Music 5000 FPGA

[hoglet]

Hi Tricky,

What would need to be added to the co-pro board to support something like the Music 5000 Synthesiser

There are a few difficulties in doing this:

1. There is a very limited address range available on the Tube FEE0-FEFF I think. The Music 5000 hardware is expected to appear at FCxx and FDxx. That's just not possible through the tube connector. So software changes would be needed.

2. There would need to be an external SPI DAC, and a few (4 I think) signals to drive it.

I'd say this really needs to be a different board that plugs into the 1MHz bus not the tube.

Dave

[tricky]

Sorry, my mistake, I confused the tube with the 1MHz bus.

[jgharston]

A MIDI interface is very simple to build.

[RobC]

A MIDI interface is very simple to build.

I'll second that - and it works really well.

[qUE]

Thinking about what you said about two DACs causing distortion. It might be the DACs output voltages at peak are causing the output signal to double and clipping. There's a method and it depends on the timing capability, I gather it must be one of the defacto standards since I've seen Acorn use it with VIDC, various other soundcard manufacturers use it and pro. music gear manufacturers use it;

You set the update timing of the DAC to samples per second * no. of channels and then run through the channels cyclically. Acorn also mentioned with VIDC something about needing a small silence between the channels otherwise they influenced eachother.

So 2 channels at 44100 would be 88200 samples per second with order of channels as 1,2,1,2,etc.
3 channels at 44100 would be 132300 samples per second with order as 1,2,3,1,2,3,etc.

Hopefully you get my drift. I only understand this from a software point of view rather than hardware side, so I may not be entirely accurate.

[CHRISJJ]

ChrisJJ, I'd really appreciate you looking over this and seeing if there is anything that seems incorrect, or that I have missed. The bit I was least sure about is the operation of the frequency modulation.

Sorry not to have seen this at the time. My feeback is now at https://github.com/hoglet67/Music5000/issues/1 .

[CHRISJJ]

My understanding is that, when the modulate bit is set in the control byte for channel N, the sign bit of the generated waveform governs whether the primary or secondary parameter set is used for channel N+1 (or +2, from what Chris said?). Is that right?

That's correct. The modulation by channel n takes effect on channel n+2 % 16.

Also, from the schematic, it looks very much as if the stereo chopper causes the left stereo channel to use the encoder output of the AM6070D and the right stereo channel to use the decoder. If that's right

That's correct.

the waveforms from the left and right sides would be significantly different, even from the same waveform source. Is that correct?

No. There's a difference, but it is not significant.

[hoglet]

ChrisJJ, I'd really appreciate you looking over this and seeing if there is anything that seems incorrect, or that I have missed. The bit I was least sure about is the operation of the frequency modulation.

Sorry not to have seen this at the time. My feeback is now at https://github.com/hoglet67/Music5000/issues/1 .

Thanks Chris, I've updated the wiki page with your suggestions.

[CHRISJJ]

I gather it must be one of the defacto standards

I'd not heard of it being used prior to the Music 500. Which is not to say no such usage existed. If anyone finds record of one, I'd be interested to hear.

since I've seen Acorn use it with VIDC

Acorn copied it from the Music 500 circuit. You may recall Archimedes pre-release info saying the machine had "an built-in Music 500".

Acorn also mentioned with VIDC something about needing a small silence between the channels otherwise they influenced eachother.

Acorn initially omitted that bit of the Music 500 implementation. In true Acorn style, they didn't understand it so assumed it was inessential, and then put the silicon to production with apparently no listening test. Hence the dire sound quality of the first Archimedes. Intermodulation distortion turned multi-channel sound to mush.

I later heard it and I pointed out to them the error. They told me they had corrected it in a future version. But I suspect they messed that up too, since the next Archimedes model I heard didn't sound much better.

If the quality of Archimedes 8-bit log sound ever did reach Music 500 level, I'd be interested to hear a recording.

[qUE]

I gather it must be one of the defacto standards

I'd not heard of it being used prior to the Music 500. Which is not to say no such usage existed. If anyone finds record of one, I'd be interested to hear.

Ah last time I posted here was a while back, I did some research into what the actual term for it is. It's call TDM (Time Divison Multiplexing), it stems from the Telecom standards much like the (A/Mu/U-law) Log Companding. It's a pretty ancient standard.

since I've seen Acorn use it with VIDC

Acorn copied it from the Music 500 circuit. You may recall Archimedes pre-release info saying the machine had "an built-in Music 500".

I hadn't seen that advert

Acorn also mentioned with VIDC something about needing a small silence between the channels otherwise they influenced eachother.

Acorn initially omitted that bit of the Music 500 implementation. In true Acorn style, they didn't understand it so assumed it was inessential, and then put the silicon to production with apparently no listening test. Hence the dire sound quality of the first Archimedes. Intermodulation distortion turned multi-channel sound to mush.

