OLPC hardware: what if there was an SDR modem / chipset?

[david at lang.hm]

On Mon, 25 Jan 2010, Luke Kenneth Casson Leighton wrote:

> On Mon, Jan 25, 2010 at 4:24 PM,  <david at lang.hm> wrote:
>> On Mon, 25 Jan 2010, Luke Kenneth Casson Leighton wrote:
>>
>>> i've been doing some research and found a couple of companies with SDR
>>> R.F. front-end ICs.  one is 40nm and is so tiny that it will only cost
>>> about $2, mass-produced.  also thanks to being in 40nm, the speed of
>>> the (SoC / embedded) ARM9 core is so fast that it's perfectly capable
>>> of handling multiple protocols.  as you're no doubt aware,
>>
>> are you sure that your $2 SDR chip is going to be able to do a protocol as
>> complex as wifi?
>
> the $ RF front-end chip ends at an "A-D" / "D-A" converter.  after
> that, it's well known that DSPs can handle the rest of the work.

it's not just 'a RF front-end chip', it can get as bad as "a RF receiver 
chip for each band, plus a RF transmitter chip for each band, plus 
switching logic to switch between them, ...."

>> that will take quite a bit of computing power (digital
>> signal processing isn't something you do efficiantly on the ARM9 core), so
>> it may push you from the $2 chip to a $10 chip (what do you think the
>> existing wifi chipsets are?)
>
> ARM9 or ARM7 cores plus a DSP.

which drives up the cost

>> while the chip can decode all the different protocols you mention below, the
>> problem is that these cover many different frequency ranges, and getting the
>> RF portion of the device (including antennas) to handle all those different
>> bands on one device, let alone at one time, is far more complex than you
>> imagine
>
> and there is expertise in the world to take care of it.  admittedly
> it's an area where the expertise is becoming increasingly rare, but
> the expertise exists.

this is expensive expertise to hire.

>> (and would you really want a device that could be wifi, OR GPS
>> OR.... but not more than one at a time?)
>
> a) two $2 transceiver chips - knock yourself out! :)
>
> b) i believe that the RF transceiver chip is multi-channel (simultaneous)
>     so can actually handle two separate protocols at once.
>
> either way, it's just extra load on the DSP.

no, it's seperate portions of the RF spectrum that need to be tuned and 
downconverted to a frequency range the DSP can deal with. those are the 
more difficult parts of the device to get right.

>> you also ignore licensing and regulatory approval.
>
> no, i didn't, but it's worth mentioning again so that it's not
> forgotten as part of the cost.
>
> the key questions to be asking are: in light of the massive volumes
> involved with OLPC XOs, is SDR worth pursuing, given all the
> development costs, but given all the benefits especially where SDR can
> be re-programmed to do "whatever" in bands where licensing and
> regulatory approval is _not_ required, on a per-country basis?

probably not, for the same reason that OLPC isn't writing it's own OS 
anymore. it requires extensive expertise and tools that they don't have.

they funded the developement and deployment of the screen technology, but 
everything else has been assembling off-the-shelf chips and subsystems, 
and with the changeing times, they have less engineering expertise than 
they used to.

David Lang