A machine learning approach for dynamic spectrum access radio identification

Dynamic spectrum access (DSA) technologies offer solutions to the spectral crowding associated with static frequency allocation. Hierarchical DSA networks aim at allowing secondary users to efficiently utilize licensed spectrum, while still protecting primary users and ensuring them first priority to spectrum access. However, these networks are often multi-tiered and the concept of different operating policies for secondary users has arisen. In this study we consider the idea of two operating modes in a stochastically modeled DSA network. Observations from the radio frequency (RF) environment are classified using self organizing maps (SOMs). The discretized observations are then utilized to develop hidden Markov models (HMMs) of each type of radio. These models are developed for a variable number of map sizes and hidden states then sequence matched against unknown radios in order to determine identification performance. The system is shown to perform extremely well for certain combinations of SOM sizes and HMM states.