RU2667462C1 - Method of recognizing low-speed speech coding protocols - Google Patents

Abstract

FIELD: data processing.SUBSTANCE: invention is intended for recognition of low-speed speech coding (LSSC) protocols. Technical result is achieved by increasing the dimension of measured redundancy coefficient vector ϕup to ϕ, L = Z + 2 and taking into account the shift effect of the vector elements ϕby forming square reference matrix Φfor all J known LSSC protocols, j = 1, 2 …, J. To this end, a digital stream Y is received for a predetermined time interval ΔT. Rectangular information matrix Yis formed, the rows of which are information blocks placed successively under each other. Redundancy coefficient vector ϕis calculated, measured vector ϕis elementwise compared with strings ϕ(l) of all J square reference matrices Φ, deviation ΔP(l) between measured vector ϕand the rows of all J reference matrices Φ, is determined, it is decided in favor of the jLSSC protocol, for which minimum deviation ΔP(l)of measured vector ϕfrom the l-th row of j-th square reference matrix Φis provided.EFFECT: technical result is to increase the accuracy of the recognition of the LSSC protocols.1 cl, 6 dwg

Description

The invention relates to the field of communication technology, in particular to the processing of digital information, and can be used in a wide class of data transmission systems for recognizing protocols for low-speed speech coding (NSCR) and verifying the correct use of vocoders in digital communication systems.

The claimed technical solution expands the arsenal of means of a similar purpose due to the possibility of recognizing NSCR protocols with higher accuracy under the influence of interference.

A known method for recognizing radio signals (see RF Patent 2551903 Method for recognizing radio signals of the authors: Dvornikova A.S., Dvornikova S.V., published June 10, 2015), in which, at the first stage, distribution matrices are formed from sampled and quantized samples of reference radio signals energy based on their frame wavelet transforms. Then from them, starting from the second line, feature vectors are formed by row-wise concatenation of all wavelet coefficients. Next, the elements of the feature vectors are normalized and their parameters are calculated. In this case, the average value of the normalized amplitude values of the elements of the feature vectors is determined as parameters, and the decision is made based on the results of calculating the differences in the values of the parameters of the recognized radio signal and reference radio signals. The recognized radio signal is considered to be incident with the reference radio signal, the module of the difference in the parameters of the feature vectors with which will be minimal.

The disadvantage of this method is the lack of accuracy of recognition of the NSCR protocols.

, z=1,2,…, Z; Z=N_б-2, содержащих по три соседних столбца прямоугольной информационной матрицы Y_K×L, где текущий номер z прямоугольной матрицы

определяется номером столбца матрицы Y_K×L, являющимся первым столбцом матрицы

, по каждой из матриц

вычисляют коэффициент избыточности ϕ_z, размерности L, на основе полученного набора коэффициентов избыточности формируют измеренный вектор коэффициентов избыточности ϕ_z путем последовательного их размещения в соответствии с порядковыми номерами z, формируют эталонные вектора коэффициентов избыточности ϕ_{zj эт} размерности L известных протоколов НСКР, на основе эталонных векторов коэффициентов избыточности ϕ_{Lj эт} формируют квадратные эталонные матрицы Ф_{Lj эт}, j=1,2,…, J, поэлементно сравнивают измеренный вектор коэффициентов избыточности ϕ_L, со строками

, всех J квадратных эталонных матриц Ф_{Lj эт}, определяют отклонения

между измеренным вектором коэффициентов избыточности ϕ_L, и строками всех J квадратных эталонных матриц Ф_{Lj эт}, принимают решение в пользу j-го протокола НСКР, для которого обеспечивается минимальное и меньше заданного порога R отклонение

измеренного вектора ϕ_L от

-й строки j-й квадратной эталонной матрицы Ф_{Lj эт.} A known method for recognizing low-speed coding (see RF Patent 2610285 Method for recognizing protocols for low-speed coding by authors: Aladinsky V.A., Kuzminsky S.V., Smirnova P.L., Chubaty D.N., published 08.02.2016), consisting in the fact that a binary digital information stream Y is received during the time interval ΔT, based on the received stream Y, a normalized autocorrelation function A is formed, and at regular intervals with equal intervals Δτ, the extrema of the autocorrelation function A decide on the presence of a block structures in the digital information stream Y, at intervals between the extrema of the autocorrelation function A, divide the digital information stream Y into information blocks of N _b bits each, sequentially assign sequence numbers k = 1,2, ..., K to the information blocks, starting from the first information block, form a rectangular information matrix, Y _{K × L} , L = N _b , the rows of which are sequentially placed under each other information blocks in accordance with their serial numbers k = 1,2, ..., K, form a sequential rectangular matrices

