Improving the accuracy of frequency estimation of radio signal with non-uniform pulse repetition intervals

In radio engineering problems, the use of multichannel frequency filtering allows one to estimate the signal frequency. Digital fre-quency filtering based on the fast Fourier transform introduces frequency estimation error, which is associated with the discreteness of the frequency samples of the fast Fourier transform method. In the case of estimating the frequency of radio signal with non-uniform pulse repetition intervals (PRIs), the use of the classical fast Fourier transforms leads to an additional decrease in the accuracy of frequency estimation compared to estimating the frequency of radio signal with uniform PRIs. Therefore, it is relevant to consider the nature of modulation of radio signal with non-uniform PRIs in the frequency estimation algorithm. This article presents methods for increasing the accuracy of frequency estimation in multichannel frequency filtering of radio signal with non-uniform PRIs. Estimates of the frequency of radio signal with non-uniform PRIs were obtained using various methods and their comparative analysis was carried out. Based on the modified fast Fourier transform algorithm, corrections for the frequency estimate at the output of the fast Fourier transform block were calculated, taking into account the wobble parameters of the radio pulse repetition period. As an example, the results of frequency estimation are given for a signal-noise ratio of 30 dB and a wobble degree of the radio pulse repetition period of 0,2. Corrections to the frequency estimate were calculated, obtained by the maximum likelihood method, which for the 16-point and 128-point fast Fourier transform provides the minimum average value of the average attributed error 1,53⸱10–2 and 0,65⸱10–2, respectively. However, this method is only applicable to a rectangular window when calculating the fast Fourier transform. The interpolating parabola method allows you to apply different weighting windows to calculate the fast Fourier transform, it is not critical to the number of fast Fourier transform channels. Thus, with 16-point and 128-point fast Fourier transform using the Blackman-Harris window, the average values of the average attributed frequency estimation error are 4,27⸱10–2 and 4,30⸱10–2 respectively.