JPH05175853A - Error correcting encoding system for voice signal - Google Patents

Abstract

PURPOSE:To reduce rendundancy caused by error correction encoding and to improve frequency utilizing efficiency by continuously encoding voice signals in respective classes by one convolutional encoder and afterwads, puncturing the signals at various thinning rates corresponding to error sensitivity. CONSTITUTION:A voice encoder 2 compresses and encodes a voice signal 8, the signals are divided into two classes corresponding to the error sensitivity, and a digital voice signal 9 having high error sensitivity and a digital voice signal 10 having low error sensitivity are outputted. A switch 3 prepares a signal 11 by synthesizing the signals 9 and 10 and supplies the signal 11 to an error correction encoder 4. At the error correction encoder 4, a tail bit is added to the end of the signal 11 only once, and the signal is convolutionally encoded at the same encoding rate. An output signal 12 of the encoder 4 is supplied to a puncture circuit 5, and prescribed bits are erased according to puncture patterns various for each class. A punctured signal 13 is outputted from an output terminal 7.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an error correction coding system for a digital voice signal compressed by a high efficiency code.

[0002]

2. Description of the Related Art In recent years, in communication systems for transmitting voice such as car telephones and mobile telephones, in order to increase the number of channels per band, voice signals are being compressed by high-efficiency coding to reduce the bit rate. .. Along with this, the error sensitivity of each bit of the voice signal becomes high, and therefore it is necessary to protect the voice signal from a course error by using an error control code. In such an audio signal, depending on the bit, the degree to which the quality of the reproduced signal deteriorates when the bit is incorrect differs. This is called the error sensitivity of each bit. In order to effectively use the frequency band, it is desirable to set the correction capability of the error correction code according to the error sensitivity of each bit.

Therefore, a conventionally proposed error correction method is a method in which a voice signal is divided into classes according to error sensitivity and each class is coded with an error correction code having a different coding rate independently. FIG. 3 shows a schematic configuration diagram of this system.

In the figure, the audio signal 7 is input from the input terminal 1.
Is input and supplied to the speech encoder 2. The speech coder 2 compresses the speech signal 7 and classifies it into a bit string 8 having a high error sensitivity, that is, a large deterioration in speech quality due to an error, and a bit string 9 having a low error sensitivity, that is, a small deterioration in a speech quality due to an error. Then, each is supplied to the error correction encoders 3 and 4. Then, the error correction encoder 3 performs error correction coding on the bit string 8 having a high error sensitivity with the error correction code having the coding rate r ₁ , and the error correction encoder 4 performs the coding rate r ₂ (r ₁ <r ₂ ). Bit 9 having a low error sensitivity is subjected to error correction coding by the error correction code of.

Generally, the lower the coding rate, the higher the error correction capability. The coded signal sequences 10 and 11 output from the error correction encoders 3 and 4 are combined into one signal sequence 12 by the switch 5 and output from the output terminal 6. In this method, r ₁ is set lower than r _2, so that a bit string having a high error sensitivity can be protected more strongly than a bit string having a low error sensitivity. When the redundancy is distributed unevenly in this way, the frequency band can be effectively used.

On the other hand, the convolutional code has already been used as an error correction code for a voice signal, since strong error correction capability can be obtained by performing maximum likelihood decoding such as Viterbi decoding. For example, in the system adopted in the US digital cellular system, a convolutional code with a coding rate of 1/2 is applied to a plurality of bits (called class 1 bit) having high error sensitivity in a compressed digital voice signal. Applied. In this system, the error correction code is not applied to the bits with low error sensitivity. If a digital audio signal is encoded at a lower bit rate than this method,
Since the absolute value of the error sensitivity of each bit becomes high, it becomes necessary to apply some error correction code even to the bit having a relatively low error sensitivity in the encoded voice signal.

By the way, when information to be encoded such as a voice signal is input to the convolutional encoder in block units, it is desirable that the trellis of the convolutional code be terminated for each block. That is, encoding is started with the shift register of the encoder being cleared so that the beginning and the end of the trellis are merged into a specific state, and m bits (m is the memory length of the encoder) are set as information bits. It is necessary to add all zero tail bits for encoding. Encoding ends when all tail bits have entered the encoder. At this time, if the coding rate r of the convolutional code and the number of information bits are K, the length of the coded data is (K + m)
/ R bits, and the effective coding rate is r × K / (K + m), which is smaller than r.

