A Technical Explanation of how the Internal Airport Base Station Power Supply Works

Here is a excellent background on how the power supply in the ABS actually works. Three readers (Brian Whitaker, Luigi Morelli, and Hac Da Lee) sent in very thoughtful letters that are reproduced here (with a few minor edits and their permission). Letters like this make my day. A big thank you to all three authors- another proof that the Internet can be used for very useful purposes. Cheers! -CvW

Mr. Whitakers original article:

"I'm a hardware applications engineer at Maxim Integrated... I happen to specialize in RF, but I know something about power supplies (in this context, I mean dc-dc switching converters, like the National LM2596 used in the Apple ABS).

I was surprised to see 25V caps specified for the input and output... this is a 12V to 5V step-down (or "buck") converter -- the AC-DC brick is a switcher that outputs something close to 12V at something up to 1.5A. Open-circuit, I measured 12.4V, which is nice, because transformer types can float up to 2x their rated voltage.

I'll offer a bit of theory here, but I'll be brief:

The idea behind a DC-DC buck converter is that you use an inductor as a storage element, charging it up with Vin (12V) and discharging it to the output Vout (5V)... controlling how long we charge the inductor before we let it discharge to the output will determine the output voltage.

It is important to remember that current cannot change instantaneously in an inductor -- if it is 1A at one instant, it cannot be 2A the very next instant... we talk about the current 'ramping up' and 'ramping down'... how fast the current ramps up and down is a function of the inductance L of the inductor, and the voltage across it -- specifically,

V(across the inductor) = L * (di/dt),

where di/dt is the ramp-up/down speed in amps/second.

Here's what the typical circuit looks like: 12V input (with reservoir cap to ground) connected to one side of the switching FET other side of FET connected to one side of the inductor at this same point (called the LX node), we also connect a diode to ground, which allows current to flow from ground to the inductor the other side of the inductor is connected to a reservoir cap to ground, and is the 5V regulated output. The dc-dc controller is listening to the 5V output, and switching the FET on and off to keep the output at 5V.

The switching FETs (in this case, internal to the LM2596) control whether we are ramping up current in the inductor... the switching cycle goes like this:

The output is discharged when we start off... no current in the inductor...
The input goes to 12V and we turn on the dc-dc controller
The controller turns on the FET, essentially connecting the 12V input to the output... but remember that current in the inductor has to ramp up... so the output does not jump to 12V, it slowly starts climbing...
During this cycle, the current is ramping up and up in the inductor (remember our equation) -- at some point, we reach the saturation current level of the inductor (that is, we've stored all the energy we can in the finite-sized inductor), and the controller turns off the switch.
Now, after this first cycle, there is little chance that the output has been able to charge all the way to 5V... but that's ok... because it will keep charging even when the FET is off, thanks to our diode.
Remember that the current cannot change instantaneously in the inductor? well, thats still true, even when I disconnect the inductor from the input.
At the instant i shut off the FET, I had a pretty high current flowing in the inductor to the output -- and the inductor will do whatever it can to keep that current flowing, even if it means pulling current from ground to deliver to the output!
So now the inductor is ramping down, delivering its stored energy to the output by pulling current from ground to deliver to the output cap...
Soon, the current will ramp back down to zero, where we start another cycle say, after a few of these cycles where we ramp up the current all the way to Isat, the output is finally at 4.9V...
The next cycle, the controller will shut off the FET when it sees the output at 5V... from then on, it will only keep the FET on long enough to keep the output at 5V -- this will be largely dependent on how much current my 5V circuit is pulling.

In these buck circuits, the input cap must be rated to handle the input voltage, and have a series resistance that keeps the voltage ripple to an acceptable level (voltage ripple is caused by the FET switching on and off, delivering lots of current to the inductor, and then none at all)... the output cap need only be rated to the output voltage, and again be of low enough ESR [Equivalent Resistance in Series - Ed.] to keep the ripple down to whatever the circuit requires. 25V on either the input or output is overkill -- I suppose that on the input, it would give Apple the freedom to specify a 16V or 20V supply, or a cheaper 12V transformer type that floats to 24V.... the ESR is a non-issue -- since there's no circuitry running off this supply directly, there could be 2V of ripple and no one would care. At the output, all we need is a 6V cap -- and since there is always current flowing to the output (its not bursty like the input), the ESR is less of an issue...

I replaced my input cap with two 150µF 16V Sanyo OS-CONs (nice low ESR, great for switching circuits like this), and my output cap with a single 300µF 10V OS-CON. For the input caps, I mounted one on the front and one on the back -- the one on the back I had to cut the EM shielded plastic case to allow the cap to poke through, but it went back in the housing just fine.

Incidentally, my failure happened as I updated the firmware and rebooted the base station... the base station is part of the serial number batch you mentioned for crappy capacitors, but it was way out of warrantee (besides, I hacked it for the lucent range-extender). At first, I had the same one-red-led-at-left, all-amber, rinse, repeat... then I blanked the settings with the reset switch, and then the ABS would just do all-on at random intervals.

Hope this is useful information is helpful... if you'd like to provide a link to dc-dc converter how-to's, I could probably find one or two."

