reverse engineering the IR3109

From circuitbending to homebrew stompboxes & synths, keep the DIY spirit alive!

Moderators: Kent, luketeaford, Joe.

User avatar
guest
Super Deluxe Wiggler
Posts: 6589
Joined: Mon Aug 19, 2013 11:49 am

reverse engineering the IR3109

Post by guest » Thu Nov 27, 2014 8:38 am

i have a bit of free time while i wait for boards to come in, so i thought id reverse engineer the ir3109. i opened up my shruthi and popped it out, and am hoping i dont fry it. at any rate, heres what i got so far:

Image

basically its an expo converter, 4 OTAs, and 4 buffers. im assuming the OTAs are ba662 cores, and will verify that next. the expo converter will be a bit tricky to exactly nail down, as little of it is exposed, but from looking at the roland filters this chip replaced, its most likely a pnp/npn pair. i will have to think of a good test to convince myself of that. and, the datasheet says it uses P-channel output buffers.

so, i just finished with the output buffers. they are indeed P-channel, with a Vgs of around 3V. there are two diferent schemes used. buffers 1/3 use 20k pullup resistors, and buffers 2/4 use 1mA current sources (2 pnp current mirror). this was suprising, but after looking at roland synth schematics it makes sense. the 2/4 buffers are used as the output stage of the 2/4 pole filters, and need extra drive to feed the next stages.

the input current on the FETs is less than 20nA, and im not sure i can measure any more accurately than that as the gate is directly connected to the output stage of the OTA, so its hard to disambiguate where the current is going. even with the OTA off, there may still be some leakage in that direction.
openmusiclabs.com

User avatar
guest
Super Deluxe Wiggler
Posts: 6589
Joined: Mon Aug 19, 2013 11:49 am

Post by guest » Thu Nov 27, 2014 9:20 am

the OTA sections are ba662/lm3080 style cores. just that pesky expo converter left.
openmusiclabs.com

User avatar
Electric Druid
Common Wiggler
Posts: 115
Joined: Sat Mar 31, 2012 1:20 pm
Location: UK
Contact:

Re: reverse engineering the IR3109

Post by Electric Druid » Thu Nov 27, 2014 9:21 am

Very cool!

I'd guessed the buffers were FETs, since Roland seem to love FET buffers in many of their OTA filters. That's a part of the Roland sound, for sure. The info about the current sources for the other two stages is very interesting - makes sense, like you say.

The expo convertor probably isn't that crucial for the sound of it, though obviously more octaves and less error would be nice...

Anyway, good work. Can you do a schematic of the equivalent circuit?

Thanks,
Tom

User avatar
guest
Super Deluxe Wiggler
Posts: 6589
Joined: Mon Aug 19, 2013 11:49 am

Post by guest » Thu Nov 27, 2014 9:52 am

yeah, the expo isnt so critical to the sound, but im still curious. right now its looking like the pnp/npn setup, although im not sure how they do the current sharing, and probably will never be able to say for certain.

the OTAs could be easily replaced with LM13700s for this purpose, as the only difference is the current mirror input, and that is driven with a current source, so its unaffected. finding the right FET will be a bit of a needle in a haystack, though.

ill definitely do a full drawing once ive got the last piece.
openmusiclabs.com

User avatar
minisystem
Common Wiggler
Posts: 168
Joined: Wed Dec 26, 2012 5:16 pm
Location: Toronto
Contact:

Post by minisystem » Thu Nov 27, 2014 10:23 am

Very excited about this!

Luap
Super Deluxe Wiggler
Posts: 1211
Joined: Fri Nov 30, 2012 6:41 pm
Location: North'ish London, UK

Post by Luap » Thu Nov 27, 2014 10:44 am

Its too technicated for me to be of any help, but I sure am glad you are doing this, Mr Guest :tu:

User avatar
baloo
Common Wiggler
Posts: 129
Joined: Sat Aug 17, 2013 7:07 pm
Location: Jungle

Post by baloo » Thu Nov 27, 2014 4:08 pm

interesting.....

synthnerireland
Learning to Wiggle
Posts: 7
Joined: Fri Jul 25, 2014 4:21 pm

Post by synthnerireland » Thu Nov 27, 2014 10:55 pm

As part of his replacement module for the Juno 106, analogue renaissance has used 13700's and a matched transistor pair to accurately replicate the 3109, with another 13700 to handle resonance and VCA duties, which leads to the obvious inference, the 3109 is not a special chip, just 4 OTA's and a transistor pair on board. As Roland used the 3109 to replace cascaded BA662's in the Jupiter 4 with a saving on board space and hopefully improved matching, then it would be a reasonable assumption that the BA662 is just an OTA with a transistor pair driving the current. Nothing magical.

User avatar
guest
Super Deluxe Wiggler
Posts: 6589
Joined: Mon Aug 19, 2013 11:49 am

Post by guest » Fri Nov 28, 2014 3:37 am

very true, OTAs are almost all identical. and, one of the great properties of them, is that the transfer function is not too dependent upon transistor beta, so it doesnt matter too much which transistors you use, as long as they are matched. but, there is more going on than just the OTA.

i looked at the analogue renaissance work while doing background research for this project. i would consider it an "improved IR3109". the current rev seems to use a standard opamp/diffpair expo converter (which will track better, and therefore be different than the original), and uses opamp buffers after the OTAs (which whill have way less distortion).

since the OTAs are running with negative feedback, the buffer is not too important for normal operation, but will make a world of difference when over-driven, or used with lower supply rails. so it all depends upon what the IR3109 is going to be used for.
openmusiclabs.com

User avatar
guest
Super Deluxe Wiggler
Posts: 6589
Joined: Mon Aug 19, 2013 11:49 am

Post by guest » Sat Nov 29, 2014 8:31 am

ok, here are the output stages:

Image

still working on the expo converter. it uses a voltage independent current source to set the bias current, and has a gain of around -290dB/volt. the output is relatively linear over 100dB of its range, which is pretty good. it saturates internally at 700uA, and measuring on the low end is difficult, and quite noisey. i went down to about 2nA before i couldnt discern the signal anymore.
openmusiclabs.com

User avatar
guest
Super Deluxe Wiggler
Posts: 6589
Joined: Mon Aug 19, 2013 11:49 am

Post by guest » Sun Nov 30, 2014 6:36 am

ok, here are the results:

Image

the top drawing is a schematic representation of whats in there, and the bottom drawing is a best guess at how its implemented.

as it turns out, they use an opamp/diff-pair for the expo converter. this gets mirrored off the top rail and sent into a the OTAs with the base currents buffered to ground via a transistor. the bias current is 3uA, and this is set with a generator that runs between +/-Vcc. there is an opamp of some sort that holds this 3uA constant through the reference transistor. for a number of reasons, i believe they used another BA662 for the opamp, and have its bias current set at 3uA as well. exactly how this current is reflected into the BA662 is unclear, but the version shown would use the fewest transistors. im pretty confident about the 1.6k resistor, and that they set the 3uA with a zener source biased with a 20k resistor. the 3uA doesnt have the greatest stiffness with either temperature or Vcc, so something like what is drawn was probably used.

the OTAs on mine were matched to about 5% of each other, so not the greatest tracking.
openmusiclabs.com

User avatar
Tim Stinchcombe
Uncommon Wiggler
Posts: 1421
Joined: Fri Mar 13, 2009 1:55 pm
Location: Cheltenham, Glos, UK

Post by Tim Stinchcombe » Sun Nov 30, 2014 1:16 pm

I'm a little confused as to what you are actually looking at - if you decapsulated an actual Roland IR3109 chip and could see the die, I assume you wouldn't be asking about transistor configurations, as you'd be able to see them, so are you actually reverse-engineering some other 'interpretation' of the chip, as implemented in the Shruthi?!

Tim

User avatar
guest
Super Deluxe Wiggler
Posts: 6589
Joined: Mon Aug 19, 2013 11:49 am

Post by guest » Sun Nov 30, 2014 1:20 pm

no, i have an actual IR3109, and ive been probing all the pins and seeing how it responds to figure out the configuration of the transistors inside.
openmusiclabs.com

User avatar
minisystem
Common Wiggler
Posts: 168
Joined: Wed Dec 26, 2012 5:16 pm
Location: Toronto
Contact:

Post by minisystem » Sun Nov 30, 2014 7:50 pm

Fascinating. I'm interested to see where you take this. As already mentioned, it seems the AR 80017A takes a simpler approach to produce what by all accounts is a very authentic sound, whereas a direct, discrete clone of the IR3109 has a more complicated (ie. higher part count) expo converter and output stages. So we're looking at 5 OTAs, 4 FETs, almost 2 dozen BJTs and a handful of resistors/diodes? I guess the chances of getting it to fit on the original IR3109 footprint are small...

User avatar
guest
Super Deluxe Wiggler
Posts: 6589
Joined: Mon Aug 19, 2013 11:49 am

Post by guest » Mon Dec 01, 2014 12:21 am

i dont think i could make a clone that fits into the IR3109 footprint. maybe if it was a sandwich of 2 boards, each double sided, but that doesnt seem worth it. ultimately, i dont think the exponential converter being "discrete" makes much of a difference. i think it might be nice to put in the saturation effect of the OTA driving the diff-pair, but that could be accomplished with a zener diode. i would probably use a 2.5V voltage reference and a resistor/opamp to set the 3uA. exactly how to mirror the output current into the OTAs is an interesting question, but not all that critical to the sound.

im not sure what the AR 80017A uses exactly, but im guessing the only major difference between something i would do and that, would be to use the discrete P-channel buffer output stages.
openmusiclabs.com

User avatar
Dimitree
Veteran Wiggler
Posts: 677
Joined: Mon Aug 01, 2011 9:59 am

Post by Dimitree » Sun Dec 21, 2014 2:06 pm

Just wondering, is this project going to end in a clone chip like you did on the BA662 or it was just a research/experiment?

