AU2020101634A4

AU2020101634A4 - Filtering dimension reduction decoding method

Info

Publication number: AU2020101634A4
Application number: AU2020101634A
Authority: AU
Inventors: Xiujuan DU; Chong Li; Lijuan Wang
Original assignee: Qinghai Normal University
Current assignee: Qinghai Normal University
Priority date: 2019-08-22
Filing date: 2020-08-03
Publication date: 2020-09-10
Anticipated expiration: 2028-08-03
Also published as: CN110535562A; CN110535562B

Abstract

The invention provides a filtering dimension reduction decoding method, and solves the problem that a decoding algorithm of RLT depends on a coding packet of degree 1, the loss or error code of the coding packet of degree 1 occurs in a channel propagation process, or a short cycle problem exists in a generated matrix, so that the decoding is easy to fail. According to the method, mathematical analysis is carried out on the short cycle problem, the strict short cycle problem is defined and analyzed on the basis, the strict short cycle is further utilized, an FDR decoding algorithm is provided, decoding defects of RLT are overcome, and the decoding success rate is increased. DRAWINGS 51 0 1 1 S2 0 10 01 53 4l1 L1 S4 5 1 L 0 1 Y Y2 Y3 Y4 FIG. 3 Decoder Entry (Store Coding Packets F of d E (4,k]) (Store Coding Packets \ of d= 4) (Store Coding Packets of d= 3) (Store Coding Packets of d = 2) Flow Direction of Coding Packets (Store Original Packets) Flow Direction of Secondary Coding Packets FIG. 4 2

Description

DRAWINGS

51 0 1 1

S2 0 10 01

53 L1 4 l 1 S4 5 1L 0 1 Y Y2 Y3 Y4 FIG. 3

Decoder Entry

(Store Coding Packets F of d E (4,k])

(Store Coding Packets \ of d= 4)

(Store Coding Packets of d= 3)

(Store Coding Packets of d = 2) Flow Direction of Coding Packets

(Store Original Packets) Flow Direction of Secondary Coding Packets

FIG. 4

DESCRIPTION FILTERING DIMENSION REDUCTION DECODING METHOD FIELD OF THE INVENTION

[0001] The invention relates to the technical field of network communication, in particular to a filtering dimension reduction decoding method, which is used for the transmission of multimedia information in an underwater sensor network.

DESCRIPTION OF THE RELATED ART

[0002] The underwater sensor network (UWSN) has wide application prospects in environmental monitoring, resource exploration, disaster prevention, etc., which has attracted great attention of researchers. Due to different applications, people's demand for underwater data formats is not limited to lightweight text information. The transmission of different forms of multimedia information, such as pictures of several K in size and even up to several M of underwater real-time video, has gradually become the research interest of scientists. The larger the amount of information, the stricter the real-time and reliability requirements for data transmission; however, the reliability of the underwater link deteriorates due to the existence of more adverse factors. Therefore, the multimedia information transmission of underwater data is closely related to the reliable transmission technology. Digital foundation codes (DFC) is an efficient code-free error correction code; compared with the coding method with fixed code rate, it has more reliable and excellent performance, and shows good adaptability to the underwater acoustic network with harsh communication environment.

[0003] There are problems in the existing technology: the decoding algorithm of the digital foundation codes mainly includes two types: BP algorithm and GE algorithm. BP (belief propagation) algorithm depends too much on the coding packet of degrees, which makes the utilization of coding information low. Compared with solving linear equations, the decoding process of GE (Gaussian elimination) algorithm has a higher computational complexity.

[0004] The applicant has proposed a simple and efficient digital foundation code -RLT

coding and decoding algorithm; RLT uses recursive coding, which has high coding and decoding efficiency and a greatly increased probability of successful decoding. However, a decoding algorithm of RLT depends on a coding packet of degree 1; if the loss or error code of the coding packet of degree 1 occurs in a channel propagation process, or a short cycle problem exists in a generated matrix, the decoding is easy to fail. According to the method,

DESCRIPTION mathematical analysis is carried out on the short cycle problem, the strict short cycle problem is defined and analyzed on the basis, the strict short cycle is further utilized, an FDR decoding algorithm is provided, decoding defects of RLT are overcome, and the decoding success rate is increased.

