Method and system for controlling electronic device within multi-device environment

Abstract

Embodiments of the disclosure describe a method for controlling at least one electronic device (e.g., 100A, 100B,..,100N) within a multi-device environment. The method includes receiving Wi-Fi CSI data from each electronic device (e.g., 100A, 100B,..,100N) associated with the multi-device environment. The Wi-Fi CSI data may include at least a CSI signal and an ACF threshold. The method includes generating a plurality of clusters of the received Wi-Fi CSI data based on a ACF threshold-based dissimilarity learning mechanism and a threshold clustering mechanism with integration of a UCIE loss. The method includes determining a proximity of a user within the multi-device environment based on the plurality of generated clusters and the received Wi-Fi CSI data. The method includes controlling the at least one electronic device (e.g., 100A, 100B,..,100N) within the multi-device environment based on the determined proximity of the user.

Description

Description TITLE OF INVENTION: METHOD AND SYSTEM FOR CONTROLLING ELECTRONIC DEVICE WITHIN MULTI-DEVICE ENVIRONMENT Technical Field

[0001] The present invention generally relates to the field of signal processing, and more specifically relates to a method and a system for controlling at least one electronic device within a multi-device environment. Background Art

[0002] Proximity detection is a technology that can sense presence and a distance of objects or people in a specific area. In a smart home or multi-device environment, the proximity detection is used to enhance user experience and improve device automation. Wi-Fi Carrier Sense Multiple Access (CSMA) signals, also known as Wi- Fi Channel State Information (CSI), can be used for the proximity detection. The Wi- Fi CSI signal reflects one or more changes in a wireless environment, including the presence and movement of objects or people. By analyzing variations in these Wi-Fi CSI signals, a system can determine the proximity of nearby electronic devices or individuals. In a smart home environment, the proximity detection using Wi-Fi CSI signals can be leveraged to automatically adjust lighting, temperature, or media playback based on the user’s location within the smart home. The proximity detection can also be used for security, for example triggering alerts when unauthorized movement is detected. Specifically, the proximity detection technology provides a seamless and intelligent interaction between the user and the smart home or multi- device environment. However, several problems are encountered in the existing proximity detection, which are mentioned below.

[0003] Certain existing methods have explored the use of Wi-Fi-based solutions for presence detection, proximity detection, and similar applications. These approaches can be broadly categorized into two main groups, i.e., statistical methods and deep learning-based methods. Each category has associated advantages and disadvantages.

[0004] The statistical method generally exhibits a low inference time, meaning the time required to process and generate output from input data is relatively quick. Additionally, this method requires less memory resources compared to deep learning-based methods. However, the statistical method is limited in its capability to perform complex tasks, and it often demonstrates poor generalization, necessitating calibration for adaptation to different environmental conditions.

[0005] On the other hand, the deep learning-based method enables the execution of more complex tasks, leveraging the inherent capabilities of neural networks. This method also exhibits strong generalization abilities, eliminating the need for extensive calibration when adapting to new environments. However, the deep learning-based method typically incurs a higher inference time compared to statistical methods, and it requires a higher memory usage to accommodate the complexity of the neural network architecture.

[0006] Researchers face a trade-off when enabling tasks such as presence and proximity detection. They can either employ computationally expensive Deep Learning (DL) models, which require substantial amounts of data for generalization, resulting in significant initial costs, or utilize statistical methods, which entail lengthy calibration processes. Traditionally, many proximity detection methods based on an Auto Correlation Function (ACF) incorporate a threshold adaptability mechanism to adjust to environmental requirements. However, this calibration procedure can be excessively time-consuming, ranging from 15 minutes to 1 hour, posing a considerable inconvenience to users.

[0007] In addition, the existing methods for user proximity detection using Wi- Fi CSI have significant limitations in terms of quick environmental adaptation. This poses challenges in developing a generic solution that can work efficiently across different types of environments. In a home setting with multiple electronic devices like TVs, air conditioners, family hubs, microwaves, and washing machines, the presence detection becomes more complex when multiple electronic devices try to sense the user’s presence simultaneously. Scenarios, where the requirement is to detect the user’s proximity or area of movement concerning a specific electronic device, are difficult to develop using the existing methods, as illustrated in FIG.1.

[0008] For instance, in a first scenario 10, the common threshold-based presence detection of the user (10a and 10b) might not work consistently across different home environments (Home 1 and Home 2) with the same accuracy.

[0009] For instance, in a second scenario 11, the limitations of the Wi-Fi CSI- based presence detection solutions can lead to false alarms, where the electronic device 11a can detect the user’s presence 11b and turn on even without the user’s intention to use it. This can result in a family hub display screen 11a being turned on when the user is not in close proximity to the electronic device.

[0010] Furthermore, in a third scenario 12, multiple electronic devices at home can detect the user’s presence irrespective of their proximity to the electronic devices. For instance, the user 12a watching TV in a living area (Room-1) might cause the TV in a bedroom (Room-2) to turn on due to the presence detection, leading to unnecessary energy consumption.

[0011] As per the aforementioned scenarios, the existing methods have certain limitations, which may include: a. User presence sensing solutions are prone to false positives, with electronic devices detecting presence even when the user is not in close proximity. b. User presence and absence detection cannot be effectively enabled in multi-device environments. c. Without the proximity detection solution that can work quickly and efficiently across all types of environments, it is challenging to deploy proximity-based services at a commercial scale. d. Device-based services cannot operate at 100% efficiency without sensing the user’s presence near the electronic device, such as displaying the family hub screen only when the user approaches it. e. Adaptive proximity solutions are necessary to enable effective energy- saving scenarios across different home layouts.

[0012] Thus, it is desired to address the above-mentioned disadvantages or other shortcomings or at least provide a useful alternative for controlling at least one electronic device within the multi-device environment. Summary of Invention

[0013] This summary is provided to introduce a selection of concepts in a simplified format that are further described in the detailed description of the invention.This summary is not intended to identify key or essential inventive concepts of the invention, nor is it intended for determining the scope of the invention. Solution to Problem

[0014] According to one embodiment of the present disclosure, a method for controlling at least one electronic device within a multi-device environment is disclosed herein. The method includes receiving Wireless Fidelity (Wi-Fi) Channel State Information (CSI) data from each electronic device associated with the multi-device environment, where the Wi-Fi CSI data comprises at least a CSI signal and an Autocorrelation Function (ACF) threshold. The method further includes generating a plurality of clusters of the received Wi-Fi CSI data based on an ACF threshold-based dissimilarity learning mechanism and a threshold clustering mechanism with integration of an Unsupervised Class Information Entropy (UCIE) loss. The method further includes determining a proximity of a user within the multi-device environment based on the plurality of generated clusters and the received Wi-Fi CSI data. The method further includes controlling the at least one electronic device within the multi- device environment based on the determined proximity of the user.

[0015] According to another embodiment of the present disclosure, the electronic device for controlling the at least one electronic device within the multi-device environment is disclosed herein. The electronic device includes a multi device controlling module coupled with a memory, a processor, and a communicator. The multi device controlling module is configured to receive the Wi-Fi CSI data from each electronic device associated with the multi-device environment, where the Wi-Fi CSI data comprises at least a CSI signal and an Autocorrelation Function (ACF) threshold. The multi device controlling module is further configured to generate the plurality of clusters of the received Wi-Fi CSI data based on the ACF threshold-based dissimilarity learning mechanism and the threshold clustering mechanism with integration of the UCIE loss. The multi device controlling module is further configured to determine the proximity of the user within the multi-device environment based on the plurality of generated clusters and the received Wi-Fi CSI data. The multi device controlling module is further configured to control the at least one electronic device within the multi-device environment based on the determined proximity of the user.

[0016] To further clarify the advantages and features of the present invention, a more particular description of the invention will be rendered by reference to specific embodiments thereof, which are illustrated in the appended drawings. It is appreciated that these drawings depict only typical embodiments of the invention and are therefore not to be considered limiting of its scope. The invention will be described and explained with additional specificity and detail in the accompanying drawings. Brief Description of Drawings

[0017] These and other features, aspects, and advantages of the present invention will become better understood when the following detailed description is read with reference to the accompanying drawings in which like characters represent like parts throughout the drawings, wherein:

[0018] FIG.1 illustrates problem scenarios associated with the existing methods for user proximity detection using Wi-Fi CSI, according to prior art;

[0019] FIG.2 illustrates a block diagram of an electronic device for controlling at least one electronic device within a multi-device environment, according to an embodiment as disclosed herein;

[0020] FIG.3 is a flow diagram illustrating a method for receiving Wi-Fi CSI data from each electronic device associated with the multi-device environment for database creation, according to an embodiment as disclosed herein;

[0021] FIG. 4 illustrates a block diagram of a threshold learning unit of the electronic device for controlling the at least one electronic device within the multi- device environment, according to an embodiment as disclosed herein;

[0022] FIGS.5a to 5d illustrate one or more operations associated with an ACF threshold-based dissimilarity learning module of the threshold learning unit, according to an embodiment as disclosed herein;

[0023] FIGS.6a-6b illustrate one or more operations associated with a UCIE loss-based clustering module of the threshold learning unit, according to an embodiment as disclosed herein;

[0024] FIG. 7 illustrates one or more operations associated with a centroid detection module of the threshold learning unit, according to an embodiment as disclosed herein;

[0025] FIG.8a and FIG.8b illustrate one or more operations associated with a threshold adaption unit of the electronic device for applying a Quick Adaptation Strategy (QAS), according to an embodiment as disclosed herein;

[0026] FIG.9a and FIG.9b illustrate one or more operations associated with a human proximity tracking unit of the electronic device for determining the proximity of the user within the multi-device environment, according to an embodiment as disclosed herein

[0027] FIGS. 10a to 10C illustrate example scenarios associated with the disclosed method for controlling the at least one electronic device within the multi- device environment, according to an embodiment as disclosed herein; and

[0028] FIG.11 is a flow diagram illustrating a method for controlling the at least one electronic device within the multi-device environment, according to an embodiment as disclosed herein.

