Solving the UA726 problem . Let's get it done for good

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

Moderators: Kent, Joe., analogdigital, infradead, lisa, parasitk, plord, sduck

Post Reply
raph36
Learning to Wiggle
Posts: 44
Joined: Wed Aug 05, 2015 8:14 pm
Location: France

Solving the UA726 problem . Let's get it done for good

Post by raph36 » Mon Jan 18, 2016 5:29 pm

Ok guys,

As some of us already know, getting the infamous UA726 is becoming nearly impossible or damn expensive. Since the recent shortage of chinese clones wich went for about 15$ (and were mere replacement of the original), getting the last stocks will cost you about 90$ (a buchla 208 needs two and a lot of us are building those), if you already have some stock , think about an eventual failure, or think about the rest of us!

the idea of this thread is to develop a permanent solution for the diy community once and for all. I would do it myself alone if i was totally sure of what i'm doing but i'm not....

since i don't think it's been done the idea of this thread is to end up with a working and elegant PCB that can replace any UA726 , i will make a batch of them to sell to the community for a few bucks if some of you want to, as well as providing the eagle files

SO
I've been looking for what's been said and it seems that using a CA3046 is the way to go, i've found two schematics, one seems to be a direct rewiring of a CA3046 with no added circuitry, the other one seems a little more complex with the use of an op-amp and a few resistors, if somebody knows what are the main differences (precision ?) ?
i've started to replicate the more complicated circuitry in eagle, but i want to be sure so let's start from that.

Image

Image

cheers

Grumskiz
Wiggling with Experience
Posts: 341
Joined: Thu Apr 03, 2014 12:51 pm
Location: Germany

Post by Grumskiz » Mon Jan 18, 2016 6:04 pm

I'm no expert, but here is the gist of the differences of the two replacement circuits you mentioned.

Both should be equally accurate in theory, since they rely on the matched transistor pair inside the CA3046. The uA726 also really just is a matched transistor pair.
However matched transistors alone don't give you much temperature stability. Many designs nowadays use tempco resistors to counteract temperature changes.
The uA726 takes a different approach. It has an internal heater which keeps the matched pair at a certain temperature, which should be well above the expected operating temperature, but of course within the ratings of the transistor.
The more complex replacement circuit also uses this approach. In the CA3046 are 3 more transistors besides the matched pair. One is used as a thermometer, another as a heater. I think one transistor remains unused.
AFAIK the op-amp is the controlling element that activates the heater until the temperature you want is reached. It can be set via the trimmer.

Here are some things you should check out for more information:

Rene Schmitz has written an article on temperature compensation with tempco resistors and explains why it is necessary to do so.

Another wiggler by the name "guest" (of OpenMusicLabs?) has written a really great paper on various temperature compensation methods. The one you show here is also discussed in it.

Also you already seem to know that Doepfer uses a heated CA3046 in one of their designs. There is also the System X VCO by Frequency Central in which a uA726 of the original Roland design was replaced by a heated CA3046. It is available as a DIY module if you want to take a closer look at it.

One last thing: This all isn't going to get anyone very far if one wants to build replacement circuits with the CA3046, because this part is no longer being manufactured. (I think there is a SMD version which still is...not sure, never seen it in any online stores)
Maybe you could go for the THAT300 part? It's rather expensive though and I'm not sure whether it could be used in this context.

User avatar
BugBrand
Knowledge of Bugs
Posts: 7082
Joined: Tue Jul 15, 2008 11:59 am

Post by BugBrand » Mon Jan 18, 2016 6:14 pm

I actually got interested in this sometime last year - the internal schematic of the uA726 seems quite cloneable. I haven't progressed fully due to general busys (and perhaps as I've no direct desire for the chips - it was more curiousity) - but I did revise the design down to:
- quality matched pair LS318 (SOIC) combined with 0605 SMD parts, tiny dual transistors and some zeners - all mounted one side of a small PCB that overhangs the original footprint slightly, but not too much (may be possible to shrink further once the concept has been proved or not). Thermal goop would cover the top to ensure stable temp. (not just epoxy - proper epoxy heat compound)
I have had a test PCB (or several) for a month now and haven't had a chance to fully stuff it, yet alone try to run tests.
Actually - maybe someone in UK has a suitable test circuit already hooked up? // someone who has used proper ua726s plenty before - perhaps if I finished off a couple and sent them to someone to test (prefer UK at this stage).

slow_riot
Super Deluxe Wiggler
Posts: 4002
Joined: Tue Jun 30, 2009 8:58 pm
Location: uk

Post by slow_riot » Mon Jan 18, 2016 7:28 pm

Now I might get into trouble for saying this, but support original design and stop putting out Buchla clones!

