JPS5848916B2 - Audio transmission processing method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to improvements in transmission and processing of audio waveform signals using a zero-crossing wave transmission method.

A zero-crossing wave, which is the amplitude of an audio waveform signal that is binarized at its zero-crossing position, is the simplest form of digitizing an audio waveform signal.
Ability to understand words at a level similar to that of Tori.

In addition, as an example of applying this method to wireless communication, an audio waveform signal is subjected to single sideband modulation with carrier wave suppression (hereinafter referred to as "SS BJ"), and the amplitude is made constant using a limiter before being transmitted. A demodulation method has been proposed.

This constant amplitude SSB (hereinafter referred to as JCSBBJ) is
It can be regarded as a zero-crossing wave of SSB.

Incidentally, since the waveform amplitudes of both the zero-crossing wave and the CSSB are constant, their processing is simple and efficient transmission is possible.

In particular, zero-crossing waves can be applied to digital information processing because digital technology can be used.

However, with zero-crossing waves and CSSB, the low-level noise in the pause section is at the same level as the signal, making it extremely difficult to hear, and there is also a lot of distortion due to waveform shaping, so it is not suitable for practical communication. Not available.

Therefore, it is often used for feature extraction, etc., and when used for communication, amplitude information is transmitted through a separate channel, as seen in CNL (link computation).

In order to reduce the effects of noise and distortion as mentioned above and improve the quality of zero-crossing waves and CSSB, we invented a speech processing method and transmission method using short-time autocorrelation functions ("speech processing method"). : Japanese Patent Application No. 50083139, ``Audio Digital Transmission System'': Japanese Patent Application No. 50-083143, JSSB Transmission System'': Japanese Patent Application No. 50-083144, both filed on July 8, 1975).

These basic principles are based on SPAC (Speech P
processing system by use of
AutoCorrelation function
) (Masashi Suzuki: "Speech processing method using autocorrelation function - SPAC-J, Journal of the Institute of Electronics and Communication Engineers, Vol. J59A, No. 5, May 1978)."

When SPAC is applied to zero-crossing waves and CSSH, the noise level is reduced by 10 dB or more, making it very easy to hear, but it is still unpleasant to the ear and is not necessarily satisfactory.

The present invention mixes a pulse waveform of a periodic pseudo-random sequence of a certain level with an audio waveform signal, converts it into a zero-crossing wave or CSSB, transmits it, performs processing using an autocorrelation function on the receiving side, It is characterized by reproducing audio waveform signals, and its purpose is to transmit (record and reproduce) audio waveform signals at the highest possible 8N ratio.

Since the present invention utilizes the properties of an autocorrelation function (hereinafter referred to as "R"), first, the characteristics of R related to the present invention will be listed.

(4) The frequency component of the periodic wave is stored in R of the periodic wave.

(B) In the periodic wave R, the fundamental period can be easily detected because the phases of certain parts of each frequency match at the origin.

(Q The amplitude difference of R of a random waveform is concentrated near the origin.

(C) For quasi-periodic signals that change over time, such as speech, a short-time autocorrelation function (hereinafter referred to as "S") with a limited integration time is used in the same way as R.

(D) R of a periodic pseudo-random sequence (hereinafter referred to as "PN sequence") has values at the origin, period, and multiples thereof, and the amplitude value at other times is almost O.

Further, S, whose integration time is equal to the period of the PN sequence and its multiple, becomes the same function as R.

Note that when the audio waveform signal sampled at the sampling period ■ is represented by a, an example of the formula for calculating S is shown in equation (1).

Here, j corresponds to the delay time, and the time interval is ■.

Also, the product-sum time of S (corresponding to the integration time of a continuous system) L・I
(L is an integer).

The present invention will be explained below with reference to the drawings.

FIG. 1 is a block diagram of an embodiment of the first invention, in which 21 represents a random pulse oscillator, 22 a mixing circuit, 23 a binarization circuit, 24 a modulation/transmission/demodulation process, or a recording/reproduction process. 25 is a sampling circuit, 26 is a memory circuit, 27 is a correlator, 28 is a waveform editor, 29 is a sampling pulse oscillator,
20 is an input terminal, and 40 is an output terminal.

Figures 2 and 3 are diagrams showing the waveforms of each part in Figure 1.
is an audio waveform signal, 2 is a zero crossing wave (a,) of 1, 3 and 4 are S waveforms of 2, 5 is an audio waveform signal cut out from S and connected, while 6 is a PN series pulse waveform, 7 is a low-level human noise waveform, 8 is a mixed waveform of 6 and 7, 9 is a zero crossing wave of 8, and 10 is an S waveform of 9.

The numbers 1 to 10 in FIG. 1 indicate the waveform numbers in FIGS. 2 and 3.

