KR101412036B1 - Method for predicting decoding performance of approximate decoding of network coded source data using position similarity information - Google Patents

Abstract

The present invention relates to network coding, and more specifically, to a decoding performance prediction method for approximate decoding utilizing the position resemblance information of network-coded source data. The decoding performance prediction method according to an embodiment of the present invention receives network-coded data and uses the correlation of the received data to approximately recover lost data. The decoding performance index of the present invention is computed to have a higher value if the success probability of the decoding is higher. When the packet loss ratio is 50% or less, the decoding performance index converges to the maximum saturation level as the position resemblance information decreases. When the packet loss ratio is more than 50%, the decoding performance index converges to the minimum saturation level as the position resemblance information decreases.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention [0001] The present invention relates to a method of predicting decoding performance of approximate decoding using position similarity information of network-

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to network coding, and more particularly, to an approximate decoding technique for network coded data.

As mobile devices with wireless networking capabilities are routinely used, the importance of the technology to transmit information in a timely manner within such a dynamic network is becoming more and more dynamic as the wireless network continues to change by a number of different types of devices It is gradually getting higher.

Network coding is increasingly being used as a methodology for constructing an efficient distributed transmission algorithm in dynamic networks with various paths and sources. The existing routing-based network scheme simply forwards the data to point-to-point, so if the network configuration changes in the middle, the existing routing path is cut off or the speed is lowered to affect the quality of service and to maintain the routing path There is not much overhead.

On the other hand, network coding is a process by which neighboring nodes in the network repeatedly receive and output packets, thereby allowing data to flow to a destination without a predetermined routing path. Therefore, network coding can efficiently use bandwidth, power resources, And shows robustness to network fluctuations. As the source and path vary, that is, the number of nodes having the same information and the number of nodes requesting, the performance of such network coding can be maximized.

However, since the information transmitted through network coding is not transmitted as it is in the original data, but the network node modulates it through proper operation, there is a problem that the data can not be reconstructed when the symbols are incompletely received within a predetermined time have.

In particular, in the case of time delay sensitive data such as multimedia data, if the packet does not arrive within the allowable delay time, data decoding can not be performed by the conventional decoding methodologies, which is pointed out as a big limitation of network coding.

In order to deal with these problems, the proposed approximate decoding method is a technique that approximates lost data by using correlation of received data when receiving network coded data. For example, since the temperature or voltage measured in the real world has a high correlation between measurement data close in time, even when the data received by the data collection device is not sufficient for network decoding, Can be restored.

[1] Ho, T., M'edard, M., Shi, J., Effros, M. and Karger, D.R., "On randomized network coding, in Proc. Allerton Annual Conf. Commun., Control, and Comput., Oct. 2003 [2] H. Park, N. Thomos, and P. Frossard, Approximate decoding approaches for network coded correlated data, Signal Processing (Elsevier), vol. 93, no. 1, pp. 109 to 213, Jan. 2013

SUMMARY OF THE INVENTION It is an object of the present invention to provide a method of optimizing a position similarity information size for approximate decoding of network coded source data.

According to an aspect of the present invention, there is provided a decoding performance prediction method of approximate decoding using position similarity information of network-

과

의 차이

에 관한 각각의 후보

들 및 실제로

인

의 위치 인덱스들에 기초하여 생성되는 위치 유사성 정보가 소스 데이터에 부가된 경우에, 상기 위치 유사성 정보를 기초로 네트워크 코딩된 소스 데이터를 근사 복호화할 때의 복호 성공 확률에 따라 복호화 성능 지수를 연산하는 단계;Continuous source data in GF ( ^2M )

and

Difference

Each candidate for

And actually

sign

When the position similarity information generated based on the position indices of the positional similarity information is added to the source data, the decoding performance index is calculated according to the decoding success probability at the time of approximate decoding of the network coded source data based on the position similarity information step;

If the packet loss rate of a given channel is 50% or less, if the calculated decoding performance index is less than the upper saturation level, the upper saturation level is used instead of the calculated value of the decoding performance index. Otherwise, the calculated decoding performance index is used as the predictive decoding performance Outputting as an exponent; And

If the packet loss rate of a given channel exceeds 50%, then if the calculated decoding performance index is below the lower saturation level, the lower saturation level is used instead of the calculated value of the decoding capability index, otherwise the calculated decoding performance index is predicted And outputting it as a decoding performance index.