The Open Music Labs exponential converter can be done with an AD633 (with a slight dent in linearity of input voltage) and a quad transistor. Once you've trimmed out the offset errors on the AD633 you can get around 10mV drift over a 40 degree change in temperature.

User avatar
qp
Veteran Wiggler
Posts: 516
Joined: Sat Apr 27, 2013 4:58 pm
Location: United States

Post by qp » Mon Jan 18, 2016 10:29 pm

So the picture you posted is the replacement 726 I came up with for my 208r. This is actually wrong. It should be using the pair on pins 1-5 (pin 3 are tied together). Mine works fine the way I have it wired up though.

I recently came across the top image and decided to make a pcb of it. It just came in the mail the other day and I plan on testing it out this week. I'll be sure to post here how it turns out.

Don T
Super Deluxe Wiggler
Posts: 1235
Joined: Sun Feb 02, 2014 3:11 pm

Post by Don T » Mon Jan 18, 2016 11:16 pm

As far as the 208r goes, I already have a solution drawn up for the uA726 problem, I just want to test it out before going public with it. It will involve an alternate card 6 and card 7 though.

When will I test it? Waiting for the $$ to buy all those accursed switches and molex connectors! Well, waiting for the panel to come in also.

User avatar
BugBrand
Knowledge of Bugs
Posts: 7082
Joined: Tue Jul 15, 2008 11:59 am

Post by BugBrand » Tue Jan 19, 2016 5:50 am

slow_riot wrote:Now I might get into trouble for saying this, but support original design and stop putting out Buchla clones!

The Open Music Labs exponential converter can be done with an AD633 (with a slight dent in linearity of input voltage) and a quad transistor. Once you've trimmed out the offset errors on the AD633 you can get around 10mV drift over a 40 degree change in temperature.
Hear-Hear!

Very interesting paper - I should read it properly.
As mentioned in the conclusion, the AD633 approach isn't entirely viable for regular production due to excessive $$$. (though I know your feelings on such things!)

Quick pic of the unfinished (maybe this'll urge me to get a couple finished this evening)
Image
The circular pinout is only roughly what the ua726 should be, but should be fine with standard wire legs (not proper pins) - getting the exact angles/placement was making my head hurt.

User avatar
oberkorn
Veteran Wiggler
Posts: 603
Joined: Tue Jan 18, 2011 6:30 am
Location: sweden

Post by oberkorn » Tue Jan 19, 2016 6:20 am

BEMI has SSM2212's in some of their easels, as spotted by Logan Erickson.
any pros/cons regarding that substitute?

ImageImage
Do not buy from x0xsh0p.de! they will not send your goods and will not answer your calls/emails

User avatar
BugBrand
Knowledge of Bugs
Posts: 7082
Joined: Tue Jul 15, 2008 11:59 am

Post by BugBrand » Tue Jan 19, 2016 6:43 am

It is hard to tell whether they just have the matched pair or also the active heating. Certainly, if the heating parts are present and are mounted underneath, then there is no thermal coupling between them..


SSM2212 or LS318 = quality matched pair -- but that's only one half of the ua726 approach, the other half being the heating to give constant temperature.

raph36
Learning to Wiggle
Posts: 44
Joined: Wed Aug 05, 2015 8:14 pm
Location: France

Post by raph36 » Tue Jan 19, 2016 6:49 am

Nice work Bugbrand ! seems like a good solution if it works !!
BEMI has SSM2212's in some of their easels, as spotted by Logan Erickson.
any pros/cons regarding that substitute?
:eek: Buchla seems to use a quite simple solution.... does anyone know if they use this as a direct replacement solution? No added circuitry? no tempco ? seems almost too simple...like everyone has been scratching their head way too much....
if so we just have to clone this little board, which shouldn't take more than an hour...
Image

seems like there is a spot for a tempco (R1) but it is unused, like they wanted to put a tempco but they found it was useless

so we should be able to start from this, right

Image

Image
edit : this seems to be a wrong diagram here is the fairchild one ::

Image
Last edited by raph36 on Tue Jan 19, 2016 1:05 pm, edited 1 time in total.

