CN100452693C - AMR method for effectively guaranteeing speek voice quality in wireless network - Google Patents

Abstract

The present invention relates to the field of mobile communication, particularly to a method for effective guarantee of quality of speech in a wireless network. The method is characterized in that at the sending end, SNR is used as an index for balancing the quality of a channel; when the SNR is smaller than a threshold value namely the condition of the channel is poor, coding is carried out according to a lowest mode of ARM, the bit in the type A of the core frame of the AMR is kept to be invariant, and then the bit in the type B of the core frame of the AMR is compressed; in this way, the bit number of the type B is reduced, redundant bit number is added, and thus, the erroneous frame ration in transmission is lowered; when the SNR is bigger than the threshold value, the coding is carried out according to a standard AMR method. Besides, a new frame type is proposed, and the index of the frame type is 12; at the receiving end, if the type index value of the frame received is 12, the decoding is carried out according to the method of the present invention; if the type index value of the frame received is not 12, the coding is carried out according to the standard AMR method.

Description

A kind of in wireless network the AMR method of effectively guaranteeing speek voice quality

Technical field

The present invention relates to moving communicating field, be specifically related to a kind of in wireless network the AMR method of effectively guaranteeing speek voice quality.

Background technology

Adaptive multi-rate speech encoding and decoding (AMR) have become the voice compression in GSM and the W-CDMA system.It can support 8 kinds of different patterns, and code rate is from 4.75kbps to 12.2kbps.AMR can dynamically switch code rate according to channel quality, can improve the utilance of resource like this.When bad channel quality, adopt low code rate, the redundant bit in the chnnel coding will increase like this, thereby to the information better protection; When channel quality is good, can adopt high code rate to improve the quality of voice.But when channel quality was very poor, speech had very high frame error rate in transmission, had a strong impact on the quality of speech.In this case, promptly use minimum code rate pattern 4.75kbps to carry out encoding and decoding speech, speech quality still can not get guaranteeing.In order under this condition, still to carry out speech transmissions, must further reduce code rate, the raising redundant bit reduces the frame error rate in the transmission.

According to 3GPP TS 26.101, the AMR frame is divided into three parts, i.e. AMR header, AMR supplementary and AMR core frames.Comprise the indication of frame type and frame quality in the AMR header; Comprise the pattern indication in the AMR supplementary, mode request and encoder CRC check; The AMR core frames comprises speech parameter, and according to the importance of parameter, these parameter bits are divided into three type: type A, type B and Type C.If comfortable background noise frames, these parameters are parameters of comfortable background noise, all are placed among the type A.

Bit among the type A is the most responsive to mistake, and in these bits make mistakes in any position, all must adopt suitable error concealment mechanism to handle after decodable code, and this type bit will be protected with encoder CRC.If the bit in type B and the Type C is made mistakes, speech quality reduces, but subjectivity is an acceptable, and the speech frame of mistake also can direct decoding.

Though above standard A MR encoding and decoding speech method comes into operation, when channel quality was very poor, voice quality can not get guaranteeing, can not realize proper communication.

Summary of the invention

(1) technical problem that will solve

The purpose of this invention is to provide when channel quality is very poor, still can realize proper communication a kind of in wireless network the AMR method of effectively guaranteeing speek voice quality

(2) technical scheme

In order to achieve the above object, the present invention takes following method step:

1) at transmitting terminal, at first with signal to noise ratio (snr) as the index of weighing channel quality, when signal to noise ratio (snr) during less than a threshold value, when promptly channel condition is very poor, in order to guarantee the transmission quality of speech, encode earlier, keep the bit among the type A of AMR core frames constant, compress the bit in the type B of AMR core frames then by lowest mode among the AMR, by reducing the bit number in the type B, increase the redundant bit number, reduce the frame error rate in the transmission, thereby improve the quality of speech; When signal to noise ratio (snr) during, still encode according to standard A MR method greater than threshold value; At receiving terminal, be not 12 o'clock if receive the types index value of frame, still decipher according to standard A MR method.

