KR101135062B1 - Signal compression apparatus and method for power analysis attacks - Google Patents

Abstract

The present invention relates to a signal compression apparatus and method for a power analysis attack. The signal compression method according to the present invention extracts a signal for at least one or more specific clocks from a power waveform signal, and weights each of the elements of the extracted signal. Is calculated based on the characteristic of the power waveform signal to calculate the weight vector, and the compressed signal is generated by multiplying the weight vector calculated by the elements of the extracted signal.

Description

Signal compression apparatus and method for power analysis attacks

The present invention relates to a signal compression apparatus and method for a power analysis attack, which is a subchannel attack method for encryption communication. And a recording medium recording the same.

In order to securely transmit information requiring confidentiality to other parties, many scholars have studied how to encrypt and decrypt information, and modern cryptography can provide communication security through integrity, confidentiality and authentication in communication. However, it has been known that even algorithms known to be mathematically safe have an additional leak of information that has not been taken into account at the implementation stage, and side channel attacks have been introduced that can obtain confidential information from these weaknesses. Side channel attacks obtain information from the side channel related to the secret key contained in the encrypted information by measuring physical characteristics such as communication runtime, power consumption, electromagnetic radiation, and using statistical techniques. .

Such subchannel attacks include fault insertion attacks, timing attacks, power analysis attacks, and electromagnetic emission attacks, particularly in the implementation of devices. Differential power analysis (DPA) attacks, which find and analyze correlations between measurements and other practices, are known as the most powerful methods.

The differential power analysis attack is an attack method for finding a key using signal processing and general characteristics of a plurality of power waveforms, and may have a large performance difference depending on the signal processing method.

The technical problem to be solved by the present invention is to overcome the limitations of the performance degradation of the power analysis attack by not considering the characteristics of the power waveform signal, the important components associated with the power analysis in the power waveform signal is lost in the signal compression process To solve this problem. Furthermore, the present invention seeks to solve the problem that excessive computation and resources are required by performing signal compression for a power analysis attack on a plurality of clocks in a power waveform signal.

In order to solve the above technical problem, a signal compression method for a power analysis attack according to the present invention comprises the steps of extracting a signal for at least one predetermined clock (clock) from the power waveform signal; Calculating a weight vector by determining a weight of each of the elements of the extracted signal based on the characteristic of the power waveform signal; And generating a compressed signal by multiplying the calculated weight vector by elements of the extracted signal.

In order to solve the above technical problem, the step of calculating the weight vector in the signal compression method for power analysis attack according to the present invention is such that the variance of the compressed signal for the elements of the extracted signal is maximized using principal component analysis. It is desirable to determine the weight vector.

In order to solve the another technical problem, the signal compression method for a power analysis attack according to the present invention further comprises generating a compressed signal for all clocks of the power waveform signal using the calculated weight vector. .

In addition, the following provides a computer-readable recording medium having recorded thereon a program for executing the signal compression method for the power analysis attack described above on a computer.

In order to solve the above technical problem, a signal compression device for a power analysis attack according to the present invention includes an input unit for receiving a power waveform signal; And extracting a signal for at least one predetermined clock from the input power waveform signal, calculating a weight vector by determining a weight of each of the elements of the extracted signal based on a characteristic of the power waveform signal, and extracting the weight vector. And a signal processor for generating a compressed signal by multiplying the calculated weight vector by elements of the processed signal.

In order to solve the above technical problem, in the signal compression device for power analysis attack according to the present invention, the weight vector is a value that maximizes the variance of the compressed signal with respect to the elements of the extracted signal using principal component analysis. It is preferred to be determined.

The present invention improves the performance of power analysis by calculating a weight vector using principal component analysis in consideration of the characteristics of the power waveform signal in a signal compression method for power analysis attack. It is fully preserved during signal compression, and can provide effective power analysis results with fewer operations and resources.