는According to one embodiment,

The

Here, 1? N? M-k and M is a characteristic value of GF,

인

가

일 확률은in reality

sign

end

The odds are

Lt; / RTI >

According to one embodiment, the decoding success probability is

이 n 개의 후보

들을 이용하여 원본 심볼

과 동일하게 복원될 확률일 수 있다.Is approximated by, it dropped packets recovered symbol in the real space from a N _l individual condition

These n candidates

The original symbol

As shown in FIG.

According to one embodiment, the decoding capability index may be one of a decoding success probability, a normalized error rate, or an average square error rate.

According to the decoding performance prediction method of approximate decoding using the position similarity information of the network coded source data of the present invention, decoding performance is predicted by using the position similarity information between consecutive source symbols or packets, Can be performed.

According to the decoding performance prediction method of approximate decoding using the location similarity information of the network coded source data of the present invention, the actual decoding performance is statistically determined according to the packet loss degree of the channel, the number of encoding packets of the source node, It can be predicted without tracking.

1 is a diagram conceptually illustrating a distributed data transmission system to which the present invention can be applied.
FIG. 2 is a graph illustrating approximate decoding performance according to the number of lost packets and the size of position similarity information in a decoding performance prediction method of approximate decoding using position similarity information of network coded source data according to an embodiment of the present invention.
3 is a flowchart illustrating a decoding performance prediction method of approximate decoding using position similarity information of network coded source data according to an embodiment of the present invention.

For the embodiments of the invention disclosed herein, specific structural and functional descriptions are set forth for the purpose of describing an embodiment of the invention only, and it is to be understood that the embodiments of the invention may be practiced in various forms, The present invention should not be construed as limited to the embodiments described in Figs.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. The same reference numerals are used for the same constituent elements in the drawings and redundant explanations for the same constituent elements are omitted.

The following description is largely divided into three sections, Sections 1 and 2, which illustrate the basic principles of approximate decoding along with an introduction to a distributed data transmission system to which the present invention may be applied, The decoding method will be explained.

Note that some of the notations for vector or variable appearing in Section 3 may differ from those shown in Section 1 and Section 2.

It should also be noted that the terms symbol, packet, and data in the following description are used interchangeably, unless the context clearly implies otherwise.

1. Random linear network coding

1 is a diagram conceptually illustrating a distributed data transmission system to which the present invention can be applied. And the source data is transmitted to the decoder (decoder) through the network coding node. The data transmitted in this system has a correlation with each other. The correlated data may have external correlation, such as data measured over the sensor network, or may be highly correlated data, such as images within a video sequence.

Research on the transmission of such correlation data has been widely performed in distributed systems. Network coding techniques have been studied as a technique for transmitting correlation data over a network.

In FIG. 1, an overlay ad hoc network is composed of source nodes, intermediate nodes, and receiving nodes, and the source data is transmitted from the source node to the receiving node through intermediate nodes responsible for network coding operations do. The intermediate node may transmit the received data as it is, but may combine and output the data received from the other source nodes.

Random Linear Network Coding (RLNC) techniques used in the present invention among various network coding schemes are described in Ho, T., M'edard, M., Shi, J., Effros, M. and Karger, DR, "On randomized network coding," in Proc. Allerton Annual Conf. Commun., Control, and Comput., Oct. 2003.

내의 L 개의 소스 심볼들은

라고 표시되며, 각각 이산(discrete)이면서 상관되어(correlated) 있고, 1≤i≤L일 때에

이다.

는

의 알파벳 집합(alphabet set)이고 |

|는

의 크기를 의미한다. 네트워크 코딩 연산은 갈로와 필드(GF: Galois Field) 내에서 수행되기 때문에, 각각의

은 GF 내의 원소로 고려되어야 한다.

