Method for adaptive processing of hydroacoustic signals for increasing noise immunity of hydroacoustic devices

Abstract

FIELD: radio electronics; hydroacoustics.SUBSTANCE: the invention relates to methods of signal detection and processing in the presence of interference. The method uses an adaptive algorithm that allows the system to automatically configure and change its parameters in real time depending on the input data and detected signal characteristics, and also provides a higher order of accuracy compared to the analogue for solving the filtering-detection problem. To achieve this, the adaptive system adjusts the coefficients to changes over time. The adaptive filter is configured so that its output signal matches the unknown source signal according to the criterion of the best standard deviation between the simulated random process and its estimate. The adaptive system may function in both the time and frequency domains. This approach helps avoiding the influence of a number of factors, such as the spatiotemporal variability of the ocean's hydrophysical characteristics, noise suppression (highlighting the useful signal against the background of noise), extracting useful information, adjusting parameters in real time, improving resolution with lower computational costs, and implementing an adaptive filter to solve the filtering-detection problem. The invention consists in implementing a generalization of the filtering problem using an adaptive filter to a filtering-detection problem. For this purpose, the adaptive filtering system includes a filtering block that generates an estimate of the process and an adaptation circuit that generates an estimate of the unknown parameters. The filtering block includes: an equation for estimating the state under the hypothesis of the presence of a signal, an equation for estimating the state under the hypothesis of the absence of a signal, equations for estimating the variance of the filter error for each hypothesis, and the corresponding equations for the filter gain coefficients. The adaptation circuit includes: equations for the evolution of the parameter matrix under the hypothesis of the presence of a signal and under the hypothesis of the absence of a signal, equations for the evolution of the correlation matrix of the error in determining the parameters under each hypothesis, equations for the evolution of the derivative of the process with respect to the parameters under the hypothesis of the presence and absence of a signal, equations for the evolution of the derivative of the correlation matrix of the error in determining the parameters for each hypothesis. In addition, a likelihood equation is additionally solved, which includes these estimates, and based on the calculation results, a decision is made about the detection or non-detection of the signal.EFFECT: reduction in computational costs, ensuring high efficiency of signal detection under various interference conditions, obtaining a more efficient filtering algorithm and the ability to operate in real time.1 cl, 4 dwg

Description

[0001] The present invention pertains to the field of radio electronics, specifically to methods for processing and detecting signals against a background of noise interference. The claimed invention is aimed at solving the problem of optimally detecting a hydroacoustic signal under noise conditions described by linear stochastic differential equations using a developed method. To solve this problem, the algorithms propose using an adaptive method for processing hydroacoustic signals.

[0002] Existing adaptation methods are aimed at optimizing the system by reducing the destabilizing influence of a number of factors, such as spatiotemporal variability of the hydrophysical characteristics of the ocean, noise suppression (highlighting the useful signal against the background of noise), extracting useful information, adjusting parameters in real time, improving resolution, and adaptive methods are used for data processing, detection and classification of objects.

[0003] A method for adaptive processing of noise emission signals is known [1]. This method involves receiving the noise emission signal of an object, creating a static fan of vertical directivity characteristics, broadband filtering the signal, measuring the interference level and setting a threshold, detecting the signal in each spatial channel, measuring the signal amplitude in the channels where the noise emission signal of the object was recorded, and making a decision on the presence of a signal in the spatial channel where the signal is maximum. A disadvantage of this system is the need to obtain sound velocity sections along the path, the absence of which significantly reduces the reliability of the obtained estimate of the position of the object emitting noise.

[0004] There is a type of adaptive signal processing known as the angular tracking method using broadband passive systems [2]. This method uses two antenna halves, each with a wideband split-aperture correlator at the output to determine the direction of the noise source and generate an error signal, as well as a Hilbert transform, a multiplier, an integrator, and a directional error signal generation circuit. The phase difference at the output of the processing system is proportional to the deviation of the target position relative to the direction characteristic axis. A disadvantage of this system is the requirement for a high signal-to-noise ratio at the input. Furthermore, under multipath reception conditions, interference between the beams occurs, which distorts the phase difference estimate.

[0005] A known adaptive processing method [3] involves receiving an object's noise emission signal with a single directional characteristic, changing the directional characteristic's position in the vertical plane, determining the directional characteristic's position, and searching for the directional characteristic's position that provides the maximum signal-to-noise ratio. A disadvantage of this method is the need for manual directional characteristic adjustment and its operation only at low speeds of the noise direction-finding station's carrier.

