GUIDE: Telling apart good Genesis 1s and Genesis 2s from bad ones.

[Eep386]

I should clarify what I was talking about with Ace, by using those 270pF caps. I was referring to the VA2+'s 315-5684's first filter stage there. The VA0~1.8 uses a Sallen-Key topology on the FM unity-gain buffers, which is a bit different from the simple RC low-pass filters and active filters used on the 315-5684.

Here is the schematic for the VA0~1.8's sound mixing/filtration logic:
VA0 Output Stage.jpg

And yes, I can say from first-hand experience that changing the 5600pF caps to 270pF in the VA0~1.8's Sallen-Key filter is NOT a good idea - I always just remove them, that eliminates the ugly squeaky distortion and heavy muffling the poorly thought out Sallen-Key filter induces. At least now I know why removing those caps works as it does.

Also, from personal experience adding caps to the FM inputs (eg. replacing R22 and R24 with 0.1uF caps and replacing R23 and R25 with 20K) only causes MASSIVE problems unless you adjust the rest of the FM buffers accordingly. (Part of the problem is the fact that the 20K total input resistance is simply much too low. 47~51K works much better. I tried 47K input resistance with a 10uF cap, and that seems to work okay.)

The VA2 and VA2.3 are the only models to have LM358-type op-amps as FM pre-amplifiers feeding into the 315-5684, the VA3 and later just interface the FM (via 0.1uF coupling caps) directly into the 315-5684. Sega added 4.7K resistors on the outputs of the LM358 FM pre-amp to bias it into Class-A, to avoid crossover distortion.

[JamesF]

The VA0~1.8 uses a Sallen-Key topology on the FM unity-gain buffers, which is a bit different from the simple RC low-pass filters and active filters used on the 315-5684.

I believe you're wrong there.
VA2+ have exactly the same topology as VA0-VA1.8,, VA2+ has everything integrated into the 315-5684 besides the capacitors which are external.
I can also see additional buffer stage in the 315-5684 before the 10K resistor which is not on the VA0-VA1.8.
The VA0-1.8 lack a coupling capacitor between the analog FM outputs and 10K inputs, in simulation this completely removes the bottom half of the waveform hence creating terrible distortion.

Also, from personal experience adding caps to the FM inputs (eg. replacing R22 and R24 with 0.1uF caps and replacing R23 and R25 with 20K) only causes MASSIVE problems unless you adjust the rest of the FM buffers accordingly.

No no, you should ADD 10uF capacitors before the first 10K resistor, not replace them, the topology should not be changed.
added caps.jpg

If you could please try to add these caps, maybe it will fix the distortion for all the 'bad' consoles.

[Eep386]

Show me where the Sallen-Key topology is on the 315-5684, cause I'm not seeing it at all:
Vref Correction.jpg
315-5684 VA3 application.jpg
This is the amplifier/mixer used by the VA2 and later.

As I understand it, the PSG is not filtered at all by the first and second filter stages, interestingly(!).

Also, adding caps WILL cause distortion, unless you increase the input resistance. Trust me, I've tried it. (Replacing the two 10K resistances with a single 20K resistance achieves mostly the same effect.) 47K+ has no distortion with the 10uF cap added.

[JamesF]

315-5684 with external caps.jpg
Just follow the signal path, it is the same as the discrete components in VA0.
The last 27k+1800pF has the same cutoff frequency as 7.5k+5600pF.

You can't just replace the two 10k resistors with a single 20k or 47k resistor, that completely ruins the 2nd-order lpf topology.
Please try adding a 10uF-100uF cap before each 10k, and report back.

PSG and CD are filtered by the second 1st-order LPF (27k+1800p), just like in VA0.

[Eep386]

I've already tried that on my VA0; it saturated until I increased the resistance to 47K. After you throw out the 5600pF caps at C20 and C22 (therefore removing the Sallen-Key topology), the resistors at R23/R23 and R24/R25 are effectively rendered as simple series resistors. Therefore 10K + 10K = 20K.

