RU2343563C1 - Way of transfer and reception of coded voice signals - Google Patents

Abstract

FIELD: physics, communication.

SUBSTANCE: invention relates to conversion and transfer of voice signals and can be used for transfer of coded speech over voice-frequency channels. For this, relative quantity of transmitted amplitudes and phases is redistributed so as to reduce an amplitude share and allow reducing the quantity of transmitted signals and, at a time, maintaining a qualitative restoration of speech. Note also that excessive noise-immune encoding of confidential amplitudes is spaced over various harmonics in transmission and aforesaid amplitudes are noise-immune decoded in reception to increase the level of transmitted signals and to reduce duplication of errors in sequences of amplitudes. On the transmitting side, the voice signal is, first, subjected to quadrature modulation and, prior to its encoding, it is subjected to uniform decimation over, at least, two voice-signal amplitude count-downs. Now, a reference amplitude of the magnitude subject to encoding is added to every encoded amplitude, both amplitude magnitudes being transmitted into the channel by two harmonics as a part of total signal shaped allowing for all amplitudes and encoded phases. On the reception side, the levels of every pair of received harmonics are corrected by comparing the sum of encoded and reference amplitudes with the fixed-threshold. Also, harmonic amplitudes are normalised and target amplitude corrected magnitude is computed. After decoding amplitudes and phases and prior to forming low-frequency quadrature components and to voice-signal time representation, decoded amplitudes are computed by interpolation between adjacent count-downs and computed count-downs are between aforesaid amplitudes.

EFFECT: higher noise immunity of transmission and quality of voice signals without expansion of channel signal range.

6 dwg, 1 tbl

Description

The invention relates to the conversion and transmission of speech signals. Its use for transmitting encoded speech over tonal frequency channels allows providing a technical result in the form of increased noise immunity and quality of speech signals.

Known methods for secreting a speech signal in a secured manner are based on converting it into a digital form, followed by applying a pseudo-random sequence generated on the receiving and transmitting sides by synchronized generators using the same cryptographic key unknown to unauthorized persons [1].

The bandwidth of the communication channel necessary for the transmission of the secret speech signal in binary form, depending on the speech quality requirements, can be tens of kbit / s [2, pp. 112-113] and [3].

To transmit digital classified speech signals over tonal frequency (PM) channels with a bandwidth of 300-3400 Hz, it is necessary to reduce the transmission speed by compressing speech signals, for example, by using linear prediction coding and the use of corresponding modems with digital information inputs / outputs.

These methods are characterized by a sharp decrease in the intelligibility of restored speech due to propagation of errors, up to the loss of communication, with a deterioration in the quality of the communication channel below a threshold value, which is especially noticeable on short-wave radio channels with multipath signal propagation. Moreover, in such PM channels, the transmission speed of classified speech signals is practically no higher than 1200-2400 bit / C, which requires a greater degree of compression of speech signals compared with the transmission of speech over wire channels.

This drawback is less characteristic of methods that do not use or use to a small extent compression of the speech signal before transmitting it to the communication channel. This class of technical solutions includes the methods described in US patent No. 4179586 [4] and RF patent No. 2221284 [5]. The last of them is closest to the proposed method and is therefore selected as a prototype.

Voice transmission by a known method is carried out after transmission of a preamble to the communication channel containing the necessary information for correcting the working signals at the receiving side and synchronizing the encoders. On the transmitting side, the transmitted analog speech signal is converted to digital form by the method of analog-to-digital conversion (ADC). The digitized speech signal is subjected to bandpass filtering with the release of low-frequency quadrature components. Two time sequences are formed from these components from samples of amplitudes and phases with the number M of possible phase values equal to the power of number 2. Then, by thinning the samples in the indicated sequences, the repetition rate of the samples is brought to a value equal to the total spectrum width of the quadrature components.

Using the compression operation, the amplitude levels are reduced to values lower than the M number by the value of the double guard interval, designed to reduce the multiplication of errors during decryption. The values of the amplitudes with the added guard interval and the phase values are classified modulo M. Then, differential encoding of the secret values of the amplitudes and phases is carried out.

To do this, the samples of classified amplitudes and phases are divided separately into packets of N samples in each packet, each previous packet is stored, and coding is carried out by adding modulo M the corresponding samples from the current and previous packets in order of sequence.

Then, the secret values of amplitudes and phases encoded in this way are converted using the inverse Fourier transform from the frequency domain to the time domain in the form of a sum of N harmonics placed in the passband of the PM communication channel. A protective time interval coinciding with the initial part of this stream is connected to the time stream of samples, thereby forming a frame group signal with a duration T for transmission to the communication channel, in particular, using digital-to-analog conversion (DAC).

On the receiving side, the analog signals received from the channel are converted into a sequence of digital signals. This sequence, in turn, is converted from the time domain to the frequency domain. For this, a direct Fourier transform is used and the real and imaginary parts of the received digital signal are calculated. Then, the amplitudes and phases of the harmonics of the received signal are determined from these parts, differential decoding of the amplitudes and phases of the signals, and declassification of the phases and amplitudes is carried out with a decrease in the value of the latter by the value of the guard interval. After that, the amplitudes are expanded and the low-frequency quadrature components of the declassified signal are calculated by multiplying the amplitudes by the cosine and sine of the corresponding phases.

Then the signals of the quadrature components are subjected to low-frequency interpolation and the spectrum of the interpolated signals is transferred up to a frequency equal to the average frequency of the frequency band of the speech signal in transmission. By adding the interpolated signals from the quadrature components transferred over the spectrum, a single time stream of the digital speech signal is formed and converted to an analog form before being output to the telephone using digital-to-analog conversion. The prototype method provides the transmission of a speech signal with a bandwidth equal to N / T hertz at T values expressed in seconds.

