WO2006111063A1

WO2006111063A1 - A method for protecting data transmitting based on erasure codes

Info

Publication number: WO2006111063A1
Application number: PCT/CN2006/000389
Authority: WO
Inventors: Zhong Luo; Bin Song; Yilin Chang
Original assignee: Huawei Technologies Co., Ltd.
Priority date: 2005-04-21
Filing date: 2006-03-14
Publication date: 2006-10-26
Also published as: CN1855794A; CN100446452C

Abstract

The invention discloses a method for protecting data transmitting based on erasure codes, which core is: making the erasure codes only having one layer of the parity nodes, and performing the data transmitting protecting with respect to said erasure codes. In the case of the invention ensuring that the capability of protecting the data transmitting does not observably reduce, it sets only one layer of the parity nodes for erasure codes, so that it effectively reduces the amount of calculation for generating the layer of the parity nodes by the erasure codes, reduces the delay of the data transmitting greatly, improves the cost performance of the data transmitting protecting, and further makes the method according to the invention for protecting data transmitting being applied in the technical field of H.263 or H.264 ,etc. The invention improves the efficiency of the data transmitting, and promotes the application of the new technology about H.264,etc.

Description

Data transmission protection method based on erasure code

The present invention relates to an anti-drop packet error technique in a communication system, and in particular to a data transmission protection method based on an erasure code. Background of the invention

The erasure code is a widely used anti-lost error technique. Because the Tornado code in the erasure code is relative to other erasure codes, such as Reed-Solomon code, it has the characteristics of simple structure, high efficiency, strong protection, etc., so in the actual data transmission process, especially in audio and video. The real-time transmission of data has been widely used.

The so-called Tornado code is to divide the data stream sequence into segments of the same size (Unit), also called data nodes (Data Nodes), and then calculate these data nodes according to certain mathematical rules to generate the first. Layer check nodes (Parity Nodes or Check Nodes), and then continue to operate on the check nodes of the first layer according to the same or different mathematical operation rules, generate check nodes of the second layer, and so on, and generate third Layer, layer 4, check nodes up to the second layer.

Since the number of nodes in each layer of the Tornado code is decremented by a fixed ratio, the structure of the Tornado code is like a pyramid rotated 90 ^{Q to} the right, as shown in Figure 1. According to the summation law of the equal series, the number of all nodes (including data nodes and check nodes) in the entire pyramid structure is p times the number of data nodes, where p is not necessarily an integer. The above multiple relationship makes the Tornado code in the process of data transmission, and finally the number of nodes to be processed at the encoding and decoding end is linear with the number of data nodes n, which means that the time required for the encoding and decoding end to process the Tornado code is complicated. The degree has a linear relationship with the number n of data nodes, that is, has a linear-time characteristic. Since the Tornado code has the linear time characteristic described above, it has high computational efficiency and is therefore suitable for real-time communication in a communication system. During the data transmission process using the Tornado code, each check node and data node of the Tornado code are sent by the sender to the receiving end through the network. If some nodes are lost during network transmission, since the upper node participates in the generation of the lower node, that is, the information of the upper node is already included in the lower node and the lower node, so the lost node can pass a sufficient number. The lower node or lower node is fully recovered. If each node is a packet, the lost packet can be completely recovered by other packets received correctly, thereby realizing data transmission protection.

At present, the Tornado code is mainly used in data transmission protection based on H.261 H.263/H.263+/H.263++/H.264 video compression coding. The method includes the following steps:

Step 1. Design the structure of the Tornado code. Specifically, the data node is determined to have a size of L1 according to a given protection capability, other requirements, specific data types, such as audio or video, rate, and the like, and the data node is determined according to factors such as a maximum network delay acceptable in a specific application. Number n, the total number of check nodes L, the number of houses in the middle check layer m, the scale factor of the number of nodes between layers, etc., determine the bipartite graph between any two adjacent nodes The left degree distribution vector and the right degree distribution vector.

Step 2. Generate a random bipartite graph between each adjacent two-layer node by using a random graph matching method according to the left-hand degree distribution vector and the right-side degree distribution vector.

Step 3. When starting to transfer data, the sender

The H.261/H.263/H.263+/H.263++/H.264 video encoder generates a data stream that requires data transmission protection.

Step 4: The data stream to be protected is divided into data nodes D having equal sizes of L 1 bits. , DD ₂ , : ₃ ....D _T .

Step 5. Let t=0. Step 6. Obtain D _t , Dtw — A+^n data nodes.

Step 7. According to the random bipartite graph between the adjacent two-layer nodes determined in step 2, each of the n data nodes D _t , D _t+1 ....D _{t+I>1 is} generated layer by layer. Node layer MC ^(G) , MC ⁽¹⁾ ... MC ^(m) and finally check node layer FC.