For example, if r = 1/2, K = 50, and m = 6, the encoded data has 112 bits,
The substantial code rate is 50/112, which is smaller than the code rate 1/2 of the convolutional code. As described above, when convolutional coding of block data is performed, particularly when the block length K is small, the actual coding rate becomes smaller than the coding rate of the original code, that is, the redundancy increases. , The frequency utilization efficiency decreases.

Since the convolutional code has such a property, when the convolutional code is used as the error correction code in the circuit shown in FIG. 3, two information blocks each having a short block length are independently convolutionally coded. This leads to a problem that the redundancy for terminating the trellis becomes large. In a system where the frequency band is limited, it is necessary to increase the coding rate of the convolutional code or reduce the number of bits of the speech signal by this amount, and in either case, the quality of the decoded speech deteriorates.
In addition, if the number of classifications according to the error sensitivity of the voice signal is increased, the block length is further shortened and the number of blocks is increased, so that there is a problem that redundancy is further increased and frequency utilization efficiency is reduced.

[0010]

As described above, in order to effectively use the frequency band, when the digital voice signal is subjected to error correction coding, the voice signal is classified into classes according to the error sensitivity of each bit. It is desirable to code with error correction codes having different coding rates. However,
When the convolutional code is used as the error correction code, the redundancy for terminating the trellis is required in each encoder, and there is a problem that the frequency utilization rate is lowered. Also, in a system where the frequency band is limited, it is necessary to lower the coding rate of the convolutional code or reduce the number of bits of the voice signal by that amount,
There was a problem of degrading the voice quality.

The present invention has been made to solve such a conventional problem, and an object of the present invention is to provide an error correction coding system for a voice signal capable of reducing redundancy by error correction coding. Especially.

[0012]

In order to achieve the above object, an error correction system for a voice signal according to the present invention divides a voice signal composed of a plurality of bits into a plurality of classes according to the error sensitivity of each bit, and The speech signal of a class is continuously input to the same convolutional encoder for each class, and the output signal of the convolutional encoder is punctured with a different punching pattern for each class.

Further, the above error correction method is characterized in that a puncture pattern is used for the output of the convolutional encoder so that the puncture decimation rate is in ascending or descending order.

[0014]

In the error correction coding system of the present invention, the speech signals of each class are continuously coded by one convolutional coder and then punctured at different decimation rates depending on the error sensitivity. Therefore, in the encoder, it is not necessary to terminate the trellis for each class, and the redundancy can be made smaller than in the conventional method using the convolutional code having the same coding rate. In addition, since the thinning rate for puncturing can be set arbitrarily, it is easy to increase the number of classes and change the coding rate that is flexible with respect to error sensitivity.

[0015]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of an error correction circuit to which an error correction coding method for a voice signal according to the present invention is applied.

In FIG. 1, a voice signal 8 before compression is input to a voice encoder 2 from an input terminal 1. The voice encoder 2 compression-codes the voice signal 8, divides it into two classes according to the error sensitivity, and outputs a digital voice signal 9 having a high error sensitivity and a digital voice signal 10 having a low error sensitivity. Here, the K ₁ bit number of the signal 9 is {I ₁ ,
I ₂ , ..., I _K1 }, the K ₂ bit of the signal 10 is {I
_{K1 + 1} , I _{K1 + 2} , ..., I _{K1 + K2} }.

The switch 3 puts the signals 9 and 10 together,
Continuous K ₁ + K ₂ bit signal 11 {I ₁ , I ₂ ,
, I _K1 , I _{K1 + 1} , I _{K1 + 2} , ..., I _{K1 + K2} },
It is supplied to the error correction encoder 4. The error correction encoder 4 adds m bits (m is the memory length of the encoder) tail bits to the end of the signal 11 only once, and performs convolutional coding at the same coding rate. Now, assuming that the coding rate of the convolutional encoder is 1/2, the output signal 12 of the error correction encoder 4 has 2 × (K ₁ + K ₂ + m) bits. Here, the input I _j to the encoder 4 (j = 1, 2, ..., K ₁ +
K ₂ + m. However, the 2-bit output signals corresponding to I _j = 0, j ≧ K ₁ + K ₂ +1) are C _{j, 0} and C _{j, 1} . The output signal 12 of the encoder is supplied to the puncture circuit 5, and predetermined bits are erased according to the puncture pattern of FIG.