가 실수 필드(

, field of real numbers)에 있는 수인지 아니면 GF에 있는 수인지 여부를 명확하게 특정하기 위해서, 수학식 1과 같은 식별 함수(identity function)을 정의할 수 있다.Source set

Lt; RTI ID = 0.0 > L &

And are respectively discrete and correlated, and when 1? I? L

to be.

The

Of the alphabet set (alphabet set) and |

| Is

. Since the network coding operation is performed in a Galois Field (GF), each

Should be considered as elements in the GF.

Is a real field (

, field of real numbers, or the number in GF, it is possible to define an identity function as shown in equation (1).

가

을 의미하는 GF 내의 원소임을 의미한다.

인 경우에,

는 L 개의 요소를 가지는 N 번째 소스 데이터 집합이다.

는

의 알파벳 세트를 의미한다. 또한 본 명세서에서 행렬 또는 벡터의 위 첨자로 쓰이는 T는 행렬 또는 벡터의 전치(transpose)를 의미하며, 실수값 T 또는 아래 첨자로 쓰이는 T와 구별됨을 유의한다.this is

end

Which means that it is an element in GF.

in case of,

Is the Nth source dataset with L elements.

The

≪ / RTI > Also note that T, used as a superscript of a matrix or vector herein, is a transpose of a matrix or vector, and is distinguished from a real value T or a T used as a subscript.

기호는 생략된다.

와

연산자들은 각각 GF 내에서 정의되는 합산 연산과 승산 연산을 의미한다.In the formula describing with the operation symbols in the GF, it implies that all operations occur in the GF,

The symbol is omitted.

Wow

The operators refer to summation and multiplication operations defined in GF, respectively.

The intermediate node k using the RLNC generates and transmits a packet according to Equation (2).

와 코딩 계수

의 GF에서의 선형 결합(linear combination in GF)을 의미한다. i는 1≤i≤L인 임의의 정수이고, T는 하나의 패킷 내에 결합된 데이터의 개수이다.

와

는 항상 GF 내의 원소로 간주될 수 있다. 코딩 계수들은 크기 2^M을 가지고 GF로부터 균일하고 랜덤하게 선정되는데, 이를

라고 한다. 이는 GF 크기가 M에 의해 결정되며,

임을 의미한다.Equation (2)

And a coding coefficient

Means a linear combination in the GF. i is any integer with 1? i? L, and T is the number of pieces of data combined in one packet.

Wow

Can always be regarded as an element in the GF. The coding coefficients are chosen uniformly and randomly from GF with size 2 ^M ,

. This is because the GF size is determined by M,

.

에서 합산 연산은 XOR 연산으로 수행될 수 있다. GF의 크기는 소스 심볼에 관하여 수행될 수 있는 코딩 연산들의 집합을 결정한다. 따라서 입력 알파벳의 크기가

라고 가정할 수 있다. 만약

인 경우에는 입력 집합은 예를 들어 소스 결합(source binning) 또는 양자화 등을 이용하여 축소되고 그럼으로써 입력 집합은 GF 크기, 즉 2^M을 초과하지 않게 된다.In the present invention, in particular, a GF having a characteristic value of 2

The summation operation can be performed by an XOR operation. The size of GF determines the set of coding operations that can be performed with respect to the source symbol. Therefore, the size of the input alphabet is

. if

The input set is reduced using, for example, source binning or quantization, so that the input set does not exceed the GF size, i.e. 2 ^M.

이 수신되어 연산에 쓰일 수 있다면, 선형 시스템

는 다음 수학식 3과 같이 표현될 수 있다.The packets generated at each node are transmitted to neighboring nodes and finally to the receiving nodes. If at the decoder of the receiving node there are K innovative or linearly independent symbols or packets,

Lt; / RTI > can be received and used in an operation,

Can be expressed by the following equation (3).

연산자는 유한 필드(finite field)에서 행렬들 사이의 승산 연산을 의미한다. here

The operator denotes a multiplication operation between matrices in a finite field.

는 코딩 계수 행렬(coding coefficient matrix)이라고 하는데, 컬럼 벡터들

로 구성되며, 수신된 심볼들의 유사도 모델에 기초하여 생성된다. 코딩 계수 행렬

를 유사도 모델에 기초하여 생성하는 방법은 네트워크 코딩의 복호화 기법에 숙련된 자들에게 주지된 방법이므로 설명을 생략한다.K × T matrix

Is called a coding coefficient matrix, and column vectors < RTI ID = 0.0 >

And is generated based on the similarity model of the received symbols. Coding coefficient matrix

Based on the degree-of-similarity model is a method well known to those skilled in the network coding decoding technique, and thus a description thereof will be omitted.

가 풀-랭크(full-rank), 즉 K=T인 경우에, 수학식 3의 선형 시스템의 해(solution)인

는, 코딩 계수 행렬

의 역행렬

가 고유하게 결정되고 또한 이에 상응하여

이기 때문에, 다음 수학식 4와 같이 고유하게 결정될 수 있다.If such a coding coefficient matrix

Is a full-rank, that is, K = T, the solution of the linear system of equation (3)

, A coding coefficient matrix

Inverse of

Is uniquely determined and correspondingly

, It can be uniquely determined as shown in the following equation (4).

The inverse matrix of the coding coefficient matrix can be easily obtained by a known method such as performing a Gaussian elimination on GF.

을 결정하기에 충분하지 않은 수의 심볼들만 수신된 경우에, 다시 말해, K<T인 경우에, 코딩 계수 행렬

가 풀-랭크가 아니기 때문에 수학식 3의 선형 시스템의 해

는 유일하게 결정되지 못할 뿐 아니라, 무한한 개수로 얻어질 수 있다. The problem is, if a unique inverse matrix

In the case of K < T, the coding coefficient matrix &lt; RTI ID = 0.0 &gt;

Is not a full-rank, the solution of the linear system of equation (3)

Not only can not be determined uniquely, but can be obtained in an infinite number.

The approximate decoding method is introduced to solve this problem, and the approximately determined source symbols are reconstructed with insufficiently received data.

2. Approximate decoding method

이 특이 행렬(singular matrix)인 경우에는, 코딩 계수 행렬

의 역행렬

은 얻어질 수 없고 수학식 3의 선형 시스템에서 해가 유일하지 않을 수 있다.The decoder can recover the source data upon receipt of the symbol set y. If there are external factors such as packet delay or loss during the transmission procedure, or a randomly selected coding coefficient matrix

In the case of this singular matrix, the coding coefficient matrix

Inverse of

Can not be obtained and the solution may not be unique in the linear system of equation (3).

To solve this problem, the approximate decoding method can make the coding coefficient matrix full-rank using additional constraints.

와

를 구축하는 데에 이용될 수 있다. 추가 제한 조건들

및

의 모든 요소들(elements) 역시 GF 내의 값이다.The additional constraints that the correlation of the input data adds in the decryption procedure

Wow

Can be used for constructing. Additional Restrictions

And

All the elements of GF are also values in GF.

및

을 추가함으로써 선형 시스템은 유일한 해를 가질 수 있게 되고 유일한 해

는 다음 수학식 5와 같이 표현될 수 있다.Additional Restrictions

And

By adding the linear system can be the only solution and the only solution

Can be expressed by the following equation (5).

및

내의 계수들은 소스의 모델들에 기초하여 결정될 수 있다.

가 결정되면, 원본 데이터의 근사 심볼

은 수학식 1의 식별 함수들에 의해

과 같이 얻어질 수 있다. Additional Restrictions

And

The coefficients in the matrix can be determined based on the models of the source.

Is determined, an approximate symbol of the original data

Is determined by the identification functions of Equation (1)

Can be obtained.

H. Park, N. Thomos, and P. Frossard, Approximate decoding approaches for network coded correlated data, Signal Processing (Elsevier), vol. 93, no. 1, pp. 109 to 213, Jan. 2013 thesis can be referred to.