User avatar
snaper
Ultra Wiggler
Posts: 975
Joined: Mon Feb 10, 2014 12:29 am
Location: Magyarország

Post by snaper » Tue Jan 19, 2016 7:00 am

raph36 wrote:Nice work Bugbrand ! seems like a good solution if it works !!
BEMI has SSM2212's in some of their easels, as spotted by Logan Erickson.
any pros/cons regarding that substitute?
:eek: Buchla seems to use a quite simple solution.... does anyone know if they use this as a direct replacement solution? No added circuitry? no tempco ? seems almost too simple...like everyone has been scratching their head way too much....
if so we just have to clone this little board, which shouldn't take more than an hour...
Image

seems like there is a spot for a tempco (R1) but it is unused, like they wanted to put a tempco but they found it was useless

so we should be able to start from this, right

Image
Image
Oh...thats game changer :D

User avatar
papz
Super Deluxe Wiggler
Posts: 2832
Joined: Sat May 18, 2013 6:52 pm
Location: in a suitcase
Contact:

Post by papz » Tue Jan 19, 2016 7:06 am

On another forum, MechaSeb wrote that he tested a simple pair of 2N3904's in place of the UA726 in his 208r's MO, it tracks perfect on 6 octaves and the pitch is stable over hours.

User avatar
BugBrand
Knowledge of Bugs
Posts: 7082
Joined: Tue Jul 15, 2008 11:59 am

Post by BugBrand » Tue Jan 19, 2016 7:26 am

snaper wrote:Oh...thats game changer :D
Putting a matched pair where a matched pair was already but ignoring the temperature parts?

I'll say again - there are 2 parts to the 726:
- a pair of NPN transistors - should be matched.
- heating.

You can, of course, just replace with the matched pair (or 2N3904s etc) but.. isn't that missing something?! The exp-converter circuitry will work without the heating part, of course..

[just to reiterate I have no experience with ua726s - this is idle curiousity]

raph36
Learning to Wiggle
Posts: 44
Joined: Wed Aug 05, 2015 8:14 pm
Location: France

Post by raph36 » Tue Jan 19, 2016 7:39 am

BugBrand wrote:
snaper wrote:Oh...thats game changer :D
Putting a matched pair where a matched pair was already but ignoring the temperature parts?

I'll say again - there are 2 parts to the 726:
- a pair of NPN transistors - should be matched.
- heating.

You can, of course, just replace with the matched pair (or 2N3904s etc) but.. isn't that missing something?! The exp-converter circuitry will work without the heating part, of course..

[just to reiterate I have no experience with ua726s - this is idle curiousity]

Well... the UA726 needed a heater because of its transistors instability over temperature change, so the idea was to heat them at the same temp so they would stay at equal spec, thus providing a better matched pair....
but if we have no trouble with the transistors staying matched without a heater, i guess that's it ...
question is will this replace the UA726 in any given situation, meaning like in a minimoog
It's seems to be working for the 208 though, i guess Don Buchla wouldn't sell a synth that doesn't track well ?

User avatar
BugBrand
Knowledge of Bugs
Posts: 7082
Joined: Tue Jul 15, 2008 11:59 am

Post by BugBrand » Tue Jan 19, 2016 7:57 am

I've never really put enough effort into trying to understanding the underlyings of exponential converters & temp compensation, but this page gives what seems like a good overview:
http://www.schmitzbits.de/expo_tutorial/index.html

summary relating to this discussion - I surmise that you do need temperature compensation (tempco or heater or..) unless you are cutting corners.

Would no doubt be good to hear from Mr.Fritz or Mr.Hinton!

User avatar
magman
Veteran Wiggler
Posts: 501
Joined: Wed Aug 26, 2009 4:28 am
Location: Liverpool, UK

Post by magman » Tue Jan 19, 2016 9:06 am

I've been thinking of a development of bugbrands idea, using two PCB's stacked on top of the each other. This way, you could put the heater transistor(s) in close proximity to the matched pair on one side of the 2nd PCB, giving better thermal coupling. A bit if thermal compound or even thermal epoxy and this could end up being a neat package.

Ideally these would be round PCB's, close to the size of a ua726, but this depends on having enough PCB real estate to fit all the components in, using 3 of the 4 PCB sides (leaving the base clear for mounting).

Just an idea at the moment, but I might just dig into it a bit more when I have a bit more time.

Regards

Magman

User avatar
neil.johnson
Super Deluxe Wiggler
Posts: 1272
Joined: Thu Apr 08, 2010 3:51 pm
Location: Cambridge, UK
Contact:

Post by neil.johnson » Tue Jan 19, 2016 9:07 am

The expression for the exponential current in a transistor junction has TWO termperature-sensitive terms: "T" and "Is". Using two matched transistors removes the Is term, leaving the T term. The common approach has been to use a temp.co resistor to measure T so that it can be compensated, but it's always been a bit of a kludge. The alternative approach, as taken the uA726 and the heated CA3046 approach, is to keep T constant so you don't need to measure it - once the system is tuned it won't drift as the ambient temperature changes.

