Phase reset should have been given its own activation bits like key on bits for each operator. When I was making my OPN2 emulator I accidentally had the phase not reset on new notes and things were really cool. No clicks and nice dynamic sound, no two notes were the same haha. This would have been cool to have as real feature on the chips...
One thing that is possible in emulation is to lowpass filter the phase output, so that sudden changes such as what produces the clicks will get smoothed out. The filter will need to take account the phase increment, otherwise you're gonna destroy higher freq sounds but overall I think the idea should work nicely.

Tiido wrote:

One thing that is possible in emulation is to lowpass filter the phase output, so that sudden changes such as what produces the clicks will get smoothed out. The filter will need to take account the phase increment, otherwise you're gonna destroy higher freq sounds but overall I think the idea should work nicely.

Sorry, this seems to be a bit more than my knowledge reserve. Is that the de-click function of some commerial audio repairing tools? Thanks.

Stuff in commercial audio tools is some voodoo since it works in entire audio range not a single sawtooth generator output (that's what the phase generator is). Phase generator output goes into sine and power table, ADSR is applied and final output is produced for the operator. Some of that final output can go back into the phase generator to modulate it, which is the key to the entire synthesis method.

Wow! It sounds great! Thanks so much!
So... how do we start experimenting?

We know the A of envelope could be 15 to 0. Higher value comes with faster attack. 15 means immediately. This could cause the sudden change of the output.

Does it happen instantaneously on the real chip? Or still takes some measurable time?

I was coding a mod player and had problems with popping on certain mods where a looped sample would stop abruptly. I did try a simple low-pass filter with averaging consecutive samples together which helped a little bit. But later I solved the problem entirely by interpolating the volume level over a number of samples. So when it drops from 64 to 0 for instance, I keep playing the loop and drop the volume by 1 each sample until it reaches 0. So it goes silent very quickly (64 samples = ~1.5 milliseconds) but not instantly so no more pop.

bakemono wrote:

Does it happen instantaneously on the real chip? Or still takes some measurable time?

I was coding a mod player and had problems with popping on certain mods where a looped sample would stop abruptly. I did try a simple low-pass filter with averaging consecutive samples together which helped a little bit. But later I solved the problem entirely by interpolating the volume level over a number of samples. So when it drops from 64 to 0 for instance, I keep playing the loop and drop the volume by 1 each sample until it reaches 0. So it goes silent very quickly (64 samples = ~1.5 milliseconds) but not instantly so no more pop.

Sorry, I don't have a real OPL 2/3 chip at hand.
Moreover, I think maybe there is no way to test it since I have no idea how to make a note with a non-zero start (there is always a phase reset).
Maybe we could take a look at NukedOPL core, since it is based on chip scanning.

Yes! That's right!
Actually, I have tested a similar solution: use another group of channel, making notes alternately together with the original channles, to give each note more release time .
However, I found some notes with audible longer A time. We know, each channel works as a monophonic synthesizer. I found that the start point of the envelope of the new note is affected by the previous envelope level. So simply changing the working mode to polyphonic is not OK.

A key-tracking lowpass filter would work, but you might still have some anomaly, limited to the frequency band of the waveform. If you're already implementing a temporal solution, why not watch for a zero-crossing instead?

I'm rezzing this thread for a question about the possibility of eliminating the clicks in games that support Adlib when using a dual OPL2 card such as the SB Pro 1 or PAS 8-bit.

Would it be possible to send every other note to separate OPL2 chips? Would this help at all?