Skip to content

Appendix 1: Channel Capacity with State Constraints

Detailed Proof of Theorem 1

This appendix proves the determinant method for computing channel capacity when symbol sequences are constrained by state transitions.


The Setup

States: \(a_1, a_2, \ldots, a_m\)

From state \(i\), symbol \(s\) with duration \(b_{ij}^{(s)}\) leads to state \(j\).

Let \(N_i(T)\) = number of allowed signals of duration \(T\) starting from state \(i\).


The Recurrence

\[N_i(T) = \sum_{j,s} N_j(T - b_{ij}^{(s)})\]

For large \(T\), assume \(N_i(T) \approx A_i W^T\). Substituting:

\[A_i W^T = \sum_{j,s} A_j W^{T - b_{ij}^{(s)}}\]

Dividing by \(W^T\):

\[A_i = \sum_{j,s} A_j W^{-b_{ij}^{(s)}}\]

Rearranging:

\[\sum_j \left(\sum_s W^{-b_{ij}^{(s)}} - \delta_{ij}\right) A_j = 0\]

For a non-trivial solution \((A_1, \ldots, A_m) \neq 0\), the determinant must vanish:

\[\det\left|\sum_s W^{-b_{ij}^{(s)}} - \delta_{ij}\right| = 0\]

The largest real root \(W\) gives capacity \(C = \log W\).


Example: Telegraph Revisited

States:
- \(a_1\): last symbol was not a space
- \(a_2\): last symbol was a space

Transitions from \(a_1\):
- dot (duration 2) → \(a_1\)
- dash (duration 4) → \(a_1\)
- letter space (duration 3) → \(a_2\)
- word space (duration 6) → \(a_2\)

Transitions from \(a_2\):
- dot (duration 2) → \(a_1\)
- dash (duration 4) → \(a_1\)

Matrix:

\[ \begin{vmatrix} W^{-2} + W^{-4} - 1 & W^{-3} + W^{-6} \\ W^{-2} + W^{-4} & -1 \end{vmatrix} = 0\]

Expanding:
\((W^{-2} + W^{-4} - 1)(-1) - (W^{-3} + W^{-6})(W^{-2} + W^{-4}) = 0\)

Simplifies to:
\(W^{-2} + W^{-4} + W^{-5} + W^{-7} + W^{-8} + W^{-10} = 1\)

(The characteristic equation from §1!)


Generalization

This method works for any regular language constraint on sequences. The state graph encodes the grammar, and the determinant equation captures the growth rate of allowed sequences.


Next: Appendix 2 — Maximum Entropy Derivations