Jump to content

CEM 3372 "Xpander" VCF/VCA for MB SID


gavgomad

Recommended Posts

I had intended to get more done on this about six months ago, but my feeble skills on layout and lack of time have kept me at bay....

....BUT, now that I am slowly moving forward on my 6582 project (tentatively called the 6582X for "Xpanded"  ;) ), I want to start getting things moving again - still busy, but need to push myself a little!! :)

I seem to recall from my travels that a few others were looking to put some Xpander filters in their boxes, and was curious how many of us are out there? Perhaps with a bit of collaboration on this one could get things done faster, and in the end perhaps we can do a small run of quality boards (if we don't just home brew them)?

Anyway, this post has two goals - 1) found out who's out there and interested in working on this (I recall quite a few, like me, already have 3372s....) and 2) start some discussion as to the basic design of the PCB.

To get things started, I have been looking at some other implementations of this filter (there was an EFM variant a long time ago, but I don't believe it is verified?), and it looks as though the 3372 can be run on +5/-12V (whereas the Xpander, as I recall, ran on -5/+12V). It seems to me that it should be quite easy to clone the basic Xpander circuit as noted in the service manual, compensate for the voltage change, and maybe put two or four 3372s on each board, with extra switching for true bypass of the 3372?

I think it might be worth the extra $$$ to design around the Maxim line of switches as opposed to the good ol' 4051/4053 combo, as they are capable of handling a full +/-12V supply and apparently "pop" less when switched - which would be nice if someone decides to take TK up on his suggestion of wave-tabling the mode switch states, and modulating through them on the fly....

So to quote Pink Floyd, Is There Anybody Out There?  ;)

Gav.

Link to comment
Share on other sites

So to quote Pink Floyd, Is There Anybody Out There?  ;)

Gav.

Yes! I have 8 3372's at home waiting to for just this project. I was really into doing it about half way through my 6582 but then I burnt out a little. But I'm feeling ready now. I don't know much about layout at all, but would be happy to help if I can. I have done a little research on the original xpander filter. I'll dig around my files and see what I can come up with.

Luke

Link to comment
Share on other sites

Heh, I thought I didn't have any of these chips, turns out I have 12 :D

I wouldn't want the switching circuit on-board though... IMO that is a separate job and would be better on a separate board.

Otherwise I'm interested :)

Link to comment
Share on other sites

I wouldn't want the switching circuit on-board though... IMO that is a separate job and would be better on a separate board.

Otherwise I'm interested :)

Welcome aboard our madness stryd! (Although I hope this won't pull you away from the switcher project!  ;) )

Do you mean moving all of the switching circuitry, including the mode switches and buffers necessary to keep the cores at optimum level? I guess the concern on the mode switches is how long the lines can be without affecting the tone adversely....

Marc Bareille has done an excellent job with his Extrapole board - http://m.bareille.free.fr/modular1/vcf_multimode/extrapole.htm), although I think that for our purposes we need something more tailored to the CEMs (ie. includes the core buffers etc.), and by pairing two circuits on one board (which is then closely mated to a dual 3372 board) we should also save some space and $$$.

It could be that there would be a broader interest in a multimode switch board with true bypass switch than a dedicated 3372 board.... I'm not a guru, so someone with more EE background may be able to add more here, but the 3378/3379 that others are working with in the MB domain have a similar basic core structure to the 3372, although I expect the maths would be different (resulting in different cap and resistor values). No guarantees there though, and as I have 3372s I'm not too interested in spinning too much on that end of things. I don't believe it will be of any use to SSM2044 folks, as the access to the cores isn't brought out in quite the same way, as I understand....

If we can do it without sacrificing stability and quality, it would certainly be easier to design a separate CEM part of the PCB (which can either be a standalone 4P LP or, with jumpers, connected to DIL connectors to the switch board), then graft on a PCB for the switches and resistor networks. We'd need a tight mating of the boards though....

Do the more "EE" guys see a prob with separating the switch circuitry like this? Precautions we need to take?

G

Link to comment
Share on other sites

Hey Stryd!