Step 8. The data nodes D _t , Dt+ A^ and the check nodes in each check node layer generated by step 7 are all packaged by network packing methods such as UDP/IP or TCP/IP, and then sent to the receiving end.

Step 9. The transmitting end determines whether all the data nodes are processed according to the numerical relationship of t and T. If yes, execute step 10; otherwise, let t=t+n, and then perform step 6.

Step 10. The communication process of the sender ends.

Since the sender is often in the process of real-time communication, it is sent in the order of encoding compression.

The above process describes the sender. For the receiving end, after receiving a batch of data nodes and check nodes at the receiving end, it should first determine which data nodes are data nodes that should be received but are actually lost and recoverable, and then, for the above, have been lost but may be The recovered data node is decoded and restored according to the general decoding process of the Tornado code. The receiving end repeats the above receiving and recovery process until the communication process ends.

Although the Tornado code has the advantages of simple structure, high computational efficiency and strong protection ability compared to other erasure codes, since the Tornado code needs to generate multiple intermediate node layers, layer calculation is required, and more complicated calculations such as Reed-Solomon are needed. The method generates a check node in the final check node layer, thus adding a lot of computational work to the sender. If the video data needs to be transmitted, such as: H.264 compression coding itself has a large amount of computation, the sender is required to have a strong computing power, and the result limits the application of H.264 and other technologies.

It can be seen from the structure of the above Tornado code that if the number of nodes in each layer of the Tornado code Relative to the decreasing scale factor of the number of nodes in the previous layer 3⁄4o=l/2, 'and the number of layers of the intermediate check layer is m=3, since the number of check nodes of the 笫3 layer is at least 2, the data node can be deduced The number of n = 16. That is to say, if a frame image is used as a data node, the use of the Tornado code will bring a time delay of 16 frames. In practical applications, since the video communication frame rate does not exceed 30 s, 16 frames correspond to a time delay of more than 0.5 seconds, and other time delays, resulting in extremely large video data transmission time delay. And even if m = 2, the video data transmission time delay brought by the Tornado code will reach 0.25 seconds or more.

In addition, since the data format of NAL (Network Adaption Layer) of H.264 is just completed by IETF (Internet Engineering Task Force), the existing method for data transmission protection based on Tornado code is not directed to H. The 264 NAL function does not work best with H.264 NAL. Summary of the invention

In view of this, an object of the present invention is to provide a data transmission protection method based on an erasure code, which is overcome by setting an erasure code having only one layer of check nodes to achieve a significant decrease in data transmission protection capability. In the prior art, the shortcomings of the computing power of the transmitting end are high, the data transmission is extended, and the technical application such as H.264 is limited.

The method of the invention mainly comprises the following steps:

a, setting an erasure code having only one layer of the face node layer;

b. Perform data transmission protection according to the erasure code.

In the above method, the step a includes:

Al, determining a size of a data packet corresponding to the data node in the data node layer of the erasure code, a number n of data nodes included in the data node layer, and a number L of check nodes included in the check node layer;

A2, determining a random bipartite graph and a predetermined calculation between the data node layer and the check node layer Law.

In the above method, the step a1 is: determining, according to a data transmission rate, a data type, a data protection capability requirement, and/or a network delay requirement, a size of the data packet, a number n of data nodes included in the data node layer, and Check the number L of check nodes contained in the node layer.

In the above method, the step a2 is:

Determining the left side distribution vector and the right degree distribution vector of the bipartite graph;

Generating, according to the left degree distribution vector and the right degree distribution vector, a random bipartite graph between the data node layer and the check node layer by using a random matching manner, and determining the data node layer and the check node layer The association relationship of the XOR operation is performed according to the random bipartite graph.

In the above method, the step b includes:

Bl, at the transmitting end, dividing the data stream into n data nodes according to the size of the data packet corresponding to the data node in the data node layer of the erasure code;

B2, generating, according to the predetermined algorithm and the random bipartite graph of the erasure code, only one layer of a check node layer having L check nodes for the n data nodes;

B3, transmitting the n data nodes and L check nodes to the receiving end;

B4. At the receiving end, when it is determined that the lost data node can be recovered, the lost data node is recovered according to the received check node, the random bipartite graph, and the XOR operation relationship.

