What the PSG also can sound like...

Here is some more information about the SID player on the ZX-Spectrum. It does not play directly a sid file (Like NYYRIKKI's sidplay or RoboSID) but a converted preprocessed file (.m64).

https://www.worldofspectrum.org/forums/discussion/54160/c64-sid-player-coming-soon-to-specy-128

Apparently only the 50hz timer is used:

Now that you mention it, in Auf Wiedersehen Monty you can also clearly hear that they not only implemented PWM (often called “the SID effect” on Atari ST) but also triangle waves… that’s pretty nice, I haven’t heard that before in ST music I think.

One of my favourites is the Double Dragon 2 theme. Uff, that PSG, beautiful.

(Sorry, off-topic :D)

(Slightly more on-topic, MSX version of Auf Wiedersehen Monty with correct speed is here @ 6:18.)

On the subject of Gremlin, let’s not forget Death Wish III and Venom Strikes Back .

But anyway, back to “what the PSG can also sound like” as-in-unexpectedly, actually a pretty easy and little CPU-consuming (and underused) technique is to use audio-rate frequencies for the envelope generator! This way you can get nice buzzy sounds like in Six Miles High and Raccoonerie by Gasman from AY Riders.

What is called the "SID effect" simply refers to pulse waves with variable duty cycle (either static or modulated over time).

The PSG produces square waves by outputting 1 for a period of n clock ticks (/16), and then outputting 0 for another n clock ticks (/16). The output of 1 is also modulated by the volume setting.

The more general case of square waves is pulse waves, which specify the 1 period relative to the 0 period as the duty cycle. For square waves, because the 1 and 0 periods are the same, the duty cycle is 50%.

So the goal is to make the PSG generate a pulse waveform with duty cycles < 50%. This can be achieved by prematurely cutting off the 1 output using the volume register.

For example, when playing a tone with period 100, the PSG will output a 1 for 1600 cycles and a 0 for another 1600 cycles. Now if you let the CPU control the volume synchronously, setting volume to 15 initially, wait 800 cycles, set volume 0, wait 2400 cycles, set volume back to 15, and repeat. Then the actual waveform that is produced will be a pulse wave with 25% duty cycle.

As a slight simplification, because the volume can be turned back on anywhere within the latter 1600 cycles (the output will remain 0 either way), you can actually modulate the volume at the exact same rate as the PSG does, and the initial time delay will determine the pulse width. E.g. initially wait 800 cycles, set volume 0, wait 1600 cycles, set volume 15, wait 1600 cycles, set volume 0, etc. will produce the very same 25% duty cycle.

The difficulty here is that as the tone frequencies change, the CPU needs to cut the volume at different rates as well. If you add support for 3 channels and pulse width modulation, it further complicates matters.

For the Atari ST this is relatively easy to do because it has DMA whose rate can be controlled by a timer (at least I think so). For MSX, quite a bit more complicated.

Perhaps the easiest way is to use a high frequency timer interrupt and decrement counters for all three channels, every time they cross 0 you toggle the volume and re-add the tone frequency to the counter. The duty cycle will be a bit jittery, but if the update frequency is high enough it’ll probably sound ok.

Note that duty cycles > 50% don’t need to be supported, because a pulse wave with duty cycle 80% sounds identical to one with duty cycle 20%, except that the waveform is inversed.