RU2439679C1 - Method of nonparametric signal recognition - Google Patents

Abstract

FIELD: information technologies.

SUBSTANCE: method to recognise a nonparametric signal includes normalisation of survey data, performance of a wavelet conversion operation, signal and noise extraction on a survey data spectrogram, besides, in process of formed matrix training, the normalised survey data is averaged in a sliding matrix window, numbers of lines and columns in which are found on the basis of experiment planning with maximum signal-to-noise ratio, and in process of examination the normalised survey data is averaged in a sliding matrix window with numbers of lines and columns with a maximum signal-to-noise ratio.

EFFECT: higher reliability of signal recognition by a wavelet conversion spectrogram.

3 dwg, 2 tbl

Description

The invention relates to processing a non-parametric radio signal in the case of relatively small values of the signal-to-noise ratio.

A known method of detecting a signal based on nonparametric methods [1].

However, if part of the noisy signal or part of the interference is included in the analyzed interval, this can lead to the erroneous acceptance of the hypothesis H ₀ (no signal) or hypothesis H ₁ (there is a signal). Therefore, it is necessary to distinguish the time interval of the noisy signal and the time interval of the interference in order to apply nonparametric methods separately to the selected observation intervals.

Closest to the expected result is a method for processing a nonparametric signal, including normalization of the observation data, the operation of the wavelet transform, signal extraction and noise on the spectrogram [2].

The disadvantage of this method of signal processing is the decrease in recognition accuracy at relatively low values of the signal-to-noise ratio.

The objective of the proposed technical solution is to develop such a signal processing method, in the implementation of which it would be possible to isolate the signal from the wavelet transform spectrogram even with a signal-to-noise ratio of approximately equal to unity.

The technical result of the invention is to increase the reliability of signal recognition by spectrogram wavelet transform.

The technical result is achieved by the fact that in the method for recognizing a nonparametric signal, including normalizing the observation data, performing the wavelet transform operation, extracting the signal from the spectrogram, normalized observation data are averaged in the “sliding window” matrix, the number of rows and columns of which are found on based on the planning of experiments from the condition of the maximum signal-to-noise ratio, and during the exam normalized data is averaged in a “sliding window” -matrix with the number of rows and one hundred Centers at which there is a maximum signal to noise ratio.

The method is implemented as follows. In the learning process, the observation data, including observations related to interference, and observations related to the signal, are presented in the form of a one-dimensional series. The latter is converted into a multidimensional series using a one-parameter shear procedure.

Averaging is performed in a “sliding window” matrix, the number of rows and columns of which is determined based on the planning of experiments from the condition of the maximum signal to noise ratio. The smoothed signal is normalized, subjected to wavelet transform. The signal is isolated by the spectrogram of the wavelet transform, on which the noise corresponds to the darkened interval.

During the exam, the observation data is normalized, converted into a multidimensional series using a shear procedure. Conduct matrix averaging in a "sliding window" -matrix, the number of rows and columns of which are determined during the training period from the condition of the maximum signal-to-noise ratio. The smoothed signal is subjected to wavelet transform. The signal is isolated by the spectrogram of the wavelet transform, on which the noise corresponds to the darkened interval.

An example of a specific implementation of the method is illustrated by observation data on interference (Table 1) and signal (Table 2).

According to the Wilcoxon criterion, the first sixteen observations are a hindrance, the remaining observations are a signal. The signal-to-noise ratio is 1.14. After normalizing the initial sample of observations and continuous wavelet transform using the Mexican hat wavelet, a wavelet spectrum was obtained (Fig. 1), from which it is not possible to extract a signal.

According to the proposed technical solution, the normalized sample of observations is converted into a multidimensional series using a shear procedure. Conduct matrix averaging in a "moving" window-matrix, the number of rows and columns of which is determined from the condition of the maximum signal-to-noise ratio.

If the averaging matrix has seven columns and five rows, then the signal-to-noise ratio rises to a value of 4.6. The average values of the sequence of observations of interference and signal are shown in figure 2, where the first eleven values refer to the interference, the remaining eleven to the signal. After normalizing the converted sample and the continuous wavelet transform using the Mexican hat wavelet, a wavelet spectrum was obtained (Fig. 3), on which the interface between the interference and the signal is clearly traced.

Information sources

1. Shakhtarin B.I. Signal Detection. M .: Helios ARV, 2006, pp. 201-276.

2. Baranova A.V., Kudryavtseva E.A., Shilova M.N. Determination of the signal localization interval against the background of noise. The collection of materials of the VI Interregional scientific-practical conference "Modern problems of the informatization of education, science and technology." M., 2009, pp. 348-352.

Table 1 Normalized discrete time one 2 3 four 5 6 7 8 X _i 1,5 -0.7 -2.1 0.6 1,1 3.4 0.2 -one 9 10 eleven 12 13 fourteen fifteen 16 0.4 -2.3 2,4 0.4 -2.1 -3 -0.6 1.7

table 2 Normalized discrete time 17 eighteen 19 twenty 21 22 23 24 X _i 1,6 -2.4 2.2 0.6 1,2 3.6 0.2 -1.1 25 26 27 28 29th thirty 31 32 0.4 0.7 2.5 0.5 -2.1 3.6 1,1 1,5

Claims (1)

A method for recognizing a nonparametric signal, including normalizing the observation data, performing the wavelet transform operation, isolating the signal and noise on the spectrogram of the observation data, characterized in that during the training of the generated matrix, the normalized observation data is averaged in a "moving" matrix window, the number of rows and columns which is found on the basis of the design of the experiment from the condition of the maximum signal-to-noise ratio, and during the exam normalized observation data is averaged in the “moving” a matrix window with the numbers of rows and columns at which the maximum signal-to-noise ratio takes place.

Images