(Hey Luke as well! Forgot to say hey in my last!  ;) )

I actually was planning on doing a nice generic switch board for those kinds of duties. For example, you could wire it as a DPDT to select between ground and an input for a SID in, or more interesting to me for routing purposes, either true bypass of the VCF/VCA or routing each SID pair to either a main L/R or alt L/R output, so that there are two post processing options, selectable for each voice pair (can't really see needing more on the fly!).

Well, once I figure out the final circuit we're going to need (ie. whether the "EFM-ized" version actually works with its altered supply, input structures and cv controls etc.), hopefully I can take a crack at a PCB (bear with me folks, it'll be my first and I'm going to be learning Eagle along the way and looking for comments! ;) ) I think I'll try and give it a whirl this coming weekend - will of course start brushing up on eagle in advance, maybe with an attempt to layout a simple switch board as above using the Maxim chips and running on +/- 12V? Might make a useful contribution to the group as well!  ;D

Will post a couple specific design Q's shortly. I had asked around on electro-music for some tips, but didn't get too far in respect of some things, so it'll be a repro of that post (about a month or so old).

Keep trucking on the switcher, stryd, and best of luck - that baby is going to be the centre of my studio and live rig when it's done!!  ;D

Gav.

Link to comment
Share on other sites

OK, so here is my previously posted (in electro-music forums) surface comparison of the original Xpander circuit, and the "EFM-ized" version. If all goes well, the page from the Xpander Service Manual, and a "cut-and-paste" version of the EFM version should be attached at the end (they're not too large - my first attachments here, so hope I'm within the rules!).

Here are the differences I noted, and what (in all my naive glory!) I think it means - EE folks, please stand up and help me out here! ;) :

Audio Inputs

EFM inputs have 1k resistor pulling to ground after the 2.2uF electro, whereas the Xpander only has 680 ohms to ground. Not sure how critical this would be, and may be a "convenience" choice by TomG, as I'm sure he had far more 1k's around....

VCA CV Input

EFM has 10k resistor on the input pin and 0.1uF cap to ground. Xpander has an 18k resistor instead, with same 0.1 uF cap to ground. Presumably all this would do is alter the sensitivity of the input, no? The EFM would be a tad more sensitive?

VCF FM CV Input

BOTH have FM CV going through 39k resistor, and Cutoff CV through 38.3k to input pin. EFM also feeds the input with apparently -12v through a 196k resistor to input, with 1k resistor to ground. Xpander does the same, but with -5v.

This is a big one for me.... Why the difference in voltage, or is this an error in the EFM schem? Frankly, -12V will sit better with my supply design in this instance (as my supply is based on the NorthernLightX design). It seems like this network is some sort of biasing of the CV inputs....

Also, a general Q - if only 1 input is needed from DAC, what should the CV input config look like like? I assume the -5V/196k and 1k to ground network would stay, but are we safe just pinning a single input with a 38.3k from the DAC? Seems like a simple summing job here, but we don't want to destabilize anything....  

VCA Output

Both feed into inverting op ap stage, non-inv input grounded, with 2.2uF on out. But EFM has a 33k resistor in feedback loop, whereas Xpander has 10k resistor in feedback loop. Guessing the Xpander value should trump, as the balance is critical as the gain cells are mixed, no? Maybe this also has something to do with the re-biasing of the input stage and/or supply changes....

VCA Input

Both feed through 2.2uF cap into a resistor before input pin, but the EFM has a resistor value of 33k and the Xpander has a 10k. Will this perhaps compensate for the difference noted on the VCA output above? Would seem that it is doing exactly the same thing as the VCA output op-amp....

Resonance Input

Both take feed right after 2.2uF cap from the VCF output. From there, the EFM feeds through a 20k resistor to input, with 1.2k resistor pulling to ground. The Xpander feeds through a 27k resistor to input, with the same 1.2k to ground.

Gain Cells 1,2,3

Generally, the EFM uses inverting inputs on all op amp buffers from the cells, grounding the non-inverting input via 1k resistors. The Xpander on the other hand uses non-inverting inputs, and grounds inverted inputs via 1k resistors. The exception is the second stage buffer from Cell 2 on the Xpander which, as with the EFM, uses the inverted input. Again, to perform the cell mixing, is this unusual inversion not going to throw off the magic?

Anyway, food for thought while we ponder this one!

Gav.

