Prof. Matthieu Bloch
Thursday, November 9, 2023
The capacity of the AWGN channel with average power constraint \(P\) \[ C = \frac{1}{2}\log_2 \left(1+\frac{P}{\sigma^2}\right) \]
The distribution that maximizes the relative entropy suject to a second moment constraint \(P\) is the Normal distribution with zero mean and variance \(P\)
A \((n,M_n)\) code \(\calC\) for soft covering over a discrete memoryless source \(P_{Z|X}\) consists of an encoding function \(f_n:\{1,\cdots,M_n\}\to\calX^n\)
For a discrete memoryless channel characterized by \(P_{Z|X}\) and an input \(P_X\), \(C_{\textsf{r}} = \mathbb{I}(X;Z)\)
Consider a generic channel \((\calU,P_{V|U},\calV)\) with message set \(\{1,\cdots,M\}\)
For \(\gamma>0\), let
\[\begin{align*} \calC_\gamma \eqdef \left\{(u,v)\in\calU\times\calV:\log\frac{P_{V|U}(v|u)}{P_V(v)}\leq\gamma\right\} \end{align*}\]
Encoder
For any \(\gamma>0\),
\[\begin{align*} \E[C]{S(C)} \leq \P[P_UP_{V|U}]{(U,V)\notin \calC_\gamma} + \frac{1}{2}\sqrt{\frac{2^{\gamma}}{M}}. \end{align*}\]
Consider a sequence of codes \(\set{(f_n,g_n)}_{n\geq 1}\) such that \(\lim_{n\to\infty}S^{(n)}=0\) and \(\lim_{n\to\infty}\frac{\log M_n}{n}\geq R\). Then \[ R\geq \min_{P:P\circ P_{Z|X}=P_Z}\mathbb{I}(X;Z) \]