JPH0481139A - Voice privacy call device - Google Patents

Abstract

PURPOSE:To prevent remarkable deterioration in decoded voice quality due to privacy call processing and to improve the quality of a decoded voice signal by providing a frequency inversion means before and after a transposition means, interleaving a signal in a normal order and employing a high speed Fourier transformation device. CONSTITUTION:A voice signal is subjected to interleave processing in the reverse order and signals are made coincident on a time axis by a delay element, the phase is shifted by a desired quantity at a phase shifter. Then a high speed inverse Fourier transformation device 4 applies inverse FFT processing to the signals to obtain signals in an even number band in signals at each small band in the ascending order and signals of an odd number band are obtained in an inverted form in the descending order. Only the inverted signal is inverted by frequency inversion circuits (IV)10-16-10-31, a transposition section applies transposition processing to the signal, signals of the same band are subjected to inverse processing at frequency inversion circuits (IV)11-16-11-31, subjected to inverse FFT processing at a high speed inverse Fourier transformation device 6. subjected to phase shift processing, inputted to polyphase sub filters of identical characteristic, in which the signal is subjected to interpolation processing and a privacy call signal is obtained.

Description

[Detailed Description of the Invention] [Summary] This invention relates to a voice privacy device that polarizes a signal in a predetermined frequency band including a voice band by dividing and replacing the signal for the purpose of improving decoded voice quality. If the division unit of the frequency band is n divisions, the input sampling signal is thinned out every n in the reverse order of input, with the thinning interval of n being
1/n of the sampling frequency of the human sampling signal
an input signal thinning section that thins out n sampled low-speed signals at a frequency of n and converts them into complex signals of n frequency bands; A first complex polyphase digital filter section, a first phase shift section, and a first fast inverse Fourier transformer are used to convert the signal into a small band signal.
a first frequency band inverting means for inverting a part of the frequency of the small band signal in order to unify all frequency bands of the small band signal to either normal rotation or inversion; transposing means for exchanging each subband signal between the output of the small band signal outputting means that is not processed by the first frequency band inverting means and the output of the frequency band inverting means; a second frequency inverting means for inverting the frequency of the band inverted by the first frequency band inverting means among the small band signals; from a second fast inverse Fourier transformer, a second phase shift section, a second complex polyphase digital filter section and a signal interpolation section for combining the subband signal with that which is not processed by the second frequency inversion means; and second small band signal output means.

[Industrial application field]

The present invention relates to a voice encrypting system that performs division and replacement of an audio band to encrypt speech, and more particularly to an audio enciphering device that performs frequency division and transposition on an analog audio signal by digital signal processing.

As information communication networks become more sophisticated and wide-area, and as communication services become highly diversified, social demands for privacy protection of communication information are increasing. In particular, there is an urgent need to develop superior confidential communication technology for wired and wireless analog voice communications that are widely used in our daily lives.

In addition to high communication privacy, these confidential communication techniques have the following characteristics: there is little deterioration in decoded voice quality or processing delay due to signal processing for communication privacy and decoding, and there is little effect on decoding quality due to line characteristics. Furthermore, low economic cost is an important point for realizing the device.

[Conventional technology]

Prior art documents include the following.

(1) N, S, Jayant: “Analog S
cramblers for 5peech Pr1v
acy, Computers & 5ecurit
y1. pp, 215-289. Nort-Ho
lland (1982).

(2) R, V, Cox, D, E, Bock, K.
B., Bauer, J.D., Johnston, J.H.
, 5nyder: “The Analog Voic
e Privacy System, AT&T
Tech, J., 66, 1. pp, 119
-13L (Jan, Feb, 1987).

(3) L, S, Lee, G, C, Chou,
and C, S, ChaB:”ANew Freq
uency Domain 5peech Scrum
bring 5ysteta which does
not Require Frame 5ynchr
onizatiOn”+IEEE Trans, Co
mmun, C0M-32+ 4+ pp, 444-4
56 (April 1984).

(4) Matsubara, Oyo, Sakurai, Koga: “Full-duplex analog secret communication device using FFT and its basic operation” IEICE theory (A),
J72-A, 4. pp. 692-702 (1989)
.

