UNIFORM BASIC MODEL FOR VARIOUS WI-FI SURVEILLANCE TASKS USING CHANNEL CONDITION INFORMATION

Abstract

This disclosure provides an approach for receiving a wireless data stream containing channel state information (CSI) from an entity in a computer network. The approach performs a tokenization process on the CSI to generate input embeddings associated with a task. This tokenization process operates independently of the entity's hardware configurations, parameter configurations, and wireless communication standards. The approach trains a basic model based on the input embeddings, which is then trained to understand the task. The approach generates an activity prediction associated with the task.

Description

CROSS-REFERENCE TO RELATED REGISTRATION(S)

[0001] This application claims the benefit and priority of the preliminary US application with serial number 63 / 700,055 entitled “A Unified Foundational Model for Diverse Wi-Fi Sensing Tasks using Channel State Information”, filed on September 27, 2024, which is expressly incorporated herein by reference in its entirety. TECHNICAL AREA

[0002] Embodiments of the present disclosure relate to the acquisition and, in particular, to various acquisition tasks using channel state information (CSI). BACKGROUND

[0003] The proliferation of wireless networks has opened new avenues for the passive acquisition of environmental and / or human activity detection. Passive environmental and / or human activity detection can leverage the ubiquitous nature of wireless infrastructure. Wireless detection can be used for a wide range of applications. Channel characteristics of a wireless link can be used to obtain information about the propagation environment, which can enable the detection of environmental perception. BRIEF DESCRIPTION OF THE DRAWINGS

[0004] The described embodiments and their advantages are best understood with reference to the following description in conjunction with the accompanying drawings. These drawings in no way restrict modifications in form and detail that a person skilled in the art may make to the described embodiments without departing from the spirit and scope of the described embodiments. Fig. Figure 1 is a block diagram illustrating an exemplary system for various data acquisition tasks using CSI according to some embodiments of the present disclosure. Fig. Figure 2A is a block diagram illustrating an example of a preprocessing module according to some embodiments of the present disclosure. Fig. Figure 2B is a block diagram illustrating an example of an output module according to some embodiments of the present disclosure. Fig. Figure 3 is a block diagram illustrating an example of a deep clustering module according to some embodiments of the present disclosure. Fig. Figure 4 is a block diagram illustrating an example of a basic model according to some embodiments of the present disclosure. Fig. Figure 5 is a flowchart of a procedure for various task capture using CSI according to some embodiments. Fig. Figure 6 is a block diagram illustrating an exemplary system for capturing various tasks using CSI according to some embodiments of the present disclosure. Fig. Figure 7 is a block diagram of an exemplary computing device that can perform one or more of the operations described herein, according to some embodiments of the present disclosure. Fig. Figure 8 is a block diagram 800 illustrating an exemplary tokenization process according to some embodiments of the present disclosure. DETAILED DESCRIPTION

[0005] Wireless sensing, particularly Wi-Fi® sensing using CSI, has proven to be a powerful, non-invasive technique for a wide range of applications, such as capturing human activity and / or environmental changes. For example, some sensing applications include gesture recognition, human pose estimation, vital sign monitoring, and human activity detection. CSI characterizes the channel characteristics of a wireless link and captures fine-grained information about the propagation environment, enabling the detection of human activity and environmental changes.

[0006] Despite advances in Wi-Fi capture, current approaches face several challenges. For example, some methods are designed for specific tasks, resulting in limited generalizability and the need for task-specific model development. In another example, the performance of these models can degrade when deployed in new environments or when confronted with unseen activity, requiring extensive data collection and model fine-tuning. Yet another example: the increasing complexity of capture tasks necessitates more sophisticated models, which in turn require larger labeled datasets and computational resources.

[0007] This disclosure addresses the challenges of Wi-Fi® capture by employing a novel basic modeling approach that scales to multiple Wi-Fi capture tasks using CSI. This disclosure adapts and modifies the deep learning architecture (e.g., Mamba) for processing CSI data. CSI data, like natural language, exhibits complex temporal and spatial dependencies that can be effectively captured by state-space models and selective mechanisms. This disclosure includes CSI-specific enhancements to better handle the unique characteristics of Wi-Fi signals.

[0008] This disclosure employs an innovative foundational model for scaling to multi-task Wi-Fi acquisition using a hybrid architecture that combines transformer-based self-attention layers with state-space layers. This disclosure enables efficient processing of high-dimensional CSI data across multiple subcarriers and time steps. It scales to multi-task learning while simultaneously addressing various Wi-Fi acquisition applications, such as gesture recognition, gait analysis, human activity detection, and occupancy detection. This joint representation approach enhances generalization and facilitates efficient scaling to new scenarios with limited labeled data. Furthermore, this disclosure can employ a pre-training strategy that mixes supervised and unsupervised learning objectives on a large corpus of unlabeled CSI data.

[0009] In one embodiment, the present disclosure uses a processing device for receiving a wireless data stream comprising CSI. In one embodiment, the processing device can receive the wireless data stream comprising the CSI from an entity in a computer network.

[0010] The processing device performs a tokenization process on the CSI to generate input embeddings associated with a task, with the tokenization process operating independently of the entity's hardware configurations, parameter configurations, or wireless communication standards. In some embodiments, the task comprises one or more specific tasks. For example, activity prediction determines the specific task based on the activity forecast.

[0011] The processing device trains a basic model based on the input embeddings. The basic model can be trained to grasp the task. In some embodiments, the basic model includes one or more state-space layers that maintain a state of the basic model during training. In some embodiments, the basic model includes multiscale integration, which involves parallel processing of different scales to obtain weights for the basic model associated with grasping the task.

