Phase-Type Distribution

A phase-type distribution is defined by a continuous-time Markov chain with transient states (phases) and a single absorbing state. The time until absorption—representing an event like a primary user returning to a channel—follows the phase-type distribution. The process evolves through a sequence of exponential sojourn times in each phase before eventual absorption, allowing the modeling of durations with coefficients of variation greater than or less than 1, unlike the rigid exponential distribution.

Every continuous phase-type distribution is fully specified by two mathematical objects:

• α (initial probability vector): A row vector defining the probability of starting in each transient phase.
• T (subgenerator matrix): A square matrix governing transition rates between transient phases.

The cumulative distribution function is given by F(t) = 1 − αe^(Tt)1, where e^(Tt) is the matrix exponential. This compact parameterization enables efficient likelihood computation for statistical inference on channel holding times.

The class of phase-type distributions is dense in the space of all continuous distributions on [0, ∞). This theoretical property guarantees that any primary user activity pattern—whether exhibiting heavy tails, multimodality, or burstiness—can be approximated arbitrarily closely by a phase-type distribution of sufficient order. This makes it a universal approximator for channel holding time and inter-arrival time modeling in cognitive radio networks.

A Coxian distribution is a canonical, minimal-structure phase-type where transitions flow only from phase i to phase i+1 or directly to absorption. Key properties:

• Bidiagonal T matrix reduces parameter count and prevents overfitting.
• Any acyclic phase-type distribution can be transformed into a Coxian representation.
• Widely used in spectrum mobility prediction for fitting empirical primary user ON/OFF period data via expectation-maximization algorithms.

The k-th moment of a phase-type distribution has a closed-form expression: E[X^k] = k! α (−T)^(−k) 1. This enables analytical computation of mean channel holding time, variance, and higher-order statistics directly from the (α, T) representation. The distribution also exhibits a matrix-geometric structure in its probability density function, facilitating efficient numerical evaluation and integration into partially observable Markov decision process (POMDP) frameworks for optimal handoff policy derivation.

Parameter estimation for phase-type distributions from empirical spectrum occupancy data typically employs the Expectation-Maximization (EM) algorithm. The procedure treats the sequence of visited phases as latent variables:

• E-step: Compute expected sojourn times and transition counts given current parameters.
• M-step: Update α and T using closed-form maximum likelihood estimators. This iterative method converges to a local maximum of the likelihood function, yielding a fitted model suitable for proactive spectrum handoff prediction engines.

A stochastic framework representing the temporal behavior of licensed spectrum users, typically modeled as an alternating ON/OFF process. The ON period represents channel occupancy, while the OFF period represents idle opportunities for secondary users.

• Exponential ON/OFF models are common but fail to capture heavy-tailed behavior
• Phase-Type distributions provide a tractable alternative that matches empirical measurements
• Accurate activity models are critical for minimizing forced termination probability

The statistical duration a secondary user can occupy a specific frequency channel before a primary user's return forces a spectrum handoff. This metric directly determines link maintenance probability and quality of service.

• Modeled as the time until absorption in a Phase-Type distribution
• Directly impacts proactive handoff target channel selection
• Empirical measurements often show heavy-tailed distributions that Phase-Type models can approximate

A square matrix defining the probabilities of transitioning between channel states (idle, busy, or intermediate phases) in a discrete-time Markov chain. Forms the core computational structure of Phase-Type distributions.

• Rows sum to 1, representing all possible next states
• The absorbing state represents a primary user arrival event
• Eigenvalues of the matrix determine the distribution's tail behavior and moments