SUMMARY OF THE INVENTION

[0005] The key technical problem to be solved by the invention is that a decoding algorithm

of RLT depends on a coding packet of degree 1, if the loss or error code of the coding packet

of degree 1 occurs in a channel propagation process, or a short cycle problem exists in a

generated matrix, the decoding is easy to fail.

[0006] In order to solve the technical issues above, the invention adopts the following

technical solutions: a filtering dimension reduction decoding method, comprising the

following steps:

[0007] (1) the receiving end using FDR decoding algorithm starts the decoding process when

it receives the first coding packet; as long as there is a properly including relationship

between the corresponding set of original packets participating in coding and the set of

original packets participating in coding corresponding to the coding packets received

subsequently, there is no need to wait for the coding packet of degree 1, which shortens the

decoding time to a certain extent;

[0008] (2) in order to increase the success rate of FDR decoding, increasing the probability of

coding packets of large degree under the premise that the number of coding packets of degree

1 is appropriate, so that the average degree of coding packets is small; at the same time, the

problem of imperfect coverage is taken into account; in the case of the number of original

packets is large, it can not only control the decoding to be not too complicated, but also

ensure a higher success rate of decoding;

[0009] the degree distribution of the coding packet is designed as follows: Ink vrk(d+1) Q~d fp(d)+-r(d) d 2,3,4

1 d =k

p(d) (d = 2,3,4)

DESCRIPTION

(d)= d(d+1),(d = 2,3,4)

[0010] wherein k indicates that each data block comprises k original data packets, and the coder codes the k original data packets; d is the degree of the coding packet, d E{1,2,...,k}; Q(d) represents the probability distribution of the coding packets of degree d;

[0011] (3) the FDR decoder adopts a layered design, and there are 5 kinds of degrees of the coding packet, which are: d = 1, d = 2, d = 3, d = 4, and d = k, and the decoder is accordingly

designed as 5 layers: li, /2, 13, 14, and lk, which respectively store the coding packet of the corresponding degree; the coding packet in the layer comprises both the received coding packets and the secondary or multiple decoding packets generated after XOR processing; it should also be noted that the layer Ik stores coding or decoding packets of a degree range of d E(4,k].

[0012] The FDR decoding method above adopts the following decoding process:

[0013] (1) the FDR algorithm stipulates that after the transmitting end completes data coding, it first transmits the coding packet Yn of d = k, and then transmits the coding packets of d= 4, d= 3, d= 2, and d= 1 in sequence; the decoding process of the receiving end is as follows;

[0014] (2) receiving the first coding packet Y, for d(Yn )= k, putting it in the layer lk;

[0015] (3) comparing the received coding packets Yi from the layer lkwith the coding packets in the layer one by one, to judge whether the XOR condition is met;

[0016] (4) if Yi and a certain coding packet Y meet the XOR condition and d(Y) < d(Y), then Ysec = Yi 1 Y, and putting Yi of a smaller degree into the corresponding layer; discarding Yi of a larger degree directly and no longer saving; calculating the degree d(Ysec) of the generated secondary coding packet and putting it into the corresponding layer; if the coding packet does not meet the XOR condition until the layer li, then putting it into the corresponding layer according to the degree;

[0017] (5) repeating steps 3-4 until all the coding packets are received; at this time, if the vector S of the layer li contains all the original packets, the decoding is successful;

[0018] (6) if there are still original packets that have not been recovered, then performing XOR on the original packet of the layer li with the coding packets of layers 2,/3, and4 layer by layer until the decoding is successful.

DESCRIPTION BRIEF DESCRIPTION OF THE DRAWINGS

[0019] FIG. 1 is two cases of stopping set;

[0020] FIG. 2 is the decoding termination phenomenon of LT code; wherein white circles are original data packets, and black circles are coding packets;

[0021] FIG. 3 is an example of a 4-membered short cycle;

[0022] FIG. 4 is a conceptual diagram of the FDR decoder;

[0023] FIG. 5 is formula 1-6;

[0024] FIG. 6 is the optimized degree distribution functional operation;

[0025] FIG. 7 is a comparison of coding and decoding complexity.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0026] Unless otherwise specified, the methods and devices used in the following embodiments of the patent of the invention are conventional methods and devices; the equipment and reagents used are the conventional equipment and reagents purchased by the reagent company. In order to make the purpose, technical solution and advantages of the patent of the invention clearer, the specific implementation of the patent of the invention will be described in detail hereinafter with reference to specific embodiments. Examples of these preferred embodiments are exemplified in specific embodiments.

