Power Supply Modding
Moderators: altitude, adafruit_support_bill, adafruit, hamburgers, phono

Please be positive and constructive with your questions and comments.

Power Supply Modding

by Nordcore on Sun Jul 30, 2017 5:42 am

There was always a lot of talking about the "unstable" power supply of the 303 circuit design.
Some people even think that there is a magic component, the "sumida coil" in the original 303, which they believe as necessary to reproduce the original sound.

Doing some thinking about the circuit, some Google searching about what others had already found out, and some trials on my x0x, it gets pretty clear what is happening.

First we need the schematics:
psu.png (27.52 KiB) Viewed 2148 times

The 9V goes into the voltage doubler, so we get round about 11V at C3 and about 24V at C5. As the current from the higher voltage is much lower (about 20mA) than from the first stage (some 200mA), this is not exactly doubled.

For the 5V and 6V we see pretty boring standard regulators.

For the analogue supply it gets more interesting:
First there is R1/C8 acting as low pass. Corner frequency is 16Hz. That is reasonably low to filter out any high frequency noise. Which *we* don't have here in our x0x-es, but which might be there in an original 303 supply, where this point is supplied by a switching DC/DC converter.

See 303 schematics:
303psu.png (50.95 KiB) Viewed 2144 times

Within the 303 DC/DC converter there is a regulation, consisting of D42 and Q42. This regulates the voltage at R172 to about 15.6V.

Filtered by R1/C8 this 15.6V delivers the supply for IC16, an OP-Amp (ab)used as voltage regulator. First there are the 5.333V, used as virtual ground and as reference for the pitch CV. There is a proper reference diode (D43) and a trim (TM6) for the 5.333V, where this voltage is set to its exact value.
The 5.333V is than multiplied by (1+R178/RR179)=(1+2200/1800)=11.85V, giving the "12V" supply for the analogue voltage.

So, where is the difference?
Up to the description above the only difference is the height of the supply for IC16.
As the current draw from the analogue circuitry is about 10 to 20mA, there is 1 to 2V voltage drop across R172 and we have about 14V supply for IC16 within the 303 and about 22.5V within the x0x.
As the AN1458 (Note: AN6562 is the same, Panasonic has used both numbers for the same chip) does a pretty good job at regulation, there is no difference in the output voltage.

Until we consider component tolerances *and* the minimum difference IC16 needs between supply and output voltage.
a) The AN1458 needs 1.5V minimum voltage difference between supply and output, else the output follows the supply.
b) The 11.85V for the 12V are multiplied by 5% resistors from the (pretty exact) 5.333V.
c) the 15V Z-Diode D42 has another 5% (or so) Tolerance.
d) voltage drop across R172 is dependent of the current consumption by the analogue circuitry, and about 1 to 2V. The envelope generator draws a little more current during attack, as everything else is not fully constant - slightly more during a tone.

If, by chance of the tolerance of our assembled components in our specific 303, the 12V is a little on the high side, the 15V from D42 is a little on the low side, and the voltage drop on R172 is on the high side (during attack), than our 12V will see a small break in.

We will never see this on a x0x. Increasing R171 (to, as sometimes suggested, 220Ohms) does not change that, as there still is plenty of voltage to drop before IC16 might lose regulation.

To mimic this possible original behavior a different approach is needed:
a) add an adjustable regulator (e.g. LM317) between C5 and R1, so one could mimic any voltage tolerance from the original supply.
b) add a trim to R178/R179 divider, so the 12V could be adjusted.

I did that on my x0x.
One interesting point is, that adjusting the 12V in the range 5% resistors (all with regulation maintained) might give, there really is some change in the sound behavior of the x0x. So if you think "no two boxes sound the same" - here definitely is one of the sources.
The boring thing: most of that is "Accent Amount", if you adjust the settings carefully you could bring back a "higher 12V sound" back to the "nominal 18.85V" sound. All what really is left is an slightly increased setting range.

I'm still not sure about the 12V "dip" (regulation lost) sound. If overdone the attack gets weak, just add a little (which is a not very stable setting) and it might get a little more "living".
Last edited by Nordcore on Sun Jul 30, 2017 10:55 am, edited 4 times in total.

