That's some pretty clean emulator code! Sticks fairly closely to the hardware design, and it's clear what's going on for the most part. In addition to practical use as an emulator, it might make a good learning tool for someone who wants to understand how the Atari VCS's weird hardware works. Perhaps read along with the canonical Stella Programmer's Guide (http://www.alienbill.com/2600/101/docs/stella.html) and the less technical but still informative book Racing the Beam (http://www.amazon.com/gp/product/026201257X/ref=as_li_ss_tl?...).
A minor quibble is that the audio emulation, while accurate from the perspective of what the audio hardware does, produces a bunch of harsh-sounding aliasing when used to generate digital audio. That's a common problem with 8-bit emulators, though. The short explanation of why: in the hardware implementation, the Atari feeds square waves directly into speakers. Square waves are in a sense the "natural" electrical signal for digital circuits, and from simple bit-patterns you can get complex harmonics. But when, instead of feeding them as voltages directly to a speaker, you generate digital audio samples that look like square waves, the harmonics above the Nyquist frequency (1/2 the sample rate) fold over into the audible range as aliasing artifacts.
The fix is to generate band-limited square waves, which are approximations to square waves w/ no frequency content above the Nyquist frequency. There's an active research literature on how to best do that efficiently (it's important if you're trying to digitally replicate the sound of classic analog synths like the Moog), but it mostly boils down to replacing the discontinuity in the square wave (which is where the harmonics come from) with something more like a ripple, designed to only produce sub-Nyquist harmonics. One fairly simple approach, integrated into several emulators, is explained here (w/code): http://www.slack.net/~ant/bl-synth/
Thats an amazing observation mjn!!!
I knew a little about the Nyquist frequency, but I didn't really hear much aliasing during development. Surely that is a good point to improve!
Thanks.
To be fair it's not really a show-stopper in the context of Atari emulation, just a "nice to have". Most of the Atari's sounds are so noisy to begin with (e.g. the famous Combat tank rumble) that the aliasing is much less noticeable than if you're digitally emulating simpler waveforms, like the ones from analog synthesizers.
The most-used 2600 emulator, Stella, doesn't do bandlimited synthesis either, without many complaints. I believe z26 is the only Atari emulator that does, so listening to differences between how things sound in z26 and other emulators might be a way to figure out if it really matters. Though as far as implementation, z26 is 100% asm, so not very easy to use as a reference.
Probably the easiest way to solve the aliasing problem is to pre-calculate a series of square waves, each with successively fewer harmonics. You can then just select the right square wave for any given note and play it back via linear or polynomial interpolation. This is what I do in my iPad synth app.
The main downside of this approach is that it takes more memory, but it's very easy on the CPU.
Yes... A real Atari does the same thing! That's what the processor and the other chips produce when the machine is started with no ROM to be executed.
I tried to implement the emulation at the hardware component level as accurately as I could. It happened spontaneously!
When the processor tries to read the game program from the empty cartridge slot, it ends up reading random bytes from the data bus. Its like a random program commanding the video and audio hardware!
They don't like clones of games, but I haven't heard of emulators having any trouble. Stella (http://stella.sourceforge.net/) has been around for years without problems. Hopefully it'll stay that way!
A minor quibble is that the audio emulation, while accurate from the perspective of what the audio hardware does, produces a bunch of harsh-sounding aliasing when used to generate digital audio. That's a common problem with 8-bit emulators, though. The short explanation of why: in the hardware implementation, the Atari feeds square waves directly into speakers. Square waves are in a sense the "natural" electrical signal for digital circuits, and from simple bit-patterns you can get complex harmonics. But when, instead of feeding them as voltages directly to a speaker, you generate digital audio samples that look like square waves, the harmonics above the Nyquist frequency (1/2 the sample rate) fold over into the audible range as aliasing artifacts.
The fix is to generate band-limited square waves, which are approximations to square waves w/ no frequency content above the Nyquist frequency. There's an active research literature on how to best do that efficiently (it's important if you're trying to digitally replicate the sound of classic analog synths like the Moog), but it mostly boils down to replacing the discontinuity in the square wave (which is where the harmonics come from) with something more like a ripple, designed to only produce sub-Nyquist harmonics. One fairly simple approach, integrated into several emulators, is explained here (w/code): http://www.slack.net/~ant/bl-synth/