(5) Torii, Higashi, Akita: “Prototype of constant envelope spectral scrambler” IEICE Technical Report, 1N84-121 (1985)
-03) The secret technology for analog audio signals can be roughly divided into:
There are digital confidential communication systems and analog confidential communication systems (refer to the above document (11)). It cannot be said that it has reached a sufficiently practical level yet.As for the latter method, there are various methods such as spectrum inversion method, frequency shift method, sample sequence recombination method in the time domain, noise addition method, and methods that combine these methods. There are several methods, but all of them have problems such as privacy, processing delay, and quality deterioration due to line characteristics.On the other hand, the band division replacement method is a method that achieves high privacy by increasing the number of divisions, and has been widely used. Research has been conducted (see references (3) to (5) above).
A method using the filter bank summation method that does not require a synchronization circuit has been proposed (see document (3) above), but the decoding quality is insufficient (by overlapping DFT frames, the signal processing format There have been problems such as the complexity of the process, and the low calculation efficiency in the complex signal processing process in band division.Furthermore, the method of subdividing the band using FFT (see reference (4) above) has been problematic. proposed,
Although some devices have been put into practical use, they are susceptible to line characteristics and require equipment such as group delay, frequency offset cancellers, and high-precision synchronization circuits, which poses problems in terms of device costs.

As a conventional technology of the present invention, there is a band division permutation scrambler that applies the trans multiplexer technology (abbreviated as TMUX) used in TDM-FDM conversion, and this secret method is a method consisting of a filter bank using DFT. However, by adopting a signal processing method unique to TMUX, good decoding quality can be achieved without overlapping DFT frames, and each subband signal obtained by band division has complex It has features such as high calculation efficiency because it does not include a conjugate part.

A technique employing this TMUX is described in Japanese Patent Application No. 01-059611, which was proposed before the present application.

FIG. 13 is a block diagram showing a conventional voice secret communication device described in the above-mentioned Japanese Patent Application No. 01-059611.
FIG. 14 is a frequency allocation diagram showing a small band signal when the input audio band in FIG. 13 is divided into four for the sake of simplicity.

[Problem to be solved by the invention]

As shown in FIG. 14, in the conventional voice secret communication device using TMUX (Japanese Patent Application No. 01-059611), odd-numbered band #3. #1 is output with its frequency inverted and flows directly to the line. That is, even-numbered band #0. #2 is transposed with normal rotation, and odd-numbered band #3. #1 is transposed while being inverted and output to the line as is.

Therefore, if there is no frequency offset in the line, the odd-numbered band will be inverted once again on the receiving side, returning to the original band, but if there is a frequency offset, the receiving side will change the offset frequency. The band will be shifted by the second frequency, causing a significant deterioration in decoded audio quality.

SUMMARY OF THE INVENTION An object of the present invention is to prevent decoded voice quality from significantly deteriorating even in the presence of a frequency offset in a voice privacy device, and to improve the decoded voice quality as a whole.

[Means to solve the problem]

FIG. 1 is a block diagram of the principle of the present invention. In the figure, the input signal X (Z) is a signal in a predetermined frequency band including an audio signal, and this predetermined band is divided into n parts. 1
The input sampling signal is thinned out every n in the reverse order of input, with the thinning interval of n, and is thinned out so as to output n sampled low-speed signals at a frequency of 1/n of the sampling frequency of the input sampling signal. Te n
This is the thinning part of the input signal that is converted into a complex signal of several frequency bands. 12 passes each of the n low-speed sampling signals to a first complex polyphase digital filter section 2-
0 to 2-(n-1), first phase shift section 3-0 to 3-
(n-1) and the first fast inverse Fourier transformer 4 to convert into n-divided small-band complex signals.

10i (i=n/2) to 10(n-1) are provided according to the present invention, and are used to normally rotate all frequency bands of n complex signals obtained at the output of the small band signal output means 12. This is first frequency band inverting means for inverting a part of the frequency of the small band signal. 5 is a first frequency band inverting means 10i (i=n/2) among the outputs of the small band signal outputting means 12.
.about.10(n-1) and the output of the frequency band inverting means. 11 i (i=n/2) to 11(n-
1) is the first frequency band inversion unit 10i (i=n/2) of the small band signal transposed by the transposition unit 5.
10(n-1) is a second frequency inverting means that inverts a frequency corresponding to the frequency of the band inverted by 10(n-1). 1
3 is for synthesizing a small band signal of the output of the second frequency inversion means 11i to 11(n-1) and the output of the transposition means 5 which is not processed by the second frequency inversion means,
Second fast inverse Fourier transformer 6, second phase shifter 7
-0 to 7(n-1), second complex polyphase digital filter sections 8-0 to 8-(n-1), and a signal interpolation section 9.

