Author Topic: Yaugi 4: GPIB Ethernet PoE Adapter (Read 18960 times)

essele · « **on:** October 16, 2020, 05:40:29 pm »

Just about a year ago I first posted about a project I was working on to produce a USB to GPIB adapter along the lines of many others you could buy, but with a goal of being somewhat lower cost.

https://www.eevblog.com/forum/metrology/yet-another-usb-gpib-interface-(yaugi)/msg2744372/#msg2744372

Anyway ... I have continued developing this, but have switched away from USB in favour of ethernet, and specifically PoE (802.11af) ... it's not ready yet, but I've seen a few things over the last few days that's made me think I should talk about what I'm doing.

It's designed to be a single device controller, so not intended to run a long chain of GPIB cables, the idea being that they should be low cost enough that you just throw one in each device. I understand that might not be ideal from a triggering perspective, but given most new devices are ethernet based and not GPIB I'm sure it's not the end of the world.

I've redesigned the GPIB handshake and moved it from software bit twiddling into the programmable logic on the PSoC device, so it's much faster, it also has a hardware timeout mechanism and so far I've not managed to lock it up at all.

It uses a W5500 ethernet controller and I've implemented (from scratch) a DHCP client, a mDNS responder, a VXI11 rpc based client, and a web server to handle config and even firmware updates. (Now you know why it's been a year!) It also accepts TCP and UDP based queries, so can be used with things as simple as nc. It also seems to work ok with LabView and other NI tools although I've only done very rudimentary testing.

I'm using a Silvertel Ag9900 PoE module, which really feels like the lowest cost way of achieving PoE (and now are available from Mouser, the first few I had to get direct!) Still mulling whether to add a USB port for power in for non-PoE scenarios.

I've got a version with the GPIB connector directly on the PCB (see photo in 3d printed case) and I've just built another version with an IDC connector and ribbon cable, which I think is more practical for use behind kit in tight spaces etc.

So far it's largely working, I still need to implement the VXI11 abort and interrupt channels, and there is still a fair amount of work to do on the web browser and web based UI.

My best guess at this point (as I've just found some lower cost ethernet magjack options) is that at small scale this is likely to be $30 to $40 of board and components. My last order of 5 assembled IDC ones from JLCPCB was $40 + shipping/tax and you just need to add the PoE module, PSoC, MagJack, and ribbon cables/connectors.

Interested in thoughts/feedback. No commercial intent - doing it for my own use, and will be very happy to share once it's working reasonably well.

(And yes, there is a bodge wire in the photo ... my new MiniProg4 seems to be missing a pull-up on the UART so receives garbage during reset and programming, so this will be added on the next revision, but only needed for debug)

Lee.

ve7xen · « **Reply #1 on:** October 16, 2020, 07:55:55 pm »

I really like the concept. PoE power is a nice idea, just get a small PoE switch and power several of these on the bench from one power supply. Much prefer it to janky USB solutions for this, even if it is a bit more expensive.

I'd be concerned about availability of the PoE module, especially if you're presenting this as a project for DIYers and not a product for sale, but for your own personal use, who cares! It doesn't seem to be available from any of the normal suppliers; at a glance I can't even find anywhere to buy it online in singles. I assume it's also fairly expensive - curious where you got them from? There are lots of integrated PoE-PD / DC-DC converter ICs available on the market (e.g. NCP1083, Si3404, TPS23757), even if it's a somewhat larger footprint, I think I'd rather a discrete implementation. Probably cheaper too, if you're doing this in quantity.

essele · « **Reply #2 on:** October 17, 2020, 07:28:16 am »

Thanks ve7xen.

I'm fairly confident the PoE modules are reasonably available, when I first started looking at this I had to get them "direct" (was actually through a distributor, but I was able to buy 2 modules at a reasonable price even with shipping to the UK), but now they are available at Mouser...

https://www.mouser.co.uk/ProductDetail/silvertel/ag9905lp/?qs=OlC7AqGiEDm9ZoM6zJQCoQ==&countrycode=GB&currencycode=GBP

I did look at a discrete implementation as I think I would prefer it, but I couldn't find any way of doing it that was going to be lower cost, and definitely would have used more board space. I even looked at the IC's used in the silvertel module, but couldn't figure out what they are.