The audio waveform signal 1 applied to the input terminal 20 is added to the PN series pulse waveform 6 of period N generated by the random pulse oscillator 21 in the mixing circuit 22 .

Generally, the amplitude of No. 6 is lower than No. 1 by 20 dB or more, so the output waveform of No. 22 is also almost the same as No. 1.

The output of 22 is led to a binarization circuit 23 and converted into a zero-crossing wave 2.

After this zero-crossing wave passes through a modulation, transmission, and demodulation (or recording/reproduction) process 24, a sampling circuit 25 generates a sampled signal a. is converted to

i indicates the sampling number.

Note that the method 24 may be of any type, such as FM, SSB, or AM, and may be directly connected.

a. is the same as 2 as a waveform, and the sequential memory circuit 26
is memorized.

The a read from 26 starting at time t is sent to the correlator 27, and the short-time autocorrelation function St(
Figure 2 3) is calculated.

3 is sent to the waveform editor 28, and first the period T1 of the waveform is determined using the peak value of 81.

This process is extremely easy due to the feature (B) of R.

Next, T1' corresponding to one period T1 is cut out from S1 at a sampling period ■ and sent to the output terminal 40.

Since S has the feature of R (0), use of information near the origin (j=o in equation (1)) is avoided to remove the noise level.

In the example shown in FIG. 2, S is cut out from a point close to the origin, where the amplitude value is 0 and its differential coefficient is positive.

On the other hand, the information on T1 determined in 28 is passed through 31 to 26
, move the sample for T1, and change the starting point to t2-t1+
27a as T1. send.

As a result, at 27, 82 (FIG. 2, 4) is calculated for t2, and at 28, the period T2 is determined and a sample T2' corresponding to T2 is cut out.

The next S3 is calculated as t3 two t2 + T2 (0).

These operations are repeated, and at 40 T1', T2
' , T7, . . . , the audio waveform signal shown in FIG. 25 is obtained, in which one cycle of S continues.

By the way, when the voice is paused, what is applied to 20 is a low-level noise waveform.

This is shown in FIG. 37.

The random pulse oscillator 21 generates a PN sequence pulse waveform 6 with a period N.

6 and 7 are mixed in a mixing circuit 22.

The mixed waveform is shown in FIG. 38.

At this time, if the amplitude of 6 is thicker than the amplitude of 7, then 8
The zero-crossing wave of is as shown in Figure 3, 9, which is the same P as 6.
This is an N series of pulse waveforms.

As mentioned in 9's S is the characteristic of R), the period of 9 and (
1) When the product-sum times in the equation are made to match, the values other than the periods of J20 and 9 are almost 0.

In FIG. 3, 6 takes as an example a PN sequence with a code interval of ■, N231, and since M.■ is more dog-like than N.■, the S of 9 is 1.
It has a peak at N・■ like 0.

Moreover, the amplitude value of S is almost O at other points.

Note that in actual processing, the peak value in the period of the PN series is unnecessary, so N・■ is set larger than M・■, and S
Avoid peaks.

In other words, if processing is performed with the product-sum time (L・■) equal to the period of the PN sequence and M・■ set to be shorter than the period of the PN sequence, the amplitude of the waveform applied to 20 will be 6. When the amplitude is smaller than , the waveform is completely suppressed at 40, and when the periodic waveform has an amplitude larger than 6, a waveform as shown in FIG. 2 is obtained.

The output audio waveform signal obtained in this manner has a better signal-to-noise ratio and is much easier to hear than a zero-crossing wave.

Note that when determining the period T from S, the peak of S is used, but when no peak is observed as in the case of a PN sequence, T is set to a constant value.

The period of the PN sequence is determined by taking into consideration the characteristics of the voice and the PN sequence, and a period of 20 to 30 ms is appropriate.

Furthermore, if the code interval of the PN sequence is made to match the sampling period {circle around (2)} as in the example shown in FIG. 3, distortion due to non-synchronization between the two can be prevented.

In this case, if L=N, the product-sum time and the period of the PN sequence become equal.

Note that even if the product-sum time does not match the period of the PN sequence, sufficient practical effects can be achieved as long as the two are close.

Examples of the PN sequence include an M sequence, an L sequence, and a twin prime number sequence.

Although any series may be adopted, the hardware for generating the M-sequence signal can be manufactured extremely easily.

Next, the second invention will be explained.

FIG. 4 is a block diagram of an embodiment of the second invention, in which numeral 33 represents a carrier suppression SSB modulator, numeral 34 represents an amplitude limiter, and numeral 35 represents a process of transmission, transmission, and demodulation.

Note that blocks having the same numbers as those in FIG. 1 have the same functions as in FIG. 1.

FIG. 5 is a diagram showing the waveforms of each part in FIG.
2 is the CSSB waveform, and 14 is the demodulated waveform of 13.

