oversampling

• Messages sorted by:[ date ][ thread ][ subject ][ author ]

On Tue, 6 Jan 1998, Kragen wrote:

> Penio was talking about. It sounded like his 1-bit A/D was a comparator
> that compared the input level to a fixed, nonzero voltage. 

Well, I used fixed voltage to extend an example that was given. A better
way of saying that is "known in advance" by the DSP behind the comparator. 

Lets revisit the A*sin(w*t) example. When we have sampled time finely
enough that we have the 22222111111111111122222 sub-stream, then we gain
new information from further subdivision of the time only at the two
transitions, if we keep the comparator levels fixed, that is. Obviously,
this is a waste of resources. Is we had the freedom to change the gain in
the 1111111 region, we would uncover a wealth of information there,
otherwise hidden from us.

Revisiting the spiking neuron example, it is a 0/1 stream and if we want
to be efficient, we _need_ to change the threshold level continuously. 
This is some times called desensitization, adaptation, or pre-spike
generation dynamics. 

But who needs efficiency -- we have so many neurons! 

Well, being twice as inefficient as currently would mean a head twice as
big as now (ask your mommy how would she like to give that birth), and
50% more food -- currently the head dissipates about half the calories we
eat (for those in the USA this argument may be moot, but evolution did not
happen in 20-th century USA, remember :-)

--
Penio Penev <Penev@pisa.Rockefeller.edu> 1-212-327-7423

.