Re: [avr-chat] will an avr be fast enough...

[Matthew MacClary]

On Fri, Jun 03, 2005 at 10:48:47AM -0500, Alan Kilian wrote:
>     Can anyone else come up with an IR transmission scheme that will   
>     work?

    Well I can brainstorm a little.  One option is multiplexing the IR
link so either have an AVR driving two IR transmitters or have two
avrs that each read 8 channels worth of ADC and transmit on their own
IR transmitter. Putting the data back together then becomes a little
harder but it this might be doable depending on what is receiving the
data.
    You could choose to take some data loss by giving up bits of
precision or sampling rate. Also transmitting packed bytes would save
bandwidth; don't transmit 10 bits of data as two bytes rather transmit
all bits of data back to back and then tack on a checksum for the
whole transmission.  That way 10 bits of data will only require
slightly more than 10 bits of transmission. Since the IRs are going to
be well shielded the error rate should be low enough to make this a
win.
    A common technique for reducing bandwidth requirements is
transmitting only changed data. This does make your communication
protocol more complex but it significantly reduces the amount of data
you must transmit. To simplify such a protocol from needing to deal
with variable transmission segment lengths, you could look at the data
coming off of the ADC and find that usually no more than X bits change
between samples. You could then just transmit fixed blocks of X bits
after sending an initial reading. In this scenario, turning up the
sample rate may (or may not) reduce the number of bits you need to
send.
    One more way to encode the data for transmission is a "model
predictive" approach where you assume a model for the ADC data on both
sides of the IR link and then you transmit just enough data so that
the receiver can use the model to predict what the ADC reading is. For
example if you could reasonably assume that a quadratic function in
time would fit your ADC data then you could transmit enough points to
reconstruct a quadratic on the receiver side.  This technique would
result in the lowest bandwidth requirements but would require extra
latency because the model is fit to many successive samples of data.

-Matt


---------------------------------
Where did the universe come from?