# US 2021/0241173 A1 · Device-free localization robust to environmental changes

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2021/0241173 A1
- Document ID: US20210241173A1
- Status: Published application
- Kind: A1
- Application number: 17/008,218
- Filed: 2020-08-31
- Published: 2021-08-05
- Earliest claimed: 2020-02-05
- Assignees: Samsung Electronics Co., Ltd.
- Applicants: Samsung Electronics Co., Ltd.
- Inventors: Hang Li; Xi Chen; Di Wu; Xue Liu
- Web page: https://patentgrep.com/patent/US20210241173A1

## CPC

- G01S 5/0252: {Radio frequency fingerprinting}
- G06N 20/00: Machine learning
- G01P 13/00: Indicating or recording presence, absence, or direction, of movement (electric switches H01H; counting moving objects G06M7/00)
- G01S 5/0284: {Relative positioning}
- G06N 3/045: Combinations of networks
- G06N 3/048: Activation functions
- G06N 3/0499: Feedforward networks
- G06N 3/084: Backpropagation, e.g. using gradient descent
- G06N 3/0895: Weakly supervised learning, e.g. semi-supervised or self-supervised learning
- G06N 3/09: Supervised learning
- G06N 3/094: Adversarial learning
- G06N 3/096: Transfer learning
- G06N 5/04: Inference or reasoning models
- H04W 4/02: Services making use of location information
- G01P 3/66: using electric or magnetic means (G01P3/80 takes precedence; measuring short time intervals G04F8/00, G04F10/00)
- H04W 84/12: WLAN [Wireless Local Area Networks]
- G01S: RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G01P: MEASURING LINEAR OR ANGULAR SPEED, ACCELERATION, DECELERATION, OR SHOCK; INDICATING PRESENCE, ABSENCE, OR DIRECTION, OF MOVEMENT (measuring or recording blood flow A61B5/02, A61B8/06; monitoring speed or deceleration of electrically-propelled vehicles B60L3/00; vehicle lighting systems adapted to indicate speed B60Q1/54; determining position or course in navigation, measuring ground distance in geodesy or surveying G01C; combined measuring devices for measuring two or more variables of movement G01C23/00; measuring velocity of sound G01H; measuring velocity of light G01J7/00; measuring direction or velocity of solid objects by reception or emission of radiowaves or other waves and based on propagation effects, e.g. Doppler effect, propagation time, direction of propagation, G01S; measuring speed of nuclear radiation G01T; measuring acceleration of gravity G01V; {measuring or recording the speed of trains B61L23/00; speed indicators incorporated in motor vehicles B60K35/00; measuring frequency or phase G01R; traffic control G08G})
- H04W: WIRELESS COMMUNICATION NETWORKS (broadcast communication H04H; communication systems using wireless links for non-selective communication, e.g. wireless extensions H04M1/72)

## Abstract

A method of location determination with a WiFi transceiver and an AI model includes jointly training, based on various losses: a feature extractor, a location classifier, and a domain classifier. The domain classifier may include a first domain classifier and a second domain classifier. The losses used for training tend to cause feature data from the feature extractor to cluster even if a physical object in an environment has moved after training is completed. Then, the location classifier is able to accurately estimate the position of, for example, a person in a room, even if a door or window has changed from open to close or close to open between the time of training and the time of estimating the person's position.

## Claims

### Claim 1

A method of location determination by a system including an electronic device with a WiFi transceiver and an AI model, the method comprising:
jointly training, based on a cross-domain loss, a clustering loss, and a triplet loss:
a feature extractor comprising at least a feature extraction layer of the AI model,
a location classifier comprising at least a location classifier layer of the AI model, and
a domain classifier comprising at least a domain classifier layer of the AI model;
classifying, using the location classifier layer of the AI model and based on a first set of feature data from the feature extraction layer, first channel state information (CSI) data to obtain a first estimated location label of a first person in a first environment;
receiving, via the WiFi transceiver, second CSI data associated with a second person at a first physical location and a second environment;
obtaining, using the feature extraction layer of the AI model and based on operating on the second CSI data, a second set of feature data;
classifying, using the location classifier layer of the AI model and based on the second set of feature data from the feature extraction layer, the second CSI data to obtain a second estimated location label of the second person, and
outputting, by a visible indication or an audible indication, a second indicator of the second estimated location label of the second person.

