Unsupervised learning for simultaneous localization and mapping in deep neural networks using channel state information

Patent metadata

Field	Value
Publication number	US-12200660-B2
Application number	US-202117461927-A
Country	US
Kind code	B2
Filing date	Aug 30, 2021
Priority date	Aug 31, 2020
Publication date	Jan 14, 2025
Grant date	Jan 14, 2025

Abstract

A method of training an artificial neural network (ANN), receives, from a base station, signal information for a radio frequency signal between the base station and a user equipment (UE). The artificial neural network is trained to determine a location of the UE and to map the environment based on the received signal information and in the absence of labeled data.

First claim

What is claimed is:

1. A method of localization and mapping using an artificial neural network (ANN), comprising:
receiving, by the ANN, input signal information from one or more base stations for radio frequency signals between the one or more base stations and a user equipment (UE) in an environment;
determining, by the ANN, a location of the UE based on the input signal information; and
decoding, with a channel model, the location of the UE to generate reconstructed signal information based on a set of learned environment parameters corresponding to propagation effects in the environment and the location information determined by the ANN, the propagation effects including one or more of reflections or scattering of the radio frequency signals, the learned environment parameters based on a training loss computed from a difference between the input signal information and the reconstructed signal information the learned environment parameters being input to the channel model.

2. The method of claim 1 , in which the input signal information is channel state information, a time of flight or an angle of arrival.

3. The method of claim 1 , further comprising training the ANN using back propagation.

4. The method of claim 1 , in which the input signal information is unlabeled data.

5. The method of claim 1 , further comprising:
determining one or more virtual anchors based on the reflections of the radio frequency signals; and
jointly learning a position of the one or more virtual anchors and parameters of the ANN.

6. A method of localization and mapping using an artificial neural network (ANN), comprising:
receiving, via a base station, input signal information from one or more user equipments (UEs) for radio frequency signals between the one or more UEs and the base station in an environment;
determining locations of the one or more UEs based on the input signal information; and
decoding, with a channel model, the locations of the one or more UEs to generate reconstructed signal information based on a set of learned environment parameters corresponding to propagation effects in the environment and the determined locations of the one or more UEs, the propagation effects including one or more of reflections or scattering of the radio frequency signals, the learned environment parameters based on a training loss computed from a difference between the input signal information and the reconstructed signal information the learned environment parameters being input to the channel model.

7. The method of claim 6 , in which the input signal information is channel state information, a time of flight or an angle of arrival.

8. The method of claim 6 , further comprising training the ANN using back propagation.

9. The method of claim 6 , in which the input signal information is unlabeled data.

10. The method of claim 6 , further comprising:
determining one or more virtual anchors based on the reflections of the radio frequency signals; and
jointly learning a position of the one or more virtual anchors and parameters of the ANN.

11. An artificial neural network comprising:
an encoder configured to determine a location of one or more user equipments (UEs) based on input signal information from the one or more UEs for radio frequency signals between the one or more UEs and a base station in an environment; and
a decoder, comprising a channel model and configured to parameterize the environment of the one or more UEs and generate reconstructed signal information based at least in part on the location of the one or more UEs and a set of learned environment parameters corresponding to propagation effects in the environment for mapping the input signal information to location information, the propagation effects including one or more of reflections or scattering of the radio frequency signals, the learned environment parameters based on a training loss computed from a difference between the input signal information and the reconstructed signal information the learned environment parameters being input to the channel model.

12. The artificial neural network of claim 11 , in which the encoder comprises a second artificial neural network and the decoder includes the channel model.

13. The artificial neural network of claim 12 , in which the channel model comprises a function of the set of learned environment parameters.

14. The artificial neural network of claim 13 , in which the set of learned environment parameters comprise locations of one or more virtual nodes representing signal propagation effects associated with the radio frequency signals between the one or more UEs and the base station.

15. The artificial neural network of claim 12 , in which the channel model comprises a generative artificial neural network.

16. The artificial neural network of claim 12 , in which the encoder and the decoder are trained simultaneously using backpropagation.

17. A method of operating an artificial neural network (ANN), comprising:
receiving transmissions of radio frequency (RF) signals from one or more reference nodes in an environment;
inferring from the RF signals or input signal information derived from the RF signals, a location of a receiver of the RF signal transmissions relative to the one or more reference nodes; and
decoding, with a channel model, the location of the receiver to generate reconstructed signal information based on a set of learned environment parameters corresponding to propagation effects in the environment for mapping the input signal information to location information, the propagation effects including one or more of reflections or scattering of the RF signals, the learned environment parameters based on a training loss computed from a difference between the input signal information and the reconstructed signal information, the learned environment parameters being input to the channel model.

18. The method of claim 17 , in which the learned environment parameters are associated with one or more virtual node locations, the one or more virtual node locations corresponding to reflected signal transmissions from the one or more reference nodes received by the receiver.

19. The method of claim 18 , further comprising generating a mapping of the environment of the receiver of the RF signal transmission based on the location of the receiver and the learned environment parameters associated with the one or more virtual node locations.

Assignees

• Qualcomm Inc

Inventors

• Behboodi Arash
• Ghazvinian Zanjani Farhad
• Soriaga Joseph Binamira
• Ferrari Lorenzo
• Amjad Rana Ali
• Welling Max
• Yoo Taesang

Classifications

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/0475

  Generative networks · CPC title

• G06N3/0455

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

• G06N3/0895

  Weakly supervised learning, e.g. semi-supervised or self-supervised learning · CPC title

• G06N3/098

  Distributed learning, e.g. federated learning · CPC title