User avatar
guest
Super Deluxe Wiggler
Posts: 6589
Joined: Mon Aug 19, 2013 11:49 am

Post by guest » Sun Dec 21, 2014 3:09 pm

ive been thinking about that as well. i could never get it down to the size of the original IC. if i were to make a clone, i would use 2 x LM13700 for the core, as it would be the same, then single MOSFET buffers with matched pair current mirrors on 1/3. then its another matched pair/opamp/voltage reference for the exponential converter. so its getting pretty big (maybe 1.2"x3"). do you think there would be interest in that sort of thing?
openmusiclabs.com

User avatar
Dimitree
Veteran Wiggler
Posts: 677
Joined: Mon Aug 01, 2011 9:59 am

Post by Dimitree » Sun Dec 21, 2014 3:30 pm

You mean as direct replacement of the original? I don't know if there would be interest, since I don't have any Roland synth so I don't know how much those chips fail.
I can speak about me: my only interest is to replicate old technology with modern parts (and SMD), and I'd like to build a roland filter with such philosophy, so pcb dimensions are not really an issue for me.

User avatar
guest
Super Deluxe Wiggler
Posts: 6589
Joined: Mon Aug 19, 2013 11:49 am

Post by guest » Sun Dec 21, 2014 3:43 pm

if i were to just release a design file and bill of materials, would that work as well?
openmusiclabs.com

User avatar
minisystem
Common Wiggler
Posts: 168
Joined: Wed Dec 26, 2012 5:16 pm
Location: Toronto
Contact:

Post by minisystem » Sun Dec 21, 2014 4:20 pm

Do it for the Boss PH2s! ;)

If you release the design files, then perhaps projects that require IR3109s could be rejigged to accept a slightly larger format module. Also, someone could eventually create a through-hole version for DIYers that aren't comfortable with SMT projects.

The Roland filter sound is perennially popular. Having an open source clone of that filter, even if it can't be used as a drop in replacement, would be an amazing resource I think.

User avatar
Dimitree
Veteran Wiggler
Posts: 677
Joined: Mon Aug 01, 2011 9:59 am

Post by Dimitree » Sun Dec 21, 2014 4:24 pm

guest wrote:if i were to just release a design file and bill of materials, would that work as well?
Of course for me even the single schemaic would be great. I like to design my own smd PCBs

User avatar
jsleeio
Super Deluxe Wiggler
Posts: 1530
Joined: Thu May 29, 2014 7:52 pm
Location: Sydney, Australia

Post by jsleeio » Sun Dec 21, 2014 5:32 pm

I would be down for some PCBs, whether from someone else's manufacturing run, or my own. I'd also happily buy some assembled units
Zap him again! ZAP THE SONOFABITCH AGAIN!

User avatar
xpmtl
Knob Twiddler
Posts: 368
Joined: Wed Mar 16, 2011 8:42 am
Location: Belgium

Post by xpmtl » Mon Dec 22, 2014 10:16 am

the single schematic would be great!

User avatar
guest
Super Deluxe Wiggler
Posts: 6589
Joined: Mon Aug 19, 2013 11:49 am

Post by guest » Mon Dec 22, 2014 6:20 pm

ok, apologies for this being such a large image, but i didnt want to lose too many details.

Image

this is my best guess at whats inside. the OTA cores, buffers, exponential pair, and 1mA current source i am pretty certain about. i am also fairly certain about the general topology of the exponential current source, but the current mirror, diff amp, and 3uA current source could be done differently. subparts of those i am also pretty certain about:

1. the 1.6k resistor and pnp to drive the exponential pair.
2. the 3uA value for the current source.
3. the diff pair is driven with 3uA as well, and has a current output (ota).
4. the 1.2V zener for the 3uA current source.
5. the use of a single mirror to reflect the exponential converter current into the OTAs (rather than generating 4x the current and dividing it amongst the OTAs).
openmusiclabs.com

User avatar
minisystem
Common Wiggler
Posts: 168
Joined: Wed Dec 26, 2012 5:16 pm
Location: Toronto
Contact:

Post by minisystem » Tue Dec 23, 2014 11:44 am

Bonkers. I find this kind of electronic archeology fascinating. Nice work.

So, what is your next step? Would you prototype a fully discrete version of this or start by replacing the OTA stages with a pair of LM13700s? Simplify the current source?

How crucial is the choice of FET for the buffers?

Post Reply

Return to “Music Tech DIY”