### Claim 2 (depends on 1)

The method of claim 1 further comprising:
receiving, via the WiFi transceiver, the first channel state information (CSI) data associated with the first person at the first physical location and the first environment; and
obtaining, using the feature extraction layer of the AI model and based on operating on the first CSI data, the first set of feature data.

### Claim 3 (depends on 2)

The method of claim 2, further comprising outputting, by the visible indication or the audible indication, a first indicator of the first estimated location label of the first person.

### Claim 4 (depends on 2)

The method of claim 2, further comprising:
back-propagating a first gradient of the cross-domain loss to update the domain classifier layer; and
back-propagating a reversal of the first gradient to train the feature extraction layer, thereby conditioning the feature extraction layer to map both the first CSI data of a source domain and the second CSI data of a target domain into a same feature space associated with the first physical location.

### Claim 5 (depends on 2)

The method of claim 2, wherein the first environment is associated with a source domain, the second environment is associated with a target domain, the cross-domain loss is associated with the source domain and the target domain, the clustering loss is associated with only the target domain, the triplet loss is associated with the source domain and the target domain, the AI model is a neural network based AI model, the second environment is different than the first environment, and the jointly training is configured to update the domain classifier layer of the AI model without a knowledge of a third location label of the second CSI data, thereby permitting domain adaptation even under an environmental change.

### Claim 6 (depends on 2)

The method of claim 2, further comprising determining, using a reconstruction layer of the AI model and based on the first set of feature data from the feature extraction layer, a reconstruction loss associated with the first person and the first CSI data.

### Claim 7 (depends on 1)

The method of claim 1, wherein jointly training further comprises jointly training based on a location classification loss in a source domain.

### Claim 8 (depends on 1)

The method of claim 1, wherein the cross-domain loss is a first objective function with a first term and a second term so as to perform adversarial learning, whereby the first term is configured to decrease the cross-domain loss of the domain classifier layer of the AI model and the second term is configured to increase the cross-domain loss of the feature extraction layer of the AI model.

### Claim 9 (depends on 1)

The method of claim 1, wherein the clustering loss is a second objective function with an entropy term based on only a target domain.

### Claim 10 (depends on 1)

The method of claim 1, wherein the triplet loss is a third objective function comprising a first term and a second term, the first term being a distance measure in a feature space between source domain anchor points and first target domain points, and the second term being the distance measure in the feature space between the source domain anchor points and second target domain points, thereby tending to align the first target domain points with a first location and tending to align the second target domain points with a second location different than the first location.

### Claim 11 (depends on 1)

The method of claim 1, further comprising registering the first person, wherein the registering comprises:
obtaining, via a user interface of the system, an identifier of the first person,
obtaining, via the user interface, a ground truth location label of the first person, and
obtaining, via the WiFi transceiver, third CSI data associated with the ground truth location label; and
providing an application output based on the first estimated location label or the ground truth location label.

### Claim 12 (depends on 1)

The method of claim 1, further comprising training weights of the system, based on third CSI data associated with the first environment and with a second person who is a first registered user of the system, to obtain updated weights.

### Claim 13 (depends on 11)

The method of claim 11, wherein the application output comprises the electronic device automatically shutting off based on recognizing that a room associated with the first environment and with the second environment is empty.

### Claim 14 (depends on 11)

The method of claim 11, wherein the application output comprises beamforming, by the electronic device, sound to the first estimated location label of the first person.

### Claim 15 (depends on 11)

The method of claim 11, wherein the application output comprises pan-tilting, by the electronic device, a display screen of the electronic device depending on the first estimated location label of the first person.