[0029] Further, skilled artisans will appreciate that elements in the drawings are illustrated for simplicity and may not have necessarily been drawn to scale. For example, the flow charts illustrate the method in terms of the most prominent steps involved to help to improve understanding of aspects of the present invention. Furthermore, in terms of the construction of the device, one or more components of the device may have been represented in the drawings by conventional symbols, and the drawings may show only those specific details that are pertinent to understanding the embodiments of the present invention so as not to obscure the drawings with details that will be readily apparent to those of ordinary skill in the art having the benefit of the description herein. Description of Embodiments

[0030] For the purpose of promoting an understanding of the principles of the invention, reference will now be made to the embodiment illustrated in the drawings and specific language will be used to describe the same. It will nevertheless be understood that no limitation of the scope of the invention is thereby intended, such alterations and further modifications in the illustrated system, and such further applications of the principles of the invention as illustrated therein being contemplated as would normally occur to one skilled in the art to which the invention relates.

[0031] It will be understood by those skilled in the art that the foregoing general description and the following detailed description are explanatory of the invention and are not intended to be restrictive thereof.

[0032] Reference throughout this specification to “an aspect”, “another aspect” or similar language means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present invention. Thus, appearances of the phrase “in an embodiment”, “in one embodiment”, “in another embodiment”, and similar language throughout this specification may, but do not necessarily, all refer to the same embodiment.

[0033] The terms “comprise”, “comprising”, or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process or method that comprises a list of steps does not include only those steps but may include other steps not expressly listed or inherent to such process or method. Similarly, one or more devices or sub-systems or elements or structures or components proceeded by “comprises... a” does not, without more constraints, preclude the existence of other devices or other sub-systems or other elements or other structures or other components or additional devices or additional sub-systems or additional elements or additional structures or additional components.

[0034] The embodiments herein and the various features and advantageous details thereof are explained more fully with reference to the non-limiting embodiments that are illustrated in the accompanying drawings and detailed in the following description. Descriptions of well-known components and processing techniques are omitted so as to not unnecessarily obscure the embodiments herein. Also, the various embodiments described herein are not necessarily mutually exclusive, as some embodiments can be combined with one or more other embodiments to form new embodiments. The term “or” as used herein, refers to a non-exclusive or unless otherwise indicated. The examples used herein are intended merely to facilitate an understanding of ways in which the embodiments herein can be practiced and to further enable those skilled in the art to practice the embodiments herein. Accordingly, the examples should not be construed as limiting the scope of the embodiments herein.

[0035] As is traditional in the field, embodiments may be described and illustrated in terms of blocks that carry out a described function or functions. These blocks, which may be referred to herein as units or modules or the like, are physically implemented by analog or digital circuits such as logic gates, integrated circuits, microprocessors, microcontrollers, memory circuits, passive electronic components, active electronic components, optical components, hardwired circuits, or the like, and may optionally be driven by firmware and software. The circuits may, for example, be embodied in one or more semiconductor chips, or on substrate supports such as printed circuit boards and the like. The circuits constituting a block may be implemented by dedicated hardware, or by a processor (e.g., one or more programmed microprocessors and associated circuitry), or by a combination of dedicated hardware to perform some functions of the block and a processor to perform other functions of the block. Each block of the embodiments may be physically separated into two or more interacting and discrete blocks without departing from the scope of the invention. Likewise, the blocks of the embodiments may be physically combined into more complex blocks without departing from the scope of the invention.

[0036] The accompanying drawings are used to help easily understand various technical features and it should be understood that the embodiments presented herein are not limited by the accompanying drawings. As such, the present disclosure should be construed to extend to any alterations, equivalents, and substitutes in addition to those which are particularly set out in the accompanying drawings. Although the terms first, second, etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are generally only used to distinguish one element from another.

[0037] Throughout this disclosure, the terms “device”, “edge device” and “electronic device” are used interchangeably and mean the same.

[0038] The present disclosure introduces a method for the rapid collaborative adaptation of a CSI-driven ACF signal threshold in contextual and dynamic situations. The method utilizes CSI signal-to-feature conversion and unsupervised classification entropy loss-based cluster formation on a server to facilitate this adaptation. The quick collaborative adaptation strategy is based on a comparison between the predicted ACFthreshold and a dynamic continual threshold. Significant discrepancies between these values help reinforce the dynamic continual threshold through continual learning. Furthermore, the disclosed method enables human proximity tracking near intended devices through a Intention Embedding Generation technique. This approach generates embeddings specific to the intended proximity device, stores these embeddings, and subsequently infers proximity with the intended device in new scenarios, as discussed throughout the disclosure (FIGS.2 to 11).

[0039] Referring now to the drawings, and more particularly to FIGS.2 to 11, where similar reference characters denote corresponding features consistently throughout the figures, there are shown preferred embodiments.

[0040] FIG. 2 illustrates a block diagram of an electronic device 100 for controlling at least one electronic device (not shown in FIG., e.g., “100A”, “100B”,…., “100N”) within a multi-device environment, according to an embodiment as disclosed herein.

[0041] In one or more embodiments, the electronic device 100 comprises a system 101. The system 101 may include a memory 110, a processor 120, a communicator 130, and a multi device controlling module 140. In one or more embodiments, the system 101 may be implemented on one or multiple electronic devices (not shown in FIG.).

[0042] In one or more embodiments, the memory 110 stores instructions to be executed by the processor 120 for controlling the at least one electronic device within the multi-device environment, as discussed throughout the disclosure. The memory 110 may include non-volatile storage elements. Examples of such non-volatile storage elements may include magnetic hard discs, optical discs, floppy discs, flash memories, or forms of electrically programmable memories (EPROM) or electrically erasable and programmable (EEPROM) memories. In addition, the memory 110 may, in some examples, be considered a non-transitory storage medium. The term “non-transitory” may indicate that the storage medium is not embodied in a carrier wave or a propagated signal. However, the term “non-transitory” should not be interpreted that the memory 110 is non-movable. In some examples, the memory 110 can be configured to store larger amounts of information than the memory. In certain examples, a non-transitorystorage medium may store data that can, over time, change (e.g., in Random Access Memory (RAM) or cache). The memory 110 can be an internal storage unit, or it can be an external storage unit of the electronic device 100, a cloud storage, or any other type of external storage.

[0043] In one or more embodiments, the processor 120 communicates with the memory 110, the communicator 130, and multi device controlling module 140. The processor 120 is configured to execute instructions stored in the memory 110 for controlling the at least one electronic device within the multi-device environment, as discussed throughout the disclosure. The processor 120 may include one or a plurality of processors, maybe a general-purpose processor, such as a Central Processing Unit (CPU), an Application Processor (AP), or the like, a graphics-only processing unit such as a Graphics Processing Unit (GPU), a Visual Processing Unit (VPU), and / or an Artificial Intelligence (AI) dedicated processor such as a Neural Processing Unit (NPU).

[0044] In one or more embodiments, the communicator 130 is configured for communicating internally between internal hardware components and with external devices (e.g., server) via one or more networks (e.g., radio technology). The communicator 130 includes an electronic circuit specific to a standard that enables wired or wireless communication.

[0045] In one or more embodiments, the multi device controlling module 140 is implemented by processing circuitry such as logic gates, integrated circuits, microprocessors, microcontrollers, memory circuits, passive electronic components, active electronic components, optical components, hardwired circuits, or the like, and may optionally be driven by firmware. The circuits may, for example, be embodied in one or more semiconductor chips, or on substrate supports such as printed circuit boards and the like.

[0046] In one or more embodiments, the multi device controlling module 140 may include a CSI data processing unit 141, a threshold learning unit 142, a threshold adaption unit 143, and a human proximity tracking unit 144.

[0047] In one or more embodiments, the Wi-Fi CSI data processing unit 141 is configured to receive Wireless Fidelity (Wi-Fi) Channel State Information (CSI) data from each electronic device associated with the multi-device environment, and performone or more operations, as described in conjunction with FIG.3. The Wi-Fi CSI data may include at least a CSI signal and an Autocorrelation Function (ACF) threshold.

[0048] In one or more embodiments, the threshold learning unit 142 is configured to generate a plurality of clusters of the received Wi-Fi CSI data based on an ACF threshold-based dissimilarity learning mechanism and a threshold clustering mechanism with integration of an Unsupervised Class Information Entropy (UCIE) loss, as described in conjunction with FIG.4, FIGS.5a to 5d, FIGS.6a-6b, and FIG.7.

