KR100394146B1 - The duplex multilevel adaptive compression voice signal transmission method over internet protocol - Google Patents

Abstract

The present invention relates to a "Duplex Multilevel Adaptive Compression Algorithm Transmission Method" for a real-time voice signal over an Internet protocol. The present invention relates to a resolution level by measuring network delay characteristics at the time of signal input.

Determining an analysis level; Right angle filter for full restoration

Adaptive decomposition with dual multi-resolution through quantum level control and quantum level control; Adaptive quantizing the decomposed signal by band division adaptive differential pulse code modulation (SB-ADPCM); Transmitting the decomposed dual multi-resolution real time voice data and resolution state information through a primary channel and an auxiliary data channel; Dequantizing the transmitted signal; Synthesizing the original signal through inverse quantum level control and inverse rectangular filter; It includes. In other words, the real-time speech data encoding and its transmission method proposed through the present invention are characterized by adaptively transmitting speech using multi-resolution by two-dimensionally changing the encoding method and its transmission speed according to the delay state of the network. . The present invention provides a method for transmitting real-time voice data that is remarkably improved in a network environment where a packet delay problem is serious.

Description

The duplex multilevel adaptive compression voice signal transmission method over internet protocol

The present invention is a special adaptive speech compression algorithm and its transmission method for real-time voice communication technology, ie Voice over Internet Protocol (VoIP) technology, which is expected to play an important role in the communication revolution under the rapidly changing communication environment. To present. The present invention, which is named as the Duplex Multi-Low Adaptive Compression Algorithm, provides a more efficient and robust real-time voice signal transmission method in the case of network delay in the Internet.

Voice over Internet Protocol (VoIP) refers to packet-based voice transmission over the Internet, and is used interchangeably with Internet telephony, IP telephony, packet-voice, and packetized voice. With the advent of this VoIP technology, the Internet and telephone networks are being integrated.

Internet Protocol Network (IP Network) is basically a network designed to transmit general data, not real time, and is not suitable for an environment that delivers real time information such as voice. This is because the voice traffic transmitted in real time on the IP network is affected by the amount of other data transmission, so the voice data may be lost or not properly delivered. In other words, User Datagram Protocol (UDP) on the Internet used for real-time information delivery basically provides end-to-end information transfer function, but it does not guarantee that the information is correctly delivered, and problems occur depending on the network status during transmission. The real-time information is then unavailable and eventually discarded.

In this respect, the effective transmission of voice information through the Internet is a technical problem to be solved and requires a high level of coding and efficient transmission techniques for voice signals.

However, using the Internet to deliver real-time voice information is economically more advantageous than using a conventional telephone network, and because it is widely available and can be conveniently used anywhere, voice transmission techniques and devices using IP networks are constantly Has been developed.

Efforts to develop these techniques and devices began at the beginning of communication between computers with Internet Protocol (IP) addresses, that is, PC-TO-PC. There was a limitation that the other party must be connected to the Internet, so it was not widely used. However, due to the development of related technologies connecting the Internet and the general telephone network and the rapid spread of the Internet, the application of the PC-TO-PHONE and PHONE-TO-PHONE methods by interworking the Internet and the general telephone network became common.

Common factors for determining call quality in the current VoIP service schemes include delay of packets carrying voice information, difference in delay, loss of packets, echo, and the like. Among these, problems related to the delay of each packet are dealt with by applying a technique such as adjusting the size of the buffer, and many techniques have been developed to eliminate the echo problem. If packet loss occurs, it directly affects the decryption of the called party, and if it is severe, it affects the call quality of the called party. Even if about 10% of the voice packets are lost at the current technology level, high quality voice is restored. However, if more loss occurs, the voice is not restored, causing a problem such as a broken voice. Such packet loss is closely related to network reliability and stability. With the expansion of the Internet, the reliability and stability of the network are improving, which has a good effect on the implementation of VoIP technology.

