External Audio DSP Processor

[williamk]

Status: Just an idea
License: GPL

The idea is to use the http://www.xmos.com/products/developmen ... opment-kit as it is, no hardware changes.
In the computer side, the user could use a VST that handles data transfer to the XC-1A device.
Them you can use the device as an effects processor, or even VA (Virtual Analog) synthesis devices.
We don't know if the USB connection can be used as it is, or if extra drivers are required, or even if the current XTAG USB driver will do the trick. If not, we will scratch this idea and go with something else.
There's also the XC-2 (http://www.xmos.com/products/developmen ... hernet-kit) option, which the network can be used to transfer data, but a large delay would be added that way.

[Bianco]

The XC-1A uses an FTDI FT2232 chip for JTAG.
It also emulates a serial port. The emulated serial port and JTAG are the only ways to transfer data to the processor. This is not really ideal.

[nassim]

HI? can you put Somme details and it's similar of the Blackfin DSP?

thanks

[williamk]

Well, Blackfin seems to be only 400mhz and a single chip, so the quad-xmos chip would do much more with much less money, from what I could understand.

[williamk]

I still haven't looked at how the XC-1A talks back to the computer via USB, but I guess I will need help from someone figuring out stuff out.

There's another kit with a RJ45 connection, which I assume is much faster compared to the USB version? But I don't know if it wouldn't add too much latency to the whole thing.

Still researching. But if the XC-1A kit USB connection won't do it, I guess we will need to MOD it to be able to handle it.

[williamk]

Here's the link for the topic at the forum:

http://www.xcore.com/forum/viewtopic.php?t=754

[tknesel]

Not necessarily. While XMOS chips are attractive because of their DSP capabilities keep in mind that they are fixed rather than floating point. Also, DSP such as Blackfin, Sharc, Sigma, etc are designed with audio and image processing in mind and have a large database of reference code and virtual programming libraries. So depending on your application, channel count, AVB, etc. it might be easier (and more cost effective) to develop your project on the XMOS L2 combined with Sigma DSP for the tough stuff such as EQ, reverbs, etc. Keep in mind that with any XMOS chip you will largely have to develop your own audio DSP algorithms or manually port them from another language.

Right now there is not much more available than some of the filters in the XDK and USB audio reference code.

[williamk]

I get the part that other DSP devices have much more power, BUT, they are harder to program, and not open-source like XMOS is. XMOS is 32 bits integer, which CAN do audio related processing. Take a look at this: http://www.chameleon.synth.net/
Its based on a very old DSP processor: Motorola DSP56303, 24 bit @ 100 MHz - Heck, listen to the audio examples, good stuff in there. So I wonder what a quad-core XMOS chip can do, even if it has to run in stand-alone mode, maybe with only midi been sent to it. I'm sure that with time I can build my own advanced project that will allow users to use it in the way I want, but for now, I guess I will have to stick with something simpler for starts.

[williamk]

I know someone already mentioned the XC-2 kit, but I was worried about speed and latency. But the guys from FxTeleport has something like this over the net and latency wasn't that bad, at least not for effects.

So I wonder if maybe the XC-2 kit would do better?

http://www.xmos.com/products/developmen ... hernet-kit

From what I understand its $149 USD and includes the XTAG connector and USB cable too, right? If that's so, its good. I still like the XC-1A better for the price, but heck, this is just a test anyway...

[tknesel]

Hi William ..
Trust me...NOT at all saying that XMOS can't do audio. In fact my company is actively developing multiple audio products on the G4 platform with heavy implementation of the DSP for audio processing. For bi-quad filters, mixers, compression and other things, yes, XMOS is pretty powerful and quite capable. I'll disagree on the dedicated DSP being harder to program as Cirrus and AD both have very powerful tools for programming that are as close to drag and drop as you can get AND there are a lot of excellent algorithms that are ready to port. So - really - depending on what you want / need to do it could be easier to use a bespoke DSP for audio processing vs. rolling your own or porting code from another platform. Things like reverbs, all pass, multiband compressors, impulse, room tuning - MIGHT be better handled by a processor designed and optimized for those algorithms. You can't really compare MIPS, threads or processor speed and say one is more powerful. It might also be a much shorter product development cycle to use tools optimized for particular functions. Some of the Sigma processors, for instance, can be $2 or $3 at production quantities. So from a BOM cost standpoint an L2 + Sigma could be a cheaper solution than a single G4.

You've got some good ideas. Interested in seeing where you take them!