1 is a flowchart illustrating a signal compression method for a power analysis attack according to an embodiment of the present invention.
2 is a flowchart illustrating a process of calculating a weight vector in a signal compression method for a power analysis attack according to an embodiment of the present invention in more detail.
3 is a diagram illustrating one clock waveform of a power waveform signal to explain a signal compression method for a power analysis attack according to an embodiment of the present invention.
4A to 4D are diagrams illustrating eigenvectors of a clock signal calculated in a signal compression method for a power analysis attack according to an embodiment of the present invention.
5A and 5B illustrate power waveform signals and signals compressed by a signal compression method for power analysis attack according to an embodiment of the present invention, respectively.
6A and 6B are flowcharts illustrating methods of generating a compressed signal for all clocks other than one clock initially selected in the signal compression method for power analysis attack according to another embodiment of the present invention.
7 is a diagram illustrating a signal compression device for a power analysis attack according to an embodiment of the present invention.

Prior to describing embodiments of the present invention, a signal processing procedure for a power analysis attack will be briefly introduced.

The signal processing technique consists of three stages: signal alignment, noise cancellation, and signal compression. The main advantages of applying signal compression techniques to differential power analysis (DPA) attacks introduced earlier are shorter analysis times and reduced memory. However, if the signal compression loses important components associated with power analysis in the original power waveform signal, it may be a cause of deterioration in the analysis performance. Thus, in performing signal compression to process high-dimensional data into low-dimensional data, there is a need to ensure that significant significant components of the original power waveform signal are not lost as much as possible. In connection with such signal compression, methods such as raw integration, maximum extraction, and sum of squares are known.

This signal compression technique is derived from a rather intuitive approach that does not reflect the characteristics of the power waveform signal. More specifically, these compression methods are as follows.

First, Raw Integration is the most basic compression technique that adds all the signals in each clock and compresses them into a single point. In terms of power analysis, this technique is equivalent to all of the signals in one clock. That is, the signals in one clock are all given an equal weight.

Second, the Maximum Extraction technique is a compression technique that extracts a signal for one point having a maximum value for a signal in each clock. This technique determines the power of the location that consumes the most power from the signal within one clock as the most meaningful signal from the power analysis point of view.

Third, the Sum of Squares technique adds the squares of all signals in each clock and compresses them into a single point. This technique combines the characteristics of the two compression techniques. That is, it is a method of judging the power of the position which consumes the most power from the signal in one clock as the most meaningful signal, and giving meaning to the signal with small consumption. As a result, one clock signal is added by weighting each signal size.

However, the three compression techniques described above can be a cause of degrading the performance of differential power analysis (DPA) because the characteristics of the signal are not considered at all in a fairly intuitive manner. In particular, the compression techniques described above have limitations in that weights are not differentially and individually assigned to respective signal elements according to the characteristics of the signal.

Specifically, in a signal compression technique for power analysis, a signal in one clock may be represented by a set of m signal elements (means points of a signal) as shown in Equation 1 below.

In addition, an optimal weight vector for each element of the signal x may be expressed as Equation 2 below.

Here, the optimal weight vector means weights for respective signal elements for optimally compressing the corresponding signal. Therefore, the problem of which compressed signal or which weight vector is optimal is ultimately a problem of whether the characteristics of the signal can be accurately understood and reflected. In other words, it will be very difficult to produce an optimal compressed signal unless you can accurately characterize the original signal. Thus, the optimal signal compression technique for power analysis means finding the optimal weight vector ω taking into account the characteristics of the power waveform signal for each signal element of signal x in one clock.

The original power waveform signal vector may be compressed into one point signal as shown in Equation 3 by the weight vector thus found.

Equation 3 indicates that after multiplying m weight vectors corresponding to m elements of the power waveform signal by each, the multiplied results are added and compressed into one signal.

Now look at Table 1 below to see how the weight vector is applied in the three signal compression techniques introduced earlier. These three signal compression techniques also compress signals using weight vectors, and each signal compression technique uses different weight vectors.

