Block code decoding device using OSD and autoencoder in mobile communication system environment

Abstract

The present invention relates to a block code decoding device using OSD and an autoencoder in a mobile communication system environment and, more specifically, to a block code decoding device using OSD and an autoencoder which is a machine learning technique. The decoding device proposed in the present invention decodes a reception vector by using an autoencoder, thereby enabling decoding having excellent performance in a Rayleigh fading environment similar to an actual mobile communication environment.

Description

Block code decoding device using OSD and autoencoder in a mobile communication system environment

[0001] The present invention relates to a block code decoding device using an OSD and an autoencoder in a mobile communication system environment, and more specifically, to a block code decoding device using an ODS and an autoencoder, which is a machine learning technique.

[0002] Mobile communication systems experience wireless channel environments such as Additive White Gaussian Noise (AWGN), Rayleigh fading, and Rician fading. The Rayleigh fading model is a reasonable model for environments where radio signals are scattered or dispersed due to numerous objects along the radio path before reaching the receiver. According to the central limit theorem, if there is sufficient signal scattering, the channel response is modeled as a Gaussian distribution, regardless of the distribution of individual elements. The Rayleigh fading model is suitable for modeling signal propagation in the Earth's troposphere and ionosphere, as well as for wireless signal effects in urban environments with many tall buildings. The Rayleigh fading model is most appropriate when there is no line-of-sight (LOS) propagation between the transmitter and receiver. If the main signal is transmitted via the LOS path, the Rician fading model is considered a relatively more appropriate model.

[0003] Ordered statics decoding (OSD) is one of the error-correcting code decoding methods, and it exhibits the best performance in AWGN channel environments. However, OSD does not perform optimally in other channel environments, including Rayleigh fading and Rician fading. Therefore, a decoding method that demonstrates good decoding performance in various channel environments, including AWGN, is required.

[0004] The problem to be solved by the present invention is to propose a decoding method that shows good performance in various channel environments including AWGN channel environments.

[0005] Another problem that the present invention seeks to solve is to propose a decoding method that shows good performance in an environment similar to an actual wireless communication environment.

[0006] For this purpose, the decoding device of the block code using the OSD (ordered statics decoding) and autoencoder of the present invention receives the received vector ( ), AE output reception vector ( ) and generating matrix ( ) is input, and the input reception vector ( ) is an array sequence arranged according to set criteria. ), AE output reception vector ( ) AE output array sequence arranged according to set criteria ( ) and shape generating matrix ( ) a permutation function sorting unit that outputs a sequence of arrays ( ), AE output array sequence ( ) and shape generating matrix ( ) is input, and the input array sequence ( ) from and creates, and creates and shape generating matrix ( ) using the test vector ( ) to generate a test vector; a test vector ( ) and AE output array sequence ( ) and calculates the error, and the minimum error test vector ( with the smallest difference in the calculated error) ) for producing a minimum error selection unit; and a minimum error test vector ( ) and a permutation function inversion unit that generates an estimated codeword vector by estimating the codeword vector.

[0007] Here, Is

[0008]

[0009] For this purpose, the decoding device of the block code using the OSD (ordered statics decoding) and autoencoder of the present invention receives the received vector ( ), AE output reception vector ( ) and generating matrix ( ) is input, and the receiving vector ( ) is an array sequence arranged according to set criteria. ), AE output reception vector ( ) AE output array sequence arranged according to set criteria ( ) and shape generating matrix ( ) a permutation function sorting unit that outputs a sequence of arrays ( ), AE output array sequence ( ) and shape generating matrix ( ) is input, and the input AE output array sequence ( ) from and creates, and creates and shape generating matrix ( ) using the test vector ( ) to generate a test vector; a test vector (input from the test vector generation unit) ) and array sequence ( ) and calculates the error, and the minimum error test vector ( with the smallest difference in the calculated error) ) for producing a minimum error selection unit; and a minimum error test vector ( ) and a permutation function inversion unit that generates an estimated codeword vector by estimating the codeword vector.

