Author Topic: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...  (Read 76051 times)

0 Members and 1 Guest are viewing this topic.

Offline marmad

  • Super Contributor
  • ***
  • Posts: 2979
  • Country: aq
    • DaysAlive
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #125 on: June 05, 2013, 12:20:46 am »
The thing that's a bit of a pity is that there are many nice features of this DSO - but either through some intentional or unintentional blunders, Instek has made it much more difficult to focus on those things.

Setting aside the mess-up with the published information of the Short record length, I don't think they should have pushed the 80k wfrm/s rate (with the reduced sample rate) as their banner rate - they should have just stated clearly that it can do ~50k wfrm/s  w/1k @ 2GSa/s - and it can get as high as 80k at lower sample rates - which is totally fine in the price class. The same with the intensity grading - why not just publish the actual specifications? It's not as good as the Rigol or Agilent, but it's better than the lower cost DSOs - and you get other features in exchange.

But the perception, intended or not, of being either a little misleading or reticent with some of the information - is, IMO, worse than just honestly stating from the start, "Here are all the specs: both the strong and weak points."
 

Offline marmad

  • Super Contributor
  • ***
  • Posts: 2979
  • Country: aq
    • DaysAlive
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #126 on: June 05, 2013, 12:28:52 am »
Moderately annoying since it was one of the things I was looking at when I was evaluating.

BTW, Greg, I suspect you might be able to turn that 'annoyance' into some kind of tangible 'settlement' from Instek (while pointing out that even their FW Help information was misleading).
 

Offline Hydrawerk

  • Super Contributor
  • ***
  • Posts: 2599
  • Country: 00
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #127 on: June 05, 2013, 01:05:02 am »
I am not a Rigol fan, but I think that there was no critical error in their manual or specifications of DS2000. There was clearly stated when the waveform update rate reaches 50000.
Amazing machines. https://www.youtube.com/user/denha (It is not me...)
 

Offline Yaksaredabomb

  • Regular Contributor
  • *
  • Posts: 124
  • Country: us
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #128 on: June 05, 2013, 06:30:20 am »
OK, this table is for 1kpoints per channel.
...
Now what is it like for Auto record length?
I'd be curious to see the auto record length chart, too.  Please post a link or the info itself if anyone has it.
 
Edit June 6th: My display name has recently changed from "jneumann" to "Yaksaredabomb"
« Last Edit: June 06, 2013, 12:37:51 pm by Yaksaredabomb »
My display name changed June 6th from "jneumann" to "Yaksaredabomb"
 

Offline marmad

  • Super Contributor
  • ***
  • Posts: 2979
  • Country: aq
    • DaysAlive
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #129 on: June 05, 2013, 10:32:11 am »
OK, this table is for 1kpoints per channel.
...
Now what is it like for Auto record length?
I'd be curious to see the auto record length chart, too.  Please post a link or the info itself if anyone has it.

No one does.
 

Offline tinhead

  • Super Contributor
  • ***
  • Posts: 1918
  • Country: 00
    • If you like my hacks, send me a donation
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #130 on: June 05, 2013, 10:56:06 am »
DS2000 must be better when it has two FPGAs.

the amount of FPGA is from pure technical point of view not really important, you can have one bigger FPGA doing
the job better that 3 other small FPGAs.

In case on Instek (who is using Altera FPGAs since 5+ yrs), they picked up high pin count Cyclone IV instead of
two smaller Spartan 6 like Rigol did. A high pin count Spartan 6 costs more money, Xilinx license costs as well
money (and not to forget IPs).

We would need to see design sources to judge about which DSO is better (from FPGA vs. design point of view).

For the use of DSO there are other things that important, but not pin count or size or amount of FPGA, etc.
I don't want to be human! I want to see gamma rays, I want to hear X-rays, and I want to smell dark matter ...
I want to reach out with something other than these prehensile paws and feel the solar wind of a supernova flowing over me.
 

Offline marmad

  • Super Contributor
  • ***
  • Posts: 2979
  • Country: aq
    • DaysAlive
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #131 on: June 05, 2013, 11:34:17 am »
We would need to see design sources to judge about which DSO is better (from FPGA vs. design point of view).

Well, I think we can draw some conclusions about the power/speed of each approach from the fact that the Instek uses a 1k sample length for it's high wfrm/s, the low number of levels of it's intensity grading, and from the evidence provided in Dave's video of the display updating freezing and stuttering. These are all tangible byproducts of their design.
 