In the above method, the data code stream includes: a data code stream based on a data unit encapsulated by a network adaptation layer or a data unit encapsulated based on a network abstraction layer;

And the step b1 is: at the transmitting end, forming a predetermined number of data units based on the network adaptation layer or the network abstraction layer encapsulation into a bit block, and then according to the size of the data node in the data node layer of the erasure code The bit block is segmented into data nodes. In the above method, the data unit based on the network adaptation layer encapsulation includes:

H.263 network adaptation layer encapsulated data unit; the network abstraction layer encapsulation-based data unit comprises: a data unit based on an H.264 network abstraction layer encapsulation.

In the above method, the step bl further includes: when determining that the size of the bit block is not an integer multiple of a data node size in the data node layer of the erasure code, performing bit filling on the bit block, The size of the bit block after padding is made an integer multiple of the data node size in the data node layer of the erasure code.

In the above method, the step a further includes:

Determining the correspondence between each data type and each data transmission protection level;

Setting an erasure code having a check node layer and having different check nodes for each data transmission protection level;

The step b includes:

Determining, according to predetermined information carried by each data in the data stream, a data type of each data; determining, according to the data type and a correspondence between the data type and the data transmission protection level, data transmission protection corresponding to each data in the data code stream Grade

Dividing the data stream into corresponding substreams according to the data transmission protection level; and performing data transmission protection on each substream according to the erasure code set for the data transmission protection level corresponding to each substream.

In the above method, the predetermined information comprises: non-texture information and texture information of the data unit based on the H.264 network abstraction layer.

In summary, the present invention can be seen from the description of the technical solution, the present invention reduces the correction by setting an erasure code having only one layer of check nodes, while ensuring that the data transmission protection capability is not significantly degraded. The code generation generates the operation amount of the check node layer, reduces the delay of the data transmission, and improves the cost performance of the data transmission protection, so that the data transmission protection method of the present invention can be applied to new technologies such as H.264; Layer or network abstraction layer The encapsulated data unit performs data node segmentation, so that the data transmission protection method of the present invention can be well matched with technologies such as H.263 and H.264; by using erasure codes with different protection capabilities for different data types, For example, an erasure code with high redundancy is used for important data, and an erasure code with low redundancy is used for ordinary data, so that the present invention can target different types of data, such as: important image data and ordinary image data. Realize non-equal weight protection, so that all kinds of important data can be restored as much as possible in a high packet loss environment, so that the data stream can obtain maximum data transmission protection under the same average redundancy rate of the erasure code. It has realized the purpose of improving data transmission protection cost performance, improving data transmission efficiency and promoting the application of new technologies such as H.264. BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a schematic structural diagram of a Tornado code in the prior art.

2 is a schematic structural view of a modified Tornado code of the present invention.

FIG. 3 is a schematic diagram of the improved Tornado code and the H.264 NAL unit in the present invention for data transmission protection.

Figure 4 is a schematic diagram of the data format of the H.264 NAL unit. Mode for carrying out the invention

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described in detail with reference to the accompanying drawings.

In the process of data transmission protection using Tornado code, setting the multi-layer Tornado code check node layer will enhance the data transmission protection ability to a certain extent, however, setting the multi-layer Tornado code check node layer will also make the Tornado code. The amount of computation is large, so that the data pays a large delay in the process of transmission protection. If you can reduce the number of layers in the check node layer without ensuring a significant drop in data transmission protection, you can effectively reduce it. The amount of operation of the Tornado code is reduced, thereby greatly reducing the delay in the data transmission process, and thus seeking higher data transmission protection performance - cost performance.

Therefore, the core of the present invention is: setting an erasure code having only one layer of check nodes, and performing data transmission protection according to the erasure code.

The technical solution of the present invention is further described in detail below.

The erasure code of the present invention only has one layer of check nodes, and the intermediate check node layer of the existing Tornado code is removed. Similarly, the inherent requirement of generating the last check node according to Reed-Solomon code in the existing Tornado code is also removed. . Thus, the erasure code of the present invention has only one layer of data nodes and one layer of check nodes as shown in FIG. 2. It can be said that the erasure code of the present invention is a Tornado code with a simplified structure, which is an improvement. Tornado code.

In the improved Tornado code structure of the present invention, the size of the data node L1, the number n of data nodes in the data node layer, and the number L of check nodes in the check node layer can be determined based on actual requirements. For example, the size of the data node in the data node layer, the data node layer, can be determined according to factors such as data transmission rate, data type, such as audio/video data, data protection capability requirements, and maximum network delay that can be accepted. The number n of data nodes included in the check node and the number L of check nodes included in the check node layer.