[0010] Here, Is

[0011]

[0012] For this purpose, the block code decoding device using the OSD (ordered statics decoding) and auto-encoder of the present invention receives the AE output vector ( ) and generating matrix ( ) is input, and the AE output reception vector ( ) AE output array sequence arranged according to set criteria ( ) and shape generating matrix ( ) outputting a permutation function sorting unit; an AE output array sequence ( from the permutation function sorting unit ) and shape generating matrix ( ) is input, and the input AE output array sequence ( ) from and creates, and creates and shape generating matrix ( ) using the test vector ( ) to generate a test vector; a test vector ( ) and AE output array sequence ( ) and calculates the error, and the minimum error test vector ( with the smallest difference in the calculated error) ) for producing a minimum error selection unit; and a minimum error test vector ( ) and a permutation function inversion unit that generates an estimated codeword vector by estimating the codeword vector.

[0013] Here, Is

[0014]

[0015]

[0016] Here, the AE output reception vector ( ) is the received vector ( ) that the codeword vector received through the wireless channel. ) is a vector output by an autoencoder (AE), and the generator matrix is ​​a matrix that converts a message vector into a codeword vector, and the shape generator matrix ( ) is a matrix that rearranges the identity matrix and parity matrix, and the parity matrix is ​​a matrix that generates the parity part of the codeword vector.

[0017] A block decoding device using an OSD and an auto decoder in a mobile communication system environment according to the present invention shows good performance in various wireless channel environments including an AWGN channel environment, and in particular, has good decoding performance in an actual wireless communication environment and a Rayleigh fading environment.

[0018] FIG. 1 illustrates an OSD-based decoding device according to one embodiment of the present invention.

[0019] FIG. 2 illustrates a decoding device that implements a decoding method (OSD+AE Method-1) using machine learning based on OSD according to one embodiment of the present invention.

[0020] FIG. 3 illustrates a decoding device that implements a decoding method (OSD+AE Method-2) using machine learning based on OSD according to another embodiment of the present invention.

[0021] FIG. 4 illustrates a decoding device that implements a decoding method (OSD+AE Method-3) using machine learning based on OSD according to another embodiment of the present invention.

[0022] Figure 5 is a graph comparing the BER (bit error rate) for the decoding method proposed in the present invention in an AWGN channel environment.

[0023] Fig. 6 is a graph comparing the BER of OSD, Method-1 to Method-3 in a Rayleigh fading channel environment.

[0024] Figure 7 shows a probability distribution diagram for the output of the reception vector and AE in an AWGN channel environment.

[0025] Figure 8 shows a probability distribution diagram for the output of the reception vector and AE in a Rayleigh fading channel environment.

[0026] The aforementioned and additional aspects of the present invention will become more apparent through preferred embodiments described with reference to the accompanying drawings. Below, these embodiments of the present invention will be described in detail to facilitate understanding and reproducibility by those skilled in the art.

[0027] The present invention proposes a method for decoding block codes using only OSD in a mobile communication system environment, or a method for decoding block codes using the ODS technique and an AE (auto-encoder), a machine learning technique. In other words, the present invention proposes a block decoding method that utilizes OSD or OSD and AE in a Rayleigh fading channel environment most similar to an actual mobile communication environment, thereby achieving relatively improved decoding performance compared to existing methods.

[0028] The length of the code word and the length of the message is The sign of ( , ) is called a block code and a generating matrix Encoding is performed using the generator matrix. That is, the generator matrix encodes the message matrix (vector) into a codeword matrix (vector). In BPSK, the codeword of the block code Is is generated as a bipolar sequence Is can be obtained from here, is a message, It consists of. is the length of the message part of the codeword vector and is a natural number, is the length of the codeword vector and is a natural number.

[0029] Received vector received through AWGN channel silver , and the received vector is received through the Rayleigh fading channel. Is It is. At this time is a Gaussian random variable with a mean of 0 and has a variance of , is a Rayleigh random variable. is the noise power spectral density.