If anyone has a suggestion that's going to beat £6 then please let me know.

dietert1 · « **Reply #3 on:** October 19, 2020, 03:03:24 pm »

This is a great project, if you implement a nice subset of VXI and if it works reliable. About ten years ago i made some USB-GPIB adapters based on a XILINX CPLDs and FTDI USB-parallel chips. Those GPIB adapters have been running for months. Usually until the DLINK USB hub gets stuck, maybe when someone touches the cables.

As far as i remember POE is something in the making. It started as something simple to supply IP Phones, but nowadays it can supply a PC. Meanwhile power conversion got sophisticated rules and signalling, that can't be implemented without special chips. As far as the chip implements all the rules it should be/will be tolerant to future POE upgrades. The situation compares well to the different USB standards, if you consider those USB-C connectors Apple introduced.

Regards, Dieter

ve7xen · « **Reply #4 on:** October 21, 2020, 02:17:14 am »

Oh yeah, I didn't realize the part number is slightly different than the series number

. Looks to be available at Mouser at least, so that's not so bad.

About $9 USD for the module doesn't seem too bad in small quantity. In larger quantity I'd be confident the discounts are going to ramp much faster on the discrete components, especially the magnetics. And sometimes magnetics availability is an even bigger problem! In onesy-toosies, wouldn't be surprised if it's similar / more to do it yourself. Fair enough decision.

As a quick thought experiment, I took a look at the Si3404 EVB isolated flyback configuration. Assuming passives are free, adding up the transformer and active components I get about $8 USD at qty 1. Add a dollar or so for passives and it seems slightly more expensive at low quantity (looks like the folks at Silvertel did their homework lol). At high quantity sourcing most of the stuff from Asia, the gap widens pretty quickly, at only qty 100 it's down to about $5.50.

Technically isolation is required for this, since it connects to an external device (the test equipment), but in practice you could probably get away without it and it greatly simplifies / shrinks the design. If you did that, I think you'd be closer to $4 at qty 1 and $3 and qty 100.

Doing PD-powered gadgets (or just a breakout board) has been something on my mind for a while so it's interesting to evaluate the options, and good to know about those Ag modules.

essele · « **Reply #5 on:** October 21, 2020, 01:07:24 pm »

It's interesting you picked the Si3404 ... I did look at that as a potential option, however I always get stuck when it comes to the magnetics. I can't find a source for the TOEP13-0155S1, and whilst the datasheet is available it doesn't appear to be off-the-shelf. What source are you using?

There are some HanRun and TNK parts on LCSC that look "similar" (with a small 's') but it really not an area I'm familiar with so don't understand the implications of different choices. Maybe an interesting learning exercise to build a few different variants??

EDIT: Hmmm ... have just found some low-cost/low-power coilcraft transformers ... I fear a rabbit hole appearing!

dietert1 · « **Reply #6 on:** October 21, 2020, 03:02:32 pm »

Just noticed the Silvertel module AG9905M requires a minimum load current of 200 mA, otherwise it will emit audible noise and may cause POE failure. See 8.3 in the datasheet.

Regards, Dieter

essele · « **Reply #7 on:** October 21, 2020, 06:17:50 pm »

Hi Dieter,

Yes indeed, I was slightly concerned when I first saw that but the W5500 uses 128mA when it has a 100M link, and then you add the PSoC and a couple of led's and it's all getting pretty close anyway.

Also, in my testing I never heard a peep even with the W5500 in power down mode ... with various versions I'm on about module 6 from three different batches as well all with no issues and no audible noise .. so don't think it's an issue, at least in this application.

Edit: hadn't really registered the MPS "maintain power signature" comment properly before, but on investigation this is actually only 10mA, so not a problem here.

ve7xen · « **Reply #8 on:** October 21, 2020, 07:25:09 pm »

Quote from: essele on October 21, 2020, 01:07:24 pm

It's interesting you picked the Si3404 ... I did look at that as a potential option, however I always get stuck when it comes to the magnetics. I can't find a source for the TOEP13-0155S1, and whilst the datasheet is available it doesn't appear to be off-the-shelf. What source are you using?

There are some HanRun and TNK parts on LCSC that look "similar" (with a small 's') but it really not an area I'm familiar with so don't understand the implications of different choices. Maybe an interesting learning exercise to build a few different variants??

Good point. It seems like these flyback converters are generally designed with a very specific transformer in mind. At least they never seem to give you the applications information needed to select another one, and I'm not bona fide enough to be able to do it from first principles.

I had originally looked at the Si3402, but switched my post to Si3404 when I realized it's pretty similar but a slight bit cheaper. However, the flyback design in the 3404 EVB at least uses the AUX winding from the special transformer, while I was still using the CoilCraft FA2924/FA2805 from the Si3402 EVB. Use of the AUX winding is optional, and the parameters of the CoilCraft transformers seem similar enough that they could be a drop in replacement if you cut the AUX winding from the design, but not confident about it. Not really too sure what the main differntiators between the two are other than that the 3402 is in a slightly larger package and seems to be an older design with a simpler DC-DC converter; doesn't seem like enough to justify a new part, but I don't know the market.

Quote

EDIT: Hmmm ... have just found some low-cost/low-power coilcraft transformers ... I fear a rabbit hole appearing!

Oops

nightfire · « **Reply #9 on:** October 24, 2020, 10:07:11 pm »

Thats a nice way of getting new life into old gear- I also was thinking of getting some old hardware for usage in my home lab, but the most interesting stuff had only GPIB, not even a RS232 interface...
As I personally try to avoid having proprietary interface cards (and drivers) ready, some standardized interface like ethernet/scpi is most welcome for a moderate price.

mjkuwp · « **Reply #10 on:** October 27, 2020, 12:57:38 pm »

+100 for this idea.

I've recently been doing some automation with power supplies and electronic loads. I started out with USB or RS-232 interfaces but found it a real hassle to develop on Windows and then implement these in Linux. There are enough differences to make this a real pain.

I switched to all Ethernet by buying different equipment or the LXI firmware upgrades and discovered this solution is far better for ease of use. I've set my router to assign IP addresses based on the MAC address of the device and now I can seamlessly develop on Windows but then run the scripts from a Raspberry pi that is running on PoE.

i've never used GPIB but I have instruments that have this and no other interface. I would also consider buying an older instrument if I could upgrade it with such a device.

Quote from: essele on October 16, 2020, 05:40:29 pm

.......
So far it's largely working, I still need to implement the VXI11 abort and interrupt channels, and there is still a fair amount of work to do on the web browser and web based UI.
....

what is the web browser and web based UI for? without these features would it be useable by writing code? A person can use the python-vxi11 library to send IVI /scpi commands?

I guess there must be some amount of work to port GPIB things to IVI commands.

uski · « **Reply #11 on:** October 27, 2020, 11:51:14 pm »

Pretty cool ! I developed a commercial USB-GPIB interface years ago as part of my work, so it is cool to see some activity in this space still in 2020
I would love to build a few units for my own use too, especially if the firmware is open source.

essele · « **Reply #12 on:** October 28, 2020, 11:33:19 am »

Thanks for all the positive comments.

The web browser was the easiest way to enable config ... dhcp vs static, names and values for discovery etc, plus any GPIB configs (EOI management etc.) Also, it turned out to be the easiest way to enable to firmware update.

I've also just got a simple "send GPIB command" page working as well ... not really the main way of using it, but useful to test things.

It's already usable by writing code, the vxi11 interfaces work already, so I'm testing with python, but it also works with netcat (nc) and the lxi command line tools.

I have just discovered that the cheap plastic 24-way IDC Centronics connector doesn't actually fit a 3458A -- way too loose, although it fits a 34401A perfectly! Very strange ... but the central bit is quite a bit smaller. Need to rethink this.

enut11 · « **Reply #13 on:** November 16, 2020, 08:56:24 am »

Yes, the plastic 24 PIN is also loose on my HP3456A. The workaround for me was to open the area around the mounting screws. This allows the connector to sit further in. To hold it in place I use short screws with large flat washers.

essele · « **Reply #14 on:** November 16, 2020, 07:41:30 pm »

I’ve come up with a small adapter board that converts to a standard IDC header. That way I can use the higher quality (but still low cost) GPIB connectors.

Also ... I’m just waiting for the second version of my PoE DC/DC converter, if it works as expected then it’s between about $4 and $5 total cost, so a viable replacement for the Silvertel module (although will use slightly more board space.)

essele · « **Reply #15 on:** November 23, 2020, 12:27:34 pm »

Quick update on the DC/DC converter...

The test boards have just arrived, and with an extra 100uF cap seem to work perfectly (and the Yaugi already has a 100uF onboard) ... nice regulation and low ripple at 300mA, which is my target for now (also seems good at 500mA.)

