FastBynase reduces this number dramatically by weighting more recent samples more than older ones. When the each sample is weighted 2x the previous sample, only Log2(N) samples are required to produce N possible values.
This is similar to PositionalNotation, except that the weights of each postion change with each sample.
We studied idealized behavior of FastBynase through computer simulation. Here is a frequency sweep that shows the improved frequency response at the expense of considerable dither noise passing through the reconstruction.
Here we look at the same signal in frequence over time at two different resolutions.
Many thanks to PeterCunningham who helped us with these visualizations.
|Last edited April 16, 2006
Return to WelcomeVisitors