[0030] Table 1 shows the structure of the overcomplete AE model proposed in the present invention.

[0031] Layer(type)Output ShapeParameterHidden layer1(Dense)[-1,1,46]1,104Hidden layer2(Dense)[-1,1,92]4,324Hidden layer3(Dense)[-1,1,46]4,278output layer(Dense)[-1,1,23]1,081Total params: 10,787

[0032] Each layer uses a Dense layer, which is a fully connected layer, and the number of parameters indicates the complexity of the model. As shown in Table 1, the Output Shape of hidden layer 1 is [-1, 1, 46], the Output Shape of hidden layer 2 is [-1, 1, 92], and the Output Shape of hidden layer 3 is [-1, 1, 46]. Accordingly, the complexity of the model also increases to a certain extent and then decreases.

[0033] The AE model is the received vector Since it is trained with the goal of noise reduction, the ELU (exponential linear unit) is used as the activation function of each hidden layer so that data with negative values ​​can be considered, and the tanh function is used as the activation function of the output layer so that a value between -1 and 1 can be output. The model's loss function is was used, and MSE is the mean square error. The training of this AE was performed for 10 to handle various input data of AWGN and Rayleigh fading channels. 8 It consists of a random code word.

[0034] Epoch and batch size are 100 and 10 respectively. 3 By training on these diverse data sets, AE can effectively remove noise from the received signal, thereby improving decoding accuracy. The AE model's approach is useful when the characteristics of the channel noise are uncertain or particularly variable.

[0035] FIG. 1 illustrates an OSD-based decoding device according to an embodiment of the present invention. Hereinafter, the OSD-based decoding device according to an embodiment of the present invention will be described in detail using FIG. 1.

[0036] According to FIG. 1, the OSD-based decoding device (OSD decoding device) (100) includes a permutation function sorting unit (102), a test vector generation unit (104), a choose minimum discrepancy unit (106), and an invers permutation function unit (108). Of course, other configurations than the above-described configurations may be included in the OSD-based decoding device proposed in the present invention.

[0037] According to Fig. 1, the permutation function sorting unit (102) sorts the received vector according to a set criterion. , is the received vector as a permutation function It is a function that sorts the reliability in descending order. The permutation function sorting unit (102) generates a matrix and permutation function , Using Creates.

[0038] The permutation function sorting unit (102) Through elementary row operations (Elementary Row Operations), the organizational behavior get. At this time Is It is a unit matrix, Is It is a parity matrix. The array sequence is is. The permutation function sorting unit (102) , Prints out. Silver of It refers to a shape generating matrix that has an organized form.

[0039] The test vector generation unit (104) and test vector Creates. Here, Is

[0040]

[0041]

[0042] It can be obtained through . In the leaflet The dog symbol is the most reliable independent symbol (MRI). Therefore, The message portion consists of MRI symbols and parity part are distinguished. The OSD of the test vector generation unit (104) ) is a text vector is created. At this time is the length and the Hamming weight is 0. is an arbitrary vector with a range of up to . At this time It is. That is Is and possible occurrence It is generated through the operation of . is the order of OSD, for example, if it is 1, a vector with a Hamming weight from 0 to 1 is used, and if it is 2, a vector with a Hamming weight from 0 to 2 is used. The Hamming weight is the number of non-zero components of the vector. Therefore, is at least 1 can be, , Is Each one exists.

[0043] The minimum error selection unit (106) and Find the error between, and in particular, the one with the minimum error Select . That is Is and The difference between is obtained. The minimum error selection unit (106) is the minimum error test vector. Prints out.

[0044] The permutation function inversion part (108) is the smallest Having cast And Estimated codeword vector through That is, the permutation function inversion part (108) is selected from the minimum error selection part. The estimated codeword vector is calculated using .