3003_EFM3372MM_jpgece34eebd9b528b065033f

3005_27_xpander_service_manual_copy_jpg1

Link to comment
Share on other sites

  • 2 weeks later...

I'd been thinking about this too, but I quit it because I had no idea how to switch the filters with the midibox software.

Heheh.. I've had this thread open ever since the 1st post, waiting for me to get time to study the schematics....Because I know how to switch the filters with the midibox, but the filter itself I have no idea about ;)

Link to comment
Share on other sites

Hey all!

I actually HAVE been pondering this some more in my absence from the boards.... REALLY!  ;D

Frankly, I appreciate the bump though.... Unless its for the day job, I can REALLY procrastinate!  ::)

I'd been thinking about this too, but I quit it because I had no idea how to switch the filters with the midibox software. Would be cool though.

As stryd_one indicated, the switching in the software isn't too bad.... 4 bits needed, so 4 external switches ought to do (nicely provided for in the SID patch architecture).... For my purposes, I'm more than happy to switch in pairs, so that only 4 of the 8 external switches are used per SID engine, leaving 4 per engine for other things (gates, output or effects switching, etc.). TK already offered the sweet idea of creating a wavetable to "count the bits" so that you can arrange the modes in any order you want and, if desired, modulate through the sweep (recalling that you are still going to be hard switching between modes - no interpolation here!).

It would be nice to have a dedicated parameter in the i/o to give visual feedback of filter mode though (save a LOT on getting individual 7-seg displays and counters to let you know what's going on in there....)

As for the filter, I think I really need to experiment a bit with the filter sans the switching circuitry first. I was actually re-considering whether or not to do this in two parts - one PCB with a basic buffered 3372 4-Pole Lowpass/VCA, but use SIL headers with jumpers to configure it so that removing the jumpers opens the circuit, and with a few connections, hook up to a dual switching daughterboard with the CMOS switches, extra stage buffers and resistor networks. It seems that this should be doable in light of the daughterboard used for the MOTM 440 filter....

I first thought that it would be better all on one board, but as I need to "suss-out" the 3372 a bit without that circuitry anyway, and there may be some folk simply looking for yer basic SSM2044 style circuit (potentially more compact board as no linearization is needed!), this approach might be best. Also, you can simply upgrade things at a later date by adding the switch board! I'm open for comments and cautions here....  ;)

Certainly haven't decided one way or other on this one yet!

My first concern on the filter is the supply and biasing (the 3372 circuit seems to want to see some -V at the Cutoff Freq CV In).... The Xpander filter (and the datasheet example) uses -5V/+12V. It would be MUCH easier to run this on -12V/+5v, but I don't know how safe this is, and it WILL change the CV input scheme (unlike the EFM Xpander example - I think -12V, even through the 196k etc., is still not good anywhere near an input which is looking for 0V to +5V....)

That being said, I don't know I'd want that +5V rail near the 3372s, as there is bound to be a LOT of potential for digital hash on that line? The 3372 apparently can take 25V total supply swing. Maybe -12V and +9V? Keeps us in a more "safe" zone, and on a less noisy rail (I hope? How much hash is there in the SID supply lines?)

Anyway, I'm off to dig up some other 3372 designs to see how they were done.... Maybe some other designs will give me a bit more insight.... Will report back soon!  ;)

Gav

Link to comment
Share on other sites

I was actually re-considering whether or not to do this in two parts....

I'm open for comments and cautions here.... 

....

How much hash is there in the SID supply lines?

I'd be using mine as part of a filterbank and not restricted to the SID, so I like the separate boards thing.

Link to comment
Share on other sites

I'd be using mine as part of a filterbank and not restricted to the SID, so I like the separate boards thing.

Well, maybe some form factor discussion is in order. This will be a long one, but hopefully will spark some discussion on where we want to go....

I am definitely a fan of doing a two-board system. Board A = Dual 3372 VCF/VCA; Board B = Dual Multimode Switch Board. I think I may opt to put the cell buffers for the Xpander mode on the 3372 board, to be jumpered in and out, that way we have a buffered signal before it travels along any wires to the switch board.

Let's discuss Board A first (I think Board B will come naturally, as really its a direct copy of the 4051/4053 circuit from the Xpander service manual).

