EEVblog #800 - Siglent 1000X Oscilloscope Teardown

[EEVblog]

Teardown of the new Siglent 1000X Series Oscilloscope

[lukier]

Nice teardown.

This is strikingly similar to DS1000Z, just better frontend (but only two of them) and Blackfin instead of iMX ARM. Everything else seems the same (Hittite ADC, Analog VCO PLL, Flow SRAM + DDR3 for the main FPGA memory, some CPLD for display/glue logic).

[Lukas]

The H1K IC is a ADA4932-1 Low Power Differential ADC Driver.

[rs20]

Just to clarify, the "sequence" mode on the Siglent is not about some difference between waveforms-per-second stored in memory vs displayed on the screen (as might be suggested by the message at 7:26), nor is it useless because it doesn't improve detection rates of runt pulses, nor is it even designed for that purpose.

It should probably have been named "sprint" or "burst" mode to hint at its usefulness. Contrived example, if you have a MCU that blurts out 1000 SPI packets separated by 3us each, and you want to capture every single packet in segmented memory mode, you need a scope that can retrigger on 3us intervals. And since a typical scope has no sort of "sprint" or "burst" capability, that would imply a requirement for 333k waveforms per second -- so, forget about any Rigol on the market, even the DS6000 series. Any scope that can't capture 333k waveforms per second, or (less deceivingly) waveforms on a 3us spacing, will miss packets. But Siglent's scope, in sequence mode configured to 1000 packets, will catch every single packet and allow the entire transaction to be replayed.

In short, it's about capturing 400 waveforms in a single millisecond, which very few scopes can do. The 400,000 waveforms per second does not mean that the Siglent can capture 400,000 waveforms in a single second, it clearly cannot. Just like a human can run 100m in 10s but that doesn't imply a human can run 100km in 10,000s. I think that feature has been misunderstood, even mocked a little in these videos, especially with the "woah, look at that huuuge blind time" commentary. Nothing's happening during the blind time in the intended use case for this feature!!!

[PA0PBZ]

Oops? 



That could give a little bit of fireworks...

[rdl]

The chassis metalwork is an embarrassment. All crooked, bent and misaligned. At least it didn't have factory installed rust.

[EEVblog]

In short, it's about capturing 400 waveforms in a single millisecond, which very few scopes can do. The 400,000 waveforms per second does not mean that the Siglent can capture 400,000 waveforms in a single second, it clearly cannot. Just like a human can run 100m in 10s but that doesn't imply a human can run 100km in 10,000s. I think that feature has been misunderstood, even mocked a little in these videos, especially with the "woah, look at that huuuge blind time" commentary. Nothing's happening during the blind time in the intended use case for this feature!!!

Yes, I should have given a usage example like that. What I was trying to get across is that for most uses people should not confuse this mode/spec with the screen update rate and regular capture mode that everyone is used to to capture random events etc.
Yes it's very useful if you have a specific requirement that can take advantage of it.
Siglent don't even highlight this feature in their datasheet.

[Tothwolf]

At least four of the filter capacitors are Rubycon. Note the difference in the top vents. Can't tell for certain what the others are, but two of them could be Nichicon. The Rubycon parts they used aren't the high end ones either, about middle of the road. They might have gone with a Lelon brand on the main input filter so they could get those custom right-angle terminals instead of a standard 10mm snap-in.

[rs20]

Yes, I should have given a usage example like that. What I was trying to get across is that for most uses people should not confuse this mode/spec with the screen update rate and regular capture mode that everyone is used to to capture random events etc.
Yes it's very useful if you have a specific requirement that can take advantage of it.
Siglent don't even highlight this feature in their datasheet.

Cool -- you're absolutely right that's it's not exactly an everyday usecase! My example was very contrived. Also a difficult thing for the Siglent marketers to pitch without deceiving/disappointing people; which is perhaps why they didn't put it in the datasheet.

[dentaku]

What I nice retro Texas Timepiece you have on your bench

[silvas]

Just to clarify, the "sequence" mode on the Siglent is not about some difference between waveforms-per-second stored in memory vs displayed on the screen (as might be suggested by the message at 7:26), nor is it useless because it doesn't improve detection rates of runt pulses, nor is it even designed for that purpose.