The disadvantage of the prototype method is to significantly reduce the noise immunity of reception during fast fading on channels with multipath propagation of signals. Under these conditions, the effect of error propagation inherent in differential (differential) encoding-decoding, especially in the sequence of amplitudes carrying the energy component of the speech signal, begins to noticeably affect. In fact, the amplitude coding method used in the prototype well compensates for distortions during slow changes in the amplitude characteristics of the communication channel, which are characterized by a small difference in the magnitude of the distortions of the amplitudes in adjacent frames, while the level of transmitted amplitudes is restored with small errors. With the rapid nature of amplitude distortions, when the difference in the magnitude of the distortions of the amplitudes in adjacent frames becomes large, the effect of error propagation when decoding the amplitudes is manifested, which is further amplified due to anomalous errors arising from the declassification of distorted amplitudes with values close to zero or the modulus encryption M.

The objective of the proposed method is to increase the noise immunity and transmission quality of classified speech signals while maintaining a given bandwidth of the transmission channel.

The technical result achieved by the proposed method is as follows.

Firstly, in reducing the number of transmitted samples of classified signals by thinning out samples of speech amplitudes and the associated reduction in the number of harmonics used to transmit these signals, while maintaining the quality of speech. This is ensured by a change in the transmitted sequence of samples of the ratio between the number of samples of amplitudes and the number of samples of phases and the restoration of the reception of equality in this ratio by interpolating decimated samples of amplitudes.

Secondly, in providing noise-resistant coding of samples of all classified harmonics amplitudes with the addition of control amplitude samples to each classified harmonic and transmitting them instead of freed harmonics as part of a common group signal, as well as noise-free decoding without multiplying errors of each received pair of samples of classified and control harmonics amplitudes .

To achieve this, on the transmitting side, after calculating the amplitudes and phases of the speech signal and before classifying them, uniform thinning of the samples of the amplitudes of the speech signal is carried out at least twice. Then, for each sample of the classified harmonic amplitude, a control amplitude sample is formed with a value equal to the addition of the value of the secret amplitude to the classification module M, and both amplitudes are increased by the guard interval.

After that, the conversion of classified and reference amplitudes of harmonics with the same number of differential classified phases from the frequency domain to the time domain is carried out using the inverse Fourier transform. As a result of this, a group signal is formed in the form of the sum of all harmonics with amplitudes corresponding to the values of classified and control amplitudes, and with initial phases corresponding to classified phases, prepared for transmission to the communication channel.

On the receiving side, after calculating the amplitude levels and phase values of the harmonics of the received signal, the amplitudes of each pair of classified and control harmonics are decoded with the formation of one final amplitude. Decoding consists of adjusting the levels of the original amplitudes using a correction factor, which is calculated by comparing the sum of these amplitudes with a fixed threshold. The value of the final amplitude is calculated as half the difference of the corrected amplitudes of the classified and control harmonics increased by the value of the module M.

After declassification of the samples of amplitudes and phases restored by decoding, and before the formation of samples of low-frequency quadrature components and the temporary presentation of the speech signal, the order of phases and amplitudes is restored. Then, the values of the amplified samples thinned out by the transmission are calculated by interpolating them from the neighboring samples of the sequence of declassified amplitudes and inserting the calculated samples between them into the sequence of interpolated samples.

In the proposed method, firstly, there is no propagation of errors in the amplitudes from one frame to another due to the lack of communication between the classified amplitudes in adjacent time frames inherent in the prototype method.

Secondly, the value of each classified amplitude is transmitted to the channel by two different samples and restored at the reception using samples of two received harmonics, the main and control, and not one harmonic, as in the prototype, which actually increases the received signal level by 2 times. It also provides compensation for changes in amplitudes during slow and fast group fading of signals.

In addition to these positive effects, the proposed method is also characterized by an improvement in comparison with the prototype of the characteristics of the group signal transmitted to the channel in terms of reducing its dynamic range and peak factor. This is due to the constant level of the sums of the amplitudes of each pair of harmonics conjugated in coding, equal to the value of M-1, increased by a double guard interval, which is inherent in the introduced method of encoding harmonics of amplitudes. While in the prototype the classified values of the amplitudes of all N harmonics are independent of each other and the sum of the amplitudes of any two harmonics can have up to 2M different values. The decrease in the dynamic range reduces the level of nonlinear distortion in the group signal due to the possible trimming of the peak bursts of signals in the transmission path, especially in radio transmitters, and also creates more stable conditions for the operation of nodes for automatic level control at the reception, which helps to increase the noise immunity of the transmission of classified speech signals .

The considered method of increasing the noise immunity of the transmission of classified speech signals while maintaining a given bandwidth of channel signals involves reducing the number of transmitted samples of speech signals, which would reduce the intelligibility and quality of restored speech reception.

However, the practically proposed method for reducing and restoring samples of speech signals only in the sequence of amplitudes without exception of phase samples does not, under certain conditions, reduce the intelligibility and quality of the restored speech in comparison with the prototype.

This is because the high-frequency rapidly changing components of the speech signal are concentrated mainly in the phase sequence, the number of samples in which according to the proposed method remains unchanged or even increases depending on the selected mode of decimation of amplitudes. At the same time, the sequence of amplitudes, which are the amplitude envelope of the input signal, changes much more slowly [6, p. 55-59]. Therefore, it can be thinned to a certain extent without loss of quality of its restoration at the reception by the method of interpolation from neighboring samples.

If we choose the degree of thinning slightly more than two (3 or 4), then in this case the decrease in the accuracy of restoration of amplitudes for speech quality will be largely compensated (and even with excess losses) due to the possibility of transmitting in each frame a larger number of phase readings than in prototype method. This option, by increasing the frequency of the phase samples, provides an extension of the frequency band of the transmitted speech signal, which leads to an improvement in the quality of the reconstructed speech compared to decimation of every second sample.

Show an existing relationship between the number L _s transmitted in the frame classified amplitudes number L classified phases _phases, the number of harmonics N channel and the value K decimation factor which must be considered when implementing the subject method. The determining condition for this dependence is the obvious assertion that the total number L of secret amplitudes and phases (L = L _am + L _phases ) in the frame should not exceed the value N + N / 2.

Indeed, in the presence of N channel harmonics, it is possible to transmit in each frame using only the harmonics amplitudes no more than N / 2 classified samples, since the amplitudes of the remaining harmonics will be occupied by the samples of the amplitudes of the reference harmonics calculated from the indicated N / 2 classified samples. The remaining N classified samples from the total number L can be transmitted in the form of the initial phases of all N harmonics with amplitudes corresponding to both classified and control amplitudes.