, z = 1,2, ..., Z; Z = N _b -2, containing three adjacent columns of a rectangular information matrix Y _{K × L} , where the current number z of the rectangular matrix

determined by the column number of the matrix Y _{K × L} , which is the first column of the matrix

, for each of the matrices

calculate the redundancy coefficient ϕ _z , dimension L, on the basis of the obtained set of redundancy coefficients form the measured vector of redundancy coefficients ϕ _z by sequentially placing them in accordance with serial numbers z, form the reference vectors of redundancy coefficients ϕ _{zj et} dimension L of the known NSCR protocols, based on the reference the redundancy coefficient vectors ϕ _{Lj et} form the square reference matrix Φ _{Lj et} , j = 1,2, ..., J, compare the measured vector of redundancy coefficients ϕ _L , with rows

, all J square reference matrices Φ _{Lj et} , determine the deviations

between the measured vector of redundancy coefficients ϕ _L , and the rows of all J square reference matrices Φ _{Lj et} , decide in favor of the j-th NSCR protocol for which a minimum and less than a given threshold R deviation is provided

measured vector ϕ _L from

ith row of the jth square reference matrix Φ _{Lj et.}

This method, in the absence of interference in the digital information stream Y, has a high probability of correct recognition of the NSCR protocols. The disadvantage of the prototype is to reduce the likelihood of correct recognition of the NSCR protocols when single and group bit errors occur in the digital information stream Y, which are typical for radio channels.

, в которые входит z-й столбец, что приведет к искажениям соответствующих им вычисленных значений коэффициентов избыточности ϕ_z-2, ϕ_z-1, ϕ_z. Последнее приводит к троекратному учету одного и того же ошибочно принятого бита при формировании измеренного вектора ϕ_L, и, как следствие, к снижению вероятности распознавания протоколов НСКР. Кроме того, аналог требует значительных вычислительных затрат при своей реализации, что определяется троекратным использованием столбцов матрицы Y_KxL для вычисления признаков распознавания.The appearance of bit errors in the digital information stream Y is random. Therefore, the influence of even one bit error in the Y _{K × L} matrix on the recognition efficiency of the NSCR in the analogue method is significant. Obviously, the existing one bit error in the _zth column of the matrix Y _KxL appears in three rectangular matrices

, which includes the z-th column, which will lead to distortions of the corresponding calculated values of the redundancy coefficients ϕ _z-2 , ϕ _z-1 , ϕ _z . The latter leads to three-fold accounting of the same erroneously received bit during the formation of the measured vector ϕ _L , and, as a result, to a decrease in the probability of recognition of NSCR protocols. In addition, the analogue requires significant computational costs in its implementation, which is determined by using the columns of the matrix Y _KxL three times to calculate recognition features.

, z=1,2,…, Z; Z=N_б-2, содержащих по три соседних столбца прямоугольной информационной матрицы Y_K×L, где текущий номер z прямоугольной матрицы

определяется номером столбца матрицы Y_K×L, являющийся первым столбцом матрицы

, по каждой из матриц

вычисляют коэффициент избыточности ϕ_z, на основе полученного набора коэффициентов избыточности формируют вектор измеренных коэффициентов избыточности ϕ_z путем последовательного их размещения в соответствии с порядковыми номерами z, формируют эталонные векторы коэффициентов избыточности ϕ_{zj эт} известных протоколов НСКР, вектор измеренных коэффициентов избыточности ϕ_z сравнивают с эталонными векторами коэффициентов избыточности , ϕ_{zj эт} j=1,2,…, J, известных протоколов НСКР.The closest in technical essence to the claimed one is a method for decoding a cyclic noise-tolerant code (see Timofeev D.I., Tavalinsky D.A., Chubaty D.N. Analysis of parameters of low-speed speech encoders in the context of structural and parametric uncertainty // High-tech technologies, No. 8, 2011. P. 4-9), consisting in the fact that the digital information stream Y is received during the time interval ΔТ, based on Y, a normalized autocorrelation function A is formed, according to regular extrema of the autocorrel at equal intervals Δt function A make a decision about the presence of a block structure in the information stream Y, at intervals between the extrema of the autocorrelation function A, divide the information stream Y into information blocks of N _b bits each, sequentially assign sequence numbers k = 1,2, ..., K to the information blocks, starting from the first information block, form a rectangular information matrix Y _{K × L} , L = N _b ; the rows of which are successively placed under each other information blocks in accordance with their serial numbers k = 1,2, ..., K, form a sequence of rectangular matrices

