RU2701465C1 - Method of selecting digital streams - Google Patents

Abstract

FIELD: radio engineering and communications.

SUBSTANCE: invention relates to radio engineering, particularly to radio networks for transmitting data and high-frequency voice messages. Technical result is achieved by additional determination of linear relationships between elements of truncated sampling of autocorrelation function r_M={r₀, r₁, r₂, …,r_M-1}, M<N, N is number of bits in DS y, based on calculating mean-square error (MSE) E_m linear prediction (LP) and values of linear prediction coefficients (CLP) {a_m}_M, m = 1, 2, … M, where M determines order LP, and decision on used in DS form of communication is received on a set of events: exceeding value MSE LP E_M threshold value E_thr1=0.2 and presence in set CLP {a_m}_M global minimum with value

, serial number m of which coincides with one of the values of volume N_b, N_b=M-1, packet DS, which corresponds to receiving a voice message. Otherwise, the decision to receive text or an image is received based on threshold processing results with E_thr2=0.1. Fulfillment of condition E_thr2>0.1 corresponds to receiving an image, otherwise – to receive text.

EFFECT: technical result consists in improvement of probability of correct selection DS, under conditions of a priori uncertainty about their parameters and structure, shape of control frame.

1 cl, 6 dwg, 1 tbl

Description

The invention relates to the field of radio engineering, in particular, to radio networks for transmitting data and voice messages in the high-frequency range (HF) and can be used to analyze and isolate (select) digital streams generated by low-speed speech encoders, to determine the fact of changing the type of communication by a radio station.

The claimed technical solution expands the capabilities of similar means due to the higher accuracy of the selection of digital streams generated by low-speed speech encoders, based on the analysis of linear relationships between the elements (bits) of the studied digital stream (CPU) channel level (KU), formed by the radio station vocoder when transmitting voice messages.

Known methods for selecting digital streams proposed in Pat. RF №2480932, IPC H04L 25/03, publ. 04/27/2013, bull. No. 13. Analogs suggest the simultaneous reception of a CPU by a set of decoders, determining, according to a given criterion, the number of the decoder that successfully received the CPU, generating an output information signal based on the use of the signal at the output of the most successful decoder.

As disadvantages of analogues, the following should be noted:

high hardware redundancy;

require a priori knowledge of the characteristics of all speech encoders used in the CPU.

- n-й коэффициент ковариации, D(y)=М[у²]-(М[у])² - дисперсия ЦП у, М[.] - стандартная функция вычисления математического ожидания,

- центрированный вектор,

- центрированный вектор, сдвинутый на n=0, 1, 2, …, N-1 значений относительно

, по наличию регулярных с равными интервалами Δn экстремумов АКФ r принимают решение о наличии блочной битовой структуры в ЦП у, что соответствует наличию в ЦП речевого сообщения, подвергнутого низкоскоростному кодированию.Closest to the claimed one is a method for selecting digital streams (see Timofeev D.I., Tavalinsky D.A., Chubaty D.N. Analysis of parameters of low-speed speech encoders under structural and parametric uncertainty // High-tech technologies, No. 8,2011. - S. 4-9.). It consists in the fact that during the time interval ΔT, the digital information stream y = (y ₁ , y ₂ , ..., y _n , ..., y _N ) is received, where y _n is the nth bit of the CPU y containing N bits, based on the CPU y form the normalized autocorrelation function (ACF) r = {r ₀ , r ₁ , r ₂ , ..., r _N-1 }, where r _n = c _n / D _n - n ^_th correlation coefficient,

is the nth covariance coefficient, D (y) = M [y ² ] - (M [y]) ² - the variance of the CPU y, M [.] is the standard function for calculating the mathematical expectation,

- centered vector

- the centered vector shifted by n = 0, 1, 2, ..., N-1 values relative to

, by the presence of regular at equal intervals Δn extrema of the ACF r, they decide on the presence of a block bit structure in the CPU y, which corresponds to the presence in the CPU of a voice message subjected to low-speed encoding.

The prototype provides a sufficiently high accuracy of the selection of digital information flows KU with the presence of a block structure based on the autocorrelation method, including in the absence of a control frame, in the considered communication protocols. However, the CPU has a block structure not only when exchanging voice messages, but also when transmitting data. The autocorrelation function of the CPU containing the text message (Fig. 1a) has significant similarities with the ACF of the CPU of the voice message (Fig. 1b). As a result, the disadvantage of the prototype is manifested through a decrease in accuracy characteristics with an increase in the share of digital streams with block structure containing data. Ultimately, this does not allow to correctly determine the type of communication used by the radio station and to allocate digital streams generated by low-speed speech encoders (NKR).