Since the number of phases L _phases in the frame prior to classification is equal to the number of non-decimated amplitudes, the number of decimated amplitudes issued for classification will be L _am = L _phases / K. Based on the foregoing, we obtain the following expression (L _phases + L _phases / K) ≤ (N + N / 2), which determines the relationship between the values of the considered parameters. It follows from this that the number L of _phases must satisfy the condition

Using relation (1), we determine that for values of K = 2, 3, 4, and 5, the number of L _phases will be 1.0N, respectively; 1.125N; 1.20N; 1.25N.

Since at K> 2 the permissible number of phases L _phases in the frame is greater than the number N, while in the prototype L _phases = N, with the indicated values of K it will be possible to form a larger number of phases and unshrunked amplitudes in the time interval T compared with the prototype . This makes it possible to increase the sampling frequency of samples in sequences of amplitudes and phases and, thereby, expand the transmitted frequency band W _{pc of the} speech signal from W _pc = N / T to W _pc = L _phases / T. In absolute terms, this is equivalent to expanding the frequency band of the speech signal at K> 2 by several hundred hertz and an additional increase in speech intelligibility. The most effective application of the proposed method of transmission and reception with the values of the coefficients of thinning is not higher than 3 or 4. With more frequent thinning, the effectiveness of the method decreases due to the deterioration of the recoverability of the thinned amplitudes.

The proposed method contains known from the description of the prototype and scientific literature of the operation. Such operations include, for example, operations of bandpass filtering of signals by decomposing them into low-frequency quadrature components with the formation of amplitudes and phases, operations of decimation (decimation) and interpolation of signals, multi-frequency transmission of signals by orthogonal harmonics, difference coding and decoding of signals, and others. However, their combination in the proposed method is new.

A new combination of known operations allowed, firstly, by reducing the relative proportion of amplitudes and increasing the proportion of phases without reducing the quality of speech, redistribute and reduce the total number of transmitted samples of amplitudes and phases of open speech signals. Secondly, due to the harmonics freed up in the group channel signal, it was possible to transmit noise-resistant coding of the secret amplitudes of the harmonics on the transmission, and at the reception, they were able to decode them without error propagation and qualitatively restore the amplitudes of the speech signal. This all made it possible to fulfill the task of increasing the noise immunity of transmission and the quality of speech signals without expanding the spectrum of channel signals.

In the proposed and known methods, common features are the transmission of encoded speech over a previously prepared transmission and reception path. This is ensured by the preliminary transmission and detection of a special preamble for correction on the receiving side of the useful signals and synchronization of the encoders.

At the same time, on the transmitting side, the speech signal is converted from an analog form to a digital form, its band-pass filtering is performed with the selection of sequences of samples of low-frequency quadrature components. Then, by decimation, the repetition rate of the samples in these sequences is brought to a value equal to the total spectrum width of the quadrature components, and the values of amplitudes and phases are calculated from the decimated samples of the quadrature components. In this case, the value of the number M of possible phase values is equal to the power of number 2. After this, by compressing the values of the amplitude levels are reduced to a value smaller than the number M, the guard interval is added to them, then the phase values and amplitude values are classified modulo M and the classified samples are distributed into two groups. The samples in the first of the groups determine the amplitudes of the orthogonal harmonics with frequencies in the passband of the communication channel. The samples in the second group, previously subjected to differential modular coding, specify the initial phases of these harmonics. The samples of the amplitudes and phases of N harmonics are converted from the frequency domain to the time domain. To do this, use the inverse Fourier transform and after the conversion, a protective time interval is attached to the generated group signal, thereby forming a frame time group signal with a duration T, which is transmitted to the communication channel.

On the receiving side, the frame signals received from the channel are in turn converted from the time domain to the frequency domain by using the direct Fourier transform, the output signals of which determine the amplitudes and phases of the orthogonal harmonics of the received signal. Then, differential decoding of the harmonic phase values is carried out and the sequence of samples of amplitudes and phases is declassified modulo M, declassified amplitudes are reduced by the value of the guard interval. After that, the amplitudes are expanded and the low-frequency quadrature components of the declassified signal are calculated by multiplying the amplitudes by the cosine and sine of the corresponding phases. Then the signals of the quadrature components are subjected to low-frequency interpolation and the spectrum of the interpolated signals is transferred up to a frequency equal to the average frequency of the frequency band of the speech signal in transmission. By combining the interpolated signals of the quadrature components, a single time stream of the digital speech signal is formed before it is converted to analog form.

The features of the proposed method, different from the features of the prototype method, are the following features.

On the transmitting side, after the operation of compressing the amplitudes and before the start of the secret operation, uniformly thinning the samples in the time sequence of amplitudes is performed at least twice. After that, the remaining samples of the amplitudes and the samples of the phases are combined in one time sequence in a certain order and so that their total number does not exceed the value N + N / 2 for time T. Then, the classified samples are classified and distributed, classified in groups with the inclusion in the first group of N / 2 classified samples, which is supplemented by the same number of control amplitude samples. The value of each control reference is set equal to the addition to module M of the value of the corresponding classified reference of this group. After that, the output of all N classified and control samples from the first group, increased by the guard interval, and the same number of coded classified samples from the group of phases for their joint conversion from the frequency domain to the time domain are carried out.

On the receiving side, after calculating the amplitude levels and phase values of the harmonics of the received signal, each pair of classified and control amplitudes of the harmonics is decoded with the formation of one final amplitude. The decoding operation consists in adjusting the levels of the source amplitudes using the correction factor and calculating the value of the resulting amplitude. The correction factor is calculated by comparing the sum of the initial amplitudes with a fixed threshold. The value of the final amplitude is defined as half the difference of the corrected amplitudes of the classified and control harmonics increased by the value of the module M. After the decoded samples of amplitudes and phases are declassified, and before the formation of samples of low-frequency quadrature components, the sequence of phases and amplitudes is restored and the values of amplified samples sampled by transmission are interpolated by interpolating the sequence of received declassified amplitudes from adjacent samples and inserting the calculated values of samples between them into the sequence of interpolated samples .

