I looked into it a bit, and I must say; the long memory depth acknowledges the waveform/s.
Putting in large amounts of memory is ofcourse at an expense. The Uni-T has 16Mpts. Interesting.. is it just to beat Owon (they put 10M in so it's more than Rigol's and Hantek's), or?
First make it clear; waveforms/second is fully pronounced as the
display of waveforms/second. They can acquire the data and throw it away. But that does every scope I think. How can you otherwise get 50% pre and post trigger data? You can't predict a trigger, so it must continuously be sampling data. So basically, to count as a wf/s, it needs to be on the screen.
To display all waveforms on the screen, you either need a 50kHz, 150kHz or 1MHz refresh rate screen.. don't think they exist. And otherwise I'm sorry mr. scope
Other way: you need to 'remember' all the waveforms and process them onto 1 screen at once.
If you read along in the same application note, chapter 4, they talk about the memory technology involved.
If you refresh the screen at 50Hz, you need to store 3000 waveforms (150k/50) into memory. With 6000 samples, thats 18Mpts of memory required.
Obviously 50Hz is not right, if you take 60Hz, you need 15Mpts. So that sounds doable, because the scope must contain 16Mpts to get to it's specs anyway.
At one time you'd need to downsample all that data to 1 screen. R&S say that's where up to 90% of the scope is busy with (at 10ns/div). So apparently the Uni-T is quick in doing that.
Obviously, you can't use on-chip memory of a FPGA for this (maybe only to continuously sample and once triggered & sampled, sent it to the 'main' RAM banks).
If you look at the Rigol DS2000; they can acquire 50k waveforms/second. The Rigol does this with a record length of 14kpts. If they refresh the screen at 50Hz, you need to store 1000 waveforms. 1000*14kpts = 14MPts. What a coincidence, the maximum memory depth of the scope is 14MPts.
There is an upgrade to 56Mpts, but the scope can't process the data fast enough to make use of that. Otherwise you can use 56MPts to get 200k waveform/s (in theory).
If you take the DS4000: it can do 110k waveform/s and has 140Mpts of memory.
Once again, it can do 14kpts minimum. So you can fit 10000 waveforms in memory, which means you only have to refresh the screen 11x a second. I think the processing speed is limiting there; I would have thought you must be able to get 500k wf/s on that memory length.
I have no idea what Agilent is doing with their 2000X and 3000X series. The 2000X series has 100kpts of memory. What? How can you 'store' 1k waveforms for 50k wf/s? At a length of 1k, times 1k, it's still 1Mpts.
The 3000X has 2Mpts. It does 1M wf/s, so to make that work with a memory depth of 1k you need 1M*1k/50 = 20Mpts. Obviously I guessed the record length of only 1k here.
I can't find the 2000X or 3000X series minimum memory depth.. These calculations suggests it's only 100 samples per trigger. That's very very small, but probably sufficient to fill 1 screen wide. It's not single shot so you don't need the zooming capability.
100 samples * 1Mwf/s = 100MS/s. The scope can do 1Gs/s+, so that's fine.
So either Agilent must be doing something else (fast screen? they drive it in an unique manner) or just polished up some figures.. (yeah we can capture 1 million things a second, you can't see anything on it, but whatever)
R&S RTO series: 20Mpts maximum, 1M wf/s. That's what was required to make the 3000X series work with 1ks per trigger. So that sounds right.
First I was sceptic, but the R&S would actually confirm the Uni-T performance a bit.
Still though, if it's too good be true, it probably is.
They could be lying here and say they can pulse the trigger output 150k wf/s and can't figure out/too expensive how to make an ASIC/FPGA deal with the intense memory operations (R&S talked about 20 memory lanes each running at 500Mbps to store 8x10Gbps of ADC data). I would be very impressed if it did indeed work out. Keep us updated if you can get your grabs on a Uni-T + a screwdriver