As can be seen from Table 1, the three signal compression techniques introduced above are very intuitive, but they do not take into account the characteristics of each signal element, and even some techniques do not reflect any difference in importance between signals. not. Thus, since the characteristics of the original signal have not been applied to the signal compression technique, there is a fundamental limitation in generating the optimal compressed signal as a result.

Hereinafter, through various embodiments of the present invention to propose a signal compression technique that can overcome the limitations of the above-described signal compression techniques. The signal compression technique to be introduced through embodiments of the present invention uses principal component analysis (PCA) and corresponds to a signal compression technique in consideration of characteristics of a power waveform signal.

Principal component analysis in statistics is one of the techniques for analyzing a data set. When mapping data to one axis, the axis with the largest variance is located as the first coordinate axis, and the second axis is positioned as the second coordinate axis. In this way, it refers to a technique for linearly converting data into a coordinate system sequentially. By placing the most important components of the data on each axis like this, principal component analysis is characterized by an optimal linear transformation that preserves the subspace with the largest variance. In particular, principal component analysis does not have a predetermined weight vector, unlike conventional linear transformations, and has the advantage that it can be determined according to characteristics of given data. Therefore, as described above, the signal compression technique for power analysis may be appropriately used to find an optimal weight vector considering the characteristics of a signal. That is, in the embodiments of the present invention, the principal component analysis is used to determine a signal having significant variation in power analysis from the original signal as a large signal value depending on the data value.

Hereinafter, various embodiments of the present invention will be described in detail with reference to the drawings.

1 is a flowchart illustrating a signal compression method for a power analysis attack according to an embodiment of the present invention, and includes the following steps.

In step 110, a signal for at least one predetermined clock is extracted from the power waveform signal. Here, the predetermined clock may be one clock selected from a plurality of clocks. Signal compression may be more efficient on one clock operation because the signal inherent linearity of converting power consumed by multiple clocks with a principal component vector at a time may be lost. That is, in step 110, one clock of the power waveform signal is selected to extract a signal for the corresponding clock.

In particular, the software-implemented cryptographic operation operates for multiple clocks, which results in too many computations if a similar power analysis is repeated every clock, resulting in excessive computing power and resources. Done. Therefore, in this case, the weight vector may be calculated only for a specific clock in terms of economy, and then signal compression may be performed using the weight vector calculated for the remaining clocks.

Meanwhile, when the original power waveform signal is input, the signal extraction step described above corresponds to a process of processing the received signal information in an electronic form, and thus a processor capable of performing signal extraction and a storage space for such an operation. It can be implemented through memory. In addition, such a processor and a storage space may be appropriately selected by those skilled in the art in consideration of the utilization environment and operating environment of the technical field to which the present invention belongs. Furthermore, this signal extraction step can be implemented not only by the hardware illustrated above, but also by additional software code that can be utilized in the above signal extraction process or control the illustrated hardware.

In step 120, a weight vector is calculated by determining weights of the elements of the signal extracted in step 110 based on the characteristics of the power waveform signal. Unlike the three signal compression techniques described above, step 120 may calculate an optimal weight vector by considering the characteristics of the power waveform signal, and give an individual weight instead of the same weight for each signal element.

In particular, in the process of calculating the weight vector in step 120, the weight vector is determined to maximize the variance of the compressed signal with respect to the elements of the signal extracted in step 110 using principal component analysis. Therefore, as mentioned above, the weight vector corresponding to each of the elements of the signal is not a predetermined or fixed value, but corresponds to a value that is individually determined in consideration of the characteristics of the signal.

The process of applying the concept of principal component analysis to find the optimal weight vector is as follows. The n clock signals used for the power analysis attack may be expressed as Equation 4 as follows. Here, a signal corresponding to one clock of each clock signal is shown.

를 찾는 것을 의미한다.In the embodiment of the present invention, finding an optimal weight vector is a weight vector satisfying the following Equation 5 for the n clock signals described above.

Means to find.