On figa presents a structural diagram of a transmitting part of a system for transmitting and receiving encoded speech by the proposed method (implementation example), on figb and figv respectively examples of the implementation of the amplitude encoder and phase encoder.

On figa presents a structural diagram of the receiving part of the system for transmitting and receiving encoded speech by the proposed method (implementation example), on figb and figv examples of the implementation of the circuits of the amplitude decoder and, accordingly, the phase decoder.

The transmitting part contains blocks:

1 - microphone

2 - speech ADC,

3 - quadrature demodulator,

4 - decimator in-phase component,

5 - decimator quadrature component,

6 - calculator of amplitudes and phases,

7 - block thinning amplitudes,

8 - amplitude compressor,

9 - adder

10 - multiplexer,

11 - encoder,

12 - demultiplexer,

13 - amplitude encoder,

14 - phase encoder,

15 - block serial-parallel conversion (ON / PA) sequences of samples of amplitudes,

16 - block software / PA sequences of phase readings,

17 - block inverse fast Fourier transform (OBPF),

18 is a block of a temporary protective interval (VZI),

19 - channel DAC,

20 - channel output.

Amplitude encoder 13 contains adders 13.1 and 13.2. The phase encoder 14 comprises an adder 14.1 modulo M and a delay line 14.2.

The receiving part contains blocks:

21 - channel input

22 - channel ADC and demodulator,

23 - block direct fast Fourier transform (PBPF),

24 - calculator of amplitudes and phases,

25 - amplitude decoder,

26 - phase decoder,

27 - multiplexer,

28 - decoder,

29 - demultiplexer,

30 - amplitude interpolator,

31 - expander,

32 - calculator quadrature components,

33 - interpolator of quadrature components,

34 - quadrature modulator,

35 - speech DAC,

36 is the telephone.

Amplitude decoder 25 (Fig. 26) contains adders 25.1, 25.2 and 25.4, a multiplier 25.3, a divider 25.5. The phase decoder 26 (pigv) contains a delay line 26.1 and the adder 26.2 modulo M.

Most of the blocks indicated on the diagrams of examples of the implementation of the method in question coincide in functionality with similar blocks in the examples of the prototype method [2]. The exception is mainly only the newly introduced units of decimation, coding, decoding and interpolation of amplitudes (blocks 7, 13, 25, 30).

Some blocks in figa and 2a, for example, blocks 3 and 34 in order to simplify the description of the method are made in the form of combining in one block several blocks from the prototype in full compliance with their functional purpose. So in the transmitting part (figa), a quadrature demodulator 3 with two inputs (signal - x (t ₁ ) and frequency - f _c ) performs the functions of 4 blocks of the prototype transmitting part of the circuit, generating a sine and cosine of the middle frequency, forming quadrature components of the speech signal and their filtering. A similar combination in one block of the quadrature modulator 34 is carried out in the receiving part (figa) when implementing the inverse function of the above, i.e. the formation of a speech signal by low-frequency quadrature components.

The operation of the encryption system - decryption of speech by the proposed method is as follows. Before the transmission of speech to the receiving side, a synchronization preamble is transmitted. It consists of several frames of unmodulated and phase-modulated harmonics. They are designed to configure the automatic gain control system, compensate for the frequency offset between the transmitter and the receiver, possible adjust the corrector of the amplitude-frequency and phase characteristics of the transmission and reception path, as well as provide personnel phasing of the receiver and the encoder and decoder trigger signal consistent in time and content.

Details of the parameters of these signals are given in the prototype and can be refined based on specific requirements. The end result of these operations is the preliminary transmission and detection of the preamble for correction on the receiving side of the useful signals and synchronization of the encoders, which is not the subject of the invention, therefore, the nodes implementing the above operations are not shown in FIG. 1a and 2a, although their presence implied. The same applies to the blocks for generating sampling frequencies of analog signals and other clock and modulating frequencies used in signal conversion processes.

The values of the sampling frequency f _d1 in the ADC and DAC and the average frequency f _c for frequency modulation and demodulation depend on many factors. But the determining ones are the selected values of such channel parameters as the sampling frequency f _d2 channel signals, the number N and the values of the frequencies of the channel harmonics, as well as the duration of the T frame and the protective time interval. The value of these parameters determines the number L of secret samples of speech signals transmitted during one channel frame, and the corresponding bandwidth and decimation coefficient K of the transmitted bandwidth W _{pc of the} transmitted speech signal and the spectrum width of the channel signal W _k .

In table 1 below, for several possible implementations of the proposed method, the values of the main parameters and characteristics are given that complement the description of the schemes in the implementation examples.

Table 1. No. p.p. f _d1 Hz f _d2 Hz K _d N K Channel frames Speech Packing Spectrum Width (Hz) Counts 1 / T Hz L L _phases L _am W _pc W _k one 9600 8000 four 48 2 160 fifty 72 48 24 2400 3000 2 10800 8000 four 48 3 160 fifty 72 54 eighteen 2700 3000 3 10800 7200 four fifty 2 160 45 75 fifty 25 2250 2812.5 four 10800 7200 13/3 48 3 156 46.1 72 54 eighteen 2492 2700 5 10800 7200 four fifty four 160 45 75 60 fifteen 2700 2812.5

The transmitted speech signals from the output of the microphone 1 are sent to ADC 2, which converts continuous speech signals into a sequence of digital samples (samples) following equal intervals of 1 / f _d1 , measured in seconds, using analog-to-digital conversion with a sampling frequency f _d1 . From the output of the ADC 2, the sequence of samples x (t ₁ ) corresponding to time instants t ₁ = n / f _d1 , where n = 1, 2, 3, ... is the number of samples, enters the quadrature demodulator 3, which is similar in function to the combination of several prototype blocks. Demodulator 3 from the input stream forms two low-frequency quadrature components: in-phase x _I (t ₁ ) and quadrature x _Q (t ₁ ). For this, the well-known method [6] is used, which consists in multiplying the input signals by a complex exponent

and low-pass filtering resulting from the multiplication of signals. The value of the frequency f _c in expression (3) is equal to the value of the average frequency of the frequency band W _pc allocated for transmitting the speech signal. The bandwidth of the low-pass filters is taken equal to W _pc / 2.