는 유사도 모델에 기초하여 생성될 수 있는데, 예를 들어 가장 유사한 데이터

및

의 위치에 상응하는 자리에 값 "1" 및 "1"을 가지는 두 개의 요소들을 제외하고 각 행이 0으로 구성되도록 생성된다. 또한

은 T-K 길이를 가지는 벡터로서 결정되는데, 예를 들어 모든 값이 0일 수 있다(

).(TK) x T matrix

May be generated based on the similarity model, for example, the most similar data

And

Quot; 1 "and" 1 "in the place corresponding to the position of " 1 &quot; Also

Is determined as a vector having a TK length, e.g., all values may be zero (

).

은 다음의 수학식 6과 같이 얻어질 수 있다.Thus, an approximate symbol of the original data

Can be obtained by the following Equation (6).

This allows the decoder to approximate the original symbols even though some symbols are insufficient for complete decoding.

를 가역적으로(invertible) 만든다는 점이다. 이는 네트워크 코딩된 데이터에 관하여 데이터 복원 방법론의 새로운 패러다임을 제시한다. The core idea of these existing algorithms is to add a surplus equation based on the source correlation,

To make it invertible. This presents a new paradigm of data restoration methodology for network coded data.

와

가 경험론적 방법론(heuristic approach)에 의존하는 경우가 대부분이어서 소스 유사성을 제대로 재현하지 못하였고, 이는 제한된 성능으로 이어졌다.
However, the above-described approximate decoding methodologies are not limited to the additional constraint matrix

Wow

Is largely dependent on the heuristic approach, which does not reproduce the source similarity well, leading to limited performance.

3. Approximate Decoding and Performance Prediction Techniques Using Positional Similarity of Source Data

3.1 Location similarity information of source data

인

가 실수 공간에서는 하나의 값을 가지지만, GF(2^M) 내에서는 복수의 후보 Δ들, 즉 Δ₁, Δ₂, Δ₃, ..., Δ_n을 가질 수 있다는 점은 주지의 사실이다. 종래에는 이러한 특성이 근사 복호화법에 이용되지 않았지만, 본 발명과 동일 출원인이 2012년 6월 29일자로 출원한 한국특허출원 제2012-0071282호 "위치정보행렬을 활용한 네트워크 코딩된 데이터의 근사 복호화 방법"에서는 이러한 복수의 후보 값들과 그 위치 인덱스를 이용하는 근사 복호화 방법을 개시하며, 상술한 한국특허출원 제2012-0071282호는 본 발명의 내용의 일부로서 병합된다.Meanwhile,

sign

It is well known that the GF (2 ^M ) can have a plurality of candidate values Δ ₁ , Δ ₂ , Δ ₃ , ..., Δ _n in the GF (2 ^M ) . Although this characteristic has not conventionally been used in the approximate decoding method, Korean Patent Application No. 2012-0071282, filed on June 29, 2012, which is the same as the present invention, describes an approximate decoding method of network coded data using a positional information matrix Method "discloses an approximate decoding method using a plurality of candidate values and a position index thereof, and Korean Patent Application No. 2012-0071282 mentioned above is incorporated as part of the present invention.

을 생성하는 데에 PIM 정보들을 활용할 수 있게 한 발명이 개시된다.In the above-mentioned Korean Patent Application No. 2012-0071282, information on the position indexes about the positions of symbols having the candidate Δ values with respect to each candidate Δ value and the immediately preceding symbol are listed in the order of candidate Δ, Quot; Position Information Matrix " (PIM), and transmits it to the receiving node together with the source data, when performing approximate decoding on the missing symbols in the receiving node,

The present invention discloses an invention enabling PIM information to be used for generating the PIM information.

은 M-k 개의 후보들을 가지는데,

에 속하는 n 번째 후보 Δ_n는 다음의 수학식 7과 같이 표현되고, 후보 Δ_n가 실제로

인

와 일치할 확률은 수학식 8과 같이 표현된다.In GF (2 ^M ), if Δ = 2 ^k (0 ≤ ^k ≤ ^M )

Has Mk candidates,

Δ _n n-th candidate that belongs to is expressed as follows: Equation (7), the candidate is actually Δ _n

sign

Is expressed by Equation (8). &Quot; (8) &quot;

는 후보 Δ들 중 어느 하나이게 되므로, 수학식 8과 같은 확률식이 도출된다.Here, 1? N? Mk. The probability of Equation (8) means that the number of times a particular Δ _n appeared during the full potential of 2 ^(MK) -1 meetings candidate Δ 2 ^(Mkn) in GF (2 ^M). Actual

Is one of the candidates?, A probability equation such as Equation (8) is derived.