[0006] An adaptive processing method using the difference-range method for determining the spatial position of a noise-emitting object is also known [3]. This method analyzes the time delays between signals during multipath propagation in the vertical plane, calculates the structure of the acoustic field based on the measured speed of sound, and determines the location of the noise-emitting object based on the obtained time delays received in the vertical plane. This system is capable of forming a vertical fan of static directivity characteristics and receiving a noise-emitting signal in the vertical plane.

[0007] Adaptive antenna arrays are known that employ an algorithm for maximizing the output ratio of the useful signal power to the sum of the interference and noise powers [4]. Operation of this type of adaptive antenna array requires a priori information about the direction of arrival of the useful signal. During operation, a complex covariance matrix of the signal and interference is formed. After inverting this matrix, the original vector of complex weighting coefficients of the adaptive antenna array is transformed into a vector of complex weighting coefficients that ensure the maximum value of the signal-to-interference + noise ratio.

[0008] When implementing this method to form a complex covariance matrix of the signal and interference, it is necessary to isolate two quadrature components of the signal in each channel of the adaptive antenna array and perform cross-covariance processing of the signals received by the channels of the adaptive antenna array, which complicates the system. In addition, an increase in the number of channels leads to an increase in the processing time of the covariance matrix. To eliminate these drawbacks, the number of adaptive channels is reduced by combining antenna channels into receiving modules [5]. In the extreme case, the adaptive array degenerates into a two-channel adaptive interference cancellation system, which contains only one correlator and is capable of resisting only one interference. In addition to the above-mentioned drawbacks, the implementation of signal processing in adaptive antenna arrays requires control of the amplitudes and phases of signals in the channels of the adaptive antenna arrays.

[0009] There is a method for adaptive signal processing in a modular phased antenna array. Signals from a known direction are received by an even number of modules arranged symmetrically relative to the antenna array's phase center. These modules have a symmetrical amplitude and complex-conjugate phase distribution. For each pair of symmetrically arranged modules, sum and difference signals are generated. By summing the power of the sum signals of these pairs, we obtain the original sum signal of the antenna array. The covariance matrix of the difference signals of the pairs and the vector of covariance coefficients for the original sum signal and the difference signals are then calculated. The covariance matrix of the difference signals is summed with a diagonal matrix, where a higher diagonal weight corresponds to lower values ​​of the correction phases.Next, a coefficient matrix is ​​formed, and the correction phase vector is determined by multiplying the inverse coefficient matrix by the covariance coefficient vector. After this, the module signal phases are modified in accordance with the determined correction phase vector, and the signals from the pairs of modules with the modified phases are summed, resulting in the output signal of the antenna array.

[0010] A method of space-time adaptive signal processing is known, used in a monopulse shipborne radar with an active phased antenna array (APAA) [6]. The method involves receiving radio signals from a target against a background of active and passive interference using the APAA, after which the received signals are converted into digital format. Then, a two-stage adaptive processing of the received radio signals is carried out simultaneously on the sum and difference channels, and threshold processing and target detection are carried out under active and passive interference conditions. The first stage of processing involves processing the "active interference plus noise" signals, which are received by the APAA elements and signals free from passive reflections. A spatial-covariance matrix "Φ" of the processed signals is formed, based on which an adaptive radiation pattern (RP) of the APAA is created in digital form with deep nulls in the direction of the active interference.At the second stage of processing, passive interference is suppressed using Doppler filtering and multi-window signal processing with the help of the adaptive pattern generated at the first stage.

[0011] This approach to two-stage space-time adaptive signal processing enables sequential suppression of active jamming, and then uses the results of the first stage to suppress passive jamming. This increases the suppression rate of active and passive jamming in monopulse AESA shipborne radars by 15% or more, while also simplifying the signal processing algorithm and reducing jamming time by more than half.

[0012] Also known is a useful model of a system for adaptive modeling of random process filtration, the purpose of which is to expand the functionality of adaptive filters for working with random signal models. Adaptive filters have long been used in the modeling of unknown physical dynamic systems. The same signal is fed to the input of both the unknown system and the adaptive filter, and the filter is configured so that its output signal corresponds to the output signal of the unknown system [7]. The synthesis of an algorithm for generating an adaptive model of a random process is based on the Wiener estimator, which leads to the solution of the Wiener-Hopf equation for the vector of weight coefficients according to the criterion of the least standard deviation between the random process being modeled and its estimate.The key feature of the adaptive model generation algorithm is the forced "mixing" of white noise into the original process, which, in turn, excites an N-dimensional space of weight coefficients. The adaptive filter is configured so that its output signal matches the unknown original signal using the best mean-square approximation criterion.

[0013] Thus, the adaptive filter's impulse response will tend to match the useful message. To create a model of the original process, it is necessary to rewrite the weighting coefficients in another adaptive filter, which is fed modeling white noise to its input, allowing it to generate a model of the original process at the output.