Rtotal = R1 + R2 + R3 + ….. Rn etc.
Source: http://www.electronics-tutorials.ws/resistor/res_3.html

Or do you mean, try adding the caps before the two 10K resistors *with* the 5600pF caps in place? I haven't tried _that_ yet. (That would require me to do a little rerouting...)

Regarding the PSG on the VA2+ with the 315-5684, yeah, it's eventually filtered by the output LPF. I think I mentioned that it's not filtered by either of the first two LPFs.

[JamesF]

Or do you mean, try adding the caps before the two 10K resistors *with* the 5600pF caps in place?

Yes.

Thanks for the tremendously basic electronics lesson.

By the way, do you see that VA2+ has exactly the same topology as VA0 but integrated?

[Eep386]

C'mon, man, no need to get fresh here. I'm just trying to figure this stuff out.

I suppose I could solder the 10K resistors at R22/R24 in perpendicular, so that one lead is left floating, so I can simply hook a capacitor between it and the unconnected pad. I don't want to cut traces if I can avoid it.

Regarding the VA2+ with the 315-5684, well, I see some similarities looking VERY closely at it, but then there's the 1200pF caps coupled to 1/2 Vcc that throws me off. Is it just me or do I see some kind of feedback network for a volume control? (The 315-5684 is a so-called 'Sound Network/Phones Amplifier', after all.)

I could try throwing out the 5600pF/4700pF caps between FM LP OUT1/2 and LFB1/2 and seeing (hearing?) what happens. But wouldn't that cause some kind of problem too? It'd be absolutely fantastic if I could just remove those caps entirely and use 2200pF (2.2nF, slightly better than VA2 MD1-level 3.8kHz cut-off) caps at the first LPF stage, as that would cut down the BOM of my own VA2/2.3 sound mods a fair bit. I'll try that on my VA4 Model 2 (the caps in question are at C50 and C51 on the VA4).

[Eep386]

Bleh, what the hell am I doing? I should explain why I went with the 10uF cap + 47K route. It's because I replaced the 1K resistors at R72 and R73 with jumpers, which makes the FM sound *much* louder. But I would have to make sure that I leave those 1Ks alone on the next VA0/1/1.8 I try your idea on. (I'll still have to get another VA0~1.8, as I changed too much on this one to make a really valid comparison.)

But I did try removing the 5600pFs on my VA4 and replacing the 270pFs I put on the first LPF with 1200pF (the closest I have to 2200pF on hand), and, wouldn't you know, it sounds pretty dang good! Though with 1200pFs it has a 6.6 kHz cut-off, so it sounds a fair bit clearer than the Model 1. I haven't noticed any problems with the 5600pF FM filters removed, it sounds like the LPF1/2 stage does a fine job of filtering the FM by itself.

And guess what, that got rid of the clipping I was hearing on this VA4 when I picked up health in Streets of Rage 3!

Anyway, regarding the 315-5684's internal layout, there IS one notable difference I saw right off the bat: the mono amps are tapped BEFORE the stereo output amps on its inputs, not after on its outputs like on the discrete VA0~1.8 circuit. I actually tried replicating that setup, and it got rid of the tinniness on the output sound, and made Mono Out sound much better.

[Eep386]

OK, I'll make a formal apology to JamesF. I just tried his idea after reverting IC9's configuration on my VA0 to the way it was (though I kept the crossover-free TL074 on IC9 along with the 2.5v VRef feeding the op-amps - thank goodness I managed to find the 15Ks I had on R34 and R37!), and wouldn't you know, it works just fine without crunching or distorting. I'll have to make a note of this.

Thank you JamesF for your help!

[JamesF]

Eep386 can you confirm that adding a 10uF cap before the 10k resistors fixed the dreaded VA0-VA1.8 distortion and makes it sound just as good as VA3-VA4?
Is the PSG channel balanced?

Feeding an opamp with lower voltage result in less available headroom for the signal to swing, hence sooner distortion with lower signal.
I think you should put back the 5v to the opamp.