In Equation 5, argmax means a weight vector that maximizes the variance of the compressed signal. The task of finding the weight vector is performed based on the principal component analysis of the original power waveform signal, which is determined by the data value of the signal that is meaningful for power analysis, depending on the data value. That is, calculating the optimal weight vector means giving a relatively larger weight to a signal having a large variation in the power waveform signal. Therefore, when the ω value is found by the principal component analysis, the g value can be determined by Equation 5. A more specific method and principle of determining the weight vector such that variance of the compressed signal is maximized will be described later with reference to FIG. 2.

The calculating of the weight vector corresponds to a process of calculating the weight vector from elements of the signal extracted from the power waveform signal processed in the electronic form. It can be used in a similar way. In addition, it is natural that additional software code for controlling such processors and storage space may be utilized.

In step 130, a compressed signal is generated by multiplying the weight vector calculated in step 120 by elements of the signal extracted in step 110. As described above through Equation 3, after multiplying the weight vectors calculated in step 120 to the elements of the original power waveform signal, one compressed signal may be generated by adding all the multiplication results. In this case, since the compressed signal is a result of an optimized weight vector calculated in consideration of the characteristics of the power waveform signal through step 120, an important element of the signal elements corresponds to a compressed signal that is not lost.

The process of generating such a compressed signal corresponds to a process of generating a compressed signal, which is a final product, by using signal elements processed in an electronic form and a weight vector processed in an electronic form, and thus a processor and memory capable of performing these operations. Space can be utilized. In addition, it is natural that additional software code for controlling such processors and storage space may be utilized.

According to the above-described embodiment of the present invention, in the signal compression method for the power analysis attack, the weight vector is calculated using principal component analysis considering the characteristics of the power waveform signal, thereby improving the performance of the power analysis and power in the power waveform signal. Important components associated with the analysis are preserved intact during signal compression, and it is possible to perform effective power analysis with fewer operations and resources.

FIG. 2 is a flowchart illustrating in more detail a process (step 120 of FIG. 1) of calculating a weight vector in a signal compression method for power analysis attack according to an embodiment of the present invention. Is omitted for convenience. That is, in FIG. 2, the process of inputting the extracted signal will be described. The input signal at this time is a signal corresponding to one clock for each of the n power waveform signals. In addition, below, the variable m refers to the clock size and the variable p refers to the selected number of principal components.

In order to perform a power analysis attack, even if signal compression is performed, a power consumption pattern based on a Hamming weight or Hamming distance of data values must be maintained. As mentioned earlier, this linearity can be lost if the power dissipated in multiple clocks is converted using principal component vectors at one time. Thus, in embodiments of the present invention the signal compression must be done on the operation of one clock.

라 할 때, 추출된 신호들에 대한 압축 신호들로 이루어진 분포

에 대해 Var(Z)의 값은A covariance matrix is calculated from the elements of the signal extracted in step 121. Extracted signals

Is a distribution consisting of the compressed signals for the extracted signals.

The value of Var (Z) for

It may be modified as in Equation 6 below.

Here, the covariance matrix and the mean vector are arranged as in Equation 7 below.

In step 122, eigenvectors and eigenvalues are calculated from the covariance matrix calculated in step 121. In the above-described equation (7), the task of finding the vector ω such that Var (Z) has the maximum value is performed as in Equation 8 below.

In Equation 8, the vector ω which makes Var (Z) partial by ω to be 0 may be a candidate vector for Var (Z) to be maximum. This results in the problem of finding the eigenvectors and eigenvalues of the covariance matrix of the data vectors.

As shown in Equation 8, the variance value when the linear transformation is performed on the corresponding eigenvectors becomes a corresponding eigenvalue. Therefore, the eigenvector with the largest eigenvalue becomes g which satisfies Equation 5.

In step 123, a predetermined number of eigenvectors corresponding to eigenvalues having a relatively large value are selected from the eigenvalues calculated in step 122.