[0014] To solve the problem, the algorithms propose to use an adaptive method for processing hydroacoustic signals.

[0015] The results are correct provided that the filtered process is described in state space up to a vector of constant parameters. With this approach, the structure of the process being evaluated plays a decisive role in the design of the filtering algorithm. In this case, the solution to the filtering-detection problem can be obtained using adaptation procedures.

[0016] The stated problem is solved by the fact that the declared method for processing hydroacoustic signals is based on adapting parameters to changing environmental conditions and signals. This allows the system to automatically adjust and change its parameters in real time depending on input data and detected signal characteristics. The process of optimally restructuring the receiver is carried out in accordance with a quality (optimality) criterion. The choice of optimality criterion is naturally determined by the system's purpose.

[0017] The main property of an adaptive system is the adjustment of coefficients to changes over time. An adaptive system can operate in both the time and frequency domains. The adaptive method is more flexible and capable of operating in a variety of conditions. This adaptive algorithm can adapt based on observed data and adjust to changing conditions without predetermined models.

[0018] STATE OF THE ART - characterized by the presence of the closest analogue:

[0019] - Multidimensional discrete Kalman-Bucy filter [8], [9],

[10] ;

[0020] From Doob's theorem

[11] , it follows that if a random process has a fractional-rational spectral density, then it can be represented as a multicomponent Markov process

[12] , the spectral density of which is described by a fractional-rational function [10, pp. 174-176]. As a result, the mathematical model of the SP is represented by linear vector-matrix stochastic equations of state and observation. As indicated above, such a description is a direct description of the SP, which forms realizations of a random process from white Gaussian noise with zero mean and a delta-shaped correlation function (G(ω)).

[0021] It is known [10, pp. 174-176] that the spectral densities of real signals, with any degree of accuracy, are approximated by a fractional rational function:

[0022]

[0023] where:a j, b k ~ constant coefficients;

[0024] When dynamically representing a signal in the state space, the spectral density of a random process (1) corresponds to a differential equation of the form:

[0025]

[0026] where: f1 is the setting action.

[0027] Denoting the right side in the last relation by n k (t), it can be represented by a system of first-order differential equations:

[0028]

[0029] The solution to system of equations (3), under given conditions, defines a random process that is Gaussian and linear, since the change in state is linearly dependent on the state of the dynamic system itself, describing this process as a stochastic differential equation. In matrix form, expression (3) can be written:

[0030]

[0031]

[0032]

[0033] where: v(t) is a column vector of a random process of dimension n;

[0034]

[0035]

[0036] N1 is a square non-negative definite matrix of two-sided spectral densities.

[0037] Detecting signals against a background of interference is a generalization of the filtering problem. This, in turn, leads to dividing the tasks into two stages. The first stage involves solving the filtering problem, which involves finding an estimate of the interference-to-noise ratio using the minimum mean square error criterion. The second stage involves the detection problem, which involves estimating the mixture ratio of the useful signal and interference to noise using the Neyman-Pearson criterion

[10] ,

[13] ,

[14] ,

[15] .

[0038] Technically, the noise filtering process can be represented as a device that implements a weighted summation of a priori and a posteriori data. This approach is reflected in the Bucy Kalman filter algorithm [9]:

[0039]

[0040] Thus, the system of equations (6), as shown in Fig. 1 (functional diagram of the implementation of a priori and a posteriori data in the Kalina Bucy filter algorithm) can be implemented as a technical device that performs a recurrent algorithm, in which there is a computing unit that performs data extrapolation, i.e. a priori

[0041] information and meter providing a posteriori data Z k . In this case, the assessment at each stage represents the instantaneous output voltage

[0042] The optimal estimate is based on the Multivariate Discrete Kalman-Bucy Filter algorithm (analog), which is a classic method for solving a linear filtering problem. The Multivariate Discrete Kalman-Bucy Filter method is based on a recursive Bayesian approach to calculating the posterior distribution of the system state [9, 16]. This filtering method provides the best linear unbiased estimate (optimal estimate) of a stochastic process. The filter takes into account measurement uncertainty, making it robust to noise and interference.

[0043] The equations for the formation and observation of the process are described by a mathematical model defined by stochastic differential equations [17, 18]. When filtering in discrete form, the signal is observed:

[0044]

[0045] Where is Z i - column vector of dimension m;

[0046] m - number of monitoring channels

[0047] H i- observation matrix of size (m × n);

[0048] n 0,j( . is a column vector of additive BGS in channels of dimension m.

[0049] In this case, the following relationships take place:

[0050]

[0051] N 0,i - symmetric (m × m)-matrix of two-sided spectral densities of additive noises acting in the channel n0(t).

