Quiet & low cost +/-V rails

[fcb]

I'm working on a project at present that is powered from 4xAAA cells (i.e. 3.7 to 5.7V after diode drop), the analog sections need +/-6V at around 15-25mA (less on the negative rail).

My first iteration used 3x ICL7660's and an LDO in the following config:
2x ICL7660's as voltage doublers in parallel (giving a +6.8v to +10.8v rail)
1x LP2951 as regulator fed from new rail giving 6.0V
1x ICL7660 as voltage inverter on the output of the LP2951 (giving -6V rail)

Most of the electronics in the unit is audio, so I'm keen to keep the noise down.  It just seems like an expensive way of doing things.

I did contemplate building a pair of anti-phase charge pumps out of 74HC04's (driven off 24KHz, derived from a 48KHz clock)? Anyone got a smart (low cost and small board space!) idea...?

[Precipice]

I've just done a design using an LT1930ES5 to generate similar rails. P8 of the datasheet has the +-generator I ripped. I dropped the diodes down to BAT54S, since I don't need the full current and they're already on my BoM.
No idea yet how noisy it is - but switching frequency is out at 1.2MHz (or you can get the 2.2MHz 'A' part), so it should be safely out of the way of audio.
$2 at 100-off, so not super-cheap (but unusually good for an LTC part!). Certainly compares well with a pile of ICL7660s unless they've suddenly got cheap. It's tiny, too (SOT23) - and the support components are small, which was a driver for me, but you don't say.


Edit: you do say small. In that case, I reckon this is worth a look.  If you're doing audio stuff, I really don't think you want to be deriving your PSU from 48KHz. That way lies much annoyance.

[void_error]

I did contemplate building a pair of anti-phase charge pumps out of 74HC04's (driven off 24KHz, derived from a 48KHz clock)? Anyone got a smart (low cost and small board space!) idea...?

You could indeed use a 74HC04 (or MC14584, same thing but works up to 15V) and boost its output current using a PNP-NPN pair or low power MOSFETs.

Add the voltage doubler / voltage inverter bits of circuitry to that and you only have to use one chip. 48kHz is a waste of space because you'll need higher value capacitors, go for around 300kHz.

[cellularmitosis]

I recently worked trying to filter out the noise produced by an ICL7660: https://www.eevblog.com/forum/projects/an-evening-with-the-icl7660/

[fcb]

Precipice:  That's a nice part, I think it's a little bit too expensive though (I can't find in stock and for $2).  I have to say, the idea of the small footprint does appeal.

void_error: I'm probably going to try a 74HC04 on the bench today, I choose 48KHz as I have that clock (with a nice 50% duty-cycle) available nearby, certainly I could take the BCK (3.072Mhz) and feed it into a /8 and get 384KHz which might push the noise up the band - I worry about diode losses increasing..  Perhaps i'll try that as well.

cellularmitosis: Nice other thread link, interesting stuff.  I like the faraday pan...

One other idea is to use a 555 as the driver, nice powerful output stage.  Will report back my chosen solution when i've spent some time on the bench.

[Precipice]

Precipice:  That's a nice part, I think it's a little bit too expensive though (I can't find in stock and for $2).  I have to say, the idea of the small footprint does appeal.

Ah - I've got a reel of 2000 here, if you want a smallish number to prototype with.

And, from bitter experience - running your switchers well outside the audio range (and dodging AM radio, to be kind to others), makes filtering so much easier (smaller, cheaper).

Also - bear in mind there are better versions of the 7660, which can actually deliver useful current. Nat Semi's LM2662MX f'rinstance.

[fcb]

LM2662MX runs upto 5.5v, I have to cater for at least 5.7v (probably safer to assume someone inserts freshly charged NiMh's in the unit which can sit higher than 1.5v).  Nice part though.

Thanks for the offer of a few parts to prototype with, I should be OK - it'll require some serious deadbugging or a PCB to prototype that SOT23-5 properly.  Any other 'goto' HF SOT23 boost regulators out there?

Also, is there a 'goto' LDO -V adjustable regulator (negative version of the LM1117)?

[Precipice]

LM2662MX runs upto 5.5v, I have to cater for at least 5.7v (probably safer to assume someone inserts freshly charged NiMh's in the unit which can sit higher than 1.5v).  Nice part though.

Ah, yeah, that would be brave...

>Thanks for the offer of a few parts to prototype with, I should be OK - it'll require some serious deadbugging or a PCB to prototype that SOT23-5 properly.  Any other 'goto' HF SOT23 boost regulators out there?

