TO220 rail splitter?

[0xdeadbeef]

I recently stumbled over a Chinese dual voltage module on eBay (see attached picture) that seems to consist only of a cheap boost converter (LM2577 clone), some capacitors and a five pin TO220.
As it claims to create a dual voltage output (with up to 3A), I guess this TO220 must be a rail splitter. Like a beefy version of the TLE2426 which would be very handy for rail splitters demanding more than 80mA.
So is there such a thing?

[capt bullshot]

I'd guess it's a power OpAmp. One can buy them in said 5 pin TO220 and make a rail splitter out of them using two additional resistors.

[0xdeadbeef]

As far as I can tell, there is only one resistor on the PCB and that is used for the LED.

[Zero999]

As far as I can tell, there is only one resistor on the PCB and that is used for the LED.

There's only one through hole resistor, but there appears to be many surface mount resistors on that PCB.

[0xdeadbeef]

Ah, crap. While there are no SMD resistors on the top (of the main PCB), there are indeed four resistors (3x10k, 1x1k) on the bottom.
So yeah, it's probably really an OpAmp.

[EDIT]
Indeed the "typical application - split supply" circuit in the LM675 datasheet shows a circuit with four resistors (well actually five, but let's ignore the 1Ohm resistor).

[CJay]

Curious how much that costs, have you a link?

I'm wondering if it's worth buying for the chips.

[0xdeadbeef]

It's sold as "DC 3V-30V USB 5V To 3V-18V 3A Adjustable Dual DC Regulator Power Supply Module" on eBay for 10.35€ or around £9 in the UK.
German eBay link
https://www.ebay.de/itm/DC-3V-30V-USB-5V-To-3V-18V-3A-Adjustable-Dual-DC-Regulator-Power-Supply-Module/183453477426?ssPageName=STRK%3AMEBIDX%3AIT&_trksid=p2057872.m2749.l2649
UK eBay link
https://www.ebay.co.uk/itm/DC-3V-30V-USB-5V-To-3V-18V-3A-Adjustable-Dual-DC-Regulator-Power-Supply-Module/262489885410?hash=item3d1d9dcae2:g:z8YAAOSwUl5a~pIv:rk:1:pf:0

[spec]

Hi  0xdeadbeef,

There is no rail splitter as such. The circuit accepts a 3V to 30V input and outputs two power rails. One power rail is plus and the other power rail is minus but they are both the same voltage magnitude of between 3V to 18V. The circuit will be a simple flyback converter, probably oscillating at 50kHz, using the LM2577 'simple switcher' SMPSU chip.

It is a very useful function and used widely. There are numerous similar functional units on the net, all at very reasonable prices.

[0xdeadbeef]

I'm not sure what you're trying to tell me. Of course this is a rail splitter, it's just non an integrated one like the TLE2426 which is actually also just an OpAmp and a trimmed voltage divider on the same die.
So the positive and negative voltage (related to the virtual ground) won't be as symmetrical as those created with a TLE2426, maybe there's even a temperature drift. It will work, but worse than an integrated solution.
Or do you question the term "rail splitter" (which also TI uses for the TLE2426)?
And yes, as mentioned in my original post, there is a simple boost converter (soldered) on that Chinese module which is probably based on some LM2577 clone. Actually I have a bunch of them lying around, some even in a SEPIC configuration (with two inductors).

[Zero999]

I'm not sure what you're trying to tell me. Of course this is a rail splitter, it's just non an integrated one like the TLE2426 which is actually also just an OpAmp and a trimmed voltage divider on the same die.
So the positive and negative voltage (related to the virtual ground) won't be as symmetrical as those created with a TLE2426, maybe there's even a temperature drift. It will work, but worse than an integrated solution.
Or do you question the term "rail splitter" (which also TI uses for the TLE2426)?
And yes, as mentioned in my original post, there is a simple boost converter (soldered) on that Chinese module which is probably based on some LM2577 clone. Actually I have a bunch of them lying around, some even in a SEPIC configuration (with two inductors).

I don't see why the LM675 will be inferior to the TLE2426. The TLE2426 is specified to have a tolerance of <1%, which should be easily achievable with the LM675, given close enough tolerance resistors.