2) propose a new frame type, the frame type index is 12; At receiving terminal, be 12 o'clock if receive the types index value of frame, explanation is a frame type after treatment, in order to make the decoder correct decoding, need recover compressed information, then according to the 4.75kbps mode decoding of standard; If receiving the types index value of frame is not 12 o'clock, still decipher according to standard A MR method.

Wherein, the method for the bit in the type B of the AMR of compression described in step 1) core frames comprises: for the information of 3 bits, the pulse position in the fixed codebook is divided into two groups: 000,001,010,011 and 100,101,110,111; At transmitting terminal,,, pulse position comes the position of index pulse if in second group, transmitting 1 if pulse position in first group, only transmits 0 position that indicates this pulse; At receiving terminal,, think that then the position of original pulse is 010 if the indicating bit of receiving is 0; If the indicating bit of receiving is 1, think that then the position of original pulse is 110; For 2 bit informations, be divided into two groups: 00,01 and 10,11; At transmitting terminal, if pulse position in first group, then indicating bit is set to 0, if pulse position in second group, indicating bit is set to 1.At receiving terminal,, think that then these two is 01 if indicating bit is 0; If indicating bit is 1, then think 10.

Wherein, described coding comprises the convolution code that provides in described TS 25.212 standards of employing 3GPP as chnnel coding, after having encoded, for satisfied fixing transmission rate, carries out rate-matched again.

Wherein, chnnel coding adopts rate-matched deletion convolution code (RCPC).

(3) beneficial effect

1) owing to obtain protection to type A bit by the bit in the type B that reduces the AMR core frames, the present invention still can effectively guaranteeing speek voice quality for channel quality when very poor.

2) easy and AMR operating such.

Description of drawings

Fig. 1 is a chnnel coding block diagram of the present invention;

Fig. 2 is a workflow diagram of the present invention;

Fig. 3 is 4.75kbps pattern of the present invention and 3.60kbps pattern and does not have under the punching situation frame error rate with the change curve of signal to noise ratio;

Fig. 4 is the MOS value comparison diagram of the present invention's 4.75kbps pattern and 3.60kbps pattern under different signal to noise ratios.

Among the figure: SNR, signal to noise ratio; FER, frame error rate; MOS, Mean Opinion Score; Bits, bit; Classes ordering classification and ordination; CRC-8,8 bit cyclic redundancy; RCPC, rate-matched deletion convolution code; Rate Matching, rate-matched; MUX, multiplexing; TX, transmitting terminal; RX, receiving terminal.

Embodiment

Following examples are used to illustrate the present invention, but are not used for limiting the scope of the invention.

In order fully disclose content of the present invention, before specific implementation method of the present invention is described, the principle of description standard AMR encoding and decoding speech method at first.

The basic principle of AMR encoding and decoding speech method is: the 8kHz that is input as of encoder samples, the linear PCM coding of 16 bit quantizations, and encoding operation is a frame with the voice of 20ms, i.e. 160 sample points.Encoder extracts the parameter of algebraic codebook Excited Linear Prediction (ACELP).These parameters comprise parameter, the adaptive codebook of linear prediction filter (LP), the index and the gain of fixed codebook.Be transmitted after these parameter codings.In decoder end, these parameters are extracted from the bit stream that receives, and construct composite filter and pumping signal then, and the reconstruct voice also will and carry out ratio and amplify through postfilter.

Represent frame type with 4 bits in the AMR core frames, 16 kinds of states altogether, promptly in addition 8 kinds of AMR speech coding patterns and 4 kinds of comfortable background noise patterns and empty frame, also have 3 frame type index to be retained standby.When channel condition is good, adopt the higher pattern of code rate to improve speech quality; And when bad channel conditions, adopt the lower pattern of code rate to guarantee the quality of speech.Yet, when channel condition is very poor, adopt minimum rate mode 4.75kbps can not guarantee the quality of speech.

Below specific implementation method of the present invention is described.

