RaspberryPi + SPI driver #42
Replies: 22 comments 44 replies
-
|
Funny you should mention this. I don't know if you are aware but Linuxcnc-RIO does what you are suggesting, https://github.com/multigcs/LinuxCNC-RIO Concerning the TSX0108s....I'd be hestiant to use them unless they are not driving a signal via a cable. From the Data sheets the capacitance of a cable can have an affect on the signal. I would tend to go with something like SN74CB3T3245, which is basically an 8 bit version of what mesa uses of there boards, they use a big 56 pin device that has 2 independent 12 bit ports. |
Beta Was this translation helpful? Give feedback.
-
|
Thank you for pointing out the The big IC on the 7C81 board are |
Beta Was this translation helpful? Give feedback.
-
|
What Pete from Mesa does is supply the PI5C16211 with 4.2v on the vcc, this combined with the voltage drop across the switch (it was mentioned as 1v or so) drops it to a safe voltage for the Spartan chips. I just can't find the page which explains it. I'm pretty sure if you just followed the schemtics for the 7i81c, specifically page 7c81bu26.pdf & 7c81io26.pdf you'll be right. |
Beta Was this translation helpful? Give feedback.
-
|
I already saw something which was something like a voltage divider and and a follower in schematics. On I think I prefer the |
Beta Was this translation helpful? Give feedback.
-
|
Agree with the 74CB3t3245 This page is what I was thinking about with the 1v drop (Under voltage translation) |
Beta Was this translation helpful? Give feedback.
-
|
Actually that may not have been the best advice. |
Beta Was this translation helpful? Give feedback.
-
|
The trick with For the design of the HAT, I think I would prefer:
|
Beta Was this translation helpful? Give feedback.
-
|
I just received a batch of the bypass boards. The eyes aint so good as they were 20 years ago so I have to do the soldering in small batches, tho I'm thinking of using the mill with a pickup tool in the chuck to help align the boards hahaha. One thing that could be an option is to have one group of connectors, say J1 to J8 as input (48 lines) and leave the other side as the outputs which are 5v anyways. Then just check that outputs are only J9 to J16 and inputs are on J1 to J8 prior to building. Not as flexible but somewhat simpler. Only 6 bypass boards to solder. |
Beta Was this translation helpful? Give feedback.
-
|
Sneak preview: Capabilities:
Either the RPi is powered by USB3 and the 5A-75B is powered from the GPIO header, or the 5A-75B is powered with 5V and the RPi is recieving power from the card. The card and RPi should never be powered simultaneously, as this will result in damage. |
Beta Was this translation helpful? Give feedback.
-
|
The board layout has finished 🎆
The final capabilities are:
Also started on making a manual on how to hook this up. ❓ Question, before ordering the board, should I add these components or not:
|
Beta Was this translation helpful? Give feedback.
-
|
Looks great Last 2 options I'd say yes. |
Beta Was this translation helpful? Give feedback.
-
|
Those two options are also on the board now! I'm not sure when to order a proto-type, either before or after the SPI-module has been developed... |
Beta Was this translation helpful? Give feedback.
-
|
I have a question about the E-STOP. |
Beta Was this translation helpful? Give feedback.
-
|
I agree with you for all lines which go off the cards. It is always good to use one of the other I/O pins as a connection to the physical emergency stop circuit. The ENA/E_STOP would be useful for example to convey a driver error. |
Beta Was this translation helpful? Give feedback.
-
|
Work on the driver has been started in #48 . The boilerplate for the code is ready and it opens a connection to the SPI on the RPi. |
Beta Was this translation helpful? Give feedback.
-
|
Also had to change the configuration JSON, to distinguish between ...
"connection": {
"connection_type": "etherbone",
"tx_delay": 0,
"ip_address": "10.0.0.10",
"mac_address": "0x10e2d5000000"
},
...Above will replace the current ...
"connection": {
"connection_type": "spi",
"MOSI": "j1:0",
"MISO": "j1:1",
"CLOCK": "j1:2",
"CS": "j1:3",
},
... |
Beta Was this translation helpful? Give feedback.
