I. INTRODUCTION
In unknown signal detection, the task of an intercept receiver is to decide whether only noise or noise and signal(s) is present. The significance of reliable signal detection is growing rapidly in the communications due to the increasing need for flexible spectrum usage and spectrum sensing, for example, by the means of cognitive radio [1], [2]. Signal detection is based on the statistical hypothesis testing. Let's assume that the received sequence consists of statistically independent complex valued samples . Now, a general detection problem hypothesis can be presented as a selection between the noise-only hypothesis denoted as [3] H_{0}:x_{k}=n_{k},\quad k=1, \ldots, K \eqno{\hbox{(1)}} and the signal-plus-noise hypothesis denoted as H_{1}:x_{k}=\theta s_{k}+n_{k},\quad k=1, \ldots, K. \eqno{\hbox{(2)}} Herein, is a complex zero-mean proper white Gaussian process with variance per dimension, and is a parameter which determines the power level of the signal. We assume additive, white, Gaussian noise (AWGN) channel, i.e., remains constant. If the threshold is exceeded, the signal is declared to be present. If this is caused by the noise-only samples a false alarm occurs. In Neyman-Pearson detection the goal is to maximize the probability of detection while maintaining the false alarm probability on the desired level. According to Neyman-Pearson theorem, the likelihood ratio is the optimal decision statistic but it usually requires too much information to be practically realizable. The conventional and widely studied solution to the unknown signal detection problem is the total-power radiometer, which uses the total energy of the received signal as a decision metric. Unknown signal detection using total-power radiometer was originally studied in [4]. In [5], [6], the performance of the radiometer was studied for measurement purposes. The threshold setting for a digital radiometer was studied in [7]. We assume that the noise variance is perfectly known. In practice, it can be estimated using antenna-off samples or, more accurately, using constant false alarm rate (CFAR) methods [8, Section 6]. Since 1980's cyclostationarity based detectors [9] have been studied extensively. The cyclostationarity based detectors outperform energy detectors if the signals are cyclostationary but are not useful otherwise.