The first and second frequency inverting means 10i to 1 rotate normally.

If the order of thinning in the thinning means 1 is set in the forward order and a fast Fourier transformer is used in place of the first inverse fast Fourier transformer 6, the phase characteristics of each band signal will match, and the − layer decoded audio quality will be improved. do.

If cosine roll-off filters are used as the first and second complex polyphase digital filter sections, code amount interference in signal processing can be reduced, and decoded audio quality can be further improved.

[Effect]

The odd-numbered or even-numbered real parts and imaginary parts of the band-divided complex signal output from the fast inverse Fourier transformer 4 are inverted on the frequency axis by alternately inverting their polarities. By including this band inversion means before and after the transposition means 5, all the band signals are processed in the normal rotation state or all processed in the inversion state, and therefore, the state in which the band signals are also transposed in the normal rotation or inversion state. is output.

In the present invention, even if polarized voice passes through a line with a frequency offset, the only effect is the shift by the offset frequency, and the polarization does not cause significant deterioration in decoded voice quality.

Further, by performing thinning in the normal order and using a fast Fourier transformer, the phases of the small band signals after dividing the input signal match, and the quality of the decoded audio signal is further improved.

Furthermore, by using a cosine roll-off filter as the complex polyphase digital filter, code amount interference in signal processing is reduced.

〔Example〕

FIG. 2 is a block diagram showing a first embodiment of the present invention. In the figure, an example of 32 divisions is shown, and the first
Parts corresponding to those in the figures are given the same reference numerals.

Parts 10-16 to 10-31 and 11-1 marked IV
6 to 11-31 are frequency inverting circuit portions. Z-1~
-2-31 is a delay element, and Z-1 indicates a delay of one cycle. H0 to I"131 are complex polyphase digital filters. Its structure is disclosed in the above-mentioned patent application No. 01-05.
It is similar to that described in No. 9611. E0~
E” is a phase shift device and E=exp(-j2π/4n)
It is. The audio signals are thinned out in reverse order and matched on the time axis using a delay element. They are processed with a polyphase subfilter and phase shifted by the desired amount with a phase shifter. Then, by subjecting them to inverse FFT processing (IFFT) in the fast inverse Fourier transformer 4, the signals of each small band are obtained first, and the signals of the even band are obtained in ascending order.
Furthermore, signals in odd-numbered bands are obtained in descending order in inverted form. Only these inverted signals are sent to the frequency inverting circuit (IV
) 10-16 to 10-31 perform inversion processing, and the transposition unit 5
The signals in the same band are also transposed using a frequency inversion circuit (IV
) 11-16 to 11-31 perform inversion processing, fast inverse Fourier transformer 6 performs inverse FFT processing, phase shift processing,
A secret signal is obtained by inputting it to a polyphase sub-filter with the same characteristics and performing interpolation processing.

FIG. 3 shows frequency inversion circuits 10-16 to 10-31 and 11-16 to 11- for inverting odd-numbered bands.
31. FIG. As shown, the band-divided input complex signal X (n) = an + j
The real part a and the imaginary part of bn are 1. are inverted on the frequency axis by alternately inverting the polarity. By including this circuit before and after the transposition processing, all of the small band signals are processed in the normal rotation state and outputted to the line in the transposed state while maintaining the normal rotation. As a result, even if the polarized voice passes through a line in which a frequency offset exists, the effect is only due to the offset frequency, and polarization does not cause significant deterioration in decoded voice quality.

Frequency inversion circuits 10-16 to 10-31 and 11-1
6 to 11-31 may invert even-numbered small bands. In this case, all the small bands are output to the line in an inverted state, and the influence of frequency offset is also reduced.

FIG. 4 is a block diagram showing a second embodiment of the present invention. 2, the difference from the first embodiment shown in FIG. 2 lies in the thinning section 1a and the fast Fourier transform section 4a, and the other configurations are the same as in FIG. 2. The thinning section 1a thins the input signal in the normal order, and also uses a fast Fourier transformer instead of the fast inverse Fourier transformer to modulate the fundamental band before dividing the audio band. After that, filter processing will be performed. That is, since the signals after thinning have the same phase characteristics, the S/N is further improved than in the first embodiment.

By adopting this method, it becomes possible to match the phase of the input audio signal in each band, and even better decoded audio quality can be obtained.

As yet another embodiment of the present invention, there is a method using a filter having a cosine roll-off characteristic as shown in FIG. 5 as a complex polyphase digital filter.