Offline marmad

  • Super Contributor
  • ***
  • Posts: 2979
  • Country: aq
    • DaysAlive
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #132 on: June 05, 2013, 12:39:55 pm »
To clarify:

 

Offline grego

  • Frequent Contributor
  • **
  • Posts: 330
  • Country: us
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #133 on: June 05, 2013, 06:08:51 pm »
Moderately annoying since it was one of the things I was looking at when I was evaluating.

BTW, Greg, I suspect you might be able to turn that 'annoyance' into some kind of tangible 'settlement' from Instek (while pointing out that even their FW Help information was misleading).

I'm working on that now.
 

Offline tinhead

  • Super Contributor
  • ***
  • Posts: 1918
  • Country: 00
    • If you like my hacks, send me a donation
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #134 on: June 05, 2013, 08:10:16 pm »
To clarify:

ehm, these 656600000 wfms/s at 20ns/DIV on Rigol, where they coming from  :-//

yeah, i see it now, somehow i saw on right side (Instek) max 81 330 (where you wrote 81 330 000)
which was like "wfms/s" result and on left side trilions of milions :)
« Last Edit: June 05, 2013, 09:05:33 pm by tinhead »
I don't want to be human! I want to see gamma rays, I want to hear X-rays, and I want to smell dark matter ...
I want to reach out with something other than these prehensile paws and feel the solar wind of a supernova flowing over me.
 

Offline marmad

  • Super Contributor
  • ***
  • Posts: 2979
  • Country: aq
    • DaysAlive
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #135 on: June 05, 2013, 08:28:58 pm »
ehm, these 656600000 wfms/s at 20ns/DIV on Rigol, where they coming from

Wfrm/s? As it states, it's total samples captured per second (with the DSO's fastest sample length): @20ns/div, it's ~46,900 acquisitions * 14k = 656.6M samples per second.**

For the Rigol, it's the wfrm/s * 14k; for the Instek, it's wfrm/s * 1k.

** = Hypothetical (see post below)
« Last Edit: June 05, 2013, 10:23:02 pm by marmad »
 

Offline Hydrawerk

  • Super Contributor
  • ***
  • Posts: 2599
  • Country: 00
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #136 on: June 05, 2013, 08:54:22 pm »
EDIT: I am not sure about this.

At DSOX2002A you get up to 50 000 wfrms/s ×50kpoints = 2.5×10^9 =
2 500 000 000 total samples captured per second.  :-+
(Yes, the memory length is not great.)
« Last Edit: June 06, 2013, 02:29:51 pm by Hydrawerk »
Amazing machines. https://www.youtube.com/user/denha (It is not me...)
 

Offline marmad

  • Super Contributor
  • ***
  • Posts: 2979
  • Country: aq
    • DaysAlive
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #137 on: June 05, 2013, 09:32:27 pm »
At DSOX2002A you get up to 50 000 wfrms/s ×50kpoints = 2.5×10^9 =
2 500 000 000 total samples captured per second.

Well, we were discussing what 2 FPGAs vs. 1 FPGA might mean in terms of acquisition throughput - not really ASICs. But yes, the Agilent is the fastest, although I've never seen any published numbers yet for the wfrm/s when using the 1M (or is it 500k?) added memory.
 

Offline marmad

  • Super Contributor
  • ***
  • Posts: 2979
  • Country: aq
    • DaysAlive
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #138 on: June 05, 2013, 10:20:17 pm »
At DSOX2002A you get up to 50 000 wfrms/s ×50kpoints = 2.5×10^9 =
2 500 000 000 total samples captured per second.

Actually, after thinking about it more, it's impossible that the Agilent ever captures that much data (see below). Below 500ns on the Rigol or Agilent (or 50ns on the Instek), there is no way of really knowing exactly what the DSO is doing about the excess (offscreen) samples. I mean, if you STOP the DSO and zoom out, the full sample size is there - but what it does precisely while in RUN mode is unknown. My total samples chart is just hypothetical (and likely wrong) at those smaller time base settings because the DSOs quite possibly (in some cases, they absolutely must) limit the acquisition size until stopped.

