Table 1 Summary of spectrum prediction technique
From: Statistical spectrum occupancy prediction for dynamic spectrum access: a classification
Category | Model | Research works | State space/state dependency | Occupancy decision criteria |
|---|---|---|---|---|
Memory less stochastic | Bernoulli/ binomial | x∈[0,1,..,S]/p(x t ) | Channel status | |
source models (Section 5) | Poisson | |||
Exponential | \( x \in \mathcal {R}/p(x_{t}) \) | Duty cycle | ||
Log-normal | Signal/power | |||
Uniform | [69] | \( x \in \left [ 0,1,..,S \right ] \,, x \in \mathcal {R}/p(x_{t}) \) | ||
Finite order Markov models (Section 6) | 2-state Markov chain | x∈[0,1]/p(x t |p(xt−1) | Channel status | |
3-state Markov chain | [59] | x∈[0,1,2]/p(x t |p(xt−1) | ||
High-order Markov chain | [84] | x∈[0,1]/p(x t |xt−m),m>1 | ||
Semi-Markov | \( x \in \left [ 0,1,... \right ] /p(x_{s+t}\,,s>0|x_{t}) \, s \in \mathcal {R} \) | Duty cycle | ||
Continuous time MC | ||||
Hidden Markov model | x∈[0,1,...,S]/p(x t |xt−1) | Channel status | ||
Bayesian models | \( x \in \mathcal {R}/p(x_{t}|x_{t-1}) \) | Signal/power | ||
Finite order linear regression models (Section 7) | Autoregressive | \( x \in \left [ 0,1\right ] \, or \, x \in \mathcal {R}/p(x_{t}|x_{t-1},..,x_{t-m}) \) | Channel status or signal/power | |
Moving average | [100] | |||
ARIMA | [66] | |||
Random walk | [102] | |||
Machine learning statistical based techniques (Subsection 4.3) | Neural networks | \( x \in \left [ 0,1\right ] \, or \, x \in \mathcal {R}/p(x_{t}|x^{t-1})\) | Duty cycle, channel status, or signal/power | |
Support vector machine | [47] | |||
Pattern mining |