I later heard it and I pointed out to them the error. They told me they had corrected it in a future version. But I suspect they messed that up too, since the next Archimedes model I heard didn't sound much better.

If the quality of Archimedes 8-bit log sound ever did reach Music 500 level, I'd be interested to hear a recording.

Yeah, the sound isn't to professional standards, but then the Archimedes was somewhat production cost driven. I have an A3000 with the horrible LPF filter bypassed, I can post some samples of it if you wish (!Maestro is probably what Acorn wanted show off), although RISC OS's audio routines aren't exactly silent on the background noise. It's something I rectified in my Music Sequencer (amcs.3rdevent.net) although the Archimedes versions are just PoC and not the destination architecture for the system.

[lazarusr]

Managed to get this working on my Papilio One. It really is brilliant. Thanks Dave.

One thing (one of the many things) I don't understand - why have you used a 23.7832 MHz clock when the original Music 5000 had a 12 MHz oscillator?

[hoglet]

One thing (one of the many things) I don't understand - why have you used a 23.7832 MHz clock when the original Music 5000 had a 12 MHz oscillator?

All of the Music 5000 stuff is running of clk6, which ideally should be exactly 6MHz.

The board I was using, the GODIL GOP, has a 49.152MHz oscillator, which isn't a nice multiple of 6MHz.

The best I could do was:
- use a DCM to multiply this by 15/31 (23.7823MHz) which is as near as I could get to 24MHz.
- use a counter to divide by 4

On a Papilio One, with a 32MHz oscillator, you should be able to get exactly 6MHz.

What sort of DAC are you using?

Dave

[lazarusr]

OK. Yes, it sounds a bit closer to the original with the precise clock.

I am using the same MCP4822 you used.

[BigEd]

The best I could do was:
- use a DCM to multiply this by 15/31 (23.7823MHz) which is as near as I could get to 24MHz.
- use a counter to divide by 4

(Hmm, so 6 = 49.152 * 125/1024 but the spartan 3 clock gadgets only support multipliers and divisors up to 32. But it looks like you are allowed to cascade two DCMs. Otherwise, perhaps you could you export a 25/32 clock, re-import and scale by 5/32? Or use direct digital synthesis, as described here?)

[hoglet]

Hi Ed,

But it looks like you are allowed to cascade two DCMs.

Agreed this is possible, but I don't believe it's actually recommended when using the CLKFX (DFS) output of the first DCM, because this can have sufficient jitter to prevent the second DCM from locking. How much jitters depends on the multiplier and divider settings of the first DCM:
https://www.xilinx.com/support/document ... df#page=59

That said, the only time I've actually had problems getting a DCM to stay locked was when I tried to drive one from Phi2 on the Beeb. It definitely didn't like the 1MHz slow down.

Dave

[BigEd]

Yes, it's evidently something to do with care and only when necessary. But I think your frequency is 1% off, which is I think noticeable to a musician.

I did see a note that the second DCM should have the reset input fed by the LOCK output of the first.

[hoglet]

But I think your frequency is 1% off, which is I think noticeable to a musician.

I will give this a try I think.

I do need to update this design anyway, as I recently found a much more straightforward way of including DCMs than using Xilinx's Core Gen "DCM Wizard" thingy. You can just instantiate them directly:
https://github.com/hoglet67/opc/blob/co ... opro.v#L85

I did find one place I have used cascaded DCMs, so you are right it can work:
https://github.com/hoglet67/AtomVGAWing ... g.vhd#L179

Dave

[jonb]

I've only just got my hands on a 5000. Haven't had time to even connect it when I stumble cross this thread.

Kicking myself now!

Nice one Dave!

[Elminster]

I can kick you in a week or so if you like

Looks like something to add to the diy part of the hardware list.

[jonb]

I can kick you in a week or so if you like

No ta...

I'm going to be pretty shattered anyway. Just found ANOTHER defect with my car, so I have to swing by the garage on the way to yours. Probably have to borrow one of their old bangers

[Elminster]

I can kick you in a week or so if you like

No ta...

I'm going to be pretty shattered anyway. Just found ANOTHER defect with my car, so I have to swing by the garage on the way to yours. Probably have to borrow one of their old bangers

Pick up a sinclear c5 for that extra retro flare

[jonb]

Pick up a Sinclair C5 for that extra retro flare

No, ta...

(There's an echo in here!)

I used to have one BITD. Useless thing. Ended up selling it to a "retro collector" for a sizeable chunk of cash, so it panned out well in the end.

[lazarusr]

Managed to get this working on my Papilio One.

OK, don't try this at home. The Papilio One is not 5V tolerant so you could damage it. Luckily, mine seems OK.