, z = 1,2, ..., Z; Z = N _b -2, containing three adjacent columns of a rectangular information matrix Y _{K × L} , where the current number z of the rectangular matrix

determined by the column number of the matrix Y _{K × L} , which is the first column of the matrix

, for each of the matrices

calculate the redundancy coefficient ϕ _z , on the basis of the obtained set of redundancy coefficients form a vector of measured redundancy factors ϕ _z by sequentially placing them in accordance with serial numbers z, form reference vectors of redundancy coefficients ϕ _{zj et} known NSCR protocols, compare the vector of measured redundancy coefficients ϕ _z with reference vectors of redundancy coefficients, ϕ _{zj et} j = 1,2, ..., J, of the well-known NSCR protocols.

This method allows you to recognize the NSCR protocols.

However, the disadvantage of the prototype is the relatively low recognition accuracy of the NSCR protocols in the absence of N _p characters in the first block of the digital information stream Y. The probability of the absence of characters in the first block of the digital information stream Y is determined by an expression of the form

.

.

, что вызывает сдвиг значений вектора коэффициентов избыточности ϕ_z на величину

. В этом случае в векторе ϕ_z будут отсутствовать значения коэффициента избыточности ϕ_z с номерами z=N_п, N_п+1, а также добавляются значения коэффициента избыточности ϕ_z с номерами z=N_б-1, N_б. Значения N_б цифровых потоков, формируемых вокодерами, которые разработаны для радиостанций диапазона ВЧ, могут составлять 45, 48, 54, 60, 64, 66, 81 или 96 бит. Таким образом, высока вероятность искажения значений в векторе коэффициентов избыточности ϕ_z, что приводит к ошибочным решениям в способе-прототипе. Дополнительно точностные характеристики прототипа снижаются из-за некорректного сокращения длины измеренного и эталонного векторов ϕ_z и ϕ_{zj эт} соответственно на два элемента. Последнее приводит к неполному использованию информации о структуре принятого сигнала, и как следствие, к снижению точности распознавания протоколов НСКР.When the condition 0≤N _p <N _b is fulfilled, in the information matrix Y _{K × L} there is a shift of the columns to the left by the amount

, which causes a shift in the values of the vector of redundancy coefficients ϕ _z by

. In this case, the redundancy coefficient ϕ _z with numbers z = N _p , N _p +1 will be absent in the vector ϕ _z , and the redundancy coefficient ϕ _z with numbers z = N _b -1, N _b will be added. The values of N _b digital streams generated by vocoders, which are designed for radio stations in the high frequency range, can be 45, 48, 54, 60, 64, 66, 81 or 96 bits. Thus, there is a high probability of distortion of values in the vector of redundancy coefficients ϕ _z , which leads to erroneous decisions in the prototype method. Additionally, the accuracy characteristics of the prototype are reduced due to incorrect reduction in the length of the measured and reference vectors ϕ _z and ϕ _{zj et} respectively by two elements. The latter leads to the incomplete use of information about the structure of the received signal, and as a result, to a decrease in the recognition accuracy of the NSCR protocols.

In addition, the implementation of the prototype involves a large amount of computation on the formation of the vector of the measured redundancy coefficients, which reduces its performance.

The purpose of the claimed technical solution is to develop a method for recognizing NSCR protocols, which provides increased recognition accuracy under the influence of interference and its speed.

, из матрицы Y_KxL с номерами

, вычисляют значения математического ожидания

появления определенных импульсов по каждому столбцу

, информационной матрицы Y_KxL, формируют вектор измеренных значений математического ожидания

последовательным размещением полученных значений математического ожидания

в соответствии с их порядковыми номерами

, формируют набор М векторов m

значений математического ожидания

на основе сформированного вектора значений математического ожидания m(0) путем последовательного циркулярного сдвига его значений на величину

, формируют эталонные векторы значений математического ожидания m_jэт, j=1,2,… J, по каждому цифровому информационному потоку Y_jэт, соответствующем j-му известному протоколу НСКР, последовательно сравнивают каждый вектор значений математического ожидания

оцениваемого протокола НСКР с эталонными векторами значений математического ожидания m_jэт, j=1,2,… J, вычисляют значения вероятности правильного распознавания

j-го протокола НСКР по каждому

-му вектору значений математического ожидания

, принимают решение в пользу j-го протокола НСКР, для которого обеспечивается максимальное значение вероятности правильного распознавания