[David Hess]

There is such a thing as a switching rail splitter.  A buck converter with synchronous rectification will produce half of its input voltage at a 50% duty cycle whether sourcing or sinking current.  In practice, feedback is used to tightly control the output but this is not absolutely necessary.

For practical purposes, an integrated class-D audio amplifier is probably the easiest way to implement this now.

Linear Technology made some switching regulators for termination applications which worked this way including some high voltage ones.  Note that when the output current is negative, the switching regulator pumps power back into the supply.

https://www.analog.com/en/ltc3717

[GigaJoe]

exactly what I thought , bump voltage to +40, then use audio power amp to split it,
but I not sure D-class in TO220 , not enough pins.   it maybe TDA2030A clone , that cost $0.10
then we definitely fit in $2.5 budget

[0xdeadbeef]

I don't see why the LM675 will be inferior to the TLE2426. The TLE2426 is specified to have a tolerance of <1%, which should be easily achievable with the LM675, given close enough tolerance resistors.

First of all I'm not even sure which OpAmp is used there. I picked the LM675 just as an example.
Anyway, for sure the resistors on this Chinese board are neither hand selected nor pre-aged or whatever, so this circuit will behave worse than a TLE2426 for sure.

About the TLE2426 tolerance:  there's a table in the datasheet of the TLE2426 which says the error (i.e. deviation of Vout/Vin from 50%) is practially 0 for input voltages around 5V (Vin) and then linearly raises to 1% at 15V but keeps raising to 3.6% for 40V.  So I'd say the initial total error is supposed to be <1% only for voltages below 16V. Anyway, this linear voltage dependency of the total error seems to indicate that it's not caused by the trimming of the resistors. Also note that the temperature dependency is very low which is certainly also a result of the integrated voltage divider.

So yes, I'm still skeptical that it's a piece of cake to achieve the same error with a discrete voltage divider, specifically under temperature/load changes and considering aging.

[Zero999]

I don't see why the LM675 will be inferior to the TLE2426. The TLE2426 is specified to have a tolerance of <1%, which should be easily achievable with the LM675, given close enough tolerance resistors.

First of all I'm not even sure which OpAmp is used there. I picked the LM675 just as an example.
Anyway, for sure the resistors on this Chinese board are neither hand selected nor pre-aged or whatever, so this circuit will behave worse than a TLE2426 for sure.

About the TLE2426 tolerance:  there's a table in the datasheet of the TLE2426 which says the error (i.e. deviation of Vout/Vin from 50%) is practially 0 for input voltages around 5V (Vin) and then linearly raises to 1% at 15V but keeps raising to 3.6% for 40V.  So I'd say the initial total error is supposed to be <1% only for voltages below 16V. Anyway, this linear voltage dependency of the total error seems to indicate that it's not caused by the trimming of the resistors. Also note that the temperature dependency is very low which is certainly also a result of the integrated voltage divider.

So yes, I'm still skeptical that it's a piece of cake to achieve the same error with a discrete voltage divider, specifically under temperature/load changes and considering aging.

The op-amp makes very little difference. The offset errors will be tiny, compared to the supply voltage.

The resistors don't need to be hand selected or anything special to get equal performance to the TL2426. A power op-amp with 1% resistors will give similar performance, at low currents and will easily beat it, at high currents.

The op-amp isolates the load changes from the potential divider and both resistors should have a similar temperature co-efficient.

[0xdeadbeef]

The op-amp makes very little difference. The offset errors will be tiny, compared to the supply voltage.

Well, not all OpAmp have small output offsets. E.g. the LT1010 (which is also used as rail splitter and the datasheet shows an according circuit) defines a maximum of 150mV (even 220mV over full temperature range).
Also bias currents etc. differ. As a side note, the 1mV offset given for the LM675 seems to be the input voltage offset (i.e. the voltage needed to get 0V at the output). IMHO this is not a definition of the worst case output offset.

The resistors don't need to be hand selected or anything special to get equal performance to the TL2426. A power op-amp with 1% resistors will give similar performance, at low currents and will easily beat it, at high currents.

Well, this adds another 1% (or so) error to the total error.  My understanding is that the trimming of the TL2426 removed this error more or less completely (i.e. to 0.1% or so).

The op-amp isolates the load changes from the potential divider and both resistors should have a similar temperature co-efficient.

But there are other errors introduced by a real OpAmp. E.g. there are bias currents going into the inputs which are voltage and temperature dependent.
Actually I would think that error increasing linearly with the voltage on the TL2426 is mainly caused by the input bias current(s).

[David Hess]

Unlike the earlier and slower LM302 and LM310 voltage followers which internally use a symmetrical differential input stage for low offset voltage, the LT1010 power buffer relies on matching of an NPN and PNP pair and is not an operational amplifier and does not function like one.  Widlar's design takes advantage of a "stored charge" PNP and "isolation base" NPNs to wring as much performance out of a junction isolated NPN process as possible.

National also made some even faster buffers but they used hybrid construction to take advantage of complementary circuit designs and were much more expensive.  Even National made the distinction between voltage followers and buffers.  Grey beards might recognize these as the "Fast" and "Damn Fast" buffer amplifiers.

[0xdeadbeef]

You're right that the LT1010 is not a normal OpAmp but a unity gain buffer which is also why there is no negative input pin. So it's an OpAmp with a fixed negative feedback (and thus a gain of 1).
Again, if you look for TO220-5 OpAmps on Mouser, there are not too many choices. Actually it boils down to the LM675, the LT1010 and the OPA544, where the OPA544 is the only one that doesn't have a rail splitter circuit as application example. Doesn't seem to be meant for unity gain circuits.

Anyway, it seems that the TLE2426 has exceptionally high input bias currents which degrade its performance at higher voltages. So yeah, as the LM675 has much lower input bias currents, it might outperform the TLE2426 for input voltages higher than 15V even if only using 1% resistors. E.g. at 30V, the TLE2426's deviation from 50% is already >2%.

[Zero999]

Have you done any calculations?

If the total power supply voltage is 20V for a +/-10V rail, a 1% error is 100mV. Look at the the data sheets for half decent op-amps and you'll find that the offset voltage error will be an order of magnitude less than that. The effects of the bias offset currents can be catered for by not using stupidly high value resistors. The error due to the op-amp will  be insignificant.

The error due to the resistors will be under 1%, if they're in the same batch, they will be fairly well matched.

An op-amp + potential divider can easily beat the TLE2426.

[0xdeadbeef]

Hm. I feel I will repeat myself at some points but in a nutshell: the idea was not to replace the TLE2426 with some OpAmp that can deliver even less current but with a beefy TO220 one.
When looking for TO220-5 OpAmps on Mouser, it comes up only with the LM675, the LT1010 and the OPA544.
The OPA544 doesn't seem to be unity gain stable and the LT1010 (which is actually a unity gain buffer) has a significant offset error (and can only deliver only +/-150mA anyway). Actually there is a 0.3V offset when you simulate the recommended application circuit in LTSpice with the official LT1010 model.
So this leaves us with the LM675. I simulated the recommended application circuit in LTSpice and it doesn't show any significant error there but then again I can't tell how realistic the model is (taken from some opamp.lib with an author tag "rperez"). And of course there is no temperature model and no worst case model and what not.
But yes, as stated in my last posting, it looks like the total error of the TLE2426 will exceed that of an LM675 with 1% resistors for input voltages >15V.
So for a typical +/-12V or +/-15V scenario, the LM675 should give better results if the total error stays below 2% as the simulation etc. suggest(s).

[David Hess]

You're right that the LT1010 is not a normal OpAmp but a unity gain buffer which is also why there is no negative input pin. So it's an OpAmp with a fixed negative feedback (and thus a gain of 1).

No!  The LT1010 is not an operational amplifier and does not function as one!

The LM302 and LM310 voltage followers are operational amplifiers internally configured for unity gain.  The LT1010 is a series connected complementary emitter follower.  The LH0002 is a parallel pair of series connected complementary emitter followers operating in class-AB which some will recognize as a diamond buffer; these are easy to implement with 4 discrete transistors and form the input structure of many current feedback operational amplifiers and some high slew rate voltage feedback operational amplifiers.