The audio waveform signal 11 applied to the input terminal 20 is a PN series pulse waveform (
6) in FIG. 3 in a mixing circuit 22.

In general, the amplitude of 6 is 20 dB compared to the amplitude of 11.
Since it is set lower than this, the output waveform of 22 is almost the same as that of 11.

The output of 22 is modulated by a carrier suppression SSB modulator 33 and becomes 12.

Further, the waveform 12 is shaped into a waveform of a constant amplitude by the amplitude limiter 34 to become the waveform 13.

The function of 34 is substantially the same as that of the binarization circuit 23 in FIG.

For the waveform 14 obtained by transmitting and demodulating the waveform 13, the same correlation processing as that applied to the zero-crossing wave 2 in the method of the invention shown in FIG. 1 is performed in blocks of 25 or less.

The operation is the same as that explained using FIG. 1, so the explanation will be omitted.

Regarding noise, as shown in Fig. 3, 20
If the amplitude of the noise waveform 7 added to is smaller than the amplitude of the PN series pulse waveform 6, the output of 34 is the CSSB of the PN series pulse waveform.

When this CSSB is demodulated and S is calculated, the result is 10 in Figure 3.
Almost the same waveform is obtained.

Therefore, when transmitting audio using CSSB, the system of the second invention can suppress noise during pauses.

In the first and second inventions, it is necessary to appropriately set the amplitude level of the PN series pulse waveform 6.

If this is thicker than the noise level during pause, the noise can be completely suppressed in the output, but if the amplitude of 6 is made too large, the quality will deteriorate.

Further, if the amplitude of 6 is made small, noise suppression becomes insufficient.

Practically, an appropriate amplitude is about -40 dB with respect to the maximum amplitude of the audio waveform signal (1 or 11).

Next, an embodiment of the method of the present invention will be described.

For a voice waveform signal with an S/N ratio of approximately 40 dB and a maximum amplitude of 1000, the amplitude is 10 and the code interval is 0.1 ms, N = 2.
After applying 55 M-sequence signal pulses, I = 0.
It was converted into a zero-crossing wave in 1 ms.

This corresponds to 10 kbps digital transmission.

When this zero-crossing wave was processed with L=255 and M2170, the noise during pauses was completely suppressed, and the distortion in the voice section was also reduced, making it very easy to hear.

As described above, according to the transmission processing method of the present invention, 10k
Digital voice communication at bps level and efficient SSB
Because it enables communication and calls with higher quality voice than conventional methods, it can be widely used in mobile radio, satellite communication, long-distance voice communication, etc., and can also be used for recording voice response devices.

[Brief explanation of drawings]

1 and 4 are block diagrams of an embodiment of the present invention, and FIGS. 2, 3, and 5 are waveform diagrams. 20...Input terminal, 21...Random pulse oscillator, 22...Mixing circuit, 23...
...Binarization circuit, 24...Modulation/transmission/demodulation (recording/reproduction), 25...Sampling circuit, 26.
...Memory circuit, 27...Correlator, 28.
... Waveform editor, 29 ... Sampling pulse oscillator, 3-3 ... Carrier suppression SSB modulator,
34... Amplitude limiter, 35... Transmission
Transmission/demodulation, 40... Output terminal, ■...
- Sampling period, T... Period of waveform.

Claims (1)

[Claims] 1. After mixing the audio waveform signal and the PN sequence pulse waveform, converting it into a zero-crossing wave, transmitting (or recording/reproducing) this, and then calculating the short-time autocorrelation function 81 at time t1. Then, determine the period T1, cut out one period of waveform T1' from S1 avoiding the vicinity of the origin, and then cut out the waveform T1' of one period avoiding the vicinity of the origin, and then at time t2-t
1+T1, calculate the short-time autocorrelation function 82, determine the period T2, cut out the waveform T2' of one period from S2, connect it to the previously cut out T, /, and then t3-t2+T2
Repeat the process of cutting out T3' and connecting T3' next to T2' to obtain T,', T2
' , T3' , . . . are successively performed to suppress noise in the pause section and obtain an audio waveform signal of excellent quality. 2. Mix the audio waveform signal and the PN sequence pulse waveform, then modulate this waveform into a carrier-suppressed single sideband wave, transmit it with a constant amplitude, and demodulate the waveform signal at the receiving side at time t. Calculate the short-time autocorrelation function 81, determine its period T, cut out one period of the waveform T1' from S1 while avoiding the vicinity of the origin, and then calculate the short-time autocorrelation function 82 at time t2=t1+T1. Then, determine the period T2, cut out one period of waveform T2' from S2, connect it to the previously cut out T1', then perform the same process for t32+T2, cut out T3', and connect T3' to T2.
' Repeat the process of connecting next to T1', T2', T3
A voice transmission processing method characterized by suppressing noise in the pause section and transmitting a voice waveform signal of excellent quality by making the pause interval continuous.