EEVBlog #1119 - Designing a 1kV Isolated Oscilloscope

[NiHaoMike]

I'm surprised it's not Thunderbolt, which would handle the bandwidth easily.

[BrianHG]

The 1Gbit might be a bandwidth limitation of the Altera Aria LVDS channels as well as memory bandwidth.  They might be using only 2-4 pairs LVDS feeding a network PHY.  This wouldn't have been an issue with a Stratix, but, I suspect putting in an 4-8k$ FPGA would double the price of the scope.

[nctnico]

The 1Gbit might be a bandwidth limitation of the Altera Aria LCDS channels as well as memory bandwidth.  They might be using only 2-4 pairs LVDS feeding a network PHY.  This wouldn't have been an issue with a Stratix, but, I suspect putting in an 4-8k$ FPGA would double the price of the scope.

Wouldn't the FPGA have an internal gigabit transceiver which can drive the SFP port (network) directly? A Xilinx Spartan6 which costs a few tens of dollars has no problem with that.

[BrianHG]

The 1Gbit might be a bandwidth limitation of the Altera Aria LCDS channels as well as memory bandwidth.  They might be using only 2-4 pairs LVDS feeding a network PHY.  This wouldn't have been an issue with a Stratix, but, I suspect putting in an 4-8k$ FPGA would double the price of the scope.

Wouldn't the FPGA have an internal gigabit transceiver which can drive the SFP port (network) directly? A Xilinx Spartan6 which costs a few tens of dollars has no problem with that.

Yes, and they are using a 2 pair gigabit transceivers for the gigabit Ethernet/optical networking.
They didn't want to go further than that.

They are using the ' 5AGXBA1D4F31C5N ', a 260$ fpga in lots of 27.
It has 6.5536 Gbps transceivers, not the 10.3125 Gbps or 12.5 Gbps required for a 10Gbit lan connection.
The DDR3 ram controller works at 533Mhz or 667Mhz depending on configuration and wiring.
They also started development around 3 years ago, so, back then, there were no Aria FPGA in the series which had the 10Gbps and were also affordable.  Like it or not, we are stuck with 1Gbps lan and would need to wait for the next version of the scope for the 10Gbps lan support.

I'm not sure but the speed and width of the ram and simultaneous sampling with network transfer may also be a factor.  I only saw 4 ram chips on the PCB, however, pausing the sampling during the network transfer would make things faster.  They would need 40 gigabit per second transfer rate for real time transfer of all 4 channels at full sampling speed.

Since they have full firmware update-ability, they wont be using Altera's cheap quick-asic mask version.

[schmitt trigger]

This chap Bart comes across not only as a very talented individual, but also as a down to earth, almost folksy individual.
His enthusiasm is contagious.
In my book, a great man.

Don’t have a need for such high performance instruments in the foreseeable future, but I hope all of his efforts are successful. In economic terms.

[GigaJoe]

Transfer 40Gb data in overall a humangous job calculation,  using a  standard IP stack ... and entirely inefficient; Ideally something like infiniband   interface more suitable , so it would be a separate board with ram storage then PC will have access to this storage and manipulate with a data.    That alone would be approx $1500 to $4500 - depend on ram capacity ... BUT possible to connect a multiple scopes .... is any one need 16 channels ?,   consider of synchronization can be done over infinband as well

[LapTop006]

These fibre links remind me of something i read about a couple of years ago. Intel developed something very similar to what is used here. Maybe it is the same thing?

It was supposed to be used for high speed internal busses in servers and workstations. After those one or two articles i read, i never heard anything about it again. Seems to have found a use 

I guess you're thinking of what became Intel Omni-Path which is only really interesting for the supercomputing folk, haven't heard from my friends in that space what they think.

[ali_asadzadeh]

Dave, that's selling
I wonder what part numbers are used for the fiber isolation?

[mjs]

The fiber link looks like muRata MMC3. There's not too much data available about it, though. Its main market is apparently in active optical cables.

https://www.murata.com/about/newsroom/techmag/metamorphosis20/productsmarket/module

[nctnico]

Transfer 40Gb data in overall a humangous job calculation,  using a  standard IP stack ... and entirely inefficient; Ideally something like infiniband   interface more suitable , so it would be a separate board with ram storage then PC will have access to this storage and manipulate with a data.    That alone would be approx $1500 to $4500 - depend on ram capacity ... BUT possible to connect a multiple scopes .... is any one need 16 channels ?,   consider of synchronization can be done over infinband as well

Products like that already exist. I've worked on such a project/product for one of my customers: a multi-channel distributed data acquisition system for use in high voltage testing labs (this needed slightly more than 1kV isolation...). Synchronisation is done using White Rabbit sub-nanosecond time synchronisation developed by CERN (which they use to synchronise data acquisition in the LHC). It is pretty cool to sample the same signal using two seperate acquisition units and have the resulting waveforms overlap flawlessly.

[rrinker]

This chap Bart comes across not only as a very talented individual, but also as a down to earth, almost folksy individual.
His enthusiasm is contagious.
In my book, a great man.

Don’t have a need for such high performance instruments in the foreseeable future, but I hope all of his efforts are successful. In economic terms.

 Totally agree. That was a great interview, and a very clever bit of design. Too bad I have no use for something like that, because he sure as heck sold me.

[lordvader88]

I u took a conductor, and energized it with +500V, then -500V, at whatever freq like in the motor, with the 10ns rise time....

why can't they just build a probe and probe the EM waves just above it in free space ?

[T3sl4co1l]

An e-field probe works just fine for high frequencies... however, can you calibrate it accurately, given that the probe can be any distance above any charged shape?

And it doesn't do anything for differential sensing, of course.

Tim

[lordvader88]

wouldn't "they" do a bunch of number crunching on whatever some probe detects, in order to isolate the signal they are after ? , like in any number of branches of science

I have no idea what it would cost anyways, the scope in the video is like +10,000