Skip to content

16. The Channel Capacity in Certain Special Cases

Simplified Capacity Formulas

Shannon derives closed-form capacity expressions for several important channel classes.


1. Noiseless Channel

If the channel is noiseless with \(n\) input symbols and no output constraints:

\[ C = \log_2 n \]

Achieved with uniform input distribution.


2. Independent Noise Per Symbol

If each symbol has independent noise (memoryless channel):

\[ C = \max_{p(x)} \sum_i p(x_i) \sum_j P_{x_i}(y_j) \log \frac{P_{x_i}(y_j)}{p(y_j)} \]

This is the standard discrete memoryless channel (DMC) formula. Optimization is convex in \(p(x)\).


3. Gaussian Approximation to Discrete Channels

For channels with many output levels, the noise can be approximated as Gaussian. Capacity approaches:

\[ C \approx \frac{1}{2} \log_2\left(1 + \frac{S}{N}\right) \]

per channel use, where \(S\) is signal power and \(N\) is noise power. This foreshadows the continuous Gaussian channel result in Part IV.


4. Channel with Symbol Costs

If different input symbols have different transmission costs \(c_i\) and total cost is constrained:

\[ C(\beta) = \max_{p: \sum p_i c_i \leq \beta} I(X;Y) \]

This is a constrained optimization problem solvable with Lagrange multipliers.


5. Sum Channel

If the channel is a mixture of subchannels used with probabilities \(q_i\):

\[ C = \max_{q_i} \sum_i q_i C_i \]

subject to resource constraints on the subchannels.


Summary Table

Channel TypeCapacity FormulaOptimal Input
Noiseless\(\log n\)Uniform
BSC(p)\(1 - H_2(p)\)Uniform
BEC(\(\alpha\))\(1 - \alpha\)Uniform
Z-channel\(\log(1 + (1-p)p^{p/(1-p)})\)Non-uniform
Gaussian (preview)\(\frac{1}{2}\log(1 + S/N)\)Gaussian

Next: §17 — An Example of Efficient Coding