[0049] In one or more embodiments, the threshold adaption unit 143 is configured to perform one or more operations associated with a Quick Adaptation Strategy (QAS) to control the at least one electronic device within the multi-device environment, as described in conjunction with FIG.8.

[0050] In one or more embodiments, the human proximity tracking unit 144 is configured to determine a proximity of a user within the multi-device environment based on the plurality of generated clusters and the received Wi-Fi CSI data, as described in conjunction with FIG.9a and FIG.9b. The human proximity tracking unit 144 is further configured to control the at least one electronic device within the multi- device environment based on the determined proximity of the user.

[0051] In one or more embodiments, the system 101 may include a display module (not shown in FIG.). The display module is configured to accept user inputs and is made of a Liquid Crystal Display (LCD), a Light Emitting Diode (LED), an Organic Light Emitting Diode (OLED), or another type of display. The user inputs may include but are not limited to, touch, swipe, drag, gesture, and so on.

[0052] In one or more embodiments, the system 101 may include a camera module not shown in FIG.). The camera module may include one or more image sensors (e.g., Charged Coupled Device (CCD), Complementary Metal-Oxide Semiconductor (CMOS)) to capture one or more images / image frames / video to be processed included in the image.

[0053] A function associated with the various components of the electronic device 100 may be performed through the non-volatile memory, the volatile memory, and the processor 120. One or a plurality of processors controls the processing of the input data in accordance with a predefined operating rule or AI model stored in the non-volatile memory and the volatile memory. The predefined operating rule or AI model is provided through training or learning. Here, being provided through learning means that, by applying a learning algorithm to a plurality of learning data, a predefined operating rule or AI model of the desired characteristic is made. The learning may be performed in a device itself in which AI according to an embodiment is performed, and / or may be implemented through a separate server / system. The learning algorithm is a method for training a predetermined target device (for example, a robot) using a plurality of learning data to cause, allow, or control the target device to decide or predict. Examples of learning methods include, but are not limited to, supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning.

[0054] The AI model may consist of a plurality of neural network layers. Each layer has a plurality of weight values and performs a layer operation through a calculation of a previous layer and an operation of a plurality of weights. Examples of neural networks include, but are not limited to, convolutional neural network (CNN), deep neural network (DNN), recurrent neural network (RNN), restricted Boltzmann Machine (RBM), deep belief network (DBN), bidirectional recurrent deep neural network (BRDNN), generative adversarial networks (GAN), and deep Q-networks.

[0055] Although FIG.2 shows various hardware components of the electronic device 100, but it is to be understood that other embodiments are not limited thereon. In other embodiments, the electronic device 100 may include less or more number of components. Further, the labels or names of the components are used only for illustrative purposes and do not limit the scope of the invention. One or more components can be combined to perform the same or substantially similar functions to control the at least one electronic device within the multi-device environment.

[0056] FIG.3 is a flow diagram illustrating a method 300 for receiving Wi-Fi CSI data from each electronic device associated with the multi-device environment for a database creation, according to an embodiment as disclosed herein. The method 300 may execute multiple operations for a database creation, which are given below.

[0057] At operation 301, the method 300 includes receiving the Wi-Fi CSI data from each electronic device associated with the multi-device environment during an initial ACF calibration process. The initial ACF calibration process involves a processof determining and setting an appropriate value of the ACF threshold for each electronic device based on the received Wi-Fi CSI data. The appropriate value of the ACF threshold is determined using a mathematical model. At operation 302, the method 300 includes creating the database on a server to store the received Wi-Fi CSI data and the ACF threshold.

[0058] In other words, the data collection process outlined in the previous operations is a crucial component for the subsequent blocks of the system 101 (e.g., 142, 143, 144, etc.). The data is gathered from multiple homes, utilizing various devices situated in different environmental conditions, while the presence detection solution is actively running on each device. The input for this process consists of the collected CSI data from the diverse homes and devices during the calibration of the thresholdparameter ^^ ∈ ^^(ே∗ெ)×^×்), as well as the resulting Autocorrelation Function (ACF)threshold after the calibration has been completed. a. The first step in the data collection process is the calibration process (initial ACF calibration process). When a motion detection method is activated in each home, it initiates a calibration period. This calibration period is crucial for adapting the threshold parameter in the method to the specific environmental conditions present in the environment. b. During this calibration period, raw Wi-Fi CSI information, along with the final calibrated threshold value, is transmitted from each electronic device (e.g., 100A, 100B, 100C, etc.) to an edge device (e.g., 100) connected to it, utilizing the Wi-Fi CSI solution. c. Once the data has been collected from all the electronic devices, the edge device (e.g., 100) then sends the aggregated data to the server for the purpose of building the data frame. d. Finally, the server consolidates the data collected from the various households and environments, forming a comprehensive database for the subsequent cluster module training process.

[0059] The desired output of this data collection process is the formation of a comprehensive database on the server, which will be utilized for further processing in the subsequent block of the system 101 (e.g., 142, 143, 144, etc.).

[0060] FIG.4 illustrates a block diagram of the threshold learning unit 142 of the electronic device 100 for controlling the at least one electronic device within the multi-device environment, according to an embodiment as disclosed herein.

[0061] In one or more embodiments, the threshold learning unit 142 may include an ACF threshold-based dissimilarity learning module 142a (i.e., ACF threshold-based feature extractor module), a UCIE loss-based clustering module 142b, and a centroid detection module 142c.

[0062] In one or more embodiments, the ACF threshold-based dissimilarity learning module 142a is configured to determine whether a motion detection mechanism is activated within the at least one electronic device. The ACF threshold- based dissimilarity learning module 142a is further configured to initiate a calibration period in response to determining that the motion detection mechanism is activated within the at least one electronic device. The Wi-Fi CSI data is received from each electronic device during the calibration period. The ACF threshold-based dissimilarity learning module 142a is further configured to partition the received Wi-Fi CSI data into a predetermined quantity of segments, where each segment indicates an environmental condition. The ACF threshold-based dissimilarity learning module 142a is further configured to allocate a calibrated ACF threshold value label to each segment, as described in conjunction with FIG.5a.

[0063] In one or more embodiments, the ACF threshold-based dissimilarity learning module 142a is configured to perform one or more data pre-processing operations on the received Wi-Fi CSI data, as described in conjunction with FIG.5b. The one or more data pre-processing operations comprise an amplitude extraction, a phase extraction, a phase sanitization, an outlier removal, and a detrending.

[0064] In one or more embodiments, the ACF threshold-based dissimilarity learning module 142a is configured to transform the received Wi-Fi CSI data into a feature space using a dissimilarity loss to enable a formation of the plurality of clusters that accurately represent one or more relationships among the received Wi-Fi CSI data with corresponding ACF threshold, as described in conjunction with FIGS.5c-5d. The received Wi-Fi CSI data with similar ACF thresholds are positioned in a close proximity within the feature space. The dissimilarity loss indicates an ability of dissimilaritylearning to identify at least one of differences and similarities between the received Wi- Fi CSI data based on the corresponding ACF threshold.

[0065] To identify the at least one of the differences and similarities, the ACF threshold-based dissimilarity learning module 142a is configured to perform one or more operations, which are given below. a. Select a set of anchor samples from training data; b. Identify one or more positive samples for each anchor sample based on a first absolute difference. The first absolute difference between a first anchor sample of ACF threshold (A_i^th) and a second anchor sample of ACF threshold (A_j^th) is less than a predefined threshold (γ); and c. Identify one or more negative samples for each anchor sample based on a second absolute difference. The second absolute difference between the first anchor sample of the ACF threshold (A_i^th) and the second anchor sample of the ACF threshold (A_j^th) is higher than the predefined threshold (γ).

[0066] In one or more embodiments, the UCIE loss-based clustering module 142b is configured to convert the Wi-Fi CSI data into a feature space using an ACF threshold-based feature extractor module. The UCIE loss-based clustering module 142b is further configured to pass the converted CSI data through a deep learning based module to extract one or more cluster probability distributions. The UCIE loss-based clustering module 142b is further configured to train a clustering module based on the one or more extracted cluster probabilities using the UCIE loss. The UCIE loss-based clustering module 142b is further configured to utilize the trained clustering module for inference on new CSI data to assign at least one cluster based on the training, as described in conjunction with FIGS.6a-6b.

[0067] In one or more embodiments, the centroid detection module 142c is configured to determine a centroid of each of the plurality of generated clusters, as described in conjunction with FIG.7. The centroid is determined by averaging one or more ACF threshold values of all Wi-Fi CSI data within a corresponding cluster. The centroid is subsequently employed during an inference stage to swiftly initialize the ACF threshold.

[0068] FIGS.5a to 5d illustrate one or more operations associated with the ACF threshold-based dissimilarity learning module 142a of the threshold learning unit 142, according to an embodiment as disclosed herein.