It is assumed that the Tornado code in the prior art has an m-layer intermediate check node layer, and from the data node layer to the m-th intermediate check node layer, the ratio of the number of nodes between adjacent two layers is ίθ, and finally The equal scaling factor of the number of nodes between the layer and the mth layer is "o ^m+1 /(l - 3⁄4t>), then the total number of nodes of the Tornado code in the prior art is Total _N . _De is:

Total _N. _De =n[l+80+80 ² + ..... + 3⁄4o ^m +io ^m+1 /(l-3⁄4o)]=[l/(l - 3⁄4o)]n„

Thanks to Total _N. _De = n + L, therefore, the setting of L is limited, L = [3⁄4o / (l - 3⁄4o)] n, L can not be arbitrarily set. Since it is necessary to ensure that the number of nodes of each node in the Tornado code is an integer, η ϊ θ, η ϊΰ ² , η κ ³ n3⁄4o ^m and nl o ^m+1 /(l-«o) are required to be integers. This condition is called hidden. Contains the number of integer nodes. According to this condition, if given in the Tornado code With 01 and 3⁄40, you can calculate the conditions that n needs to satisfy. For example, when m=3, ΪΟ = 1/2, then n=16k can be calculated, where k is any natural number. It can be seen that the minimum value that n can obtain is 16, and the code rate r of the Tornado code in the prior art is: r=n/(n+L)=l-3⁄4o=1/2, and the redundancy rate 1 - r is: 1 - r = 3⁄4o = l/2.

The improved Tornado code of the present invention eliminates the need for the implicit integer node number condition described above because of the absence of the intermediate check node layer, and the check node of the check node layer of the improved Tornado code of the present invention. The number L is: L = «on, so the equal-ratio scalar factor of the number of nodes of the data node layer and the check node layer of the improved Tornado code of the present invention can be arbitrarily set, given the number n of data nodes, L can be flexibly set.

The code rate r of the improved Tornado code of the present invention is: r = l / (l + so); The redundancy rate of the improved Tornado code of the present invention 1 - r is: 1 - r = so / (l + «o).

The improved Tornado code of the present invention can be expressed as TN (n+L, n), such as TN (30, 20), indicating the number of data nodes in the data node layer n=20, and the number of check nodes in the check node layer. =10. At this time, the improved Tornado code of the present invention has £O = L / n = 10 / 20 = 1 / 2, and the code rate r = 2 / 3 = 66.7%.

By comparing the code rate and the redundancy ratio of the improved Tornado code in the present invention with the code rate and redundancy rate of the Tornado code in the prior art, it can be known that, under the condition that the same number of data nodes n and the equal scaling factor BO, The code rate of the improved Tornado code in the present invention is increased by 1/6 compared with the code rate of the Tornado code in the prior art, the percentage increase is: 33%, and the redundancy rate is reduced compared with the Tornado code in the prior art. It is. Therefore, the transmission efficiency of the improved Tornado code in the present invention is higher. In the prior art, the data transmission delay introduced by the Tornado code is at least 0.5 second, and the data transmission delay introduced by the improved Tornado code in the present invention is at a normal level of the video frame rate, that is, at 30 s, the minimum can be 1 /15=0.07 seconds, according to the video frame rate of Sfps, the number of transmission delays introduced by the improved Tornado code does not exceed 0.25s. After determining the number of data nodes n, the number of check nodes L, and the data node size L1 of the improved Tornado code of the present invention, the next step is to determine a predetermined algorithm for generating a check node by the data node, and a data node layer and a check node layer. The relationship between the two. The relationship between the data node layer and the check node layer also refers to which data nodes in the data node layer participate in the generation of check nodes in the check node layer.

In the present invention, determining the relationship between the data node layer and the check node layer in the improved Tornado code is the same as the relationship between the data node layer and the check node layer of the Tornado code in the prior art, and the specific The process is:

The association between the data node layer and the check node layer is essentially a topological relationship of graphs, which can be described by a Bipartite Graph. In the branch of the discrete mathematics "Graph Theory", the bipartite graph is also called the even graph, which is a special graph. This kind of graph is divided into two sets of left and right nodes, or a set of two vertexes (vertex), a connecting arc between the left and right nodes, that is, an edge (Edge) represents the relationship between the left and right nodes, and the edge exists. There is no edge between any two nodes in the same node set between a pair of nodes belonging to the left and right node sets.

For example, the relationship between the data node layer and the check node layer is: If there is an association between the data node i and the check node j, that is, the data node i participates in the generation operation of the check node j, then the data There is an edge between node i and check node j; vice versa. If the relationship between the data node i and the check node j can be expressed in the form of sufficient necessary conditions: if and only if there is an edge between the data node i and the check node j, the data node i participates in the school. Check the generation operation of node j.

The improved Tornado code in the present invention has only one bipartite graph, that is, a bipartite graph between the data node layer and the check node layer, and the Tornado code in the prior art has at least two bipartite graphs.