=Δ₁=2^k일 경우가 가장 확률이 높은데, 만약 소스 데이터 중에서 각각의 직전 심볼에 관하여

=Δ₁=2^k로 연산되는 일부 심볼들의 위치 인덱스들에 관한 정보를 수신 노드가 알고 있다면, 소스 데이터 집합 내의 심볼들은 선형 상관 관계가 있으므로, 일부 심볼이 전송 중에 누락되더라도, 알고 있는 위치 인덱스를 이용하여 추가 제한 조건

를 생성한다면 좀더 높은 확률로 근사 복호할 수 있다는 것이 상술한 한국특허출원 제2012-0071282호의 기술적 사상 중 하나라고 할 수 있다. 그러한 위치 인덱스는 소스 노드가 생성하여 수신 노드에 부가 정보로서 전달한다.E.g,

= Δ ₁ = 2 ^k is the most probable. If for each immediately preceding symbol in the source data

= DELTA ₁ = 2 ^k , the symbols in the source data set are linearly correlated, so that even if some symbols are missing during transmission, the known position indexes Additional Restrictions Using

It is one of the technical ideas of the above-mentioned Korean Patent Application No. 2012-0071282 that the approximate decoding can be performed with a higher probability. Such a location index is generated by the source node and is transmitted to the receiving node as additional information.

Further, if Δ ₁ then the probability is added with high-degree position index of the symbols computed as Δ ₂ transferred to, and further, if the transmission in addition also the position index of the symbol on the further up Mk of candidates Δ _n, the receiving node It is possible to more precisely approximate the missing symbols.

In the present invention, the information on the Δ ₁ and the information on the position indexes about the symbols represented by the difference Δ _n between the consecutive symbols are arranged in the order of probability to be referred to as position similarity information (PSI) .

3.2 Approximate Decoding and Performance Prediction Using Location Similarity Information

The concept called PIM in Korean Patent Application No. 2012-0071282 mentioned above will be referred to as position similarity information in this specification.

The PSI information may be increased or decreased according to the size of n, but it is bandwidth added because it is added to the original source data and transmitted. Thus, increasing the size of the PSI reduces the bandwidth that the data can occupy in the entire bandwidth instead of the more accurate approximate decoding. On the other hand, if the size of the PSI of the data is reduced, the approximate decoding becomes somewhat inaccurate instead of increasing the bandwidth that the data can occupy over the entire bandwidth.

를 PSI로부터 생성할 수 있다. Assuming that N _l symbols are lost during transmission of the symbols combined in the number of encodings at a time to know the factor that determines the size of n, the receiving node has a vector having a length equal to the number of symbols lost, Condition

Can be generated from the PSI.

내에 약 1/2의 확률로 내에

₁가 나타날 것이고, n이 2이면 약 1/2²의 확률로

₂가

내에 나타난다. 따라서, n 개의 후보 Δ₁, Δ₂, Δ₃, ..., Δ_n들을 PSI에 포함시켰다면 하나의 손실 심볼에 대한 길이 1의

내에 Δ₁, Δ₂, Δ₃, ..., Δ_n들이 모두 나타날 확률은 각각의 확률들을 모두 더한 값이다.If n is 1,

Lt; RTI ID = 0.0 &gt;

₁ will appear, and if n is 2, a probability of about 1/2 ²

₂

Lt; / RTI &gt; Therefore, if n candidates? ₁ ,? ₂ ,? ₃ , ...,? _N are included in the PSI, the length 1

The probability that all of Δ ₁ , Δ ₂ , Δ ₃ , ..., Δ _n appear in the _matrix is the sum of all the probabilities.

If it is N _l loss symbols, it can be approximated by Equation (9).

