8-bit Sine tables to not describe sines, but rather negative cosines

sensorium / Mozzi

sound synthesis library for Arduino

https://sensorium.github.io/Mozzi/

GNU Lesser General Public License v2.1

1.09k stars 187 forks source link

8-bit Sine tables to not describe sines, but rather negative cosines #279

Closed eclab closed 2 weeks ago

eclab commented 1 month ago

Most of the sine tables do not start at the origin, but rather at -128. This is bad because it can cause a click if the wave is played raw. They should start at 0, then increase from there, then back down to -128.

tomcombriat commented 1 month ago

Hi, I am not sure I completely understand the issue. Looking at the sine tables, only sin1024_int8.h starts at -128, the others correctly start at 0. I agree that the stated types of the tables are not always coherent (and tables/sin1024_uint8.h looks a bit weird with a lot of padding). A bit of cleanup might be needed.

Or are you talking about the cos tables? In that case, the cosphase are also of interest.

eclab commented 1 month ago

sin1024_int8.h sintest_int8.h sin1024_uint8.t sin8192_uint8.t

These four start at the wrong values. The uint ones should be starting at 128 (that's their origin).

I use sin1024 heavily for 8-bit sounds and this discovery was unfortunate.

So let me change "most" to "several".

tomcombriat commented 1 month ago

Okay, thanks for the clarification. Here is my take on these tables:

sin1024_int8 should be fixed I agree, hopefully no-one is using the wrong behavior. I quite don't understand how that bring some click as the error seems only to be on the phase, hence the wave is still continuous, just not starting at the correct spot. I have not plotted it though.
sintest should be deleted. All its content is commented anyway
sin1024_uint8 and sin8192_uint are, IMO, more confusing than anything else, their content is not matching the said type, and having a trigonometric function which is not actually 0-centered sounds a bit off. I would be in favor of removing them (or deprecating them).

I am interested to have the opinions of @tfry-git and @sensorium before further action though.

eclab commented 1 month ago

sin1024_int8 creates a click when you start the wave in an Oscil because it's not starting at 0 so it pops from 0 to -128 for its first sample output. As to the uint variants, I presumed they're for unipolar modulation.

sensorium commented 1 month ago

I am all for a clean up if someone has the attention to do it!
No objections to re-aligning tables to match their titles. The scripts in /extras for generating tables were written while I was re-aquainting myself with Python, and quite likely adapted and re-used with many inconsistencies etc. They're only a guide and obviously they can also be tidied up if anyone wishes to do it. PR's welcome! @tomcombriat, you do too much already...

I think deprecating old tables would be good to give others time to adapt their software in case they depend on particular (erroneous) behaviours.