Since the association relationship is essentially a many-to-many relationship, there may be multiple nodes associated with the data node i and the check node j in the above example. That is, the data node i can participate in the generation operation of verifying a plurality of check nodes in the node layer, and the face node j can be generated by a plurality of data nodes in the data node layer.

According to the terminology in graph theory, the number of edges associated with a node is called the degree of the node.

(degree), for the two parts of the improved Tornado code of the present invention, there are two degree sequences: a left side sequence and a right degree sequence. The left degree sequence is a sequence formed by the left node, that is, the degree of each node in the generated node set, and the right degree sequence is a sequence formed by the right node, that is, the degree of each node in the generated node set.

The degree of the edge can be defined according to the left degree sequence and the right degree sequence of the node. For example, for the edge e, if the degree of the data node i on the left side is 2, and the degree of the check node j on the right side is 8, it can be said that the left degree of the edge e is 2, and the right degree is 8.

In the improved Tornado code of the present invention, the more edges exist in the bipartite graph, that is, the more the relationship between the two sets of nodes on the left and right, the stronger the data transmission protection capability, that is, the data transmission protection capability and the left and right. The association relationship between the node sets is related, and there is no direct relationship between which data node is associated with which check node. More strictly speaking, the left and right distribution vectors are an important factor in determining the data transmission protection of the improved Tornado code.

The present invention determines the left degree distribution vector of the bipartite graph of the data node layer and the check node layer after determining the data node size of the improved Tornado code, the number of data nodes of the data node layer, and the number of check nodes of the check node layer. And the right-degree distribution vector, and randomly generating a random bipartite graph of the data node layer and the check node layer. In the random bipartite graph, the nodes and the nodes are randomly associated, but in the macroscopic aspect, the edges The distribution of degrees conforms to the distribution law given by the left degree distribution vector and the right degree distribution vector.

The predetermined algorithm between the data node layer and the check node layer of the improved Tornado code of the present invention is an exclusive OR XOR (exclusive OR) operation, also called addition modulo 2 . The XOR operation is a bit operation, and the data nodes of the improved Tornado code are equal in length. Bit block, the check node is a bit block formed by bit XOR operation between corresponding bits of two data node bit blocks. For example: Two bit blocks of length L are: A=[a. , a! , a ₂ , ..... , a _L ] , B=[b ₀ , b _{1 ;} b ₂ , ..... , b _L ] , then A and B are obtained after XOR operation The result is: A/B = [ a _Q /b. , _ai /b! , a ₂ /b ₂ ,

The XOR operation is used because XOR has the following mathematical properties: For any binary variable a, b, the following relationship exists: (a/b)/ba. Similarly, for any binary block variable A=[ao, a _{1 ;} a ₂ , ....., a _L ], B=[b ₀ , b _u b ₂ , ..... , b _L ] , also exists: (A / B) / B = A. If A and B are data nodes, then (A/B) is the check node. Thus, during data node transmission, when data node A is lost, data node B and check node (A/B) are received. If it is received correctly, the receiving end can recover the lost data node A by (A/B)/B=A.

The above-mentioned receiving end recovers the lost data node according to the formula (A/B)/B=A, which is only the most simple and basic example. According to the formula (A/B)/B=A, a series of more can be constructed. Complex and involved recovery equations for multiple missing nodes. For example, data nodes A, B, and C are all lost, but check nodes A/B, A/C, and A/B/C are correctly received by the receiving end, and the receiving end can be based on the formula (A/B) I (A /C) I (A/B/C)=A to recover data node A.

In summary, when the receiving end recovers the lost data node in the decoding process of the Tornado code improved according to the present invention, the principle of restoring the data node is: For the bipartite graph, if a right node is correctly received and associated with the right node Only one of the left nodes of the join is lost, then the lost left node can be recovered by the right node being different from all the left nodes that are not lost. The erasure code based data transmission protection method of the present invention.

The H.264 video compression coding standard was developed by ITU-T (International Telecommunication Union) in conjunction with ISO/IEC (International Organization for Standardization/International Electrotechnical Commission) MPEG (Moving Picture Experts Group) International Standards Organization, and H.264 has gradually become a multimedia communication. The mainstream standard in China. Because H.264 uses a variety of efficient coding algorithms, the video stream is very sensitive to channel errors. Taking IP networks as an example, although IP networks use many quality of service (QoS) policies in the bearer layer, they still appear frequently. Network bandwidth fluctuations, packet loss, and packet delay. In actual video communication, the image quality degradation caused by IP network packet loss is very serious, sometimes even unbearable. In severe cases, it will lead to decoding. The end system crashed. In other words, H.264 is more powerful, more efficient, and more versatile than other video compression coding standards, even if it is less able to withstand network deletion errors, even a single primary error. It may also cause a sharp drop in the quality of the recovered video.