[0045] FIG. 2 illustrates a decoding device that implements a decoding method (OSD+AE Method-1, Method-1) using machine learning based on an OSD according to an embodiment of the present invention. Hereinafter, the decoding device using machine learning based on an OSD according to an embodiment of the present invention will be described in detail using FIG. 2.

[0046] A decoding device (200) implementing Method-1 proposed in Fig. 2 includes a permutation function sorting unit (202), a test vector generation unit (204), a minimum error selection unit (206), and a permutation function inversion unit (208).

[0047] According to Fig. 2, the permutation function sorting unit (202) is a criterion for generating a permutation function. This is the output of AE, not To this end, the permutation function sorting unit (202) is used from AE. is input. That is, , Is It is a function that sorts the reliability in descending order. The permutation function sorting unit (202) , , Prints out.

[0048] Additionally, the minimum error selection unit (206) reduces noise due to AE. with the permutation function can be used in calculations to improve decoding performance. To elaborate, the minimum error selection unit (206) and Find the error between, and in particular, the one with the minimum error person is selected. The minimum error selection unit (206) Prints out.

[0049]

[0050] FIG. 3 illustrates a decoding device that implements a machine learning-based decoding method (OSD+AE Method-2, Method-2) based on an OSD according to another embodiment of the present invention. Below, using FIG. 3, a decoding device using machine learning based on an OSD according to an embodiment of the present invention will be described in detail.

[0051] A decoding device (300) implementing Method-2 proposed in Fig. 3 includes a permutation function sorting unit (302), a test vector generation unit (304), a minimum error selection unit (306), and a permutation function inversion unit (308).

[0052] According to Fig. 3, the permutation function sorting unit (302) is a criterion for generating a permutation function. . Of course, the permutation function sorting unit (302) is from AE. is input. The permutation function sorting unit (302) is the same as the permutation function sorting unit (202) of Fig. 2. , , Prints out.

[0053] According to Fig. 3, the test vector generation unit (304) generates a test vector To create a Generated as output of non-AE Use . Noise is reduced through AE Create using, In the process of comparing to calculate, it has high reliability. The decoding performance is improved. To elaborate, the test vector generation unit (304) and test vector is created. Here, Is

[0054]

[0055]

[0056]

[0057] Additionally, the minimum error selection unit (306) Use .

[0058] FIG. 4 illustrates a decoding device that implements a decoding method (OSD+AE Method-3, Method-3) using machine learning based on an OSD according to another embodiment of the present invention. Hereinafter, using FIG. 4, a decoding device using machine learning based on an OSD according to an embodiment of the present invention will be described in detail.

[0059] A decoding device (400) implementing Method-3 proposed in Fig. 4 includes a permutation function sorting unit (402), a test vector generation unit (404), a minimum error selection unit (406), and a permutation function inversion unit (406).

[0060] According to Figure 4, in the decryption process This is the output of AE, not . Noise is reduced due to AE. Using it, we can expect an improvement in decoding performance, and AE becomes a means to verify decoding performance.

[0061] According to Fig. 4, the permutation function sorting unit (402) is a criterion for generating a permutation function. This is the output of AE, not To this end, the permutation function sorting unit (402) is used from AE. is input. That is, , Is It is a function that sorts the reliability in descending order. The permutation function sorting unit (402) , Prints out.

[0062] The test vector generation unit (404) generates a test vector To create Generated as output of AE, not Use .

[0063] The minimum error selection unit (406) By using and Find the error between, and in particular, the one with the minimum error person is selected. The minimum error selection unit (406) Prints out.

[0064] In the present invention, an optimal decoding method is proposed by comparing performance in an AWGN channel and a Rayleigh fading channel using four methods as described above.

[0065] For performance comparison, the present invention uses the (23, 12) Golay code. The Golay code is a representative linear block code that can be applied with OSD, boasting a short code length and capable of correcting errors of up to 3 bits. OSD is known to exhibit optimal decoding performance regardless of the block code type in AWGN channel environments. Therefore, rather than using various codes, we evaluate its performance using only the (23, 12) Golay code.