It should probably have been named "sprint" or "burst" mode to hint at its usefulness. Contrived example, if you have a MCU that blurts out 1000 SPI packets separated by 3us each, and you want to capture every single packet in segmented memory mode, you need a scope that can retrigger on 3us intervals. And since a typical scope has no sort of "sprint" or "burst" capability, that would imply a requirement for 333k waveforms per second -- so, forget about any Rigol on the market, even the DS6000 series. Any scope that can't capture 333k waveforms per second, or (less deceivingly) waveforms on a 3us spacing, will miss packets. But Siglent's scope, in sequence mode configured to 1000 packets, will catch every single packet and allow the entire transaction to be replayed.

In short, it's about capturing 400 waveforms in a single millisecond, which very few scopes can do. The 400,000 waveforms per second does not mean that the Siglent can capture 400,000 waveforms in a single second, it clearly cannot. Just like a human can run 100m in 10s but that doesn't imply a human can run 100km in 10,000s. I think that feature has been misunderstood, even mocked a little in these videos, especially with the "woah, look at that huuuge blind time" commentary. Nothing's happening during the blind time in the intended use case for this feature!!!

It still doesn't seem to add up. At 1GSa/s the scope's 14Mpts of memory can straight-up capture 14ms. So why bother triggering on each packet?

Maybe this is making up for not having a way to inspect/navigate raw recorded data after the fact in a structured way (a la Tek's Wave Inspector), so instead it actually triggers on and captures multiple segments (and then you are able to navigate segment by segment)?

[rf-loop]

It still doesn't seem to add up. At 1GSa/s the scope's 14Mpts of memory can straight-up capture 14ms. So why bother triggering on each packet?

Perhaps this data packet is not best possible example, exept if there is some times long time gap between packets.

I have situation where equipment give out 10ns double pulses every 1 - 100ms and there rising edge time distance vary 40 - 100ns.   I need capture these so that I need  measure/watch  double pulses distance from each others and pulse shape anomalies also I'm somewhat interested about double pulses time from other next or previous double pulses.

Using 14M memory and 1GSa I may get one or up to 14 double pulses to memory.
Using seqmented acquire I can get 1000. Even if all doublepulses time distance is 100ms as in this example. Even if time period is 1 hour, no matter. In this case limiting factor is not single shot acquisition memory lenght if it is 1M or 100M

Btw, if need 1GSa/s and if this example best case 1ms period is between all pulses, there need 1000ms time what means  it need 1Giga points sampling memory if continuous capture.  If worst case in this exaple 100ms... it need 100Giga points sampling memory.

This kind of things are very typical use of segmented acquisition. Still, what more fast it is it can inside this time also capture if there exist some glitch what pass for trigger setting and trig after very short trigger rearm time.

[rs20]

It still doesn't seem to add up. At 1GSa/s the scope's 14Mpts of memory can straight-up capture 14ms. So why bother triggering on each packet?

A) The UI complexity required to give the user the ability to take one long waveform, and then segment it after the fact seems horrendous
B) Even then, there may be 10 bursts of 100 packets, or a bunch of packets separated by random intervals down to 3us but up to 1s or more. Genuine live retriggering is required for this case. But I concede things are getting increasingly contrived here...

[Godzil]

Am I the only one to have spotted this soldering oopsie ?


Ho, I was too quick, Dave spotted it later in the video

[AF6LJ]

I can see this is going to be a good video....
Shame I cannot watch it today, club newsletter to get finished first.

[rsjsouza]

Having several DSP options on the marketplace that supersede BlackFin DSPs with lower cost tools and higher performance (and longer roadmaps), I strongly suspect this device was used first by one of the big brands and somehow the code/schematics have become available and are being replicated to no end (the black box approach or sheep mentality - I can't tell which one is prevalent). Something similar as to what happens in audio as well.

Unfortunately that is one of the reasons why entry level scopes have a very limited FFT performance (and probably limits the math functionality as well).

[AF6LJ]

Good video Dave.
The sheet metal work is crap .

[vlad777]

EEVBlog #497 - Siglent SDG5000 Function Generator Teardown
time: 17:46

In this video same LATTICE MachXO is used. LCMXO640C
I only know because I checked the price and decided to buy LCMXO1200C.

But I still have it laying around because for god's sake I can't etch 0.5mm pitch.
I can solder 0.5mm , but can't etch with printed transparent foil technique.

[Bud]

Either these morons steal each other's design or the design was done by a same Yingineer. Look at this Siglent 1000X vs Rigol DS2072A, this is hilarious.
Similarities anyone ?

 



Unless it is the same company just pretending to be two.

[rf-loop]

Either these morons steal each other's design or the design was done by a same Yingineer. Look at this Siglent 1000X vs Rigol DS2072A, this is hilarious.
Similarities anyone ?