At the output of the filters of the quadrature demodulator 3, two quadrature sequences x _I (t ₁ ) and x _Q (t ₁ ) are formed. The highest signal frequency in these sequences does not exceed W _pc / 2 Hz. Therefore, these signals can be completely reconstructed from decimated (decimated) samples made with a frequency of W _pc , which is K _d times less than the repetition rate of the input signals, i.e. on samples of each To _d- th reference. This corresponds to the sampling frequency f _d1 / K _d . The sampling rate is reduced with the help of decimators 4 and 5, at the output of which quadrature components x _I (t) and x _Q (t) are obtained, the readings in which correspond to time instants t = K _d t ₁ .

After that, the signals of the quadrature sequences x _I (t) and x _Q (t) are converted in the calculator 6 into samples of the amplitudes and phases of the corresponding sequences A (t) and ϕ (t), which is necessary for further transformations and transmission over the communication channel.

The values of the amplitude and phase are determined by the relations:

and lead to an integer form, while the phase readings will have values from 0 to M-1. In our case, the value of M is 128.

When implementing the considered method of transmitting and receiving speech on a microprocessor to perform calculations using expressions (4) and (5), the operations of finding the values of approximating polynomials given in [7, §§4.4, p. 57-60] can be used to calculate the square root and in [7, §§4.3, p. 54-57] - arctangent.

Next, the sequence of samples of amplitudes A (t) in block 7 is subjected to additional thinning with a coefficient K, and the remaining samples are compressed by level in the compressor 8.

Examples of such an implementation of the process of companding signals to reduce their dynamic range during transmission over channels (compression) and reverse recovery at reception (expansion) are widely known, for example, in [2, pp. 113-128], where companding according to the laws of A and µ is considered . In our case, compression can be applied according to the modified law A [2, p.126, 127] with limiting the level of input signals of amplitudes A (Kt) to 2047 and reducing the number of bits used from 7 to 6 by eliminating the least significant bit in the quantization step code (4 column code table [2]). In this case, the signals A _c (Kt) at the compressor output will have values in the range from 0 to 63.

In order to increase the noise immunity of signal transmission over the communication channel, the converted amplitude values in the adder 9 are shifted, by analogy with the prototype, by the value of the protective interval C ₁ = 32 to reduce the level of anomalous errors that occur when decrypting distorted amplitudes with values close to zero or to the module M.

The new values A _c (Kt) + C ₁ , taking integer values from 32 to 95, are fed to one of the inputs of multiplexer 10, where they are combined in one sequence with phase readings ϕ (t) received at its other input, the values of which vary in the range from 0 to 127. From the output of multiplexer 10, samples of the sequence V _k (T), k = 1, ..., L, are input to the encoder for classification. For each time interval T equal to the duration of the channel frame, L samples, including L _{phases of phase} samples and L _am samples of amplitudes, are input to the encoder.

In the encoder 11, the samples V _k (T) received at its input are classified by adding modulo M with a sequence of pseudorandom numbers generated in the encoder, and issued for further transformations to the input of the demultiplexer 12 in the form of a sequence of classified samples S _k (T), divided by packets of length L.

The demultiplexer 12 distributes L classified samples into two groups for their preliminary coding, which is carried out with amplitudes and phases of harmonics before transmitting them to the communication channel. The first group includes samples S _i (T), i = 1, ..., N / 2, which determine the harmonics amplitudes, the second sample includes S _j (T), j = 1 + N / 2, ..., 3 / 2N, which determine the phases harmonics.

Amplitude encoder 13 (Fig. 1b) attaches to each input sample S _i (T) a control sample with a value equal to the addition of the input sample to module M, after which the values of both samples increase the value of the protective interval C ₂ . From the output of the encoder 13, each pair of samples a _n (T) = S _i (T) + C ₂ and a _{n + 1} (T) = MS _i (T) + C ₂ are output to the block of parallel-parallel software / PA conversion 15 for accumulation of all N samples per frame. The guard interval of C ₂ amplitudes is intended to exclude the transmission of harmonics with zero amplitude, which may be among classified samples, which would create uncertainty in the restoration of the initial phase of such a harmonic at the reception. In our case, the value of the protective interval C ₂ is 32.

The secret phase values S _j (T) are, by analogy with the prototype, differential encoding in phase encoder 14, an implementation example of which is shown in FIG. The encoder comprises an adder 14.1 modulo M and a delay line (register) 14.2. The adder receives the input phase samples S _j (T) and phase samples ϕ _n * (T) from the output of the adder, delayed by a time equal to the duration of the frame T. The values of the encoded phases ϕ _n (T) at the output of the encoder 14 are determined by the relation ϕ _n ( T) = (S _j (T) + ϕ _n * (T)) modM for n = jN / 2. Such phase encoding, together with appropriate reception decoding, ensures the correct restoration of the transmitted phase values regardless of the amount of signal delay in the transmission path. It also provides compensation for slow changes in the phase characteristics of the communication channel. From the output of the encoder 14, the samples of the encoded phases ϕ _n (T) are output to a block of parallel-parallel software / PA conversion 16 for accumulating all N samples per frame.

From the parallel outputs of blocks 15 and 16 N pairs of samples of encoded amplitudes a _n and phases ϕ _n , n = 1, ..., N, are entered at the corresponding inputs to block 17 of the IFFT. The unit generates a group signal by converting signals from the frequency domain to the time domain in the form of the sum of harmonics with the corresponding amplitudes and initial phases. The harmonic numbers and the corresponding inputs of the IFFT block are selected in such a way as to ensure matching of the spectrum of the resulting signal with the bandwidth of the communication channel. With the Fourier transform dimension equal to 128, and the sampling frequency fd ₂ equal to 7200 Hz or 8000 Hz, the harmonic frequencies must be a multiple of 56.25 Hz and, accordingly, 62.5 Hz.

