A three-step adversarial based indoor positioning method

By combining three-step adversarial training with multi-gradient descent, the positioning accuracy problem of indoor positioning methods under interference such as multipath effect and channel noise is solved, and high-precision indoor positioning is achieved in dynamic environments and high-noise conditions.

CN116684820BActive Publication Date: 2026-05-29SICHUAN HYBRID POSITIONING TECH CO LTD
0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
SICHUAN HYBRID POSITIONING TECH CO LTD
Filing Date
2023-06-09
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

Existing indoor positioning methods are not ideal in terms of positioning accuracy under dynamic environmental changes and various interference factors, especially the effects of multipath effect, channel noise and human shadow effect. Furthermore, traditional methods fail to effectively utilize the similarity and difference information between multiple measurement data.

Method used

A three-step adversarial training method combined with multi-gradient descent is adopted. Through an autoencoder, feature extractor and regressor, the feature extractor is trained to transform the source domain and target domain data into a feature space with the same distribution and make it lie in the 'flat' part of the regression hyperplane of the regressor. Multiple measurement data are used to reduce the influence of interference. At the same time, the weights of the loss function are self-learned to balance the differences in domain distribution and multiple measurement differences.

Benefits of technology

It improves the robustness and accuracy of indoor positioning, adapts to long-term environmental changes and high-noise environments, and reduces the hyperparameter requirements for model training.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116684820B_ABST
    Figure CN116684820B_ABST
Patent Text Reader

Abstract

The present application belongs to the technical field of indoor positioning, and particularly relates to an indoor positioning method based on three-step confrontation. The present application solves problem a) by minimizing the multi-core maximum mean difference between offline and online data, and solves problem b) by three-step maximum minimization confrontation learning. In addition, the present application adopts a multi-gradient descent algorithm to balance the processing capacity of the model for the above two problems, and enables the present application to not require any hyperparameters. The present application enhances the generalization of the model to the differences between different samples in the target domain class by using three-step confrontation learning, overcomes the problem that traditional domain adaptation techniques only align the inter-domain marginal distribution and ignore the differences between different samples in the class, and thus achieves a high-precision positioning effect.
Need to check novelty before this filing date? Find Prior Art