So, in summary, you:
* compensate for T with a tempco, or
* keep T constant with a heater

Neither is ideal (compensation is so-so, and heating is power-hungry).

Neil
Modular stuff :: http://www.njohnson.co.uk

User avatar
Graham Hinton
Super Deluxe Wiggler
Posts: 2948
Joined: Fri Jul 02, 2010 6:28 pm
Location: England
Contact:

Post by Graham Hinton » Tue Jan 19, 2016 9:29 am

raph36 wrote:... the UA726 needed a heater because of its transistors instability over temperature change, so the idea was to heat them at the same temp so they would stay at equal spec, thus providing a better matched pair....
but if we have no trouble with the transistors staying matched without a heater, i guess that's it ...
It is nothing to do with instability. All transistors are subject to the Ebers Moll equation giving the relationship between collector current and Vbe. This contains a term in the exponent of k.Vbe/q.T, where q and k are constants, but T is the absolute temperature in degrees Kelvin. This means that ALL transistors can be used to make exponential converters and ALL are temperature sensitive. Differences in the transistor die structure can introduce errors, and a low bulk emitter resistance is desirable for a wide range as this causes a significant voltage drop at higher currents. Most matched transistor pairs and quads have been designed with this in mind, the 3046 was not. HF compensation is sometimes used to correct this aspect.

A better converter can be made by having two identical transistors at the same temperature for temperature balancing and this is the most common circuit. The transistors have to be at the same temperature and that means the silicon itself, not the outside of the package, which means a monolithic pair. The temperature of the transistors will change with self heating as the current changes so two transistors stuck together do not give good results.

If that circuit remains at a constant temperature it will function as required, but if it is subject to temperature changes of a few degrees the output will be in error proportional to the change in absolute temperature. It will still be exponential and will track another, but the scale will have changed and an octave CV will not be exactly an octave change. At 25 deg C a 1 degree change is 1/(25 + 273.15) = 0.003354 so if you can multiply the Vbe by something that changes by the same factor it will be temperature compensated. Alternatively the temperature can be prevented from changing. Neither technique is as easy as it sounds. The problem with temperature compensating resistors is that the coefficients have a +/-10% tolerance whereas the transistors don't, so you can't just drop one in and expect it to work. If compensation was really working you would be able to change the temperature with a hot air blower or a freezer spray and hear very little change.

The problem with temperature ovens is that it has to run at a temperature higher than any that will be encountered in use, say above 60 degC, and the servo loop will have a time lag so it will be cycling between a maximum and minimum temperature.

In practise several VCOs of the same design in the same synthesizer at the same temperature will change pitch a similar amount and appear to track each other, but have an absolute error. Trying to get different VCOs in different places in a modular synthesizer to track is far more difficult and is only going to work if the temperature compensation is exact or if the temperature can be held above any ambient changes within a few degrees.
Also two identical synthesizers may not track together if built at different times using a different batch of tempco resistors
question is will this replace the UA726 in any given situation, meaning like in a minimoog
Only the last Minimoog models used µA726s, most of them used 3046s and Q81s.
It's seems to be working for the 208 though, i guess Don Buchla wouldn't sell a synth that doesn't track well ?
Try playing a Buchla with a five octave keyboard...
I would put more trust in Dave Smith.

Hasn't anybody noticed that the BEMI pcb pictured above does not match the µA726 pinout?

User avatar
neil.johnson
Super Deluxe Wiggler
Posts: 1272
Joined: Thu Apr 08, 2010 3:51 pm
Location: Cambridge, UK
Contact:

Post by neil.johnson » Tue Jan 19, 2016 11:02 am

Well it doesn't help that the diagram given for the uA726 is wrong compared to the Fairchild datasheet. It looks like the PCB is laid out correctly, in so far as the tracks I can see go to the right places.
Modular stuff :: http://www.njohnson.co.uk

User avatar
BugBrand
Knowledge of Bugs
Posts: 7082
Joined: Tue Jul 15, 2008 11:59 am

Post by BugBrand » Tue Jan 19, 2016 11:15 am

neil.johnson wrote:Well it doesn't help that the diagram given for the uA726 is wrong compared to the Fairchild datasheet. It looks like the PCB is laid out correctly, in so far as the tracks I can see go to the right places.
Well noticed - wherever that ua726HC image comes from appears to have pins 3 & 4 swapped?! (compared with fairchild)