This goal is achieved by the fact that in the known method for recognizing NSCR protocols, including receiving a digital information stream Y for a time interval ΔT, forming a normalized autocorrelation function A based on the received stream Y, deciding on the presence of a block structure in the digital information stream Y according to regular equal intervals Δτ to the extrema of the autocorrelation function A, according to the intervals between the extrema of the autocorrelation function A, the division of the digital information stream Y into information He blocks with a volume of N _b bits each, sequential assignment of information numbers to the information blocks k = 1,2, ..., K, starting from the first information block, the formation of a rectangular information matrix Y _{K × L} , L = N _b , the rows of which are sequentially placed each under another information blocks in accordance with their serial numbers k = 1,2, ..., K, alternately highlight the columns

, from matrix Y _KxL with numbers

calculate math values

the appearance of certain impulses for each column

, information matrix Y _KxL , form a vector of measured values of mathematical expectation

sequential placement of the obtained values of the mathematical expectation

according to their serial numbers

form a set of M vectors m

expectation values

based on the generated vector of mathematical expectation values m (0) by successive circular shift of its values by

, form the standard vectors of values of mathematical expectation m _jet , j = 1,2, ... J, for each digital information stream Y _jet corresponding to the j-th known NSCR protocol, each vector of values of mathematical expectation is sequentially compared

of the evaluated NSCR protocol with reference vectors of the values of mathematical expectation m _jet , j = 1,2, ... J, the values of the probability of correct recognition are calculated

j-th protocol of the NSCR for each

vector of mathematical expectation values

, decide in favor of the j-th NSCR protocol, for which the maximum value of the probability of correct recognition is provided

Due to the new set of essential features in the claimed method, an increase in the probability of correct recognition of the NSCR protocols is achieved by forming a vector of measured values of the mathematical expectation m (0), which eliminated the repeated influence of the same bit error on the recognition results.

The reduction in the dimension of the source data for the formation of the vector of the measured values of the mathematical expectation allowed to achieve an increase in the speed of the proposed method.

The inventive method is illustrated by drawings, in which:

in FIG. 1 - a generalized algorithm for recognizing NSCR protocols;

in FIG. 2 - appearance of the autocorrelation function of the information flow Y;

in FIG. 3 - the order of formation of the information matrix Y _KxL ;

информационной матрицы Y_KxL;in FIG. 4 - column selection order

information matrix Y _KxL ;

in FIG. 5 - simulation results;

in FIG. 6 is a generalized block diagram of an NSCR protocol recognition device.

An analysis of the use of modern low-speed speech encoders in communication systems indicates that, despite the desire of the international telecommunication union to switch to the use of standard encoders, a new compression algorithm and the corresponding NSCR protocol are often developed, which leads to a variety of low-speed speech codecs (see Bondarenko M. .F., Rabotyagov AV, Schepkovsky SV Modern methods of encoding a speech signal. - Kharkov: KHNURE Bionics of intelligence, No. 2 (69), 2008). Under conditions of structural and parametric uncertainty, in connection with an increase in the number of NSCR protocols, the task of recognizing them becomes more complicated, the solution of which is necessary for subsequent decoding of speech signals. This task is further compounded by a complex signal-to-noise environment in the high-frequency range.

Due to the fact that different encoders of speech signals form different structures of blocks of digital streams (CPUs), the analysis of the latter consists in determining the number of bits in one of them and identifying the structure of a block of CPUs based on an analysis of the conditional probabilities of transitions of parameter values from block to block over the entire length linear prediction windows.

In vocoders used in foreign radio stations of the HF range, implemented are the coding method (MK) with ten linear prediction coefficients (LP) LPC-10 (Linear Predictive Coding), MK with LP and mixed excitation MELP (Mixed Excitation Linear Prediction), MK with mixed multiband excitation MICEX (Mixed-Multiband Excitation). The main parameters of the CPU are: Speed _Wok CPU (600, 800, 1200 or 2400 bps), the number of N _b bits in one information block (the length of the information block) and the duration t _{b of} its transmission. The latter is asked at the stage of vocoder development (see Bykov S.F., Zhuravlev V.I., Shalimov I.A. Digital telephony. Textbook for high schools. - M .: Radio and communications, 2003. - 144 p.). The speed V _{wok of the} output CPU is related to the parameters N _b and t _{b by the} expression B _wok = N _b / t _b , from which N _b = V _wok ⋅t _b (see Aladinsky V.A., Kuzminsky S.V. Analysis of digital streams at the outputs of vocoders used on foreign radio lines of the high frequency range. - M .: Successes of modern radio electronics. No. 7, 2015. P.73-76.). The duration of one voice message is determined by the formula ΔT = K⋅t _b .