In addition, when using the prototype method, the studied digital streams with a block structure containing data are falsely defined as implementations formed on the basis of new NKR protocols, which reduces the likelihood of correct recognition of NKR protocols.

The aim of the invention is to develop a method for selecting digital streams, which increases the likelihood of correct selection of CPUs, under conditions of a priori uncertainty about their parameters and structure, shape of the control frame, by performing additional analysis of linear relationships between elements of one CPU implementation.

и наличию в наборе КЛП {a _m}_М глобального минимума

, порядковый номер m которого совпадает с одним из значений объема N_б, N_б=М-1, пакета ЦП, что соответствует приему речевого сообщения, в противном случае решение о приеме текста или изображения принимают по результатам пороговой обработки с

где значение Е_М>0,1 соответствует приему изображения, иначе - приему текста.This goal is achieved by the fact that in the known method for the selection of digital streams, including the reception during the time interval ΔT of the digital information stream y = (y _1; y ₂ , ..., y _n , ..., y _N ), where y _n is the nth bit of a digital stream y containing N bits, the formation on the basis of the CPU of the normalized autocorrelation function r = {r ₀ , r ₁ , r ₂ , ..., r _N-1 } by the presence of regular extrema of the function r with intervals between them equal to Δn, making a decision about the presence of a block bit structure in the CPU y, additionally determine the linear relationships between elements truncated th sample ACF r _M = values {r _0, r _1, r _2, ..., r _M-1}, M <N, based on calculating the mean square error (MSE) E _M linear prediction (LP) and the values of linear prediction coefficients (LPC) { a _m } _M , m = 1, 2, ... M, where M determines the linear prediction order, and the decision on the type of communication used in the CPU is made according to the totality of events: exceeding the value of the RMSE L _M E of a given threshold value

and the presence in the set of CLP { a _m } _{M of a} global minimum

, the serial number m of which coincides with one of the values of the volume N _b , N _b = M-1, of the CPU packet, which corresponds to the receipt of a voice message, otherwise the decision to receive text or image is made according to the results of threshold processing with

where the value of E _M > 0.1 corresponds to the reception of the image, otherwise - the reception of the text.

, определения факта наличия глобального минимума

, совпадающего с одним из известных значений m объема N_б пакета, характерных для цифровых потоков, которые содержат речевые сообщения. Кроме того, для ЦП, имеющего блочную структуру, но не содержащего речевое сообщение, сравнение значения Е_М с пороговым уровнем

позволяет определить наличие в ЦП изображений или текста.Thanks to the new set of essential features in the claimed method, an increase in the probability of correct selection of the CPU containing the encoded voice messages is achieved by calculating the mean-square deviation of the meander PL E _M and CLP values { a _m } (m = 1, 2, ..., M), comparing the mean-square deviation E _M LP with a threshold value

determining the existence of a global minimum

coinciding with one of the known values m of volume N _{b of the} packet, characteristic of digital streams that contain voice messages. In addition, for a CPU having a block structure but not containing a voice message, comparing the value of E _M with a threshold level

allows you to determine the presence of images or text in the CPU.

The claimed method is illustrated by drawings, which show:

in FIG. 1 - autocorrelation functions generated by digital streams that contain text (1a) and voice message (1b);

in FIG. 2 - KLP values { a _m } for digital streams containing speech and formed by vocoders of types LPC-10-2400 (2a) and MELPe-600 (2b);

in FIG. 3 - values { a _m } for digital streams containing text (3a) and image (3b);

in FIG. 4 - generalized (for L> 1) algorithm for selecting digital streams;

in FIG. 5 is a digital stream selection algorithm in accordance with the claimed invention;

in FIG. 6 - dependence of the gain ΔТ in accuracy on the proportion of false (incorrectly selected) CPU b determining the probability of false alarm.

The main direction of development of the radio technology of the HF range is the transition to discrete information transmission systems. Replacing obsolete analogue radio communications has led to the widespread use of digital radio stations, including in the high frequency range (see Aladinsky V.A., Kuzminsky S.V. Analysis of digital streams at the vocoder outputs used on foreign high-frequency radio lines // Successes modern radio electronics, No. 7, 2015. P. 71-74), with built-in or external (plug-in) vocoders developed on the basis of LP methods. At the same time, the same radio standards with the same parameters of radio signals are used to transmit messages from the NKR or data.

