16-Feb-2015, 23:24
(16-Feb-2015, 23:15)Jean-Marie Wrote: It is even better than that: this is floating point. I don't know exactly how the 40 bits are partitioned, I'm more familiar with 32bits where you have 24bits of mantissa and 8 bits of exponent. I suspect that 40 bits is 32 bits of mantissa and 8 bits of exponent. bottom line is that -60dB for instance is 'just doing' -10 on the exponent, leaving the 24 or 32 bits of the mantissa intact, therefore not loosing anything.
Yes, that makes perfect sense - SHARC does have a 40-bit floating-point format with 8-bit exponent and 32-bit mantissa as you said. It's like IEEE-754 single-precision with an extra 8 bits in the mantissa.
Roon (Mac Mini), Wilson Benesch Full Circle, Expert 1000 Pro CI, Kaiser Chiara
Warwickshire, UK