[0052]

[0053] The message is given by a vector-matrix differential equation:

[0054]

[0055] Here X i - a column vector of a message of dimension n;

[0056] A i . - matrix of system coefficients

[0057] n 1,i - a column vector of forming BGS with zero and correlation matrices:

[0058]

[0059] where N 1,i - a symmetric matrix of two-sided spectral densities of size (n × n).

[0060] The linear filtering system, optimal according to the criterion of minimum mean square error, is described by the equations

[0061]

[0062] where K i - matrix gain;

[0063] R i - correlation matrix of filtering errors.

[0064] The Kalman-Bucy filter is an optimal procedure for estimating Gaussian processes [9]. Moreover, unlike the Wiener-Kolmogorov filter, it is also optimal for non-stationary processes. It also applies to processes in which the matrix A in the equation of state changes over time.

[0065] The multivariate continuous Kalman-Bucy filter is a widely used and quite effective recursive computational algorithm that solves a number of problems such as estimating the multivariate dynamic state of a system based on incomplete noisy measurements, smoothing and predicting the values ​​of the system state over time, etc. However, its application and implementation are associated with a number of limitations.

[0066] These restrictions include:

[0067] • The filter assumes that the system changes gradually, and with sudden changes, the filter may not accurately assess the state;

[0068] • The filter estimates may be highly dependent on the initial values, and the filter cannot handle missing data;

[0069] • The filter requires knowledge of the process and observation covariance matrices. If the noise covariance matrix is ​​chosen incorrectly or the system is nonlinear, the filter will diverge and produce unreliable estimates;

[0070] • The filter assumes that both the state model and the observation model are linear and the Gaussian noise has a normal distribution.

[0071] The functional diagram implementing the algorithm of the multidimensional discrete Kalman-Bucy filter is shown in Fig. 2, where:

[0072] Block 1 - Setting Initial Conditions

[0073] Block 2 - matrix of system coefficients;

[0074] Block 2.1 - multiplier;

[0075] Block 3 - Calculation of a priori (predicted) assessment values and the correlation matrix of the filtering error

[0076] Block 3.1 - matrix transposition block;

[0077] Block 3.2 - multiplier;

[0078] Block 3.3 - adder;

[0079] Block 4 - monitoring functions;

[0080] Block 4.1 - matrix transposition block;

[0081] Block 5 - device for measuring (observing) process Z i ;

[0082] Block 6 - Correction of forecast data, calculation of the current estimate ;

[0083] Blocks 6.1, 6.5 - multipliers;

[0084] Blocks 6.2 - gain multiplier [-1];

[0085] Block 6.3, 6.6 - adders;

[0086] Block 6.4 - gain;

[0087] Block 7 - filter error correlation matrix R i ;

[0088] Block 7.1, 7.2, 7.4, 7.5 - multipliers;

[0089] Block 7.3 - adder;

[0090] Blocks 8, 9 - correlation matrices of observation noise N0 and forming noise N1;

[0091] Block 8.1 - matrix inverse transformation block.

[0092] At the current stage of development of filtering methods, adaptive filtering of processes under conditions of parametric uncertainty (prototype) can be distinguished.

[0093] The structure of the evaluated process plays a decisive role in the synthesis of the filtering algorithm

[19] . The filtered process is described in the state space with an accuracy of a vector of constant parameters. With this approach, the structure of the evaluated process plays a decisive role in the synthesis of the filtering algorithm. The solution to the filtering-detection problem in this case can be obtained by using adaptation procedures. An adaptive filtering system, as a rule, can include a filtering block, which generates an estimate of the process, and an adaptation loop, which generates an estimate of the unknown parameters. With this approach, the solution to the optimal filtering problem also includes the identification of the model of the filtered process

[20] .

[0094] The formal statement of the problem will look as follows. Let x(t) be a vector random process representing the filtered noise, accessible from indirect measurements of Z k at discrete moments in time is described by the following system of stochastic differential equations written in difference form:

[0095]

[0096] where x k - is the interference state vector, and A is the forming matrix (5), consisting of the parameters a to be adapted, n i - discrete vector white noise of formation with zero mathematical expectation and correlation matrix N1, n0 - discrete vector white noise of observation with zero mathematical expectation and correlation matrix N°, Z k - input implementation dimension.

[0097] It is necessary to obtain an unbiased estimate of the filtered noise x(t) at time instants based on the results of all observations available at this point, and obtain a vector of parameter estimates and the interference estimation vector minimizing error variance using an adaptive algorithm, θ=0 is the detection parameter under the hypothesis of absence of a signal, θ=1 is the detection parameter under the hypothesis of presence of a signal.