### Claim 16 (depends on 11)

The method of claim 11, wherein the application output comprises updating an augmented reality presentation to the first person based on the first estimated location label of the first person.

### Claim 17 (depends on 11)

The method of claim 11, wherein the application output comprises providing an advertisement to a personal electronic device of the first person based on the first estimated location label of the first person.

### Claim 18 (depends on 11)

The method of claim 11, wherein the application output comprises a first indication of an estimated location of the first person.

### Claim 19 (depends on 11)

The method of claim 11, wherein the application output comprises a second indication of whether the first person is stationary or moving within a predetermined period of time.

### Claim 20 (depends on 11)

The method of claim 11, wherein the application output comprises a third indication a direction of motion of the first person within a predetermined period of time.

### Claim 21 (depends on 11)

The method of claim 11, wherein the application output comprises a fourth indication of a speed of motion of the first person within a predetermined period of time.

### Claim 22 (depends on 1)

The method of claim 1, further comprising:
obtaining, via a user interface of the system, an identifier of the first environment as a source domain or the second environment as a target domain.

### Claim 23 (depends on 1)

The method of claim 1, further comprising:
determining, based on a passage of a predetermined time, that a current environment has changed from a source domain to a target domain.

### Claim 24 (depends on 1)

The method of claim 1, further comprising:
determining, based on a reconstruction loss measure, that a current environment has changed from a source domain to a target domain.

### Claim 25 (depends on 1)

The method of claim 1, wherein the domain classifier is configured to align a first marginal distribution of a source domain and a second marginal distribution of a target domain.

### Claim 26 (depends on 21)

The method of claim 21, wherein the domain classifier includes a second domain classifier, wherein the second domain classifier is configured to match a first conditional distribution of a source domain and a second conditional distribution of a target domain, and a second AI model comprising the second domain classifier is an extended form of the AI model.

### Claim 27 (depends on 26)

The method of claim 26, wherein an input to the second domain classifier comprises a concatenation of a first output of the location classifier and a first output of the feature extractor.

### Claim 28

An electronic device comprising:
one or more memories, wherein the one or more memories comprise instructions;
a WiFi transceiver; and
one or more processors, wherein the one or more processors are configured to execute the instructions to:
jointly train, based on a cross-domain loss, a clustering loss, and a triplet loss:
a feature extractor comprising at least a feature extraction layer of an AI model,
a location classifier comprising at least a location classifier layer of the AI model, and
a domain classifier comprising at least a domain classifier layer of the AI model;
classify, using the location classifier layer of the AI model and based on a first set of feature data from the feature extraction layer, first channel state information (CSI) data to obtain a first estimated location label of a first person in a first environment;
receive, via the WiFi transceiver, second CSI data associated with a second person at a first physical location and a second environment;
obtain, using the feature extraction layer of the AI model and based on operating on the second CSI data, a second set of feature data;
classify, using the location classifier layer of the AI model and based on the second set of feature data from the feature extraction layer, the second CSI data to obtain a second estimated location label of the second person, and
output, by a visible indication or an audible indication, a second indicator of the second estimated location label of the second person.

### Claim 29

A non-transitory computer-readable storage medium storing instructions configured to cause a processor to:
jointly train, based on a cross-domain loss, a clustering loss, and a triplet loss:
a feature extractor comprising at least a feature extraction layer of an AI model,
a location classifier comprising at least a location classifier layer of the AI model, and
a domain classifier comprising at least a domain classifier layer of the AI model;
classify, using the location classifier layer of the AI model and based on a first set of feature data from the feature extraction layer, first channel state information (CSI) data to obtain a first estimated location label of a first person in a first environment;
receive, via a WiFi transceiver, second CSI data associated with a second person at a first physical location and a second environment;
obtain, using the feature extraction layer of the AI model and based on operating on the second CSI data, a second set of feature data;
classify, using the location classifier layer of the AI model and based on the second set of feature data from the feature extraction layer, the second CSI data to obtain a second estimated location label of the second person, and
output, by a visible indication or an audible indication, a second indicator of the second estimated location label of the second person.