Posts: 99
Joined: Sat May 23, 2015 3:14 pm

Re: Power Supply Modding

by antto on Sun Jul 30, 2017 6:42 am

i don't know where the difference is, probably the current flows in different patterns and creates different kind of noise
after all, the PSU circuits are different

but i think the more important question is whether that matters at all

the 303 doesn't have a dedicated digital 5V, and this is one difference
since the LEDs are driven by a matrix, which frequently turns them on and off one by one (or in groups, doesn't matter)
i've heard that the overal noise floor of the x0x is lower than on the 303, and i think that's a good thing
but some folks think that the exact behaviour and level of the noise floor in the 303 is a desirable component

tbh i'm sick of argueing with those people, so i'll try to stay at a distance
beware that some of them may understand that this is all bullsh*t but might have interest in keepying the myth alive for profit

Posts: 1636
Joined: Thu Apr 15, 2010 3:21 pm

Re: Power Supply Modding

by antto on Sun Jul 30, 2017 6:58 am

if we look at this differently

the PSU provides a few voltage sources to the rest of the circuit

so for each of those voltages, it has a given amount of current that it can provide, and this might also change under actual stress (usage)
additionally, it will have some noise, which may get worse under stress

now, the 303 analog section draws a little current normally, so things are quite calm
except when notes are played.. charging (and maybe also uncharging) the envelopes happens in a short time frame during gate-on and gate-off
but there are reservoir capacitors pretty much everywhere including in the envelope sections
in any case, this causes stress, specifically the gate-on trigger, since it's audible afterwards

the filter is vulnerable to stress on the 5.333V line (and maybe on the 12V line also) because it has a low-freq resonance point (about 8Hz) besides the normal variable cutoff frequency
so when you cause stress to the supply voltages, you excite (more or less) the low-freq instability of the filter (this is only valid when the resonance is actually in effect)
as a result, the 5.333V line begins to resonate

thing is, in some units, this resonation is small and decays out quickly
while in other units it's much bigger and may not even decay down but grow
when that happens not only the main cutoff freq of the filter starts "warbling" but also the VCO pitch starts going up and down

the instability can be adjusted by adjusting (via modification) the cutoff frequencies of the 4 or 5 highpass filters (capacitors) inside the VCF feedback path
so the people who think this "warbling" is a desirable effect - they could play with the caps and get their filter unstable
i wouldn't
in fact, my second x0xb0x started warbling (as well as having all kinds of other nasty side effects) after i (stupidly) poured nail-polish-remover all over it
the warbling means that your filter is getting unstable.. if you like it - good

Posts: 1636
Joined: Thu Apr 15, 2010 3:21 pm

Re: Power Supply Modding

by Nordcore on Sun Jul 30, 2017 1:45 pm

Neither the 303 nor the x0x-es I had here did have any noise or any other strange supply behavior.
The supplies where always clean.

What I described in the the first post are two relative simple influences:

One, which every 303 and every x0x has:

That is the absolute height of the 12V power supply.
That has as an astonishing large influence on the sound. Nothing magic: just the modulation amounts gets larger.
That might make up for a very impressive sound difference between two boxes, if one box is on the upper end, the other on the lower end, of what ist to be expected from 5% tolerance resistors.
This could be explored very easily by adding a trimm to R178/R179.
(I just used a 50kohms 20 turn trim parallel to the voltage divider. Not fool proof, but works well if you adjust the trim middle before soldering and than only turn it a little with the DMM connected. )

The other sound influence is "starving the regulator", which can't happen in a x0x but *might* happen in some 303.
That requires the supply of IC23 being lower than 1.5V above the "12V" output voltage. As the voltage drop at R1 is about 1 to 2 V that leads to 11.85V+1.5+1...2=14,35V-15.35V at the C5 side of R1.
If there is more voltage, as in the x0x, the starvation will never happen.
If there is less voltage, the regulation would not work at all, leading to unusable behavior.

The sound effect could be easily tested by supplying an adjustable voltage to R1. As stated above I used a simple LM317 for that.

What you have observed on the 5.333V is interesting, let me think about that.
(For the first time in engineering we are trying to make a regulation less stable... )

Posts: 99
Joined: Sat May 23, 2015 3:14 pm

Re: Power Supply Modding

by antto on Sun Jul 30, 2017 3:13 pm

Nordcore wrote:(For the first time in engineering we are trying to make a regulation less stable... )

i'm not

but you're certainly not alone

Posts: 1636
Joined: Thu Apr 15, 2010 3:21 pm

Re: Power Supply Modding

by 3phase on Thu Oct 05, 2017 6:29 am

Hi there and thanks for analytic look on the power supply issues,

There is a little wrong thinking in the general view on the topic, biased or prejudice in regards of audio myth in general and a too statical view on current draw. how fast does a power supply delivers to the current draw? only matters in high power amplification? Why? small signals of a 303 draw power too.. a 303 gets warm..