[0098] We will not specify the mathematical description of the process at the input of the processing system. This problem formulation is somewhat related to the classical problem formulation of synthesizing an adaptive filtering system for a given model of the process being filtered [20, 21, 22] and is well suited to practical needs, since in practice the same filtering system is used to solve various problems. In this case, the adaptation problem is solved as an adaptation of the given processing system for filtering and detecting various processes.

[0099] The estimate of the vector of adaptable unknown parameters and the variance of this estimate are obtained from stochastic differentials:

[0100]

[0101]

[0102] Equations (12) and (13) describe the block of parameter estimates a.

[0103] Equations describing the non-adaptive block, i.e. equations for noise estimates × and the filtering error correlation matrix R i , coincide with those of the analogue (formula above) In addition, equations for derivatives must be added to the system: which are obtained by differentiating with respect to a, the corresponding equations for the estimates × and the correlation matrix of the filtering error R i .

[0104] The equations for the derivatives included in the system of equations are as follows:

[0105]

[0106] The optimal filter is described by differential equations (Kalman-Bucy filtering equation):

[0107]

[0108] Then the adaptive system of equations describing the optimal filter will look like:

[0109]

[0110] Equations (16) completely describe an adaptive filter defined by its structure and parameters. For unknown parameters, the adaptive system is constructed from two blocks: the main non-adaptive block (the filter block), described by equations (15), where estimated values ​​of the parameters are used instead of their true values, and a block of estimates of the unknown parameters, equations (12, 13).

[0111] The differential equations that determine the variances, together with the equations for the derivatives, describe the block for generating estimates of unknown parameters (spectral width and intensity). The generated estimates are entered into the main filtering block in place of the true values ​​of the unknown parameters.

[0112] The obtained adaptive filtering algorithms correspond to a non-Gaussian approximation of the joint probability density of the distribution of Markov noise and unknown parameters, since its derivation took into account the dependence of the variance of the filtering error of Markov noise on the unknown parameters.

[0113] The functional diagram implementing adaptive filtering of processes under conditions of parametric uncertainty is shown in Fig. 3, where:

[0114] Block 10 - Setting Initial Conditions

[0115] Block 11, 12, 13 - multiplier;

[0116] Block 14 - Gain Multiplier [-1];

[0117] Block 15, 16 - adders;

[0118] Block 17, 18 - multipliers;

[0119] Block 19, 20 - adders;

[0120] Block 21 - gain multiplier [-1];

[0121] Block 22 - matrix transposition block;

[0122] Block 23, 24 - multiplier;

[0123] Block 25 - adder;

[0124] Block 26 - Gain Multiplier [-1];

[0125] Block 27 - gain multiplication block;

[0126] Block 28 - surveillance functions;

[0127] Block 29 - matrix transposition block;

[0128] Block 30 - multiplier;

[0129] Block 31 - matrix transposition block;

[0130] Block 32 - adder;

[0131] Block 33, 34, 35, 36, 37 - multipliers;

[0132] Block 38, 39, 40 - adders;

[0133] Block 41 - matrix transposition block;

[0134] Block 42 - multiplier by coefficient [2];

[0135] Block 43, 44 - multiplier by coefficient [2];

[0136] Block 45, 46 - multipliers;

[0137] Block 47 - Gain Multiplier [-1];

[0138] Block 48 - adders;

[0139] Block 49 - Calculating the Score ;

[0140] Blocks 50, 51 - correlation matrices of observation noise N0 and forming noise;

[0141] Blocks 52 - matrix inverse conversion block.

[0142] The essence of the invention lies in generalizing the filtering problem under conditions of parametric uncertainty, using an adaptive algorithm under conditions of reverberation interference, to the problem of adaptive filtering-detection in vector-matrix form.

[0143] Above, an optimal receiver of a broadband deterministic hydroacoustic signal was synthesized using the Markov nature of reverberation interference. It was assumed that the statistical characteristics, namely the correlation function, are known (the spectral density of the reverberation interference is proportional to the energy spectrum of the emitted signal). However, in general, the spectral density of the reverberation may not coincide in shape with the energy spectrum of the signal. Some of the parameters describing the spectral density of the reverberation interference (RI) may be unknown a priori. Thus, the problem of optimizing the processing system under conditions of parametric uncertainty arises [23, 24].