 



Unless it is the same company just pretending to be two.

Take two car, you can see front window, back window, doors, front lights, back lights,  seats, four wheels etc.  Today if you look more far it is difficult to tell name. They looks so much same...

When I look these two oscilloscopes internals I can see nearly only differencies. Nearly all is different, exept yes, both have  2 front ends but even when they are nearly in same position (you know where are car wheels) they are totally very different. Whole circuit design is really so different that where you see that engineers in other sides of world (perhaps north and South China) have never say hello to each others.  But yes, I can see both have two inputs, on board sub SMPS circuits, processing units, memories, relays, ADC, connectors, PCB and so on.   If you open SDS2000 and SDS1000X you perhaps feel that perhaps least some members in developmenst/design teams know each others.

[Godzil]

Yeah I don't see any notable similarity between the two scopes, Rigol use two FPGA +  one DSP, Siglent use only one FPGA + one DSP, and all the component are not placed in the same position and are not the same in fact.

[rs20]

Either these morons steal each other's design or the design was done by a same Yingineer. Look at this Siglent 1000X vs Rigol DS2072A, this is hilarious.
Similarities anyone ?

 

Who claimed they were identical? Who are you satirizing here?

[Bud]

I did not realize I need to chew the cheese for people, then put it in their mouth and then explain how they should swallow it. But let me fix my mistake.

In both scopes two low voltage switching power supplies located in the left top corner
both these power supplies are oriented west-east in both scopes
same RF shielding can size over the input stage
same type of relays in the input stage attenuators at same location inside the housing
same type of input buffer as in rigol DS1000 made from 4 transistors and the same topology: FET follower cascaded with a BJT follower, both loaded with current sinks.
same COSMO relays to switch AC/DC input stage coupling
the unpopulated FPGA footprints to the right of the RF shield are at exact same X-Y location
input stage LDO regulators in both scopes located to the righ to the RF shield
input stage LDO regulators use the same topology: 2x +5V, -5V
input stage LDO regulators are exactly the same part number, LP3878-ADJ (soic-8), TPS72301 (sot-23-5). How's that possible that two different companies and two different designers used exactly the same parts out of the full variety of parts available on the market?
same ADF4360-7 PLL for the ADC clock
battery holders are at almost the same location in mid-top of the board
same ADI Blackfin processor part
top three connectors - USB, LAN and BNC on the right are at exact same location (USB and LAN swapped)

And we were not shown the bottom of the Siglent PCB, surely the list would grow.

So unless the Chinese government dictates that power supplies must be placed at the top left corner of the PCB and requires their orientation must be west-east, and mandates all the other things listed above must be the same for all scopes, I do not see how this is not a copied design.
Who copied whom I do not care. They may as well have used same 3rd party source, a person or a company who did the designs for them or used a common design template pulled from Chinese iNet or passed around privately.

[rs20]

The external plugs need to be close to the edge, since the power supplies block everywhere else. The scope BNCs are located at the bottom left by convention across all oscilloscopes, which leaves the unpopulated logic analyser plug & FPGA down the bottom and middle-rightish, again purely due to proximity of where the LA plug is conventionally placed on the front panel. My god, they use 5V!? I take your point, one of them should have used 3.857V. This is only tentative, but ADF4360 looks like a fairly one-of-a-kind chip, I've come across that chip multiple times before (and not just in oscilloscopes). Honestly, maybe you have a point, but by half your list being dead reasonable design decisions rather than crazy aberrations, I simply fail to be convinced. It'd be more compelling if you could find identical faults, because it seems to me that if two designers both go to digikey to find the cheapest shield/voltage reg/PLL for the same basic specs (5V, 100MSps, whatever), and rank by price, they're both going to see the same items a the top of the list. I'm not a professional electronic engineer, but I suspect that a lot more effort goes into selecting components when thousands and thousands of units are being made than our hobbyist projects, and all that effort is going to converge towards the same solutions; especially with designs that aren't based on custom ASICs.

Also, who cares if they're cooperating? Seems like a win-win.

I did not realize I need to chew the cheese for people, then put it in their mouth and then explain how they should swallow it. But let me fix my mistake.

Cheers for the childish remarks; always strengthens an argument.

[vlad777]

I did not realize I need to chew the cheese for people, then put it in their mouth and then explain how they should swallow it. But let me fix my mistake.

You are totally correct , and this is not a childish remark.
Not even PC mobos are this alike.