91Ó°ÊÓ

In simple terms, a stationary process is a statistical process whose probabilistic properties do not alter over time. For this concept to apply, the process must have:

• Consistent mean across different time points
• Variance that remains constant
• Constant autocorrelation over time

In our exercise, we look at the process \( \{T(t): t \geq 0\} \). This process is derived from a Poisson process \( N(t) \) which counts the number of random events occurring over time. Since Poisson processes have a consistent rate, and the random initial direction \( T_0 \) is equally likely to be +1 or -1, the overall behavior of \( T(t) \) remains consistent over time.
If you calculate the statistical properties at any time \( t \) and compare them to another time \( t' \), you'll find no difference. That's why \( T(t) \) is stationary.

The auto-correlation function is a tool used to measure how a signal correlates with itself over different time lags. In the exercise, we're tasked with finding the auto-correlation between \( T(s) \) and \( T(s+t) \).
Mathematically, the correlation is computed with the expectation \( \mathbb{E}[T(s) T(s+t)] \). Given that \( T(t) = T_0 (-1)^{N(t)} \), our job is to understand how this expression behaves as we ''shift'' time by \( t \).
Recall that \( T_0 \) is either 1 or -1, hence \( T_0^2 = 1 \). This simplifies the expression, leaving us to evaluate \( \mathbb{E}[(-1)^{N(s+t)}] \).
The trick here is to use independence properties of \( N(t) \), finding:

• \( \mathbb{E}[(-1)^{N(t)}] = e^{-2\lambda t} \)

This leads to the final correlation function: \( \rho(T(s), T(s+t)) = e^{-2\lambda t} \), which gives the strength of correlation based on how much the two points of interest are separated by time \( t \).

Expectation and variance are fundamental concepts in probability, indicating averages and dispersion. For the random function \( X(t) = \int_{0}^{t} T(s) \ ds \):
Expectation (Mean):
The mean \( \mathbb{E}[X(t)] \) tells us the average value of \( X(t) \). For \( T(s) \), having \( T_0 \) equally likely \( +1 \) or \( -1 \), the mean value is zero. Thus, integrating for any time \( t \) gives an expectation of zero: \( \mathbb{E}[X(t)] = 0 \).

Variance:
The variance \( \text{Var}(X(t)) \) measures how much the values of \( X(t) \) deviate from the mean or expectation. We use the correlation function found earlier to compute this:

• \( \text{Var}(X(t)) = \int_{0}^{t} \int_{0}^{t} \mathbb{E}[T(s)T(u)] \, ds \, du = \frac{t}{2\lambda}(1 - e^{-2\lambda t}) \)

This formula shows that as time increases, the variances of outcomes we observe from our process also increase, proportionally scaled by the rate \( \lambda \).

The random telegraph process is a fascinating example in stochastic processes where the signal flips back and forth between two states at random intervals. This randomness can model systems such as binary communication signals or switching states in electronics.
In this exercise, \( T(t) = T_0(-1)^{N(t)} \) represents a telegraph process where:

• \( T_0 \) is the initial state ( either +1 or -1)
• \( N(t) \) increases when events (or 'switches') occur
• \((-1)^{N(t)}\) indicates switching between states

Thus, the state at any time \( t \) simply depends on the number of underlying Poisson events. These Poisson-triggered flips make \( T(t) \) appear like a signal jumping at exponential random intervals.
The random telegraph process here is stationary due to its reliance on stationary components. Its random nature leads it to mimic real-world signals, providing a valuable model for telecommunications and other applications.