[0144] For the signal filtering-detection problem, the corresponding linear stochastic differential equation describing reverberation interference with unknown bandwidth and intensity

[25] can be written as follows:

[0145]

[0146] where: A is a vector of unknown parameters characterizing, respectively, the width of the spectrum of Markov interference and its intensity, dimension M×M;

[0147] р(t) _ additive interference, dimension N×1;

[0148] Z(t)- input realization of dimension M ×1;

[0149] N0 - spectral density of observation noise, dimension N×1;

[0150] N1 - spectral density of the noise formation, dimension N×1;

[0151] n0(t) - column vector of generating white noise, dimension N×1;

[0152] S(t) is a deterministic signal of dimension M × 1;

[0153] θ=0- detection parameter under the hypothesis of no signal;

[0154] θ=1- detection parameter under the hypothesis of signal presence.

[0155] Let us assume that the observation equation (17) is:

[0156]

[0157] where: - assessment of interference under hypotheses of presence and absence of a signal, respectively;

[0158] - interference assessment, under the hypothesis of the presence of a signal.

[0159] A realization is observed that is described by the equation:

[0160]

[0161] With precisely known parameters, the optimal filter is described by differential equations (the filtering equation under the hypothesis of the presence and absence of a signal):

[0162]

[0163] where R 2 - variance of the Markov noise filtering error.

[0164] When the detection parameter θ values =0 or θ =1The system of differential equations (19) will determine the filtering block in the absence or presence of a signal, respectively. For unknown parameters, the adaptive system is constructed from two blocks: the main non-adaptive block (the filtering block), where estimated values ​​are used instead of the true parameter values, and a block of estimates of the unknown parameters. The equations modeling the operation of the second block under various hypotheses are as follows:

[0165]

[0166] The equations for the derivatives included in the system of equations (20) are obtained by differentiating (19) with respect to the parameter vector:

[0167]

[0168] Equations (19-21) completely describe the adaptive filter, defined by its structure and parameter vectors. Moreover, equations (20), together with the equations for derivatives (21), describe the block for generating estimates of unknown parameters (spectral width and intensity). The generated estimates are entered into the main filtering block (20) in place of the true values ​​of the unknown parameters. To implement an optimal detector [3, 5], equations (19-21) must be supplemented with an equation for calculating the likelihood ratio:

[0169]

[0170] The improved adaptive algorithm will be as follows:

[0171]

[0172] Where:

[0173] - interference estimate, under the hypothesis of the presence of a signal, M × M;

[0174] - interference estimate under the hypothesis of no signal, M × M;

[0175] R i- correlation matrix of the filtering error under the hypothesis of the presence and absence of a signal, M × M;

[0176] K i . _ filter gain, M × M;

[0177] H - matrix of observation functions, M × M;

[0178] A 1,j - equations of evolution of the parameter matrix under the hypothesis of absence of a signal, M × M;

[0179] A 0.i - equations of evolution of the parameter matrix under the hypothesis of the presence of a signal, M × M;

[0180] R A0.i - equations of evolution of the correlation matrix of the error in determining the parameters under the hypothesis of absence of a signal, M × M;

[0181] R A1,i - equations of evolution of the correlation matrix of the error in determining the parameters under the hypothesis of the presence of a signal, M × M;

[0182] - equations of evolution of the derivative of the process with respect to parameters under the hypothesis of absence of a signal, M × M;

[0183] - equations of evolution of the derivative of the process with respect to parameters under the hypothesis of the presence of a signal, M × M;

[0184] - equations of evolution of the derivative of the correlation matrix of the error in determining the parameters under the hypothesis of absence of a signal, M×M;

[0185] - equations of evolution of the derivative of the correlation matrix of the error in determining the parameters under the hypothesis of the presence of a signal, M×M;

[0186] L - likelihood ratio evolution.

[0187] Thus, the resulting equations completely describe the optimal detector of a deterministic signal against a Markov reverberant noise background with unknown parameters. The detector's structure includes blocks for filtering the noise under various hypotheses, a block for estimating the noise model parameter, and a detection indicator.

[0188] The functional diagram of the implementation of the adaptive detection algorithm is shown in Fig. 4, where:

[0189] Blocks 53, 67 - block for setting initial conditions under the hypothesis of no signal and under the hypothesis of the presence of a signal

[0190] Blocks 54, 68 - blocks for extrapolation of the estimate in the presence and absence of a signal;

[0191] Blocks 55, 69 - blocks for extrapolation of the estimate dispersion in the presence and absence of a signal;

[0192] Blocks 56, 70 - blocks for assessment correction, if available and absence signal;

[0193] Blocks 57, 71 - blocks for correcting the dispersion of the filtering error under the hypothesis of the presence of a signal and the absence of R i ;

[0194] Blocks 58, 72 - observation noise correlation matrices N0;

[0195] Blocks 59, 73 - correlation matrices of the forming noise N1;

[0196] Blocks 60, 74 - correlation matrices of observation functions H in the presence and absence of a signal;

[0197] Block 61 - a meter that provides observed data;