At 300mA it's 75% efficient, which is exactly what I see with the Silvertel board (and 78% at 500mA, but all very coarsely measured for now.)

This is with a HanRun HR051067 transformer, which is $1.01 in single quantities! I've also got some Pulse and Bel transformers to try, but if the Hanrun works then it's the lowest cost.

There's a large amount of ripple and audible noise without the extra cap, which I don't get on the various other PoE modules I've had, so I will try to get to the bottom of that, likely a school boy error, plus I want to spend a bit of time looking at thermals. I've not included any thermal considerations on this test board so need to understand that before building this into the base Yaugi board.

All of the components are part of the JLCPCB SMT service other than the PWM chip and the transformer ... the 100V caps and the zeners cause quite a consumption of "extended components" which adds to the PCBA cost for small runs, so I need to look at options there too. Final board space permitting, is there any reason not to use two 50V in series?

Not tested the actual PoE signature bit yet, that's pretty simple, so hopefully will be ok.

Schematic attached ... this is all based on the FAN7601 application note and what I see on some of the existing modules. Very keen to hear feedback if anyone can help (although probably getting off topic for the metrology section, happy to start another project thread if better done that way.)

Lee.

MiDi · « **Reply #16 on:** November 23, 2020, 01:11:41 pm »

Quote from: essele on November 23, 2020, 12:27:34 pm

...
the 100V caps and the zeners cause quite a consumption of "extended components" which adds to the PCBA cost for small runs, so I need to look at options there too. Final board space permitting, is there any reason not to use two 50V in series?
...

If voltage accross those 100V caps exceed 40Vp, it is recommended to use balancing resistors if 2 series caps are used instead.
That adds to board space and usually to bom cost.
Usually it does not make sense to put 2 caps in series.

essele · « **Reply #17 on:** December 07, 2020, 06:53:33 pm »

Very quick update ...

New board with integrated DC/DC converter is back ... so far all seems ok.

The transformer is too close to the opto-coupler (wouldn't be a problem with the Bel-Fuse ones), but other than that initial power-up was good and everything appears to be working.

I'll look at thermals tomorrow, and actually test the GPIB functionality (although that's not changed) ... I've also included the required USB connections since the PSoC has the capability as per my previous designs, so this board should work with PoE or USB power, and be able to communicate over both USB and ethernet.

I'll probably switch to the 1.27mm pitch shrouded headers and 0.635mm ribbon cable for the main GPIB connectivity to make the cable a bit more manageable.

This is all manufacturable using JLCPCB's PCBA service with just a few things to add: the transformer, the PSoC, the PWM controller, the FRAM (I hate the PSoC eeprom emulation), ethernet jack, and the various headers.

essele · « **Reply #18 on:** January 27, 2021, 04:45:04 pm »

Ok ... final board spin looks good, there are few changes that I think work well...

1. The debug port is back to standard 10-way 1.27mm pitch shrouded header, but at right angles and brought out so it can be accessed from outside the case ... this is probably mostly for my benefit during development as I need to keep re-flashing them!
2. The serial port is now a right angled header 4-pin (+V, RX, TX, GND) also brought out from the case, this will double as a mechanism for resetting to defaults (not 100% sure yet, jumpering something) and also provides another method of powering.
3. I've switched to 0.625mm pitch ribbon and 1.27mm pitch IDC connectors. This is much more manageable and shaved a bit off the board. I've also reworked the adapter for the GPIB connector, which also seems to work ok. You can have the ribbon cable any size you like to suit your needs.
4. The USB port is tested and working for both power and data.

Main focus for a bit now is going to be the software side ... lots to do, including a complete re-write of the USB stack, but I'm going to focus on the ethernet/VXI first.

There are some "mapping" things I can do to make things easier with some devices ... for example I've added an option to intercept a VXI "*IDN?" request, so I can return whatever the device returns (i.e. no change), I can return a fixed string, or I can return the response to a different GPIB request ("END ONCE; ID?" for a 3458A for example.) ... [the lxi tools do a *IDN? when discovering!]

I have plenty of rom and fram available, so I could also easily have a mapping table so "non-standard" devices could be somewhat mapped to standard commands.

I'm also thinking about some "automated" measurement options, where the Yaugi configures a specific measurement and then periodically measures and sends the results to a grafana type solution.

