What to do with noisy poorly regulated USB +5V?

[Smokey]

Hey guys,

The issue of the USB +5V having a low tolerance (5% according to wikipedia) strikes me as a problem that many people will end up having to deal with when they try to use USB to directly power an analogy project.  I thought it would be cool if we could go over some of the options to solve it and find out what people thinks works best.

My current situation:
Up until now the only time I ever used the +5V straight from a USB cable was to power some digital chips that didn't really care how clean the rail was. 
I have a new board I want to run off USB power but it has an analog chip that internally uses 1/2 the 5V supply as a reference for the analog output.  The low regulation tolerance of the USB 5V is going to cause lots of problems in this case since I'll never know where the output midpoint will be.  I know in this case there are ways to shift the analog output based on the actual 5V voltage, but having a properly conditioned 5V to start with is probably a better solution.

There appear to be a couple popular options for getting a solid 5V from the unpredictable USB power rail.
1)  Use a boost converter to bump the USB 5V up to something around 6.5V that an LDO can handle and then use another regulator to drop that back to a solid 5V.
2)  Use a dedicated USB power conditioning chip like the linked semtech
3)  Something completely different.

There are obviously trade offs with each solution like cost, parts count, board space required, and source current levels.  What you end up using will vary depending on your project requirements, but having a lot of possible options in one place on this thread will make it easier to evaluate what's out there.
Who has some experience with this issue?  What solution have you used and what would you recommend?

Here are a some interesting app notes for solutions I found after a quick search.  I haven't used any of these so I can't say how well they work.
http://www.intersil.com/content/dam/Intersil/documents/an13/an1317.pdf
http://www.semtech.com/images/datasheet/an-pm-1007-5v-to-5v_ag.pdf

[amyk]

I have a new board I want to run off USB power but it has an analog chip that internally uses 1/2 the 5V supply as a reference for the analog output.

Does that chip support an external reference? You could just use a 2.5v linear regulator in that case.

Also, from what USB source are you going to be powering from? A cheap AC adapter will certainly be no good, but if it's plugged into a PC the tolerance tends to be much better than 5%. The PC I'm currently writing this on has the 5V rail a little less than 2% high (5.08V).

[Bored@Work]

it has an analog chip that internally uses 1/2 the 5V supply as a reference for the analog output.

You need to quantify that in more detail. You need to know how good the power supply needs to be, or how bad it can be, while still getting the desired performance, in order to be able to decide which of the possible ways to stabilize/regulate some USB power supply is sufficient.

[dannyf]

The issue of the USB +5V having a low tolerance (5% according to wikipedia) strikes me as a problem that many people will end up having to deal with when they try to use USB to directly power an analogy project.

Why?

You don't want to end up solving that problem that doesn't exist.

[Frant]

There appear to be a couple popular options for getting a solid 5V from the unpredictable USB power rail.
1)  Use a boost converter to bump the USB 5V up to something around 6.5V that an LDO can handle and then use another regulator to drop that back to a solid 5V.
2)  Use a dedicated USB power conditioning chip like the linked semtech
3)  Something completely different.

There are obviously trade offs with each solution like cost, parts count, board space required, and source current levels.  What you end up using will vary depending on your project requirements, but having a lot of possible options in one place on this thread will make it easier to evaluate what's out there.
Who has some experience with this issue?  What solution have you used and what would you recommend?

Requirements and constraints of the particular application (which are currently unknown to us) will determine the way of dealing with the problem. Generally, there are at least two possibilities:

1. If you can get away with 3.3 V instead of 5 V, do it. There are excellent low noise voltage regulators such as TPS7A4700, TPS7A49xx, ADP150, TPS796xx etc. Be careful not to use older chips like 78xx or LM317, which are neither low dropout nor low noise and therefore wouldn't do the job.

2. If you absolutely need 5 V, what would be the maximum current? For a low current demand you can use a switched-capacitor solution such as the TPS60241 (up to 25 mA). For a higher current you can use a step-up (boost) converter followed by a fast, low noise linear regulator (one of the above mentioned or similar).

[peter.mitchell]

how about just good ole filtering to clean it up and use a shunt regulator.

[richcj10]

I am doing a simaler project.
I used  a pie filter on the input at then used a low noise 3.3v reg.

[Smokey]

good points.
I probably shouldn't have given my example at all.  I was trying to be vague on my particular application on purpose to collect a range of some general purpose solutions to a problem that I thought more people would be having.  I guess I didn't turn up many search results on the topic for a reason.  Wouldn't be the first time I was wrong.  I've already changed my circuit and selected some new components to step around the problem by running everything on 3.3V, but I can see situations where that would not be possible and a solid consistent 5V would be needed.  Not everything has a 3.3V equivalent yet.

Even if one specific PC is dead nuts on 5.0000V doesn't mean you can rely on that.  The spec is (once again according to wikipedia, I've read 10% elsewhere) 5% tolerance so you should really design for the worst case.  If it comes in high there are more ways you can filter and bring it back down but if it comes in low, which according to the spec it can (wikipedia "no less than 4.75 V"), there is less you can do without some tricks.

Generally, there are at least two possibilities:

1. If you can get away with 3.3 V instead of 5 V, do it. There are excellent low noise voltage regulators such as TPS7A4700, TPS7A49xx, ADP150, TPS796xx etc. Be careful not to use older chips like 78xx or LM317, which are neither low dropout nor low noise and therefore wouldn't do the job.

2. If you absolutely need 5 V, what would be the maximum current? For a low current demand you can use a switched-capacitor solution such as the TPS60241 (up to 25 mA). For a higher current you can use a step-up (boost) converter followed by a fast, low noise linear regulator (one of the above mentioned or similar).

Cool ideas.  The TPS60241 looks like a really good small low parts count solution for under under 25mA.  Good call on the newer regulators as well.  Thanks!

[hlavac]

The 5V from USB is not really 5V, its only 5V right at the connector in PC.
There are large allowed voltage drops in the cable, and bus powered hubs, etc so you may end up with something like 4.2V at your device without USB spec being violated...

You may want for example a SEPIC DCDC converter to get reliable 5V.
Or run at 3.3V and use LDO.

[DutchGert]

If u only need 100-200mA @5V u could also use a verly low drop out 5V regulator from TI. We use one to filter a 5V from a switcher that is used for a touchpanel controller (very noise sensitive). In practice u end up with only 30-75mV drop at 150mA and worst case perhaps 4.85V at high load/high temperature.

Also, Micrel has something that they call Ripple Blocker. I don't have experience with them but it's basically the same thing, a very low dropout LDO with a high PSRR.

Ofc, if 3V3 is enough go for a LDO with a very high PSRR. That'll solve it.

[tszaboo]

Switched cap buck-boost is probably your best solution, as it is plain simple.
http://www.ti.com/product/tps60241
I would use this. If you need more than 25mA, then you can use some more crude chip for the digital stuff.
http://nandblog.com/chip-review-mcp12523/
i would use this two chip to get higher current more noisy 5V (with the MCP but that doesn't matter) for digital and the Ti for the analog. If this doesn't work than we need more information about the currents.

[brabus]

http://www.ti.com/product/tps63000

Simply one of the best ICs I have ever used.

[DutchGert]

http://www.ti.com/product/tps63000

Simply one of the best ICs I have ever used.

True, that's a nice beast. Used it for a Lipo powered system ones, worked a treat.