[0069] Referring to FIG. 5a: FIG. 5a illustrates an exemplary process 501 of collecting and preparing Wi-Fi CSI data for the subsequent application of threshold- based dissimilarity learning. This process aims to leverage the calibration data from the motion detection method to construct the feature space that enables clustering based on ACF thresholds. The data collection and calibration process involves the following operations: a. Calibration process: When the motion detection method is activated, the calibration period spanning T minutes is initiated. During this time, raw Wi- Fi CSI data is gathered from various households, along with the corresponding calibrated threshold values (final calibrated threshold value Aj). b. Temporal chunking of data: To ensure compatibility with the subsequent inference stages, the entire T-minute duration of CSI data is partitioned into smaller T' chunks. Each chunk represents a snapshot of the environmental conditions and is used as input for feature extraction and clustering. c. Labeling with calibrated ACF thresholds: A structured data frame is created, where each temporal chunk is assigned the label of the calibrated ACF threshold value. This labeled data serves as the basis for the threshold-based dissimilarity learning.

[0070] This process leverages the calibration data from the motion detection method to construct a feature space that facilitates clustering based on the ACF thresholds, which can be valuable for various applications, such as indoor localization, activity recognition, and smart home automation.

[0071] Referring to FIG. 5b: FIG. 5b illustrates the one or more data pre- processing operations 502, which are given below. a. Amplitude extraction: CSI amplitude variations in a time domain have been extracted as it is found that they are different for different humans, activities,gestures, etc., which can be used for various tasks. Given a complex CSI value z= ^^ + ^^^^. Amplitude can be extracted by below-mentioned equation.A = √aଶ + bଶ (1)b. Phase extraction: The phase of the CSI signal can provide valuable information about channel characteristics, such as the delay spread, and Doppler shift with proper processing. For a complex CSI value ^^, it can be extracted using the below-mentioned equation.c. Phase sanitization: The phase information cannot be directly used as it contains random noise due to unsynchronized timing between the transmitter and receiver. The aim of the phase sanitization is to mitigate the effects of phase noise and the random phase offsets by implementing a linear transformation, using the below-mentioned equation.Where ^ఋ^ is frequency difference of adjacent subcarriers, ^^(^^, ^^, ^^) is an unwrapped phase of ^^௧^ subcarrier transmitted from ^^௧^transmitter and received at ^^௧^ receiver. ^^, ^^ are slop and offsetrespectively. The calibrated phase can be extracted by below below- mentioned equation. ^^ᇱ(^^, ^^, ^^) = ^^(^^, ^^, ^^) − 2^^^ఋ^(^^ − 1)^^ (4)d. Outlier removal: Outliers can have a significant impact on the accuracy of the features extracted and can cause significant errors in the estimation of channel frequency response, leading to degradation of the system performance. Hampel filter, for example, is used to substitute outliers. For each window, the filter replaces values that surpass a certain threshold based on median absolute deviation. e. Detrending: To remove the long-term trends in the data caused by various factors such as signal reflections, interference, and hardware imperfections. Linear De- trending is used for this purpose. It calculates a least-square linear regression of the data and subtracts the resulting line from the data to de-trend it.

[0072] Referring to FIG. 5c: FIG. 5c illustrates feature extraction through dissimilarity learning process 503, a technical aspect of training a robust feature extractor using dissimilarity loss within the context of ACF threshold-based clustering. The core objective is to position data points with similar ACF thresholds in close proximity within the feature space, thereby enhancing the subsequent clustering performance, as described below. In other words, the process involves selecting anchor, positive, and negative samples from a batch of training data to ensure that the feature extractor positions data points with similar ACF thresholds in close proximity within the feature space. a. Sample selection: The initial step involves the meticulous selection of anchor, positive, and negative samples from a given batch of training data. For a target ACF threshold ^^௧^^, the positive samples are defined by the condition |^^௧^^- ^^^௧^| < γ, where γ represents the predefined threshold. Conversely, samples that fail to meet these criteria are categorized as negative samples. b. Deep learning architecture: A state-of-the-art deep learning architecture is employed to process the selected batch of data and facilitate training in a dissimilarity learning manner. This sophisticated model is designed to effectively capture the intricate patterns and dependencies within the Wi-Fi CSI data, thereby enhancing the feature extraction process. c. Optimization with NTXentLoss: The training process incorporates the NTXentLoss function, which is a variant of the multi-class N-pair loss with the addition of a temperature (T) parameter. NTXentLoss is known for its unique characteristics, such as scale sensitivity, enhanced discrimination, and improved stability. The mathematical representation of NTXentLoss is as follows, ^^ supervised ^^୮൫^୧୫൫௭^,௭ೕ൯ / ఛ൯ ே்ି Xent∑మೖಿసభ^“sim(. )” denotes the cosine similarity function.^ -^^^^^^ ^^^are the encoded features of samples ^^^^^^^^^ ^^^respectively.^ - 2^^௬^ − 1 denotes the set of indices of all positives in themultiview batch distinct from ^^.

[0073] Referring to FIG. 5d: FIG. 5c illustrates a justification for ACF threshold-based dissimilarity learning (504 (a and b) and 505 (c and d)). The adoption of threshold-based dissimilarity learning within the research framework is substantiated by the following technical considerations. a. Inherent separability limitations in raw CSI Data: The raw CSI data may inherently lack the necessary separability, compelling the application of pre- processing techniques to facilitate effective cluster formation. The unprocessed CSI data can exhibit complex patterns and interdependencies, making it challenging to identify meaningful clusters solely based on the raw CSI values. b. Potential clustering artifacts: Even if clustering methods were applied directly to the raw CSI data, there is a likelihood that the resulting clusters would be based on alternate features rather than the desired ACF thresholds. This unintended clustering outcome undermines the goal of accurately representing the relationship between data points in terms of their ACF thresholds. c. Requirement for feature space transformation: Consequently, there arises a need to transform the raw CSI information into the feature space that exhibits enhanced separability based on the final calibrated ACF thresholds. By leveraging ACF threshold-based dissimilarity learning, the framework aims to extract relevant features from the raw CSI data, ensuring that data points with similar ACF thresholds are positioned in close proximity within the feature space. d. Enabling effective clustering: This conversion process enables subsequent clustering methods to effectively identify and differentiate clusters based on the desired ACF thresholds. The transformed feature space accurately reflects the relationships among the Wi-Fi CSI data points, allowing for the formation of clusters that closely align with the underlying ACF threshold characteristics.

[0074] FIGS.6a-6b illustrate one or more operations associated with the UCIE loss-based clustering module 142b of the threshold learning unit 142, according to an embodiment as disclosed herein.

[0075] Referring to FIG.6a: The objective of this UCIE loss-based clustering module 142b is to leverage the effectiveness of deep learning techniques 601 in an unsupervised learning paradigm to create distinct clusters of Wi-Fi CSI data based on their ACF thresholds. This is achieved by utilizing a loss function called UCIE loss, as described in conjunction with FIG.6b.

[0076] The input to this UCIE loss-based clustering module 142b is a set of ACFthreshold-based features, denoted as ^^ ∈ ^^௭బ, extracted from the previous processingoperations. The output of this UCIE loss-based clustering module 142b is a set of ^^ clusters generated by the deep learning-based clustering module, along with their precomputed centers of ACF thresholds. The key components of this UCIE loss-based clustering module 142b are as follows:

[0077] Unsupervised training with UCIE loss: This section (which may relate to FIG.6b) focuses on the unsupervised training of the deep learning module for ACF threshold-based clustering. The proposed UCIE Loss function is employed to guide the training process, ensuring the formation of distinct clusters based on the ACF thresholds.

[0078] Centroid pre-computation of clusters: This section (which may relate to FIG. 7) involves the precomputation of the ACF thresholds for each cluster. These precomputed cluster centers may be utilized in the inference stage to enable a quick initialization of the ACF thresholds, thereby enhancing the overall efficiency of the system 101.

[0079] Referring to FIG.6b: FIG.6b illustrates an unsupervised training with UCIE loss. The training of the UCIE loss-based clustering module 142b consists of two key components: the feature extraction layer and the classification head. a. Feature extraction layer provides a mapping from Raw WIFI CSI space to ACF threshold-based feature embedded space ^^ఏభ : ^^^^.b. After this, the classification head provides a mapping of feature space i.e. ^^ఏమ : ^^ → ^^.

[0080] The UCIE loss-based clustering module 142b utilizes a deep learning module trained in an unsupervised learning paradigm, with the UCIE loss as the guiding objective. It is represented as follows:where, ^ ^^ : Final UCIE (Unsupervised class Information Entropy) Loss; ^ ^^^: Classification Entropy; ^ ^^ஏ: Class Entropy; ^ ^^^: Interclass Distance; ^ ^^^^: Last Layer Features; ^ ^^ : Kronecker delta function; ^ ^^^^: Probability output of ^^௧^sample belonging to the ^^௧^class; ^ ^^ : Total Number of Samples in the Batch.