и m_jэт соответственно, и представляемых в виде последовательности столбцов

матрицы Y_KxL. В результате битовые ошибки оказывают влияние на результаты только по одному разу каждая. Влияние сдвига в строках информационной матрицы Y_KxL (из-за отсутствия N_п символов в первом блоке цифрового информационного потока Y) устраняется путем формирования набора измеренных векторов значений математического ожидания

последовательным циркулярным сдвигом значений вектора m(0) на величину

и сравнением всех элементов этого набора.A positive effect in the proposed method is achieved by reducing the dimension of the source data used to form the vectors of the measured and reference values of the mathematical expectation

and m _jet, respectively, and represented as a sequence of columns

matrices Y _KxL . As a result, bit errors affect the results only once each. The influence of the shift in the rows of the information matrix Y _KxL (due to the absence of N _p characters in the first block of the digital information stream Y) is eliminated by forming a set of measured vectors of mathematical expectation values

successive circular shift of the values of the vector m (0) by

and comparing all the elements in this set.

Implementation of the claimed method can be carried out as follows (see Fig. 1). At the preparatory stage, reference vectors of mathematical expectation values m _{jet are calculated} . For a better understanding, this operation is considered after the description of the action to form the measured mathematical expectation vector m (0).

In the process, take a binary digital information stream Y of length N _cp bits, the number of which is determined based on the expression

After that, the normalized autocorrelation function A is calculated, which characterizes the linear dependence of the original sequence Y on the sequence Y (τ), circularly shifted by τ = 1,2, ..., N _cn -1 bits with respect to Y. The normalized autocorrelation function of a real speech signal, the view of which is shown in FIG. 2 has distinct local maxima located at periodic intervals Δτ. Therefore, Δτ corresponds to the length of the information block. The presence of local maxima of the autocorrelation function located with a period Δτ is explained as follows. The parameters of the speech signal are located inside the information block strictly in a certain place. At shifts Δτ, which are multiples of the length of the information block, the ACF value will be the largest. The desired autocorrelation function (see Fig. 2) is considered a set of values of the form

где α(τ)=c(τ)/D(Y) - коэффициент корреляции; с(τ) - коэффициент ковариации; D(Y) - дисперсия цифрового информационного потока Y.

where α (τ) = c (τ) / D (Y) is the correlation coefficient; c (τ) is the covariance coefficient; D (Y) is the dispersion of the digital information stream Y.

The result of the correlation analysis is the detection of a periodic component in the digital information stream Y. In this case, the frequency value is equal to the length of the information block of the speech signal.

Using regular extrema of the normalized autocorrelation function A at equal intervals Δτ, they decide on the presence of a block structure in the information flow Y. The absence of extrema of the normalized autocorrelation function A with equal intervals Δτ indicates that the digital information flow Y does not contain a speech message, and its analysis ends.

If there are extrema of the normalized autocorrelation function A with equal intervals Δτ, the number of information blocks contained in the digital information stream Y is calculated as follows:

At the next stage, at intervals Δτ between the extrema of the normalized autocorrelation function A, the information stream Y is divided into information blocks of N _b bits each (see Fig. 3). Sequentially assign information blocks to serial numbers k = 1,2, ..., K, starting from the first information block. A rectangular information matrix Y _KxL , L = N _b , is formed (see Fig. 3), the rows of which are information blocks sequentially placed under each other in accordance with their serial numbers k = 1,2, ..., K.

из информационной матрицы Y_KxL с номерами

(см. фиг. 4). По совокупности элементов каждого столбца

вычисляют математическое ожидание (МО) их появления.Next, the columns are selected in turn

from the information matrix Y _KxL with numbers

(see Fig. 4). By the totality of the elements of each column

calculate the expected value (MO) of their appearance.