For a good example, what would the Agilent 3000 X-Series do? At 10ns/div, it's supposed to do 1,030,000 wfrm/s. But it can't possibly be acquiring the fixed memory size (1M) - it's impossible because (1 / 4G[Sa/s]) * 1M[Pts] * 1.03M[wfrms] = 257.5 seconds. In fact, it must only be acquiring a tiny subset of the full memory (possibly just the amount of samples for the display) - and only capturing the full length when STOP is pushed.
« Last Edit: June 06, 2013, 12:01:03 am by marmad »
 

Offline Hydrawerk

  • Super Contributor
  • ***
  • Posts: 2599
  • Country: 00
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #139 on: June 05, 2013, 11:18:49 pm »
Well, we were discussing what 2 FPGAs vs. 1 FPGA might mean in terms of acquisition throughput - not really ASICs. But yes, the Agilent is the fastest, although I've never seen any published numbers yet for the wfrm/s when using the 1M (or is it 500k?) added memory.
I tried it. At Run/Stop mode DSOX2002A uses 500kpoints memory per channel (when one or all channels used.) The Trig Out frequency is as specified by Agilent in their PDF (up to 50 000 kHz). http://cp.literature.agilent.com/litweb/pdf/5989-7885EN.pdf
I measured it.
Well, DSOX2000 is as fast as possible... so you cannot set the memory length, interpolation or even dots/vectors.  :palm: No scope is perfect.
Amazing machines. https://www.youtube.com/user/denha (It is not me...)
 

Offline Hydrawerk

  • Super Contributor
  • ***
  • Posts: 2599
  • Country: 00
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #140 on: June 05, 2013, 11:34:08 pm »
At DSOX2002A you get up to 50 000 wfrms/s ×50kpoints = 2.5×10^9 =
2 500 000 000 total samples captured per second.

For a good example, what would the Agilent 3000 X-Series do? At 10ns/div, it's supposed to do 1,030,000 wfrm/s. But it can't possibly be acquiring the fixed memory size (1M) - it's impossible because (1 / 4G[Sa/s]) * 1M[Pts] * 1.03M[wfrms] = 257.5 seconds. In fact, it must only be acquiring a tiny subset of the full memory (possibly just the amount of samples for the display) - and only capturing the full length when STOP is pushed.
Yes, at 10ns/div timebase you will never see the whole record on the screen. (The picture is as Sagan would draw it.  :) :)  )

DSOX2002A triggers 50 000 times in a second, but the whole record update rate might be much lower.
« Last Edit: June 05, 2013, 11:37:57 pm by Hydrawerk »
Amazing machines. https://www.youtube.com/user/denha (It is not me...)
 

Offline marmad

  • Super Contributor
  • ***
  • Posts: 2979
  • Country: aq
    • DaysAlive
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #141 on: June 05, 2013, 11:43:44 pm »
I tried it. At Run/Stop mode DSOX2002A uses 500kpoints memory per channel (when one or all channels used.) The Trig Out frequency is as specified by Agilent in their PDF (up to 50 000 kHz). I measured it.

Yes, at 10ns/div timebase you will never see the whole record on the screen.

You miss my point. It has nothing to do with what you see on the screen - the DSO can't capture that much data in the given time. Your DSOX2002A can't capture 500k samples 50,000 times a second - it's impossible: at 2GSa/s, it takes the DSO 250us to fill 500k; if you multiply 250us * 54k you get 13.5 seconds.

Your DSO (and the Agilent 3000 X models) must be capturing MUCH less than 500k - and only filling the full memory on the FINAL capture - when stopped.

In fact, it's an impossibility that the 2000 X-Series is EVER capturing 50k at any time base setting below 200ns/div, because if the wfrm/s rate is > 50k the total acquire time > 1s. Again - impossible.
« Last Edit: June 05, 2013, 11:59:51 pm by marmad »
 

Offline Hydrawerk

  • Super Contributor
  • ***
  • Posts: 2599
  • Country: 00
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #142 on: June 05, 2013, 11:53:10 pm »
Well, only Agilent guys know how their scope really works. Thats it... On the other hand, one sales manager from czech distributor of Agilent knew a shit. He told me, that DSOX2000 has a real 230V switch.  He told me that he even replaced one of them. :-DD :-DD :-DD But we know that.
Even my scope consumes 5 Watts when turned off. I have a 230V wattmeter.
Amazing machines. https://www.youtube.com/user/denha (It is not me...)
 

Offline marmad

  • Super Contributor
  • ***
  • Posts: 2979
  • Country: aq
    • DaysAlive
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #143 on: June 05, 2013, 11:56:06 pm »
Well, only Agilent guys know how their scope really works. Thats it...