Therefore, in H.264-based video communication, it is necessary to adopt an effective anti-drop error technique and combine various video error-resistance methods to ensure the quality of the restored image.

In order to cooperate with H.264 and other technologies as much as possible, the present invention adopts the following two practical cooperation measures:

Measures 1. After the data stream outputted by the H.264 video encoder is encoded and encoded by the network abstraction layer (NAL), a predetermined number of consecutive H.264 NAL units (NALUs) are set. As a large block, as shown in FIG. 3, the data node is divided according to the large block and the check node is generated, and then the network transmission is performed. That is, the present invention is implemented based on the H.264 NAL unit when dividing the data node.

In practical applications, in order to ensure that the data length of the large block is an integer multiple of the data node size, the present invention may also perform bit positioning on the large block when the data length of the large block is not an integer multiple of the data node size. Filling, for example, may fill a number of zeros at the end of the large block, such that the data length of the large block is an integer multiple of the data node size.

H.264 NAL is equivalent to a network adaptation layer, which can ensure that the data stream on it is independent of the specific transmission network. The H.264 NAL unit adapted by H.264 NAL can be adapted to many specific network protocols. Make a package transfer. Similar in existence to H.264 NAL Other techniques of the network adaptation layer, such as H.263, can also be applied to the above measures. Step 2: When determining the number of nodes in the check node layer of the improved Tornado code of the present invention, it may also be determined according to the content of the predetermined information carried in the H.264 NAL unit. That is, according to the non-texture information and texture information carried in the H.264 NAL unit to determine whether it is ordinary data or important data, and then use the improved Tornado code with relatively high redundancy for important data, and the redundancy for common data. A relatively low improvement Tornado code.

Here, it should be pointed out that: The distinction between ordinary data and important data is relative rather than absolute, since strengthening protection for important data is at the expense of efficiency, so more redundancy needs to be introduced. Therefore, in specific cases, the trade-off principle, that is, the importance of protection and efficiency, can be used to determine which data is important data and which data is ordinary data. Generally speaking, in the video information, the information constituting the pixels of the image itself is called texture information, and in addition, the structural information of the image, such as: header information, motion vector, various message information, etc., are not pixels of the image itself. The information is therefore non-texture information. Non-texture information belongs to the scope of important information, while the opposite texture information belongs to the scope of ordinary information. But at some point, the texture information of the reference frame is also important information.

The specific structure of the H.264 NAL unit is shown in FIG. 4. In FIG. 4, the H.264 NAL unit is composed of NAL header information and NAL payload. The AL payload information is the texture information, and the NAL header information is the non-texture information. The NAL header information includes a 1-bit forbidden_zero-bit, a 2-bit Nal_ref_idc, and a 5-bit Nal-unit. — type, where forbidden—zero—bit is always zero; Nal—unit—type indicates the type of NALU. Currently, there are 12 types defined in H.264 technology; Nal— ref—idc has the meaning: If Nal— If the value of ref_idc is zero, it means that a slice (Slice) or slice data of the non-reference image is stored in the H.264 NAL unit, and the H.264 NAL unit does not affect its subsequent H.264 NAL unit. Decoding; if the value of Nal_ref_idc is non-zero, that is, the value is 1, indicating that the H.264 NAL unit stores a sequence/image parameter set or a reference image. A slice or slice data segmentation, the Η.264 NAL unit will seriously affect the decoding of its subsequent H.264 NAL unit.

Therefore, when the data transmission protection of the H.264 code stream is performed by using the improved Tornado code of the present invention, the video data can be first classified into two categories according to the value of Nal_ref_idc in the H.264 NAL unit: The image data, that is, the H.264 NAL unit whose value is 1 for Nal_ref_idc, and the H.264 NAL unit whose value is 0 for Nal_ref_idc. Then, the important image data is protected by data transmission using a modified Tornado code with higher redundancy, that is, a lower code rate and strong protection capability, such as an improved Tornado code using TN (30, 20); Data transmission protection can be performed using an improved Tornado code with less redundancy, ie higher code rate and weaker protection, such as the improved Tornado code using ΤΝ(24, 20).

In other technologies, it is also possible to determine the type of data according to the content in the predetermined information carried by the data, thereby determining the degree of importance according to the type of the data, and then using different modified Tornado code redundancy for data of different importance levels. Degree to achieve non-equal weight data transmission protection methods.

