CEM3396 ?

[Jaicen]

EDIT: Forget all that!

I just read the datasheet, this looks like a brilliant little mono-synth on a chip man!

I'm up for helping in any way I can, I'm pretty good at PCB layouts if that helps?

[Rowan]

Ok,

I'm back from my hiatus from MB Land. I had to take a well deserved brake from my every day life...........I've spent a few very nice weeks on a beach soaking up sun (and rain).

My chips arrived while I was away so I to have a stash to cook while we try to develop a MB1000, well I think that it will end up better than a Matrix 1000.

So I guess it is down to us now to nut out some ideas on how to get this to work. As the data sheet states the chip runs off CV's and some timing signals for the waveform converter. After looking at the datasheet the suggested configuration uses 4051's to multiplex the CV's........how about using the AIN module in reverse (i.e. running the PIC pins as DAC's rather than ADC's) ? I don't know how this could be done, but 4051 can drive the CV inputs of the chip (as done on the matrix 1000).

Just a hair brained scheme....?

Rowan

[AndrewMartens]

Sample-and-hold is your friend here.  In order to make the necessary number of CVs, you could probably get by with one 12-bit resistor ladder, and a dozen S&H circuits.  I think you should be able to make that with three TL074s (one opamp per S&H), though you'll need an opamp buffer on the output of your R2R ladder as well.  And a bunch of transistors (to gate the signal to the necessary opamp), capacitors, etc.

It's the approach that I'm using on my revised wavetable synth in order to control the analog filter (!) circuitry.  Not as accurate as using a bunch of DACs, but a LOT cheaper, and far easier to acquire parts for.

[Jaicen]

Can I put in a vote here  for a traditional 'knobby' synth?? Ideall something that works like a Juno 60, where you can store patches to buttons, and then when you touch a knob or whatever, it jumps to that value.

I have a real hankering for some real hardware right now!

[stryd_one]

My first thoughts are with the mixing/routing and modulation sections, which obviously would be external to these chips.....

[modularkomplex]

@Ilmenator:

the one that bidded you out this time was me  :o, couldn't resist to get this babies. As you already got 6 of them, I thought it isnt so bad bidding for them.

Mike

[ilmenator]

Yeh, we really should coordinate our activities in some way  ;D

I got an "instant buy" offer from the seller, so if I want I can still buy them. Currently I am bidding on the switches he offers - depending on whether I get them or not I will decide on the CEMs.

Best regards, ilmenator

[modularkomplex]

Sorry for that, ilmenator. Good for my conscience you got that "instant buy" offer.

Mike

[modularkomplex]

I've got my delivery yesterday. Well packed in tubes  :D

Some thoughts about the design:

- We need 2 or 3 ADSR and a LFO (per chip) which could be done by the PIC.

- An additional noise-source would be fine

- I suggest a modular concept like the MB SID: Someone can build just a monosynth or a couple of them (up to 6) for polyphonic one.

- In a polyphonic setup most of the cv could be routed to all 3396s. If I'm right only gain, filter frequency and both timing signals have to be seperat for all 3396.

I have to learn more about sample-and-hold and multiplexing cv. Any suggested links?

Mike

[modularkomplex]

Reading a bit about multiplexing and sample-and-hold I think I've realized how that works.

Attached is a simple flow-chart for one 3396. Could anybody have a look at it? Does that work this way?

Mike

[stryd_one]

I have to learn more about sample-and-hold and multiplexing cv. Any suggested links?

I dunno I'm thinking more of digipots...

[Jaicen]

I'd rather see real pots to be honest.

I think it would be great to homebrew a real analogue style synth. The way I see it working is to not have an LCD, but a number of button pre-sets and banks to select patches. Once a patch is selected, moving a knob will make it active.

Of course, that's just what I want so it's unlikely to be to everyone's taste, especially with the flexibility that encoders allow.

[modularkomplex]

I dunno I'm thinking more of digipots...

So whats the benefit using them? I don't know much about digipots, but have seen only up to 8bit ones.

I'd rather see real pots to be honest.

.

.

.

The way I see it working is to not have an LCD, but a number of button pre-sets and banks to select patches. Once a patch is selected, moving a knob will make it active.

Once you want presets you will have to use AINs. You will need a way to send the cv to the 3396, too.

It makes no real difference using real Pots or Encoders on the input side. The application should be able to handle both of them. The 3396 interfacing is the same.

Mike

[stryd_one]

Well if the circuitry supporting the 3396's is designed to use pots, then an analog interface could quickly be attached to it (just wire up the pots), or digipots could be substituted for a digital interface, which could then use pots or encoders. I do think it's best to choose one method or the project can become too scattered, but this way doesn't create a branching off of the project, but a split, into the synth circuitry around a 3396, and the CS.

I'm going to go read the datasheet again, because I obviously missed the part with the 4051's... I'm all for following the datasheet, I am so clueless at circuitry.

