I spent the afternoon today with Dave/hoglet, digging through some mysteries in the Music 5000.

It struck me as a very interesting bit of hardware - not that we have one, although there are some around, instead we were looking at and listening to Dave's FPGA model, the published schematics for the Issue 1 board, and the software model for this synthesiser in the B-em BBC Micro emulator.

ETI article about the instrument, by Chris Jordan who designed it:






Some highlights:
- high quality phase-accumulating wavetable synthesiser, 16 note polyphonic.
- low part count, all TTL with two RAMs, plus a DAC.
- very clever logical design: free-running 8-beat state machine at 6MHz producing 47kHz samples.
- two-nibble adder does the 24-bit phase accumulation and 8-bit gain control (in the logarithmic domain)
- companding DAC converts from 8 bits of sign+logarithm to audio output, giving 12 bits of output quality.
- very nice schematic.
- it's an 8-bit peripheral, occupying 257 addresses in the 1MHz bus area of the Beeb's memory map - one address is 'owned' and the other 256 can be shared with other peripherals.
- as such, it could attach to many 6502 systems or other 8-bit micros.
- the interface is memory-mapped, with a command byte which maps in one page of a 2k RAM.
- the RAM is write-only for the host, with a single config page and the rest to write the 14 wavetables.

Much more info in Dave's wiki here.



Thanks to Chris Jordan for encouraging the investigations.

Oh, a couple of videos demoing the Music 5000:
https://www.youtube.com/watch?v=snjSIG7bb5U
https://www.youtube.com/watch?v=3S54O2ecRAY

Just to add: the 16 channels are added by being time-multiplexed into the DAC, which is driven at 16x the advertised sample rate.

And, the stereo panning is done by pulse-width modulating whether a sample goes into the DAC or whether a zero goes in, on successive beats of the 6MHz clock.

And, the companding DAC/ADC is used to drive stereo by using the ADC mode as a second DAC channel - it's meant to be used as a binary-search ADC.

That's quite cool! Sounds like a cross between the Fairlight's synthesizer facilities and Konami's SCC chip. I'm curious how the noise is being produced in those demos, since the wiki mentions nothing about a noise generator - is it using heavy FM to create a pseudo-white noise sound?

I think channel modulation is heavily used - although there are 16 phase accumulators, they appear to be presented, mostly, as 8 voices, so the software and presets seem generally to use paired oscillators. Either they are used in tandem to produce a richer tone, or one will modulate the other.

I haven't fully understood what the modulation can and can't do - it's not FM in the Yamaha sense, as far as I can tell, it uses only the sign bit of the modulating channel to swap the target channel between one configuration and another. That could be as often as many times a cycle or as seldom as some tens of cycles. (At even lower modulation rates, it could be done by the controlling software, which is how envelopes are managed.)

Or, possibly, I've misunderstood what the hardware can do... page 70 of the manual implies that the FM mode can do a lot.




(When it says "available", that's a combination of what the hardware can do, and what the included software offers.)



At least one of the samples used is a noise waveform.

The user manual (here) talks about an RM mode - ring modulation. But there is no true multiplication possible: what there is, is the ability to use the sign bit of the controlling channel to invert the controlled channel. I suppose it must be that this is a musically useful approximation.





The software is mostly controlled in text, it's almost like a Forth, the manual speaks of defining words.

You are right, there is nothing in the hardware like a PRBS based noise generator.

But one of the "standard" waveforms looks very noise-like:

The full set of "standard" waveforms is on the wiki:
https://github.com/hoglet67/Music5000/wiki

Dave

Have you any idea Dave, what the system is doing to offer a variable-depth frequency modulation? I can't see that the hardware is any help, unless one channel's output value can be added into the low byte of the next channel's phase update - and even then, I can't see how the depth could vary. Software could manage FM at some rate, but I don't know if it could keep up at an audio rate, and I don't see any way for the software to synchronise phase with the hardware.

I think it must be all done in software.

Take a look at the section in the Music 5000 User Manual titled "Making Instruments":
http://www.retro-kit.co.uk/user/custom/ ... df#page=59

The notion of pitch and amplitude envelopes has to be 100% software.

