From my experience, half the noise comes from power rails and other half is result from pollution on the silicon level and most particularly in the mixer sections. Supplying VREF for a chipset capable of accepting external VREF from a high PSRR voltage regulator can do wonders and also increase dynamic range a bit in case of things where VREF normally sits elsewhere than half of the power supply. Clean analog AND digital rails help too, but ultimately on-chip crosstalk is what sets the limits. All the junk from ISA signals bouncing around will couple into analog section one way or other on the most integrated designs.

For small gains you can do :
* Opamp changes can help a bit but only if the power rails are decent already and opamp has high PSRR, most of the noise is well beyond ultrasonic range where PSRR falls rapidly of typical components, plus such circuits are generally on inputs side and not outputs. If the design uses single supply rails there's no point messing around, there are no gains to be had without redoing everything. Dual rail (+/-12V that's regulated down to something lower) is absolute necessity for good peformance.
* All the speaker/headphone amps and whatnot should be removed from signal path, inputs disconnected from main circuit and VREF decoupled from the sound chip side if it is used such way. Clean VREF is a necessity for good performance and extra chips hooking onto it will cause pollution due to their poor PSRR and crosstalk figures (again silicon level problems).
* Messing with capacitors and other passive components usually doesn't have whole lot of effect on anything but distortion levels which is usually low enough that there's not much bother with it. Films caps are a double edged sword, physical size makes them whole lot more vulnerable to all the RFI and EMI pickup inside the computer, if you really want to go that route you'll have to build a metal box around the analog section.
* It does help a little to lowpass all inputs and any loopback signals a litle bit above audible range (i.e 25KHz) to reduce any ultrasonics which will put much strain on the not so great analog bits in the chips resulting in extra noise and possible distortions but gains are minor here.

Tiido wrote:

From my experience, half the noise comes from power rails and other half is result from pollution on the silicon level and most particularly in the mixer sections. Supplying VREF for a chipset capable of accepting external VREF from a high PSRR voltage regulator can do wonders and also increase dynamic range a bit in case of things where VREF normally sits elsewhere than half of the power supply. Clean analog AND digital rails help too, but ultimately on-chip crosstalk is what sets the limits. All the junk from ISA signals bouncing around will couple into analog section one way or other on the most integrated designs.

Interesting. So the card I would subject to science is one with a Yamaha YMF719E. I have a datasheet to the YMF715E. It has VREF in and separate analog and digital voltage supply inputs. I think you have seen that datasheet and commented in other threads perhaps. I was poking around on vogons before starting this thread.

There are no voltage regulators on-board. I think it just establishes a +5v plane of sorts straight from the ISA slot, filters it to ground with a couple of 10uf caps, and pumps it into what needs it. I want to say +12 and -12 are used exclusively for the wavetable header, and that is about it.

So let's say you wanted to hot rod it up a bit - would you run three new regulators and pull their supply from +12V (what sort of draw are we talking)? For this card, you could bolt the regs to a perf board and then double side tape that to the back of the card or into an available section (the unused QS1000 section is nice and flat), cut the lines to the VREF, analog power source, and digital power source, and then pipe in each of the three new VRs to the appropriate places?

This card isn't too noisy to begin with... but doing plenty of pre-mod and post-mod recordings/measurements for science might be fun.

There is a voltage regulator on the card, that transistor like thing near gameport which is used for analog side. Now if you added another regulator for digital side, and a regulator for VREFI there will be good improvements. Power draw is something near 100mA at most, heatsink wouldn't be necessary as seen already on the 78L05. Analog side is the most sensitive one though and 7805 will not provide much PSRR so lot of junk from power rails will still pass through. There's also the problem of poor power delivery, decoupling capacitors are not very close to the chip and traces used are thin and long.
Using the QS1000 side would work for digital reg but not analog things, any lenght of wire will diminish the gains very quickly, inductance and EMI are the big killers, regulator would need to be very close to the chip, closest point is pretty much the other side of the card.
+/-12V isn't used at all on the card other than waveblaster header and that should get it as is, though it would be nice to run it though some ferrite beads to reduce noise in the lines as they snake through the analog section.
You can increase value of the WSS DAC loopback capacitors to slightly enhance low freq range and also stick some filter caps on the DAC output to reduce ultrasonic noise, there will be slight improvement on mixer peformance then too.

This card would be lot better option for enhancements as it does more things right the first time : http://www.tmeeco.eu/BitShit/YamahaCards/Yama … %20Wave%202.jpg

CT2940 has best SNR while still having a real OPL3 chip (if you can find one with the OPL3 chip installed).
That card sounds beautiful. But keep the games volume in check because it's prone to clipping. I'm talking from my own experience.

CkRtech wrote:

I am certainly not going to assume Hi-Fi from a mid-90s ISA sound card - SNR was not that great back then.

Oh but you can, you just need the right sound card. Heck, I got an IBM Music Feature Card from 1986 (made by Yamaha), and it is completely quiet. It's just the cheap Asian stuff such as Sound Blasters and their clones that are horrible.
Pro Audio Spectrum is very quiet, Gravis UltraSound is very quiet, and then there's the (semi-)professional stuff from Turtle Beach, Roland etc.
It might be worthwhile to study their designs.
At least some of these cards get 90+db SNR.

The whole stack had to be sufficient for high quality audio. This was during the time of fake wattage multimedia speakers (aka the time when bass was better).

Of course, which explains why there was no serious focus on noise reduction in the low or even mid-range - but even back then some people hooked their sound cards up to HiFi and we certainly can do so now.

I'm constantly struck by how noisy all the old stuff is - but I was spoiled - I went straight from PC speaker to GUS Max in 1995, which sounded great over my Sherwood HiFi set. The difference (in negative terms) when I 'upgraded' to SBLive some 5 years later was noticeable.

dionb wrote:

I'm constantly struck by how noisy all the old stuff is - but I was spoiled - I went straight from PC speaker to GUS Max in 1995, which sounded great over my Sherwood HiFi set. The difference (in negative terms) when I 'upgraded' to SBLive some 5 years later was noticeable.

I was spoiled by having owned a C64 and a Commodore Amiga (which I both connected to a HiFi set as well).
When I got my first sound card for PC, a Sound Blaster Pro 2, I literally went back to the store to return it, because it was so noisy I thought it was broken.
They explained that this is just how it is, I listened to a card they had set up in the shop as a demo, and it had the same problem.

The C64 isn't exactly super-quiet, but it is not as bad as the SB Pro 2, where I could hear every bit going over the ISA bus, through my speakers. The Amiga however, was super-quiet. So, when you go from a computer designed in 1985 to the best Sound Blaster in 1991 (which was even aimed at semi-professional use, since I got the version with MIDI kit and Voyetra Sequencer software), and it is that much more noisy, it's a bit of a shock. Especially since I literally paid as much for the SB Pro 2 as I would for a complete new Amiga 600 at that time (and in many cases, the 4-channel digital sound of the Amiga actually sounded better than the SB Pro 2 which was only used as a poor OPL2 MIDI synth in most games).

The irony of it all is that I got used to the noise as an indicator of how loud my HiFi set was. Before you start a game, you hear noise, especially if you do a dir-listing or such. So when the noise is loud, you'd think: "Hum, I'd better turn it down before I start the game".
Once I upgraded to a GUS MAX, and finally got a silent PC, I often got unpleasant surprises when starting a game, because the volume was left up high, but I didn't notice it, because there was no noise.