But first ... a bit of work on the config side, and then I need to complete the VXI stack. I'm also looking into how I can do some rudimentary EMI testing, (a) because it interests me, (b) because I've just got a shiny new SSA3000X (inc VNA thanks to this board!), and (c) because I'm going to have these things next to sensitive equipment and I still don't trust my DC/DC converter.

MilkmanCDN · « **Reply #19 on:** February 19, 2021, 11:22:24 pm »

This is exactly what I've been looking for. Are you planning on making these available via eBay or some other outlet?

essele · « **Reply #20 on:** February 21, 2021, 05:05:16 pm »

Quote from: MilkmanCDN on February 19, 2021, 11:22:24 pm

This is exactly what I've been looking for. Are you planning on making these available via eBay or some other outlet?

Hi ... I wasn't actually planning on selling these, I'm mostly doing it for fun and the learning experience, and my own use of course, but if there's interest then I could certainly make a few available.

I've just submitted another rev of the board ... I wanted to tighten up some of the loops on the DC/DC converter which should hopefully improve noise and ripple, it's wasn't too bad, but there was an easy optimisation I could do. But more importantly this version includes a TQFP64 version of the PSoC so that I have the option of more memory and flash (32k/256k) .. that way there should be enough space to try to embed eLua and have an onboard scripting capability. I still need to look at "screen dumps" and I have a feeling that might need some more memory.

I've also switched to 3.3V rather than 5V for the main supply (there's still a 3.3V regulator for when you are USB powered), this gives me a greater choice of FRAM so I can also get a larger device in there as script storage.

I did prepare a board using the QFN64 but I'm yet to really get to grips with hand soldering these, so thought I'd stick with the more forgiving TQFP's for now.

psynautic · « **Reply #21 on:** February 21, 2021, 06:19:16 pm »

I would be interested in at very least, a board or two and the STL's for the enclosure.

MilkmanCDN · « **Reply #22 on:** February 21, 2021, 08:23:52 pm »

Hey there,

Drat. I'd be happy to pay you whatever you think it's worth for one of these to test. PM me if you're interested. I'm looking to control my older Espec SH-241 oven (GPIB only) via ethernet, and this seems like the cleanest solution.

H.O · « **Reply #23 on:** February 22, 2021, 05:51:12 am »

I'd certainly be interested would you make them available.

essele · « **Reply #24 on:** March 24, 2021, 03:56:11 pm »

Just a quick update ... the 3.3V change all seems to be fine, and the DC/DC converter seems stable and pretty good from a ripple and noise perspective -- although I do want to re-test with a Bel Fuse 3.3V transformer as I would hope it should be better than the HanRun 5V one for this use case.

The switch to the 64 pin PSoC also worked fine, this has enabled quite a few improvements given the extra memory, so I've been focusing on building some good capabilities into the firmware.

The biggest change is the incorporation of Lua as a scripting language, I had originally intended to use eLua, but it's quite a major codebase diversion from the main Lua distribution and is therefore stuck on a fairly old version. So I've created my own embedded Lua version which is focused on minimal code changes and really just adding supporting for read-only strings and tables ... it seems to work really well, just over 3K (compared to over 20K normally) of used memory on the PSoC with all the relevant modules loaded.

I also incorporated a Cortex-M0+ math library (used by Lua) which saved a huge amount of flash compared to the standard library functions.

There are also DNS lookups supported, so you will be able to send data to data logging services by name.

I've also added the PSoC "real time clock" ... which is a challenge since I don't have an accurate crystal, so I've also incorporated a simple form of NTP and some clock skewing to keep the time reasonably accurate for scheduled activities (i.e. run an ACAL at 1am every morning.)

In working with the web server code, to support running Lua scripts and getting streaming results, and also looking at how this is going to work with blocking GPIB read/writes, I've decided that my DIY cooperative scheduling approach is going to be a problem, so I'm now working on getting it all up and running on FreeRTOS ... this enables much simpler code paths for complex activities which should also make for much easier debugging.

I am definitely in the scope creep phase ... but I am enjoying it ... and I do think the move to FreeRTOS will ultimately make everything much more manageable.

Lee.


EEVblog Main Site	EEVblog on Youtube	EEVblog on Twitter	EEVblog on Facebook	EEVblog on Odysee

Author Topic: Yaugi 4: GPIB Ethernet PoE Adapter (Read 18960 times)

Share me