Why are the software-generated sounds of pre-Rock Gottlieb games lo-fi and muffled?

[smaybius]

Until either 2019 or several years before, the sound of Stern games with the M6840 PTM as the sound generator, as well as something like the GamePlan MSU-1 sound board, had the same filtered sound quality. Everything's through a low-pass filter and/or has sinc interpolation. But this isn't the case for games with the AY or YM synths, where the softsynths have the clearest quality. The softsynths for Williams and Bally games don't have the quality either that I'm reporting as problematic.

[toxieainc]

Could you please try to find the exact version that introduced the 'worse' sound quality for you? That would help.
(btw: 'sinc interpolation' as PinMAME uses it, will not harm sound quality, as its the perfect upsampling from existing data to an output device, and will not introduce any noticable change. BUT of course there could have been some other change that affected the quality)

[smaybius]

3.6.0.1227. It's the accuracy of the emulation that I'm worried about such resampling affecting. Is the sample rate for such games an adjustable setting, or determined by the hardware or ROM? The Stern SB-300, GamePlan MSU-1, and maybe also the Hankin games also used to have such upsampling quality back in 2012 for example, muffled back then but perfectly clear as of 2019-ish, even though the given sound boards use a PTM (such as the Moto 6840) instead of a software synthesizer.

The attached file shows an example from Rack 'em Up, with the SFX and music in order by sound commands.
gottlieb rackemup sfx and moozik.zip

[toxieainc]

It depends. Stern SB-300 definetly has issues in general, the others i will have to double check. As for sampling rate: Back in the days, the sound emulation for some games/soundchips did depend on what the user selected as output rate. This lead to all kinds of issues that users reported.
Nowadays we have either native rates (for most chips) or some artificial oversampling rate, which is the best tradeoff to avoid all kinds of issues we saw back then. But of course there is always room for improvement, especially for older HW.

Thanks for the more detailed analysis!! I will investigate soon.

[essence25]

How about we start developing PC sound boards that have various chipsets on them and run audio on real hardware, interfaced to piname/vpx. Or even using FPGA to simulate audio boards etc.

[toxieainc]

The first suggestion would be nice, but too exotic that it would be worth it, i don't think that many would buy this. Also a lot of the original/very old chips are hard to find nowadays (remember that PinMAME features A LOT of audio chips being emulated, including some that are extremely exotic).
Second one (FPGA) will only make sense if all these sound chips are decapped and then transferred to logic, AND if these (or the soundboards) don't feature any analogue effects (so chips must work fully digital and the boards mostly, too). Either way, the quality will not be automatically better just due to it being run on a FPGA.
What makes FPGAs mostly shine is, that if one simulates ALL chips and wiring of a system on a FPGA, then the result can be bit-exact on all outputs.
Just picking one component and FPGAing that one will not help with syncing, timing or other similar issues.

[mjrgh]

Toxie - that's an excellent explanation. I'd just like to add one more point, which is kind of a corollary to what you said: in the cases where we DO have enough information that you could build a perfect FPGA replica of the chip, we can almost always create an equally perfect replica in software on a PC. Almost any chip from the 1970s or 1980s or even 1990s is going to be simple enough that it won't tax a modern Intel or ARM CPU to replicate the chip's behavior in C/C++ code in the digital domain. I think most of the cases where the MAME audio emulation has problems are cases where there's not enough information available in the data sheets and original schematics to figure out exactly how the original boards worked.

On the other hand, a Universal Pinball Audio board sounds like a fun project that could easily occupy an ambitious hardware designer for years and years and years. I'd estimate that a couple hundred programmer-years worth of time have gone into creating MAME and PinMame, with a lof of that time spent poring over schematics for old machines and experimenting with the ROMs. But I know from my little bits of experience working in isolated corners of the code that the fine engineers who did all of that work were largely allergic to documenting their reverse-engineering work on the old boards, and that the MAME/PinMame source code is pretty much the only documentation that came out of it. If you just created FPGAs and other circuits based on the PinMme code, the hardware would inherit the same deficiencies that the code has, so you'd really have to go back to the original schematics and repeat the entire reverse-engineering enterprise.

[toxieainc]

Decapping will be the solution long-term, after all a lot of other chips already underwent that process (see results in some of the MisterFPGA project cores, and other efforts), and a lot of the process to transferring the logic from the images is nowadays already automated to some extent. So maybe at some point somebody will be brave (insane? ;)) enough to start something like MisterFPGA for pinball simulation??

[smaybius]

Been wondering what's up with the SB-300 despite being pretty much an AY-3-8910 as discrete logic, and also probably the same thing as what Exidy used. From what I remembered, it's only main components are a 6840 for the tone generation, a 128-bit shift register for the noise generation, and an adder for volume modulation.

[toxieainc]

It has been quite a while since i last fiddled with the stern sound boards. Do you have some pointers to 'new' documentation or the like as you seem to know some of the details?

[smaybius]

I looked up schematics and images of the board, and there's also some documentation like https://www.pinball4you.ch/okaegi/rep_soundsb300.html which is for servicing instead of really explaining how everything works. IPDB has schematics that are mostly hard to read, but still hopefully enough in conjunction with photos of the board to remake it in eeschema or KiCad

[mjrgh]

[toxieainc] Do you have some pointers to 'new' documentation
[smaybius] I looked up schematics and images of the board, and there's also some documentation like...

That's basically where the MAME and PinMame code came from, from looking at schematics from the Operations Manuals and service manuals, the original manufacturer data sheets for the ICs, and in some cases maybe even physical specimens of some of the boards. So the IPDB scans of the manuals probably don't have anything new as far as PinMame goes. In fact, I've found that the PinMame source code contains a great deal of information that you can't infer from looking at the schematics, since a lot of the boards have parts that aren't labeled properly in the schematics, jumpered circuit variations, and programmable logic chips that aren't documented anywhere (except for what has been discovered by reverse-engineering by PinMame and other projects).

[smaybius]

Back to the original topic, I remember many years ago pointing out the muffled quality on a VPForums thread, and someone responding about it being at a 16-ish KHz sample rate. I also kept wondering why it's called "custom" instead of "DAC" like probably almost all other softsynths except for the Williams-Bally DCS sound system which is also called "custom".