A Comparative Study of Parametric Spectrum Estimation Techniques for Cognitive Radio Using Testbed Prototyping

Cognitive radio (CR) has become an emerging field to rescue wireless communication applications from the spectrum scarcity problem. Spectrum estimation (SE) has been a key ingredient for faster and efficient network implementations using the concept of CR. In this work, we have performed a comparative study of SE technique for the CR systems by employing the null hypothesis approach. Autoregressive (AR), autoregressive moving average (ARMA) and autoregressive integrated moving average (ARIMA) modelling based on optimal data length and goodness of fit (GoF) has been utilized for optimal spectrum modelling. The optimization of the modelling has been achieved through the Akaike information criteria (AIC) and Bayesian information criteria (BIC). Validation and optimization of the time-series data samples have been accomplished using Fit (%) along with \(\chi^{2}\) test GoF. The entire process of SE along with the validation of data samples has been verified on the RICE University’s FPGA-based WARP radio testbed in association with MATLAB. A thorough statistical analysis of variance and Standard Error (SER) of the received samples has been carried out for the optimization of sample time-series data length for optimal performance of the receiver or users. It is noteworthy that, we could achieve a much accurate and frugal SE with a data length of 250 only using ARIMA (3,1,2) model of the data samples with a significant improvement in power spectral density (PSD) compared to other conventional approaches. Extensive experimental work has been incorporated to establish the work.