-
|
i would prefer boards like the 'Colorlight i5', you don't have to re-solder anything, but a nice board ! where to get the bridges for the output drivers ? I would also like to build the board when it is ready and adapt it to the LinuxCNC-RIO project. |
Beta Was this translation helpful? Give feedback.
-
|
Can you provide a link to the Rio project? With the bridges, do you mean the ICs on the HAT or the flexboards to replace the buffers on the HAT. For the BOM for the HAT I will supply links to both EU shop and Aliexpress. |
Beta Was this translation helpful? Give feedback.
-
|
Whilst they seem like a good idea the bridges are a bit of a pain to work with. The SN74CBT3245CPWR are a good option, as long as you supply them with ~4.2v you get 3.3v on the FPGA side. The colorlight boards run off 4.2v ok, I had a mate over and did some testing on one of his boards. I've been playing around with a Spartan 6 dev board from aliexpress using mesa 7i90 firmware and am using SN74CBT3245CPWR with the non FPGA pullen up to 5v olts via 3.3v resistors. Iniitial testing has also shown that: Aliexpress FPGA board > SN74CBT3245CPWR > Aliepxpress RS422 to TTL Adapter wokrs for smart serial. (7i90 from MESA is a cheaper option) Next thing is to mod a lisn RV901T to run MEas firmware. Swap out 8 74hct245's & replace th3 75 ohm resistor networks with zero ohm equivs. This is more to see if it can be done. |
Beta Was this translation helpful? Give feedback.
-
|
The buffers SN74CBT3245CPWR are part of the HAT. You don't have to put them on the FPGA, which makes powering the RPi from the 5A-75B much easier. Personally I don't like the bridges, as they are quiet expensive to make. I prefer therefore wire bridges. I like that there are many options going on. |
Beta Was this translation helpful? Give feedback.
-
|
They weren't all that expensive (about AUD$9 for 9 cards of 4 (36 pieces) delivered), but they are fiddle to trim, you need to get basically in the middle of the via, and there's not a lot to actually solder to. If you have the eyes of a 20 - 30 yro they aren't a bad option. But if you're the wrong side of 50, not so much fun. |
Beta Was this translation helpful? Give feedback.
-
|
I'm going towards the 'wrong' side. Think I'll stick with the wires, getting trained in that one. About the Tang Nano, those are also nice boards with a good price-point. If the board is supported by LiteX, it can run LitexCNC as well. Gets especially interesting when the SPI driver is finished, as it does not have Ethernet. The i5 has the advantage of not having to replace the buffers, but it is a whole lot more expensive ($40 on AliExpress). |
Beta Was this translation helpful? Give feedback.
-
|
thanks for you great work |
Beta Was this translation helpful? Give feedback.
-
|
You can also connect the JTAG pins to the raspberry GPIO's for FPGA bitfile programming: |
Beta Was this translation helpful? Give feedback.
-
|
The JTAG pins are connected, however I use OpenOCD for programming. Went wrong today, burnt out a card unfortunately. The RPi survived luckily |
Beta Was this translation helpful? Give feedback.
-
|
New update:
Todo before I send teh board for first proto-type:
|
Beta Was this translation helpful? Give feedback.
-
|
Done! Going to send the order tomorrow 😄 ...
|
Beta Was this translation helpful? Give feedback.
-
|
Bummer, got two set-backs:
|
Beta Was this translation helpful? Give feedback.
-
|
Tried today with a FT232R to connect to the card and it worked like a charm. This means the card is not broken, so it must be the Pi? Looking at the output of OpenOCD I noticed that the received data is consistent (every time the same), but all bits seems to be shifted one position. So also the Pi did survive, it is mere coincidence that the problems arose when changing the jumper cables. My conclusion is that the jumper cables are causing this problem with timing. I tried many settings, but cannot find a remedy for this behavior. As an insurance I will add a JTAG header on the HAT, so the JTAG can be used at all times, even when the problems with the RPi as JTAG cannot be resolved. EDIT: |
Beta Was this translation helpful? Give feedback.
