Design flaw? Polyfuses for USB current limiting

Yes soldering the +5V pins of the USB ports together would be safer and easier, we need a picture of that.

These should be the right pins, I think. Please confirm.

Vcc of both USBs connected. (Picturepart taken from http://elinux.org/File:Raspberry-pi-bottom.jpg by Selsinork)

connectedUsbVcc.jpeg (24.91 KiB) Viewed 1646 times

[abishur]

Of course the pi has a 1A limiting fuse on the input

Apparently 750mA ( 700mA ? ) fitted to the boards being shipped.

You sure about this? The schematics they provided here just a month ago seem to indicate that the fuse is still 1A.

[abishur]

I think bridging F1 and F2s outputs helps. So one USB device could draw up to 200mA or two USB devices can share the current as they need (e.g. 150mA + 50mA). This would be at least an easy home-made way. (Image embedded.)

Great hardware hack, Timo, thanks!

I think we need a site for comprehensive articles on useful hacks for the community. The ideal place would have been this forum, but it isn't any longer because last night the admins disabled post editing after 30 minutes, so it's no longer possible to work on a well-considered post and hone it into perfection. Pity, but the forum won't be usable for extended descriptive articles improved incrementally.

Morgaine.

Not to go too off topic here, but lamentably this is a necessity due to trolls and the like who get an argument going and then go back and edit their posts to make it seem as if the other person is being overly aggressive and a bully. We've tested both ways (unlimited editing and limited) and despite the inconvenience caused by not being able to eternally edit a post, we've found limiting it in some fashion helps tremendously (hence the 30 minute editing window)

Of course the pi has a 1A limiting fuse on the input

Apparently 750mA ( 700mA ? ) fitted to the boards being shipped.

You sure about this? The schematics they provided here just a month ago seem to indicate that the fuse is still 1A.

There's your quote from Pete Lomas himself; "If you remove them then all you have for protection is the 700mA inbound fuse" and I've also found the earlier thread stating 750mA fitted ...

viewtopic.php?f=63&t=5068&p=67725
jojopi » 22 Apr 2012 14:17

But who knows ? Should it be 1A, 750mA or 700mA ? I asked for a definitive answer later in that thread but I don't recall receiving one and the confusion continues. As it does with other issues and that's why I, and I suspect others, get a little cheesed off and frustrated at times

Thanks to the author of this thread and to those who have pitched in to help make the RPi a better product.

[abishur]

There's your quote from Pete Lomas himself; "If you remove them then all you have for protection is the 700mA inbound fuse" and I've also found the earlier thread stating 750mA fitted ...

viewtopic.php?f=63&t=5068&p=67725
jojopi » 22 Apr 2012 14:17

But who knows ? Should it be 1A, 750mA or 700mA ? I asked for a definitive answer later in that thread but I don't recall receiving one and the confusion continues. As it does with other issues and that's why I, and I suspect others, get a little cheesed off and frustrated at times

Shesh me, pay attention to what you just posted!

I would be very interested in getting a firm answer on this. I feel like my pi is using more than 700mA, the schematics show 1A, I'll try and get a good look at the fuse on my Pi so I can provide a link to the data sheet on it.

..The ideal place would have been this forum, but it isn't any longer because last night the admins disabled post editing after 30 minutes

No, they didn't. This has been the case since inception of the new board over a week ago, but you only noticed it today. Please stop making things up to further your own agenda.

[rurwin]

The fuses kick in hard around 280mA and fold back and limit to 140mA. If you remove them then all you have for protection is the 700mA inbound fuse. The tracking on the board is good for 500mA+ so you could if you really wanted too. What about a powered hub – to power the Pi and bigger USB devices.

But if they are limiting to 140mA, then they do so by increasing resistance and hence dropping voltage. If the USB devices were linear then the fuses would drop 2.5V and only supply 2.5V to the devices. In practise the devices will fail to work at that voltage. The result will be fairly random, but not in any way usable. If you remove the fuses then the inbound fuse will kick in hard at 1.4A, which is rather too high for the tracking, but the Raspi should take over 400mA leaving less than 1A for the USB. Divided by the two power paths the tracking should be capable, but the current probably wont split equally. It will probably work, but it is close to the limit.

But if they are limiting to 140mA, then they do so by increasing resistance and hence dropping voltage. If the USB devices were linear then the fuses would drop 2.5V and only supply 2.5V to the devices.

I assume you mean ohmic, rather than small-signal linear. In any case, you are interpreting Pete's description too literally. PPTCs do not monitor current and they do not switch gradually between the hold and trip currents. Rather, they are guaranteed to stay low resistance up to the hold current and guaranteed to go high resistance after some time above the trip current. Between those values is a forbidden area where they may or may not trip. Once they go high resistance they tend to stay that way until the load is removed completely and they are given time to cool off.

See http://www.te.com/content/dam/te/global ... entals.pdf

(Why is this thread still going?)

(Why is this thread still going?)

Waiting for it to cool down.

[mahjongg]

These should be the right pins, I think. Please confirm.

Well, its the Pin 1 of the USB connector that carries the 5V supply.
If you hold the board so that you look into the USB connector(s), with the PCB under the connectors, then the leftmost pin would be pin 1.
So if I mentally rotate the picture in my mind correctly I can conform it is correct.
On these pins, using the metal of the USB connector as ground, there should be +5 Volt.

[mahjongg]

Of course the pi has a 1A limiting fuse on the input

Apparently 750mA ( 700mA ? ) fitted to the boards being shipped.

You sure about this? The schematics they provided here just a month ago seem to indicate that the fuse is still 1A.

Well if you look at this high res picture of the bottom of a R-PI you can see F3 in the top left corner:
http://elinux.org/images/b/b8/Raspberry-pi-bottom.jpg
it is marked "07", which seems to indicate 0.7A, which confirms the quote that a 700 mA fuse is fitted.

The fuses fitted for the USB ports are "Raychem" (Tyco) fuses, but this one doesn't have any logo.
However on similar (1812 0.75A) polyfuses from lttlefuse the marking LF075 doesn't indicate a tripping current of 0,75A, but instead a holding current! The tripping current is 1,5A!

For such a Polyfuse its resistance is between 0.1 and 0.45 Ohm, so at 600mA in worst case condition it will drop U=I*R = 0.6 x 0.45 = 0,27 Volt, and that is in the "danger zone"!

Why do we still discuss this?! Well for one we are in the dark, as we can only make educated guesses. And on the other hand, this is important stuff, as it can explain many of the problems we see happening.

Unfortunately I do not have a PI to experiment with, so I cannot do any test and measurements myself, I do have the nagging feeling that using the rated fuses is a mistake, and they should have rated them at double or triple of their current ratings. the many problems I see in the troubleshooting area do seem to all point at the same problems (combined with insufficiently stable "charger" power supplies). The problem seems to be present in both the power input fuse and the USB fuses, both seem to have been rated such that in worst case scenario's their internal resistance can increase to unacceptable levels.

As picture "figure 3" in the tyco white paper shows, the internal resistance does increase with temperature, even if it only escalates toward the tripping point. Also it doesn't do it linearly, as the graph shows, but logarithmically! But I guess that is why the resistance of the littlefuse (which would behave the same way) is rated as being between 0.1 and 0.45 Ohm.

..The ideal place would have been this forum, but it isn't any longer because last night the admins disabled post editing after 30 minutes

No, they didn't. This has been the case since inception of the new board over a week ago, but you only noticed it today.

My mistake, sorry. You are right that I only just noticed last night that there was a block on previous edits.

What I wrote still holds true though, despite my mistake about when this block was enabled. Forum contributors will not be able to create detailed articles written over a period of time about how to do things with Raspberry Pi as a result of this editing policy, which I think is a reasonable thing to highlight.

PS. Ending this discussion here, as it's not on topic. I gave Timo my encouragement to make a hard-hacks article ... if he makes one, he'll just have to write it perfectly on first attempt.

Morgaine.

This is a pic from my dead Pi - it confirms the F3 value - Mine's gone back to RS

I agree with mahjongg that we're still discussing this matter because it's a very important issue for stable Pi operation, and also because there are still so many unknowns surrounding it. What's more, it's in the nature of engineers to want to understand their devices fully, particularly when there might be a fault here --- we want to fix it!

On a simple practical note, the information we discuss here provides input to thousands of interested people in the Pi community, and it also provides input to the Foundation. There are orders of magnitude more engineers among us here than they have in the small RPF team, and our sheer numbers means we can catch things that they might not. It's just like in open source software --- the power of a million eyeballs.

Morgaine.

