On the Sample Size for the Estimation of Primary Activity Statistics Based on Spectrum Sensing

Dynamic spectrum access (DSA)/cognitive radio (CR) systems can benefit from the knowledge of the activity statistics of primary channels, which can use this information to intelligently adapt their spectrum use to the operating environment. Particularly relevant statistics are the minimum, mean and variance of the on/off period durations, the channel duty cycle and the governing distribution. However, most DSA/CR systems have limited resources (power consumption, memory capacity, computational capability) and an important question arises of how many on/off period observations are required (i.e., the number of observed on/off periods, referred to as observation sample size in this paper) to estimate the statistics of the primary channel to a certain desired level of accuracy. In this paper, closed-form expressions to link such sample size with the accuracy of the observed primary activity statistics are proposed. A comprehensive theoretical analysis is performed on the required number of observed on/off periods to obtain a specific estimation accuracy. The accuracy of the obtained analytical results is validated and corroborated with both simulation and experimental results, showing a perfect agreement. The analytical results derived in this paper can be used in the design and dimensioning of DSA/CR systems in which the spectrum awareness function relies on spectrum sensing