-го столбца

находят по известной формуле (см. Вентцель Е.С., Овчаров Л.А. Теория вероятностей и ее инженерные приложения. - М: Наука, 1988. - 480 с. - ISBN 5-02-013748-0):In the general form of MO

column

find according to the well-known formula (see Ventzel E.S., Ovcharov L.A. Probability theory and its engineering applications. - M: Nauka, 1988. - 480 p. - ISBN 5-02-013748-0):

где

- вероятность появления i-го значения символа в

столбце

, i=1,2, …, I. В связи с тем, что цифровой поток Y является бинарным (т.е. I=2, а значения символов: 1 или 0), тогда из (4) имеем:

Where

is the probability of occurrence of the ith value of the symbol in

column

, i = 1,2, ..., I. Due to the fact that the digital stream Y is binary (i.e., I = 2, and the values of the characters are 1 or 0), then from (4) we have:

столбце

вычисляется по формуле:The probability of occurrence of a value of 1 in

column

calculated by the formula:

- количество символов со значением 1 в

столбце

.Where

- the number of characters with a value of 1 in

column

.

элемент эталонного вектора m_jэт j-ого класса определяют из выраженияIn turn, in the formation of the reference vectors of values of mathematical expectation

the element of the reference vector m _{jet of the} j-th class is determined from the expression

-

элемент w-й реализации обучающей выборки j-го класса, w=1,2,…, W, w - порядковый номер реализации обучающей выборки; W - объем обучающей выборки.Where

-

element of the wth implementation of the training sample of the jth class, w = 1,2, ..., W, w - serial number of the implementation of the training sample; W is the size of the training sample.

j-го протокола НСКР по каждому

-вектору значений математического ожидания

Для рассматриваемого случая J>2 используют следующее выражение вероятности распознавания (см. Ту Дж., Гонсалес Р. Принципы распознавания образов / Пер. с англ. - М.: Мир, 1978, - 411 с.)In the next step, the probability of correct recognition is calculated

j-th protocol of the NSCR for each

vector of mean values

For the case under consideration, J> 2, use the following expression of the probability of recognition (see Tu J., Gonzalez R. Principles of pattern recognition / Transl. From English. - M .: Mir, 1978, - 411 pp.)

- плотность распределения j-класса. Обычно плотность

описывают нормальным законом распределения (ЗР). Выбор последнего является приемлемым допущением и единственным вариантом, при котором обеспечивается анализ многомерного ЗР. Нормальная плотность распределения, описываемая многомерным набором признаков

принадлежащая j-му классу, имеет вид (см. Ту Дж., Гонсалес Р. Принципы распознавания образов / Пер. с англ. - М.: Мир, 1978, - 411 с.)

where p _j

is the distribution density of the j-class. Usually density

describe the normal distribution law (SP). The choice of the latter is an acceptable assumption and the only option in which the analysis of multidimensional SR is provided. Normal distribution density described by a multidimensional set of features

belonging to the j-th class, has the form (see Tu J., Gonzalez R. Principles of pattern recognition / Transl. from English. - M .: Mir, 1978, - 411 pp.)

- усредненная ковариационная матрица набора эталонных векторов значений математического ожидания m_jэт; w=1,2,…, W, w - порядковый номер реализации обучающей выборки; W - объем обучающей выборки;

- матрица, обратная

- математическая операция транспонирования.Where

- averaged covariance matrix of a set of reference vectors of values of mathematical expectation m _jet ; w = 1,2, ..., W, w - serial number of the implementation of the training sample; W is the volume of the training sample;

is the inverse matrix

- mathematical operation of transposition.

принимают решение в пользу j-го класса, для которого значение вероятности распознавания

является наибольшимAccording to the resulting set of recognition probability values

decide in favor of the j-th class for which the recognition probability value

is the largest

.In addition, expression (10) allows us to determine the probability of correct recognition of the NSCR protocol:

.

Simulation using digital information streams received from vocoders LPC-10-2400 (STANAG 4197) and MELPe-2400 (STANAG 4591) showed the following (see Fig. 5). In the absence of bit errors in the digital information stream Y, the probabilities P _{r of the} correct recognition of the NSCR protocols for the claimed recognition method and prototype are approximately equal (P _p ≈ 0.93). When the number of errors in the digital information stream Y, determines the probability of an error P _err, there is a relatively higher value of the probability P _p of correct recognition NCIS protocols for the proposed technical solution (see. FIG. 5, curve 2). The latter indicates its higher noise immunity compared to the prototype, ceteris paribus. In the limiting case, when it is still possible to restore the voice message (P _OS = 0.15), the gain in the probability of correct recognition of the NSCR protocols P _p when using the proposed method is ~ 0.14.

In addition, the proposed method has a higher speed due to the reduction in the volume of calculations. In the prototype, the redundancy coefficients ϕ _z calculated from the additionally generated z = N _b −2 rectangular matrices are used as initial data. In the proposed method, the columns of the information matrix Y _KxL act as the initial data, based on which the MOs are calculated. As a result, a threefold reduction in the volume of calculated data is achieved.