User avatar
Graham Hinton
Super Deluxe Wiggler
Posts: 2948
Joined: Fri Jul 02, 2010 6:28 pm
Location: England
Contact:

Post by Graham Hinton » Tue Jan 19, 2016 12:17 pm

neil.johnson wrote:Well it doesn't help that the diagram given for the uA726 is wrong compared to the Fairchild datasheet. It looks like the PCB is laid out correctly, in so far as the tracks I can see go to the right places.
I wasn't referring to the bec pin sequence, but the grouping. The BEMI board has an unused pin between the two transistors whereas the IC package has the transistors consecutive: 9, 10 , 1-4, and then four pins on the other side, one unused.

User avatar
neil.johnson
Super Deluxe Wiggler
Posts: 1272
Joined: Thu Apr 08, 2010 3:51 pm
Location: Cambridge, UK
Contact:

Post by neil.johnson » Tue Jan 19, 2016 12:25 pm

Graham Hinton wrote:
neil.johnson wrote:Well it doesn't help that the diagram given for the uA726 is wrong compared to the Fairchild datasheet. It looks like the PCB is laid out correctly, in so far as the tracks I can see go to the right places.
I wasn't referring to the bec pin sequence, but the grouping. The BEMI board has an unused pin between the two transistors whereas the IC package has the transistors consecutive: 9, 10 , 1-4, and then four pins on the other side, one unused.
I think you're referring to pin 1 on the BEMI board (it has the square pad), which is one of the bases, and it looks like they've assumed that it is connected to ground on the main board and so connected that to the ground plane on that little board. So you won't see a specific trace for it, but you can just about see that it looks like it and the matching pin on the SSM chip are tied to the plane.

Neil
Modular stuff :: http://www.njohnson.co.uk

raph36
Learning to Wiggle
Posts: 44
Joined: Wed Aug 05, 2015 8:14 pm
Location: France

Post by raph36 » Tue Jan 19, 2016 1:08 pm

mmmmh s*** you guys are right this diagram is wrong and it is a screenshot of the one that is on that CA3046 replacement circuit image posted earlier
so whoever is working with this make sure you use this one instead

Image

User avatar
guest
Super Deluxe Wiggler
Posts: 3589
Joined: Mon Aug 19, 2013 11:49 am

Post by guest » Tue Jan 19, 2016 1:43 pm

its my guess that the missing R1 in the replacement board is actually under the matched pair. its the only way it would work.

as for making a replica uA726 with discretes, ive looked into this a bit as well, and decided that the opamp heater version was better. for either, you will need 4 transistors on the same die. the issue, as graham pointed out, is thermal cycling. it takes some time for the temperature to stabilize between your exponentiating transistor and your sensing transistor. the further these are apart from one another, the longer this process takes. and, to complicate things further, thermal processes move really slow to begin with (at least compared to how fast electrons move around). then you have the heater, which also needs to pass its heat to both the expo pair and the sensing transistor. there is necessarily some oscillation that will occur when doing this, and the longer it takes to equilibriate, the worse this will be. you can dampen the system, but then you arent cancelling out faster moving temperature differences.

one improvement that could be made, since we know the topology the matched pair will be used in (one side being driven with a current source), is to use that transistor as the temperature sense transistor as well. that way there is no lag or mismatch to the sensing side of things.
openmusiclabs.com

ttown23
is banned
is banned
Posts: 1157
Joined: Fri Oct 12, 2012 5:20 pm
Location: Los Angeles, CA

Post by ttown23 » Tue Jan 19, 2016 1:48 pm

raph36 wrote:does anyone know if they use this as a direct replacement solution? No added circuitry? no tempco ? seems almost too simple...like everyone has been scratching their head way too much....
if so we just have to clone this little board, which shouldn't take more than an hour...
As far as I know, the BEMI solution is a direct drop-in replacement. There is no TEMPCO on that board, and I have seen high-resolution images, too, and no TEMPCO was present there, either. There is no added circuitry, either.

My vote would be to go this route as opposed to the CA3046. Ultimately, when CA3046 stocks dry up, we would be in exactly the same boat as with the UA726. At least this way, we are using currently in-production SSM 2212s.

@ Bugbrand... my offer stands to do VCO testing of prototypes. I could do these side-by-side with UA726 and record values as well.
I'm banned for dishonest and unethical behaviour.

Post Reply

Return to “Music Tech DIY”