(21)

Alternatively, a parallel computation structure can be used, with the advantage of reducing the overall latency, especially for large numbers of input values. Other area-latency trade-offs are obtained resorting to approximate formulations of ${\mathrm{max}}^{\ast}$. In [25], it is shown that the Chebyshev inequality can be exploited to approximate (21) as

${\mathrm{max}}^{\ast}\left(a\text{,}b\text{,}c\right)\cong \mathrm{max}\left(a\text{,}b\text{,}c\right)+\mathrm{ln}\left(1+{e}^{-\delta}\right)\text{,}$ (22)

(22)

where $\delta $ is the difference between the first and the second maximum. ...

Start Free Trial

No credit card required