LM302 Voltage Follower 10V/us 50nA 20mV 0.9985V/V
LM310 Voltage Follower 30V/us 10nA 10mV 0.999V/V
LT1010 Power Buffer 75V/us 800uA 220mV 0.995 V/V
LH0002 Diamond Buffer 200V/us 10uA 30mV  0.95V/V

Again, if you look for TO220-5 OpAmps on Mouser, there are not too many choices. Actually it boils down to the LM675, the LT1010 and the OPA544, where the OPA544 is the only one that doesn't have a rail splitter circuit as application example. Doesn't seem to be meant for unity gain circuits.

There used to be a few more.  There are ways to wire dual and quad operational amplifiers in parallel for more output current but this does not help much with limited power dissipation.  Integrated class-AB linear audio amplifiers were particularly useful for this application but have largely been replaced with class-D devices.

Circuit stability is often a problem.  A power splitter should have low AC output impedance which implies adding an output capacitor to either or both supplies but some devices will become very unhappy with this unless it is carefully done.  The standard techniques for driving a capacitive load help but raise mid frequency impedance.  Swamping the output with like a 10 microfarad solid tantalum or aluminum electrolytic capacitor often provides the best performance.

[CJay]

Given that those power op-amps are not cheap, what's the betting it's actually something like an LM1875 or TDA2030...

[Zero999]

Hm. I feel I will repeat myself at some points but in a nutshell: the idea was not to replace the TLE2426 with some OpAmp that can deliver even less current but with a beefy TO220 one.
When looking for TO220-5 OpAmps on Mouser, it comes up only with the LM675, the LT1010 and the OPA544.
The OPA544 doesn't seem to be unity gain stable and the LT1010 (which is actually a unity gain buffer) has a significant offset error (and can only deliver only +/-150mA anyway). Actually there is a 0.3V offset when you simulate the recommended application circuit in LTSpice with the official LT1010 model.
So this leaves us with the LM675. I simulated the recommended application circuit in LTSpice and it doesn't show any significant error there but then again I can't tell how realistic the model is (taken from some opamp.lib with an author tag "rperez"). And of course there is no temperature model and no worst case model and what not.
But yes, as stated in my last posting, it looks like the total error of the TLE2426 will exceed that of an LM675 with 1% resistors for input voltages >15V.
So for a typical +/-12V or +/-15V scenario, the LM675 should give better results if the total error stays below 2% as the simulation etc. suggest(s).

Then use a standard op-amp with pair of transistors on the output.

Circuit stability is often a problem.  A power splitter should have low AC output impedance which implies adding an output capacitor to either or both supplies but some devices will become very unhappy with this unless it is carefully done.  The standard techniques for driving a capacitive load help but raise mid frequency impedance.  Swamping the output with like a 10 microfarad solid tantalum or aluminum electrolytic capacitor often provides the best performance.

Yes, I would think a large capacitor on the output is a good idea. An inductor with a resistor in parallel might also help to isolate the output capacitance from the op-amp, at high frequencies.

[0xdeadbeef]

For the record: GigaJoe was right: the mystery TO220 is a TDA2030 class A/B audio amplifier.

[Zero999]

For the record: GigaJoe was right: the mystery TO220 is a TDA2030 class A/B audio amplifier.

That doesn't surprise me.

Incidentally, it should be easy to obtain better accuracy than the TLE2426 using the TDA2030 and 1% resistors.

Did you try reverse engineering it? If so, please post the schematic.

[0xdeadbeef]

It just arrived today and I only quickly tested it. I will post a schematic of the virtual ground circuit as soon as I find the time.

The embedded step up converter is based on the XLSemi XL009 with a frequency of 400kHz and a built-in N-FET switch.
So it actually seems to be an enhanced version of the LM2577 with an official branding and not a simple counterfeit. I actually like that since thus there's a real datasheet etc.
There's a 220µF cap at the input and a 100µF cap at the output. The voltage feedback is done with a voltage divider formed by a 330Ohm SMD resistor and a 10k trimmer (W103).
Anyway, there's nothing totally surprising there I guess as these modules are flooding eBay etc.

What's quite interesting is that the DC/DC board is actually placed in a cutout of the virtual ground board. So you could even remove the DC/DC completely.