Why have you investigated in the power supply myth in the first place? You hear differences?
Many do, and the Power supply is big part of that experienced differences. Obviously since the circuit of a xox is otherwise very close to a real 303. But when we add just the facts stated in this thread we must assume a huge impact on the sound just by that static measurable differences.. Thats progress..some years ago even that so very audible fact was called a myth too.

But back to the main question.. How do we emulate the effect of the 303 power supply ???

The postings here clearly state and prove that a simple resistor in one of the power lines wont do that. So a myth busted here.

One for sure, power supplies do have audible fx in audio circuits.
Even subtile and difficult to measure differences have audible results. That applies especially to power amplifiers. But the biggest effort in silent super stable power supplies you find in mic and measurement amps.

A voltage might be stable, but limiting effects of the current supply abillity are very audible. The transients too fast and current demand after the transient actually rising to second long powerdraw demands on the power supply on bass frequencies.
You hear weakness in the current supply ability way before you measure any voltage drop on your power rails. And a slow current draw reading wont show you much what happend here..since we have a very dynamical signal and load here..

We have statements here like resonating reference voltage of the filter..that gets multiplied to become the analog supply voltage....
Hello? what a huge fx that must have on a synth circuit.

But why would that make the sound elastic? warbeling? ok.. the 303 typical rubber band sound is there way before it marbels.. its there on on any real 303, even on the ones that dont marbel at all ! All 303´s sound different..true.. but all of them have that elastic quality.
So its not the marbeling alone we have to look into here.

Where does that come from? what posible role can the sumida coil play here?

I dont think the sumida coil is magical, its just a little transformator, ..nothing special, just far from a mathematical ideal multiplier. A component with complex dynamical nonlinearities.
Its too easy to just call it an inferior supply.. There is a certain beauty in that inferiority and we need to emulate the beauty of that , not just the inferiority of it.
What is causing that beauty? Current draw issues? likely, but what issues? how do they fx the sound?

There are modern guitar amps that still use rectification valves..instead of cheap and way better diodes. And there are other modern guitar amps that model the behavior of a rectification valve in dsp to achieve a more realistic amp simulation. A rectification valve actually could be a thing that gives some properties of the 303 power supply.. but thats even less practical than sourcing a sumida coil. But the dffference between the dynamic current supply issues of a valve, versus the way faster diode show very much what kind of issues we might be into here. The properties of valve rectified amps versus teh same amp with diode rectification , reminds very much to the percieved differences between a xox and a 303 style power supply.

So its quite likely that we miss out on the dynamical attributes of that 303 power supply. resonance on the supply rail is a point that must be very audible.. The mentioned 16hz filter.. such a filter has time domain properties..,.. And how does the coil reacts to that? Has such a resonating coil current storage abilities? Big coils in power transformators do have that.

But that tiny high freq coil of the sumida? According to dogma such miniscule fx always have to be inaudible... But according to reality the miniscule differences of the xox powersupply are very audible,
Why? we are getting closer to answer that question.. But no replacement circuit that works in sight yet.

Just to call the sumida coil magical is not enough to dismiss its properties, there is probably a little more to analyze about it...

greets, Sven
Posts: 203
Joined: Wed Apr 22, 2009 2:06 pm

Re: Power Supply Modding

by antto on Thu Oct 05, 2017 1:39 pm

that was a very long and poetic post, but it didn't clear up the fog around the myth

Nordcore (or anyone else): i've been thinking about modding the IO board.. i mean.. making a modified IO PCB
the main reason was to fix the MIDI issue (which is due to poor circuit design at schematic level, for some reason), and possibly that might fix the MIDI<->USB(UART) data corruption that i've noticed when both communications are "busy" at the same time
and while changing the IO board, i could also change the PSU itself, including changing the input voltage to DC, and so on
also, adding solder points for an ON/OFF switch (so that you don't have to cut a trace), and maybe adding some more optional input/output jacks which can be ommited by other people (but will probably be used in my own firmware)

this is just an idea, but i got the itch.. and it itches..
so suggestions are welcome
i was pretty much thinking about changing to a DC adapter, and then a regulator, and maybe with the option to use an internal battery as well.. without having conflicts if you have both a battery and plug in the DC adapter

Posts: 1636
Joined: Thu Apr 15, 2010 3:21 pm

Re: Power Supply Modding

by Nordcore on Fri Oct 06, 2017 1:14 pm

oh ... of course I have already thought about "making things better"...

Going DC: if 6V DC supply is acceptable:
low drop regulator to make 5V.
A small 1W standard SIP-module, 5V to 15V DC/DC for the analogue part.