By using this method, it is possible to reduce code amount interference in signal processing, and good decoded audio quality can be obtained.

The following describes the signal processing configuration, filter design method, evaluation conditions for computer-based simulations, and the evaluation scale used, as well as decoding quality in the case of fixed and variable transposition, the influence of line characteristics, and secrecy based on subjective evaluation. The evaluation results will be explained in more detail.

(1. Configuration of ITMUX Scrambler The audio signal is regarded as a frequency multiplexed signal, and in order to divide it into subband signals, FDM/SSB conversion in TMUX is used, transposition processing is performed in the frequency domain, and scrambled. After that, SSB/FDM conversion is used to combine the subband signals and generate a secret signal sequence.For example, in band 32
FIG. 4 shows the configuration of the TMUX scrambler in the case of division. In the same figure, Z-'Hi, E-'' (0≦i≦31
) are a delay element, a subfilter obtained by decomposing the reference filter, and a phase shift term, respectively. Further, IV is the frequency band inversion processing section mentioned above.

The audio input signal strings are thinned out in the order of input, and 32 signal strings are simultaneously subjected to filter processing, phase shift processing, and FFT processing to obtain each subband signal 30 to 5ell.

The processing up to this point corresponds to an audio band division section using FDM/SSB conversion. Of the obtained subband signals, the latter half SI6 to S3+ signals are obtained in an inverted form on the frequency axis, so they are processed by the frequency band inversion processing unit IV.
After inversion processing is performed and all subband signals are brought into a normal rotation state, transposition processing is performed. Of the transposed signals, the latter half of the signal is similarly inverted, and the obtained subband signals P0 to P31 are subjected to inverse FFT processing,
A secret output signal is obtained by performing phase shift processing, filter processing, and interpolation processing in order. Here, the process from inverse FFT processing to interpolation processing corresponds to a voice band synthesis section using SSB/FDM conversion. Regarding processing on the receiving side,
They are exactly the same except that the transposition pattern in the transposed part is reversed. Below, we will outline the principles of each processing unit and the filter design method.

(11-1" Voice range section This section is the same as the conventional one. For simplicity, Figure 14 shows the principle of the voice band dividing section when the band is divided into four. The low-pass implementation serves as the standard for cutting out subbands. Let H(z) be the transfer function of the filter, and let H,(z) be the filter for cutting out channel IIO in the figure, which is shifted by 172 bands.In order to cut out the audio signal sequence with Ha(z), By performing thinning processing by shifting the band for channels (K=0.1, 2.3) in the negative direction on the frequency axis, subband signals SO to S3 are obtained.Here, subband signals SO to S3 are obtained. By performing band shifting, it is possible to use an FFT with the same number of points as the number of band divisions in filter processing.As a result, each subband signal after thinning processing is separated by one band as shown in the figure. Since it is repeated, the influence of adjacent bands is small and the decoded audio quality is good.It also has the advantage that the number of filter taps for cutting out each band can be configured with a relatively small number.About the explanation of this principle will be explained in more detail later.

(11-2-B The latter half of the sub/<nd signals (S2 and 33 in FIG. 14) extracted by the audio band dividing section are obtained in a state inverted on the frequency axis.This signal is If it is transmitted as is, if a frequency offset exists in the line, decoding quality will deteriorate.

For example, if you set the center frequency of a certain band to be transmitted as fc
Let the frequency of the signal to be transmitted be fc+df.

Also, let fO be the frequency offset of the line. When forward rotation processing is performed on the transmitting and receiving side, the receiving frequency is fc + df + f
It becomes O. On the other hand, when inversion processing is performed on the transmitter/receiver side, the transmission frequency is fc-df because it has been inverted, and the reception frequency is fc+df-fo because an offset is added and the band is further inverted. Therefore, if the normal inversion band and the inversion band coexist, the decoded audio band on the receiving side will include a band shifted by fo in the positive direction and a band shifted by fo in the negative direction, resulting in deterioration of decoding quality. happen. Therefore, the frequency band inversion circuit 10-1 shown in FIG.
is used to return all subband signals to normal rotation, and then perform transposition processing. This circuit performs frequency band inversion processing by multiplying a subband signal by a signal with a frequency that is half the decimation frequency and extracting a complex conjugate portion.

(1)-3” voice.