What I'm describing is math - the Agilent DSOs MUST shorten their acquire length in order to achieve the wfrm/s rates - it's a simple fact, easily proved with the numbers.
 

Offline Hydrawerk

  • Super Contributor
  • ***
  • Posts: 2599
  • Country: 00
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #144 on: June 06, 2013, 12:54:29 am »
OK
Amazing machines. https://www.youtube.com/user/denha (It is not me...)
 

Offline tinhead

  • Super Contributor
  • ***
  • Posts: 1918
  • Country: 00
    • If you like my hacks, send me a donation
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #145 on: June 06, 2013, 10:07:14 am »
For a good example, what would the Agilent 3000 X-Series do? At 10ns/div, it's supposed to do 1,030,000 wfrm/s. But it can't possibly be acquiring the fixed memory size (1M) - it's impossible because (1 / 4G[Sa/s]) * 1M[Pts] * 1.03M[wfrms] = 257.5 seconds. In fact, it must only be acquiring a tiny subset of the full memory (possibly just the amount of samples for the display) -

1 * samples * wfms/s / samples/s = second ? i need koffee ^^

At 4GS/s sample rate and 1Mpoint buffer depth the DSO need 250us to capture data into memory, then it need
to do the postprocessing. The total acquisition time is sum of both (i'm writing down all steps for those who get lost).
But total time is 1us (at 1M wfms/s), so the postprocessing need to be smaller than that, at least twice that small
(assuming Megazoom IV ASIC can do same data throughput during postprocessing). This give us 2000 data point per wfms.
Now let's calculate back, these 2000 points sampled with 4GS/s at 1M wfms/s give us then 0.5 which is less than
a second, which is what it should be because it need to be "per second".

The real data point value is probably equal to display resolution (visible area size, is it 600 dots?).

and only capturing the full length when STOP is pushed.

right, when writing once the full 1M buffer there is no blind time, so this part can work like that.

It can be as well as on TEK DPOs (not sure if on latest model as well), where the buffer is one time fully written
at begin of sampling (so when you do single shot it is once, and when you do RUN it is once ++++) and then
refreshed with what the hardware allows at max. (e.g. DPO3000 with 10k point, TDS700 with 500point)
 
When that's the case for Agilent, you would not even see (even with 4M memory) any gap between
"button pushed vs. data displayed", that' only 1ms to get full buffer once sampled, the refresh
data with 600-2000 dots per wfms is sufficient (as you can anyway see only that data and nothing else on screen).

So Rigol DS2000, with 21us total acquisition time from which 7us is used to capture the 14k points is not that bad,
that's (worst case) 7 time the buffer size of Agilent DSOX3000 (based on the findings/calculations above),
that's worst case third of DSOX 3000 wfms/s rate ^^ yeah, maybe.
I don't want to be human! I want to see gamma rays, I want to hear X-rays, and I want to smell dark matter ...
I want to reach out with something other than these prehensile paws and feel the solar wind of a supernova flowing over me.
 

Offline EEVblog

  • Administrator
  • *****
  • Posts: 37661
  • Country: au
    • EEVblog
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #146 on: June 06, 2013, 10:28:13 am »
Well, DSOX2000 is as fast as possible... so you cannot set the memory length, interpolation or even dots/vectors.  :palm: No scope is perfect.

The dots/vectors thing aside, isn't the Agilent essentially "perfect" in this case?
i.e. it automatically selects whatever memory it needs to get blindingly fast update rate in run mode, and then gives full memory in stop mode when you want to analyse it, right?
After all, you really only care about memory after you've stopped it and want to analyse it. Because features like Hires mode, and peak detect mode are things that work independent of memory depth, so essentially there is no benefit to having a deep memory while in run mode.
It obviously might operate different (and slow down?) in dual timebase zoom mode?
But if I'm not in dual timebase mode, and only have limited data on screen, then any extra deep memory being captured in run mode is essentially being wasted and is just slowing down the update rate, so it's pointless tradeoff. Update rate would win every time in normal run mode, there is no usage case were deep memory would be an advantage in run mode.
« Last Edit: June 06, 2013, 10:31:54 am by EEVblog »
 

Offline marmad

  • Super Contributor
  • ***
  • Posts: 2979
  • Country: aq
    • DaysAlive
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #147 on: June 06, 2013, 10:30:01 am »
1 * samples * wfms/s / samples/s = second ? i need koffee ^^

 ;D  Perhaps my formula is written a little strange, but it means (1 / 4,000,000,000) * 1,000,000 * 1,030,000  - which basically means it takes 257.5 seconds to fill a 1M buffer 1.03M times (without adding in postprocessing time).