[0027] It should also be noted here that in order to avoid obscuring the technical solutions of the patent of the invention due to unnecessary details, in the embodiments, only technical solutions and/or processing steps that are closely related to the solutions according to the invention are shown, and other details that are not relevant are omitted.

[0028] Embodiment 1

[0029] The recursive RLT coding and decoding proposed previously has high efficiency. However, the decoding algorithm of RLT depends on a coding packet of degree 1, the loss or error code of the coding packet of degree 1 occurs in a channel propagation process, the decoding will fail. When the decoding is terminated in advance, there are still useful coding packets (coding packets of a degree greater than 1) in the received coding packets, and the set of these coding packets is the stopping set. There are two phenomena to explain the stopping set problem: 1) there is no connection between some original packets and any coding packet, as shown in FIG. 1 (a); the original packet symbols Si and S2 can be obtained through XOR operation, but S3 does not participate in the coding of any coding packet, so S3 cannot be obtained anyway; 2) the degrees of the remaining unresolved coding packets in the decoding process are all greater than 1, that is, there is no coding packet of degree 1, which

DESCRIPTION results in the termination of the decoding process. As shown in FIG. 1 (b), the degrees of the remaining coding packets Yi, Y2 , and Y3 are all greater than 1, and the decoding cannot be continued at this time.

[0030] The first case is caused by randomly selecting the original packets for XOR in the coding process; because the random selection method makes certain original packets have a certain probability of not being selected, which thereby directly leads to these original packets to be failure to participate in the coding of any coding packet, and this situation is also called the problem of imperfect coverage (IC). The second case is the short cycle problem to be analyzed next.

[0031] Short Cycle Problem

[0032] Definition 1: in a generated matrix, if there are two columns, wherein the two rows (or more than two rows) at the same position are both "1", then these rows composed of "1" form a closed cycle, which is called a "short cycle".

[0033] If there are two rows that satisfy the definition of short cycle, then the short cycle formed by these two rows is a 4-membered cycle, and if there are three such rows, then the short cycle formed by these three rows is a 6-membered cycle, and so on; assuming that there are k'(2 k'<k) rows that satisfy the definition of short cycle, then the short cycle formed thereby is a (2k')-membered cycle. The termination phenomenon in the decoding process shown in FIG. 2 is regarded as an example to explain "short cycle".

[0034] In FIG. 4-3, Yi = S2, Y2 = S2 @S3 @S4, Y3 = Si o S2 @ S3, Y4 = Si S2 @ S3 S4.

According to the LT code decoding process, first finding the coding packet Yi of degree 1, at this time S2 can be directly coded, and removing the connection between Yi and S2; next, performing XOR operation on all coding packets {Y 2 ,Y 3,Y 4 } connected to S2 with S2, and updating the values of Y 2, Y 3, and Y4 to be the results of XOR operation, and deleting the connection therebetween. At this time, the degrees d of the remaining coding packets Y2 , Y3 ,

and Y4 are d(Y2) = 2, d(Y3) = 2 and d(Y 4) = 3, respectively. According to the LT code decoding rules, it is necessary to find the coding packets of degree 1, but the degrees of the three coding packets are all greater than or equal to 2, so the decoding is forced to terminate. At this time, the generated matrix corresponding to Y 2, Y3 , and Y4 is shown in FIG. 3.

[0035] The generated matrix contains 2 short cycles with a cycle length of 4: the two columns corresponding to Y3 and Y4 are both "1" in the first and third rows, so the first and third rows form a 4-membered short cycle, as shown by the dotted line in FIG. 4-4; similarly, the two columns corresponding to Y2 and Y4 are both "1" in the third and fourth rows, so

DESCRIPTION these two rows form the second 4-membered short cycle, as shown by the dotted line in FIG. 4-4. Taking the 4-membered cycle as an example: assuming that the coding parameter of the LT code is (n, k, Q), wherein k is the number of original packets, n is the number of coding packets, and Q is the degree distribution function of the coding packets; in the generated matrix G of order n X k, the generating probability of the symbol of degree i is Qi, and the

generating probability of the symbol of degree is Qj; then, the probability that a certain two columns of degree i and j (i>1, j>1) in the matrix G form a 4-membered short cycle is shown as formula (1) in FIG. 5.