The delay of the packet varies greatly depending on the traffic volume on the network. This is directly affected by the network capacity, routing algorithm, and encoding method of voice signals. As the number of Internet users increases, the delay increases. The latency of packets on the Internet is between approximately 70 and 160 ms. Due to the asynchronous nature of the Internet, the delay is inconsistent, but if the network is heavily loaded, it is well over 300ms. Most delays of less than 100ms do not recognize the delay. At delays of 100ms to 300ms, voices are weakly perceived, while delays of more than 300ms clearly impair call quality. In recent years, the seriousness of the problem is increasing as the type of information passing through the Internet rapidly changes from character data having a small amount of data to graphics and video having a large amount of data.

In order to solve the problem of packet delay, it is necessary to expand the bandwidth by expanding the investment in the infrastructure of the network, but in order to minimize the delay of the packet in a given environment, the packet size is reduced to 10-30 bytes or Various methods are used, such as improving the encoding method of the.

However, the existing method still has a problem that the quality of the call is significantly reduced under excessive packet delay environment. If the network environment is bad, the information of the voice information itself is blocked and the call is almost impossible.

Therefore, when the network delay becomes severe due to traffic increase, the conventional method has a problem that it is difficult to be an appropriate solution in the environment of delivering voice in real time.

The first object of the present invention is to solve a problem in which real-time voice transmission is not performed properly due to delay in data due to an increase in network traffic.

Another object of the present invention is to provide a method for monitoring the traffic of an internet protocol network and delivering voice data of higher quality than normal sound quality in real time when traffic conditions improve in view of the delay rate of packet data according to the increase and decrease of such traffic. It is.

Other purposes of the present invention will be apparent from the following examples.

1 is a view showing the structure of the decomposition QMF,

2 is a view showing the structure of a synthetic QMF,

3 is a diagram showing the frequency response characteristics of the low pass QMF,

4 is a diagram showing the frequency response characteristics of the high pass QMF,

5 is a diagram illustrating multi-resolution decomposition of a one-dimensional signal,

6 is a diagram illustrating a tree structure obtained by decomposing a one-dimensional signal into multiresolution,

7 shows waveforms of a voice signal "chat.wav",

8 is a diagram illustrating a signal decomposed in three stages,

9 is a block diagram illustrating the decomposition of an audio signal;

10 is a view showing network latency measurement data,

11 is a block diagram showing the synthesis of speech signals;

12 is a diagram illustrating a part of an input audio signal waveform;

13 is a view showing SNR when G.722 transmission method is applied.

14 is a diagram showing the results of SNR measurement when the time domain is fixed-level decomposition and the frequency domain is 7-bit fixed quantization.

15 is a diagram illustrating an SNR measurement result when the time domain is adaptively decomposed and the frequency domain is 7-bit fixed quantized.

FIG. 16 is a diagram showing an SNR measurement result when the time domain is not decomposed and the frequency domain is adaptive quantized.

FIG. 17 shows the SNR measurement results when the time domain is adaptively decomposed and the frequency domain is adaptive quantized. FIG.

In order to achieve the above object, the present invention comprises the steps of measuring the network delay characteristics at the time of signal input (Analysis Level) to determine; Adaptive decomposition to dual multi-resolution through a quadrature filter and quantum level control to ensure complete restoration; Adaptive quantization of the decomposed signal by band division adaptive differential pulse code modulation (SB-ADPCM); Transmitting the decomposed dual multi-resolution real time voice data and resolution state information through a primary channel and an auxiliary data channel; Dequantizing the transmitted signal; Synthesizing the original signal through inverse quantum level control and inverse rectangular filter; It includes. In other words, the real-time speech data encoding and its transmission method proposed through the present invention are characterized by adaptively transmitting speech using multi-resolution by two-dimensionally changing the encoding method and its transmission speed according to the delay state of the network. . The present invention provides a method for transmitting real-time voice data that is remarkably improved in a network environment where a packet delay problem is serious.