[0198] Block 62, 75 - blocks for correction of the forming matrix in the presence of A 1,i and in the absence of signal A 0.i ;

[0199] Blocks 63, 76 - filter gain K1;

[0200] Blocks 64, 77 - blocks for correcting the dispersion of the adaptation error of the parameters of the forming matrix in the presence of R A1,j and the absence of R A0,j signal;

[0201] Blocks 65, 78 - blocks for calculating the derivative of the process evaluation based on the parameters of the forming matrix in the presence of and absence signal;

[0202] Block 66, 79 - blocks for calculating the derivative of the dispersion of the parameter adaptation error in the presence of and absence signal;

[0203] Block 80 - Signal Implementation Block S i ;

[0204] Block 81 - adder;

[0205] Block 82 - adder

[0206] Blocks 83 - Gain Multiplier (-1)

[0207] Blocks 84 - adder;

[0208] Block 85 - multiplier;

[0209] Block 86 - gain multiplier block [2];

[0210] Block 87 - Gain Multiplier (-1)

[0211] Block 88 - adder;

[0212] Block 89 - adder;

[0213] Block 90 - adder;

[0214] Block 91 - matrix transposition block;

[0215] Block 92 - gain multiplier (-1);

[0216] Block 93 - Matrix to Inverse Conversion Block

[0217] Block 94 - threshold device.

[0218] Blocks 53-66 are combined into functional unit 1 implementing the adaptive filter algorithm, in accordance with the functional diagram of figure 3 under the hypothesis of absence of a signal {θ=0);

[0219] Blocks 76-79 are combined into functional unit 2 implementing the adaptive filter algorithm, in accordance with the functional diagram of figure 3 under the hypothesis of the presence of a signal (θ=1) in the input implementation;

[0220] Blocks 80-94 are combined into functional unit 3, which implements the operation of solving the likelihood ratio evolution equation, in accordance with the thirteenth equation of system (22);

[0221] As shown in Figure 4, the invention involves adding functional units 2 and 3 to functional unit 1 (the prototype). Functional unit 2 essentially replicates the prototype, solving the problem of filtering out interference for the signal presence hypothesis. Functional unit 3, on the other hand, is a device whose computational results are used to determine whether a signal has been detected or not.

[0222] Otherwise, the invention can be interpreted as adding a functional unit 3 to the functional units 1, 2 of Figure 4, solving filtering problems using an adaptive filter for opposite hypotheses of the presence or absence of a signal.

[0223] Four drawings are attached to the description:

[0224] Figure 1 - Functional diagram for explanation of the recurrent algorithm.

[0225] Figure 2 - Analogue. Functional diagram of the multidimensional discrete Kalman-Bucy filter.

[0226] Figure 3 - Prototype. A functional diagram implementing adaptive filtering of processes under parametric uncertainty is presented.

[0227] Figure 4 - Proposed method. Functional diagram of the implementation of the adaptive detection algorithm.

[0228] List of references

[0229] 1. Russian patent of 2015 under IPC G01S3 / 80. Description of the invention patent RU 2561010 C1. Method for adaptive processing of a noise emission signal.

[0230] 2. Burdik VS: "Analysis of hydroacoustic systems". L.: Shipbuilding, 1988, p. 372.

[0231] 3. Koryakin Yu.A., Smirnov S.A., Yakovlev G.V.: “Ship hydroacoustic technology.” St. Petersburg: Nauka, 2004, p. 81, p. 265.

[0232] 4. Monzingo R.A., Miller T.W. Adaptive antenna arrays. Introduction to the theory. Moscow: Radio and Communications, 1986, pp. 80-86, 179-240.

[0233] 5. Russian patent of 2013 under IPC H01Q 1 / 28. Description of the invention patent RU 2491685. System for simplifying the processing of a reconfigurable beamforming circuit in a phased antenna array for a telecommunications satellite.

[0234] 6. Russian patent of 2020 under IPC G01S 7 / 36 (2006.01), G01S 13 / 28 (2006.01) and H01Q 3 / 26 (2006.01). Description of the invention patent RU 2735216 C2. Method for spatio-temporal adaptive signal processing in a monopulse shipborne radar with an active phased antenna array.

[0235] 7. Widrow B., Stearns S. Adaptive signal processing. Translated from English. - Moscow: Radio i Svyaz, 1989. - 440 p.

[0236] 8. Kulikova M.V., Kulikov G.Yu. Numerical methods of nonlinear filtering for signal processing and measurements. Center for Applied Mathematics, Higher Technical Institute, University of Lisbon, Portugal, Computing Technologies, Vol. 21, No. 4, 2016.