Not that I've used - but stick your requirements into LTC's and TI's selection / design forms, and see what spits out. No sane person can keep track of all their offerings!

>Also, is there a 'goto' LDO -V adjustable regulator (negative version of the LM1117)?

Yeah, there's a '3' version, same as there is for most of the rest of the sizes. They're far from low noise, though! (and low noise candidates don't fit on the same footprints). If you can't face buying or building proper low noise regulators, you can clean them up a lot, for not much effort - search for noise eater. Here's a sample.
http://www.techlib.com/electronics/finesse.html
It's cheaper and smaller than trying to tame the noise with ever bigger and better capacitors...
TI have a load of low noise LDOs, if you just want to buy something and stick it down.

[David Hess]

If you are up to winding a small transformer for operation at your 48 kHz, then low noise regulated bipolar outputs can be produced by regulating the input voltage before chopping.

[fcb]

Precipice: I'm familiar with the LM337, but I can't seem to find a low-drop out version (in a SOIC package). Interesting link you posted, a "noise eater", it looks like a recipe for an oscillator - perhaps I'm reading it wrong though - does it work by creating a shunt regulator that's controlled by the noise on the line??

I suspect I will have to build a discrete regulator for the -ve side.

David_Hess: I've done a few transformer based supplies based on 48/96KHz switching (generating multiple isolated supplies in audio gear), I don't fancy winding a transformer small enough for my application though.

[Precipice]

Precipice: I'm familiar with the LM337, but I can't seem to find a low-drop out version (in a SOIC package).

LM337LM/NOPB - http://uk.rs-online.com/web/p/linear-voltage-regulators/5361322/
I don't know how low dropout you need, but it's not dreadful. TI have LM337L, with some shockingly sparse data!

Interesting link you posted, a "noise eater", it looks like a recipe for an oscillator - perhaps I'm reading it wrong though - does it work by creating a shunt regulator that's controlled by the noise on the line??

Yeah, with a series resistor before it. I've never used one - but they seem well regarded by low-noise obsessives.

>I suspect I will have to build a discrete regulator for the -ve side.

It's not the end of the world...
However, TI have a decent slew of -VE LDOs, including some claiming ultra low noise...
http://www.ti.com/lsds/ti/power-management/linear-regulators-ldo-products.page#p32=1&p182=Negative Output

Edit:URL fixed, and no, none of them are low noise.

[David Hess]

Precipice: I'm familiar with the LM337, but I can't seem to find a low-drop out version (in a SOIC package).

They are not as common and I have never found a low dropout direct replacement for the LM337 either.  This is close:

http://www.linear.com/product/LT1175

Interesting link you posted, a "noise eater", it looks like a recipe for an oscillator - perhaps I'm reading it wrong though - does it work by creating a shunt regulator that's controlled by the noise on the line??

That is exactly how it works although I have seen transformer coupled series pass versions as well.

David_Hess: I've done a few transformer based supplies based on 48/96KHz switching (generating multiple isolated supplies in audio gear), I don't fancy winding a transformer small enough for my application though.

I am sure Coilcraft makes one intended for low power isolated supplies which would be suitable.  They are used for galvanically isolating various interfaces like Ethernet, RS-232, and USB.  They are also used for isolated gate drivers in larger power supplies.

There are all kinds of dual output switching regulator designs which could be used but the filtering needed for low noise raises the cost.

[RiverTown]

Best way to search for an ICs, in this case negative LDO, is to use yours supplier parametric search.
There are dozen different - LDOs, but that don't mean that your supplier have all them in stock.
I must warn you negative LDOs aren't very cheap, I think that it is hard to find one that is under 1- 1.5 bucks in one of quantities.

[fcb]

Best way to search for an ICs, in this case negative LDO, is to use yours supplier parametric search.
There are dozen different - LDOs, but that don't mean that your supplier have all them in stock.
I must warn you negative LDOs aren't very cheap, I think that it is hard to find one that is under 1- 1.5 bucks in one of quantities.

Yes. I'm familiar with my favorite suppliers parametric search....

Yes. There are a few -ve regulators about, just not small LDO variants..

No. The bulk of the negative regulators are cheap (not quite as cheap as +ve), but they mainly seem to be LM337 or 79x variants.

[Precipice]

Out of interest, how low a dropout do you need?
(and why are we still discussing all this linear stuff? Fast switchers make for easy filters!)

