can formulate trigger conditions for that.
I will tell you, though, one simple, stupid little thing ( if you look at it superficially, that is ) that changed my life: little hysticons on measurements. That thing is just brilliant. It allows me to not only see signal on the screen, but to better understand what is going on, with one look on the screen.
I would not want to buy another scope that doesn't have that feature, ever... That is something that I miss very, very much on my 3000T.
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version