[0036] Strict Short Cycle

[0037] Now the definition of strict short cycle is given in the invention; definition 2: in a generated matrix, if there are two columns, wherein the two rows (or more than two rows) at the same position are both "1", and all the rows in one column except these two rows (or more rows) are all "0", then these rows composed of "1" form a closed cycle, which is called a "strict short cycle". Assuming that there are k'(2 k'<k) rows that satisfy the definition of strict short cycle, the short cycle formed thereby is a (2k')-membered cycle; then, the probability that the column of degree j and a certain column of degree m (m<j) in the matrix G form a strict 2m-membered short cycle is shown as formula (4-3) in FIG. 5.

[0038] Embodiment 2

[0039] There are two main types of decoding algorithms for foundation codes: belief propagation (BP) algorithm and Gaussian elimination (GE) algorithm. The BP algorithm requires the code of degree 1 as an iterative start, and codewords with reasonable design for degree distribution have excellent performance. However, due to the excessive reliance on the code of degree 1, the utilization rate of coding information is not high, and the calculation speed of medium and long codewords is slow. The GE algorithm uses the maxi-mum likelihood (ML) decoding algorithm for linear codes, which is equivalent to solving linear equations, and can be applied to the decoding of various foundation codes and can make full use of coding information; however, due to its high computational complexity, as the length of the input information increases, the efficiency will drop significantly. Each of the two decoding algorithms has its own scope of application. It is necessary to choose the decoding algorithm according to actual needs, and it is also possible to consider combining the two algorithms at the same time to improve decoding efficiency.

[0040] The invention provides a filtering dimension reduction decoding algorithm-FDR algorithm (filtering dimension reduction decoding algorithm). The FDR algorithm uses XOR

DESCRIPTION operation between the coding packets that can form a strict short cycle in the generated matrix, to generate packets of degree of 1 or to perform "dimension reduction" on the coding packets of a higher degree to reduce the degree. Unlike the BP decoding algorithm, the FDR decoding algorithm not only relieves the dependence on the coding packet of degree 1 generated by the coder at the transmitting end, but also can quickly reduce the degree of the coding packets of high degree to further reduce the decoding complexity.

[0041] Decoding Idea

[0042] In the analysis of the short cycle problem in the previous section, when the LT code decoding termination phenomenon in FIG. 2 occurs, the generated matrix formed by the three coding packets Y2, Y3, and Y4 contains two 4-membered strict short cycles. Next, it is analyzed from the perspective of the set of original packets participating in coding. The set of original packets participating in coding corresponding to the coding packet Y2 is {S3,S4}, the set of original packets participating in coding corresponding to the coding packet Y3 is {Si,S3}, and the set of original packets participating in coding corresponding to the coding packet Y4 is {Si,S3,S4}. There is a clear inclusion relationship between the three sets: {S3,S4} {Si,S3,S4}, {S,S3} {S,S3,S4}. If performing XOR operation on Y2 and Y4, we can get Si = Y2 @ Y 4 ; if performing XOR operation on Y3 and Y4 , we can get S4 = Y3 @ Y4; if performing XOR

operation on Si and Y3, we can get S3 = Si o Y3. Originally, the BP decoding algorithm can

only depend on Yi of degree I to code S2 , the decoding operation of S2 is Cs 2 = 1, while Si, S3, and S4 cannot be solved at all, so it can be considered that the decoding operations Csi, Cs3, and Cs4 of Si, S3 , and S4 are close to infinity: Csi, Cs3, Cs4 -- o, then the decoding operations of all original packets are obtained as Caii = Csi + Cs2 + Cs3 + Cs4 -- o. However, the three original packets Si, S3, and S 4 can be solved by XOR between the coding packets. The decoding operations of Si, S2, S3, and S4 are C'si = 1, C's2 = 1, C's3 = 1, C's4 = 1, respectively, then the decoding operation of successful decoding of all original packets is C', = C'si +

C's2 + C's3 + C's4 = 5 « Ca1 .