For simplicity, FIG. 15 shows the principle of the audio band synthesis section in the case of dividing the band into four. The principle of this voice band synthesis itself belongs to the prior art. In the same figure, each subband signal sequence P0 to P3, which has been transposed and the latter half of the signal sequence has been inverted, is
By cutting out and combining the bandpass filters H0 to H3 whose frequency has been converted by FFT and a phase shift term, a complex signal sequence Y (z) as shown in the figure is obtained. By extracting only the time series of the real part of this signal, the complex conjugate band signals are folded around the 0 frequency, and a secret spectrum is obtained. This principle will be explained in more detail later.

(The sampling rate of the design subband signal of the 11-4 filter is 1/N of the external rate, where N is the number of band divisions, so
In order to minimize the code amount interference on the output side, a filter having a Nyquist characteristic (see the above-mentioned document (8)) is used. However, it is assumed that the filter characteristic per stage has a cosine roll-off characteristic as shown in FIG. 5 so that FDM/SSB conversion and SSB/FDM conversion have a pair of filter characteristics. In the figure, fp is the passband frequency, fq
is the stopband frequency and fc is the cutoff frequency. At this time f
There is a relationship of p=fc-Δf, fq=fc+Δf. The amplitude characteristic of this filter is expressed as follows, and the inverse Fourier transform is as follows. Assuming that the transfer function of the FIR filter obtained by the above equation is: The coefficient of each sub-filter is obtained by thinning out every N coefficients of H(z), and the following is obtained.

(2) Computer Simulation In order to evaluate the basic performance of the confidential speech method according to the embodiment of the present invention, such as decoded voice quality and confidentiality, an evaluation was conducted by simulation using a large-scale computer.

The evaluation conditions and evaluation scale will be explained.

(2)-1 The configuration of the evaluation condition simulation is shown in FIG. In this evaluation,
As an audio signal, a male voice news audio band-limited to 300 to 3,300 Hz is used, and this is sampled at 8K) lz with a 12-bit precision A/D converter 61 to approximately 3
The data for 0 seconds is polarized by the polarization device 62 according to the present invention shown in FIG. The /A converter 65 decoded it into an analog signal. As the number of divisions of the audio band increases, steeper filters are required to cut out the band, which increases signal processing delay. In order to maintain naturalness in voice conversation, the processing delay is considered to be limited to 150 to 200 m5 in one direction, so the number of divisions was set to 32 (the recombination band was divided into 25). At this time, the processing delay is 136 m5 when using a filter with 512 taps, and 200 m5 when using a filter with 768 taps.As for the influence of line characteristics, there are synchronization shifts, group delay characteristics, and frequency that particularly affect decoded audio quality. We evaluated the offset. These characteristics are also important guidelines for device design when implementing the present confidential communication system. The group delay characteristic is 3
00Hz and 3300)1z respectively have the maximum delay2
This was realized using a fourth-order IIR all-pass filter that is a combination of second-order filters.

The frequency offset was realized using a digital Hilbert transformer in which the impulse response of a 90-degree phase shifter was multiplied by a Hamming window, and the number of taps was 41. Note that this simulation was performed entirely using double-precision floating-point arithmetic.

(2)-2''' - As an evaluation measure of decoding quality, the evaluation was performed using the segmental S/N expressed by the following formula.

5DRr”=101og+.

(dB) However, X, (k) and Z, (k) are FFT coefficient sequences of the original audio and decoded audio extracted at the r-th frame, respectively. Further, the average value for the frames of 5t)R is assumed to be SDR. In general, it is said that changes in the phase characteristics of audio signals do not have much impact on audio quality from an auditory perspective, and SD, which evaluates reproducibility in the frequency domain,
R is effective as a measure for evaluating decoded speech quality. Here, the larger the number of FFT points used in the calculation of SDR, the better, but since there is a problem with calculation accuracy, in the following evaluation, N = 4096, and the value obtained by taking the average of 16 frames for the sound section is the SDR. And so.

(3) Results of simulations conducted under the evaluation conditions above and beyond. First, we will show the effects of transposition processing on an ideal line, in which no processing is applied in the line simulator, and the effects of variable transposition, which changes the transposition pattern over time. Next, we will show the effects when line characteristics are given, such as synchronization deviation, It is shown by group delay distortion and frequency offset, respectively. Furthermore, we will show the results of subjective evaluation of confidentiality.