The 3372 is equipped with the following:

  • 2 Separate Pre-VCF Audio Inputs w. CV Control of Input Volume
  • VCF w. VC Resonance (Resonance is not exponential, so no linearization required!)
  • Final VCA w. CV Level Control

For switching the audio routing (ie. SID thru vs. 3372), a similar relay circuit to that seppoman used on the 2044 board would suffice.

I also think that it would be beneficial (and easier on the design side!) to, as seppo did with the 2044 board, make use of the scaling options on the AOUT NG, which this will probably link up to, for fine tuning and offset adjustments. This will avoid the necessity of trimmers for each of the CV functions on board the 3372 PCB.

MIXER INPUTS

Even though I don't have any particular intention of using more than one input or the volume control, it makes sense that we provide facility to use the other input and volume controls. For the audio inputs, perhaps a jumper to ground where you're not going to use the input. For the CV Mixer controls, perhaps a jumper which can be installed either to ground (to silence the input), or to app. 4.88v to use full volume (makes sense if you're strapping on to the SID for example), and if the jumper is not installed, the middle pin is your connection point for a variable CV?

Standard setup for the SID setup would be, for example, SID output (via switch or relay) to Input A. Input B jumpered to ground. Mix CV A jumpered to 4.88v (essentially full-on), Mix CV B jumpered to ground.

VCF

I have already discussed the bizarre "biasing" thing going on with the CV input. If anyone sees anything wrong with the following let me know.

It appears that all that is happening at the front of the Xpander circuit is a slight biasing of the input voltage by -.0253807.... volts by way of a resistive divider (196k and 1k for -5V). To keep this same figure with a -12V feed, we'd need to up the 196k to a 475k (if my maths are correct?) to give us something close (exact value is 471.8k).

Other than this, there is a simple divider on the front end (38.3k, 1%), using the same 1k resistor to ground. This maxes a +5V input out at 4.872V. We can keep the same value, on the presumption that the Xpander feeds a 0-5V control signal, and that this is what we will be adjusting our AOUTs to.

Resonance seems to be a standard 0-5V input, and seems wired direct to the DAC on the Xpander.

VCA

Although the VCA is independently accessible (it's own I/O), I don't see much point to adding a separate relay for accessing the VCA alone.... Personally, I don't see that it's worth the effort.

In any event, there will need to be (as with the cells - jumper = connection to the on-board filter caps, no jumper = pin to the requisite buffer and on to the switch board) a jumper here to either route the VCF in, or the Switch Matrix output....

There is a little bypass network on the Xpander circuit for the VCA level CV, so for continuity I say we follow that.

Did I miss anything???  ;)

Gav.

Link to comment
Share on other sites

Glad to see this is still happening. I wish I could offer some knowledge here, but this is pretty much outta my league, so I hope to learn a lot.

Hi Luke.... Slowly but surely! It's actually out of MY league, but I'm willing to press on and see how far we can get before we are stumped!  :D

Gav.

Link to comment
Share on other sites

I've got to let this sink in, but got no time tonight.

But I do know that the 3372 is also used in the prophet 600 and the rhodes chroma polaris. I attached the filter schematics.

Thanks much!

After reviewing a few different applications of the 3372, and the datasheet, I think that there is just no way around having a -5V supply for this filter. I still need to compile my thoughts on my own PSU for my 6582, but maybe modifying the design from northernlightx to add a 7905 as well....

In reviewing the minimum requirements for the positive supply, it has to be +9.5 or more. That rules out a -12 to +9v supply, and I think -12 to +12 is just a little too close to the 25v rail to rail incineration point of the 3372. Accordingly, must supply a -5V rail to the board.

This will eliminate some other problems though with value subs.  ;)

Now, I just need some practice with Eagle.... ;)

Gav.

Link to comment
Share on other sites

  • 2 weeks later...

Hello all,

I hope I'm not late to the party spoiling something. I just finished a filter using a stereo pair of CEM3379s and couldn't help but noticing the similarities in biasing. The VCF CV is a small signal input so all that is done using the resistor network is some biasing. For instance on the 3379/3389 the VCF input needs to stay in between -155 to +110 mV for a CV input of 0 to 5V before the resistor divider (if we go by the data sheet looking for a control range of 14 octaves). The CEM3372 has exactly the same recommended signal range and its' data sheet shows exactly the same resistor divider of 20K with 33K to Vee and 1K to GND.