[0066] Figure 5 is a graph comparing the BER (bit error rate) for the decoding method proposed in the present invention in an AWGN channel environment.

[0067] Figures 5 (a) and 5 (b) compare the BER when using OSD (1) and OSD (2) of order-1 and order-2, respectively.

[0068] Figures 5 (a) to 5 (b) show similar results, and in particular, the OSD method shows the best performance, and among the methods proposed in Figures 2 to 4, Method-2 proposed in Figure 3 shows the closest performance to the decoding process by OSD in Figure 1.

[0069] Figure 5 (c) shows a BER comparison graph for OSD (1), OSD (2), Method-2, MDD (minimum distance decoding), and AE and MDD together. MDD is a decoding method that corrects errors by adding error patterns that share the same syndrome as the received vector. Since OSD demonstrates near-optimal performance in AWGN channels, MDD, which is based on hard decision decoding, can be used as a benchmark to verify the noise reduction effect of AE.

[0070] As shown in (c) of Fig. 5, when performing MDD with the output of AE rather than the existing MDD, the decoding performance is improved, and it can be confirmed that AE reduces noise. Since OSD utilizes the soft information of the received vector, the soft information of the received vector is damaged and reduced due to the noise reduction effect of AE, and it can be seen that the decoding performance of the Method-2 decoding device proposed in Fig. 3 is lowered compared to the OSD decoding device proposed in Fig. 1. To elaborate, in an AWGN channel environment, the OSD decoding device has relatively excellent decoding performance.

[0071] Fig. 6 is a graph comparing the BER of an OSD decoder and Method-1 to Method-3 decoders in a Rayleigh fading channel environment. According to Fig. 6, it can be seen that the decoders according to Method-1 to Method-3 show relatively superior decoding performance compared to the decoder according to OSD, and in particular, the decoder according to Method-1 shows relatively the best decoding performance.

[0072] The reason decoders using Methods 1 through 3 demonstrate relatively superior decoding performance is that, while Rayleigh fading channels vary over time, AWGN channels are fixed over time. Based on AE, decoders using Methods 1 through 3 can learn and adapt to channel information, resulting in relatively superior decoding performance. Therefore, to better understand the observed results, it is necessary to analyze the mean, variance, and distribution of the vectors passing through each channel and the AE output.

[0073] Table 2 shows the mean, variance, and BER for the received vector and the output of AE when the signal-noise ratio (SNR) is 0 dB and 5 dB in each channel type. Table 2 shows the mean, variance, and BER for the all-one vector ( ) was derived assuming that the received vector and AE output were transmitted, and hard decisions were used for both the received vector and AE output in the process of calculating BER.

[0074] 0dBchannelreceived vectoroutput of AEAWGNmean0.990.78var0.480.20BER0.0740.073Rayleighmean2.420.68var13.610.45BER0.210.145dBchannelreceived vectoroutput of AEAWGNmean0.990.98var0.160.005BER0.0060.002Rayleighmean1.620.91var4.360.10BER0.060.03

[0075] According to Table 2, at 0 dB, the AE output exhibited lower dispersion than the received vector in both AWGN and Rayleigh fading channel environments. While there was no significant change in BER in the AWGN channel environment, BER improved in the Rayleigh fading channel environment.

[0076] At 5dB, the output of AE not only improved the BER in both channels, but also reduced both the mean and variance compared to 0dB, demonstrating the decoding performance of AE in noise reduction.

[0077] Figure 7 illustrates the probability distribution of the received vector and the output of AE in an AWGN channel environment. In particular, Figure 7 illustrates the case where the SNR is 0. Figure 7 (a) shows that the received vector in the AWGN channel follows a Gaussian distribution, and Figure 7 (b) shows the output distribution of AE. Since the activation function of the output layer of AE is a tanh function, values ​​between -1 and 1 are output. According to the output distribution of AE, many values ​​are distributed close to 1 due to noise reduction of AE.