Based on this, with a standard telephone channel bandwidth of 300-3400 Hz, it is possible to use up to 55 harmonics (from the 6th to 60th) for the first value of the sampling frequency, and up to 50 (from the 5th to 54th for the second value) -u), which corresponds to the examples given in table 1.

The group signal from the output of the OBPF block 17 enters the block 18 of the temporary guard interval (VZI), repeating the first 32 counts of the received time signal, increasing its duration to 160 counts. Block 18 is designed to reduce interframe distortion when changing the propagation time of signals in the communication channel and is similar to the corresponding block of the prototype.

The generated single time stream of samples following with a frequency f _d2 Hz in the form of a group signal frames with a duration T is transmitted to communication channel 20 using, in particular, preliminary conversion to analog form using DAC 19.

On the receiving side (Fig. 2a), the signal 21 received from the communication channel enters the ADC and demodulation unit 22, similar to the corresponding blocks of the prototype method implementation scheme. In block 22, analog-to-digital conversion is performed with a sampling frequency f _d2 and digital demodulation of the input signal. As a result of this, 128 averages in the frame of samples are selected from the received sequence of samples at (t ₂ ) and entered into block 23 of the FFT. In block 23, the input samples are converted from the time domain to the frequency one with the formation of N samples of the real Y _In (T) and imaginary Y _Qn (T) parts of the signal.

After that, in block 24, according to the values of Y _In (T)) and Y _Qn (T), the values of the amplitudes and phases N of the harmonics of the digital signal are calculated by the relations:

и фаз ϕ_n(Т) подвергаются декодированию соответственно в декодере амплитуд 25 (фиг.2б) и декодере фаз 26 (фиг.2в).Then the calculated values of the amplitudes

and phases ϕ _n (T) are subjected to decoding, respectively, in amplitude decoder 25 (FIG. 2b) and phase decoder 26 (FIG. 2c).

Amplitude decoder 25 decodes the amplitudes of each pair of classified and reference harmonics coupled in level in the transmitter to form one final amplitude from them. The decoding operation consists in adjusting the levels of the source amplitudes using the correction factor and calculating the value of the resulting amplitude. The correction factor is calculated in the decoder 25 using the adder 25.1 and the divider 25.5.

вводится в блок 25.5, где осуществляется сравнение полученного значения суммы Z с номинальным уровнем суммы амплитуд сопряженных гармоник на выходе кодера амплитуд 13 передатчика. Этот уровень задается константой С₃=М+2С₂, используемой в блоке 25.5 в качестве делимого, а в качестве делителя там служит сумма Z.From the output of the adder 25.1, the result of the addition of two conjugate amplitudes

is entered into block 25.5, where the obtained value of the sum Z is compared with the nominal level of the sum of the amplitudes of the coupled harmonics at the output of the transmitter encoder amplitudes 13. This level is set by a constant C ₃ = M + 2C ₂ , used in block 25.5 as a dividend, and the sum Z serves as a divisor there.

с выхода сумматора 25.2. Значения N/2 итоговых амплитуд

определяются как половина суммы модуля М и откорректированных разностей µR_i(T) амплитуд засекреченной и контрольной гармоник, т.е., в соответствии с соотношениемThe correction factor µ = C ₃ / Z from the output of the divider 25.5 is fed to one of the inputs of the multiplier 25.3, the other input of which is fed to correct the difference in amplitudes

from the output of the adder 25.2. N / 2 values of the resulting amplitudes

are defined as half the sum of the module M and the corrected differences μR _i (T) of the amplitudes of the classified and control harmonics, i.e., in accordance with the ratio

соответствующей пары принятых из канала гармоник в изменяющихся условиях ослабления или усиления принимаемых сигналов при их многолучевом распространении.Let us compare the value of S _i (T) of the amplitude of the transmitted classified harmonic with the value of the total amplitude calculated from relation (8)

the corresponding pair of harmonics received from the channel under varying conditions of attenuation or amplification of the received signals during their multipath propagation.

и

При этом уровни амплитуд переданных гармоник согласно описанию передающей части примера реализации будут равныFor this we introduce the amplification (attenuation) indicators in the communication channel of the levels of the first and second harmonics in one of the pairs of coded harmonics - coefficients

and

Moreover, the amplitudes of the transmitted harmonics according to the description of the transmitting part of the example implementation will be equal

a ₁ (T) = S ₁ (T) + C ₂ , and ₂ (T) = MS ₁ (T) + C ₂ .

Taking into account the introduced notation, we obtain the relation

which after conversion takes the form

Z ₁ = (η ₁ -η ₂ ) S ₁ (T) + C ₂ (η ₁ + η ₂ ) + η ₂ M.

The correction factor µ ₁ in this case will be determined by the relation

μ ₁ = C ₃ / Z ₁ = (M + 2C ₂ ) / ((η ₁ -η ₂ ) S ₁ (T) + C ₂ (η ₁ + η ₂ ) + η ₂ M.

с учетом введенныхAmplitude difference

in view of

coefficients is defined as

R ₁ (T) = η ₁ (S ₁ (T) + C ₂ ) -η ₂ (MS ₁ (T) + C ₂ ) = (η ₁ + η ₂ ) S ₁ (T) + C ₂ (η ₁ -η ₂ ) -η ₂ M.

When the coefficients η ₁ and η _{2 are} equal or close in value, the expressions for μ ₁ and R ₁ (T) are transformed, respectively, to the form

μ ₁ = (M + 2C ₂ ) / ((η ₁ -η ₂ ) S ₁ (T) + C ₂ (η ₁ + η ₂ ) + η ₂ M) = (2C ₂ + M) / η (2C ₂ + M) = 1 / η,

R ₁ (T) = (η ₁ + η ₂ ) S ₁ (T) + C ₂ (η ₁ -η ₂ ) -η ₂ M = η (2S ₁ (T) -M), where η = η ₁ = η ₂ .