In order to guarantee still can guarantee the transmission of certain mass under this condition, we propose a new frame type, and the frame type index is 12, and this is to reserve the frame type index in the standard.The bit of type A is very sensitive to mistake in the AMR core frames, as long as wrong in these bits, decoding end must be hidden by wrong frame and handle so.These bits must obtain better protection.And only can reduce the quality of speech for the mistake of the bit in type B, but subjective impression can receive.Therefore, we can obtain the protection to type A bit by the bit in the minimizing type B.

As shown in Figure 2, at transmitting terminal, at first with the index of signal to noise ratio (snr) as the measurement channel quality, when signal to noise ratio (snr) during less than a threshold value, be channel condition when very poor,, encode by lowest mode among the AMR earlier in order to guarantee the transmission quality of speech, bit among the type A of maintenance AMR core frames is constant, compress the bit in the type B of AMR core frames then,, increase the redundant bit number by reducing the bit number in the type B, reduce the frame error rate in the transmission, thereby improve the quality of speech; When signal to noise ratio (snr) during, still encode according to standard A MR method greater than threshold value; At receiving terminal, be 12 o'clock if receive the types index value of frame, explanation is a frame type after treatment, in order to make the decoder correct decoding, need recover compressed information, then according to the 4.75kbps mode decoding of standard; If receiving the types index value of frame is not 12 o'clock, still decipher according to standard A MR method.

Described snr threshold is to determine by the voice quality after the actual channel through the voice that relatively use the 4.75kbps pattern in the inventive method and the standard.

1) processing method of bit in the type B

When channel condition is very poor, can utilize lowest mode 4.75kbps pattern to encode earlier, compiled that the bit among the type A remains unchanged after the sign indicating number, and the bit in the type B to be further processed.

When the 4.75kbps coding mode, mostly the bit in the type B is the index of fixed codebook.The fixed codebook of every frame is made up of 2 pulses.The position of pulse is represented by 3 bits.We are divided into two groups with the position: 000,001,010,011 and 100,101,110,111.At transmitting terminal, if pulse position in first group, we only transmit 0 position that indicates this pulse, come the position of index pulse if pulse position in second group, transmits 1.At receiving terminal,, think that then the position of original pulse is 010 if the indicating bit of receiving is 0; If the indicating bit of receiving is 1, think that then the position of original pulse is 110.For the information of 2 bits in the type B, such as the adaptive codebook index back two can be compressed equally.The position is divided into two groups: 00,01 and 10,11; At transmitting terminal, if pulse position in first group, then indicating bit is set to 0, if pulse position in second group, indicating bit is set to 1.At receiving terminal,, think that then these two is 01 if indicating bit is 0; If indicating bit is 1, then think 10.By this processing method, the bit number in the type B reduces to 30 by original 54.The bit number of whole like this AMR core frames is reduced to 72 by 95, thereby code rate is reduced to 3.60kbps by 4.75kbps.

After handling like this, gain need be readjusted.Because the index of fixed codebook gain and code book has relation.Fixed codebook gain is obtained by following formula:

$g_c=\frac{x_2^{T}z}{z^{T}z}---\left(1\right)$

Wherein x2 is the target vector of fixed codebook search, is determined by following formula:

$x_2\left(n\right)=x\left(n\right)-{\hat{g}}_{p}y\left(n\right)---\left(2\right)$

Z is the convolution of the impulse response of fixed codebook vector and perceptual weighting filter

$z\left(n\right)=\underset{i=0}{\overset{n}{\mathrm{\Σ}}}c\left(i\right)h(n-i)n,(n=\mathrm{0,1},\·\·\·,39)---\left(3\right)$

Wherein, c (i) is a fixed codebook vector.

Above-mentioned compression method mainly is at fixed codebook vector, so the gain of fixed codebook need be upgraded according to new codebook vectors.In order to reappraise gain, c (i) will use c (i) ' to substitute, and c (i) ' is the value that c (i) is resumed at receiving terminal.