[0078] Figure 8 illustrates a probability distribution diagram for the output of the reception vector and AE in a Rayleigh fading channel environment. In particular, Figure 8 illustrates the case where the SNR is 0.

[0079] Figure 8 (a) shows that the received vector follows a Rayleigh distribution in a Rayleigh fading channel, while Figure 8 (b) shows a distribution similar to the results in an AWGN channel. It can be seen that AE reduces noise by learning the Rayleigh fading channel environment.

[0080] As can be seen in Table 2, in the Rayleigh fading channel, the variance of the received vector and the output of AE is reduced from 13.61 to 0.45, which leads to an improvement in decoding performance.

[0081] In AWGN channel environments, decoding using OSD is the optimal method, while decoding using machine learning based on OSD shows slightly lower performance. This is because AE provides hard information with noise removed, while simultaneously damaging and reducing the soft information used in OSD.

[0082] Conversely, considering MDD, which uses only hard information for decoding, performance is improved by using refined hard information from AE. On the other hand, in Rayleigh fading channels, decoding methods using Methods 1 through 3 all show better decoding performance than decoding using OSD. Because Rayleigh fading channels are more challenging than AWGN channels, decoding methods that include AE, which has learned the channel environment, show superior decoding performance than OSD decoding methods optimized for AWGN channels.

[0083] In particular, since the decoding method of Method-2 uses the received vector to determine the permutation order and calculate the discrepancy (error) and uses the output of AE only when generating and comparing test vectors, the noise reduction effect due to AE is relatively small compared to other decoding methods.

[0084] Since the decoding method of Method-3 uses the output of the AE throughout the entire process, inherent errors in the AE occur, resulting in lower decoding performance compared to the decoding method of Method-1. In contrast, the decoding method of Method-1 shows significant performance improvements, primarily due to two factors. First, by using the output of the AE to determine the permutation order, it provides a more reliable basis than the received vector. Second, by utilizing the hard information of the received vector during the decoding process, errors that may occur in the AE are prevented. Therefore, the present invention proposes the decoding methods of FIGS. 2 to 4 as decoding methods for block codes in a mobile communication system environment.

[0085] Although the present invention has been described with reference to an embodiment shown in the drawings, this is merely exemplary, and those skilled in the art will understand that various modifications and equivalent other embodiments are possible therefrom.

[0086] The present invention relates to a block code decoding device using an OSD and an autoencoder in a mobile communication system environment, and more specifically, to a block code decoding device using an ODS and an autoencoder, which is a machine learning technique.

[0087] The block decoding device using the OSD and auto decoder in the mobile communication system environment according to the present invention shows good performance in various wireless channel environments including the AWGN channel environment, and in particular, has good decoding performance in an actual wireless communication environment and a Rayleigh fading environment.

