Geometric deep learning for lattice reduction

Patent metadata

Field	Value
Publication number	US-12413270-B2
Application number	US-202418410660-A
Country	US
Kind code	B2
Filing date	Jan 11, 2024
Priority date	Jan 11, 2024
Publication date	Sep 9, 2025
Grant date	Sep 9, 2025

Abstract

Certain aspects of the present disclosure provide techniques for wireless communications by an apparatus. Certain techniques include receiving signals corresponding to a MIMO channel matrix; generating a first gram matrix from a basis for a first lattice corresponding to a first signal of the received signals; providing the first gram matrix to a neural lattice reduction model comprising an equivariant neural network configured to generate a current extended Gauss move; generating, with the neural lattice reduction model, a current partial changed basis based on the current extended Gauss move and the basis; executing one or more additional iterations of the neural lattice reduction model; and demapping the MIMO channel matrix based on combining the current partial changed basis and each of the additional partial changed basis generated by each of the one or more additional iterations of the neural lattice reduction model.

First claim

What is claimed is:

1. An apparatus configured for wireless communication, comprising:
one or more memories; and
one or more processors, coupled to the one or more memories, configured to cause the apparatus to:
receive signals corresponding to a multiple-input multiple-output (MIMO) channel matrix;
generate a first gram matrix from a basis for a first lattice corresponding to a first signal of the received signals;
provide the first gram matrix to a neural lattice reduction model comprising an equivariant neural network configured to generate a current extended Gauss move;
generate, with the neural lattice reduction model, a current partial changed basis based on the current extended Gauss move and the basis;
execute one or more additional iterations of the neural lattice reduction model, wherein each iteration comprises:
generating a subsequent gram matrix from the current partial changed basis,
providing the subsequent gram matrix to the equivariant neural network to generate a subsequent extended Gauss move, and
generating an additional partial changed basis based on the subsequent extended Gauss move and the current partial changed basis; and

demap the MIMO channel matrix based on combining the current partial changed basis and each of the additional partial changed basis generated by each of the one or more additional iterations of the neural lattice reduction model.

2. The apparatus of claim 1 , wherein combining the current partial changed basis and each of the additional partial changed basis form a first reduced basis for the first lattice.

3. The apparatus of claim 2 , wherein the one or more processors are configured to further cause the apparatus to:
generate a second gram matrix from a second basis for a second lattice corresponding to a second signal of the received signals;
generate, with the neural lattice reduction model, one or more second partial changed bases for the second lattice; and
generate a second reduced basis for the second lattice based on the one or more second partial changed bases, wherein the one or more processors generate the first reduced basis and the second reduced basis through parallel processing of the neural lattice reduction model.

4. The apparatus of claim 3 , further comprising a plurality of antennas, wherein the first signal of the received signals is received by a first antenna of the plurality of antennas and the second signal of the received signals is received by a second antenna of the plurality of antennas.

5. The apparatus of claim 3 , wherein the parallel processing of the neural lattice reduction model is implemented by graphic processing unit batching.

6. The apparatus of claim 3 , wherein the one or more processors are configured to further cause the apparatus to demap the MIMO channel matrix based on the first reduced basis and the second reduced basis.

7. The apparatus of claim 2 , wherein the first reduced basis is more orthogonal and shorter than the basis for the first lattice corresponding to the MIMO channel matrix.

8. The apparatus of claim 1 , further comprising a plurality of antennas configured to receive the signals, wherein the received signals are distributed across time and frequency.

9. A method for wireless communications by an apparatus comprising:
receiving signals corresponding to a multiple-input multiple-output (MIMO) channel matrix;
generating a first gram matrix from a basis for a first lattice corresponding to a first signal of the received signals;
providing the first gram matrix to a neural lattice reduction model comprising an equivariant neural network configured to generate a current extended Gauss move;
generating, with the neural lattice reduction model, a current partial changed basis based on the current extended Gauss move and the basis;
executing one or more additional iterations of the neural lattice reduction model, wherein each iteration comprises: generating a subsequent gram matrix from the current partial changed basis, providing the subsequent gram matrix to the equivariant neural network to generate a subsequent extended Gauss move, and generating an additional partial changed basis based on the subsequent extended Gauss move and the current partial changed basis; and
demapping the MIMO channel matrix based on combining the current partial changed basis and each of the additional partial changed basis generated by each of the one or more additional iterations of the neural lattice reduction model.

10. The method of claim 9 , wherein combining the current partial changed basis and each of the additional partial changed basis form a first reduced basis for the first lattice.

11. The method of claim 10 , further comprising:
generating a second gram matrix from a second basis for a second lattice corresponding to a second signal of the received signals;
generating, with the neural lattice reduction model, one or more second partial changed bases for the second lattice; and
generating a second reduced basis for the second lattice based on the one or more second partial changed bases, wherein the first reduced basis and the second reduced basis are generated through parallel processing of the neural lattice reduction model.

12. The method of claim 11 , wherein the first signal of the received signals is received by a first antenna of a plurality of antennas and the second signal of the received signals is received by a second antenna of the plurality of antennas.

13. The method of claim 11 , wherein the parallel processing of the neural lattice reduction model is implemented by graphic processing unit batching.

14. The method of claim 11 , further comprising demapping the MIMO channel matrix based on the first reduced basis and the second reduced basis.

15. The method of claim 10 , wherein the first reduced basis is more orthogonal and shorter than the basis for the first lattice corresponding to the MIMO channel matrix.

16. The method of claim 9 , wherein the received signals are distributed across time and frequency.

17. A non-transitory computer-readable medium comprising processor-executable instructions that, when executed by one or more processors of an apparatus, causes the apparatus to perform a method comprising:
receiving signals corresponding to a multiple-input multiple-output (MIMO) channel matrix;
generating a first gram matrix from a basis for a first lattice corresponding to a first signal of the received signals;
providing the first gram matrix to a neural lattice reduction model comprising an equivariant neural network configured to generate a current extended Gauss move;
generating, with the neural lattice reduction model, a current partial changed basis based on the current extended Gauss move and the basis;
executing one or more additional iterations of the neural lattice reduction model, wherein each iteration comprises: generating a subsequent gram matrix from the current partial changed basis, providing the subsequent gram matrix to the equivariant neural network to generate a subsequent extended Gauss move, and generating an additional partial changed basis based on the subsequent extended Gauss move and the current partial changed basis; and
demapping the MIMO channel matrix based on combining the current partial changed basis and each of the additional partial changed basis generated by each of the one or more additional iterations of the neural lattice reduction model.

18. The non-transitory computer-readable medium of claim 17 , wherein combining the current partial changed basis and each of the ad

Assignees

• Qualcomm Inc

Inventors

• Marchetti Giovanni Luca
• Cesa Gabriele
• Pratik Kumar
• Behboodi Arash

Classifications

• H04B7/08 Primary

  at the receiving station · CPC title

• H04B7/0413 Primary

  MIMO systems · CPC title