In this case, selecting a predetermined number of eigenvectors is to remove a signal that is not similar to a signal that is similar to the original power waveform signal. At this time, the number of selecting the eigenvector can be freely set according to the needs of the user. Therefore, when a specific number of eigen values calculated in step 122 are selected as large, a specific number of eigenvectors corresponding to the selected eigenvalues will be selected.

를 만족하는 고유 벡터 ω는 S의 rank 만큼 존재하게 되며, 각 고유 벡터들은 서로 직교하는 성질을 갖는다. 따라서, 찾고자 하는 선형 변환 벡터는 고유 값 λ중 큰 값 순서대로 고유 벡터를 선택해 사용할 수 있다. 만약,

의 순서대로 고유 값들이 정렬되고, 각각의 고유 값 λ_i 에 해당하는 고유 벡터를 ω_i라고 한다면, 고유 값이 큰 p개의 고유 벡터들 (ω₁, ω₂, ..., ω_p)을 주성분 벡터로 투영해서 사용할 벡터들로 선택하면 된다.In Equation 8 summarized above

The eigenvector ω that satisfies λ exists by the rank of S, and each eigenvector has an orthogonal property. Therefore, the linear transform vector to be searched can select and use the eigenvectors in order of the larger values among the eigenvalues λ. if,

If the eigenvalues are arranged in the order of, and the eigenvector corresponding to each eigenvalue λ _i is called ω _i , then p eigenvectors (ω ₁ , ω ₂ , ..., ω _p ) with large eigenvalues This is done by selecting the vectors to be projected as the principal component vector.

In the meantime, the selecting of the eigenvector may be performed by canceling a predetermined number of eigenvectors having low correlation with the power waveform signal. That is, the eigenvectors visually inferior to the original signal are judged as noise components and erased. This process will be described below with reference to FIGS. 3 and 4A to 4D for better understanding.

FIG. 3 is a diagram illustrating one clock waveform of a power waveform signal to explain a signal compression method for a power analysis attack according to an embodiment of the present invention, and FIGS. 4A to 4D illustrate power according to an embodiment of the present invention. A diagram illustrating eigenvectors of a clock signal calculated in a signal compression method for an analysis attack.

First, note the waveform of the signal illustrated in FIG. Next, look at the four eigenvectors of FIGS. 4A to 4D generated based on the power waveform signal of FIG. 3. Of the four eigenvector waveforms, FIGS. 4A and 4B are relatively similar in pattern to the original power waveform signal of FIG. 3. That is, the visual signal is highly related to the original signal. 4C and 4D, on the other hand, are relatively visually unrelated to the original power waveform signal of FIG. Therefore, if the predetermined number is set to two, only two eigenvectors of FIGS. 4A and 4B are selected and the remaining eigenvectors are erased.

The process of selecting a visually relevant eigenvector can be performed by comparing the calculated patterns of the eigenvector graph or by calculating a correlation between the original signal and the eigenvector pattern. Various comparison means can be utilized. The comparison means for calculating the visual correlation of such graphs may be appropriately selected by those skilled in the art according to the implementation environment to which the present embodiment is applied.

2, in step 124, the selected eigenvectors are added to calculate the weight vector. According to the above-described embodiment, the weight vector may be calculated by adding the p eigenvectors selected in step 123.

In the above, a specific process of calculating the weight vector has been described. As already introduced through step 120 of FIG. 1, it is natural that a processor capable of performing the above specific operations, a storage space, and additional software code for controlling such hardware may be utilized.

5A and 5B illustrate a power waveform signal and a signal compressed by a signal compression method for a power analysis attack according to an embodiment of the present invention, respectively. In particular, FIG. 5B illustrates power using a calculated weight vector. The compressed signal generated by compressing each signal element of the waveform signal is shown.

After the weight vector is calculated by the process listed through FIG. 2, the n clock signals of the original power waveform signal may be compressed to one point using the calculated weight vector. Equation 3 described above is used for such compression. The graph of FIG. 5B is generated by using a DES algorithm, which is widely known as a cryptographic algorithm, to a weight vector proposed through an embodiment of the present invention.

