Signal dimension reduction using a non-linear transformation

Patent metadata

Field	Value
Publication number	US-11770286-B2
Application number	US-201917753737-A
Country	US
Kind code	B2
Filing date	Sep 13, 2019
Priority date	Sep 13, 2019
Publication date	Sep 26, 2023
Grant date	Sep 26, 2023

Abstract

A method performed by a radio unit for handling a number of received radio signals over an array of antennas comprised in the radio unit. The radio unit transforms the number of received radio signals into a number of sequences of complex symbols. The radio unit further filters the number of sequences of complex symbols by inputting the number of sequences of complex symbols into a trained computational model comprising an alternating sequence of linear and nonlinear functions and thereby obtaining a reduced number of sequences. The radio unit further transmits the reduced number of sequences to a baseband unit over a front-haul link.

First claim

The invention claimed is:

1. A method performed by a radio unit for handling a number of received radio signals over an array of antennas comprised in the radio unit, the method comprising:
transforming the number of received radio signals into a number of sequences of complex symbols;
filtering the number of sequences of complex symbols by inputting the number of sequences of complex symbols into a trained computational model comprising an alternating sequence of linear and nonlinear functions and thereby obtaining a reduced number of sequences; and
transmitting the reduced number of sequences to a baseband unit over a front-haul link.

2. The method according to claim 1 , wherein transforming the number of received radio signals comprises Discrete Fourier Transform (DFT) over-sampling the number of sequences of complex symbols; and wherein filtering the number of sequences of complex symbols comprises inputting the DFT over-sampled number of sequences into the trained computational model.

3. The method according to claim 2 , wherein the DFT over-sampling is performed by using a DFT matrix multiplication on the number of sequences of complex symbols.

4. The method according to claim 1 , wherein the trained computational model comprises an auto-encoder.

5. The method according to claim 1 , wherein the trained computational model comprises a convolutional auto-encoder, a recurrent auto-encoder, a neural Turing machine, a perceptron, or any combination thereof.

6. The method according to claim 1 , further comprising training the computational model at site.

7. A method performed by a baseband unit for handling sequences from a radio unit received over a front-haul link, the method comprising:
receiving a reduced number of sequences from the radio unit;
decompressing the reduced number of sequences by inputting the reduced number of sequences into a trained computational model comprising an alternating sequence of linear and nonlinear functions and thereby obtaining a number of sequences of complex symbols; and
processing the number of sequences of complex symbols for decoding signals received over an array of antennas of the radio unit.

8. The method according to claim 7 , wherein the trained computational model comprises an auto-decoder.

9. The method according to claim 7 , wherein the trained computational model comprises a convolutional auto-decoder, a recurrent auto-decoder, a neural Turing machine, a perceptron, or any combination thereof.

10. A radio unit for handling a number of received radio signals over an array of antennas comprised in the radio unit, wherein the radio unit comprising:
at least one processor; and
a memory comprising instructions which, when executed by the at least one processor, cause the radio unit to:
transform the number of received radio signals into a number of sequences of complex symbols;
filter the number of sequences of complex symbols by inputting the number of sequences of complex symbols into a trained computational model comprising an alternating sequence of linear and nonlinear functions and thereby obtaining a reduced number of sequences; and
transmit the reduced number of sequences to a baseband unit over a front-haul link.

11. The radio unit according to claim 10 , wherein the radio unit is to transform the number of received radio signals by Discrete Fourier Transform DFT over-sample of the number of sequences of complex symbols; and wherein the radio unit is to filter the number of sequences of complex symbols by input of the DFT over-sampled number of sequences into the trained computational model.

12. The radio unit according to claim 11 , wherein the radio unit is to DFT over-sample by using a DFT matrix multiplication on the number of sequences of complex symbols.

13. The radio unit according to claim 10 , wherein the trained computational model comprises an auto-encoder.

14. The radio unit according to claim 10 , wherein the trained computational model comprises a convolutional auto-encoder, a recurrent auto-encoder, a neural Turing machine, a perceptron, or any combination thereof.

15. The radio unit according to claim 10 , wherein the radio unit is further to train the computational model at site.

16. A baseband unit for handling sequences from a radio unit received over a front-haul link, wherein the baseband unit comprising:
at least one processor; and
a memory comprising instructions which, when executed by the at least one processor, cause the baseband unit to:
receive a reduced number of sequences from the radio unit;
decompress the reduced number of sequences by inputting the reduced number of sequences into a trained computational model comprising an alternating sequence of linear and nonlinear functions and thereby obtaining a number of sequences of complex symbols; and
process the number of sequences of complex symbols for decoding signals received over an array of antennas of the radio unit.

17. The baseband unit according to claim 16 , wherein the trained computational model comprises an auto-decoder.

18. The baseband unit according to claim 16 , wherein the trained computational model comprises a convolutional auto-decoder, a recurrent auto-decoder, a neural Turing machine, a perceptron, or any combination thereof.

Assignees

• Ericsson Telefon Ab L M

Inventors

• Timo Roy
• Berg Miguel
• Wernersson Niklas
• Wang Zhao

Classifications

• G06N3/0455

  Auto-encoder networks; Encoder-decoder networks · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/09

  Supervised learning · CPC title

• H04L27/2647 Primary

  Arrangements specific to the receiver only (equalisation H04L27/01) · CPC title

• H04B7/086

  using weights depending on external parameters, e.g. direction of arrival [DOA], predetermined weights or beamforming · CPC title