[sdiy] Hardware convolution box?

[Mikko Helin]

This XMOS board designed (for stompbox use mostly) by an XMOS employee
might be a good alternative for DSP chips and better than generic
ARM's:

http://www.diystompboxes.com/smfforum/index.php?topic=114354.0



On Tue, Feb 7, 2017 at 8:12 AM,  <rsdio at audiobanshee.com> wrote:
>
> On Feb 6, 2017, at 9:20 PM, cheater00 cheater00 <cheater00 at gmail.com> wrote:
>> I understand FFT domain convolution introduces latency, which is not wanted in hardware.
>
> There are a few considerations regarding latency:
>
> 1) Most convolution represents a reverberation, especially convolutions of impulses that last a couple of seconds. In nearly all cases, the reverb involves a pre-delay before the signal appears in the impulse response. In such cases, the latency can be subtracted from the pre-delay and then there won't be any perceived latency.
>
> 2) All 24-bit converters have hundreds of samples of delay. Or at least it seems that all 24-bit converters are built on delta-sigma technology with tons of digital filtering. Unless you're going to based your design around a 16-bit or perhaps 20-bit codec, you won't be able to remove all latency regardless of the convolution algorithm used.
>
> 3) One of the responses mentioned overlap-add FFT. That technique allows for significant reduction in latency, provided that the processor has the power to calculate many more FFTs per the time available.
>
> That said, I support all designs which strive to reduce latency in hardware! It's just not as easy as you might think. You certainly want to start with a DSP, though, for the best chances.
>
> Brian Willoughby
> Sound Consulting
>
>
> _______________________________________________
> Synth-diy mailing list
> Synth-diy at synth-diy.org
> http://synth-diy.org/mailman/listinfo/synth-diy