I can see the argument for requiring a powered hub for anything that uses a lot of power. The problem with the Pi is stuff you think should probably work either doesn't or worse, seems to work and then causes random problems.

I hardly use the mouse but occasionally I startx and need it. I have a yubikey on my keyring. Don't really want my keys near a running Pi but occasionally I need to log into something. And usb flash storage to move some files about. None of these devices draws a lot of current. I doubt if any would use more than 100mA peak. I really should be able to run them all from a little unpowered octopus type hub. As it is I can run them and they sort of work sometimes. Especially nasty with the usb drive which you think is working then you get errors.

I agree that the polyfuses are too conservative. Ideally I would like to see the USB powered directly from the micro-usb power input via a single bigger 1A fuse. I think my 10W USB power supply will handle it.

@Dave_G_2: Impoverished 3rd world ? Huh, they are making anything you are using. So they have much more data on USB and blue smoke trigger conditions than any of us.

Not to say i told you so...but i posted about usb chargers possibly not being up to the task of
actually running the pi, as did several other people and it seemed to be ignored or glossed over.
If you think about it, a chargers only real job is to charge a battery safely which is usually c/10 or
a 10th of the battery's mah capacity. Its not really surprising that alot of chargers out there either wont run the pi or exhibit flakey behavior as they are not designed like a switching power supply.
It's a shame that the polyfuses only make the problem of attempting to use a charger worse.
It's also a shame that as far as i have seen, none of the actual designers have bothered to post on this thread with any new information about if they are even looking into this problem or even know about it.
I'd like to know how many weeks were spent testing various usb devices before the pi was launched?
How many different "chargers" were tested?
Just what exactly was tested?

[mahjongg]

Officially (according to USB specifications any USB charger -should- comply to) any charger should output a sufficient voltage up-to the rated current, that is a "1A 5V" charger should supply at least 4.75 Volt when loaded with 1A, which is (barely) sufficient, but at 0.7A it's output voltage should be sufficient.
However, some very cheap chargers are labeled as "1A 5V", but only output 5V when completely unloaded. These things often consist of nothing more than a simple transformer, a single diode, and an elco of a few thousand microfarad. At the rated current these might drop to something like 4 Volt, sufficient for most charging circuits (that can still charge a 3.6 Volt battery with such a small input voltage), but utterly unusable for the intended use with a raspberry.
Luckily, unless you buy your charger from e-bay, or a dollar shop, most of them are better than these, but maybe not that much better. Anyway, the R-PI's wiki lists many chargers that are known to work fine with the PI (and a simple keyboard and mouse).

If you think about it, a chargers only real job is to charge a battery safely which is usually c/10 or
a 10th of the battery's mah capacity.

That is not correct, nowadays most portable devices use lithium-ion batteries, which accept a maximum charge current of about C, i.e. some 1A or more for a typical smartphone battery.
The maximum voltage on lithium-ion batteries, while charging, is regulated to 4.2V. Linear regulation from 5V to that not much lower voltage is relatively efficient, so using 5V from (an) USB (charger) makes it possible to use compact and low-cost charger circuits in the device.

Depending on the charger circuit it may need some voltage difference between the input and output to operate, so the input voltage may not drop much below 5V. Also it may not rise much above 5V because the charger circuit might become too hot.

So if you have an USB charger rated at e.g. 5V, 1000mA, it would be incorrect to assume that it will deviate much from that rating "because it is just a charger".

Best regards,
Simon

So if you have an USB charger rated at e.g. 5V, 1000mA, it would be incorrect to assume that it will deviate much from that rating "because it is just a charger".

In a perfect world all USB chargers would comply with specification but reality is that many do not and there's no easy way to tell which will or will not nor how they will perform with long term or continuous use.

You are right; it would be incorrect to assume that a charger will deviate much from its rating ( though note what mahjongg points out from the actual specification; a "1A 5V" charger should supply at least 4.75 Volt when loaded with 1A ), so it can be equally incorrect to assume that it will not. Thus it's all somewhat hit and miss, and that's before the polyfuse issue comes into it.

To a reasonable extent, the issue will be alleviated when there is an 'official R-Pi USB charger' that is guaranteed to work with the R-Pi which can be purchased.

[rurwin]

Here's some figures; make of them what you will. So far it's working, but I think I may have been lucky and I haven't tried any adventurous USB devices yet.

