FPGA's and Development Boards

[JovianPyx]

Great idea, and thanks to whoever's idea it was

I'd like to make a point about FPGAs here. An FPGA is not at all like a traditional DSP or microprocessor. Instead it is like an ocean of initially unconnected logic blocks which I like to call "Logic Lego". With HDL you can describe extremely complex configurations or circuits. All of the logic blocks are familiar to anyone who has played with CMOS or TTL logic devices.

With a device like a microprocessor or a PIC or AVR, etc., you write a program that performs functions in a step by step manner. A drawback of this is that it is wholly bound by time. If the device contains hardware to multiply two numbers, it usually contains just one and that hardware must be "shared" by all operations that need it. This limits the amount of work the device can do within a given period of time.

An FPGA, on the other hand, has many copies of the same basic logic circuits. When you configure them, they all work "in parallel" just like all six inverters on a CD4069 work (contrasting to a microprocessor which allows an OR function, but literally one operation at a time). These copies include internal RAMs and multipliers as well as adder/subtractors, comparators, etc. This means that you can construct circuits which take advantage of parallelism to increase speed and efficiency.

One must wrap their brain around this idea of "parallelism" which may seem slightly foreign to a traditional programmer at first. One must also get past the fact that HDLs (Hardware Definition Languages) *look* like a programming language. They are nothing of the kind. Rather they literally describe hardware.

Really, using an FPGA is more like designing a huge PCB with tons of logic on it.

And FPGAs are fast too. The one I use is capable of 200 MHz operation, but I've never needed more than the crystal clock of 50 MHz provides.

Another thing to note is that while microprocessors have fixed word size, FPGAs do not. If you need to do 128 bit arithmetic, then do it. This is an obvious efficiency point. In my own work, I've designed phase accumulator NCOs that have a 48 bit wide word size. In a microprocessor, this would take repeated use of 16 bit resources and adds complexity and size to the program. In an FPGA, it's just a matter of defining the word size of the "logic objects" that you are using.

IMO, digital synthesizers can take great advantage of the parallelism and vast resources in an FPGA. The fact that they contain several multipliers means that parallel DSP is possible. That is, you can have more than one DSP circuit in an FPGA, each operating in parallel and processing completely different signals. My own synth designs take advantage of parallelism, at the very least, the synth itself is one chunk of hardware while the MIDI controller is yet another that operates in parallel with it meaning that no care needs to be paid to whether an interrupt or other microprocessor action will degrade the audio performance of the design. In fact, my MIDI controller is just HDL code that creates a microcontroller within the FPGA which runs an assembly language program to perform MIDI functions such as "turn a specific note on".

I am not trying to disuade people from building projects using other computation capable devices, only that one should consider the FPGA, especially if the project is considered rather large. After seeing what an FPGA can do in my own lab, I am glad to have abandoned the idea of attempting similarly large projects with a microprocessor (for me, this all started with attempting to replicate the AVR-synth with a 486-133 - this would have been a monosynth with little possibility of ever becoming polyphonic). I've done 8 voice polyphony using only the on-chip resources of an FPGA and there are still unused resources within the IC.

It is my hope that more people will experiment with FPGAs for music and share their designs.

I do not work for Xilinx or any other FPGA mfr, I just think they are really cool.

[State Machine]

Scott,

Well written and thank you for the insight on this remarkable subject

[Uncle Krunkus]

Are these programmers, or do you need one of these for every project?

PS I know that is possibly a silly question to those who know this stuff, but I just wanted to check.

[JovianPyx]

[State Machine]

[Scott Stites]

Thanks indeed for that write-up, Scott. I learned a heckuva lot right there. I guess the first step is to really understand what the device is, and that certainly widened my pathetically thin knowledge of what an FPGA is and how it works. Uncle K already asked my first question.

It's odd that I work with instruments that incorporate FPGAs in a big way, but I have such a dim understanding of them. It's exciting to think of a vast plain of logic gates waiting to be configured and told what to do, though the aspect of learning and using the code to control it is still a bit daunting to me.

My quest would be to create an interface to convert control voltages to MIDI data. I know Analogue Solutions (or is it Analogue Systems? - one of those Analogues, anyway) has done it, but it would be really cool to custom build one. I can't imagine applying all of the groovy control surfaces of a modular (like Ian's chaotic circuits or the Klee, or both) to MIDI controlled synths, even VAs. I think that would open up worlds for me.

I suppose such an endeavour would require something greater than 12 bits, but, again, I suppose there's nothing stopping one from adding higher resolution ADC/DAC's to the inputs and outpts of the FPGA, no?

Cheers and thanks for the new forum, electro-music.

Scott

[State Machine]

[JovianPyx]

[blue hell]

I think OSC over ethernet would be very worthwhile

[JovianPyx]

[DrJustice]

Folks, perhaps this welcome thread should be split off at the FPGA-point?

And yeah, OSC over ethernet is cool but sadly underused - it ought to be the new music industry standard! In a programmable logic device OSC really needs a CPU core, because of the protocol complexity (it's actually very simple, but would probably use far too many gates if implemented as a straight up state machine).

@ ScottG : here's some electro-music.com OSC talk.

DJ
--

[Junk Rhythm]

Thank you for this sub-forum. I have lurked on this site for a long time and this forum was the last straw forcing me to finally register.

[blue hell]

[blue hell]

[State Machine]

[State Machine]

[bugfight]

alright, i feel myself getting sucked into the fpga thing now...
but wouldn't the Spartan 3A DSP be better than the 3e for audio applications?

[frijitz]

[State Machine]

[frijitz]

[State Machine]

[State Machine]

[bugfight]

[frijitz]

[State Machine]