2) processing in the chnnel coding

Three kinds of channel coding methods have been provided among the 3GPP TS 25.212: 1/2 convolution code, 1/3 convolution code and 1/3Turbo sign indicating number.Convolution code is applied in the business of middle low rate; And Turbo code is applied in the data service at a high speed.Because speech business belongs to low rate traffic, so we adopt convolution code as chnnel coding.Stipulated in the standard that constraint length is the structure of 8 convolution coder.For 1/3 convolution code, three output octal representations are 557,663 and 711; For 1/2 convolution code, two output octal representations are 561 and 753.

As shown in Figure 1, the bit of AMR encoder output need be resequenced according to importance.To add CRC check to type A afterwards, to determine erroneous frame whether occurs in the transmission.The generator polynomial of CRC is:

G(x)＝D ⁸+D ⁶+D ⁵+D ⁴+1

Bit in the AMR frame is divided into 3 types from the above, is the bit the most responsive to mistake among the type A, and in these bits make mistakes in any position, could decode after all must adopting wrong frame hiding, and this class bit carries out verification protection by CRC; Bit is made mistakes in type B and the Type C, can influence the quality of decoding back speech, but the hearer subjective be acceptable.Therefore we need carry out unequal error protection according to the otherness of type importance.In addition, after having encoded,, need carry out rate-matched in order to satisfy fixing transmission rate.Rate-matched often adopts punching or repeats two kinds of methods.

The present invention adopts rate-matched deletion convolution code (RCPC) as chnnel coding, with the deletion matrix each type bit is realized unequal error protection.Some documents have provided the optimum RCPC family of different rates convolution code, and the present invention utilizes these optimum RCPC families exactly, and the design code rate is the deletion matrix of 3.60kbps.As list of references: P.K.Frenger, P.Orten, T.Ottosson and A.B.Svensson, " Rate-compatible convolutional codes formultirate DS-CDMA systems, " IEEE Trans.Commun., vol.47, pp.828-836, Jun.1999.

Below in conjunction with accompanying drawing effect of the present invention is analyzed.

For the validity of verifying, this paper analyzes the 3.60kbps pattern that proposes and the performance of the 4.75kbps pattern among the standard A MR under additive white Gaussian noise (AWGN) channel and compares.

At first we have compared both frame error rate during different SNR under the awgn channel.Because the code rate difference of two kinds of patterns, we carry out chnnel coding at application rate coupling deletion convolution code, and the parameters of rate matching of two kinds of patterns and coupling matrix see Table 1 and table 2 respectively.Parameters of rate matching is determined according to code rate and channel speed.Deletion matrix in the table 2 uses octal number to represent.

The parameters of rate matching of two kinds of patterns of table 1.

Decoding schema (kbps)	Bit/frame (20 ms)	Type A bit	The type B bit	The Type C bit	Type A rate-matched	Type A rate-matched	The channel-decoding bit
								4.75	95	42	53	0	8/17	8/9	192
3.60	72	42	30	0	8/21	8/9	192

The punching matrix of two kinds of patterns of table 2.

Referring to Fig. 3, in emulation, at first we change into the pulse code modulation (pcm) code stream with voice signal, send into then in the AMR encoder.Encoder output bit is resequenced according to importance, is divided into the RCPC chnnel coding that 3 different types are above described, and carries out the BPSK modulation afterwards.Signal after the modulation sends receiving terminal to through awgn channel.After receiving terminal is received signal, will carry out demodulation earlier, CRC check is carried out in Viterbi (Viterbi) decoding then at last.Need to prove that in order to reduce frame error rate, the Viterbi decoder adopts soft-decision.Fig. 3 has provided, and 4.75kbps pattern and 3.60kbps pattern be the frame error rate during different SNR under awgn channel.For relatively, give the frame error rate of not carrying out under the punching situation among the figure.