(3)-1 Influence of transposition processing FIG. 7 shows a spectrogram in the case of fixed transposition in which the transposition pattern does not change over time. In the same figure, (a) shows the original speech spectrum, which shows the characteristic of the speech signal that the spectral power is relatively concentrated in the low range, but in the secret spectrum of distributed in Furthermore, in the decoded spectrum of (C1), almost the spectrum of the original sound is restored. In FIG. 7, t is the unit time for measuring the spell power.

FIG. 8 shows the relationship between the number of filter taps and SDR. If transposition processing is not performed, S for all tap numbers
The DR is about 30 dB, but as a result of introducing the transposition process, it is thought that deterioration occurs due to the influence of the residual signal left after being recombined with the sidelobe signal. When the number of filter taps is increased, the amplitude characteristics become steeper and the influence of side lobes is reduced, resulting in better SDR characteristics. However, according to subjective evaluation of decoded audio, the sound quality is sufficiently good even when the SDR is around 15 to 20 dB. I found out that I can get it.

(31-2 Effect of variable transposition When variable transposition is performed, due to the response characteristics of the filter,
Immediately after switching the transposed pattern, residual signals in the filter are observed as noise until they are all sent out. If the number of taps is very long like an ideal filter, variable transposition is virtually impossible, but if it can be configured to be relatively short, it is possible, although some quality deterioration is observed. FIG. 9 shows the relationship between the transposition period and SDR when variable transposition is performed using the number of taps as a parameter. From this result, the transposition period is 174
It was found that the decoded audio quality is not affected much if it lasts for about seconds.

(31--3 Shadow of synchronization shift Figure 10 shows the influence of synchronization shift on decoding quality.When the number of band divisions is 32, due to the configuration of the signal processing method,
A deviation of 16 samples caused the maximum deviation, but
Even in that case, the deterioration is 2 to 3 dB, and it can be seen that in the case of fixed transposition, almost no synchronization circuit is required. Further, in the case of the variable transposition method, it is necessary to establish frame synchronization, but very high synchronization accuracy is not required.

(3)-4 Influence of Group Delay Distortion The influence of the group delay distortion of the line on decoding quality is shown in FIG. 11. In the figure, the group delay time indicates the group delay around 300 Hz and 3300 Hz. Group delay 1,
5+ws corresponds to the characteristic of the NTT l link.

From the results, the quality does not deteriorate much even if delay equalization is not performed for about 1 to 2 NTT links, and there is no significant deterioration for group delays beyond that, so simple fixing etc. But I can handle it.

(3) The influence on decoding quality due to the frequency offset of the line with -5 offset is firstly evaluated.
Due to the three-frequency offset shown in FIG. 2, a portion of each band is shifted into another band and decoded as a separate band signal, resulting in decoding quality deterioration. In this simulation, the number of band divisions is 32, and each band is 125Hz.
Since the bandwidth is wide, the rate at which this degradation occurs for each bandwidth is small. In an actual line, one frequency offset is up to 4H.
Since the range is about z, there is no need to add a particular canceling device.

(3)-6 Evaluation of Confidentiality In order to evaluate the intelligibility of the confidential conversation method, a numerical intelligibility test was conducted. This test involves reading out four-digit numbers, listening to the scrambled audio, and guessing the original number.It is widely used as a test that quantitatively and comparatively evaluates the confidentiality of scramblers. Since the target voice is a number, the probability of getting the answer correct even if you answer randomly is 1.
It is 0%. Using 15 sets of 4-digit scrambled numbers, the evaluation was conducted by 20 unskilled men and women at the institute. As a result, the average intelligibility was 13.5%. A high secret signal strength equivalent to or higher than that of other scramblers with a large number of band divisions was obtained.

(4) Conclusion We proposed a scrambler for analog audio signals that applies TMUX technology, and described its signal processing configuration and filter design method. We also conducted computer simulations using a large-scale computer to evaluate the decoded voice quality, the effects of various line characteristics, and the intelligibility of confidential speech. As a result, sufficiently good decoding quality with an SDR of 15 to 20 dB was obtained regarding the decoded audio quality. In addition, it was found that the quality deterioration due to the influence of line characteristics such as group delay distortion, synchronization error, frequency offset, etc. is only 3 to 5 dB, which is not much affected, and it is an economical method that can reduce equipment costs. I understand. Furthermore, the secrecy of speech was 13.5χ as a result of a numerical intelligibility test, indicating that the secrecy of speech is as high as or higher than that of the conventional band division and replacement method. As a result of the above, this secret method has confidentiality,
It has become clear that this confidential communication method is excellent in terms of both economy and decoding quality.