, второй вычислитель 8, предназначенный для вычисления значений математического ожидания

, формирования набора М векторов измеренных значений МО

(выражения (4)-(6)), третий вычислитель 11, предназначенный для вычисления значений вероятности правильного распознавания

на основе поэлементного сравнения векторов

значений математического ожидания из набора М с эталонными значениями векторов математического ожидания m_jэт J известных протоколов НСКР в соответствии с (8)-(10), и блок принятия решения 13, предназначенный для определения принадлежности анализируемого цифрового информационного потока Y к наиболее вероятному j-му протоколу НСКР. Кроме того, устройство содержит первый блок памяти 9 и генератор тактовых импульсов 10.The proposed method for recognizing NSCR protocols can be implemented using a device whose generalized structural diagram is shown in FIG. 6. The device contains a series-connected input selector 4, intended for registration (storing a predetermined volume of the input digital information stream Y), a first calculator 5, designed to calculate the normalized autocorrelation function A in accordance with (2) and based on it to determine the values of N _b and K (expressions (1) and (3), respectively), the first _driver 6, designed to generate the information matrix of the received signal Y _KxL , the second _driver 7, designed to generate the sequence column strengths

, the second calculator 8, designed to calculate the values of the mathematical expectation

forming a set of M vectors of measured MO values

(expressions (4) - (6)), the third calculator 11, designed to calculate the probability values of correct recognition

based on elementwise vector comparison

mathematical expectation values from set M with reference values of mathematical expectation vectors m _jet J of known NSCR protocols in accordance with (8) - (10), and decision block 13, designed to determine whether the analyzed digital information stream Y belongs to the most probable jth NSCR protocol. In addition, the device comprises a first memory unit 9 and a clock generator 10.

и

. По второй установочной шине 2 поступают значения эталонных векторов математического ожидания m_jэт для всех J известных протоколов НСКР, расчет которых осуществляют в соответствии с (7) на подготовительном этапе и далее хранят в блоке 9.At the preparatory stage, the first minimum and maximum signal sample volumes are set on the first installation bus 1

and

. The second installation bus 2 receives the values of the reference mathematical expectation vectors m _jet for all J known NSCR protocols, which are calculated in accordance with (7) at the preparatory stage and then stored in block 9.

, но не менее чем

. Эти значения задаются по шине 1. В противном случае на второй группе его выходов (первой выходной шине устройства 12) формируется сигнал об окончании работы устройства и необходимости перейти к анализу другой информационной последовательности.In the process of operation of the device (see Fig. 6), the digital information stream Y enters the input of block 4 through the input bus 3. The functions of block 4 include its storage in volume

but not less than

. These values are set on bus 1. Otherwise, a signal is generated on the second group of its outputs (first output bus of device 12) about the end of the device and the need to proceed to the analysis of another information sequence.

The memorized sample of the digital information stream Y under the influence of the clock pulses of block 10 from the output of block 4 goes to the group of information inputs of the first calculator 5. The functions of block 5 include the calculation of the normalized autocorrelation function A, its analysis to detect regular extrema of the autocorrelation function A at regular intervals Δτ , splitting the received digital information stream Y into information blocks of N _b bits each according to their number in the intervals between the extrema of the normalized autocorrelation A . With the first group of information outputs of block 5 block at a N _b bit information sequence Y is supplied to a group of information inputs of the first generator 6. Otherwise, when the unit 5 are detected regularly at equal intervals Δτ extrema of the function A, the second group of information outputs its a signal is generated about the completion of the analysis of the received information sequence, which is fed to the first output bus 12 of the device.

(см. фиг. 4).In block 6, the information matrix Y _{KxL is formed} by sequential block recording of the digital information stream Y in accordance with FIG. 3. The value of the information matrix Y _KxL in parallel code then goes to the group of information inputs of the second shaper 7. In block 7, the information matrix Y _KxL is sequentially _divided into a sequence of L columns

(see Fig. 4).

последовательно поступают на группу информационных входов вычислителя 8.The resulting column values

sequentially arrive at the group of information inputs of the calculator 8.