[Eep386]

Well, I can't yet 100% confirm that did fix anything, but I can safely say it didn't hurt the sound output on my VA0, which already has different op-amps than what is usually applied (MC3403 on IC10, TL074 on IC9). But I'll need to get me another VA0~1.8 so I can do a more definitive test. In any case it's actually an improvement over the 47K + 10uF I was using earlier, in that it sounds much the same, but this way I won't have to alter the rest of the sound circuit nearly as much (saves $$ and time!!!)

The PSG sounds fine. In fact after removing those 5600pF FM low-pass filters the PSG sounds like it's filtered almost exactly the same as the FM, which is a BIG improvement.

What I mean by Vref is the 1/2 Vcc going into the positive (non-inverting) inputs of IC9 (except for unity-gain buffers) and IC10. Normally there is 2V being fed into them. 2.5V works better for the TL074 I am using on IC9 and doesn't seem to cause any problems. Both op-amps are still running on 5v Vcc.

Tell you what though, if this fixes the sound on the 'stock' LM324-fitted systems without too much additional rework, I'll be more than happy to save money by not ordering replacement quad op-amps (especially now that I know that even the crummy LM324 can be fixed up to a usable state by biasing its outputs into class-A).

[JamesF]

I guess we should wait for another user with stock Model2 VA0-1.8 or Model1 VA7 to test this.

Removing the 5600p cap on the second 1st-order lpf inverting network disables the filtering and effectively makes it another simple buffer stage.
As a result of removing the last 5600p, CD audio (SL2/SR2) and PSG will not be filtered.

[Ace]

By removing the 5600pF feedback capacitor, you are effectively turning the second-order low-pass filter into a first-order passive RC low-pass filter followed by a voltage follower. The cutoff frequency is still fairly low at 7.958KHz, which will likely still overly muffle the FM output when cascaded down to the final output stage's first-order low-pass filter.
Quite honestly, I would throw out the op-amps, put some with at least a higher slew rate (higher unity gain bandwidth helps too) and just adjust the second-order low-pass filter for a very high cutoff frequency towards the 40-60KHz mark. I find second-order low-pass filters degrade audio quality like downsampling an audio recording from 32-48KHz down towards 16-22.05KHz or so when the cutoff frequency is too low and the audio source this is being applied to doesn't have much towards the upper part of the audio spectrum (ADPCM audio hardware seems to benefit from a second-order low-pass filter the most, in my opinion).

[JamesF]

Ace, applying a 2nd-order LPF with low cutoff frequency on a buzzy synthesizer and very-low bitrate PCM audio has a benefits of more listenable/pleasant to the ear spectrum of sound.
The Genesis audio is not Hi-Fi full range audiophile pristine music player and should not be treated as one;
It is a low quality synth with atrociously low quality and aliased PCM audio, thus lowpass filtering is needed if not essential to make it ear-palatable, from there on it is in fact a personal preference.
Personally I prefer the 3rd-order LPF at higher cutoff frequency of the Model2, over the 1st-order LPF at lower cutoff frequency of the Model1.

[Eep386]

Of course the YM2612 was never meant to be this high-fidelity PCM sample player; the PCM playback capability was obviously tacked on as an afterthought (it doesn't even do any buffering, and there's no immediate facility for DMA playback either) and comes more as a 'free lunch' with what is still primarily an FM music and sound effect synthesizer. It was obviously targeting then-midrange pinball machines, which may play the occasional PCM speech or drum sample but spend most of its time playing sproingy FM music.

I believe if Sega really wanted 100% authentic, arcade quality sound and music, they would have used an eight-voice YM2151 OPM synthesizer (which still handily stomps the FM synthesis capability of any OPN-family chip any day) and a dedicated (AD)PCM chip like the OKI MSM6295. (That said, I'm kinda glad they didn't use a 6295, as it gained a slightly trouble-prone legacy...) But they wanted to be cheap and have both in a single very cheap chip (that's still 'arcade quality' by technicality of it being a pinball music synthesizer), and the YM2612 with all its quality and capability shortcomings fit the bill. Besides, its simple integrated TDM output DAC eliminated the need for an external Y3012/3014 DAC chip.