이 n 개의 후보 Δ들을 이용하여 정확히 원본 심볼

과 동일하게 복원될 확률을 의미하며, n과 k는 소스 노드에서 결정되므로, 수학식 9의 복원 확률, 나아가 수신 노드의 데이터 복원 성능은 손실 심볼들의 개수 N_l에 지배적으로 좌우된다고 할 수 있다.This is because in the loss symbol N _l,

Using these n candidate delta values,

Since n and k are determined at the source node, the restoration probability of Equation (9), and furthermore, the data restoration performance of the receiving node dominantly depends on the number N _l of lost symbols.

Referring to FIG. 2, to see how the number of lost symbols determines the performance as approximately predicted in Equation (9), FIG. 2 illustrates the location similarity of network coded source data according to an embodiment of the present invention FIG. 3 is a graph illustrating approximate decoding performance according to the number of lost packets and the size of position similarity information in a decoding performance prediction method of approximate decoding using information. FIG.

의 크기는 8이며 GF(2¹⁰)로 가정한다. 복호화 성능은 평균 제곱 오류율(MSE)의 역수로 측정하였다.Here, the number of data to be network-coded, that is, the number of packets is nine, and the packet loss rate of the channel varies from the case where packet loss is one of nine packets to eight out of nine packets. but

Is assumed to be 8 and GF (2 ¹⁰ ). Decoding performance was measured by the reciprocal of the mean square error rate (MSE).

When the loss packet is 1 or more and 4 or less, 1 / MSE values are all near the upper saturation level (Sat _u ) when there is no PSI side information other than Δ _{1 with} n = 1, that is, when the PSI overhead is 0%.

Also, when the loss packet is 5 or more and 8 or less, 1 / MSE values are all near the lower saturation level (Sat _l ) when there is no PSI side information other than Δ _{1 with} n = 1, that is, when PSI overhead is 0%.

On the other hand, if the PSI side information is added up to Δ ₂ with n = 2, the size of the data to be transmitted is increased by about 20% from the original data size. Instead, 1 / It is somewhat improved in the case of five people. However, in cases where the number of lost packets is 2 or more and 4 or less, it is in the upper saturation level, and in cases where the number of lost packets is in the range of 6 or more and 8 or less,

If the PSI side information is added up to Δ ₃ with n = 3, the size of the data to be transmitted is increased by about 30% from the size of the original data. Instead, the 1 / MSE values are Not only that, it also started to improve in cases where the number of lost packets is 2 and 3, and cases 6 and 7, respectively. However, in the case of the loss packet number 4, it remains at the upper saturation level, and in the case of the loss packet number 8, it remains at the lower saturation level.

If the PSI side information is added up to Δ ₄ with n = 4, the size of the data to be transmitted is increased by about 36% from the size of the original data, but instead 1 / MSE values can be reduced by 1/10 or more depending on the situation . In addition, 1 / MSE values are improved in all cases. Especially, in case of loss packet number 4 and loss packet number 8, it shows abrupt performance improvement by deviating from upper saturation level and lower saturation level.

If the additional information is added at n = 5 or more, the size of the data to be transmitted is increased to about 40%, but it can be seen that 1 / MSE values are significantly improved in all cases. In particular, the 1 / MSE value, which occurs when the number of lost packets is 1 or more and 4 or less, can be said to be in the range of about 95 to 99% in terms of decoding success probability.

2, when the packet loss rate is lower than 50%, when the decoding performance index according to the number of lost packets converges and when the packet loss rate is higher than 50%, the decoding performance index according to the number of lost packets converges The difference is that it is different. In addition, it is also unusual for the cases where the packet loss rate is between 0% and 50% and the packet loss rate is similar to 50% and less than 100% according to the addition of PSI additional information.

In either case, if the size of the PSI side information is increased, that is, at the expense of bandwidth for the source data in the entire bandwidth, the approximate decoding performance index is improved. The decoding performance index can be measured not only by the decoding success probability but also by the normalization error rate or the mean square error rate (MSE) or its inverse. There is a tradeoff relationship between the target decoding performance index and the bandwidth.

