Re: [Discuss-gnuradio] OLPC - next generation with SDR?

[John Gilmore]

> The thing does appear to have sufficient horsepower to do some DSP.
> I would like to think we can make several things available to this
> project.  For example, I think a tunable HF receiver for shortwave AM
> broadcast is easiy achievable for very modest cost. Further out, I
> would to see the use of this machine and OFDM skywave to provide WAN
> capability to large areas of the world without such capability.

If we were given a square inch of circuit board space, twenty cents
for components and wires and connectors, four pins, 0.2 watts of power
when operating, and half a million gates colocated with the CPU and
memory bus, what radio capabilities could we offer to the next
generation OLPC project?

That's the fun challenge.  Here's some background.

The reason software radio hardware has always cost so much is that it
ships in low volumes.  The oscilloscope boards we started with were
$1400.  The USRP is many hundreds, and the USRP-2 will be more.  But
if the USRP's RF I/O capability was integrated onto a high volume
motherboard, it would cost a lot less -- maybe $50 or $25.  If it was
integrated into a chipset, even cheaper.  Similar but specialized
wireless capabilities are in USB wifi dongles that *retail* for $40.

Today's children's XO laptop is just the first in a series of high
volume, low cost laptops -- from a variety of vendors.  We can assume
that with each generation they will get faster, lower power, and
cheaper -- as we learn more about how to design and build in that
problem space.  (Until Dec 31, you can buy one for $400 -- and a
second one will go free to a kid in a developing country.  After that,
they won't be sold at retail.  See http://laptopgiving.org.)

For the next generation effort, if they have the design time, they are
likely to build a big custom chip that integrates a whole CPU, and a
pile of system and peripheral circuitry.  Their stretch goal is a
$50/ea laptop for kids, one that's much better than the current $200
one.

We know they will want 2-channel sound in and out.  They have
already jiggered their current hardware so that the audio biases and
filters can be switched out of the circuit, so that ordinary low
voltage sources and sensors can be plugged into the audio port and
used to sample real-world sensors.  They have full control of the
drivers, since they're basing the whole thing on Linux, and they have
real kernel hackers and real GUI hackers and such.  Their system
already uses wireless WiFi, so it has antennas, and they've done a
detailed radio analysis of the package and design.

The difference between this design effort and the other things the GNU
Radio crew has done is that the result has to cost only incremental
pennies, cost zero power when not running, and run on batteries when
running.  On the other hand, gates and connectors and small antennas
come almost free (they're making hard-tooled plastic molds anyway;
adding a connector or other wires is simple).  Assuming their basic
design provided roughly their current sound and analog input
capabilities, what could we recommend that they do in order to make
the platform much more capable for SDR?

My first thought is to just increase the sample rate and effective
bits per sample of the audio processing hardware, and increase the
number of channels so that ordinary stereo audio can happen
simultaneously with analog I/O.  I think it's a crime that cheap
analog I/O chips top out at 200 kilosamples per second.

Even making it able to receive AM-band and below, by plugging in a
wire of appropriate length as antenna, would provide years of
experimentation in the schools.  Should the analog circuitry be
wired up to the existing "cat ear" antennas on the laptop (which
are currently only used by the WiFi chip)?

The analog circuitry would need to be switchable for use in three
domains:  audio (speaker/mic), DC analog (sensors), and radio analog (SDR).

Could we make it usefully transmit?  Many of Matt's transceiver
daughterboard designs are very similar -- with only a few components
changed to set the frequency range.  If made in high volumes on an
existing board, what would the cost come down to?  Could we shrink a
single band transceiver into the above constraint?  Could we design a
cheap multi-band transceiver that lets these components be switched
under software control?  Can the CPU and the free signal processing
gates automatically measure and compensate for cheap
(signal-distorting) analog circuitry?

What kind of signal processing math hardware should we add to the
custom chip, assuming that the CPU itself would be low power/low
heat/low performance for its time?  Should this be a CPU math
accelerator, or should it be wired to the digitization hardware?
Should it do one thing well (if so, what?) or be more general like
traditional x86/PPC/etc DSP instruction sets?

Should we suggest making some of "our" gates of the custom chip into a
field programmable gate array, reconfigurable in software?  If so,
what would we *use* that capability for (as opposed to putting in hard
circuitry)?

Even a receive-only SDR platform that makes it out to millions of
third world kids would be a brilliant achievement.  (And if we could
get inside their Marvell wireless-mesh chip, we would probably end up
with lots more capabilities in the WiFi bands, too.)

        John