My x0x needs up to 250mA, as it has standard red LEDs and small resistors. And we have to allow "all LEDs on", as this might happen during power up / update, due to the random state of the LED registers. (I have a fix for that, one hardware patch, one software (new Bootloader). Best if you have both... )
Input up to 9V should be allowable = about 1W to dissipate by the regulator. That could be done by some copper on the PCB, no need for an extra heat sink.

1. a simple pin row for the serial signals would be nice. User not wanting to solder SMD could just plug an available USB-TTL-RS232 adapter cable to the pins.
2. FT230X instead of FT232. Cheaper. Less parts. Better. Less pins, so a little easier to solder.
3. Opto insulation for the USB-serial interface. As the USB comes with its own 5V and both (FT230X and ATMEGA) can drive a a LED directly, that is just two 6N138 plus two resistors. This eliminates the hum loop usually created by USB.
4. Optional: may be Micro USB jack. Sitting together with FT230X on a small piggy-back board. So that could be soldered professionally, eliminating the SMD for the DIY-part.

MIDI: use 6N138 opto coupler, the now used type does *not* work reliable.

Posts: 99
Joined: Sat May 23, 2015 3:14 pm

Re: Power Supply Modding

by 3phase on Sat Oct 21, 2017 6:44 am

there is no fog around what some people call a myth without having had any real investigation into the topic..,

That the powersupply design does effect the sound of a 303 should be out of a question by now, people that state the fx have backed it up by own experiments, while critics resort to
theoretical guess work, mostly at best half educated. Its very easy to find out yourself.. just cut the power supply traces on a real 303 and feed the voltages from lab power supplies.. and the next step would be to conect that original 303 power supply to a xox.. I did that..have you?

So that part is proven fact for me..self proven, not word of the internet prove..

HAs the coil something to do with the elastic balistics of the original roland power supply? Rather likely than unlikely.

We have to assume that the coil does something for the special 303 sound.. Until somebody finds a replacement circuit that manages to sound similar without the coil.
That has not happened sofar ! And this cant really happen when analytics ignore the exsitance of this coil..replace it with the most simple ohm law equations.
Inductor math is usually not that simple. Especially under dynamic AC loads.

Is it possible to simulate that coil behavior? thats the foggy part, but no myth. Its cheap polemics to call it a myth..its just an open question, and taliban like thinking barriers dont help to answer that question.

Some even question the audible effect of power supply specs and design on audio amplification at all.. Thats a bit like claiming the earth to be flat in the 21th century.
Posts: 203
Joined: Wed Apr 22, 2009 2:06 pm

Re: Power Supply Modding

by Nordcore on Sun Oct 29, 2017 6:40 am

I did some more investigation.
I do have the guts of a TR606, which has the same DC/DC, so I could do some measurements.
The DC/DC delivers pretty exact 15V.
Looking up the data sheet that is to be expected from the RD15EB3 Z-Diode D42.
For maintaining 11.8V the AN1458 needs 13.3V. That leaves 1.7V at R172. Bringing Ohms law in: that is 17mA.
Which is in the range we have to expect as current consumption!

So this leads to the conclusion, that a slightly "dipping" 12V is not rare, but more or less common between original TB303.
And any sound comparison, especially concerning the supply, gets completely useless w/o measured values of the nominal 12V output and the supply at pin 8 of IC16.

Please keep in mind: the tolerances, given in the data sheet, are for all allowed operating conditions. At room temperature they are already tighter.
Plus: the manufacturer does not want to have much "out of specs" parts being sorted out. So cheap mass products are typically a lot better than stated, because only then there is a high yield.
On the bad side this night be exploited by the engineer: Use cheap components who will meet specs most of the time and sort out the very few units where it hasn't worked out. Here we have the R178/R179 voltage divider as example, where you might not calculate for the full +5% -5% worst case.

Posts: 99
Joined: Sat May 23, 2015 3:14 pm

Re: Power Supply Modding

by Nordcore on Mon Oct 30, 2017 6:21 am

As we have learned, the "loss of regulation" mode might be relative common, it might be interesting for some to exploit that sound effect.

... so here is the schematics to mimic the behavior of some(!) original TB303 within a x0xb0x.

psu-mod.png (41.48 KiB) Viewed 1672 times

C8 is 100µF from the x0xb0x BOM. As long as the supply voltage for the AN1358 "regulator" IC23 is high enough to maintain regulation under all load conditions, the value does not change anything in a pretty wide range.
But as we are going to exploit the loss of regulation "feature" seen on some original TB303, the time constant R1*C8 gets relevant and you might want that to be the same as in the original circuit.