Hereinafter, a basic design of a quadrature mirror filter (QMF) for signal decomposition and synthesis will be described, and a dual multiresolution adaptive compressed voice signal transmission method according to the present invention Internet protocol will be described step by step. .

1. Design of right angle filter

Signal decomposition

The rectangular filter structure of the lattice structure without band loss is as follows. To decompose the input signal as shown in FIG transfer function Hp (z) is represented by the following formula 1.

(Equation 1)

Wherein A _'k is a 2x2 orthogonal (Orthogonal) matrix, D (z) can be expressed as Equation 2 shows.

(Equation 2)

A _'k can be expressed by equation (3) because it is an orthogonal matrix having a constant gain.

(Equation 3)

In summary, Hp (z) can be arranged to form a QMF having a lattice structure having a constant gain as in Equation 4.

(Equation 4)

Where c is the total gain of the filter and is expressed as the product of the gains in each step.

(Eq. 5)

Synthesis of Signal

The inverse for restoring can be obtained as

(Equation 6)

However, considering Equation 7

(Eq. 7)

Since D ⁻¹ (z) = D (z ⁻¹ ), it can be expressed as follows.

(Eq. 8)

Therefore, the lattice structure of the reconstruction filter can be expressed as the following equation, and the structure thereof is shown in FIG. 2.

(Eq. 9)

In summary, for example, the simplest rectangular filter can be expressed as shown in Equation 10.

(Eq. 10)

Here, the lowpass filter part is represented by c [1 1] and the highpass filter part is represented by c [1 -1].

Frequency Response Characteristics of Rectangular Filters

The frequency response characteristic of the designed filter is shown in FIGS. 3 and 4. FIG. 5 shows the multi-resolution decomposition of the one-dimensional signal, and FIG. 6 shows the structure of the tree obtained by decomposing the one-dimensional signal into the multi-resolution.

2. Dual multiresolution adaptive compression algorithm

(Duplex Multilevel Adaptive Compression Algorithm)

summary

Basically, ITU-T G.722 divides the input signal into two bands of low band (0 -4 kHz) and high band (4 -8 kHz) using a quadrature filter and then performs differential quantization and synthesizes the low and high bands. To transmit. The structure of the quadrature filter must satisfy full restoration and also satisfy the linear phase. In order to have such a structure, an appropriate filter coefficient A _k must be selected. To simplify the coefficients, 1 is selected as the overall gain of 2, and the energy is doubled at each stage. This can be adjusted again in the quantization process, so we will not consider the gain here.

In the present invention, the conventional method of G.722 is drastically changed, and the low-frequency signal portion that has been processed once is repeatedly decomposed two-dimensionally to a certain level according to the traffic state to analyze the signal with a structure having a dual multi-resolution. . In multiresolution decomposition and quantization, the sampling frequency and quantization interval are repeatedly changed to a certain level according to the network conditions. This paper focuses on preventing the sound quality degradation of the real-time voice caused by the delay of packet, especially in the Internet environment where the transmission quality is not guaranteed, and attempts to deliver a higher quality real-time voice when traffic conditions improve. In order to approach this goal, the present invention provides a method of designing and transmitting an efficient voice compression algorithm that monitors the network load condition and variably adjusts the data transmission rate according to the situation.

Voice signal decomposition

The waveform of the "chat.wav" voice data used to explain the decomposition process of the voice signal is shown in FIG.

After decomposing the signal up to three stages with multiple resolutions, the low pass filter signals of each stage are shown in FIG. 8. As can be seen in FIG. 8, as the QMF steps are passed one by one, the high pass signals disappear and only the low pass signals remain. It can be seen that the transmission rate of the signal is sampled downward with each step and continues to decrease in half.

As shown in Fig. 8 showing the decomposition process in the time domain, the input signal is decomposed into several resolutions. The resolution is determined by the delay time of the network. In the present invention, for high speed transmission, the high pass signal is not transmitted.