[0237] 9. Kudryavtseva I.A. Analysis of the efficiency of the extended Kalman filter, sigma-point Kalman filter and sigma-point particle filter. Scientific Bulletin of Bauman Moscow State Technical University. Vol. 19, No. 2, 2016.

[0238] 10. Butyrsky E.Yu. Methods of modeling and estimating random variables and processes. St. Petersburg: Strategy of the Future, 2020. 642 p.

[0239] 11. Oak J. L. Probability processes. Moscow 1956.

[0240] 12. Stratonovich R.L. Conditional Markov processes and their application to optimal control theory.

[0241] 13. Sosulin Yu.G. Theory of detection and estimation of stochastic signals. Moscow: Sovetskoe Radio, 1978. 319 p.

[0242] 14. Kazakov V.A. Introduction to the Theory of Markov Processes and Some Radio Engineering Problems. Moscow: Sovetskoe Radio, 1973. 232 p.

[0243] 15. Tikhonov V.I., Harrisov V.N. Statistical analysis and synthesis of radio engineering devices and systems. Moscow: Radio and Communications, 1991. 608 p.

[0244] 16. Sinitsyn I.N. Kalman and Pugachev filters. Moscow 2006.

[0245] 17. Gelgor A.L. Methods of modeling random variables and random processes. St. Petersburg: Publishing house of the Polytechnic University, 2012.217 p.

[0246] 18. Butyrsky E. Yu. Mathematical models of hydroacoustic signals and methods of their processing. St. Petersburg: Strategy of the Future, 2018. 650 p.

[0247] 19. Ogarkov M.A. Methods of statistical estimation of parameters of random processes. Moscow, 1990.

[0248] 20. Stratonovich R.L. Principles of adaptive reception. Moscow, Sov. Radio, 1973.

[0249] 21. Perov A.I. Adaptive filtering of messages with unknown statistical characteristics. Radioelectronics. 1980. V.23. No.4.

[0250] 22. Butyrsky E.Yu. On the issue of adaptive signal reception. Fundamental and applied research in the modern world. Proceedings of the XI International scientific and practical conference, September 29, 2015, Volume 3, pp. 19–33.

[0251] 23. Kazakov I.E., Gladkov D.I. Methods of optimization of stochastic systems. Moscow, Nauka. 1987. 303 p.

[0252] 24. Tolstoy I., Kley K.S. Ocean Acoustics. Moscow, Mir. 1969. 298 p.

[0253] 25. Butyrsky E. Yu. Signal detection against the background of Markov reverberation interference. Scientific instrument making. 2012. Vol. 22. No. 3. .87-95.

Claims

A method for adaptively detecting a deterministic hydroacoustic signal against a background of Markov reverberation interference with unknown parameters and white noise, based on adaptive filtering, including obtaining measured values ​​of the input realization Z(t), arriving at the input of the processing system and containing Markov reverberation interference, white noise and, under the hypothesis of the presence of a signal, a deterministic signal S(t), forming an estimate of the interference and the variance of the filtering error of the Markov interference according to the criterion of the minimum mean square error, estimating the vector of unknown parameters A, characterizing the width of the spectrum of the Markov interference and its intensity, and the variance of the error in estimating the said parameters, forming equations for the derivatives of the estimates of the interference and the variance of the filtering error according to the said vector of unknown parameters, solving the equation for calculating the gain coefficients of the filter in the adaptive filter, characterized in that,that in order to solve the signal detection problem, at each processing step for a detection parameter equal to 0 under the hypothesis of no signal, the noise filtering equations are solved under the hypothesis of no signal and the equations modeling the operation of the unknown parameter estimation unit under the hypothesis of no signal, with the formation of an noise estimate, a correlation matrix of the Markov noise filtering error, an estimate of the vector of unknown parameters and the variance of the estimation error of the said parameters under the hypothesis of no signal, and for a detection parameter equal to 1 under the hypothesis of the presence of a signal, the noise filtering equations are solved under the hypothesis of the presence of a signal and the equations modeling the operation of the unknown parameter estimation unit under the hypothesis of the presence of a signal, with the formation of an noise estimate, a correlation matrix of the Markov noise filtering error, an estimate of the vector of unknown parameters and the variance of the estimation error of the said parameters under the hypothesis of the presence of a signal,wherein the generated estimates of the unknown parameters are entered into the main filtering block instead of the true values ​​of the unknown parameters, after which the equation for the evolution of the likelihood ratio L is solved, including the estimates generated under the hypothesis of the absence of a signal and under the hypothesis of the presence of a signal, and based on the results of calculating the likelihood ratio L using a threshold device, a decision is made on the detection or non-detection of a deterministic hydroacoustic signal in the input implementation.

Images