[0043] Apparently, the strict short cycle is a waste of coding, and is of no value to the traditional decoding method of depending on the coding packet of degree 1; however, if the decoding method is changed to take advantage of the strict short cycle, the contribution of the strict short cycle will not be neglected, and may even become a key step in decoding the remaining unsuccessful original packets. Therefore, there are the following conclusions:

[0044] Conclusion 1: in the generated matrix, as long as the two columns that form the short cycle have different degrees, and there is a "1" in the column of a smaller degree except for

DESCRIPTION the row that forms the short cycle, and there is not a "1"in the other rows, XOR operation can be performed between the corresponding coding packets. The degree of the new coding packet (which is called the secondary coding packet) generated by XOR of the two must be smaller than one of the two, or even smaller than the degree of the both. If the degree of the two differs by 1, a packet of degree 1 is directly obtained by XOR; if the degree of the two differs more than 1, after XOR, the degree of the column of a larger degree can be reduced to the difference between the two degrees.

[0045] The LT code decoding process should be carried out after receiving a certain number of coding packets; the RLT code is improved here: after the data coding is completed, the coding packet of degree 1 is transmitted first, and the decoding process of the RLT code starts immediately when the coding packet is received. However, both of the decoding methods depend on the coding packet of degree 1 to start decoding. While the decoder with FDR algorithm can start decoding after receiving the first coding packet (regardless of whether the degree is 1). Still taking FIG. 4-3 as an example, if the first coding packet received by the receiving end adopting the FDR decoding algorithm is Y4, and the second received coding packet is Y3, the receiving end will compare the set of original packets participating in coding of the two, and if there is a properly including relationship between the two sets, then the receiving end can start the decoding process and perform XOR operation on the two coding packets.

[0046] The invention provides a filtering dimension reduction (FDR) decoding method.

[0047] Embodiment 3

[0048] Next, the design and decoding process of the decoder will be mainly introduced. First, several parameters in the FDR decoding algorithm are given:

[0049] definition 4: there are k inputting symbol vectors S: S = {Si,S2,...Sk}, coding k inputting symbols to generate n coding symbol vectors Y: Y = {Yi,Y 2,...,Yi...,Y}, and the degree of the coding packet Yi is expressed as d(Y). The FDR algorithm divides the coding packet into two types: in addition to the n coding packets Y 1 ,Y2,...,Yn generated by k original

packets through the coder at the transmitting end, it also includes secondary coding packets generated by XOR of the coding packets with other coding packets in the decoder at the receiving end. In order to distinguish, the secondary coding packet is expressed as Ysec, then the degree of the secondary coding packet is expressed as d(Ysec); the ID set of an coding packet (whether it is one Y of Y1 ,Y 2,...,Yn or a secondary coding packet Ysec) corresponding to the original packet participating in the coding is T.

[0050] Design of the Decoder

DESCRIPTION

[0051] The decoder adopts a layered design according to the degree range of the coding packet. In the degree distribution function proposed by the invention, there are five kinds of degrees of the coding packet, which are: d= 1, d= 2, d= 3, d= 4, and d= k, and the decoder is accordingly designed as 5 layers: li,/2, /3, 14, and lk, as shown in the conceptual diagram of the decoder in FIG. 4. Each layer of the decoder stores the coding packet of the corresponding degree. Here, the coding packet in the layer comprises both the coding packets from the receiving end and the secondary coding packets generated by XOR with the existing (first entered into the decoder) coding packets already in the layer after the coding packets from the receiving end enter the decoding and flow between the layers. For example, the layer 12 stores all the coding packets of degree 2; among these coding packets, there may be received coding packets of degree 2 from the transmitting end, and there may also be a secondary coding packet of degree 2 generated by XOR of a coding packet of degree 3 with a coding packet of degree 1 in the layer Ii. It should be further noted that the layer Ik stores coding packets of a degree range of d E (4,k]. The coding packet of degree k can be performed XOR with any coding packet, then the degree of the secondary coding packet generated by XOR of a coding packet of degree k (assuming k > 8) with a coding packet of degree t (t = 1, 2, 3, 4) is k - t, and k - t > 4; therefore, the secondary coding packet of a degree greater than 4 and less than k is placed in the layer k. Taking the coding packet of degree d = k = 10 of XOR operation at the layer 14 as an example, the degree of the generated secondary coding packet Ysec is d = k - 4 = 6, so the secondary coding packet Ysec is placed in the layer lk. The existence form of the coding packet in the decoder is a pair of key-value, key -the ID set T of the original packet corresponding to the coding packet participating in the