The position of the bit erased by the puncture circuit 5 is controlled by the signal given from the puncture bit control circuit 6. That is, the coded bit at the position of "0" in FIG. 2 is erased. As a result, the signal of the portion corresponding to class 1 is convolutionally coded at the coding rate r ₁ = 2/3, and the signal of the portion corresponding to class 2 is convolutionally coded at the coding rate r ₂ = 3/4. Then, the punctured signal 13 is output from the output terminal 7.

In the above embodiment, the number of bits of the class 1 speech signal K ₁ = 30, the number of bits of the class 2 speech signal K ₂ = 30, the memory length m of the encoder m = 6, and the error correction for the class 1 coding rate r ₁ = 2/3, if the coding rate r ₂ = 3/4 error correction for the class 2, the number of bits of the error correction coding the signal after 13, 30 × (3/2)
+ (30 + 6) × (4/3) = 93 bits. On the other hand, when each class is encoded independently by a convolutional code with a different coding rate as in the conventional error correction method, when the number of bits and the coding rate of the voice signal are the same, The number of bits is (30 + 6) × (3/2) +
(30 + 6) × (4/3) = 102 bits. As described above, according to the present embodiment, the redundancy can be made smaller than in the conventional case even if the coding is performed at the same coding rate.

Further, FIG. 4 shows that in the above embodiment, the number of bits K ₁ of class ₁ is 62, the number of bits K 2 of audio signal of class ₂ is 64, the memory length m of the encoder is 6, and class 1 is 1. The error rate after decoding of each speech bit when the error-correction coding rate r ₁ is about 0.54 and the error-correction coding rate r ₂ for class ₂ is about 0.75. It is shown. In FIG. 4, the error rate of the bit of class 1 is around 0.003 and the error rate of the bit of class 2 is around 0.1. As a result of using the error correction method of the present invention, the error correction is performed according to the error sensitivity. It is clearly shown that it is possible to make a difference in abilities. However,
The reasonably low error rates for the first few bits and the last few bits are due to the fact that the first and last states of the trellis are terminated in specific states. The reason why the error rate gradually changes between class 1 and class 2 is that the Viterbi decoder uses a cumulative likelihood from the past called a path metric, and the truncation length (path memory depth, here 60
This is because the process of determining the previous decoded bit by (bit) is performed. Due to such a property of the Viterbi decoder, the decoding error rate does not suddenly change according to the change of the puncture decimation rate.

In the above embodiment, the class of the digital audio signal is divided into two, but it can be divided into more classes. The larger the number of classes, the greater the improvement rate of redundancy compared with the conventional method. However, when the class is divided into three or more, it is possible to effectively differentiate the error rate after decoding by setting the puncturing decimation rate in ascending or descending order. This is because, due to the nature of the Viterbi decoder described above, the decoding error rate does not change abruptly even if the puncture decimation rate is changed. Because it will be affected.

Further, the coding rate of the error correction encoder 4 in FIG. 1 is made lower (for example, 1/4), and a part of the coding sequence is erased in the puncturing circuit, so that the coding rate becomes 1. You can also make a / 3 or 1/2 part.

[0023]

As described above, according to the present invention, when a speech signal is classified into classes according to the error sensitivity and each of them is subjected to error correction coding with a convolutional code having a different coding rate, all classes are classified. It is possible to reduce the redundancy by error correction coding by puncturing different speech signals according to the class after continuously coding the speech signal of 1 by one convolutional code. ..

[Brief description of drawings]

FIG. 1 is a configuration diagram showing an error correction circuit to which the system of the present invention is applied.

FIG. 2 is an explanatory diagram showing a puncture pattern of the puncture circuit shown in FIG.

FIG. 3 is a configuration diagram showing a conventional voice encoding and error correction circuit.

FIG. 4 is a characteristic diagram showing an error rate after decoding a voice signal according to the error correction method of the present invention.

[Explanation of symbols]

 2 speech encoder 3 switch 4 error correction encoder 5 puncture circuit 6 puncture bit control circuit

Claims (2)

[Claims] 1. A speech signal composed of a plurality of bits is divided into a plurality of classes according to the error sensitivity of each bit, and the speech signals of each class are continuously input to the same convolutional encoder for each class, and An error correction coding method for speech signals, characterized in that the output signal of a convolutional encoder is punctured with a different punching pattern for each class. 2. With respect to the output of the convolutional encoder,
2. The error correction coding system for a voice signal according to claim 1, wherein the puncture pattern is a puncture pattern in which the thinning rate of the puncher is ascending or descending.