I don't have my lab notes in front of me so please bear with my lack of maths at this time; but circuit analysis of this type of divider can be done using the fact that the sum of all current at any point in the circuit is zero - current into the point equals current out from the point. This means that current in to the VCF CV pin on the chip will give a voltage of around -150mV when the CV signal before the 20K resistor is 0V. Solving this for a +5V CV we get close to +110mV. This is the old manufacturer's data sheets so far. What has been done by Oberheim and others is probably just a re-biasing to fit their chosen CV range.

Hit me back with questions if this makes no sense so that I can retrieve my lab notes where the formulas exist.

How about the -5V needed? Couldn't there be a local regulator on your board to generate this?

Link to comment
Share on other sites

Hey Olga42! Sorry for the tardy response to your helpful post! ;-)

I hope I'm not late to the party spoiling something.

Never! We need all the help we can get! ;-)

I just finished a filter using a stereo pair of CEM3379s and couldn't help but noticing the similarities in biasing. The VCF CV is a small signal input so all that is done using the resistor network is some biasing. For instance on the 3379/3389 the VCF input needs to stay in between -155 to +110 mV for a CV input of 0 to 5V before the resistor divider (if we go by the data sheet looking for a control range of 14 octaves). The CEM3372 has exactly the same recommended signal range and its' data sheet shows exactly the same resistor divider of 20K with 33K to Vee and 1K to GND.

So theoretically, then, and in light of what others seem to be doing with the 3378s and 3379s, maybe the most sensible solution is follow the datasheet in this regard? Use the recommended rails and biasing from the datasheet, and just incorporate the switching and buffering circuits further down the line....

I don't have my lab notes in front of me so please bear with my lack of maths at this time; but circuit analysis of this type of divider can be done using the fact that the sum of all current at any point in the circuit is zero - current into the point equals current out from the point. This means that current in to the VCF CV pin on the chip will give a voltage of around -150mV when the CV signal before the 20K resistor is 0V. Solving this for a +5V CV we get close to +110mV. This is the old manufacturer's data sheets so far. What has been done by Oberheim and others is probably just a re-biasing to fit their chosen CV range.

My theory is horrid.... But a query - what would the real difference be between the OB biasing and the one on the datasheet? Is the OB circuit pumping out something more than 0-5v? It seems that careful setting of the maximums on an AOUT_NG board prior to connection ought to make more sense than scrambling around with resistor values for rebiasing purposes....

Hit me back with questions if this makes no sense so that I can retrieve my lab notes where the formulas exist.

Is there any way you can show the formulas? I'd like to plug in those values from the OB and see what the difference is.... I wonder whether or not their DAC is 0-10v or something, and accordingly they need a steeper divider....

How about the -5V needed? Couldn't there be a local regulator on your board to generate this?

I think this is the way I'm going to go now, and hang it off the -12v rail, which probably isn't under heavy strain anyway. ;-)

I was just looking at Rio's 3379 schem and board.... Maybe I'm making this too hard.... Follow the datasheet

until we get to the switches and buffers, and rely on the AOUT_NG trimmers and the AOUT setup routines in the software to set the right CV levels.... ;-)

Been bogged down at work again, but as I've had a bit more success with the ol' Eagle program, I think I may try again at this on the weekend.... Starting to find the PCB work kind of fun! ;-P

Thanks again olga42! Will try and make my next post in this thread a little sooner than my last.... Hopefully I'll have an ugly design for the pundits to comment on! ;-P

Gav.

Link to comment
Share on other sites

Hello again, sorry for the delayed reply. There was some holiday activities involved. Anyway, back to the theory:

I remembered Kirchhoff's Current Law: The sum of all currents, into and out of a point, is zero since we cannot store current in an ordinary junction point or a node if you will.

So, I ran Mathematica using the formula:

[tt]Solve[(0-v)/20+-v/1+(-5-v)/33==0, v][/tt] in order to find the voltage v at the filter input pin 2 on the CEM chip. This is for the case where the applied filter CV is zero volts.