[fcb]

Dropout voltage:
It will depend on the final PSU topology somewhat, but my *guess* is

(3.7v * 2) - 2 * Vfd = 6.8v (vfd is around 0.3v, also assume cells at 1.0v each)
So at best, the worst case scenario is -6.8v, although the output Z of ICL7660 will reduce this.

I'm after -6v, so probably 0.5v dropout as a *guess*.

[Precipice]

Oh crumbs, proper-low.
Go on, ditch the 7660s... 
Don't forget that, if your current requirements are low, diode/cap inverters/doublers/triplers are cheap and small, if not terribly efficient). You'd only need one high power driver (C555?) and a bunch of passives (and then you could maybe afford some headrooom for your regulators)

[RiverTown]

Yes. There are a few -ve regulators about, just not small LDO variants..

Who is your supplier, it is weird that they don't have any negative LDOs in stock?
Take a look at Farnell website, yes they are expensive supplier, but they have few of them.

No. The bulk of the negative regulators are cheap (not quite as cheap as +ve), but they mainly seem to be LM337 or 79x variants.

Nor LM337 nor /9xx series are the LDO regulators. Linear negative regulators- yes, LDO- not a chance.
I were using TPS72325 in my last project. You can't say that it's cheap in small quantities.
http://export.farnell.com/texas-instruments/tps72325dbvt/regulator-ldo-fixed-2-5v-sot-23/dp/2323571

[f5r5e5d]

for the V multiply, inverting there should be some single chip inductive boost solutions with off the shelf coils

I used a single SOT-23-5 boost converter chip with dual winding smt toroid like http://www.bourns.com/data/global/pdfs/pm3600_series.pdf for split supplies in a similar I,V range

you can buy dual switcher in a chip too now LT1944 is one, or yet another approach is http://cds.linear.com/docs/en/design-note/dn223f.pdf

in my application the switcher was on board, within an inch or two of 12 bit ADC, Av 30 input amp/active filters for the signal with no problems (of course with added linear Vreg of the supplies)

[David Hess]

At 10s of milliamps and below, there are a number of operational amplifiers which can be used as low dropout negative regulators directly without a pass element.  Even with an external pass transistor, short circuit protection would not be required if the current is limited at the input because of supply impedance or current limiting.

[fcb]

RiverTown:  The TPS72301 (adjustable version of the part you suggested), looks ideal - it's a bit pricey though.

Precipice: The advantage of the 7660 on the negative rail is the lack of diode losses.

f5r5e5d: Nice bourns part, a but big though - I've ordered a few for my samples draw.

David Hess: A possibility, bound to be cheaper than a TPS72301, could probably use a pass element to help in the selection, I wonder if the LM358 could be used.

[Precipice]

Precipice: The advantage of the 7660 on the negative rail is the lack of diode losses.

Yeah, but they're huge (SO-8) and slow, and need big capacitors for the flying caps and the output cap, followed by filtering out all the audio frequency noise, and even then, you've got precious little headroom for your LDO, especially given its dreadful output impedance. A tripler would give you enough headroom for a cheaper Vreg, even after the diode losses. Less of a fight that way, but you're the one doing the project

[RiverTown]

I don't know what you want to power with this supply. Maybe you can just use boost switching (high frequency) converter, positive LDO and virtual ground.
This would drastically decreased costs (no need for voltage inverter and negative LDO).

[Mechatrommer]

just a quick thought... family of mornsun A_S-2W & B_S-2W Series. one example A0505S-2W

[David Hess]

David Hess: A possibility, bound to be cheaper than a TPS72301, could probably use a pass element to help in the selection, I wonder if the LM358 could be used.

An LM358 has a Vbe drop when sinking current and is not all that powerful so its voltage drop would be about 2 volts at 20 milliamps according to its specifications.

An LM7321/LM7322 or LM7331/LM7332 (single/dual) look to be ideal because they are fast for high PSRR and have high output current and unlimited capacitive load drive.  Linear Technology has some suitable models as well.  The dual parts could be used to regulate both rails with a suitable reference.  I might try to design a true bipolar tracking supply this way if it was advantageous.

RiverTown mentioned using a single regulator and virtual ground which these could also do well although there may be better parts for that.

I am still more inclined to using a low power isolation transformer with active regulation of the sine or square drive on the input side for inherently low noise but would have to consider the above despite having designed low noise switchers and charge pumps in the past.