hello,
Le 10/05/2016 00:37, Christof Ressi a écrit :
Hi,
I've read your mail, it actually was the key source of inspiration for trying
out DSP on a microcontroller!
in fact, I did not recommend audio processing on hardware that did not have
support for float arithmetic.
(only me or crazy nerd do that).
I might be crazy as well then :-). I just wanna give it a try and see how far
I'll get. I'm just excited about the low price of these modules, given their
rather fast speed (and of course the integrated WiFi functionality!). I'm
dreaming of tiny effect/sound modules that do a certain DSP job and
receive/send messages via WiFi.
a micro controller to use is for example the cortex M4. You can find them on
the teensy hardware.
Apparantly the Teensy 3.x uses the non-FPU version of the cortex M4:
https://forum.pjrc.com/threads/25909-Teensy3-1-and-FPU-%28Floating-Poitn-Unit%29
crap.
Right now I can't see any advantage over the ESP8266. Also it costs about 4
times as much!
unless you are planning to produce 100 000 of this hardware, 20$ is nothing
regarding 1 month of work.
I think you should start using the ESP8266 for the wifi and connect them to a
teensy and it's audio feature. Then, you'll be able to optimized the teensy
with you DSP code, and eventually put it in the ESP8266.
Some guy made a benchmark test between ESP8266, Teensy 3.1 and Arduino Due:
http://forum.arduino.cc/index.php?topic=337653.0
The ESP8266 wins for floating points (and loses clearly for integers).
floating point are about 10 time slower than int. You can't use them.
according to this benchmark, on the ESP826 chip, this line of code need 431ms:
for (int i = 0; i < 100000; i++) a += a * a + 1000.7 / a - 40.4;
at 44.1KHz, you have only 22µs between audio sample.
let's say that you can use 15µs for each sample, the rest for the wifi, audio
codec communication, etc.
1 sample computation should then be faster than :
for (int i = 0; i < 4; i++) a += a * a + 1000.7 / a - 40.4;
(note the i<4) : this is not much for serious DSP stuff.
using int on the ESP826, 1 sample computation should be faster than :
for (int i = 0; i < 37; i++) b += b * b + 1000 / b - 40;
on the OCS synth, i reduce the moog filter (same algo than the one i send to
this list) to:
20 time the >> or << instruction
23 : addition or substraction
12 : 32 bit multiplication
6 : 64 bit multiplication
it is the most CPU hungry module.
Since you got DSP stuff working on an Arduino Due, it should also work on an
ESP8266, I guess... But it's my task to find out.
it work, but only if you find how an elephant can fit in a show box.
By the way, what kind of interface did you actually choose for the audio output
and have you tried several methods?
I know only 1 way to go : i2s.
i use a WM8731.
cheers
c
Thanks for sharing your experience!
Christof
Gesendet: Montag, 09. Mai 2016 um 17:39 Uhr
Von: "cyrille henry" <[email protected]>
An: [email protected]
Betreff: Re: [PD] LibPd on microcontroller
hello,
I recently send a mail in this list describing my experience of audio synthesis
on microcontroler, but i did not use libpd.
i did not recommend libpd for microcontrolers.
in fact, I did not recommend audio processing on hardware that did not have
support for float arithmetic.
(only me or crazy nerd do that).
a micro controller to use is for example the cortex M4. You can find them on
the teensy hardware.
teensy also provide a audio synthesis library, and is supported by arduino IDE.
You can also have a look at the heavy compiler if you want to embedded a pd
patch.
Cyrille
Le 09/05/2016 17:01, Christof Ressi a écrit :
Hi,
has anyone experience in using LibPd on a microcontroller?
I wanna try to run some rather basic DSP code on this guy here:
https://www.olimex.com/Products/IoT/MOD-WIFI-ESP8266-DEV/open-source-hardware
There are some existing Arduino/C++ frameworks and I thought I could either
write my own small DSP library in C++ or use LibPd.
I understood that LibPd is completely independed of audio drivers and external
libraries and since it's plain C code it should work on everything that can run
code. However, I have only encountered LibPd in the context of other
applications (processing, openFrameworks) and mobile devices so far. Do you
think it could work on an ESP8266? I'm also a bit concerned about efficiency,
as computation power is rather limited:
"ESP8266EX is embedded with Tensilica L106 32-bit micro controller (MCU), which
features extra low
power consumption and 16-bit RSIC. The CPU clock speed is 80MHz. It
can also reach a maximum
value of 160MHz. Real Time Operation System (RTOS) is enabled.
Currently, only 20% of MIPS has
been occupied by the WiFi stack, the rest can all be used for user
application programming and
development."
Christof
_______________________________________________
[email protected] mailing list
UNSUBSCRIBE and account-management ->
https://lists.puredata.info/listinfo/pd-list
_______________________________________________
[email protected] mailing list
UNSUBSCRIBE and account-management ->
https://lists.puredata.info/listinfo/pd-list
_______________________________________________
[email protected] mailing list
UNSUBSCRIBE and account-management ->
https://lists.puredata.info/listinfo/pd-list
_______________________________________________
[email protected] mailing list
UNSUBSCRIBE and account-management ->
https://lists.puredata.info/listinfo/pd-list