[0081] The UCIE loss-based clustering module 142b is trained in the unsupervised learning paradigm, guided by a loss function called the UCIE loss. The UCIE loss comprises three crucial components: a. Classification entropy (^^^): The classification entropy, ^^^, is designed to capture the network’s ability to confidently classify data points. A well-trained network may have low entropy values, indicating accurate classification of the data into the respective classes; b. Class entropy (^^ஏ): To prevent the formation of a single dominant cluster, the class entropy, ^^ஏ, is introduced. This term promotes diversity within the clustering, ensuring the emergence of multiple distinct clusters during the training process.c. Inter-class distance regularization ( ^^^ ): To encourage a larger inter-classdistance, a regularization term, ^^^, is added to the loss function. This term penalizes samples from different classes for being too close to each other, further enhancing the separation between the formed clusters.

[0082] The effectiveness of the classification performance can be assessed by examining the probability distribution of the predicted class labels. A well-trained network may exhibit a probability distribution where the probability of the correct class is near unity, indicating a confident and accurate classification 602.

[0083] This principle is captured in the classification entropy (^^^) term of the UCIE loss function. ^^^is designed to measure the network's ability to confidently classify the data points, with the goal of minimizing the entropy of the class probability distribution. Low entropy values in H_Φ suggest that the network is able to classify the data points with high confidence, which is a desirable characteristic for effective clustering. However, relying solely on ^^^may result in a skewed partitioning of the data, where all samples are assigned to a single cluster with high probability. To address this potential issue, the UCIE loss introduces the class entropy (^^ஏ) term.promotes diversity within the clustering by encouraging the formation of multiple distinct clusters during the training process. By maximizingthe network is incentivized to create a more balanced and evenly distributed clustering of the data.

[0084] Additionally, the UCIE loss incorporates an inter-class distance regularization (^^^) term. This term encourages larger distances between the clusters, ensuring better separation and enhancing the overall quality of the clustering. By penalizing samples from different classes that are too close to each other, ^^^helps to maintain well-defined boundaries between the formed clusters. The simultaneous minimization of the Classification Entropy (^^^) and maximization of the Class Entropy (^^ஏ) within the UCIE loss function enables effective unsupervised training for the clustering task. This multi-faceted approach leverages the strengths of both entropy- based metrics to achieve a robust and well-structured clustering of the data.

[0085] FIG. 7 illustrates one or more operations associated with the centroid detection module 142c of the threshold learning unit 142, according to an embodiment as disclosed herein.

[0086] In one or more embodiments, the centroid detection module 142c is configured to perform a precomputation of ACF thresholds for each cluster for later to be used in the inference stage for quick ACF threshold initialization 701. Following the formation of clusters, a centroid calculation process is applied to each cluster. The centroid is obtained by averaging the ACF threshold values of all data points belonging to the respective cluster, by below below-mentioned equation.

[0087] These computed centroids serve a critical role in the inference stage, enabling quick initialization of ACF thresholds for new data points based on their proximity to the cluster centroids. This efficient centroid-based initialization process enhances the overall performance of the pipeline / system 101.

[0088] FIG.8a and FIG.8b illustrate one or more operations (801, 802, and 803) associated with the threshold adaption unit 143 of the electronic device 100 for applying a Quick Adaptation Strategy (QAS), according to an embodiment as disclosed herein.

[0089] In one or more embodiments, the primary objective of the threshold adaption unit 143 is to employ the QAS to rapidly initialize the ACF threshold by utilizing a small chunk of raw CSI data (e.g., input: “ T small Raw CSI data chunk that serves as a snapshot of the environmental conditions”). This data serves as a snapshot of the prevailing environmental conditions, which may relate to FIG.5a. The threshold adaption unit 143 is configured to evaluate the performance of the QAS approach and subsequently enhance its efficiency, accuracy, and robustness, by performing one or more operations (801, 802, and 803).

[0090] Quick adaptation strategy for ACF threshold detection 801: The quick adaptation strategy for ACF threshold detection involves leveraging a feature extractor and a Deep Learning clustering model to quickly determine the ACF threshold without executing the full calibration process. Depending on the confidence level of the Deep Learning clustering model, the process either proceeds to the usage and evaluate stage or reverts to the original ACF threshold block.

[0091] Input: T small Raw CSI data chunk that serves as a snapshot of the environmental conditions; and

[0092] Output: Selected path determined by the QAS.

[0093] Usage and evaluate stage 802: The usage and evaluate stage represents the initial deployment phase of this strategy. In this mode, the module assesses the accuracy and reliability of the QAS by comparing its results with those obtained from the conventional calibration process. This evaluation provides the necessary Key Performance Indicator (KPI) data to guide the improvement of the strategy until it meets the required accuracy standards, at which point it can be used in the usage mode exclusively.

[0094] Continuous learning 803: The final phase focuses on continuously enhancing the performance of the QAS. By leveraging the insights gained from the evaluation, the objective is to improve the efficiency, accuracy, and robustness of the QAS approach for ACF threshold detection. To prevent catastrophic forgetting, a combination of the Elastic Weight Consolidation (EWC) strategy and experience replay-based rehearsal strategy is employed. i. Input: The collected CSI data from various homes / devices during the calibration and evaluation of the QAS-based threshold (^^ ∈ ^^and the resulting ACF threshold after calibration ^^௧^^. ii. Output: improved ACF Threshold based dissimilarity feature extractor and UCIE deep learning clustering module.

[0095] In the context of QAS for ACF threshold detection 801, this 801 operation may further include one or more operations associated with a QAS-based threshold calculation and dynamic continual threshold comparator.

[0096] For QAS-based threshold calculation: In this operation, raw CSI data chunks are passed to the Feature Extraction module and the Deep Learning-based Clustering module to predict the ACF threshold. The feature extraction module preprocesses the data chunks to prepare them for subsequent analysis and prediction. The preprocessed data chunks are then passed through a pre-trained Feature Extractor, which converts the raw CSI data into an ACF-threshold-based feature space. The feature representations from the previous operation are fed into the UCIE-based deep learning-based clustering model. This model assigns each data point to a specific cluster and provides a confidence score indicating the likelihood of correct clustering.

[0097] For dynamic continual threshold comparator: To avoid potential nuisances caused by inaccurate ACF thresholds and to prioritize a positive user experience, a threshold check mechanism is implemented. This mechanism ensures that only confident and highly likely accurate ACF thresholds predicted by the QAS are used, preventing unnecessary disruptions. In other words, nuisance mitigation and end- user considerations, given the deployment of the method in households, it is essential to avoid potential nuisances caused by inaccurate ACF thresholds. To address this concern, a threshold check mechanism is implemented, ensuring that only confident and highly likely accurate ACF thresholds predicted by the QAS are used. This consideration prioritizes a positive user experience and prevents unnecessary disruptions.

[0098] Confidence-based cluster assignment: The maximum confidence score among the clusters is determined to identify the most likely cluster assignment for each data point. If the confidence score surpasses a dynamic continual threshold (μ), as a below-mentioned equation, the data point is passed to the usage / evaluation block for further processing. Otherwise, to ensure the user’s convenience, data points with confidence scores below the threshold are redirected to the original ACF calibration stage to ensure accurate threshold estimation.

[0099] Here, ^^ is the starting threshold (i.e., threshold at deployment) and ^^ is the increment factor that decides the final threshold. E.g., Could be ^^ as 0.8 and ^^ as 0.15. t is the continual learning training episode number. It basically represents how many times the model is updated in the future in the continual learning stage. Condition for using predicted ACF threshold for given ^^:• ^^ఏభand ^^ఏమare our feature Extractor and Deep learning-based clustering module.

[0100] If the confidence score is below the threshold, then the original ACF calibration method is used, and data is collected in the same manner as the 1ststep for usage in a continuous learning stage.

[0101] In the context of usage and evaluate stage 802: this operation 802 may further include one or more operations, which are given below. a. Mode selection and deployment: This operation begins with the selection and deployment of the appropriate mode. Initially, when the model is deployed for the first time, it operates in the Usage and Evaluation mode to assess the performance of the Quick Adaptation Strategy (QAS). This evaluation phase is essential for continuous learning and further enhancement of the models. b. Usage and evaluation mode: In the usage and evaluation mode, the ACF threshold value predicted by the QAS is utilized in the motion detection method. While the motion method is in operation using the QAS- predicted ACF threshold, the original calibration process runs simultaneously in the background for performance evaluation. If the mode is Usage only, the original calibration process will not run, and the motion method will simply utilize the QAS-predicted motion threshold. c. Calibration comparison: Upon completion of the original calibration process, a comparison is made between the ACF threshold predicted by the QAS and the original ACF calibration threshold. The absolute difference between these two thresholds is calculated and evaluated against a predefined threshold value, denoted as γ. ห^^௧ொ^^ௌ − ^^ ௧^^^^^^^^^ ห < ^^ (10)d. Evaluation and decision: If the absolute difference between the predicted threshold and the original calibration threshold falls within the predetermined threshold (γ), it indicates that the solution is functioning satisfactorily. In this case, the ACF threshold obtained from the QAS continues to be utilized in the motion detection method.

[0102] However, if the absolute difference exceeds the predefined threshold, it suggests a discrepancy between the predicted and calibrated thresholds. The ACF threshold utilized in the motion method is then replaced with the threshold calibrated using the original strategy, and the results and data are transmitted to the server for future use in continuous training.

[0103] In the context of continuous learning 803: this operation 803 may further include one or more operations, which are given below.