Here's the state of the fuses before the RaspPi was powered up for the first time (except for in the factory, obviously.)
F1 and 2: 6 ohms and 7 ohms
F3: 0.5 ohms

So we apply some power:
TP1&2: 4.85V
Add a cheap keyboard and an authentic Microsoft Optical mouse (both wired).
USB pin 1&4: 4.75V and 4.54V

That is in spec for an unpowered USB hub, but only just. If I plug in anything that takes 100mA then the fuses will drop 0.6V and 0.7V and the USB supplies will be 4.25V and 4.15V. Those would be a long way out of spec.

[mahjongg]

Yes, resistances in the ohms shouldn't be there, most power train calculations are working with milliohms.

USB devices can communicate with 4.4V for the USB nodes, but most devices themselves need 5V +- 5%. So for a self powered hub 4.4V is actually enough, as the logic in the device itself is not powered from the PI's low voltage, but mice and keyboards should get 4.75 Volt. Perhaps some may work with less (most mice logic will, but i'm not sure about keyboards), but even then 6 Ohms is ridiculous!
These fuses should have had resistances below 1 Ohm, like normal USB masters.

From this datasheet:
http://www.arrownac.com/offers/bourns/Bourns_USB_Port_Protection_Final.pdf

Overcurrent protection must have a resistance no greater than 700 mOhms with a maximal voltage
drop of 350 mV

Note that 700 mOHm (milliohm) is just 0.7 Ohm. so these fuses have ten time the recommended resistance.

[mahjongg]

After careful study, I must come to the conclusion that "single current unit devices", including powered hub's should work with 4.4 Volt. Only devices requiring two or more (up-to 5) current units (200 to 500mA) are allowed to expect 4.75 Volt.

This doesn't mean however that the current fuses are an acceptable solution!, even when we try to enforce the use of 100mA devices only!

Even if we assume that the PI is designed only for single current unit devices which should keep working with only 4.4 Volt (and im content that single current devices when properly designed probably will work when they receive 4.4 volt) we are still not out of the woods concerning the current polyfuses. With their widely varying internal resistances, and with a limited number of test samples we find some that are 7 Ohms! Statistically that means that 7 ohm is not rare, and there probably will be ones that are even higher (especially if they have ever been triggered before).

Seven ohm is too much for even 4.4 Volt devices! It would mean that in a worst case scenario, when the devices really does draw near 100mA, we will have a 0.7 Volt drop, and that is too much, even under the unlikely scenario that we have exactly 5.0 Volt on the board, they will only get 4.3 Volt!

In practice we can often expect something like 4.9 Volt on the board, so in reality many USB devices will only receive 4.9 - 0.7 is 4.2 Volt, and that is well below what they can work with.

Common requirements you read about often allows only for 0.35Volt (350mV) drop between the power input and USB ports, so that would mean that the fuses must always be less than 3.5 Ohm at 100mA.

If we want to support a, for the PI very useful, device such as a keyboard with built in mouse port, which typically draws 200mA (providing the PSU can deliver the extra current) then the requirements are even much more strict, as not only the voltage drop may double, but also these devices do not require that they should work with only 4.4 Volt, they should get 4.75 Volt.
And as such (not minding the PI might not get exactly 5.0 Volt) the voltage drop may only be 0.25 Volt and at 200mA that means the fuse may never be more than R=U/I = 0.25/0.2 = 1.25 Ohm!

Well that took me nearly an hour to read through. Much of it was arguing mirco differences of opinion which while occasionally interesting were for the most part not relevant to issue so I'll move from theoretical engineering to practical.
There is an issue with the USB port power and Morgaine is correct in her solution that these should be wired directly to the power input and individually fused to 500mA. It's correct not just because it solves this problem but because it is also good engineering practice. In the meantime (while they design a new PCB) we can use powered hubs or mod the boards. So that is that issue dealt with.
Until then this info needs to be included with every R Pi because the user needs to know.

1. Your power supply must be capable of delivering a regulated 5V DC +/- 0.25V at at least 700mA.
2. The USB ports are rated at a max of 100mA each. USB devices that operate near or above this limit may operate erratically or malfunction. You are advised to use a self powered USB hub if you experience difficulties. Examples of devices which may experience problems include illuminated keyboards and wifi dongles.

That will save a lot of people a lot of grief, certainly would have saved me some.
PS I will defo be getting the soldering iron out.