Pre-Trim: Adjust RX3 for 15V. (with your DMM)

To set up RX3 for "rare" loss of regulation:
connect a scope to IC23-Pin 8. Set scope to AC input mode, about 500mV/div, pretty slow. Connect second channel of scope with same settings at IC23-Pin1.
Let the x0x play a pattern. High resonance, some accent, accent turned up fully.
You'll see the voltage at channel 1 moving slightly up and down with the varying load.
Adjust RX3 so that channel 2 shows a few dips in the flat line.

Posts: 99
Joined: Sat May 23, 2015 3:14 pm

Re: Power Supply Modding

by antto on Mon Oct 30, 2017 1:57 pm

tbh i think it might actually be a good idea to keep the adapter 9VAC.. because all of us already have one
but it would be nice if we can change the PSU so that it can safely run also on a DC adapter
specifically i'm worried about the way the 7805 is fed, from what is effectively a capacitive voltage divider

perhaps maybe we could/should change that to a switching 5V regulator that has a huge input voltage range
maybe another LM317 configured to output 5V, and feed it from the "about 24V" positive rail too
C5 and C3 will then be rearanged to the positive rail and ground (as reservoirs for the inputs of the regulators), and they could be changed to something like 3300uF/35V and 1000uF/35V respectively
hm.. 3300uF/35V would be kinda huge i suspect

EDIT: oh wait, the x0x PSU actually does some weird voltage doubling thing
9VAC rectified would give 12.73VDC peak according to my calculator

Posts: 1636
Joined: Thu Apr 15, 2010 3:21 pm

Re: Power Supply Modding

by Nordcore on Tue Oct 31, 2017 2:49 am

Yes, it is a voltage doubler.
I've annotated the measured values in my schematic: 11 and 24V.
As the load is pretty different for both paths, the voltages are not exactly where the ballpark calculation predicts them...
(On my x0x I use classic red LEDs @ 10mA each, that gives some serious(!) current consumption ... or, at least: the heat sink for the 7805 is really needed for worst case calculation ... )

Using 9V AC gives about 12V DC.
We don't need large caps than, as there is plenty drop out voltage for the regulator.
(about 470µF should be enough. To large caps might cause problems on DC wall warts, I would even suggest a small resistor in the input ... )
With 9...14V/AC/DC input we will need a switching regulator for the 5V.
(Example: https://www.mouser.de/ProductDetail/REC ... lDCSDFs%3d )

But where to get the 6V for the headphone amp? ( ... lifting the RECOM by 1V, so it delivers 6V, than low dropper to 5V? )

The 15V for the AN1458 is still easy: https://www.mouser.de/ProductDetail/CUI ... L1Mw%3d%3d
... starting from the well regulated 5V *and* having the AN1458 "regulator" following, allows the use of a non regulated switcher.

Posts: 99
Joined: Sat May 23, 2015 3:14 pm

Re: Power Supply Modding

by antto on Tue Oct 31, 2017 1:20 pm

how about the AC/AC-or-AC/DC adaptor option?
as far as i can tell, the voltage doubler causes the trouble

what if we ditch the doubler, use a normal full-rectifier, and then a step-up switching regulator to get something like 20-ish volts
then stick the LM317 for the adjustable 15-ish volts, the LM7805, etc..?

then the input voltage range will be bigger in theory, so we can hopefully use the 9VAC adapter or a 12VDC adapter or so

Posts: 1636
Joined: Thu Apr 15, 2010 3:21 pm

Re: Power Supply Modding

by antto on Fri Jan 19, 2018 6:19 pm

so i found some time and started putting together the schematic
i followed some of your suggestions

i slapped a full rectifier
- a supposedly "weak" switching regulator for the adjustable "15-ish" voltage, which will feed the opamp "regulator"
- a switching regulator for the 5V stuff
- the same old LM7806 for the 6V
changed the MIDI opto-isolator.. while looking into 6N138, i found suggestions that "PC900" might be a better option, and that "H11L1" is an equivalent, but i'm not sure if i wired it up correctly
i also added a "4050" buffer.. maybe i should have used two individual buffer units, one for driving the midi thru and another for the J3 path
used the FTDI chip you suggested (in SSOP16), and added the opto stuff, again, not sure if i wired it properly

there are a few more small questions here and there


Posts: 1636
Joined: Thu Apr 15, 2010 3:21 pm

Please be positive and constructive with your questions and comments.