Quote
The real data point value is probably equal to display resolution (visible area size, is it 600 dots?).

Yes, that would be my guess.

Quote
It can be as well as on TEK DPOs (not sure if on latest model as well), where the buffer is one time fully written
at begin of sampling (so when you do single shot it is once, and when you do RUN it is once ++++) and then
refreshed with what the hardware allows at max. (e.g. DPO3000 with 10k point, TDS700 with 500point)

This makes sense - with an added one when STOP is pushed for the last data:  once++.....+++once
 

Offline marmad

  • Super Contributor
  • ***
  • Posts: 2979
  • Country: aq
    • DaysAlive
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #148 on: June 06, 2013, 11:03:45 am »
The dots/vectors thing aside, isn't the Agilent essentially "perfect" in this case?
i.e. it automatically selects whatever memory it needs to get blindingly fast update rate in run mode, and then gives full memory in stop mode when you want to analyse it, right?

In some ways, I would say yes. But aside from the fact that (if an Agilent owner) I would just prefer to be able to turn this feature on and off (as with interpolation), some of the results of their method seem a bit questionable to me (although granted only under certain circumstances). The problem with 'automatic' features in complex technology is that if they're not extremely well-documented (in terms of the ramifications on all related sub-systems), the trade-offs are often not clear.

For example, what does the ASIC do exactly when I want my trigger position 5 divisions to the left of the screen edge? Or what does it do exactly with segments: does it capture them at the fastest speed possible while cutting down the sample length - or does it maintain the sample length while reducing it's update rate?
 

Offline jpb

  • Super Contributor
  • ***
  • Posts: 1771
  • Country: gb
Re: EEVblog #474 - GW Instek GDS-2000A Series Oscilloscope Unboxing & Fi...
« Reply #149 on: June 06, 2013, 12:52:04 pm »
The dots/vectors thing aside, isn't the Agilent essentially "perfect" in this case?
i.e. it automatically selects whatever memory it needs to get blindingly fast update rate in run mode, and then gives full memory in stop mode when you want to analyse it, right?

In some ways, I would say yes. But aside from the fact that (if an Agilent owner) I would just prefer to be able to turn this feature on and off (as with interpolation), some of the results of their method seem a bit questionable to me (although granted only under certain circumstances). The problem with 'automatic' features in complex technology is that if they're not extremely well-documented (in terms of the ramifications on all related sub-systems), the trade-offs are often not clear.

For example, what does the ASIC do exactly when I want my trigger position 5 divisions to the left of the screen edge? Or what does it do exactly with segments: does it capture them at the fastest speed possible while cutting down the sample length - or does it maintain the sample length while reducing it's update rate?

I think that they define this quite well in their manuals, particularly for the 5000/6000/7000 series :

Memory Depth/Record Length
[Run/Stop] versus [Single]
When the oscilloscope is running, the trigger processing and update rate are optimized over the memory depth.
Single
Single acquisitions always use the maximum memory available—at least twice as much memory as acquisitions captured in Run mode—and the oscilloscope stores at least twice as many samples. At slow sweep speeds, the oscilloscope operates at a higher sample rate when Single is used to capture an acquisition due to the increased memory available. To acquire data with the longest possible record length, press the [Single] key.
Running
When running, versus taking a single acquisition, the memory is divided in half. This allows the acquisition system to acquire one record while processing the previous acquisition, dramatically improving the number of waveforms per second processed by the oscilloscope. While running, maximizing the rate at which waveforms are drawn on the display provides the best representation of your input signal.

So the optimisation is really just between splitting the memory into two buffers for rapid waveform running mode or using it all as one buffer for single shot mode.

Elsewhere in the manual it explains that they have a pre-trigger buffer which is FIFO and which they fill before looking for the trigger event. Then they have the post trigger buffer. They always fill at the maximum sample rate and reduce it by decimation for slower sampling rates. For segmented memory they treat each segment in the same way (pre-trigger and post-trigger buffer). It lists the reset time after each segment as 1microsecond for the 3000X series so the update time will be 2MB at 2GS/s or 1microsec plus an extra microsec which presumably reduces it the wfm/s rate from around 1M to 500K.
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf