Capacitor to smooth out Raspberry Pi power dips?

Question

I recently was trying to do a few things with a Pi3, and it consistently hung when I tried to update the system (sudo apt-get update/upgrade). It did everything else well enough, so I wonder if that's power related, with the upgrade making the WiFi work extra hard - maybe not, but I thought I'd look into it.

I've also noticed that my older Pi1 will hang once in a while (every few months), and that's a bit of an issue now that I'm using it as my sprinkler controller - reliability has become much more important.

While searching for help online, I noticed Robert Peter Oakes did some research and made a nice blog entry explaining the role the USB cables have in the power issues. (Thanks Peter!)

In a nutshell, some cables cause a voltage drop that puts the supply too far below the ideal 5v voltage level for the Pi.

The problem is that once in a while the Pi draws enough power to make the voltage dip into the danger zone.

(Some places sell adapters with a higher voltage to compensate. AdaFruit, for example sells a 5.25v adapter for the RPi, and notes that 5.25v is still within the specifications for USB, so even with a perfect no-loss USB cable that should be safe.)

One notable item, to me, was that the Pi has some serious power dips on a regular basis, regardless of the cables - just that the better supplies+cables start with higher levels at the Pi and the dips don't take it down too far.

So here's my thought - capacitors are supposed to help against dips and spikes, right?

Is there a way to add some really big capacitor at the Pi side to help avoid such dips (and maybe spikes too) ?

I'm thinking VIN-GND with a 1,000+ uF cap? I have one rated 1,000 at 10v, also I see 1,800 at 16v, both should handle 5v-ish well.

Otherwise, maybe splice a USB cable to add the large cap near the micro-USB plug end?

Will that cause trouble? Will it help at all?

Thanks!

-Nico

ntewinkel · Answer

Here's an update... I received one of the new power adapters a few days ago and tried it last night. It's a Keyestudio 5.25v / 2.4A "highly regulated switching power supply". The first feature it mentions is compatibility with Raspberry Pi requirements. It's intended as a proper power supply - the cable is directly attached, and it says the wires are 20awg. I put my multimeter on 5v and GND and saw that at startup it was at exactly 5.25v, which then goes to about 5.20v while running, and down to about 5.16v while doing work. This is at multimeter display refresh, so there may be quick details I can't see without an oscilloscope of course. The point is that I no longer see the power-warning lightning bolt, and the WiFi connection appears to be much more stable - although it did disconnect once still during testing. Also, it wasn't my SD card as I first thought - I was able to run the updates for both SD cards I tried it with. So I think the lesson here is to use proper power supplies, not chargers (as so many people have said so many times!). And the other part of the lesson is to not blame the power supply for everything - in my case the dropped WiFi was causing my troubles, because due to it being a remote-desktop connection it became inaccessible, making it look like it was hanging. Cheers, -Nico

shabaz · Answer

Hi Nico! I'm not sure that an update/upgrade would consume a lot of power, the things that would do that would be all four cores processing, and all four NEON processors working away too. Then the Pi consumes around 7W or more. It could still be a power issue however, especially if you're running a long USB cable and not using the official Pi 3 supply (which has very fat cables and slightly higher than 5V as you say). A big capacitor might help, if the demand really is for short bursts. Any of the ones you suggest would be fine. If you do need to run it from a long distance, then a good solution is to have a power supply local to the Pi, e.g. a 12V to 5V DC-DC converter, so that you can power it from a higher voltage from a distance.

rew · Answer

.... 1000uF / A .... That rule of thumb is from the old days with linear power supplies. So... to build a 1A 5V powersupply you'd take a 9V tranformer, rectify it and place a cap there. The peaks of the 9V transformer output are near 12.7V. That would put the capacitor near 11.3V. With a linear regulator like a 7805, you n eed 8V at its input. So you have 3.3V to work with. So for a capacitor the formula is: I = C dV/dt = 1000uF * 3.3V / 10ms = 0.33A. Not quite enough yet for 1A of current! Doing the math is better than using the rule-of-thumb. When you start putting a capacitor on your 5V line, an important question is: How much drop can you tolerate? Well. I've experimented and my raspberry keeps on working fine down to about 3V. But that's the pi only. USB or HDMI peripherals (like HDMI->VGA converters) may require something above that 3V or even 4V. So we need figures like the extra current draw, how long that happens, and what voltage difference we're going to tolerate. The formula is I = C dV/dt which can be reworked to C = I . dt / dV . Put in numbers like I=2A, dt = 10ms and dV = 0.2V and you get an answer for the capacitor you need. With the numbers I've suggested you need 0.1F, or 100 times more than the 1000uF you suggested... .Math to the rescue.

mcb1 · Answer

I'm thinking VIN-GND with a 1,000+ uF cap

The rule of thumb for filter capacitors is 1,000 uf per amp as the absolute minimum.

This assumes the load is constant, but for your power dip issue, you really need to be looking at 10-20,000uF to be sure.

You can try the other values and stick it across the 5v on the GPIO pins.

You should be able to measure the 5v lines to see what you are actually getting.

The GPIO is probably a good and easy place to get to.

IMO most of the issues are the power supply themselves.

I've run them from 700mA Samsung chargers with tiny leads while other 'genuine' supplies were below 5v.

Most of the 'chargers' are simply a device to provide 5 ish volts to a phone to charge a 3.6v lithium battery, and were never designed to be a stable smooth 5.x volt power supply.

You could have a mains issue, but with a switchmode supply (as most are nowdays) it should handle it.

I don't suppose it's some dodgy software hobbled together by someone in some small backwater town ...

Mark

rew · Answer

The pi has a voltage-monitor that is used to show the lightning symbol on the screen, and play with the red led. It triggers below 4.65V nominal. I've measured the pi itself. It runs fine downto below 3V. At 3.3V and lower you're starting to go outside specs for various chips, but they behave well enough to keep things running "outside spec". But the 5V from the input is passed on to the HDMI as well as to the USB ports. It depends on what you have there if it will keep on working or not. Grab an USB harddisk enclosure, and it won't work on a pi... At all... (at least some of them...) So... it depends on what you have for experience of what you say works and what doesn't work. Below 4.65V you quickly enter the area where things are "out of spec" for the "should be 5V nominal" outputs of the pi on HDMI and USB.

gdstew · Answer

I have upon occasion found USB cables on Ebay with the wire gauges listed, however I am always wary about the accuracy of listings on Ebay. I also came across a set of 3 - 6" and 3 - 3' micro-USB connector to type A USB connector cables some from a company called Sabrent on sale at Newegg that specifically mentioned that they used 22 AWG wires. That particular set is apparently no longer available at Newegg however on a quick Google search for Sabrent USB cables it also shows up at Amazon, Fry's, and Walmart with single cable and cable sets using 22 AWG wire. One disadvantage of the thicker wire is the cables are less flexible and I think the Sabrent cables use 22 AWG on all four wires because they are very stiff. Another thing I've found useful are the inexpensive USB volt/amp display meters at Ali Express. Some have both micro-USB and regular USB type A inputs and they are accurate enough for most uses. Some even come with switchable 1 and 2 amp resistor loads that plug into the meters USB outputs (they do GET HOT). They're not particularly useful for catching fast transient type problems but are fast enough to get a reasonably good look during normal use. I bought the Sabrent cable set from Newegg specifically to use the 6" ones with one of these meters to test all of the USB output power "wall warts" I bought as well as to watch the voltage supply and current usage of several Raspberry Pis and other "Pi like" SBCs with different attached hardware and program running loads.

fvan · Answer

Those power dips sound strange! I'm able to run a Pi 3 with official touch screen on a single 5.1V/2.5A official power supply. Occasionally the little lightning symbol pops up, but certainly no hanging. What are the specs of your power supply? Have you tried another?

rachaelp · Answer

Whilst the Pi3 typically shouldn't be causing power supply dips, Nico does say that there are voltage drops in his wiring so I suspect it's a wiring issue and it's not taking too much extra current to cause additional drops. Nico, if you have confirmed a significant voltage drop across your wiring, then personally I would just fix the reason you have the large drop. As others have said, it shouldn't be drawing significant current with what you are doing. You could prop the voltage up at the Pi end with big capacitors but you're just masking the issue rather than fixing it and you're likely to see the problem reoccur if you run something that really does start to draw significant current at which point no sensible amount of added capacitance will save it from dying. Best Regards, Rachael

ntewinkel · Answer

Another update, in case people are curious... My sprinkler Pi2 went unresponsive again on the weekend (last reset was 2 weeks earlier), so I decided that was a good time to switch power supplies for it - I had been using a 5v 2A tablet charger, so I now have it on the aforementioned Keyestudio power adapter. I then took the tablet power supply and put it on my Pi1 cat-video system (it plays videos for cats, LOL yes). It right away showed the coloured block at the top right indicating poor power supply.... hmmm interesting! So for the Pi1 I went back to a 2-usb-port charger supply (a cheap eBay one) that I'd been using for it, and the video flickers constantly, sometimes even keeping a blank screen - but no power-warning icon. So that video flicker issue I was trying to fix by changing config settings appears to be due to the poor power supply too. (ps, it's an old monitor with HDMI-to-DVI adapter) I have a different power adapter still on the way, but it's taking the scenic route. I'm curious as to how it will perform. Cheers, -Nico

