Neural network computation for eigen value and eigen vector decomposition of matrices
A method performs eigen decomposition with an artificial deep neural network. The deep neural network receives an input covariance matrix. The deep neural network has a number of convolutional layers and also a number of pooling layers. The deep neural network predicts dominant eigen information of the input covariance matrix, after applying the convolutional layers and the pooling layers to the input covariance matrix. The input covariance matrix may be a real-valued covariance matrix or a complex-valued covariance matrix having a concatenated pair of matrices, including a first matrix of real components and a second matrix of imaginary components. The dominant eigen information may be absolute values of a pair of dominant eigen values and sign information of the pair of dominant eigen values, and/or absolute values of a pair of dominant eigen vectors and sign information of the pair of dominant eigen vectors.
1. A method of performing eigen decomposition with an artificial deep neural network, comprising:
receiving an input covariance matrix;
applying a first two dimensional convolutional layer to the input covariance matrix by implementing a parametric rectified linear unit (PRELU) activation function to generate a first output;
applying a second two dimensional convolutional layer to the first output by implementing the PRELU activation function to generate a second output;
applying a first two dimensional average pooling layer to the second output to generate a first pooled output;
applying a third two dimensional convolutional layer to the first pooled output by implementing the PRELU activation function to generate a third output;
applying a second two dimensional average pooling layer to the third output to generate a second pooled output; and
predicting dominant eigen information of the input covariance matrix by processing the second pooled output into an output array having a first channel representing absolute values of the dominant eigen information.
2. The method of claim 1 , in which the input covariance matrix comprises a real-valued covariance matrix.
3. The method of claim 1 , in which the input covariance matrix comprises a complex-valued covariance matrix.
4. The method of claim 3 , in which the complex-valued covariance matrix comprises a concatenated pair of matrices, including a first matrix of real components and a second matrix of imaginary components.
5. The method of claim 1 , in which the dominant eigen information comprises absolute values of a pair of dominant eigen values and sign information of the pair of dominant eigen values.
6. The method of claim 5 , in which the sign information comprises an additional channel in the output array.
7. The method of claim 1 , in which the dominant eigen information comprises absolute values of a pair of dominant eigen vectors and sign information of the pair of dominant eigen vectors.
8. The method of claim 7 , in which the sign information comprises an additional channel in the output array.
9. The method of claim 7 , further comprising applying a final two dimensional convolutional layer to a convolved second pooled output by performing matrix multiplication without an activation function.
10. An apparatus for wireless communication by a user equipment (UE), comprising:
at least one processor, memory coupled with the at least one processor; and
instructions stored in the memory and operable, when executed by the at least one processor, to cause the apparatus:
to receive an input covariance matrix;
to apply a first two dimensional convolutional layer to the input covariance matrix by implementing a parametric rectified linear unit (PRELU) activation function to generate a first output;
to apply a second two dimensional convolutional layer to the first output by implementing the PRELU activation function to generate a second output;
to apply a first two dimensional average pooling layer to the second output to generate a first pooled output;
to apply a third two dimensional convolutional layer to the first pooled output by implementing the PRELU activation function to generate a third output;
to apply a second two dimensional average pooling layer to the third output to generate a second pooled output; and
to predict dominant eigen information of the input covariance matrix by processing the second pooled output into an output array having a first channel representing absolute values of the dominant eigen information.
11. The apparatus of claim 10 , in which the input covariance matrix comprises a real-valued covariance matrix.
12. The apparatus of claim 10 , in which the input covariance matrix comprises a complex-valued covariance matrix.
13. The apparatus of claim 12 , in which the complex-valued covariance matrix comprises a concatenated pair of matrices, including a first matrix of real components and a second matrix of imaginary components.
14. The apparatus of claim 10 , in which the dominant eigen information comprises absolute values of a pair of dominant eigen values and sign information of the pair of dominant eigen values.
15. The apparatus of claim 14 , in which the sign information comprises an additional channel in the output array.
16. The apparatus of claim 10 , in which the dominant eigen information comprises absolute values of a pair of dominant eigen vectors and sign information of the pair of dominant eigen vectors.
17. The apparatus of claim 16 , in which the sign information comprises an additional channel in the output array.
18. The apparatus of claim 16 , in which the at least one processor is further configured to apply a final two dimensional convolutional layer to a convolved second pooled output by performing matrix multiplication without an activation function.
19. An apparatus, comprising:
means for receiving an input covariance matrix;
means for applying a first two dimensional convolutional layer to the input covariance matrix by implementing a parametric rectified linear unit (PRELU) activation function to generate a first output;
means for applying a second two dimensional convolutional layer to the first output by implementing the PRELU activation function to generate a second output;
means for applying a first two dimensional average pooling layer to the second output to generate a first pooled output;
means for applying a third two dimensional convolutional layer to the first pooled output by implementing the PRELU activation function to generate a third output;
means for applying a second two dimensional average pooling layer to the third output to generate a second pooled output; and
means for predicting dominant eigen information of the input covariance matrix by processing the second pooled output into an output array having a first channel representing absolute values of the dominant eigen information.
20. The apparatus of claim 19 , in which the input covariance matrix comprises a real-valued covariance matrix.
21. The apparatus of claim 19 , in which the input covariance matrix comprises a complex-valued covariance matrix.
22. The apparatus of claim 21 , in which the complex-valued covariance matrix comprises a concatenated pair of matrices, including a first matrix of real components and a second matrix of imaginary components.
23. The apparatus of claim 19 , in which the dominant eigen information comprises absolute values of a pair of dominant eigen values and sign information of the pair of dominant eigen values.
24. The apparatus of claim 23 , in which the sign information comprises an additional channel in the output array.