Now we substitute the obtained expressions for the correction coefficient μ ₁ and the difference R ₁ (T) in the right side of relation (8) and we obtain that

итоговой амплитуды равно значению S₁(T) переданной засекреченной амплитуды. При этом указанное равенство переданной и восстановленной на приеме амплитуд не зависит от уровня коэффициентов η₁ и η₂. Поэтому для повышения защищенности от селективных частотных замираний целесообразно в качестве гармоник, сопряженных по амплитудному кодированию, выбирать соседние по частотному спектру гармоники, характеризующиеся близкими условиями распространения сигналов. Эффективность этого подтверждена результатами моделирования рассматриваемого способа передачи речи на имитаторах коротковолновых каналах.Thus, with the same amplification (attenuation) of signals within a pair of harmonics conjugated by coding (main and control), the value calculated in accordance with relation (8)

the total amplitude is equal to the value S ₁ (T) of the transmitted classified amplitude. Moreover, the indicated equality of the amplitudes transmitted and restored at the reception does not depend on the level of the coefficients η ₁ and η ₂ . Therefore, to increase the protection against selective frequency fading, it is advisable to select harmonics adjacent to the frequency spectrum as harmonics conjugate in amplitude coding, characterized by close propagation conditions of the signals. The effectiveness of this is confirmed by the simulation results of the considered method of voice transmission on simulators of short-wave channels.

When the coefficients η ₁ and η _{2 are} mismatched relative to each other, for example, due to the influence of noise, distortions appear in the resulting amplitudes. However, their value does not exceed half of the total distortion of the main and control amplitudes and the resulting distortions do not extend to neighboring harmonics in the current or subsequent frames.

путем исключения операции деления на 2. Для этого уменьшим значение константы С₃ в два раза по сравнению с указанным выше. Положив С₃=М/2+С₂, можно вместо соотношения (8) для вычисления

применить соотношениеWe also note the possibility of some simplification of the implementation of the calculation of values

by eliminating the operation of dividing by 2. To do this, we reduce the value of the constant C ₃ twice as compared with the above. Putting C ₃ = M / 2 + C ₂ , we can instead of relation (8) to calculate

apply ratio

according to which the adjusted difference µR _i (T) of the amplitudes is shifted by a fixed value M / 2, which corresponds to the circuit of the amplitude decoder 25 shown in FIG.

в котором через ϕ_n*(Т) обозначены задержанные на длительность Т кадра входные фазы ϕ_n(Т).The phase decoder 26 is similar to the phase decoder in the prototype and performs differential decoding of N classified phases ϕ _n (T) in accordance with the ratio

in which the input phases ϕ _n (T) delayed by the duration of the T frame are denoted by ϕ _n * (T).

и фаз

через мультиплексор 27 объединяются в одну временную последовательность отсчетов

, k=1, …, L и поступают в дешифратор 28 для рассекречивания. Рассекреченные отсчеты V_k((T) в демультиплексоре 29 разделяются на две группы, соответствующие L_ам отсчетам рассекреченных компрессированных амплитуд A_c(Kt) и L_фаз отсчетам рассекреченных фаз ϕ(t).Decoded Amplitude Samples

and phases

through the multiplexer 27 are combined into one time sequence of samples

, k = 1, ..., L and enter the decryptor 28 for declassification. Unclassified samples V _k ((T) in the demultiplexer 29 are divided into two groups corresponding to L _s readings declassified compressed amplitude A _c (Kt) and L _phases counts declassified phases φ (t).

The latter go to the quadrature calculator 32, and the amplitudes go to the interpolation unit 30 to calculate the values of the amplitude samples missed in the transmission during thinning with a coefficient K. For these purposes, taking into account the one polarity (unsigned) of the received and restored amplitudes and a small degree of thinning, fairly good results provide well-known methods of smoothing filtering, spline and linear interpolation [8, p. 466-454].

As a result of interpolation in block 30, the number of samples of compressed amplitudes in the output sequence A _c (t) becomes equal to the number of phases in the sequence ϕ (t) from the output of block 29. Before expanding in block 31 of samples of amplitudes, their values are reduced by the value of the protective interval C ₁ = 32 according to the following rule:

A _c (t) = A _c (t) -C ₁ if A _c (t)> C ₁ -1,

A _c (t) = 0 if A _c (t) <C ₁ .

After expanding the amplitudes, the input of block 32 of the quadrature calculator receives in each frame the same number of samples of amplitudes as the number of phase samples received from the output of block 29.

The values of the phases and expanded amplitudes received at the inputs of block 32 are converted into low-frequency quadrature components in it according to the following relations:

V _c (t) = A (t) cos (2πφ (t) / M),

V _s (t) = A (t) sin (2πφ (t) / M),

where V _c (t) and V _s (t) are the in-phase and quadrature components of the speech signal, respectively.

The values cos (2πφ (t) / M) and sin (2πφ (t) / M) can be stored in 32 cosines and sines tables prepared in block 32 and selected from them at the address φ (t) = 0.1 ... M-1 .

The signals V _c (t) and V _s (t) have a sampling frequency of L _phases T Hz. The spectra of these signals are low-frequency with an upper frequency of no higher (L _phases T) / 2 Hz. To convert them into source speech signals with a spectral width in W _pc Hz, the following two processing steps are performed sequentially.

The first stage consists in calculating (interpolating) between each adjacent samples of signals V _c (t) and V _s (t) new K _d -1 samples, which corresponds to an increase in the sampling frequency of signals by K _d times without changing their spectral properties. (The values of K _d for several implementations of the method in question are given in table 1).

These operations are implemented in block 33 of the quadrature interpolation, as in the prototype based on the well-known method using two interpolating filters and sampling frequency f _d1 . From the output of block 33, the interpolated signals of the low-frequency quadrature components V _c (t ₁ ) and V _s (t ₁ ) will be output.

The second stage is carried out in the block of the quadrature modulator 34, similar in function to the corresponding blocks of the prototype, and consists in multiplying the signals from the outputs of the interpolating filters by the cosine and, accordingly, the sine of the frequency f _c in order to transfer the spectrum of these signals to the high frequencies and combine these signals into output bandpass speech signal.

The combined speech signal V (t ₁ ) is formed in accordance with the ratio

V (t ₁ ) = V _c (t ₁ ) cos (2πf _c t ₁ ) + V _s (t ₁ ) sin (2πf _c t ₁ ).

Here t ₁ = n / f _d1 , where n = 1, 2, 3, ... is the number of samples that corresponds to the time points of the formation of signal samples.

