| 摘要 |
PROBLEM TO BE SOLVED: To provide a method for learning a correlation matrix, which obtains an optimum correlation matrix at an optimum threshold to ensure a sufficient margin against bit errors among all code words by stepwise increasing the threshold at learning of the correlation matrix as the learning progresses. SOLUTION: An arithmetic section 15 obtains a product of a code word X and a correlation matrix W. A comparison section 17 compares the product from an respectively section with a threshold±TH. A learning supervisory section 18 updates the correlation matrix on the basis of the result of comparison by the comparison section. After updating all the code words, when the result of the arithmetic operation using the correlation matrix after the update is within a prescribed range, the threshold is revised and the learning of all the code words is again repeated. Furthermore, when the arithmetic result using the correlation matrix after the update is not within a prescribed range and the arithmetic result before the update is equal to the arithmetic result after the update, the update value of the correlation matrix is decreased and the learning is repeated. When the update value of the correlation matrix is zero, it is considered that an optimum threshold and an optimum correlation matrix are obtained and the learning is completed.
|