Next, the principle of the audio band division section and the principle of the audio band synthesis section will be explained in detail.

(2) Principle of the audio band division section The principle of dividing the audio band into four parts belongs to the prior art, and its method is shown in FIG. In the following description, the number of band divisions is assumed to be N. The audio input signal sequence is x(nT
), and let its Z-transformed representation be X (z). Here T=1
/f, = 1/8 KHz. Let H(
z). At this time, the β shadow signal processing method (the above-mentioned document (
Let Ho(z) be a filter for cutting out channel #0 in the figure, which corresponds to the fundamental band of (see 7). Furthermore, in order to extract the audio signal sequence as Ho(z), the sequence with the channel bandwidth shifted in the negative direction on the frequency axis is converted into Xm(
z), and the subband signal of each band thus extracted is S h (z), then 5k(z)=Ho(z) Xk(z)
−(11) Here, Ho(z) corresponds to the reference filter H(z) shifted by 1/2 band, so Ha(z)=H[Zexp (j2yr(1/ 4N)l
] −(2). Moreover, when Ho(z) is decomposed into N sub-filters, it is expressed as follows, so from equation (2), it becomes as follows. Here, E=exp(-j2π/4N).

On the other hand, the input signal sequence Xk(z) is the original human power signal sequence X (
z) shifted in the negative direction by 2 by the band, and is expressed as Since it can be expressed as decomposed, from equation (5), it becomes.

Here, W=exp( j2π/N) It is.

(4).

Substituting each equation (7) into (1) yields.

The signal sequence is obtained by thinning out 5k(z) to 1/N, and when the above equation is expressed as a matrix, S (z') -W @ E-H(z') ・X"(z)" X= [Xo Xz -Xs-'
+]”W is an N-point complex DFT (FF when N is a power of 2)
T is available). E is a phase shift term required to shift the reference filter by 172 bands. Although complex coefficients are used as the coefficients of the digital filter, since the input signal sequence has only the real part, the amount of calculation is equivalent to that of a real filter. Of the obtained subband signals, N/2-1 to N-
The second half of the signal sequence up to 1 is inverted by the inverting circuit shown in FIG. 3, and then transposed.

■Principle of audio band synthesis unit The principle itself for synthesizing four divided bands also belongs to the prior art, and is shown in FIG. 15. Note that, assuming that the number of band divisions is N, it is assumed that the second half subband signals from N/2-1 to N-1 after transposition have also been inverted in advance by the inversion circuit shown in FIG. 3.

The transfer function of the k-th channel filter and the transposed subband signal in the frequency domain are respectively expressed as Hk(z),
When P k (zN) is expressed and the Z-transformed representation of the combined output signal sequence is Y (z), it is expressed as follows. Here, Hk(z) is H as shown in Figure 2.
(z) is shifted by (2 + (1/2)) band bones, and is given by. Here, when Hk(z) is decomposed into N sub-filters, Hm(z) −ΣZ−”Hk(z′′)+++αmi@ formula, becomes.

From the 00 type, becomes.

When this is expressed as a matrix, it becomes Y (z) = H (z') -W- pT (zN), where αe P = [Po Pz -PH-1
]”W can use N-point complex inverse FFT.H(z') is a complex digital filter whose input signal is a complex number and whose output side is only the real part.The output signal Y obtained by the above processing is
(z) is a complex signal sequence, and by extracting only the time series of its real part, a secret spectrum Re[Y(2)] folded around the 0 frequency is obtained.

〔Effect of the invention〕

As is clear from the above description, according to the present invention, by providing the frequency inversion means before and after the transposition means, even if the polarized voice passes through a line in which a frequency offset exists, the offset The only effect is a shift by the frequency, and polarization does not cause significant deterioration in decoded voice quality.

In addition, by thinning out in the forward order and using a fast Fourier transformer, the input signal is processed within the audio band (within the baseband), and the phases of the small band signals after dividing the input signal match. As a result, the quality of the decoded audio signal is further improved.

Furthermore, by using a cosine roll-off filter as the complex polyphase digital filter, code amount interference in signal processing is reduced.