This fact hinders the selection of digital streams CG containing heterogeneous messages based on the comparison of parameters of the vocoder as speed in _{a wok} and the number N _b of bits in one CPU unit values which are set at the design stage of a particular product (ibid.), And can match the values of similar parameters of digital streams of various types of communication. In addition, the transmitted images and data often have a periodic structure, which leads to the appearance of errors in the selection of digital streams based on a comparison of the values of the ACF parameters r.

Based on the principle of similarity, an increase in the quality of selection of digital streams generated by vocoders with LP can be achieved by analyzing the ACF of the received CPU based on the study of such parameters of the LP as RMSE and CLP. In general terms, the value of the standard deviation of a medication of order M is determined by a well-known formula (see Wai Chu, S. Speech coding algorithms: Foundation and evolution of standardized coders. Hoboken: Wiley-Interscience. - 2003. - 558 p.):

where a _m is the m-th coefficient of the drug, m = 1, 2, ..., M.

LPC values are calculated based on a matrix equation of the form (see ibid.)

Where

t is the transpose operation.

according to various algorithms. The Levinson-Darbin iterative algorithm (see ibid.) Is considered effective for minimizing computational costs, for which the initial data are a truncated sample of ACF values {r ₀ , r ₁ , r ₂ , ..., r _m , ..., r _M-1 } and the initial value of the standard deviation is E ₀ = r ₀ . The number of iteration cycles is set by the value M. For m = 1 we have: a ₁ = -r ₁ / E ₀ , E ₁ = E ₀ (1 - a ₁ ). The values of the mth CLP and the corresponding standard deviation are calculated by the following formulas (see ibid.)

- значения КЛП, вычисленные на предыдущих i=1, 2, …, m-1 циклах.Where

- KLP values calculated on the previous i = 1, 2, ..., m-1 cycles.

The sequence of actions proposed by Levinson and Darbin was implemented on a computer in the programming environment MATLAB. In this case, the input digital streams contained voice messages generated by model vocoders of the types LPC-10-2400 (N _b = 54) and MELPe-600 (N _b = 60), which are actively used in high-frequency radio stations (see Aladinsky V.A. , Kuzminsky S.V. Analysis of digital streams at the vocoder outputs used on foreign radio lines of the high frequency range // Uspekhi modern radioelectronics, No. 7, 2015. P. 71-74), as well as text messages and images. Based on the properties of the ACF r, in order to reduce the number of iterations in the Levinson-Darbin algorithm, it was assumed that M = N _b +1. As a rule, the value of N _{b is a} priori unknown for each input implementation, therefore, during the experiment, in addition to the values of N _b = 54 and 60, the maximum possible values of N _b = 81 and 96 were also selected.

и усреднены следующим образом (см. Математический энциклопедический словарь. / Гл. ред. Прохоров Ю.В. - М.: Большая российская энциклопедия, 1995. Репринтное издание - М.: Сов. энциклопедия, 1988. - 847 с.):For each of the three types of messages, L = 200 digital KU streams were generated. For digital streams of one type of message according to formula (1), taking into account formulas (3) and (4), the values were calculated

and averaged as follows (see the Mathematical Encyclopedic Dictionary. / Ch. ed. Prokhorov Yu.V. - M .: Big Russian Encyclopedia, 1995. Reprint edition - M: Sov. Encyclopedia, 1988. - 847 p.):

The final results of the experiment are shown in table 1.

существенно больше в сравнении с цифровыми потоками, содержащими тексты или изображения. На фиг. 1 приведены результаты расчета значений АКФ по цифровым потокам, содержащим текст (фиг. 1а) или речевое сообщение (фиг. 1б).From the analysis of table 1 it follows that for digital streams containing voice messages, the values

significantly more in comparison with digital streams containing texts or images. In FIG. 1 shows the results of calculating ACF values for digital streams containing text (Fig. 1a) or voice message (Fig. 1b).