-
|
First response from the card! Was very happy to see the magic number ( The maximum speed is 8 MHz (over jumper wires). Higher frequencies give bogus values. Possibly the FPGA cannot keep up, because the SPI_Bone is running on stock 35 MHz. This will be increased to 125 MHz as soon as I figure out setting the timing constraints on a clock domain. Another thing I'm working on is reading / writing multiple addresses at once. This will significantly reduce data overhead and allow for twice as fast data-transfers. |
Beta Was this translation helpful? Give feedback.
-
|
Very nice. Do you know this discord channel? Think there are some good people who can help with the clock domains |
Beta Was this translation helpful? Give feedback.
-
|
Thanks, I didn't know that Discord server yet. Will check it out. This morning I managed to reset the board using SPI: The first byte states the command and size:
Bytes [1:5] are the address to read or write, followed with any data which has to be written. The response is 0x01 to indicate a successful write, or 0x01 followed by the data which is read. The required buffer size for reading and writing data is in any case 5 + read/write data long. I will implement an extra buffer of two bytes in case a byte boundary is missed by the FPGA. Next step: the driver for LinuxCNC |
Beta Was this translation helpful? Give feedback.
-
|
Why did your friend gave the board back? |
Beta Was this translation helpful? Give feedback.
-
|
He had a connection problem. We couldn't find the error...I tested it at home. Had also one time when I could not ping the card. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks, always looking for possibilities for improvement! The SPI version of LitexCNC does register now 😄 : Now I have to implement the read and write functions for the data, something for tomorrow. |
Beta Was this translation helpful? Give feedback.
-
|
Also implemented the read and write function, feel free to test the branch #48 . What I noticed is that the communication does work, however the used library |
Beta Was this translation helpful? Give feedback.
-
|
Is your Pi user in the gpio group? |
Beta Was this translation helpful? Give feedback.
-
|
Yes, the user is in the Didn't try to use the deamon, as that might add another layer (just like When writing SPI data to |
Beta Was this translation helpful? Give feedback.
-
|
Yes, it is working The trick was to The overhead of sending data to the FPGA is now 0.3 ms (based on a safe 1 MHz data transfer). Feel free to try it out! I modified LiteX-CNC a bit, so the driver is nicely unloaded upon exit. Otherwise the SPI device could remain busy and prevent you from starting up a second time. I'm happy 😄 ... |
Beta Was this translation helpful? Give feedback.
-
|
Wrote a small speedtest tool, which also shows the timing. The whole component is crammed into |
Beta Was this translation helpful? Give feedback.
-
|
And a small improvement which shows the average time of a transaction and the maximum within the 10,000 loops: The maximum time is basically 100 us or 0.1 ms. This is faster than the ethernet connection. I tried to run the setup for 1 hour and did not get any latency excursion. With ethernet I unfortunately did.... Running the test multiple times shows an increase in maximum time, so most likely something is not cleaned up properly (is to be expected, as the |
Beta Was this translation helpful? Give feedback.
-
|
Would you say this is more data per time than over ethernet? |
Beta Was this translation helpful? Give feedback.
-
|
Not necessarily more data, as the dataframes which can be send within each loop are limited to 256 bytes. The SPI will have less jitter and hopefully less latency excursions then Ethernet. When I ping the card, the round-trip takes anywhere from 0,1 to 0,3 ms. Which means that in each loop your waiting on a response somewhere between 0,3 and 0,9 ms. Normally you run at 1 ms period, showing that there is not so much room for this waiting. Keeping the LAN port is not really key, I use WiFi to communicate with the Pi if I have to. The RPi4 has nice built-in WiFi. The compact Pi-colorlight-Pack is a nice result if everything is working! |
Beta Was this translation helpful? Give feedback.
-
|
Nice work! |
Beta Was this translation helpful? Give feedback.
