Article Preview
TopIntroduction
Cognitive radio (CR) is a solution to the problem of radio spectrum scarcity. It gives the opportunity to a secondary user (SU) to exploit the spectrum allocated to a primary user (PU). The main function of cognitive radio is spectrum sensing (SS) which has gained new aspects in the last decades to determine opportunistic spectrum holes (Yucek & Arslan, 2009, pp. 116-130).
Many SS techniques have been proposed in literature, these techniques could be categorized into two categories: coherent and non-coherent. In coherent detection such as matched filter (MF), and cylostationary detection technique (CSD), the SU has to know some of the PU features. In non-coherent detection such as energy detection (ED), and eigenvalues based detection, the SU does not need any prior information about PU's signal. ED has low implementation complexity, but it is not effective at low SNR (Al-Hmood, 2015). In MF technique, the secondary user (SU) needs to know the perfect information about the PU signals such as the pilot, preamble and training sequence that are used for channel estimation and synchronization. In addition to its easy implementation, this method has the shortest sensing time to obtain a good detection performance. However, the complexity of MF is very high, because the SU needs receivers for different PU's signals. Another drawback of MF is the estimation error for the PU signal becomes high when SNR is low (Axell, Leus, Larsson, & Poor, 2012, pp. 101-116). Eigenvalues detection is better than ED method for many reasons. Firstly, it overcomes the noise uncertainty problem and it has better performance when the PU's signals are highly correlated (Kortun, Ratnarajah, Sellathurai, Zhong, & Papadias, 2011). Although, CSD can detect the signals with low SNR; it requires long observation time and higher computational complexity holes (Yucek & Arslan, 2009, pp. 116-130).
Because of the limitation of traditional spectrum sensing techniques and for efficient sensing IEEE 802.22 standard, multistage SS has been proposed in several literature (Latha, Gohain, & Chaudhari, 2018; Hamid, Bjorsell, & Ben Slimane, 2016, pp. 630-642; Gunichetty, Hiremath, & Patra, 2015; Anaand & Charan, 2016; Maleki, Pandharipande, & Leus, 2010; Lin, Hung, & Lu, 2019; Mu, Jing, Xie, & Zhang, 2019, pp. 1148-1154; Mourougayane, Amgothu, Bhagat, & Srikanth, 2019). The first driving feature for the multistage detector is the SNR. When the SNR values are high, the simplicity of the first stages is the main advantage. However, when the SNR goes down, it requires more sensing accuracy. Therefore, more accurate detectors with a price of their complication is needed in higher stages to achieve better sensing accuracy (Hamid et al., 2016, pp. 630-642).
Multi-stage SS algorithms can be categorized into three classes: The first class is the sequential scheme where the different detection stages are serially connected and each stage is either performed or skipped depending on the sensing results of its previous stages. The second class is the parallel scheme where different detectors are used at the same time and accordingly the final decision is based on combining these parallel decisions. Finally, the third class is a sequential or parallel detection with an SNR estimation process to decide which stage to use (Hamid et al., 2016, pp. 630-642).
Maximum Eigenvalues Detection (MED) technique provide better detection performance than the others eigenvalues techniques like maximum-minimum eigenvalues (MME) (Zeng, Koh, & Liang, 2008). In addition, MED technique is more accurate than ED technique. The combination of ED and MED techniques makes use of the advantages of each detector in each stage as it exploits the speed of detection of ED at high values of SNR and good detection performance of MED at low values of SNR)Mashta, Altabban, & Wainakh, in press). Furthermore, both techniques do not need any prior information about PU' transmission, and they do not require accurate synchronization. However, they are subject to noise uncertainty due to dependence of the test statistics on the noise power (Zeng et al., 2008).