[ilmenator]

You could also take a look at how the Oberheim Matrix-1000 is designed - see here: http://wolzow.mindworks.ee/analog/m1k.htm.

Of course, one would not just build a clone of the Matrix...

Best regards, ilmenator

[AndrewMartens]

Reading a bit about multiplexing and sample-and-hold I think I've realized how that works.

Attached is a simple flow-chart for one 3396. Could anybody have a look at it? Does that work this way?

That's my general understanding of the concept, yes.  Depending on the sample rate you may be able to get away with just using a generic opamp instead of the LM398 (which I believe can hold a voltage for quite a long time).  Also, you may not need the buffer after each S&H amp - but I haven't checked the LM398 datasheet to verify that.

Looking at the Matrix1000 schematics now, it appears that if you have a fast enough sample rate, just a simple R/C network may be enough to maintain the control voltage for the CEM:  http://wolzow.mindworks.ee/analog/files/OberheimMatrix1000Schematics.pdf  Check out page 4, they have the 4051 outputs going to a cluster of R/C networks, each of which is connected directly to the CEM.

UPDATE: Heck, going to the CEM3396 datasheet - which I should have done in the first place - they have the output of the 4051 into a single cap, and then into the CEM.  Even less components!  Hmm, those CEMs are looking like a lot of fun.  Maybe I'll have to buy one somewhere so that it can sit in my parts cabinet for years until my other MIDIbox projects are done...  ;D

[note to self:  this bodes extremely well for my unrelated project which is also planning on using a fast-sampled S&H after the output mux]

[modularkomplex]

Thanks Ilmenator for the informations regarding the Matrix1000. It seems all in all less complicate to be done than assumed.

Thanks Andrew for your comments, too.

Sure, it's not the goal to clone the Matrix1000 but we can learn much from the hardware-design.

Mike

[modularkomplex]

I thought we could use the pwm-outputs from the PIC for the dco timing-signals. But now I've read somewhere that its not possible to generate frequencies below 1khz or so.

Anyone got ideas for a workaround? Clock-dividers?

Mike

[stryd_one]

Maybe take a look at the gameboy sync thread?

[modularkomplex]

Maybe take a look at the gameboy sync thread?

Hmm, don't know if I get it right. We could use a normal I/O for generating the timing signals? Sounds plausible.

Mike

[AndrewMartens]

I thought we could use the pwm-outputs from the PIC for the dco timing-signals. But now I've read somewhere that its not possible to generate frequencies below 1khz or so.

Anyone got ideas for a workaround? Clock-dividers?

Clock dividers would probably work well if you wanted a set-and-forget solution - even a 4-bit counter would work well enough (div by 16) though you could go as high as a 12-bit one (ie 74HC4040) if you really wanted to.  Alternately you could just set a high-priority interrupt on the PIC at the appropriate rate, and just toggle an output pin.  There might be a slight amount of jitter with that approach, though it wouldn't require any extra ICs.

[modularkomplex]

Thank you Andrew for your hints.

I've read a bit more about the PWM-function of the PIC in the datasheet and did some calculations with the given equations. As far as I can see it isn't possible to generate the exact needed frequencies (or multiples) for a equally tempered scale. For example: The closest frequency to an A (440hz) I found which could be generated is 441,38hz. In addition, for a full scale, many different clock-divider settings would be needed.

I the PWM-generation of the PIC really that restricted or do I get something terribly wrong?

So, two possible ways for timing are left. Interrupt toggled pins or an external timer.

I dont know if the pin-toggling method is a good idea. If we want to cover a range of 10 octaves, the highest frequency needed is nearly 16khz. Is this possible without limiting the other tasks of the PIC to much?

Mike

[Jaicen]

I can't claim to be following all this talk of timers, but i'm excited nonetheless!

I'm definitely wanting some nice analogue control for the parameters, rather than the digipots. I can live without patch storage to be honest. Anybody want to sell me a 3396??? ;)

[moebius]

Clock dividers would probably work well if you wanted a set-and-forget solution

You want a set-and-forget solution. This is how DCO synths at that era were made. High speed clock signal divided down to audio frequencies by a 8253 or 8254 programmable counter chips.

This way You'll only need to bit bang counter registers when the note frequency is changing and the uC has time to do such non-trivial tasks as CV generation and multiplexing, midi i/o and maybe user interface handling ect.

For that "analog" drifting feeling CEM could be driven with something as simple as http://www.uni-bonn.de/~uzs159/vco4069.html VCO.

Moebius

[modularkomplex]

I would prefer the programmable counters.

I like the warm touch a small amount of drifting vco make. On the other side, if we want stable operation in - lets say - more than 4-5 octaves, we need a more precise timing. We also can do a slight drift with the software, switchable - for those who do not want it.

Additional to the matrix-design we could use one programmable timer-chip (8253 or 8254) per cem3396 (the Matrix6 / 1000 uses 4 chips for 6 3396). We would get 3 independant square-waves. Two for use with the waveshapers and the third for an additional sub-oscillator.

Mike