As shown in FIGS. 5A and 5B, the signal consumed only in the positive direction may be distributed in the positive and negative directions after signal compression. This phenomenon occurs when principal component analysis is performed on each clock signal because the components of the eigenvectors are the same but the directions may be reversed. However, this phenomenon is not a problem because when the power analysis is performed, the absolute value of the correlation coefficient is compared.

Embodiments of the present invention presented above introduced methods for generating a compressed signal for one clock. We now present two embodiments that generate compressed signals for not only one such clock, but all clocks contained within the target signal for a power analysis attack.

6A and 6B are flowcharts illustrating methods of generating a compressed signal for all clocks other than one clock initially selected in the signal compression method for power analysis attack according to another embodiment of the present invention. Steps 110 to 130 in each flowchart have already been described with reference to FIG. 1, and thus detailed description thereof will be omitted.

6A further includes step 140 in the signal compression method for the power analysis attack of FIG. 1. In step 140 and step 130, a process of calculating a weight vector for one clock from the power waveform signal and generating a compressed signal therefrom is repeatedly performed for each of the clocks of the power waveform signal. This process allows the generation of optimized compressed signals for power waveform signals including all other clocks with the same clock size.

However, since the cryptographic calculator implemented in software is repeatedly performed for a plurality of clocks, iteratively performing the calculation of the above-mentioned weight vector and the generation of the compressed signal every clock generates a large amount of computation, computing power, and resource consumption. Require. Therefore, in this case, the weight vector may be calculated for only a specific clock, and then the weight vector calculated for all remaining clocks may be applied. This embodiment is discussed below with reference to FIG. 6B.

6B further includes step 150 in the signal compression method for the power analysis attack of FIG. 1. In step 150, a compressed signal for all clocks of the power waveform signal is generated using the weight vector calculated in step 130. That is, the weight vector calculated above is applied to all clocks in the target signal for the power analysis attack as well as one clock selected for the weight vector calculation.

According to this embodiment of the present invention, while important components associated with power analysis in the power waveform signal are completely preserved in the signal compression process, it is possible to provide an effective power analysis result with less computation and resources.

7 is a diagram illustrating a signal compression apparatus 700 for a power analysis attack according to an embodiment of the present invention, which corresponds to an embodiment of the embodiment of FIG. The signal compression apparatus 700 of FIG. 7 largely includes an input unit 710 and a signal processor 720, and the signal processor 720 is more specifically a signal extractor 721, a weight vector calculator 722, and a compressor. The signal generator 723 is included.

The input unit 710 receives a power waveform signal that is an original signal. At this time, the power waveform signal may be received in the form of analog data, the data processed in a digital form may be received. If the power waveform signal is input in analog form, a configuration (ADC) for processing the signal into a digital form may be involved for the convenience of signal processing.

In the signal processor 720, more specifically, the signal extractor 721 extracts a signal for a predetermined clock from the power waveform signal input through the input unit 710. This configuration corresponds to step 110 of FIG. 1 and will not be described in detail.

Next, the weight vector calculator 722 calculates a weight vector by determining the weight of each element of the signal extracted by the signal extractor 721 based on the characteristics of the power waveform signal. In addition, the weight vector may be determined as a value that maximizes the variance of the compressed signal with respect to elements of the signal extracted by the signal extractor 721 using principal component analysis. In more detail, the weight vector calculator 722 calculates a covariance matrix from elements of the signal extracted by the signal extractor 721, calculates eigenvectors and eigenvalues from the calculated covariance matrix, and calculates the result. A weight vector may be calculated by selecting a predetermined number of eigenvectors corresponding to eigenvalues having a large value among the eigenvalues and adding the selected eigenvectors. This configuration corresponds to step 120 of FIG. 1 and will not be described in detail.