Through this non-equal weight data transmission protection algorithm, it can effectively ensure that all kinds of important information can be correctly recovered with high probability in a high packet loss environment, thereby ensuring the correctness of data transmission as much as possible, and thereby ensuring video. Image quality recovery.

The following is based on a data type in H.264, and the number of erasure-correction codes provided by the present invention includes the following steps:

Step 1: First, the improved Tornado code TN(M+N, N) of the present invention is determined, where: N is the number of data nodes and M is the number of check nodes. Then, determine a random bipartite graph between the data node and the check node.

Step 2: The sender continuously collects S H.264 NALUs, and the S H.264 NALUs Composing a large block, and when the size of the large block is not an integer multiple of the data node size, the bit block is filled with bits, so that the size of the block after filling is an integer multiple of the size of the data node. . This large block is then divided into N small blocks of size L1 bits as N data nodes.

The method of filling a bit for a large block may be to add a number of consecutive 0s at the end of the large block.

Step 3: Perform XOR operations on the N data nodes according to the random bipartite graph determined in step 1, and generate M check nodes.

Step 4: N data nodes and M check nodes are packetized according to a specific network protocol, and sent to the receiving end through the network.

Step 5: The sender determines whether all H.264 NALUs have been processed. If yes, go to step 6. Otherwise, go to step 2.

Step 6: The communication process at the sender ends.

The above process describes the sender. At the receiving end, after receiving a batch of data nodes and check nodes, it should first determine which data nodes are data nodes that should be received but actually lost and recoverable. The general decoding process of the Tornado code improved in accordance with the present invention then decodes and recovers the lost but recoverable data nodes. The receiving end repeats the above receiving and recovery process until the communication process ends.

When the present invention is actually applied to the H.264 technology, the video code stream can be divided into multiple sub-streams according to the data type, and the data stream for transmitting the corresponding data type is determined according to the correspondence between each data type and each data transmission protection level. The data transmission protection level. For example, the video code stream can be divided into two sub-streams according to the NALU attribute, for example: the NALU in which the important image data is composed can be composed into one sub-stream, the NALU of the normal image data is composed into a sub-stream, and then the important image data sub-flow is adopted. An erasure code with high redundancy and strong protection capability, and using the above steps for data transmission protection processing; and for the number of ordinary images According to the substream, an erasure code with low redundancy and weak protection capability is used, and the above steps are used for data transmission protection. Therefore, all kinds of important data can be restored as much as possible in a high packet loss environment, and thus the data stream can obtain maximum data transmission protection when the average redundancy rate of the erasure codes is the same.

In the following, it is experimentally demonstrated that when the data is protected by the improved Tornado code of the present invention, the protection capability of the data is not significantly reduced in terms of protection against transmission using the Tornado code of the prior art.

Experiment 1: Under the same conditions, the improved Tornado code of the present invention is compared with the Tornado code of the prior art for the recovery capability of the data node.

The improved Tornado code of the present invention is set to TNnew (32, 16), i.e., an improved Tornado code having 16 data nodes and a total number of nodes of 32. Set the Tornado code in the prior art to TNorg (32, 16), and the number of layers of the intermediate check layer is m=3, and the ratio of the number of nodes of the data node layer and the check node layer is equal to the ratio of the value of the node. , that is, an erasure code having 16 data nodes and a total number of nodes of 32. Since "o=l/2", the Tornado code in the prior art needs to be divided into five layers, wherein one layer is a data node layer and the fourth layer is a check node layer.

In the case of losing the number of different data nodes, the recovery probability of using the improved Tornado code of the present invention and the Tornado code of the prior art to the data node is as shown in Table 1 below. '

Number of lost nodes 1 2 3 4 5 6 7 8

TNnew 100 100% 99.9% 99.47 98.38 96.17 92.53 86.67

% % % % % %

TNorg 100 100% 100% 99.9% 99.3% 97.7% 94.1% 87.5%

%

Number of lost nodes 9 10 11 12 13 14 15 16

TNnew 78.2 67.38 53.94 39.32 5.28% 1.17%

3% % % %

TNorg 76.7 62.6% 47.4% 31.5% 18.6% 9.3% 3.5% 1.2%

% Table 1

It can be seen from the above Table 1 that the Tornado code of the prior art and the improved Tornado code of the present invention have the same number of check nodes and the same number under the condition of the same redundancy overhead.

O C ^

According to the number of nodes, using the improved 'Tornado code' of the present invention and the Tornado code in the prior art, when recovering lost data nodes, the recovery probability of the two is very small, as in the worst case. That is, 16 data nodes are also lost. The recovery probability of the Tornado code in the prior art is 1.2%, and the recovery probability of the improved Tornado code of the present invention is 1.17%.