One think I noticed yesterday when I was gathering that peak level data is that the values were varied each time you played a track.
I think this is possibly because it doesn't seem possible to write to the phase accumulators from the host. Or if it is, I haven't discovered how to do this. They don't seem to appear in the memory map.

There doesn't seem to be a need for synchronization between the hardware and the host. I think because the pitch and amplitude envelopes will be many 10s or 100s of cycles long.

Dave

It might be interesting to see what an FM sound looks like from this device: if, say, a sawtooth was chosen for both the channels, we could maybe read off what's happening.

I can see FM at the millisecond level working by software. (If it is software, does it need the controlling channel at all??)

Yeah, that sounds a lot more like proper FM (or phase modulation, which works out to the same thing in the end.) Though I'm curious as to what that switching mode thing is like as well.


Yeah, I saw that in hoglet's post as well. What gets me is that, at only 128 samples long, the noise should sound a lot more periodic (for those familiar with NES music, "periodic noise" sounds more metallic than white-noise-ey - think the Metal Blade sound effect in Megaman 2.) Unless they're running it at a much lower rate...?


This is what the SID does for its ring-modulation as well. Reminds me of a lot of non-modular analog synthesizers, where instead of a proper ring modulator they cheaped out with a XOR gate connected to the square-wave outputs of the two oscillators because with rectangular pulse waveforms that amounts to the same thing


I'm not 100% sure from a brief skim of the manual, but I'd bet that what they're doing is phase modulation. Essentially, the output of the modulator (after amplification/attenuation to allow for variable depth) gets used as an additional offset into the carrier's wavetable, causing the index (which should normally be a plain linear ramp) to cycle back and forward relative to the "true" index as the value of the modulator changes. Yamaha's FM synthesizers actually do this as well, because it's simpler and gives the same end output.

Belatedly...


Software (the M5 module) computes high and low frequencies and delivers them to the hardware's modulated channel, then hardware's modulating channel selects between high and low.


This results from one or more of that track's instrument(s) not setting the phase - with PHSET.


They may be written to zero by a channel control register bit and this is used by PHSET.


There is very much a musical need for phase-envelope sync. See PHSET documentation.

It does. Is sounds white-noisey only when modulated.


See my answers in previous reply.


No. Frequency modulation, 1-bit.


That was not affordable in 1984 due to requiring a digital multiplier.


Yamaha synths of the time used only sine waves, allowing multiplication by cheap addition through a trig identity.

Chris (Music 5000 designer)

Welcome Chris!

Thanks for filling in some of the blanks.


I'm wondering if I have missed implenting this capability.

It's quite a while since I looked at this design, but I did try to write a wiki page detailing the register level interface. I don't recall any kind of a "force reset" bit in a channel's control register:
https://github.com/hoglet67/Music5000/w ... l-register

Do you recall what path in the TTL schematic was responsible for clearing a phase accumulator?

Dave

Sorry, my error. Phase set was not in the control register, but in Freq Bit 0, as your wiki correctly states.

IIRC this is not a reset (to zero) but set (to the "freq" value) though the M5 module only ever used zero.

I'd planned to use this for e.g. timbral modulation using what at the time I dubbed wave sliding, but this was precluded by pressure on host CPU and memory. Still it would be fun to implement in an updated M5.

BTW:

* "&40 unused - process access timeslot" should read "&40 unused - processor access timeslot", though since I call the Music 5000 main block a processor, perhaps (here and later) "host bus access timeslot" would be better.

* "Bit 0 Channel Disable" should read "Bit 0 Phase set", and later "Bit 0 of Frequency Lo must be zero, or the channel is disabled." likewise corrected.

* "can be used to implement a) oscillator sync (by hardware modulation) and b) gate sync, for sharp attacks (by software in the ON GATE code of the M5 module)."

should read

"is used..." I think.

* "Bit 5 Modulate Adjacent Channel" should read next channel, modulo 8 (where "next" is in register order, not the difference processing order).

Well done for this analysis!

Thanks Chris - I've made those suggested changes

Dave

PS Another option as yet unused is modulating modulation. I.e. channel 1 A+B modulation ON, channel 2 A modulation ON, B OFF, channel 3. With sync this should again provide new wave-shaping - for harmonic timbral control.