[Brief explanation of drawings]

Fig. 1 is a block diagram of the principle of the present invention, Fig. 2 is a block diagram showing the first embodiment of the invention, Fig. 3 is a block diagram showing the configuration of frequency inversion in Fig. 2, and Fig. 4 is a block diagram showing the structure of frequency inversion in Fig. 2. A block diagram showing a second embodiment of the present invention. FIG. 5 is a graph showing the characteristics of a cosine roll-off filter used in a complex polyphase digital filter according to an embodiment of the present invention. FIG. 6 is an embodiment of the present invention. 7 is a block diagram showing a simulation system for testing the effect of Graph showing the relationship between SDR. FIG. 9 is a graph showing the relationship between transposition period and SDR in the embodiment of the present invention. FIG. 10 is a graph showing the relationship between synchronization shift and SDR in the embodiment of the present invention. is the group delay distortion and SD in the embodiment of the present invention
Figure 12 is a graph showing the relationship between frequency offset and SDR in the embodiment of the present invention, Figure 13 is a block diagram showing a conventional confidential communication device, and Figure 814 is a graph showing the principle of audio band division. The figure shown in Figure 15 is a diagram showing the principle of voice band synthesis. In the figure, 1 is a thinning means, 2-0 to 2-(n-1) and 8-0 to 8-(n-1)
are complex polyphase digital filters 3-0 to 3-(n-1) and 7-0 to 78n-1) are phase shift parts, 4 and 6 are fast inverse Fourier transformers, 4a is a fast Fourier transformer, and 5 is a fast Fourier transformer. transposition means; 10i to 10(n-1) and lli to 1(n-1); the frequency inversion circuit 12; small band signal output means; 13, small band signal synthesis means.

Claims (1)

[Claims] 1. In a voice privacy device that polarizes a signal in a predetermined frequency band including a voice band by dividing and replacing the signal, if the division unit of the predetermined frequency band is n divisions, the input sampling signal is thinned out. N frequency bands are thinned out every n in the reverse order of the input, with an interval of n, and are thinned out so as to output n sampled low-speed signals at a frequency of 1/n of the sampling frequency of the input sampling signal. An input signal thinning section converts the input signal into a complex signal, and the complex signal output from the input signal thinning section is passed through a first complex polyphase digital filter section, a first phase shift section and a first fast inverse Fourier transformer. a first small-band signal output means for converting the small-band signal into a small-band signal by converting the small-band signal into a small-band signal; A first frequency band inverting means for inverting, an output of the small band signal outputting means that is not processed by the first frequency band inverting means, and an output of the frequency band inverting means. transposing means for transposing; second frequency inverting means for inverting a frequency corresponding to the frequency of the band inverted by the first frequency band inverting means of the small band signal transposed by the transposing means; a second fast inverse Fourier transformer for synthesizing a small band signal of the output of the frequency inverting means and the output of the transposing means that is not processed by the second frequency inverting means; and a second phase shift unit. , a second small-band signal output means consisting of a second complex polyphase digital filter section and a signal interpolation section. 2. The voice secret communication device according to claim 1, wherein the first and second frequency inversion means invert frequencies of odd-numbered small bands. 3. The voice confidential communication device according to claim 1, wherein the first and second frequency inversion means invert frequencies of even-numbered small bands. 4. In a voice privacy device that polarizes a signal in a predetermined frequency band including a voice band by dividing and replacing the signal, if the unit of division of the predetermined frequency band is n divisions, then the thinning interval of the input sampling signal is n. The signal is thinned out in the same normal order as the input, and the signal is thinned out so as to output n sampled low-speed signals at a frequency of 1/n of the sampling frequency of the input sampling signal.
An input signal thinning section that converts the input signal into a complex signal in a frequency band, and a complex signal output from the input signal thinning section,
a first complex polyphase digital filter section, a first
a first small band signal output means for converting into a small band signal using a phase shift unit and a first fast Fourier transformer; and converting all frequency bands of the small band signal into either normal rotation or inversion. a first frequency band inverting means for inverting a part of the frequency of the small band signal in order to unify the frequency; an output of the small band signal outputting means that is not processed by the first frequency band inverting means; transposing means for exchanging each subband signal with the output of the band inverting means; and a second transposing means for inverting the frequency of the band inverted by the first frequency band inverting means among the subband signals transposed by the transposing means. and a second fast inverse Fourier for synthesizing a small band signal of the output of the second frequency inverting means and the output of the transposing means that is not processed by the second frequency inverting means. A voice secret communication device comprising a converter, a second phase shift section, a second complex polyphase digital filter section, and a second small band signal output means consisting of a signal interpolation section. 5. The voice confidential communication device according to claim 1 or 4, wherein cosine roll-off filters are used as the first and second complex polyphase digital filters.