Robert Peter Oakes · Answer

It still comes down to the power supply in your case. Increasing the power supply to 5.25 is compensating for the voltage drop on the cables from the brick to the PI. Also the brick can in itself be dropping volts under load. Still within specifications but still adding to the issue. A USB Tablet adapter is a funny thing, some are really good quality and have regulated outputs (Geniune Apple Tablet ones for example or Samsung) but not all. Adding a capacitor at the PI end can help but it can also hinder in that during power up it can draw too much current and cause the brick to current limit while it tries to charge those big ass caps. Depending on the power brick this can recover or spin into a loop of shut down and power up Any USB type charger that is used also has a big question over the quality of the USB lead between it and the PI, if it does not say the wire gauge on the plastic sleeving then it may be a crap one, good enough for charging a phone or for data usage but not powering something that relies on the delivered power levels, In my testing I found some horrific ones and a few good ones. also a common trick is to use thin wire and really thick insulation to give the illusion of a good lead. loading the PI end with a 5 ohm or 2.5ohm resistor will quickly identify a bad lead / power brick in a way that can easily and reliably be measured (1A or 2A load but steady load), if this does not pass then there is no way it will pass on the PI reliably either. So, I have found that the purpose built power supplies for the PI are a good way to go, I would still be hesitant to go to 5.25V as this is right on the 5V +-5% limit and any surge could cause a problem. Most of the official PI ones are 5.1V and 2.5A and only have a figure 8 cable (Two separate multi strand conductors, think speaker cable), not USB cable, and the wire is pretty thick, at a whopping 18AWG. 24AWG is 0.025 Ohms per foot 22AWG is 0.016 Ohms per foot 18AWG is 0.006 Ohms per foot so for a typical 6 foot lead (12 feet of wire, remember the return path too) at 2 Amps 24 AWG = 2 * .025 * 12 = 600 mV lost... ouch 22 AWG = 2 * .016 * 12 = 384 mV lost 18 AWG = 2 * .006 * 12 = 144 mV lost So as you can see, even without physical measurement the 24 awg cable is a lost cause to supply a PI under load unless it is less than 1 foot long and this is assuming the power supply itself maintains the 5V out under load, some dont. The official PI PSU with 18AWG can easily supply 2A and more on a 6 foot lead. So in conclusion, the PI is designed with a USB connector for its supply, this does not mean unfortunately that the PI will work with all the range of the USB specifications, not even close USB Specification for volts MAX nominal volts = 5.00 +0.25 −0.60 V so anything from 4.4V to 5.25, this is MAX not under normal operation too and if you use a 5.25V supply, your risking your pi or other USB device you may connect to it. raspberry PI is 5V +0.25 - 0.25 or to put it another way, 4.75 to 5.25V The one thing I have not yet mentioned is that if a power supply for the PI is a full functioning USB specification PSU, it may limit the power delivered even though it is able to supply more, if the PI does not signal the current requirements. I dont know if t does.

gdstew · Answer

"Also the brick can in itself be dropping volts under load" I've found quite a few of the cheap USB chargers will not produce 5V at the rated current. I've seen as low as 4.6V which is of course well out of spec. I personally want at least 4.9V - 4.95V at rated current before I will use it. "if it does not say the wire gauge on the plastic sleeving then it may be a crap one" I've found quite a few that do have the wire gauge printed on them that are still crap, usually 28 AWG on all conductors, sometimes 28/26-24 AWG. At least you can tell by looking at them. I've also noticed that some of the current "Pi rated" supplies use a 4 ft. power cable instead of 6 ft. which helps a little. "I would still be hesitant to go to 5.25V as this is right on the 5V +-5% limit and any surge could cause a problem" Most ICs have a maximum voltage rating in their specs that is well above 5% of the nominal rating, usually around 6V to 7V for 5V devices specifically to handle surges on the power supply. It is not a good idea to test this specification on a regular basis. "So in conclusion, the PI is designed with a USB connector for its supply, this does not mean unfortunately that the PI will work with all the range of the USB specifications, not even close" and "The one thing I have not yet mentioned is that if a power supply for the PI is a full functioning USB specification PSU, it may limit the power delivered even though it is able to supply more, if the PI does not signal the current requirements. I dont know if t does." No USB host device is required to work with all the range of the USB specifications. The specification was specifically designed to allow a range of minimum current requirements (the minimum amount of current required to be supplied by a USB host is 50 mA) mainly to allow battery powered or otherwise low powered devices to use USB. A USB host device should always provide 5V +- %5. The 4.4V voltage rating is for hub devices to allow for voltage drops through chained hub devices powered by the USB host. I believe that two chained hubs is the maximum and 4.4V is the minimum for the second hub in the chain.

rew · Answer

I built an USB-cable testing setup once. And I ran a bunch of my cables through the tester. The resistance of the cables came out between less than 200mOhm to more than 4.5 ohm! The best was 22CM Magnet Flat Short 5Pin Micro USB Data Charger Cable Cord For Samsung HTC XC | eBay in the ugly pink...

