budude wrote:Hmm - don't get that out of that - you don't need IGMP for multicast (yes - it's better to have it - but not required) - - none of the E68x controllers use it for example. I'm not sure how else you would read "requiring to connect directly to the show computer and gobble up an ethernet port may kill it for some" other than a physical direct connection? Whether the OP meant that is left for him to answer I guess...
My comment was based on the assumption that he was just implementing a passive hardware based packet sniffer, scraping out the ACN data he needs. That is what I gathered from reading the Ethernet Parsing section of his pdf. Just my assumption and you know what they say about those!
Zeph wrote:Could you sketch out some of your ideas for the FPGA programming, ideally for those of us not already expert with those devices?If you are curious about FPGA's there is an overview on wikipedia.
For example, to control a ws2811, you will need some kind of logic which recognizes when an ethernet packet is received, checks certain bytes to see that it's valid, sorts out the universe, then fetches bytes from the ethernet buffer and serially encodes them with appropriate clock rates, to the right output ports, right? Or...?
For my design there are two main state machines. One to parse the input data from the computer and store the results in the memory corresponding to the correct output universe. The second state machine basically reads the pixels from memory and outputs them using a shift register.
I'm guessing that you will not buffer more than a byte at a time within the FPGA, and will use buffer memory on the ethernet interface to hold the brightness data, right?
The FPGA I am using has enough internal memory for 36 universes and about 8000 logic elements. Each of the logic elements can be one bit of a counter/adder/shift register, etc.
Will you use some sort of state machine to step through the packet bytes, comparing some bytes to mandatory patterns, ignoring others, and getting universe data from other; if everything's OK, you will copy the data payload out.
Will this send to a single output string (universe) output at a time, or will it send to multiple universes overlapped? In the latter case, that might involve separate state machines for each universe, which interleave their accesses to the buffer RAM on the ethernet interface?
Each universe will have it's own memory. However, if I run out of logic I may use one output controller for two universes.
Is there any way to configure to send less than a universe out to a pixel string?
ACN 1.31 only sends the active pixels in a universe. I will also only output the active pixels for the universe.
sporadic wrote:What about slapping a micro in front so you can interface to a full TCP/IP stack in software and leaving the FPGA to pure pixel / signal generation madness? Or would throughput be lacking? If not, may give you a chanced to up the channel count even more w/ a fast enough micro feeding it. A pure FPGA implementation would be cool but requiring to connect directly to the show computer and gobble up an ethernet port may kill it for some. Haven't really messed with FPGA's, but from what I understand the implementation of something like a stateful TCP/IP machine is quite an undertaking. This may be just the project I need to start playing with FPGAs! Thanks for this!
I could use a micro controller, but it would not be fast enough to handle the full bandwidth of 100Mbit/sec Ethernet. I can easily do this within the FPGA using very few resources. Also, I can remove the limit of the Wiznet chip and handle as many ports as I want.
budude wrote:Why would you need a direct connection to a show computer - seems like normal switching like we use today should still apply?
I have not implemented the control messages to tell switches what ports I am supporting. With a direct connect, all I do is receive the incoming data. So this implementation may not work with all switches. I am thinking of using a y cable for input/output, so you can daisy chain the boards.
IGMP IS basicly what tells the switch to send broadcast packets down a port. Any ways the y cabel is not a bad idea but 10/100mbs hubs are dirt cheap so might not be nessary. Also you might not want the single failure thing where one fault(short) in the y cables bring down the entire chain.
