[hpsdr] "Taylor Corrected DDS"

[Alberto I2PHD]

Bob,

   what about precomputing a table with 2^18 = 262144 entries of cosine values between 0 and pi/2 and indexing into 
that, with a zero-order interpolation, taking the value of the nearest index ? Wouldn't that be both fast and accurate ? 
Granted, you use some RAM, but RAM is not a scarce resource nowadays...

73  Alberto  I2PHD
----------------------------

Robert McGwier wrote:
> 
> Your way will be faster,  smaller, and will have much poorer 
> performance.    The 2^12 * 2^8  will be "secants"  connecting the 2^20 
> points on the circle as defined by the trig identities and the two 
> tables.  The 2^12 linear interpolation will simple yield secants between 
> the 2^12 points in the first table.   The 2^12 * 2^8 will be an exact 
> match to the correct complex numbers on all 2^20 points.  The linear 
> interpolation way will NEVER be on the correct value between the 2^12 
> original points.   This will have significantly higher total harmonic 
> distortion, spurs, etc.
> 
> 73's
> Bob
> N4HY

 1150378657.0