A.B. Tokarev

The need for radio control in a wide range of frequencies generates interest to high-speed algorithms for detection of signals observed together with the additive noise having unknown intensity. One of the ways to reduce the computational complexity of signals detection is the usage of suboptimal algorithms of the additive noise intensity estimation instead of optimum ones. Both algorithms have been analyzed in the work are two-stage procedures. The first steps of both algorithms are identical and bases on the search for minimum value from smoothed energy spectrum. The purpose of processing is preliminary division of spectrum samples to the noise and signal subsets. The final stages of the algorithms are different. The first algorithm calculates the noise average power estimation as a result of averaging the samples, previously referred to noise. In the second algorithm final value is formed by equalization of minimum value from smoothed spectrum, adjusted with the real width of minimization area. The workability of both algorithms is demonstrated on the basis of statistical modelling. It is shown that the algorithm with averaging the samples, pre-classified as noise, provides the smaller standard deviation of the estimates. But for frequency bands, containing a subset of weak signals, this algorithm generates estimates with too much bias. For an algorithm, based on the adjustment of the minimum value of the smoothed spectrum, the influence of weak signals on the quality of the evaluation turns out to be small, but its disadvantage is significantly higher values of standard deviation of the estimations.