Answers are grounded in this paper and cite their sources.
N bit digital words can be logarithmically encoded and compressed to a word length of (log2n + m − 1) bit maintaining a relative accuracy of m bit over (n − m) octaves of signal level. A bit-serial VLSI coder is reported, which requires little more than a log2n counter and an output register and it has a latency of one wordlength. The bit-parallel coder can be built with less than n2 transistors and has less than n/4 gate delays. The decoder has similar properties and it expands the logarithm to an antilogarithm with n bit of dynamic range. Using these codecs, digital multiplication, division, powers and roots are reduced to additions, subtractions and shifts, respectively.
No takes yet. Share an insight, caveat, or question.
B. Hoefflinger (1991) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: