NI Vector Signal Transceiver

[Marsupilami]

So no plane to buy the SUBRACK(For save some cost)
And how do i test the PCI-E to SFF-8643 card had the reset signal?Just probe the RST# pin when powwer on the PC?
And ...does those connector no need to solder?I see you didn't solder it in your photo.
Many Thank

The horizontal rails can be made from other stuff too. I made some card cages for storage some time ago. This used 10mm makerbeam extrusions.


Don't worry too much about the RST#. Plug it in and see if it works. If it doesn't that's something to look at.

The connectors not soldered are press fitted into the board. It is intentional. Mind you, it is not easy to do. My solution was something like this:

[KE5FX]

I noticed in the other thread that your original board had two backplane data connectors:



... as well as two trigger/clock/power connectors, which match the connectors on the back of the PXIe-5644R. 

However, your current board revision has two power connectors but only one data connector.  Is the other one not needed by the 5644R, or optional?

[Marsupilami]

On the 5644r nothing is connected on that block. Other than maybe ground. I don’t know why does it still have to be populated, maybe there’s a signal there that gets pulled down so that the backplane detects there’s something in that slot and enables power. Power is taken from the upper connector on both slots.

[KE5FX]

Got it... I see what you mean now, looks like there are only a few power pins connecting the slot-2 daughterboard to whatever is inside the adjacent shield. 

[zrq]

May you bother to discuss a bit more on the software side? (question posted on the other thread but it seems this one is more active).

If I understand right, if the description ini files in C:\ProgramData\PXISA\Chassis are created correctly (and the pxiesys.ini and possibly also the ChassisDescriptionFilePath registry), the adapter board can get recognized as a chassis and the VST board get enumerated (if it's already enumerated at UEFI booting and seen by Windows).
Or is it compulsory to write a stub DLL like NIPXIeSMu.dll to help enumeration and implement the SMBus operations?

[Marsupilami]

May you bother to discuss a bit more on the software side? (question posted on the other thread but it seems this one is more active).

If I understand right, if the description ini files in C:\ProgramData\PXISA\Chassis are created correctly (and the pxiesys.ini and possibly also the ChassisDescriptionFilePath registry), the adapter board can get recognized as a chassis and the VST board get enumerated (if it's already enumerated at UEFI booting and seen by Windows).
Or is it compulsory to write a stub DLL like NIPXIeSMu.dll to help enumeration and implement the SMBus operations?

I didn't do any of that. You plug in the board and it works. In MAX it looks a bit wonky since it won't sit under a chassis, as there isn't one, but it is working nevertheless just fine.
I tried an SMU some time ago that didn't do anything at all and I'm wondering what you just wrote might help in that case, but again for the VST I had to do zero extra software work other than installing drivers.

[KE5FX]

Make sure you post if you get it to work!

Some excellent results with your board:



Everything was pretty much plug and play, except for how difficult the 'plug' part is.  I was afraid I was going to break something by forcing or hammering the connector into place, and I don't have any mating hardware available to create a makeshift press-fit jig as you did.  So I ended just soldering the connector pins after poking them through the board as far as they would go:



Not ideal for a standard backplane installation since the connectors sit about 2mm above the board, but good enough for proof-of-concept.

For the PCIe interface, I decided to try this adapter for use with an M.2 slot on a LattePanda:



They are sold by a couple of different vendors under the same part number (S-M28643H-1-1).  It worked nicely with no additional resistors or other modifications.  I powered up the PXIe-5644R with a mini-ATX supply prior to booting the PC, and it showed up right away in NIMax:



Thanks again for doing all the homework on this!

[Marsupilami]

Some excellent results with your board:

Yaaay!
That’s a sick setup. Are you planning to package it in a small box or something?

The press-fit is pain, yes, but I’m glad you found a solution.

The datasheet of the VST has the max current draw, make sure your psu is fine with it.

Have fun and keep posting if you tweak it!

[KE5FX]

Yaaay!
That’s a sick setup. Are you planning to package it in a small box or something?

The press-fit is pain, yes, but I’m glad you found a solution.

The datasheet of the VST has the max current draw, make sure your psu is fine with it.

Have fun and keep posting if you tweak it!

I'm not concerned about the power supply but cooling is definitely a consideration.  Just sitting on the bench in the open air, the 5644R gets too hot to hold after a few minutes.  Forced air will be mandatory in any enclosed chassis, no question there.

It's tempting to mount a touchscreen display to a PXIe-1085 or similar crate, but there goes the "bargain" part of the project...   

[Marsupilami]

I’m cooling mine with a 140mm Noctua 12V PC fan. I bought the beefy industrial version but I keep it waaaay down with the controller. Probably one of the “silent” kinds would take care of it easily too.

[Kahooli]

I ordered one of these to join the gang. Seems like a great piece for relatively cheap.

[Marsupilami]

I ordered one of these to join the gang. Seems like a great piece for relatively cheap.

Ah, the infamous VST gang, yes. Welcome... I guess?

[zrq]

May I ask what version of the LabVIEW and NI-RF stack is used. Does them have to be an older version?

[Marsupilami]

May I ask what version of the LabVIEW and NI-RF stack is used. Does them have to be an older version?

