On the noise power estimation in the presence of the signal for energy-based sensing

In Cognitive Radio, the key to overcome the spectrum scarcity is reliable spectrum sensing, which will allow efficient opportunistic spectral reuse with minimum interference to licensed users. We consider energy detection based sensing that requires efficient noise power estimation technique in order to reliably detect the primary transmissions. We consider three algorithms, viz. Akaike Information Criterion (AIC), Minimum Description Length (MDL), and Rank Order Filtering (ROF), which estimate the noise power in the presence of the signal. The ROF key property is that it attempts to track the non-flat noise floor. Algorithm probability of missed detection is evaluated using data obtained from USRP2 devices. We illustrate that ROF may gain up to 6dB when signal to be detected is in the sub-band where the noise oor is not at (e.g., at the edges of the observed band).