-
|
The SPI is hardwired to J5 on the HAT. This connector is the closest to the RPi header, so it made tracing easy. It uses the following pin-config: {
"connection_type": "spi",
"mosi": "j5:14",
"miso": "j5:12",
"clk": "j5:15",
"cs_n": "j5:11"
}These are all shared pins on the 5A-75B, which means that you can also use J1 (if you have swapped out the buffers. For the HAT I have to change a little, due to the SD-card of the RPi conflicting with the connectors for RS489. These connectors (JST) will be replaced from their current 4-pin connector to a 2-pin connector. The connector will drop their GND and +5V connections, but this gives space so one can replace the SD-card without disassembling the whole HAT. When I order the HAT, the minimum order is 5 pieces. Instead of ordering yourself, you can also take 1 or 2 from me. |
Beta Was this translation helpful? Give feedback.
-
|
Managed to put the required changes in:
The first batch of boards is ordered. Wondering how the art work is coming out. |
Beta Was this translation helpful? Give feedback.
-
|
The first batch of HATs have arrived 🎉 . The CNC-logo has turned out nicely; the silkscreen print has really improved over the last few years. I can take a picture of the board, but it is just like the picture above. The parts for the HATs are not as quick though to arrive at my doorstep 😭 . Hopefully they wil arrive coming week, because I promised to present Litex-CNC and the board at a Dutch CNC meeting. In the meanwhile I designed the first daughter board for the HAT. Connection for 4 axis stepgen + home switches. Features:
|
Beta Was this translation helpful? Give feedback.
-
|
Still waiting for more parts to arrive; only received some resistors until so far. The bulk of the components is not in a hurry. Designed a second break-out board fro differential inputs (like encoders):
|
Beta Was this translation helpful? Give feedback.
-
|
Finished with documenting the various options for connecting to the FPGA: https://litex-cnc.readthedocs.io/en/48-support-spi/connections/index.html |
Beta Was this translation helpful? Give feedback.
-
|
Hi Pete, |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for your heads up! Keep us posted on the firmware. You might be able to use some of the break-out boards (although not the serial ones ATM, because SmartSerial from MESA is difficult to replicate). |
Beta Was this translation helpful? Give feedback.
-
|
I added three more board types to Litex-CNC to support the HUB75HAT:
The version number in these strings indicate the version of the underlying 5a-75b board. In these boards the IO of the HUB75 ports is mapped to the three I/O connectors on the HAT. In this way you don't have to know how the HAT is wired, you can reference directly to the ports on the HAT. For example:
These greatly simplifies the configuration of these boards, as one can use default config which belongs to the BOB. For example, when using the BOB4AXIS: {
"board_name": "EMCO5_SPI",
"board_type": "HUB75HAT v8.0",
"clock_frequency": 50000000,
"connections": {
"connection_type": "spi",
"mosi": "spi:0",
"miso": "spi:1",
"clk": "spi:2",
"cs_n": "spi:3"
},
"modules": [
{
"module_type": "gpio",
"instances": [
{"direction": "out", "pin":"j1:1", "name":"stepgen-enable"},
{"direction": "in", "pin":"j1:17", "name":"home-x"},
{"direction": "in", "pin":"j1:16", "name":"home-x-slave"},
{"direction": "in", "pin":"j1:15", "name":"home-y"},
{"direction": "in", "pin":"j1:14", "name":"home-z"},
{"direction": "in", "pin":"j1:10", "name":"drive-alarm-x"},
{"direction": "in", "pin":"j1:11", "name":"drive-alarm-x-slave"},
{"direction": "in", "pin":"j1:12", "name":"drive-alarm-y"},
{"direction": "in", "pin":"j1:13", "name":"drive-alarm-z"}
]
}, {
"module_type": "stepgen",
"instances": [
{
"name" : "x",
"pins" : {
"stepgen_type": "step_dir",
"step_pin": "j1:2",
"dir_pin" : "j1:6"
},
"soft_stop": true
}, {
"name" : "x-slave",
"pins" : {
"stepgen_type": "step_dir",
"step_pin": "j1:3",
"dir_pin" : "j1:7"
},
"soft_stop": true
}, {
"name" : "y",
"pins" : {
"stepgen_type": "step_dir",
"step_pin": "j1:4",
"dir_pin" : "j1:8"
},
"soft_stop": true
}, {
"name" : "z",
"pins" : {
"stepgen_type": "step_dir",
"step_pin": "j1:5",
"dir_pin" : "j1:9"
},
"soft_stop": true
}
]
}
]
}
When changing the BOB4AXIS to a different pin, only |
Beta Was this translation helpful? Give feedback.