I haven't tried the current RFSA/RFSG drivers, but whatever was latest a few months ago works well. LabVIEW is not required, but the latest should work fine.

I found one issue with RFSG, going back to 15.0 or something like that.
The soft front panel hangs when starting. The workaround I found was to create a simulated RFSG device in MAX. This way the SFP will show a selection dialog and will start up fine using the simulated device. After it's already running, it is safe to switch to the VST device from the menu.
No such problem with RFSA.

[KE5FX]

May I ask what version of the LabVIEW and NI-RF stack is used. Does them have to be an older version?

I downloaded these from NI for offline installation:

   5,046,984,704 ni-rfsg_23.8.0_offline.iso
   4,653,273,088 ni-rfsg_20.5.4_offline.iso
   4,751,687,680 ni-rfsa_20.5.4_offline.iso
   4,483,399,680 ni-rfsa_23.8.0_offline.iso

As I understand it, the 20.5.4 versions are the last ones that supported Windows 7.  Those are the ones I tested with, although I didn't try the signal generator app, only the spectrum analyzer. 

[zrq]

Thanks to Marsupilami and KE5FX, I got my setup up and running (after arguing for refund from ebay for one with PXIe connectors smashed and getting the second working one). The press fit connector is really a PITA, but it seems simply soldering them from the backside of the board works fine. One have to be careful not to accidentally fill the holes for not yet populated connectors with solder, otherwise cleaning the hole to let the other connector fit is another PITA without appropriate desoldering tooling.
I think I can win a prize if there were a competition for the sickest setup for the VST. Somehow the temperature raise is moderate despite there is only two low speed fans blowing towards the card.

[Marsupilami]

I think I can win a prize if there were a competition for the sickest setup for the VST.

Dude, what is this?
But, I'm happy you made it work. 

[zrq]

There is a toolkit by NI called RFmx, which can help taking advanced measurements from IQ waveforms to phase noise with these VSTs. It's not cheap at all though, but  .

[zrq]

I got a very simple GNURadio source working to some extent, there is no fancy zero-copy design but simply niRFSA_FetchIQSingleRecordComplexI16 and memcpy to GNURadio buffers. On my old i7-3770 test bed with an equally ancient display card, gr-fosphor runs perfectly stable up to 40 Msps and 45 Msps with seldom underflows. I'm using the newest NI suite under Windows 10, which modern GNURadio requires.



What's a bit curious is the NI's TDMS streaming example (https://www.vipm.io/package/national_instruments_lib_ni_streaming_host_example_for_the_ni_pxie_5644r/ , it works on LV2023 after installing LV2015 and the instrument design library) also caps at roughly 40MSps even to a RAM disk. It's supposed to work at full ADC speed of 120Msps per NI's description and the i7-3770 was a decent CPU in 2012 when the example code comes out. Is it possible that the DIY backplane is limiting the speed in some way? Did anyone manage to run the acquisition at a faster sampling rate?

I'd like to publish a write up on this and put the code somewhere once I get more time, feel free to ask me if there is anything I may be able to help before that. Also I still have a few backplane PCBs left. If someone in Europe want one at cost price, please let me know.

[Marsupilami]

I got a very simple GNURadio source working to some extent, there is no fancy zero-copy design but simply niRFSA_FetchIQSingleRecordComplexI16 and memcpy to GNURadio buffers.

This is pretty much where I got stuck too. I started looking at the custom buffer implementations, do do a proper no-double-copy version, but I didn't feel the documentation was solid enough for me to safely understand and be able to implement so I gave up.
I would be nice to have an efficient record and playback app though.

[zrq]

I got a very simple GNURadio source working to some extent, there is no fancy zero-copy design but simply niRFSA_FetchIQSingleRecordComplexI16 and memcpy to GNURadio buffers.

This is pretty much where I got stuck too. I started looking at the custom buffer implementations, do do a proper no-double-copy version, but I didn't feel the documentation was solid enough for me to safely understand and be able to implement so I gave up.
I would be nice to have an efficient record and playback app though.

I think I found the performance bottleneck, apparently the VST is only using 1x 2.5 GT/s PCIe lane, which in principle should run up to 4x. It probably the signal integrity problem with the homemade adapter board, maybe the soldered pressfit connector? and the longer than necessary (0.8m) SFF-8643 cable make it worse?

[KE5FX]

The link training happens at power-up time, IIRC.  Maybe it's not getting initialized properly.  Try powering up the VST before the PC is turned on, if you're not already doing that. 

The signal integrity of the connector should be fine.  You have to really screw something up badly to make a differential pair perform poorly at these rates.

[Marsupilami]

Mine is good.
PCI current link width = 4 in device manager

It is possible however if any of the TX or RX of the 3 other lanes fails the enumeration the link reverts to 1x.

[zrq]

Thanks for the help, I wonder if is there is anything can help troubleshooting this, a bit more information on where the link training failed would be nice. Well, a PCIe analyzer is certainly beyond my budget. Tried with another VST device and solder the R4 on the 10Gtek board (didn't do that as it appeared to work without it), but still only learned x1.

-> So I figured it out myself, it's simply because the SFF-8643 connector didn't mate properly with the board, so simply all differential pairs except one are not fully connected. 

[Marsupilami]

the SFF-8643 connector didn't mate properly

It's always something like that