[0104] Fine-tuning using Elastic Weights Consolidation (EWC) strategy: This operation presents the detailed steps involved in data consolidation and model fine- tuning for continuous learning in Auto-Correlation Function (ACF) threshold detection. The data is consolidated and the models undergo fine-tuning using the Elastic Weights Consolidation (EWC) strategy, commonly employed in continuous learning to preserve prior knowledge and prevent catastrophic forgetting. a. Data consolidation: In the first operation, data collected from various households is consolidated to incorporate cases where the QAS fails to meet the confidence Dynamic Continual threshold, as well as instances where significant discrepancies occur between predicted and original calibration thresholds in the Usage / Usage & evaluate block. This consolidation step ensures a comprehensive dataset for subsequent analysis and training. b. Pre-processing: The consolidated data undergoes the same pre-processing used in the threshold-based dissimilarity Learning block to prepare data for fine-tuning.

[0105] Model Fine-tuning with EWC: The models employed in the ACF threshold detection system undergo fine-tuning using the Elastic Weights Consolidation (EWC) strategy. EWC is a well-established technique in continuous learning for deep learning-based models. It allows for the preservation of prior knowledge while adapting the models to new environments and data distributions. By fine-tuning the models using EWC, the goal is to enhance their accuracy, robustness, and adaptability in a continuous learning setting. The final Loss used in fine-tuning is given below:^ ^^(^^): EWC enhanced Final Loss; ^ ^^^(^^) : Loss calculated on the new data; ^ ^^ : denotes each index in the weight vector; ^ ^^ : Model Parameters; ^ ^^^∗,^: Old Model Parameters for ^^௧^index; ^ λ: Control parameter; and ^ ^^ : Training data.

[0106] Integration of rehearsal technique for catastrophic forgetting mitigation: To further mitigate the risk of catastrophic forgetting, the rehearsal technique is incorporated to enhance the models’ performance and preserve knowledge from previous training stages. The process involves two key steps: data sampling strategy and model fine-tuning, which are given below.

[0107] Data sampling strategy: In the first step, a data sampling strategy is employed to select relevant data samples that effectively capture the uncertainty and representativeness of the training dataset. This strategy involves utilizing uncertainty sampling and representative sampling techniques to ensure a well-balanced and diverse dataset for training. a. Uncertainty sampling: it focuses on selecting samples that are most uncertain or challenging to the current model. Monte Carlo Dropout or Bayesian inference to estimate model uncertainty. Data points with higher uncertainty scores are chosen for rehearsal in this. b. Representative sampling: this approach selects data that represents the distribution of the previously seen data. Methods like k-means or density-based methods like Gaussian Mixture models can be employed.c. Model Fine-tuning: The sampled data is then utilized for model fine-tuning. The models, which have undergone previous training stages, are further trained on the sampled data to enhance their performance and adaptability to new scenarios.

[0108] FIG.9a and FIG.9b are a flow diagram illustrating one or more operations (901 and 902) associated with the human proximity tracking unit 144 of the electronic device 100 for determining the proximity of the user within the multi-device environment, according to an embodiment as disclosed herein.

[0109] In one or more embodiments, a proximity detection module (not shown in FIG.) associated with the human proximity tracking unit 144 employs deep learning techniques to predict the proximity of users to various devices within the multi-device environment (e.g., home). The human proximity tracking unit 144 is configured to collect data from different electronic devices (e.g., 100A, 100B,.., 100N), data pre- processing to enhance data quality, and the application of a deep learning method for proximity prediction.

[0110] The input to the proximity detection module is a small raw CSI data chunk, which serves as a snapshot of the environmental conditions. The output of the module is the proximity detection of the user near the device.

[0111] In one or more embodiments, the human proximity tracking unit 144 may execute one or more operations to determine the proximity of the user within the multi- device environment, which are given below. a. Data collection operation involves gathering Wi-Fi CSI data from multiple electronic devices (e.g., 100A, 100B,.., 100N) in the home environment. This data captures the signal characteristics and variations associated with the user’s proximity to each electronic device 100. b. The collected WIFI CSI data undergoes pre-processing operations, such as data cleaning, noise removal, signal normalization, and feature extraction, to ensure its suitability for subsequent analysis. These pre-processing operations aim to enhance the quality and relevance of the data, enabling effective utilization in the proximity detection task.c. The pre-processed WIFI CSI data is then passed through a deep learning method specifically designed for proximity prediction. The deep learning model leverages the inherent patterns and relationships present in the data to learn and predict the proximity of the user to each device. Through a process of training and optimization, the model is able to capture complex features and make accurate proximity predictions. The specific deep learning model used in this implementation can be easily generalized to any architecture that generates embeddings based on proximity.

[0112] By leveraging the collected Wi-Fi CSI data and the power of deep learning, the proximity detection module is configured to accurately identify the proximity of the users to at least one electronic device 100, facilitating various applications and enhancing user experience.

[0113] In one or more embodiments, the human proximity tracking unit 144 may execute one or more operations 901 to determine the proximity of the user within the multi-device environment, which are given below. a. Proximity embedding generation; b. Intention embedding formation; and c. Transitioning from collection to inference mode.

[0114] In one or more embodiments, the primary objective of this human proximity tracking unit 144 is to define a methodology to reduce false positive scenarios in proximity detection-based activation solutions. To achieve this by incorporating an intention detection mechanism that can discern whether an individual’s proximity is indicative of their intention to use the device or if they are merely present due to unrelated factors or using an alternative device.

[0115] The input to this human proximity tracking unit 144 is T small raw CSI data chunk from multiple devices in the home connected to the edge device (e.g., electronic device 100) that serves as a snapshot of the environmental conditions in the proximity of the device. The output to this human proximity tracking unit 144 is the intention of embedding data cluster stored in the edge device, by performing one or more operations, as described below. I∈R^(〖N×z〗_0 ). Here N is no. of electronicdevices connected to the edge device and Z_o is the size of feature embeddings in the proximity detection solution.

[0116] In the context of the proximity embedding generation, the proximity detection module generates proximity embedding vectors for each electronic device (e.g., 100A, 100B,…,100N) in the home environment. These embeddings capture the proximity information learned from the WIFI CSI data and represent the spatial relationships between the user and each electronic device (e.g., 100A, 100B,…,100N), as illustrated in FIG.9a and FIG.9b. The proximity detection method extracts relevant features from the WIFI CSI data and transforms them into dense embedding vectors, encoding the proximity information. a. Proximity detection: Each electronic device (e.g., 100A, 100B,…,100N) in the home environment is equipped with a proximity detection solution to determine if the user is in the proximity of the electronic device (e.g., 100A) or not. This proximity detection mechanism leverages the generated proximity embeddings to make accurate determinations about the user's spatial relationship to the electronic device (e.g., 100A). b. Triggering the intention embedding generation mechanism: The user triggers the Intention Embedding Generation (IEG) mechanism by actively using the electronic device (e.g., 100A), such as pressing a button on a remote or opening a fridge. This user interaction serves as the trigger point for the IEG mechanism to initiate further processing. c. Extraction of proximity embeddings and ground truth: Once a trigger point is identified, the proximity embeddings are extracted to capture the spatial information associated with the user’s actions. Simultaneously, the ground truth information regarding the electronic device (e.g., 100A) being operated by the user is recorded. Both the proximity embeddings and the corresponding ground truth are prepared for transmission to the edge server and / or electronic device 100. d. Transmission to the edge server (e.g., electronic device 100): The proximity embeddings, along with the respective ground truth information, are transmitted to the designated edge server. The edge server architecture isresponsible for receiving and processing the data in real-time, leveraging the proximity information to enable various applications and enhance the user experience.

[0117] In the context of the intention embedding formation, to preserve the relative proximity information among the electronic devices (e.g., 100A, 100B,…,100N) while the intended device is being operated, the human proximity tracking unit 144 converts the received proximity vectors into a single intention embedding vector. This conversion is achieved by summing the ACF accentuated proximity embedding vectors of all the proximity vectors together. By merging the proximity information from multiple devices, a comprehensive representation is created that encapsulates the spatial relationships between the intended device and other electronic devices (e.g., 100A, 100B,…,100N) in the environment.

[0118] After generating the intention embedding vector, the human proximity tracking unit 144 stores it in the edge device (e.g., electronic device 100) for future utilization during the inference stage. This vector, accompanied by the corresponding device label, provides a compact yet informative representation of the user's intended device and the proximity relationships among the devices. By storing this information locally on the edge device (e.g., electronic device 100), efficient and timely access to the intention embeddings is enabled during real-time inference processes, contributing to enhanced decision-making and contextual understanding,^^ = ∑^ ^ୀ ୀ^ ^^^^^^^(14) Where, - ^^ : Intention vector; - ^^^: Proximity Embedding vector for ithdevice; and - ^^^^^^^: ACF value of ithdevice.

[0119] In the context of the transitioning from collection to inference mode, the transition comprises two operations, which are given below.

[0120] Collection operation: In the initial stage, the electronic devices (e.g., 100A, 100B,…,100N) operate in the collection stage, where intention vectors arecontinuously gathered and stored. These vectors represent the spatial information captured by the proximity detection solutions integrated into the devices. Throughout this stage, the electronic devices (e.g., 100A, 100B,…,100N) actively record and accumulate the proximity vectors associated with their respective intended devices.

