8580 Filter roll-off

[PM250B]

I recorded the following SID file from my 8580

filter_res_F.zip

The filter seems to let through significantly more overtones compared to reSIDfp. Could it be that there are substantial differences in the filter roll-off?

[Toddler-Boy]

BTW, I think it's incredibly dangerous to touch this. A LOT of modern(ish) .sid files are created using GoatTracker (and others based on reSID emulation), which would ALL suddenly sound different compared to reSIDfp. Even though with your particular 8580, it would sound more 'correct,' it would ruin the listening experience for everyone using a libsidplayfp-based player. None of the tunes would sound as intended by the original authors.

If any adjustments are to be made, they should be optional and adjustable as a parameter somewhere so the user can decide between the "accurate" and "established" modes.

There is a lot of variety between the 6581. Adding any compensation for this issue would create the same problem for the 8580.

[PM250B]

I fully understand the concerns — especially regarding compatibility and the historical development of many SID tunes created using GoatTracker with the reSID engine. In fact, my own songs are also based on GoatTracker/reSID.

However, reSIDfp comes quite close to the final sound of actual 8580 hardware. In the final mix — when all instruments play together — the difference is often barely noticeable.

If certain sonic behaviors can be reliably reproduced across multiple real 8580 chips, I believe it would be desirable to offer an emulation that reflects this as accurately as possible — ideally as an optional setting.

A selectable mode might therefore be a good compromise, respecting both the historical context and technical precision.

[PM250B]

It might not be the roll-off after all — there could be another cause. I'm sure the experts here will have a better idea of what's going on.
The behavior is the same on my C64C with the 8580 SID (The first recording comes from my c64G): in both, the sound contains significantly more high-frequency content, despite the active filter. (Filter type leak, or these effects are not caused by the SID?)
I’ve also tested several alternative SID emulations besides reSID, but none of them replicate this behavior.

c64c_8580.zip

[mrdudz]

BTW, I think it's incredibly dangerous to touch this. A LOT of modern(ish) .sid files are created using GoatTracker (and others based on reSID emulation), which would ALL suddenly sound different compared to reSIDfp.

IMHO this should not matter - at all. The only reference should be the real chip. If authors did not actually test their tunes on real chips, its their problem (every composer i know does this, no one relies on how the emulator sounds).

[PM250B]

If the phenomenon can actually be reproduced on other 8580 SIDs, it should be emulated in reSIDfp to achieve the authentic analog feel. In my experience, this would not harm or distort existing SID tunes that were created exclusively in emulation — quite the opposite. They would then sound exactly as one would expect from a real SID chip.

Of course, someone would need to have the time and motivation to implement it

[Toddler-Boy]

This is indeed a leak in the filter:

If I change this

libresidfp/src/Filter.h

Line 231 in 3500560

Vmix += solveIntegrators();

to

Vmix += solveIntegrators() + ( Vsum >> 5 );

then the output is extremely close to the provided example.

This would affect both, the 6581 and the 8580, because the filter-code is shared between them. Can anyone test/confirm if the 6581 also leaks like this, or if we need two separate filter-clock functions?

[Toddler-Boy]

Here is the output after the change:

filter_res_F.zip

[Toddler-Boy]

@PM250B could you do me a favor and do the same test on your real 8580 with resonance at 0? Does it leak as well, or does the amount of leakage depend on the resonance?

[Toddler-Boy]

This modification causes problems with digi-playback on the 6581. The digi-volume changes depending on how many filtered voices are playing and at which level.

It seems some DC offset has to be compensated for, but that would be very expensive, as each voice leaking adds a different DC offset.

Maybe @drfiemost has a better idea?

[PM250B]

@PM250B could you do me a favor and do the same test on your real 8580 with resonance at 0? Does it leak as well, or does the amount of leakage depend on the resonance?

At the moment, my C64 is packed away in the basement. I'll put it on my to-do list for the next few days. It will also be a good opportunity to record a few other things.

[drfiemost]

Things are a bit more complex, looking at the schematics all the routing switches are made with transistors, which leaks signal even when open.

1. when a voice is routed to the filter part of the signal still go to the mixer and vice-versa;
2. when a filter is disabled it still leaks to the mixer;
3. when voice 3 is off it is not completely disabled.

Tried to model this in a new branch but still needs some fine tuning, the leakage depends on the transistor size (#22)
As is, it sounds quite bad.

[Toddler-Boy]

How do tunes with samples sound with your implementation? For my much more naive approach, they got really noisy, especially:

/MUSICIANS/M/Mahoney/Stereophonik_6581_2SID.sid
/MUSICIANS/M/Mahoney/Stereophonik_8580_2SID.sid

both sounded weak and noisy.

[drfiemost]

Didn't try yet. I suspect the leak is very low level, possibly negligible, judging by the size of transistors.
Voice3 is likely more noticeable though, it should be easy to measure on a real chip.