-
|
I wrote a demo configuration for LinuxCNC based on this configuration. The timings, speeds and limits of the joints are off course not correct for any production, but it is a nice basis if you want to start with the HUB75HAT. The configuration has been pre-compiled and can be directly flashed to the 5A-75B if you have a V8.0. If you have a different version, you can modify the configuration and build your own using The HAL files should work as-is, using GMOCCAPY as GUI. |
Beta Was this translation helpful? Give feedback.
-
|
Today the mailman had a present 😄; all the parts have arrived. The headers for the HUB75 port, JTAG, power have been soldered to test fit the HAT. And it fits! Also connected some wire for JTAG to the 40-pin connector to test whether the card would boot and whether the JTAG programming would work. Apparently the jumper wires do heavily affect the flashing of the board. With the HAT I can have higher frequencies and less errors while programming. Previously the board would fail to program even when you dared to look angry at it. Tomorrow I'm going to finish soldering the buffers and headers to the board and do some testing! When everything is put together, I'll post some pictures. Too bad my soldering skills look much less fancy then the renders.... |
Beta Was this translation helpful? Give feedback.
-
|
What do you mean with Rio? The easiest way to 'desolder' the buffer is by cutting the legs with a very small pliers. Afterwards you can then easily clean up the pads with a soldering iron and apply some fresh solder. For the wiring I use the wires from a flexible Cat 5 cable; the individual strands are small enough to fit on the pads. I take a long wire and tack it onto the first pad (with ample Flux, solder is already on the pads) and then bend it to the correct second pad. |
Beta Was this translation helpful? Give feedback.
-
|
RIO i an other FPGA CNC Project: https://github.com/multigcs/LinuxCNC-RIO |
Beta Was this translation helpful? Give feedback.
-
|
That's also a cool project. Indeed, I saw that one before, but didn't remember the name. With porting, you mean getting LitexCNC running on it? Would love to support that board as well. |
Beta Was this translation helpful? Give feedback.
-
|
thank you. Do you know how much logic-cells LiteX-CNC needs without ethernet ? RIO unfortunately has no ethernet support at the moment, but runs on almost all FPGA's from 4000 LUT's upwards |
Beta Was this translation helpful? Give feedback.
-
|
If I've read the RIO documentation correctly, they do support the 5A-75B card. These boards are based on To use my hardware: please do! The HUB75HAT shown in this thread is developed in a different repository. Last Saturday I've launched this open-source project in The Netherlands during a CNC Tech Meetup, so I opened the repository. You can find it here. There is still some work to do in that repo, for example the board has the following issues:
The designs of the add-ons are published in the folder Also, documentation is required. |
Beta Was this translation helpful? Give feedback.
-
|
Now the protoype has been build and tested, the branch #48 will be closing this week (after adding the documentation) and version In the coming version I also will streamline the commands of
|
Beta Was this translation helpful? Give feedback.
-
|
The branch #48 has been merged. Thanks for the support in pulling this off! |
Beta Was this translation helpful? Give feedback.
-
Playing around to create a board driver which sends the commands over SPI from the RPi to the FPGA. Frees up the ethernet card and also possibly get some tighter timings.
Together with this I would like to develop an open-source version of the 7C81. Requirements for the board:
NOTE: the 5A-75B only has 48 + 6 available IO, of which 4 are used by the SPI interface (CS, MISO, MOSI and CLCK). To have all 17 pins free and non-shared, the SPI has to loose wires, which has an impact on the speed of the protocol
Beta Was this translation helpful? Give feedback.
All reactions