[0121] Determining data sufficiency and transition to inference mode: as the collection operation progresses, the human proximity tracking unit 144 monitors the number of proximity vectors obtained for each device connected to the edge device (e.g., 100A). Once the count of proximity vectors exceeds the Inference Sample Threshold (^) for each device individually, it is inferred that a sufficient number of data points have been gathered to create a reliable representation. This threshold ensures that an adequate dataset is available to accurately capture the intentions behind device usage. Consequently, the collection mode ceases, and the system transitions to the inference mode. The system checks the following condition for each device connected to the edge device: ^^^^^^^^^^(^^^) > ^^ (15)Where Count(^^^) represents the number of proximity vectors collected for device I; ^ is the defined Inference sample threshold.

[0122] Once this condition is met for all connected devices, the human proximity tracking unit 144 transitions from the collection stage to the inference mode, utilizing the accumulated data to make accurate inferences about user intentions and device interactions

[0123] In one or more embodiments, the human proximity tracking unit 144 may execute one or more operations 902 to determine the proximity of the user within the multi-device environment, which are given below. The input to this human proximity tracking unit 144 is T’ small Raw CSI data chunk from multiple devices in the home connected to the edge device that serves as a snapshot of the environmental conditions in the proximity of the device. The output to this human proximity tracking unit 144 is the intended device the user wants to control among the devices in proximity. a. Proximity embedding transmission: Whenever the user’s proximity is detected near any device within the environment, the corresponding devicetransmits its proximity embedding vector to the edge server. This process is repeated for all devices in the network. The transmitted proximity embeddings are then used at the edge server to form a consolidated intention embedding. b. Nearest neighbour search: Utilizing the consolidated dataset of intention proximity embeddings in the previous sub-block, a nearest neighbour search is performed when a user's proximity is detected. The K-nearest neighbours’ approach is employed to identify the n closest intention proximity embeddings based on cosine similarity to the user's current proximity pattern. The selected embeddings represent potential intention candidates, based on the below- mentioned equation. Cosine similarity =c. Ground truth extraction and voting: Following the identification of the n closest intention proximity embeddings, the associated ground truths (device usage information) are extracted from the stored database on the edge device. A max voting mechanism is employed to determine the intended usage device by aggregating the ground truth labels of the identified embeddings. The device with the highest vote count is considered the predicted intended usage device. d. Enabling collection stage for performance improvement: In scenarios where the user expresses dissatisfaction with the performance of the solution or has repositioned the devices, resulting in incorrect results, an option is provided to re-enable the collection stage. By enabling the collection stage, the user can increase the threshold limit “n” for each intention vector required from the devices during the inference stage. This increased limit allows for gathering a more extensive set of proximity vectors, enhancing the system's performance and accuracy. The flexibility to modify the collection threshold empowers the user to refine the system 101 according to their evolving needs and environmental changes.

[0124] FIGS. 10a to 10C illustrate example scenarios associated with the disclosed method for controlling the at least one electronic device within the multi- device environment, according to an embodiment as disclosed herein.

[0125] FIG.10a illustrates an energy saving based on close proximity detection using Wi-Fi CSI. In the first example scenario, consider a situation 1001 where no one is present in the room, and the family hub screen is turned off. After some time 1002, the system 101 detects user presence at a far proximity using the disclosed proximity detection method based on Wi-Fi CSI. However, since the user is not in close proximity to the Family Hub, the screen remains turned off, resulting in energy savings. Subsequently 1003, as the user comes into close proximity of the family hub, the system 101 detects the user’s presence and turns on the family hub screen. This demonstrates the effectiveness of the disclosed method in providing energy-saving capabilities by leveraging close proximity detection using Wi-Fi CSI data.

[0126] FIG.10b illustrates a distance view alert for Kids, proximity detection using Wi-Fi CSI. In the second example scenario, consider a situation 1004 where the Kids Mode is enabled on the smart TV, and content is playing. The disclosed proximity detection method based on Wi-Fi CSI determines that the user is in near proximity to the TV. Upon detecting the user’s close proximity 1005, the system 101 automatically turns off the TV screen and delivers a warning message, “Alert! You are too close to TV”. This feature aims to provide a safety mechanism for children, ensuring a suitable viewing distance from the TV by leveraging the proximity detection capabilities of the disclosed method.

[0127] FIG. 10C illustrates a seamless content continuation based on user proximity. In the third example scenario, consider the following preconditions: (a) User Context: single user at home, the user is watching TV Plus content on a TV 100a in Room 1, and other sensing electronic devices (Family Hub 100b and TV 100c) are present nearby; and (b) Device Context: The TV 100a in Room 1 is turned on. a. The user moves to the room 2 with another TV 1005: When the user moves from Room 1, the system 101 detects the absence of the user and pauses the content playing on the TV 100a in Room 1. The system 101 detects the user’s presence in proximity to the family hub 100b in room 2 and starts playing the content on the family hub 100b. b. The user moves to Room 2 with another family hub 1007: when the user moves from Room 1, the system 101 detects the absence of the user and pauses thecontent playing on the TV 100a in Room 1. The system 101 detects the user’s presence in proximity to the TV 100c in room 2 and starts playing the content on the TV 100c in room 2.

[0128] This demonstrates the seamless content continuation capability of the disclosed method, where it can accurately track the user’s location and proximity to different electronic devices, enabling a continuous and personalized user experience as the user moves between rooms.

[0129] FIG.11 is a flow diagram illustrating a method 1100 for controlling the at least one electronic device within the multi-device environment, according to an embodiment as disclosed herein.

[0130] At operation 1101, the method 1100 includes receiving Wi-Fi CSI data from each electronic device associated with the multi-device environment. The Wi-Fi CSI data may include at least the CSI signal and the ACF threshold. At operation 1102, the method 1100 includes generating the plurality of clusters of the received Wi-Fi CSI data based on the ACF threshold-based dissimilarity learning mechanism and the threshold clustering mechanism with integration of the UCIE loss. At operation 1103, the method 1100 includes determining the proximity of the user within the multi-device environment based on the plurality of generated clusters and the received Wi-Fi CSI data. At operation 1104, the method 1100 includes controlling the at least one electronic device within the multi-device environment based on the determined proximity of the user. Further, a detailed description related to the various steps of FIG.11 is covered in the description related to FIG.2 to FIG.10C, and is omitted herein for the sake of brevity.

[0131] The disclosed method has several advantages over the existing method(s), for example, which are stated below. a. Cost-effective approach: The proposed method does not require additional hardware, making it a costless technology to implement. This is particularly advantageous for scenarios with high demand for quick adaptation to the home environment, such as home monitoring, energy saving, and light automation.b. Flexibility in deployment: The method does not rely on line-of-sight requirements, enabling it to be seamlessly integrated into various multi-device environments. c. Enhanced Multi-Device Experience (MDE): By leveraging the Wi-Fi CSI data, the method can effectively identify the presence of users in both close and far proximity to the devices, enabling the delivery of personalized MDE services. d. Privacy preservation: compared to camera-based technologies, the use of Wi- Fi CSI data does not raise privacy concerns, as it does not involve capturing visual information about the users. e. Comprehensive data utilization: The method incorporates both the CSI signal and the ACF threshold from the received Wi-Fi CSI data. This holistic data input allows for a more accurate and reliable assessment of the user’s location and movements. f. Advanced clustering mechanism: The method employs a robust clustering method that combines an ACF threshold-based dissimilarity learning mechanism and a threshold clustering mechanism with the integration of the UCIE loss. This advanced clustering approach enables precise grouping of the Wi-Fi CSI data, leading to improved user proximity detection. g. Seamless device control: By accurately determining the user’s proximity within the multi-device environment, the method can intelligently control the associated electronic devices. This allows for a more intuitive and responsive user experience, where the devices adapt to the user’s location and interactions. h. Scalability and flexibility: The Wi-Fi CSI-based approach inherently supports a multi-device environment, making the method scalable and flexible. It can be applied to various smart home or office settings with multiple interconnected electronic devices.

[0132] The various actions, acts, blocks, steps, or the like in the flow / sequence diagrams may be performed in the order presented, in a different order, or simultaneously. Further, in some embodiments, some of the actions, acts, blocks, steps, or the like may be omitted, added, modified, skipped, or the like without departing from the scope of the invention.

[0133] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one ordinary skilled in the art to which this invention belongs. The system, methods, and examples provided herein are illustrative only and not intended to be limiting.

[0134] While specific language has been used to describe the present subject matter, any limitations arising on account thereto, are not intended. As would be apparent to a person in the art, various working modifications may be made to the method to implement the inventive concept as taught herein. The drawings and the forgoing description give examples of embodiments. Those skilled in the art will appreciate that one or more of the described elements may well be combined into a single functional element. Alternatively, certain elements may be split into multiple functional elements. Elements from one embodiment may be added to another embodiment.

[0135] The embodiments disclosed herein can be implemented using at least one hardware device and performing network management functions to control the elements.