для m∈{N_б}. При этом максимальное значение

из совокупности {N_б} априорно известных значений определяло величину

. На фиг. 2 приведены результаты расчета значений {a _m} при М=97 по цифровым потокам КУ, сформированным вокодерами типа LPC-10-2400 (фиг. 2а) и MELPe-600 (фиг. 2б). Здесь

имеет значения при m=54 и m=60 соответственно, m∈{N_б}. На фиг. 3 приведены аналогичные результаты для цифровых потоков КУ, содержащих тексты (фиг. 3а) и изображения (фиг. 3б), в которых отсутствуют глобальные минимумы.In the course of the experiment, it was found that the KLP vector a _M obtained by analyzing the ACF of the CPU with voice communication has a global minimum

for m∈ {N _b }. In this case, the maximum value

from the set {N _b } a priori known values determined the value

. In FIG. Figure 2 shows the results of calculating the values of { a _m } at M = 97 for digital KU streams formed by vocoders of the LPC-10-2400 type (Fig. 2a) and MELPe-600 (Fig. 2b). Here

matters at m = 54 and m = 60 respectively, m∈ {N _b }. In FIG. Figure 3 shows similar results for digital streams KU containing texts (Fig. 3A) and images (Fig. 3b), in which there are no global minima.

(см. Таблицу 1); в наборе КЛП {a _m} имеется глобальный минимум

, для которого m∈{N_б}. Введение порога

позволяет селектировать между собой текстовые сообщения и изображения.From the foregoing it follows that the criterion for making a decision about detection, and, therefore, for performing the operation of selecting voice messages from the CPU, it is advisable to consider the following conditions: value

(see Table 1); in the set of KLP { a _m } there is a global minimum

for which m∈ {N _b }. Threshold setting

allows you to select text messages and images.

Based on the foregoing, it is proposed the following generalized (for L> 1) sequence of operations (see Fig. 4) in the analysis of digital streams KU:

количество реализаций ЦП L.1. Input data: time interval ΔT; values {N _b } of known vocoders; M = N _bmax +1;

the number of CPU implementations L.

2. Setting the initial variable

3. CPU reception

4. Calculation of ACF

5. Check for block structure in the CPU. If the result is positive, go to step 6. If the result is negative, go to step 19.

размерности М, М<N6. Formation of a truncated sample of ACF values

dimension M, M <N

7. Finding by formula (3), taking into account (4) the components a _m of the CLP vector a _M.

по формуле (1).8. The calculation of the value

by the formula (1).

При выполнении условия переход на этап №11, иначе на этапе №10.9. Check for completion of the cycle

If the condition is met, go to step 11, otherwise at step 10.

10. Variable change

в соответствии с (5).11. Determination of the average value

in accordance with (5).

Если условие выполняется, то осуществляется переход на этап №13, иначе - переход на этап №15.12. Verification of the condition:

If the condition is met, then go to step 13, otherwise - go to step 15.

в наборе а _М.13. The search for a global minimum

in a set and _M.

. Если для найденного

значение m∈{N_б}, то осуществляется переход на этап №16, иначе - переход на этап №15.14. Verification of the condition for

. If for found

value m∈ {N _b }, then the process proceeds to step 16, otherwise, the transition to step 15.

Если условие выполняется, то осуществляется переход на этап №15, иначе - на этап №18.15. Verification of the condition:

If the condition is met, then the transition to step number 15, otherwise - to step number 18.

16. Formation of the result: "The CPU contains a voice message."

17. Formation of the result: "The CPU contains the image."

18. Formation of the result "the CPU contains the text."

19. Formation of the result "the CPU does not contain the specified information."

20. Information output.

Selection of digital streams, as a rule, assumes L = 1. In this regard, it is proposed (see Fig. 5) the following sequence of operations corresponding to the claimed method:

;

1. Input data: time interval ΔТ; values {N _b } of known vocoders; M = N _{b max} +1;

;

2. CPU reception y = (y ₁ , y ₂ , ..., y _n , ..., y _N ).

3. The calculation of the ACF r = {r ₀ , r ₁ , r ₂ , ..., r _N-1 }.

4. Check for block structure in the CPU. If the result is negative, go to step 15.

5. Formation of a truncated sample of ACF values r _M = {r ₀ , r ₁ , r _2, ..., r _m , ..., r _M-1 } of dimension M, M <N.

6. Finding by formulas (2), (3) taking into account (4) the components a _m of the CLP vector a _M.

7. The calculation of the value of E _M according to the formula (1).

Если условие выполняется, то осуществляется переход на этап №9, иначе - переход на этап №11.8. Verification of the condition:

If the condition is met, then the transition to step number 9, otherwise - the transition to step number 11.