Brian Whitaker

Mr. Morelli's later counterpoint:

"I just read the article by Brian Whitaker on the ABS power supply. While his explanation of how a "buck" converter works is correct, it does not explain the capacitor failure.

Rating the [input] capacitor at 25V is not an overkill, it is the minimum! All pulsewidth modulators (switchmode or switching converters are just other names for them) create ungodly voltage spikes when the transistor(s) switch. Stray capacitances, leakage inductances, diode switching times all contribute. These spikes, albeit narrow, can short out some of the layers used to make capacitors. Typically, about twice the nominal DC voltage on the primary side shows up as a narrow spike. Hence the 25V rating is really needed.

So, it is unlikely that the capacitors failed by internal arcing, unless there was a bad batch. This can only be tracked by the date code (or batch number), but typically it would happen on unit lots built at about the same time.

Now, other things that contribute to spiking could be off in the wrong direction and, although the caps were O.K., they eventually died: too many voltage spikes exceeding the rating. Changing the caps to low ESR helps, but does not fix the problem. It just fixes the symptoms.

I have troubleshot too many switching converters due to fly on spacecraft, where you don't get a chance to change anything once they are up!

Undoubtedly Apple should be held responsible for it: it can be anything from a design flaw to poor quality control (insufficient or nonexisting testing) to just plain bad parts from their vendors. However you slice it, they guarantee the product and, if it is faulty, they are responsible to fix it.

Now, if this could only be done without a lawsuit..."

Luigi Morelli

Bottom Line:

I like that the Panasonic replacement capacitors I usually used not only have a higher capacitance than stock, but that their voltage rating is 25 Volts.

In November of 2002, I received some further insights from Hac Da Le about the ABS power supply that I found very interesting.

Hello,
I would like to share this info with ABS owners. I revived 2 ABSs this week, being a HW engineer I could help but notice that ABS 12VDC to 5VDC converter is not needed, if you use a well regulated 5VDC power supply. None of ABS parts require 12VDC, therefore, if you use a well regulated 5VDC supply, you can by pass the 12VDC-to-5VDC converter, and remove one point of failure out of the design. I recommend to use a 5VDC switching power supply, since switching supplies run cooler, lighter. Here is how you can by pass it:

Lift pins 1 & 2 of the LN2596S (U17)
Jumper across pads 1 & 2 (not the pin)
While you are at it you may want to replace C52 also. This capacitor and the inductor after it took a beating from the charge pump circuit of U17.
Power on the ABS, using a well-regulated 5VDC supply, reconfigure the ABS.

I found that reconfigure the ABS varied, both of them get reset back to V3.52 firmware, but for some reason they didn't behave the same. They are both working now though. That's all you need to do, and don't forget to find something in your house that needs a better 12VDC power :)

Hac.

I replied to his letter and asked why he thought that Lucent chose this design for the first generation base station, i.e. using two power supplies instead of just one. Here is Hac's reply:

I asked myself the same question, but I can't think of any reasons. The 5VDC brick is definitely cheaper than the 12VDC brick, Apple could have saved about $5-$10 per ABS, saved us many problems, and it would be a better design without the 12V to 5V switcher (one less point of failure).

[Lucent apparently designed the hardware - Apple just packaged it differently than Lucent did. Even the firmware was outsourced to Karlnet. It seems like the only "real" work Apple did was to develop the Apple Admin Utility and Assistant. The proof is in the pudding: You can upload the Apple Firmware onto similar Lucent Access Points (RG-1000, RG1100, AP500) and vice-versa. - Ed.]

Apple [Lucent] probably used AMD reference designs, the closest ref design I found was NetSC520. The ABS was based on SC400, both of AMD ref designs used 12V for various items. Such as PCI bus 12V power or 12V programming voltage for older EEPROMs, but none of these applied to the ABS as supplied. I have a few theories:

12VDC was used in the prototype phase, perhaps with RS-232 line drivers, and never
taken out of the design.
Apple ordered trucks load of these 12VDC bricks, after they looked at AMD ref designs. [Unlikely, as Lucent supplied the hardware and probably specified the power supply as well - Ed.]
Drum roll please, this one is bad: monkey see, monkey do. :)

The ABS and Orinoco RG-1000 are based on the same reference design.You sure can post my finding, just to be clear that I won't be liable for other's ABSs.

I forgot to mention in my last email, that my 5VDC switching supply is rated at 1.5A. This is important because the ABS consumed about 1.0A, so the power supply needs to be able to deliver that much current, and then some (to be safe) or the supply won't be able to regulate at 5V.
Lastly, my ABS have been running for a week now without any problems.

Hac.

If you want to try to retrofit such a power supply, follow the repair instructions on the ABS repair page to open the base station. Then proceed as described above. If Hac's description is too technical, leave it to a professional or just repair the ABS the way I describe.

Note: At least one user has experienced severe problems with his ABS after merely pulling Cap 52 and putting the leads of the 5VDC power supply into the holes. Thus, there might be more to this circuit than meets the eye. Alternatively, this could have been an isolated incident...

That should do the trick. Now you have a ABS without a internal power supply, which entails a much lower heat-load. Please click on the following links if you'd like to contact Hac or me to add a comment or two of your own. Thanks!