coding, value -the coding packet. For example, the coding packet Yi= Si f S2 S3, the

corresponding IDs of Si, S2, and S 3 are 0, 1, and 2, respectively, Tyi= {0,1,2} then the existence form of Yi in the layer/3 is {0,1,2}-Yi. After the coding packet enters the decoder, it flows down and passes through every layers of the decoder in sequence; when it reaches a certain layer, if there is a properly including relationship between the T of the coding packet and the T of a certain coding packet in the current layer, in other words, if all the original data packets in the coding packet of a smaller degree also participate in the XOR operation (this is an XOR condition) of the coding packet of a larger degree, then after the XOR operation of the two, the coding packet of a larger degree is updated to the secondary coding packet, and the degree thereof is lowered. This process is equivalent to filtering and dimension reducing the coding packets of a larger degree, and to reduce the degree of the coding packets of a

DESCRIPTION larger degree, which accelerates the decoding progress to a certain extent. Coding packets of a larger degree or a smaller degree have a higher probability of XOR operation with other coding packets in a certain layer.

[0052] Theoretically, there is no properly including relationship between the layers 2,13, 14,

and Ik of the coder and each T set in all layers except li below it. The FDR decoder has been in the decoding state since receiving the first coding packet, and the decoding is regarded as successful until the layer li contains the entire vector S.

[0053] FDR Decoding Process

[0054] the FDR algorithm stipulates that after the transmitting end completes data coding, it first transmits the coding packet Yn of d = k, and then transmits the coding packets of d = 4, d = 3, d= 2, and d= 1 in sequence; the decoding process of the receiving end is as follows;

[0055] (1) receiving the first coding packet Y, for d(Yn )= k, putting it in the layer Ik;

[0056] (2) comparing the received coding packets Yi from the layer Ik with the coding packets in the layer one by one, to judge whether the XOR condition is met;

[0057] (3) if Yi and a certain coding packet Y meet the XOR condition and d(Y) < d(Y), then Ysec = Yi 1 Y, and putting Yi of a smaller degree into the corresponding layer; discarding Yi of

a larger degree directly and no longer saving; calculating the degree d(Ysec) of the generated secondary coding packet and putting it into the corresponding layer; if the coding packet does not meet the XOR condition until the layer li, then putting it into the corresponding layer according to the degree;

[0058] (4) repeating steps 3-4 until all the coding packets are received; at this time, if the vector S of the layer li contains all the original packets, the decoding is successful;

[0059] (5) if there are still original packets that have not been recovered, then performing XOR on the original packet of the layer li with the coding packets of layers 2,/3, and14 layer by layer until the decoding is successful.

[0060] Optimize the Degree Distribution Function

[0061] The degree distribution function of the digital foundation codes affects the coding complexity and is related to the coding efficiency. A reasonable degree distribution should increase the selection probability of coding packets of large degree under the premise that the number of coding packets of degree 1 is appropriate, so that the average degree of coding packets generated by the degree distribution is small; at the same time, the problem of imperfect coverage is taken into account to ensure good coverage of all original packets

[0062] k original data packets are given, the degree distribution is shown in the formula (4) in

DESCRIPTION FIG. 5: the expression ofp(d) in formula (4) is formula (5) in FIG. 5: the expression of r(d) in formula (5) is formula (6) in FIG. 5.

[0063] When d = 1, the number of coding packets is related to k; the larger the number k of original packets, the greater the number of coding packets of degree 1, and the more beneficial to speed up the decoding speed; when d= k, the number of coding packets is 1, to ensure that all original packets participate in the coding of the packet, which avoids the problem of imperfect coverage. The design of d = 2, d = 3, and d = 4 is based on the ideal soliton distribution, and the selection probability of degree 2, degree 3, and degree 4 is set to a number less than 1, so that the optimization of the degree distribution is transformed into a constant design problem.