[0136] The foregoing description of the specific embodiments will so fully reveal the general nature of the embodiments herein that others can, by applying current knowledge, readily modify and / or adapt for various applications such specific embodiments without departing from the generic concept, and, therefore, such adaptations and modifications should and are intended to be comprehended within the meaning and range of equivalents of the disclosed embodiments. It is to be understood that the phraseology or terminology employed herein is for the purpose of description and not of limitation. Therefore, while the embodiments herein have been described in terms of preferred embodiments, those skilled in the art will recognize that the embodiments herein can be practiced with modification within the scope of the embodiments as described herein.

Claims

Claims

1. A method for controlling at least one electronic device within a multi-device environment, the method comprising: receiving Wireless Fidelity (Wi-Fi) Channel State Information (CSI) data from each electronic device associated with the multi-device environment, wherein the Wi-Fi CSI data comprises at least a CSI signal and an Autocorrelation Function (ACF) threshold; generating a plurality of clusters of the received Wi-Fi CSI data based on an ACF threshold-based dissimilarity learning mechanism and a threshold clustering mechanism with integration of an Unsupervised Class Information Entropy (UCIE) loss; determining a proximity of a user within the multi-device environment based on the plurality of generated clusters and the received Wi-Fi CSI data; and controlling the at least one electronic device within the multi-device environment based on the determined proximity of the user.

2. The method as claimed in claim 1, wherein receiving Wi-Fi CSI data from each electronic device associated with the multi-device environment for the database creation comprises: receiving the Wi-Fi CSI data from each electronic device associated with the multi-device environment during an initial ACF calibration process, wherein the initial ACF calibration process involves a process of determining and setting an appropriate value of the ACF threshold for each electronic device based on the received Wi-Fi CSI data, and wherein the appropriate value of the ACF threshold is determined using a mathematical model; and creating the database on a server to store the received Wi-Fi CSI data and the ACF threshold.

3. The method as claimed in claim 1, wherein the ACF threshold-based dissimilarity learning mechanism comprises: transforming, by an ACF threshold-based feature extractor module, the received Wi-Fi CSI data into a feature space using a dissimilarity loss to enable a formation of the plurality of clusters that accurately represent one or more relationships among the received Wi-Fi CSI data with corresponding ACF threshold, wherein the received Wi-Fi CSI data with similar ACF thresholds are positioned in a close proximity within the feature space, and wherein the dissimilarity loss indicates an ability of dissimilarity learning to identify at least one of differences and similarities between the received Wi-Fi CSI data based on the corresponding ACF threshold.

4. The method as claimed in claim 3, wherein to identify the at least one of differences and similarities between the received Wi-Fi CSI data based on the corresponding ACF threshold comprises: selecting, by the ACF threshold-based feature extractor module, a set of anchor samples from training data; identifying, by the ACF threshold-based feature extractor module, one or more positive samples for each anchor sample based on a first absolute difference, wherein the first absolute difference between a first anchor sample of ACF threshold (A_i^th) and a second anchor sample of ACF threshold (A_j^th) is less than a predefined threshold (γ); and identifying, by the ACF threshold-based feature extractor module, one or more negative samples for each anchor sample based on a second absolute difference, wherein the second absolute difference between the first anchor sample of ACF threshold (A_i^th) and the second anchor sample of ACF threshold (A_j^th) is higher than the predefined threshold (γ).

5. The method as claimed in claim 3, comprising: determining whether a motion detection mechanism is activated within the at least one electronic device;initiating a calibration period in response to determining that the motion detection mechanism is activated within the at least one electronic device, wherein the Wi-Fi CSI data is received from each electronic device during the calibration period; partitioning the received Wi-Fi CSI data into a predetermined quantity of segments, wherein each segment indicates an environmental condition; and allocating a calibrated ACF threshold value label to each segment.

6. The method as claimed in claim 3, comprising: performing one or more data pre-processing operations on the received Wi-Fi CSI data, wherein the one or more data pre-processing operations comprise an amplitude extraction, a phase extraction, a phase sanitization, an outlier removal, and a detrending.

7. The method as claimed in claim 1, wherein the threshold clustering mechanism with integration of the UCIE loss comprises: converting the Wi-Fi CSI data into a feature space using a ACF threshold-based feature extractor module; passing the converted CSI data through a deep learning based module to extract one or more cluster probability distributions; training a clustering module based on the one or more extracted cluster probabilities using the Unsupervised Class Information Entropy (UCIE) loss; utilizing the trained clustering module for inference on new CSI data to assign at least one cluster based on the training.

8. The method as claimed in claim 1, comprising: determining a centroid of each of the plurality of generated clusters, wherein the centroid is determined by averaging one or more ACF threshold values of all Wi-Fi CSI data within a corresponding cluster, wherein the centroid is subsequently employed during an inference stage to swiftly initialize the ACF threshold.

9. The method as claimed in claim 1, further comprising: detecting an addition of another electronic device to the multi-device environment or the modification of at least one environmental context within the multi- device environment; applying a Quick Adaptation Strategy (QAS) to swiftly initialize the ACF threshold based on a threshold validation process in response to detecting that the addition of another electronic device to the multi-device environment or the modification of the at least one environmental context within the multi-device environment; regenerating the plurality of clusters based on the applied QAS; determining the proximity of the user within the multi-device environment based on the plurality of regenerated clusters; and controlling the at least one electronic device within the multi-device environment based on the determined proximity of the user.

10. The method as claimed in claim 9 wherein applying the QAS to swiftly initialize the ACF threshold based on the threshold validation process comprises: determining a QAS-based ACF threshold; determining whether the QAS-based ACF threshold’s confidence is greater than a predefined confidence threshold; and performing one of: initiating a usage-evaluation ACF calibration process in response to determining that the QAS-based ACF threshold is greater than a predefined confidence threshold; or initiating an initial ACF calibration process in response to determining that the QAS-based ACF threshold is lower than the predefined confidence threshold.

11. The method as claimed in claim 1, wherein determining the proximity of the user within the multi-device environment comprises: detecting at least one of the user within the proximity of the at least one electronic device and at least one operation is performed on the at least one electronic device basedon the plurality of clusters generated from the received Wi-Fi CSI data and the received Wi-Fi CSI data; converting proximity embeddings of the at least one detected electronic device in a form of an intention embedding along with a designated key associated with the at least one electronic device, wherein the intention embedding is stored in the at least one electronic device; and determining the proximity of the user based on the intention embedding.

12. The method as claimed in claim 1, wherein determining the proximity of the user within the multi-device environment comprises: detecting at least one of the user within the proximity of the at least one electronic device and at least one operation is performed on the at least one electronic device based on the plurality of clusters generated from the received Wi-Fi CSI data and the received Wi-Fi CSI data; converting proximity embeddings of the at least one detected electronic device in a form of an intention embedding, wherein the intention embedding is stored in the at least one electronic device; comparing the intention embedding with a ground truth; determine a key of an intention embedding vector with a highest number of votes as an actual intended device to be used based on the ground truth; and determining the proximity of the user based on the determined actual intended device.

13. An electronic device for controlling at least one electronic device within a multi- device environment, wherein the electronic device comprising: a memory; a processor; a communicator; a multi device controlling module, operably connected to the memory and the processor, configured to:receive Wireless Fidelity (Wi-Fi) Channel State Information (CSI) data from each electronic device associated with the multi-device environment, wherein the Wi-Fi CSI data comprises at least a CSI signal and an Autocorrelation Function (ACF) threshold; generate a plurality of clusters of the received Wi-Fi CSI data based on an ACF threshold-based dissimilarity learning mechanism and a threshold clustering mechanism with integration of an Unsupervised Class Information Entropy (UCIE) loss; determine a proximity of a user within the multi-device environment based on the plurality of generated clusters and the received Wi-Fi CSI data; and control the at least one electronic device within the multi-device environment based on the determined proximity of the user.

14. The electronic device as claimed in claim 13, wherein to receive Wi-Fi CSI data from each electronic device associated with the multi-device environment for the database creation, the multi device controlling module configured to: receive the Wi-Fi CSI data from each electronic device associated with the multi- device environment during an initial ACF calibration process, wherein the initial ACF calibration process involves a process of determining and setting an appropriate value of the ACF threshold for each electronic device based on the received Wi-Fi CSI data, and wherein the appropriate value of the ACF threshold is determined using a mathematical model; and create the database on a server to store the received Wi-Fi CSI data and the ACF threshold.

15. A non-transitory computer readable recording medium including a program executes a controlling method of an electronic device, the method comprising: receiving Wireless Fidelity (Wi-Fi) Channel State Information (CSI) data from each electronic device associated with the multi-device environment, wherein the Wi-Fi CSI data comprises at least a CSI signal and an Autocorrelation Function (ACF) threshold;generating a plurality of clusters of the received Wi-Fi CSI data based on an ACF threshold-based dissimilarity learning mechanism and a threshold clustering mechanism with integration of an Unsupervised Class Information Entropy (UCIE) loss; determining a proximity of a user within the multi-device environment based on the plurality of generated clusters and the received Wi-Fi CSI data; and controlling the at least one electronic device (100) within the multi-device environment based on the determined proximity of the user.

Images