[sdiy] Hardware convolution box?

[Mikko Helin]

Proper way to handle large filter kernels (IR's) is to use partitioned
kenel and various combined methods (FIR for initial response, FFT for
larger delayed partitions) to get lower overall latency, see the
answer to this questin on StackExchange forum:

http://dsp.stackexchange.com/questions/25931/how-do-real-time-convolution-plugins-process-audio-so-quickly

So your hardware should be built around this in mind. For two second
IR (mono) youll need at least (sampling at 48 kHz) 192 kbytes of RAM
(not necessary continuous). It can be done easily by the board Mark
Seel is going to put to Kickstarter soon. It uses XF232 XMOS chip with
32 cores and total of 1024 kbytes or RAM. Please to to forum and ask
any question there. One of the main uses for the board is going to be
guitar speaker simulation using IR:s (in addition to any other
conventional stompbox functions).

http://www.diystompboxes.com/smfforum/index.php?topic=114354.120

On Fri, Feb 10, 2017 at 10:27 PM, cheater00 cheater00
<cheater00 at gmail.com> wrote:
> I assumed as much - that's why I think a convolution box should be within
> reach of even a novice dev. What about libraries for: reading off sd card or
> usb stick; using data structures like ring buffers; driving lcds; text
> driven menu uis; scanning front panels... is this sort of stuff available?
>
>
> On Fri, 10 Feb 2017 19:56 , <rsdio at audiobanshee.com> wrote:
>>
>> Hello all,
>>
>> Note that Texas Instruments has open source libraries that include most
>> common DSP routines. These libraries are almost exclusively written in
>> assembly, but the routines can be called from Standard C code. It's possible
>> to write firmware mostly in C, while linking in assembly routines written by
>> the experts at TI. When developing for the TMS320 C5000 family, I looked at
>> how the different variations of FFT were optimized for different goals and
>> applications, and then designed my own hybrid FFT that was optimized for my
>> exact data. Because the libraries are open source, I was able to contribute
>> bug fixes and optimizations to the Texas Instruments library that should now
>> be available to other developers on the C5000 family.
>>
>> I assume that the C6000 family has similar open source libraries. In other
>> words, you might not have to learn TMS320 assembly to get optimal code. I
>> haven't checked, but it's possible that convolution is already available in
>> the library. You might benefit from tweaking their code, but as a starting
>> point you might be able to get up and running quickly with a DSP library.
>>
>> Brian
>>
>>
>> On Feb 10, 2017, at 8:43 AM, Terry Shultz <thx1138 at earthlink.net> wrote:
>>
>> I use the Code Composer tools from TI and I have been using these tools
>> since they were in Alpha state.
>> Assembly code is difficult on the TI DSP Sitara platform. Pipeline length
>> and such.
>>
>> I find it easier to use one DSP as the Decoder for Atmos and the 2nd as
>> post processing and Bass manager etc.
>>
>> Also found it easier to do assembly programming on the ADI 4th Generation
>> Falcon parts as the pipeline line was a bit shorter. The longer the
>> pipeline,
>> the more difficult to re-order efficiently.
>>
>> In some cases I have had to build a small library for hand tuned FFT’s and
>> IFFT’s. Can’t show this IP as I was paid to develop for Automotive apps. by
>> Semiconductor company.
>>
>> The compilers have gotten pretty good for TI and ADI and I do less
>> assembly work more and more.
>>
>>
>> _______________________________________________
>> Synth-diy mailing list
>> Synth-diy at synth-diy.org
>> http://synth-diy.org/mailman/listinfo/synth-diy
>
>
> _______________________________________________
> Synth-diy mailing list
> Synth-diy at synth-diy.org
> http://synth-diy.org/mailman/listinfo/synth-diy
>