Next, the compressed signal generator 723 multiplies the elements of the signal extracted by the signal extractor 721 by the weight vector calculated by the weight vector calculator 722 to generate a compressed signal. This configuration corresponds to step 130 of FIG. 1 and a detailed description thereof will be omitted.

The elements of the signal processor 720 described above may utilize a processor and a storage space capable of performing the above-described series of operations. In addition, it is natural that additional software code for controlling such processors and storage space may be utilized.

According to the above-described embodiment of the present invention, in the signal compression device for the power analysis attack, the weight vector is calculated using principal component analysis considering the characteristics of the power waveform signal to improve the performance of the power analysis, and the power within the power waveform signal. Important components associated with the analysis are preserved intact during signal compression, and it is possible to perform effective power analysis with fewer operations and resources.

Meanwhile, the present invention can be embodied as computer readable codes on a computer readable recording medium. The computer-readable recording medium includes all kinds of recording devices in which data that can be read by a computer system is stored.

Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disks, optical data storage devices, and the like, which may also be implemented in the form of carrier waves (for example, transmission over the Internet). Include. The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion. In addition, functional programs, codes, and code segments for implementing the present invention can be easily deduced by programmers skilled in the art to which the present invention belongs.

The present invention has been described above with reference to various embodiments thereof. Those skilled in the art will understand that the present invention can be implemented in a modified form without departing from the essential features of the present invention. Therefore, the disclosed embodiments should be considered in an illustrative rather than a restrictive sense. The scope of the present invention is shown in the claims rather than the foregoing description, and all differences within the scope will be construed as being included in the present invention.

700: Signal Compression Device for Power Analysis Attack
710: input unit
720: signal processor 721: signal extractor
722: weight vector calculator 723: compressed signal generator

Claims (11)

In the signal compression method for power analysis,
Extracting a signal for one clock from the power waveform signal;
Calculating a weight vector by determining a weight of each of the elements of the extracted signal based on the characteristic of the power waveform signal; And
Generating a compressed signal by multiplying the calculated weight vector by elements of the extracted signal,
The calculating of the weight vector may include determining a weight vector such that variance of a compressed signal with respect to elements of the extracted signal is maximized using principal component analysis (PCA). delete The method of claim 1,
The calculating of the weight vector may include applying a relatively larger weight to a signal having a large variation in the power waveform signal. The method of claim 1,
Computing the weight vector,
Calculating a covariance matrix from the elements of the extracted signal;
Calculating eigenvectors and eigenvalues from the calculated covariance matrix;
Selecting a predetermined number of eigenvectors corresponding to eigenvalues having relatively large values among the calculated eigenvalues; And
Adding the selected eigenvectors to calculate a weight vector. The method of claim 4, wherein
Selecting the eigenvectors is performed by canceling an eigenvector that is less correlated with the power waveform signal. The method of claim 1,
Generating a compressed signal for all clocks of the power waveform signal using the calculated weight vector. The method of claim 1,
And repeatedly performing calculating the weight vector and generating the compressed signal for each of the clocks of the power waveform signal. A non-transitory computer-readable recording medium having recorded thereon a program for executing the method of claim 1. In the signal compression device for power analysis,
An input unit receiving a power waveform signal; And
Extracting a signal for one clock from the input power waveform signal, calculating a weight vector by determining the weight of each of the elements of the extracted signal based on the characteristics of the power waveform signal, and the weight of the extracted signal A signal processor for generating a compressed signal by multiplying the calculated weight vector by elements;
And the weight vector is determined to have a value such that variance of the compressed signal with respect to elements of the extracted signal is maximized using principal component analysis. delete The method of claim 9,
The signal processing unit,
Calculating a covariance matrix from the elements of the extracted signal,
Eigenvectors and eigenvalues are calculated from the calculated covariance matrix,
Among the calculated eigenvalues, a predetermined number of eigenvectors corresponding to eigenvalues with large values are selected,
And calculate a weight vector by adding the selected eigenvectors.

Images