Therefore, if the decoding performance index of the network coding can be predicted, a scheme for utilizing network resources more efficiently can be derived.

3 is a flowchart illustrating a decoding performance prediction method of approximate decoding using position similarity information of network coded source data according to an embodiment of the present invention.

과

의 차이

에 관한 각각의 후보

들 및 실제로

인

의 위치 인덱스들에 기초하여 생성되는 위치 유사성 정보(PSI)가 소스 데이터에 부가된 경우를 전제로 하여, 단계(S31)에서, 수학식 9의 복호 성공 확률에 기초하여 소정의 복호화 성능 지수를 연산한다. 상술하였듯이, 복호화 성능 지수는 복호 성공 확률 자체일 수도 있고, 정규화된 오류율이나 평균 제곱 오류율일 수도 있다.Referring to FIG. 3, consecutive source data in GF ( ^2M )

and

Difference

Each candidate for

And actually

sign

(PSI) generated on the basis of the position indices of the positional indexes of the positional information of the positional relationship between the positional similarity information do. As described above, the decoding performance index may be a decoding success probability itself, or may be a normalized error rate or an average square error rate.

In step S32, it is determined whether the packet loss rate? Of a given channel is 50% or less. If the packet loss rate? Is less than 50% at step S32, the process proceeds to step S33, and if the packet loss rate? Exceeds 50%, the process proceeds to step S34.

In step (S33), If the calculated decoding performance index is higher saturation level (sat _U) upper saturation level, instead of the value of the calculated decoding performance index is less than (sat _U), or predict the calculated decoding performance index decoding And outputs it as a performance index.

In step (S34), If the calculated decoding performance index is the lower the saturation level (sat _L) sub-saturated levels in place of the value of the calculated decoding performance index is less than (sat _L), or predict the calculated decoding performance index decoding And outputs it as a performance index.

Based on the predicted decoding performance, it is possible to attempt to optimize the position similarity information size for approximate decoding.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, but, on the contrary, Modification is possible. Accordingly, the spirit of the present invention should be understood only in accordance with the following claims, and all of the equivalent or equivalent variations will fall within the scope of the present invention.

Claims (5)

Continuous source data in GF ( ^2M )

and

Difference

Each candidate for

And actually

sign

Calculating a decoding capability index when approximating and decoding the network coded source data based on the position similarity information when position similarity information generated based on position indices of the position similarity information is added to the source data;
If the packet loss rate of a given channel is 50% or less, if the calculated decoding performance index is less than the upper saturation level, the upper saturation level is used instead of the calculated value of the decoding performance index. Otherwise, the calculated decoding performance index is used as the predictive decoding performance Outputting as an exponent; And
If the packet loss rate of a given channel exceeds 50%, then if the calculated decoding performance index is below the lower saturation level, the lower saturation level is used instead of the calculated value of the decoding capability index, otherwise the calculated decoding performance index is predicted As a decoding performance index,
The decoding performance index is calculated to have a higher value as the decoding success probability is higher, converges to the upper saturation level when the size of the position similarity information is reduced when the packet loss rate is 50% or less, And converging to the lower saturation level when the size of the position similarity information is reduced. The method of claim 1, wherein the position similarity information is converged to the lower saturation level. The method according to claim 1,

The

Here, 1? N? Mk, M is a characteristic value of GF,
in reality

sign

end

The odds are

And actually

sign

end

Candidates selected in order of greater probability

And the position similarity information is generated based on the position indices of the source similarity information and the size that can add the position similarity information to the source data. The decoding performance prediction of the approximate decoding using the position similarity information of the network coded source data Way. The method of claim 2, wherein the decoding success probability is

Is approximated by, it dropped packets recovered symbol in the real space from a N _l individual condition

These n candidates

The original symbol

And estimating the decoding performance of the approximate decoding using the position similarity information of the network coded source data. The method of claim 1, wherein the decoding performance index is one of a decoding success probability, a normalized error rate, and a mean square error rate. A computer-readable recording medium having recorded thereon a program for causing a computer to perform a decoding performance prediction method of approximate decoding using position similarity information of network coded source data according to any one of claims 1 to 4.

Images