Can find out obviously that from figure the performance of 1/3 convolution code of not punching will obviously be better than two kinds of patterns through the punching processing.Because the bit number of 4.75kbps pattern core frames is 95; And the 3.60kbps pattern has only 72 bits, when these two kinds of pattern matching arrive identical channel speed, needs the number of punching just different.4.75kbps the number of the punching that pattern needs is more than the 3.60kbps pattern.So under identical SNR, the frame error rate of 3.60kbps pattern will be significantly less than the 4.75kbps pattern.Satisfy identical frame error rate, the signal to noise ratio that the 3.60kbps pattern needs is than the little 0.5～2dB of 4.75kbps.

The difference of frame error rate will directly influence the speech quality of receiving terminal.In order to weigh the speech quality under these two kinds of patterns, we adopt MOS value commonly used to weigh.We have done a large amount of experiments and have weighed two kinds of patterns MOS value during different signal to noise ratio under awgn channel.The result as shown in Figure 4.

From figure, can obviously find out, signal to noise ratio less than the situation of 3dB under the MOS value of 3.60kbps pattern to be higher than the 4.75kbps pattern, the quality of this explanation the former speech this moment is better than the latter.But difference between the two reduces along with the increase of signal to noise ratio.In this case, the 3.60kbps pattern that adopts this paper to propose can to a certain degree guarantee speek voice quality down.When signal to noise ratio during greater than 3dB, the speech quality of 3.60kbps pattern begins not as 4.75kbps, and should adopt the AMR method of standard to carry out encoding and decoding speech this moment.

Claims (4)

1, a kind of in wireless network the AMR method of effectively guaranteeing speek voice quality, it is characterized in that following steps are arranged:

1) at transmitting terminal, at first with signal to noise ratio as the index of weighing channel quality, when signal to noise ratio during less than a threshold value, when promptly channel condition is very poor, in order to guarantee the transmission quality of speech, encode earlier, keep the bit among the type A of AMR core frames constant, compress the bit in the type B of AMR core frames then by minimum standard 4.75kbps pattern among the AMR, by reducing the bit number in the type B, increase the redundant bit number, reduce the frame error rate in the transmission, thereby improve the quality of speech; When signal to noise ratio during, still encode according to standard A MR method greater than threshold value;

2) propose a new frame type, the frame type index value is 12; At receiving terminal, be 12 o'clock if receive the types index value of frame, explanation is a frame type after treatment, in order to make the decoder correct decoding, need recover compressed information, then according to the 4.75kbps mode decoding of standard; If receiving the types index value of frame is not 12 o'clock, still decipher according to standard A MR method.

2, as claimed in claim 1 a kind of in wireless network the AMR method of effectively guaranteeing speek voice quality, it is characterized in that, the method of the bit in the type B of the AMR of compression described in step 1) core frames comprises: for the information of 3 bits, pulse position in the fixed codebook is divided into two groups: 000,001,010,011 and 100,101,110,111; At transmitting terminal,,, pulse position comes the position of index pulse if in second group, transmitting 1 if pulse position in first group, only transmits 0 position that indicates this pulse; At receiving terminal,, think that then the position of original pulse is 010 if the indicating bit of receiving is 0; If the indicating bit of receiving is 1, think that then the position of original pulse is 110; For 2 bit informations, be divided into two groups: 00,01 and 10,11; At transmitting terminal, if pulse position in first group, then indicating bit is set to 0, if pulse position in second group, indicating bit is set to 1; At receiving terminal,, think that then these two is 01 if indicating bit is 0; If indicating bit is 1, then think 10.

3, as claimed in claim 1 a kind of in wireless network the AMR method of effectively guaranteeing speek voice quality, it is characterized in that: described coding comprises that the convolution code that provides in employing 3GPP TS 25.212 standards is as chnnel coding, after coding is finished, in order to satisfy fixing transmission rate, carry out rate-matched again.

4, as claimed in claim 1 a kind of in wireless network the AMR method of effectively guaranteeing speek voice quality, it is characterized in that: described coding adopts rate-matched deletion convolution code.

Images