9 shows an exploded block configuration diagram of an audio signal.

Delay measurement on the network

The present invention is to adaptively transmit voice using dual multi-resolution by changing the signal transmission rate in consideration of the delay state of the network, and exchanges information on the delay time occurring on the network through an auxiliary channel. Using this information, the data rate is limited when a certain delay occurs. The network model to be used in the experiment was constructed by repeatedly measuring the Yahoo site on the actual network with "Ping.exe". The results are shown in FIG.

As shown in Figure 10, the transmission speed of the network is constantly changing. Transmitting data in real time in a situation where transmission quality is not guaranteed due to random delay may cause significant degradation of transmission data due to the characteristics of the network. In order to improve this problem, the present invention divides the transmission delay on the network by a threshold and takes the quotient to determine the analysis level. In this case, the thresholds used were set to 300ms, 200ms, 100ms, and 50ms. In each case, the data generation rate was reduced by half in both the time domain and the frequency domain as the resolution level increased. In the time domain, the decimation method is used to reduce the sampling frequency. In the frequency domain, the quantization interval of the lowpass signal is increased to reduce the data generation rate.

Synthesis of Speech Signal

The transmitted signal is synthesized into the original signal through an inverse quantizer and a square filter. At this time, the raw signal is synthesized using only the low-band signal without considering the untransmitted high-band signal. The synthesis block diagram of such a voice signal is shown in FIG.

The present invention is explained in more detail through the following examples. However, the present invention is not limited to the contents described in the Examples, and the following Examples are described for illustrative purposes only.

Example

The theme song "My heart will go on" of the Titanic movie was used as a signal, and a part of the input signal is shown in FIG. 12. In order to observe the Signal to Noise Ratio (SNR) when the signal is transmitted in the same manner as the conventional G.722, the input signal is divided into G.722 mode 2 speed 56kbps, 1 level QMF decomposition, It was quantized into a 5-bit low-band signal and a 2-bit high-band signal. The signal was transmitted through an experimental model network, followed by dequantization and QMF synthesis. The SNR change curve of the synthesized signal is shown in FIG. 13. As can be seen in Figure 13 shows a relatively even SNR value. However, by generating data at a constant speed without considering network delay time, there is a disadvantage in that a discontinuous section occurs due to time delay of voice quality.

The decomposition level determination formula, which is an important variable used to control the data generation rate in the present invention, is represented by Equation 11.

(Eq. 11)

The signal was decomposed according to the resolution level thus determined and quantized into N bits. The N value was determined as shown in Equation 12. The maximum value of decomposition level was 6 at this time. The reason is that at 7 or more, multiresolution features hardly appear at the resolution level and the signal is distorted much.

(Eq. 12)

FIG. 14 shows each signal to noise ratio (SNR) obtained by disassembling and synthesizing a speech signal for each given fixed resolution level, ignoring the delay characteristics of the network. In the frequency domain, 7-bit fixed quantization is attempted.

Fig. 15 shows the result of determining the decomposition level in consideration of the transmission delay time occurring in the network, then attempting decimation in the time domain (adaptive decomposition), and adjusting the incidence of data by fixed quantization with 7 bits in the frequency domain.

In FIG. 16, the decomposition of the time domain is not applied. In FIG. 16, data is generated by adjusting the quantization interval (adaptation quantization) according to the decomposition level only in the frequency domain. Compared with FIG. 15, it was found that signal deterioration was relatively large.

In the voice signal transmission method provided by the present invention, the decomposition and synthesis are simultaneously applied in the time domain and the frequency domain so that the generation rate of data can be adaptively adjusted according to the decomposition level representing the delay characteristics of the network. The SNR change measured by applying this method is shown in FIG. 17, and as a result, it was found that the proposed method shows more stable SNR in the transmission delay environment than other algorithms when compared with the same level of compression ratio. In addition, even when the network load worsened, it was able to quickly adapt to changes in the environment and provide voice at a low bit rate to alleviate the disconnection state of the voice.