в наборе а _М.9. Search for a global minimum

in a set and _M.

. Если для найденного

значение m∈{N_б}, то осуществляется переход на этап №12, иначе - переход на этап №11.10. Verification of the condition for

. If for found

value m∈ {N _b }, then the transition to stage 12 is carried out, otherwise - the transition to stage 11.

11. Verification of the condition E _M > 0.1. If the condition is met, then the transition to step No. 13, otherwise, the transition to step No. 14.

12. Formation of the result: "The CPU contains a voice message."

13. Formation of the result: "The CPU contains the image."

14. Formation of the result "CPU contains text."

15. The formation of the result "the CPU does not contain the specified information."

16. Information output.

Thus, when analyzing the ACF of a CPU based on an LP using the Levinson-Darbin algorithm, the values of the CLP and RMSE are calculated. Experimental verification by computer simulation showed that for a CPU containing a voice message, the LPC vector has a global minimum, and the magnitude of the standard deviation of the LPC is significantly larger than for other types of messages (texts, images). Text messages and images transmitted over a digital channel, also have a contrasting sign of _EM, which allows them to identify.

The accuracy of the selection of digital streams is determined by the expression (see Manning K.D., Raghavan P., Schütze H. Introduction to the information search. - M.: LLC “ID Williams, 2011. - 528 p.)

where ν is the fraction of digital streams correctly selected during their selection (containing voice messages, i.e. true); b - the proportion of digital streams that were incorrectly selected (not containing voice messages, i.e. false and determining the likelihood of a false alarm) for some reason; (ν + b) is the total number of digital streams subjected to selection.

Based on the results of the selection of five thousand real and model digital streams generated by vocoders, the following was established. The proposed method provides a value of ν no worse than the prototype method, but ceteris paribus eliminates at least 90% of improperly selected digital streams from their total number (ν + b). Thus, the probability of false alarm is reduced to a value of 0.1. The proportion of improperly selected digital streams when using the proposed method is not more than 0.1b with a constant value of ν. The difference between the obtained value of the accuracy of selection T ₁

and T, determined according to (5) during selection based on the prototype method, will make a gain in the accuracy of selection of digital streams containing voice messages:

Assuming that (ν + b) = 1 for ν, b∈ [0, 1], we substitute ν = 1-b in (8) and, after some reductions, we obtain

From (9) it follows that the value of the gain ΔТ in the accuracy of selection of digital streams containing voice messages is determined by the value of the fraction b∈ [0, 1]. In real conditions, the proportion of digital streams with a block structure containing data does not exceed 10%, which corresponds to b≤0.1. In FIG. Figure 6 shows the dependence of the gain ΔТ in the selection accuracy on the value of b calculated by formula (9). The performed analysis shows that the maximum gain in selection accuracy for b = 0.1 and other conditions being equal will be ΔТ = 0.089 or 8.9%.

Claims (1)

The method of selecting digital streams, which consists in the fact that during the time interval ΔT the digital information stream y = (y ₁ , y ₂ , ..., y _n , ..., y _N ) is received, where y _n is the nth bit of the digital stream ( CPU) y containing N bits, based on the CPU y form a normalized autocorrelation function (ACF) r = {r ₀ , r ₁ , r ₂ , ..., r _N-1 }, by the presence of regular extrema of the function r with intervals between them, equal to Δn, they decide on the presence of a block bit structure in the CPU у, characterized in that they additionally determine the linear relationships between the elements of the truncated sample and ACF r _M = {r ₀ values, r _1, r _2, ..., r _M-1), M <N, based on calculating the mean square error (MSE) E _M linear prediction (LP) and the values of linear prediction coefficients ( KLP) {a _m } _M , m = 1, 2, ... M, where M determines the order of linear prediction, and the decision on the type of communication used in the CPU is made according to the totality of events: exceeding the _RMSE value of the LP E _M threshold value E _por1 = 0, 2 and the presence in the set of KLP {a _m } _{M of a} global minimum with the value

, the serial number m of which coincides with one of the values of the volume N _b , N _b = M-1, of the CPU packet, which corresponds to the reception of a voice message, otherwise, the decision to receive text or image is made according to the results of threshold processing with E _por2 = 0, 1, where the value of E _M > 0.1 corresponds to the reception of the image, otherwise - the reception of the text.

Images