[Community] GTA04 - Power management improvements

Dr. H. Nikolaus Schaller hns at goldelico.com
Fri Jan 17 09:57:55 CET 2014

Am 16.01.2014 um 22:55 schrieb NeilBrown:

> On Thu, 16 Jan 2014 10:59:02 +0100 Radek Polak <psonek2 at seznam.cz> wrote:
>> On Thursday, January 16, 2014 10:34:05 AM Dr. H. Nikolaus Schaller wrote:
>>> QtMoko is something Radek should comment about, if he wants to upgrade the
>>> kernel.
>> I will only upgrade to kernel that has better power management. The current 
>> situation is bad. IMO everyday charging will kill your battery soon. And you 
>> really cant now the device without charger for more then 24h. Sometimes GTA04 
>> goes to sleep and eats 40mA, so even in sleep it's dead after 24hours. IMO 
>> this has to be fixed so that GTA04 can be used as regular phone. My opinion is 
>> that it makes no sense to work on anything else until this is fixed.
> +1
> This is my big concern regarding the GTA04A45 and the Neo900.
> How do we know they will ever have satisfactory power usage.

We know that the N900 (using the same SoC) is quite good in power management.
So it *is* possible.

What we don't know is what the hardware changes (compared to the N900)
will result in.

> How do we know if the current issues are a consequence of the hardware design
> rather than the software?  We don't!

Well, we know some hardware related design bugs in the A4 board that are
scheduled to be fixed with the A5 (e.g. IrDA vs. RS232 power off; ITG3200 and
HMC replaced by Bosch sensors).

> Until low power usage for a new board can be  positively demonstrated, I
> think it would be very irresponsible to make very many of them.

The problem is that we currently can't even make "not many" of the GT04A5.

Anyways, for the Neo900 there will be a prototype phase with a handful of
devices (~5-10).

The question will be if the others are willing to wait 1 or 2 years until power
management has been ultimatively optimized.

Or if people rather prefer to get the same board (and hoping for software improvements).

> If there was any way that a devel-board could be made so that power feeds can
> be cut and an in-line amp meter inserted for each each module individually,
> the that seems to me (as a non-hardware guy) like it could be very useful.

Puh. It is a space constraint for adding these cut/contact points.

It might be easier to unsolder chips and check if power consumption goes down.

> I'm still very suspicious of the internal USB driver which is documented as
> now allowing the OMAP3 to go to sleep properly....

Yes, that might indeed be a key issue. But if it is the USB-PHY interface or the
EHCI within the OMAP3 SoC we can't measure anything and cut any wire
(on silicon).

One more thought why this issue is so difficult to work on. Because it needs
a reproducible setup to compare different things. This starts with a controlled
hardware setup, network coverage (because the modem may draw 100mA
even if the CPU is off when trying to answer the base station), software setup.

And, such experiments need a lot of time. E.g. compile new kernels to try
something. Flash a new SD card. Reboot. Then start measurements and
compare results.

The results may be difficult to interpret, so the person doing this research
has to study manuals, read web postings, and must get a 100% understanding
of the interworkings. 

This asks for a very big table, a test object GTA04 (not used for anything else every
day) and a separate computer to control measurements and kernel variants.
And an operator who can work on just this topic without major interruptions.

Even then it may take weeks to get to the 100%  understanding.

This is how big design labs like Nokia worked to develop the N900.

IMHO this is comparable with scientific research, trying to answer challenging questions:
* "Why do neutrinos oscillate?"
* "Why does it not go down below 10 mA in suspend"

You alternately build theories, try to verify/falsify by experiments and improve
your theory until you can say: "we have to do this and that to make it better".

And you need some luck. Sometimes by doing the wrong experiment.

Looks like topic for a master or PhD thesis...


More information about the Community mailing list