The value of the modulation frequency f _c is chosen equal to the average frequency of the allocated band of the spectrum of the speech signal with a lower frequency value of this band usually around 200 Hz. In our case, the value of the frequency f _c will be in the frequency range from 1450 to 1550 Hz.

The speech signal V (t ₁ ) with a frequency spectrum in the band from 200 to 2600 or 2900 Hz, depending on the selected mode in Table 1, is output through the DAC 35 to the recipient's phone.

The structural diagrams shown in figa and figa only explain the principles and sequence of operations of the proposed method. The specific forms of implementation of these operations and their parameters depend on the transmission conditions, operating conditions, and the used elemental base. The above examples are focused on software implementation using microprocessors.

The experimental measurements of the accuracy of restoration of thinned amplitude samples during thinning amplitudes by a factor of 2 showed that the signal-to-error ratio is about 20-30 dB for low and medium level speech signals, which corresponds to good radio channel quality, and increases with increasing signal level due to the limitation of the level of amplitudes in the compressor samples. With a decrease in the quality of communication, which is especially characteristic of radio channels, the quality and intelligibility of speech using the proposed method is higher compared to the application of the prototype method due to the higher noise immunity of the transmission of channel signals.

If you choose the degree of thinning a little more than two (3 or 4), then in this case the decrease in the accuracy of the restoration of amplitudes on the quality of speech will be largely compensated (and even with an excess of losses) due to the possibility of transmitting more speech phases in each frame than in prototype method. This option, by increasing the frequency of the phase samples, provides an extension of the frequency band of the transmitted speech signal, which leads to an improvement in the quality of the reconstructed speech compared to decimation of every second sample.

Thus, the feasibility of the claimed technical result is shown, which provides, as compared with the prototype, an increase in the noise immunity and transmission quality of speech signals without expanding a given transmission channel bandwidth.

Literature

1. Diffie W., Hellman M.E. Security and imitation resistance. Introduction to cryptography. TIIER. 1979, No. 3.

2. J. Bellany. Digital telephony. - M .: Radio and communications, 1986. - 544 p.

3. Confidential communication device. RF patent №2117401, m.cl. H04K 1/00, 08/10/98, bull. Number 22.

4. A system for transmitting and receiving encoded speech. U.S. Patent No. 4,179,586, M.C. H04K 1/00, 1979.

5. A method for transmitting and receiving encoded speech. RF patent No. 2221284, m.cl. G10L 19/00, 01/10/2004, bull. No. 1. Prototype.

6. Sergienko A.B. Digital signal processing. - SPb .: Peter, 2002 .-- 608 p.: Ill.

7. DIGITAL SIGNAL PROCESSING. Using the ADSP-2100 Family, v. 1.

8. Dyakonov V. D93 MATLAB 6: training course - St. Petersburg: Peter, 2001. - 592 pp., Ill.

Claims (1)

A method for transmitting and receiving encoded speech, which includes the preliminary transmission and detection of a preamble for correcting useful signals at the receiving side and synchronizing encoders, according to which the speech signal is converted from analog to digital on the transmitting side, it is band-pass filtered with the selection of sequences of samples of low-frequency quadrature components, then, by decimation, the sampling frequency in these sequences is brought to a value equal to the total the spectrum of quadrature components, and the thinned readings of the quadrature components calculate the amplitudes and phases, and the number M of possible phase values is set equal to the power of 2, then by applying the compression operation, reduce the values of the amplitude levels to a value less than the number M, increase the amplitudes by of the guard interval, phase values and amplitudes are classified modulo M and the classified samples are distributed into two groups, samples in one of them determine the amp ituda of N orthogonal harmonics with frequencies in the passband of the communication channel, and the samples in the second group, previously subjected to differential modular coding, specify the initial phases of these harmonics, from which they then form a group signal by converting from the frequency domain to the time domain using the inverse Fourier transform with addition to the generated group signal of the guard time interval, thereby forming a frame time group signal of duration T, which is transmitted to communication channel, and on the receiving side, the frame signals received from the channel are in turn converted from the time domain into the frequency domain by using the direct Fourier transform, the output signals of which determine the amplitudes and phases of the orthogonal harmonics of the received signal, carry out differential decoding of the harmonics phases, and then declassify them by module M, the sequence of samples of amplitudes and phases, reduce the declassified amplitudes by the value of the guard interval, after which the values of the amplitudes of exp the low-frequency quadrature components of the declassified signal are generated and calculated by multiplying the amplitudes by the cosine and sine of the corresponding phases, then the signals of the quadrature components are subjected to low-frequency interpolation and the spectrum of the interpolated signals is transferred upward by a frequency equal to the average frequency of the frequency band of the speech signal in transmission, and by combining the interpolated signals quadrature components form a single time stream of a digital speech signal before converting it into an log view, characterized in that on the transmitting side, after performing the amplitude compression operation and before the start of the secret operation, uniformly thinning the samples in the time sequence of amplitudes is performed at least twice, and after that the remaining amplitude samples and phase samples are combined in a certain order in one time sequence and so that their total number does not exceed N + N / 2 over time T, and when distributing classified samples into groups, they include amp itd of channel harmonics of N / 2 classified samples and supplement this group with the same number of control samples with the value of each of these samples equal to the addition to module M of the value of the corresponding classified sample of this group, and on the receiving side after determining the amplitude levels and phase values of the orthogonal harmonics of the accepted of the signal, each pair of classified and control amplitudes of harmonics is decoded with the formation of one final amplitude, which consists in adjusting the levels of the initial amplitudes d using a correction factor calculated by comparing the sum of these amplitudes with a fixed threshold and calculating the value of the total amplitude as half the difference in the magnitude of the corrected amplitudes of the classified and control harmonics increased by the value of the module M, and after declassification of the amplitudes and phases restored during decoding and until the declassified samples are reduced amplitudes by the value of the guard interval, the order of phases and amplitudes is restored and the value of samples of amplified samples thinned out by means of interpolation from adjacent samples of a sequence of declassified amplitudes and insertion of the calculated samples between them into a sequence of interpolated samples.

Images