The present invention decomposes an input signal into a dual multi-resolution using a rectangular filter that guarantees full restoration, and then changes the format used by the G.722 standard to perform quantization of 7 bits or less, and 1 bit auxiliary. The low-delay high-quality transmission is realized by exchanging time information and control information on transmission delay using a data channel. In the present invention, it is not limited to the 300ms transmission delay environment, but by setting the threshold value of several stages (300ms, 200ms, 100ms, 50ms) and increasing the resolution level and the quantization interval of the square filter each time the multiple of the threshold is exceeded. Reduced overall data generation. In order to transmit signals at high speed, SNR must be sacrificed in the Internet environment, but the characteristics of voice signals can be used to dynamically change the resolution of these signals without human detection, so even in a constantly changing network delay environment. At the same bit rate, relatively low noise quality was obtained. In the present invention, whenever the network delay exceeds a certain level, a method of reducing the transmitted information is halved. By simultaneously reducing the data generation rate in the time domain and the frequency domain, the signal to noise ratio is improved at the same bit rate. I was able to achieve my goal.

In the voice signal transmission method provided by the present invention, the resolution and the quantization interval are adaptively changed in the time domain and the frequency domain, thereby maintaining the sound quality of the transmitted signals relatively even in an environment where QoS is not guaranteed, such as the Internet. .

By providing algorithms that can effectively transmit real-time, high-quality voice signals over the range of telephone voices on the Internet, it is possible to prevent severe sound degradation in network environments with poor delay characteristics and to maintain sound quality throughout.

In the voice signal transmission method provided by the present invention, the decomposition and synthesis are simultaneously applied in the time domain and the frequency domain so that the generation rate of data can be adaptively adjusted according to the decomposition level representing the delay characteristics of the network. The SNR change measured by applying this method is shown in FIG. 17. As a result, it was found that the method according to the present invention showed more stable SNR in the transmission delay environment than other algorithms when compared with the same level of compression ratio. In addition, even when the network load worsened, it was able to quickly adapt to changes in the environment and provide voice at a low bit rate to alleviate the disconnection state of the voice.

Claims (8)

Adaptive decomposition to dual multi-resolution through a quadrature filter and quantum level control to ensure complete restoration; Synthesizing a dual multi-resolution signal into an original signal through inverse quantum level control and inverse rectangular filter; And adaptively transmit voice by changing a signal transmission rate according to a delay state of a network. The method of claim 1, wherein the method of varying the signal transmission rate according to the network delay state performs decomposition and synthesis in the time domain and the frequency domain so that the generation rate of the data can be adaptively adjusted according to the decomposition level representing the delay characteristic of the network. A dual multi-resolution adaptive compressed voice signal transmission method over an internet protocol, characterized by simultaneous application. 3. The method of claim 2, wherein the method of adaptively adjusting the data generation rate according to the decomposition level reduces the data generation rate by half in both the time domain and the frequency domain every time the decomposition level is increased. A method for transmitting dual multi-resolution adaptive compressed voice signals over an Internet protocol characterized by reducing the sampling frequency by repeating) and reducing the incidence of data by repeatedly increasing the quantization interval of the lowpass signal in the frequency domain. . 3. The method of claim 2, wherein the decomposition level is determined by dividing a transmission delay on a network by a threshold and taking a quotient thereof. 5. The method of claim 4, wherein the threshold value used is set among a group consisting of 300 ms, 200 ms, 100 ms, and 50 ms. 3. The formulation of claim 2 wherein the decomposition level determination formula is A dual multi-resolution adaptive compressed voice signal transmission method over an Internet protocol characterized by the above-mentioned. The method of claim 1, wherein the N value in the adaptive quantization step is A dual multi-resolution adaptive compressed speech signal transmission method over an Internet protocol characterized by the above. 8. The method according to claim 7, wherein the maximum value of the decomposition level is 6 in the N value determination equation.