This is assuming that we stick with the resistors used in the data sheet (20K, 1K, 33K). Please try this for the OB values and you'll see what they most likely use in terms of voltage range.

How did I arrive at the formula?  ???

Assume that the filter CV voltage is v and that the applied CV is 0 volts. Observing the Kirchoff law we get:

(0V-v)/20kOhm +-v/1kOhm+(-5V-v)/33kOhm = 0  :o

Also, I used Ohm's law (V = R*I) => I=V/R in order to get an expression for the currents. Sometimes V is called U, E or emf depending on your application.

Cheers!

Link to comment
Share on other sites

Cool, I love a bit of math.

So, I ran Mathematica using the formula:

Solve[(0-v)/20+-v/1+(-5-v)/33==0, v]

I never knew such a thing existed, my toolset is usually a pencil, paper, and a basic calculator. I imagine I take somewhat longer than you to do this stuff :D

Link to comment
Share on other sites

Hehe, the Mathematica thing was just me being too lazy to actually juggle the equation in order to get a tidy expression in the form v= numerator/denominator. I'm no EE by educashun so I usually measure what I calculate before connecting any CEM/SSM or expensive hardware just to be safe  :-[

Link to comment
Share on other sites

Thanks guys! Now, to plug in the numbers and figure what it means....

Firstly, the variation for the input voltage seems to be in the first part of the equation (first brackets - 0-v) correct? Accordingly switching between 0 and 5 here should give us the respective output at 0 and 5v respectively.

My math sucks, and I too am not an EE, so bear with me.... My results might look a little funny....

For the 20k/33k as in the datasheet, if I set the output to "0V" as in your equation, I get v=-.14025 (close to the -150 you were referring to?). With "5V" at the output, I get v=.0911641, which is a little more distant from the upper figure of 110....

Now, for the OB (38.3k/196k/), I get v=-.0247381 at 0v, and v=.101859 at 5v.... Seems that they have "biased" the filter a little more open, or are my maths completely rotten? OBs have always been capable of some pretty bright buzzy sounds, as well as dark deep sounds....

Does this sound right? Which is preferable to others?

Sorry for my thickness.... I'll get there in the end! ;-)

Gav

Link to comment
Share on other sites

Hi Gav! You're about to get there but there is some speculation about the OB that doesn't ring true I believe.

If you take a peek at either the 3378 reduced size data sheet at http://www.synthtech.com/cem/c3378pdf.pdf or at the CEM3389 data at http://www.synthtech.com/cem/c3389pdf.pdf page 4 it says that the upper limit of about 90mV yields an upper frequency limit of 20kHz. Also, it states that the lower limit of 5Hz sits at -150mV and that the frequency control scale is 17.5(min), 19.0(typ), 20.58(max) mV/octave. So, -140mV is still resulting in a lower frequency than the 20Hz you can hear. Low enough to rattle things I have heard  ;D

But, yes - your numbers are correct. I didn't have my notes near me so I guesstimated the +110mV in my first post - sorry for that. Since you did the math and arrived at the correct numbers (I checked them vs my notes and did the calculations again!) I could add insult to injury by writing down how I manually shortened the equations so that I am really thorough:

Freq CV = 0v: 0-v/20 - v/1 +(-5-v)/33 = 0

-v/20 -v +(-5-v)/33=0

-v/20+(-5-v)/33 = v

-5-v/33=20v+v

20(-5-v)=33(21v)

-100-20v=693v

-100=713v

v=-100/713

Freq CV = 5v gives you v=65/713 (0.091) after some shortening.

Back to the Obie: I deeply suspect a different CV range. Let me check its' schemos and service manual if you want a checked response plus calculations. However, the above thing plus the data sheet works well. Also, the VCF is very much the same-same in CEM3378/3379/3389 and the CEM3372 - close enough for the data sheets to suggest the same configuration. Also, it may be a good read to see how the recommended caps were chosen (see the 3389 pdf) and compare those with those of the Obie - I wouldn't assume they are identical. The CEM3372 PDF at Synthtech is really short and not very forthcoming. Maybe there is a more full version in the old yellow On-Chip data book I have somewhere?

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

Loading...
×
×
  • Create New...