Claims

1. AE output reception vector ( ) is the received vector ( ) that is the codeword vector received through the wireless channel. ) is a vector output by an autoencoder (AE). The generating matrix is ​​a matrix that converts a message vector into a codeword vector. Shape generating matrix ( ) is a matrix that rearranges the identity matrix and parity matrix. A parity matrix is ​​a matrix that generates the parity part of a codeword vector. Receiving vector ( ), AE output reception vector ( ) and generating matrix ( ) is input, and the input reception vector ( ) is an array sequence arranged according to set criteria. ), AE output reception vector ( ) is an AE output array sequence arranged according to set criteria. ) and shape generating matrix ( ) sorting function that outputs the permutation function; The array sequence from the above permutation function sorting part ( ), AE output array sequence ( ) and shape generating matrix ( ) is input, and the input array sequence ( ) from and creates and shape generating matrix ( ) using the test vector ( ) to generate a test vector; Test vectors received from the test vector generation unit ( ) and AE output array sequence ( ) and calculates the error with respect to the minimum error test vector ( ) with the smallest difference in the calculated error. ) for producing a minimum error selector; and The minimum error test vector received from the minimum error selection unit ( ) is used to estimate a codeword vector, and a permutation function inversion unit is included to generate an estimated codeword vector. A block code decoding device using OSD (ordered statics decoding) and an autoencoder. Here, Is 2.AE output reception vector ( ) is the received vector ( ) that is the codeword vector received through the wireless channel. ) is a vector output by an autoencoder (AE). The generating matrix is ​​a matrix that converts a message vector into a codeword vector. Shape generating matrix ( ) is a matrix generated by processing the identity matrix and parity matrix. A parity matrix is ​​a matrix that generates the parity part of a codeword vector. Receiving vector ( ), AE output reception vector ( ) and generating matrix ( ) is input, and the receiving vector ( ) is an array sequence arranged according to set criteria. ), AE output reception vector ( ) is an AE output array sequence arranged according to set criteria. ) and shape generating matrix ( ) sorting function that outputs the permutation function; The array sequence from the above permutation function sorting part ( ), AE output array sequence ( ) and shape generating matrix ( ) is input, and the input AE output array sequence ( ) from and creates and shape generating matrix ( ) using the test vector ( ) to generate a test vector; Test vectors received from the test vector generation unit ( ) and array sequence ( ) and calculates the error with respect to the minimum error test vector ( ) with the smallest difference in the calculated error. ) for producing a minimum error selector; and The minimum error test vector received from the minimum error selection unit ( ) is used to estimate a codeword vector, and a permutation function inversion unit is included to generate an estimated codeword vector. A block code decoding device using OSD (ordered statics decoding) and an autoencoder. Here, Is 3.AE output reception vector ( ) is the received vector ( ) that is the codeword vector received through the wireless channel. ) is a vector output by an autoencoder (AE). The generating matrix is ​​a matrix that converts a message vector into a codeword vector. Shape generating matrix ( ) is a matrix generated by processing the identity matrix and parity matrix. A parity matrix is ​​a matrix that generates the parity part of a codeword vector. AE output reception vector ( ) and generating matrix ( ) is input, and the AE output reception vector ( ) is an AE output array sequence arranged according to set criteria. ) and shape generating matrix ( ) sorting function that outputs the permutation function; The AE output array sequence from the above permutation function sorting unit ( ) and shape generating matrix ( ) is input, and the input AE output array sequence ( ) from and creates and shape generating matrix ( ) using the test vector ( ) to generate a test vector; Test vectors received from the test vector generation unit ( ) and AE output array sequence ( ) and calculates the error with respect to the minimum error test vector ( ) with the smallest difference in the calculated error. ) for producing a minimum error selector; and The minimum error test vector received from the minimum error selection unit ( ) is used to estimate a codeword vector, and a permutation function inversion unit is included to generate an estimated codeword vector. A block code decoding device using OSD (ordered statics decoding) and an autoencoder. Here, Is 4. In any one of paragraphs 1 to 3, The above-mentioned generating matrix ( )silver And, Is It is a unit matrix, Is A block code decoding device using an autoencoder and OSD (ordered statics decoding) characterized by a parity matrix. is the length of the message part of the codeword vector, and is a natural number. is the length of the codeword vector, and is a natural number 5. In paragraph 1, the message portion ( ) and a parity part () are characterized by a block code decoding device using OSD (ordered statics decoding) and an autoencoder.

6. In paragraph 5, the test vector ( )Is and possible occurrence It is generated through the operation of , is the length and the Hamming weight is 0. ( A block code decoder using an OSD (ordered statics decoding) and an autoencoder, characterized in that the range of the OSD is an arbitrary vector (up to the order of the OSD).

7. In the second or third paragraph, is the message part ( ) and parity part ( ) is characterized by comprising an OSD (ordered statics decoding) and an autoencoder for decoding a block code.

8. In paragraph 7, the test vector ( )Is and possible occurrence It is generated through the operation of , is the length and the Hamming weight is 0. ( A block code decoder using an OSD (ordered statics decoding) and an autoencoder, characterized in that the range of the OSD is an arbitrary vector (up to the order of the OSD).

Images