Method, device and computer program product for learning correlation matrix

A decoding system utilizing a correlation matrix that obtains an original codeword from a different codeword, said system comprising;

• an arithmetic unit for performing a predetermined calculation using the different codeword and a pre-learned correlation matrix;
• a comparison unit for comparing the calculation results output from the arithmetic unit with a threshold;
• and
• a learning state monitor for updating the pre-learned correlation matrix based on the comparison results output from the comparison unit to obtain a post-learned correlation matrix, wherein:
• the obtained post-learned correlation matrix is used as a new pre-learned correlation matrix to repeat the predetermined calculation, the comparison and the updating to obtain an optimal correlation matrix,
• every time the learning state monitor receives the comparison results, the learning state monitor judges whether or not the comparison results satisfy a first predetermined condition, and
• the system further comprises:
• a threshold controller for increasing the absolute value of the threshold at a predetermined rate when the first predetermined condition is satisfied, and
• a decoding device for decoding a block code, using the optimal correlation matrix,
• wherein the different codeword is represented by N-dimensional column vectors, and the original codeword is represented by M-dimensional column vectors so that the pre-learned correlation matrix has N rows and M columns, and
• wherein in the calculation, the different codeword is multiplied by the pre-learned correlation matrix to obtain the M-dimensional column vectors.

A computer readable storage medium having stored thereon a program for performing a correlation matrix learning for obtaining an optimal correlation matrix used for obtaining an original codeword from a different codeword, the program commanding a computer to:

• perform a predetermined calculation with the different codeword and a pre-learned correlation matrix;
• compare the calculation results with a threshold;
• update the pre-learned correlation matrix based on the comparison results to obtain a post-learned correlation matrix;
• judge whether the comparison results satisfy a first predetermined condition;
• increase the absolute value of the threshold at a predetermined rate if the comparison results satisfy the first predetermined condition;
• and
• repeat the calculation, the comparison, and the updating, using the obtained post-learned correlation matrix as a new pre-learned correlation matrix, thereby obtaining an optimal correlation matrix,
• wherein the different codeword is represented by N-dimensional column vectors, and the original codeword is represented by M-dimensional column vectors so that the pre-learned correlation matrix has N rows and M columns, and
• wherein in the calculation, the different codeword is multiplied by the pre-learned correlation matrix to obtain the M-dimensional column vectors.

A decoding method to be carried out on a computer, utilizing a correlation matrix that obtains an original codeword from a different codeword, the method comprising:

• inputting a first codeword into the computer;
• activating the computer to perform a predetermined calculation with the first codeword and a pre-learned correlation matrix;
• activating the computer to compare the calculation results with a threshold;
• activating the computer to update the pre-learned correlation matrix based on the comparison results to obtain a post-learned correlation matrix;
• activating the computer to judge whether the comparison results satisfy a first predetermined condition;
• if the comparison results satisfy the first predetermined condition, activating the computer to increase the absolute value of the threshold at a predetermined rate;
• activating the computer to repeat the calculation, the comparison, and the updating, using the obtained post-learned correlation matrix as a new pre-learned correlation matrix, until the result remains substantially the same, thereby obtaining an optimal correlation matrix;
• activating the computer to decode a block code, utilizing the optimal correlation matrix;
• and
• utilizing the decoded block code,
• wherein the different codeword is represented by N-dimensional column vectors, and the original codeword is represented by M-dimensional column vectors so that the pre-learned correlation matrix has N rows and M columns, and
• wherein in the calculation, the different codeword is multiplied by the pre-learned correlation matrix to obtain the M-dimensional column vectors.