[0064] From the optimized degree distribution function, we can see the operation in FIG. 6:

[0065] generally, Ink is a number no greater than 0.3, so X ~ 3.57, the decoding complexity

of the improved RLT code is 3.57, which is not related to the number k of original packets. The decoding complexity comparison between the RLT code, the improved RLT code and the foundation codes using the traditional degree distribution is shown in Table 1 of FIG. 7.

[0066] Advantageous effects: the invention is a filtering dimension reduction decoding method, which has the following advantageous:

[0067] The receiving end starts the decoding process when it receives the first coding packet; as long as there is a properly including relationship between the corresponding set of original packets participating in coding and the set of original packets participating in coding corresponding to the coding packets received subsequently, there is no need to wait for the coding packet of degree 1, which shortens the decoding time to a certain extent. The FDR algorithm uses XOR operation between the coding packets that can form a strict short cycle in the generated matrix, to generate packets of degree of 1 or to perform "dimension reduction" on the coding packets of a higher degree to reduce the degree. Unlike the BP decoding algorithm, the FDR decoding algorithm not only relieves the dependence on the coding packet of degree 1 generated by the coder at the transmitting end, but also can quickly reduce the degree of the coding packets of high degree to further reduce the decoding complexity.

[0068] The NS3 simulation platform is used to perform simulation comparison with the RLT code in terms of the successful rate of decoding, and the number of times that the FDR algorithm is used to generate secondary decoding in about 100 times of data transmission is counted. A large number of simulation experiments prove that the FDR algorithm has better

DESCRIPTION performance than the RLT algorithm. The invention has an ideal effect and a promising application prospect.

[0069] Embodiment 4

[0070] The invention provides a transmission method of multimedia information in an underwater sensor network, comprising the following steps:

[0071] After the problems of degree distribution and coding of digital foundation codes are solved, next, a reliable transmission control mechanism for real-time communication between nodes needs to be implemented based on the coding and decoding scheme. The underwater acoustic Modem usually works in half-duplex mode, and reception and transmission cannot be performed at the same time; an available transmission control mechanism should be able to avoid transmission-reception interference caused by transmitting packets to nodes in the transmission state. So far, most access control protocols use RTS/CTS handshake mechanism to dynamically coordinate node transmission. The rate of generating data by the underwater sensor network is 1-5 bps, the optimized packet length is 100 bytes, and the RTS/CTS packet length is tens of bytes. Therefore, compared with the frame size of data packets, that of the RTS/CTS is not very short, and the benefits brought by the RTS/CTS handshake mechanism are not significant. On the contrary, considering the narrow bandwidth and long delay of underwater acoustic channels, RTS/CTS handshake reduces the channel utilization, network throughput, and extends the end-to-end delay. Therefore, a reliable transmission control scheme based on our coding and decoding can use a transmission mechanism without RTS/CTS handshake.

[0072] The source node first divides the original data packet into k-sized blocks, that is, each data block contains k original data packets. The source node codes k original packets. In an underwater sensor network, the time required to transmit a data block containing 50 optimized packets is about 60 seconds, which can meet the limited transmission time between two nodes. The transmission time is controlled by appropriately setting the block size, so that the receiving node can receive enough coding packets to reconstruct the original data packet, and realize reliable transmission block by block and hop by hop.

[0073] In the block-by-block and hop-by-hop transmission based on coding and decoding, those nodes that are transmitting packets are considered to be in a transmitting state. In order to avoid synchronization conflicts between data frames and ACKs and reduce excessive redundancy, a compromise is made between transmission efficiency and fairness, and we define two transmission constraints: 1) the maximum number of data packets allowed to be transmitted in one transmission phase is N. 2) The minimum time interval between two

DESCRIPTION transmission phases of the same node is Ta, and the node waiting for the expiration of Ta is considered to be in a transmission avoidance stage. Considering that the conversion delay between the transmission and reception states of the acoustic Modem is relatively large, usually in seconds, here Ta = 2RTT can be set to replace Ta = RTT.