It can be seen that the improved Tornado code of the present invention does not significantly decrease with respect to the Tornado code in the prior art in terms of data transmission protection capability, and the improved Tornado code of the present invention generates the check node layer by reducing the erasure code. The amount of calculation and the delay of data transmission are reduced, thereby improving the cost performance of the data transmission protection performance, and the data transmission protection method of the present invention can be applied to new technologies such as H.264.

Since the Peak Signal-to-Noise Ratio (PSNR) is an important metric for video image quality, the following is an experimental example of the present invention in which the improved Tornado code is used in the video image using the same number of check nodes. Important data and non-important data When video data transmission is protected, and two improvements with different check nodes are used.

The Tornado code compares the PSNR of the video data transmission protection between the important data and the non-important data in the video image.

Experiment 2: The improved Tornado code for the number of single face nodes used is TN (28, 20). The improved Tornado codes with two different check nodes are TN1 (30, 20) and TN2 (26, respectively). , 20).

Among them, the code rate of TN(28, 20) is the same as the average bit rate of TN1 (30, 20) and TN2 (26, 20). Therefore, these two strategies are used to protect video data transmission, which is redundant. The overhead aspect is exactly the same.

When testing Foreman image sequences, the PSNR comparison results are shown in Table 2 below.

Table 2

Among them, PSNR1 corresponds to TM(28, 20), and PSN2 corresponds to TN1(30, 20) and TN2(26, 20).

When testing the Container image sequence, the PSNR comparison results are shown in Table 3.

table 3 Where: PSNR1 corresponds to TN(28, 20), and PS R2 corresponds to TN1 (30, 20) and TN2 (26, 20).

It can be clearly seen from Table 2 and Table 3 that under the condition that the communication network has the same packet loss rate, the important data of the H.264 and the common data are protected by the erasure codes of two different protection capabilities. The PSNR of the video data transmission protection is improved by using only one type of protection erasure code, but the redundancy overhead is completely the same, so that the invention has a good advantage in data transmission protection capability.

While the invention has been described by the embodiments of the invention, it will be understood that

Claims

Claim

A data transmission protection method based on an erasure code, characterized in that the method comprises the steps of:

a, setting an erasure code having only one layer of check node;

b. Perform data transmission protection according to the erasure code.

2. The method of claim 1, wherein the step a comprises:

A2. Determine a random bipartite graph and a predetermined algorithm between the data node layer and the check node layer.

3. The method according to claim 2, wherein the step a1 is: determining a size, a data node of the data packet according to a data transmission rate, a data type, a data protection capability requirement, and/or a network delay requirement. The number n of data nodes contained in the layer and the number L of check nodes contained in the check node layer.

The method according to claim 2, wherein the step a2 is: determining a left degree distribution vector and a right degree distribution vector of the bipartite graph;

The method according to claim 4, wherein the step b comprises: bl, at the transmitting end, according to the size of the data packet corresponding to the data node in the data node layer of the erasure code, the data code The flow is divided into n data nodes; B2, generating, according to the predetermined algorithm and the random bipartite graph of the erasure code, only one layer of check node layers having L check nodes for the n data nodes;

B3, transmitting the n data nodes and L check nodes to the receiving end;

The method according to claim 5, wherein the data code stream comprises: a data stream based on a network adaptation layer encapsulated data unit or a network abstraction layer encapsulated data unit;

And the step b1 is: at the transmitting end, forming a predetermined number of data units based on the network adaptation layer or the network abstraction layer encapsulation into a bit block, and then according to the size of the data node in the data node layer of the erasure code The bit block is segmented into data nodes.

The method according to claim 6, wherein: the data unit encapsulated by the network adaptation layer comprises: a data unit encapsulated based on an H.263 network adaptation layer; and the data encapsulated by the network abstraction layer The unit includes: a data unit encapsulated based on the H.264 network abstraction layer.

The method according to claim 6, wherein the step bl further comprises: when determining that the size of the bit block is not an integer multiple of a data node size in a data node layer of the erasure code, The bit block is padded such that the size of the bit block after padding is an integer multiple of the data node size in the data node layer of the erasure code.

The method according to any one of claims 1 to 8, wherein the step a further comprises:

Setting an erasure code having a check node layer and having different check nodes for each data transmission protection level; The step b includes:

Dividing the data stream into corresponding substreams according to the data transmission protection level;

The data transmission protection is performed on each substream according to the erasure code set for the data transmission protection level corresponding to each substream.

10. The method according to claim 9, wherein the predetermined information comprises: non-texture information and texture information of a data unit based on an H.264 network abstraction layer.