GSoC volunteer (as student)

[Patryk Szymczak]

>  It's likely that we coordinate the mentoring while we are doing the
>  rankings.

Good to know, so there is no place for private discussion with
potential mentor about how such project could be done?

>  Sounds impressive.

It is very nice to hear it :)

>  > * Ambient Noise Detection
>
>  I really like to see getting this done. Some nice demo app on top o it
>  showing us what we can do with such things in the future.

Actually I think that there are 2 solutions for ambient noise detection:
1) We will need knowledge from 2 disciplines: signal processing and acoustics.
Solution 1a) consists of:
* once per fixed time interval audio path is activated
* we take a short sound sample (~500ms)
* it is time to put in under processing:
a) filter for detecting and removing "cracks" - e.x. if you have phone
in your trouser's pocket, then there would be lots of such cracks.
b) simple general noise recognition algorithm

Solution 1b) consists of:
* once per fixed time interval audio path is activated. I assume that
whole audio path is activated (both mic and speaker)
* we begin gathering sound sample from microphone
* we generate a fixed sound 'beep' using speaker for ~300ms (a special
preprepared sound sample - a model)
* we stop gathering sound
* it is time to put in under processing:
a) filter for detecting and removing "cracks" - e.x. if you have phone
in your trouser's pocket, then there would be lots of such cracks.
b) sound goes by PCB much faster than through the air so we need to
remove this mess from recorded data
c) simple general noise recognition algorithm based on comparing to
"model sound".

Unfortunately, that solution needs lots of data processing. I can not
estimate it right now, but it might be too heavy.

There is another idea (2):
We need to build some generic algorithm and write "learning" software.
A special genetic filter will be automatically created after testing
in as many different conditions as possible.
That would be learn cases for genetic algorithm.
It will be now easy to use it for sound samples in typical use.

Again: does anyone have sufficient knowledge and is interesting in mentoring me?

>  With your experience in GSM you perhaps also have nice ideas what we
>  can do with this "feature". :) Don't hesitate to offer different
>  applications then the one you can find in the wiki. We are open for
>  ideas here.

Unfortunately, I don't have OpenMoko phone and I am not familiar with
its GSM features so it is hard for me to find something that OpenMoko
lacks of.

I have a little idea for something that probably is not useful for
anything than experiments :)
I am talking about GSM (BSS) positioning. Imagine that OpenMoko GSM
engine performs extended network survey in fixed interval time
(something like AT#CSURVEXT).
It looks for base station identification codes and receiption levels
(in dBm) around phone. Now it is time to put some algorithm (easy one)
and we can estimate direction of movement with accuracy even less that
7 meters :)
Now imagine that before processing, OpenMoko can check for precise GPS
position and after lots of samples taken, we could have some adaptive
algorithm for tuning coefficients in positioning algorithm based on
BSSes.
After some time, we ?could? (maybe) have good enough BSS (GSM)
positioning without GPS!

-- 
----------------------------------------------
Patryk Szymczak