gdstew · Answer

It's more a matter of the wire gauge used for the power pins than anything else. Cheap USB cables will use 28 AWG (gauge) wire for all connections, better ones will use two different gauges, one for power and another for the data signals. Most (but not all) of the USB cables I have bought have the wire gauges printed on them and when they do the lower number (lower gauge equals larger wire) will be the one used for power. When trying to run 1.5 amp or more through the USB cable you need to have at least 22 AWG for power. The biggest problem I have with that is that when looking to buy USB cables over the net they very rarely list the wire gauge(s) used so either a lot of searching or buying local are about the only ways to find good cables. Note that cable length can also be a problem so I try to use 1.8m (6 ft.) or shorter cables for power.

niteowl12 · Answer

Wouldn't it be easier to use an O'Scope to measure the input voltage during the "hang" to make sure that voltage drop is the actual problem? This would be my first step because I would think it highly unlikely that an update would use enough extra current to drop the voltage enough to cause a micro to stop functioning. Is the Pi controlling the sprinkler system through logic control or is it actually providing the power to a valve? 10uf Cap 1000uf Cap Notice the drop in voltage is about the same with a 2A jump in current. R1 in the schematic represents the incoming wire resistance to the Pi. If all the power to the Pi and sprinkler run through that one wire then I would focus on fixing that. Use two separate wires or one large one. That resistance will be the primary driver for voltage drop.

rew · Answer

Jason, that's a very good suggestion. Usually it is quite difficult to catch "occasional" problems. But for power supply sagging problems you can indeed put the scope in "normal" mode (as opposed to "single" or "auto") and set the trigger point at say 4.5V. Then you'd end up with a trace of "the event" when it eventually triggers. On the other hand, not everybody has an oscilloscope. The other part of your suggestion: Solenoids, relays and the likes will often generate strong EMI pulses. Those can also create havoc with the reliability of systems...

gdstew · Answer

"The datapins are not connected, so the pi will not negotiate anything (=not conforming to spec). The pi will draw whatever it wants without negotiation (=not according to spec), and it will potentially draw more than 500mA (=not allowed by the spec). " The Pi power connection is NOT a USB device nor was it ever intended to be used as one, it simply uses a USB connector. The USB connector was chosen mainly because of the availability of USB chargers which also (usually) not USB devices. The USB A type connector is capable of handling 2A/pin, the micro connector 1.8A/pin (I was surprised when the Pi 3 spec said that it could pull 2A since it is using the USB micro connector) so it is obvious that someone figured out that it might be used for other purposes. Thanks for the corrections on current negotiation, my memory ain't what it used to be and I couldn't find the chapter on power in my USB spec. But if I remember correctly, the amount of current negotiated for is actually 100 mA to 500 mA in 100 mA increments?

shabaz · Answer

Hi Nico! I'm still not sure it is a supply issue since the power consumption shouldn't increase during an update/upgrade significantly, but if all you have is a 1000uF capacitor and no oscilloscope I don't think there is anything wrong with just trying it, if you suspect that a combination of your PSU and the USB cable are making things just marginal. It all depends on how marginal, and for how long you're getting high current spikes (if that is what is occurring, because it is speculation, but sometimes we have to speculate). If the spikes were short and infrequent (e.g. 1msec spikes) and the current demand was around 1A (I don't think the Pi ever consumes 2A unless you've got extra hardware attached; and 1.4A, i.e. around 7W, takes all cores to be performing a workload), and if the supply voltage can drop from 5V by a volt and still continue operation on the Pi 3, then a 1000uF capacitor may well help. Maybe an interesting project would be a low-voltage detector LED that lights up for a second even on short drops in voltage : ) There are ICs that will approximately do this (ICs for handling brownouts).

mcb1 · Answer

ntewinkel The Ipad chargers are rated at that, but they need to sense the device at the end in order to deliver it. There was an article in Silicon Chip regarding the voltage required on the data pins, in order to force them into giving the required amount. Tablet chargers are also a hidden mystery as my daughter found out in Japan when she took a normal charger. It seems that many detect the hardware and will bump the voltage to 15v out the USB socket to facilitate charging the tablet. The rest of the time they are a plain looking 5v USB socket supply. Nice to see there is some progress and yes the devices hanging off the USB can sometimes be the problem. Mark

mcb1 · Answer

I have no doubt some cables are worse than other. I've seen the insides and the solder joints......

niteowl12 · Answer

-Nico, The datasheet from the step down DC/DC shows that input voltage operating range is 2.5 to 5.5V and thats at the dc/dc input after a reverse protection FET/Diode and the wire from the wall unit to the board. I think you are safe even in the event of a surge.