Data-Driven Channel Modeling Using Spectrum Measurement

Dynamic spectrum access has been a subject of extensive study in recent years. The increasing volume of literature calls for better understanding of the characteristics of current spectrum utilization as well as better tools for analysis. A number of measurement studies have been conducted recently, revealing previously unknown features. On the other hand, analytical studies largely continues to rely on standard models like the two-state Markov (Gilbert-Elliot) model. In this paper, we present an alternative, stochastic differential equation (SDE) based spectrum utilization model that captures dynamic changes in channel conditions induced by primary users´ activities. The SDE model is in closed form, can generate spectrum dynamics as a temporal process, and is shown to provides very good fit for real spectrum measurement data. We show how synthetic spectrum data can be generated in a straightforward manner using this model to enable realistic simulation studies. Moreover, we show that the SDE model can be viewed as a more general modeling framework (continuous in time and continuous in value) than commonly used discrete Markovian models: it is defined by only a few parameters but can be used to obtain the transition matrix of any N-state Markov model. This is verified by comparing the two-state GE model generated by the SDE model and that trained directly from the data. We show that the GE model is a good fit for the (quantized) data, thereby a fine choice when binary descriptions of the channel condition is sufficient. However, when highly resolution (in channel condition) is needed, the SDE model is much more accurate than an N-state model, and is much easier to train and store.