, осуществляют расчет значений математического ожидания т, в соответствии с выражениями (4) - (6). Далее из полученных значений m

в блоке 8 формируют вектор значений МО m(0). После этого, благодаря последовательному циркулярному сдвигу его значений на величину

, формируют набор

векторов математических ожиданий анализируемого цифрового потока Y.In block 8, as the columns arrive

, calculate the values of the mathematical expectation m, in accordance with the expressions (4) - (6). Further, from the obtained values of m

in block 8 form a vector of values of MO m (0). After that, due to the sequential circular shift of its values by

form set

expectation vectors of the analyzed digital stream Y.

поступают на вторую группу информационных входов третьего вычислителя 11. На вторую группу его информационных входов поступают значения эталонных векторов m_jэт МО для всех J известных протоколов НСКР. Поступление

и m_jэт на входы блока 11 синхронизируется импульсами блока 10. В функции блока 11 входит расчет значений вероятности правильного распознавания Р_j

для всех J протоколов НСКР по каждому

вектору значений МО

в соответствии с выражениями (8)-(10).The found values of the set of M vectors

arrive at the second group of information inputs of the third computer 11. The second group of its information inputs receives the values of the reference vectors m _{j et} MO for all J known NSCR protocols. Admission

and m _jet to the inputs of block 11 is synchronized by pulses of block 10. The function of block 11 includes calculating the values of the probability of correct recognition of P _j

for all J NSCR protocols for each

vector of values of MO

in accordance with expressions (8) - (10).

далее поступают на группу информационных входов блока принятия решения 13. В функции последнего входит поиск максимального значения max

из всей совокупности L×J полученных в блоке 11 значений вероятностей

. В качестве идентифицированного протокола НСКР принимают j-й протокол, соответствующий max

.The values obtained in block 11

then they go to the group of information inputs of the decision block 13. The function of the latter includes the search for the maximum value max

from the entire set of L × J probabilities obtained in block 11

. The j-th protocol corresponding to max

.

The implementation of the elements of the device recognition Protocol NSCR is known. The input selector 4, the first and second shapers 6 and 7, respectively, the memory unit 9 can be implemented on memory chips (see. Large integrated circuits of memory devices: Reference / A.Yu. Gordonov et al .: Edited by A.Yu. Gordonov and Yu.N. Dyakonov. - M .: Radio and communications. 1990, - 288 p.). The first, second and third computers 5, 8 and 11 can be implemented on microprocessors with sufficient speed (see Shevkoplyas B.V. Microprocessor structures. Engineering solutions: Handbook. 2nd ed., Revised and additional - M .: Radio and communications. 1990, - 512 p.). The decision block 15 is a comparison block, which can be implemented on the elementary logic of TTL chips.

Claims (1)

A method for recognizing low-speed speech coding (NSCR) protocols implemented in vocoders, which consists in receiving a digital information stream Y over a time interval ΔT, based on the received stream Y, they generate a normalized autocorrelation function A, according to regular extrema of the autocorrelation function at regular intervals Δτ And they decide on the presence of a block structure in the digital information stream Y, according to the intervals between the extrema of the autocorrelation function A, they divide the digital information flow to Y on information blocks in volume N _b bits each, sequentially assigned information units sequence numbers k = 1, 2, ..., K, from the first information unit is formed rectangular information matrix Y _{K × L, L} = N _b, which rows are sequentially placed one under the other information blocks in accordance with their serial numbers k = 1, 2, ..., K, characterized in that the columns y _l are sequentially distinguished from the matrix Y _{K × L} with numbers l = 1, 2, ..., L, for each column y _l information matrix Y _{K × L} calculate the value of mathematical about the expectation m _{l of the} appearance of certain pulses, form a vector of measured values of the mathematical expectation m (0) = {m ₁ , m ₂ , ..., m _l , ..., m _L } by sequentially placing the obtained values of the mathematical expectation m _l in accordance with their serial numbers l, based on the generated vector of mathematical expectation values m (0) by successive circular shift of its values by the value l = 0, 1, 2, ..., L-1, a set of M vectors m (l) of mathematical expectation values M = {m ( 0), m (1), ..., m (l), ..., m (L-1)}, form reference vectors of mathematical values personal expectation

, j = 1, 2, ..., J, for each digital information stream

corresponding to the j-th known NSCR protocol, each vector of mathematical expectation values m (l) of the evaluated NSCR protocol is sequentially compared with the reference vectors of mathematical expectation values

, j = 1, 2, ..., J, calculate the probability of correct recognition Pj (l) of the j-th NSCR protocol for each l-th vector of mathematical expectation values m (l), decide in favor of the j-th NSCR protocol, for which provides the maximum value of the probability of correct recognition of P _j (l).

Images