[0074] In the first transmission stage of a data block, the transmitting node transmits N coding packets. After that, the transmitting node switches to the receiving state and waits for the feedback information from the receiving node.

[0075] The feedback information from the receiving node includes the number Nr of received frames and the number ki of original packets that cannot be reconstructed. The transmitting node replaces k in the original degree distribution with ki in the second transmission stage of the data block, and performs secondary coding on the ki original packets, generates and transmits N 1 coding packets, and then switches the state to wait for feedback information. This is repeated until the decoding is successful.

[0076] The above is only the specific embodiment of the invention. It should be noted that

those of ordinary skill in the art can make some improvements and modifications without

departing from the principles of the invention, which shall fall within the protection scope of

the invention.

Claims

CLAIMS 1. A filtering dimension reduction decoding method, comprising the following steps:

(1) the receiving end using FDR decoding algorithm starts the decoding process when it

receives the first coding packet; as long as there is a properly including relationship between

the corresponding set of original packets participating in coding and the set of original

packets participating in coding corresponding to the coding packets received subsequently,

there is no need to wait for the coding packet of degree 1, which shortens the decoding time

to a certain extent;

(2) in order to increase the success rate of FDR decoding, increasing the probability of

coding packets of large degree under the premise that the number of coding packets of degree

1 is appropriate, so that the average degree of coding packets is small; at the same time, the

problem of imperfect coverage is taken into account; in the case of the number of original

packets is large, it can not only control the decoding to be not too complicated, but also

ensure a higher success rate of decoding;

the degree distribution of the coding packet is designed as follows: Ink k(d+1) Q~d) p(d)+-r(d) d- 2,3,4

1 d =k

p(d)= ,(d = 2,3,4)

(d)= d(d1) ,(d = 2,3,4)

wherein k indicates that each data block comprises k original data packets, and the coder

codes the k original data packets; d is the degree of the coding packet, d E{1,2,...,k}; Q(d)

represents the probability distribution of the coding packets of degree d;

(3) the FDR decoder adopts a layered design, and there are 5 kinds of degrees of the

coding packet, which are: d = 1, d = 2, d = 3, d = 4, and d = k, and the decoder is accordingly

designed as 5 layers: li, /2, 13, 14, and lk, which respectively store the coding packet of the

corresponding degree; the coding packet in the layer comprises both the received coding

packets and the secondary or multiple decoding packets generated after XOR processing; it

should also be noted that the layer Ik stores coding or decoding packets of a degree range of

d E(4,k].

CLAIMS 2. The filtering dimension reduction decoding method according to claim 1, wherein the FDR decoding method adopts the following decoding process: (1) the FDR algorithm stipulates that after the sending end completes data coding, it first sends the coding packet Y, of d = k, and then sends the coding packets of d= 4, d= 3, d = 2, and d= 1 in sequence; the decoding process of the receiving end is as follows; (2) receiving the first coding packet Y, for d(Yn )= k, putting it in the layerIk; (3) comparing the received coding packets Yi from the layer lkwith the coding packets in the layer one by one, to judge whether the XOR condition is met; (4) if Yi and a certain coding packet Y meet the XOR condition and d(Y) < d(Y), then Ysec = Yi 1 Y, and putting Yi of a smaller degree into the corresponding layer; discarding Yi of a larger degree directly and no longer saving; calculating the degree d(Ysec) of the generated secondary coding packet and putting it into the corresponding layer; if the coding packet does not meet the XOR condition until the layer li, then putting it into the corresponding layer according to the degree; (5) repeating steps 3-4 until all the coding packets are received; at this time, if the vector S of the layer li contains all the original packets, the decoding is successful; (6) if there are still original packets that have not been recovered, then performing XOR on the original packet of the layer li with the coding packets of layers 2, /3, and14 layer by layer until the decoding is successful.

D R AW I N G S 03 Aug 2020 2020101634

FIG. 1

FIG. 2

D R AW I N G S 03 Aug 2020 2020101634

FIG. 3

FIG. 4

D R AW I N G S 03 Aug 2020 2020101634

FIG. 5

FIG. 6

D R AW I N G S 03 Aug 2020 2020101634

FIG. 7