[0012] The processing device generates an activity prediction associated with the task. For example, the activity prediction can indicate a type of environmental change based on the CSI within the wireless data stream.

[0013] In some embodiments, the processing device processes the wireless data stream comprising the CSI to generate a dimensional vector compatible with the tokenization process. In some embodiments, transformations applied during the processing of the wireless data stream improve the robustness of the tokenization process. In some embodiments, the processing logic generates one or more views of the CSI corresponding to a feature of the task, each of the one or more views corresponding to a capture characteristic associated with the feature. Each of the one or more views can be transformed based on at least one channel shuffling, time-slicing, or affine transformation. In some embodiments, the channel shuffling performs random subcarrier permutations on the CSI.In some embodiments, the timescale adjusts the timing of the CSI to preserve motion signatures. In some embodiments, the affine transformation scales or rotates the CSI. In some embodiments, any one or more views can be provided as input for adaptive learning associated with feature acquisition based on the CSI.

[0014] As discussed herein, the present disclosure provides an approach that improves Wi-Fi sensing by using a novel basic model approach that scales to multiple Wi-Fi sensing tasks using CSI. Additionally, the present disclosure provides an improvement in Wi-Fi sensing by offering versatility across various sensing tasks, such as, but not limited to, gesture recognition, gait analysis, human activity detection, or occupancy detection, and also provides a scalable solution that outperforms task-specific models while maintaining computational efficiency.

[0015] Fig. Figure 1 is a block diagram illustrating an exemplary system for various data acquisition tasks using CSI according to some embodiments of the present disclosure.

[0016] System 100 comprises a computer network 101, a server 110, and a network 105. The computer network 101 may include an internal network 102 and an entity 103, wherein the entity 103 transmits a wireless data stream. The wireless data stream transmitted by the entity 103 may include wireless Wi-Fi® transmissions. However, in some embodiments, the wireless transmissions may be other wireless transmission protocols, and the disclosure is not intended to be limited to the examples described herein. System 100 of Fig. Figure 1 shows a computer network 101, but the computer network 101 can include any type of network (e.g., in a home, private, business, corporate, etc.) and the disclosure is not intended to be limited to the examples disclosed herein.

[0017] Entity 103 can transmit the wireless data stream, and server 110 can receive the wireless data stream over network 105. The wireless data stream can include CSI 106. Server 110 includes a preprocessing module 112, a deep clustering module 113, a base model 114, and an output module 115. The preprocessing module 112 can receive the CSI 106 and perform initial processing of the CSI. For example, the preprocessing module 112 can preprocess the CSI 106 in preparation for reception by the deep clustering module 113. The deep clustering module 113 can be configured to perform a tokenization process on the CSI within the wireless data stream. In some embodiments, the CSI 106 are not preprocessed by the preprocessing module 112 and are received by the deep clustering module 113.The Deep Clustering Module 113 can use the CSI, either preprocessed by the Preprocessing Module 112 or raw CSI, and perform feature extraction based on the channel properties of the CSI, generating input embeddings for the Basic Model 114. The Basic Model 114 uses the input embeddings from the Deep Clustering Module 113 to train or fine-tune the Basic Model to perform acquisition based on the CSI. The Output Module 115 generates an activity prediction using the results of the Basic Model 114.

[0018] Fig. 2A is a block diagram 200 illustrating an example of a preprocessing module 112 according to some embodiments of the present disclosure.

[0019] In some embodiments, the preprocessing module 112 can process the CSI in preparation for the deep clustering module 113. The preprocessing module 112 can preprocess the CSI using various procedures. For example, block diagram 200 of Fig. 2A some procedures that can be implemented by the preprocessing module 112, such as adaptive subcarrier selection 201, complex feature conservation 202, median normalization 203 or phase correction 204.

[0020] In some embodiments, the preprocessing module 112 implements a number of sophisticated techniques to extract striking features from raw CSI data. We refer to the complex channel frequency response as: H(f,t)∈CNT×NR where N T and N RThe preprocessing module 112 represents transmit and receive antennas. It incorporates adaptive subcarrier selection based on signal-to-noise ratio (SNR) thresholding, preserving complex-valued features to capture subtle phase changes. The module employs a robust, median-based normalization technique to mitigate outlier effects, followed by linear phase correction using a state-space formulation. The process culminates in a time-domain transformation using a window function (e.g., Chebyshev window function) and a fast Fourier transform (FFT). This approach yields highly accurate, noise-resistant CSI features critical for various Wi-Fi capture tasks, while maintaining adaptability across different hardware configurations and environmental conditions.

[0021] Fig. 2B is a block diagram 220 illustrating an example of an output module 115 according to some embodiments of the present disclosure.

[0022] Output module 115 can include task-specific heads or a classification of tasks predicted or identified from the CSI. For example, output module 115 can include gesture recognition 221, activity recognition 222, gait analysis 223, or presence recognition 224. Gesture recognition 221 can include a prediction regarding whether the data within the CSI corresponds to gestures recognized (e.g., human poses). Activity recognition 222 can include a prediction regarding whether the data within the CSI corresponds to a specific activity (e.g., human activity, movement). Gait analysis 223 can include a prediction regarding whether the data within the CSI corresponds to a person's gait (e.g., a person walking, running, jogging).The presence detection 224 can include a prediction relating to whether the data within the CSI corresponds to something that is recognized as present (e.g., person(s) present, obstacles, objects, cars, etc.).

[0023] Fig. Figure 3 is a block diagram 300 illustrating an example of a deep clustering module according to some embodiments of the present disclosure.

[0024] The Deep Clustering Module 113 can receive raw CSI or preprocessed CSI from the Preprocessing Module 112 and perform a tokenization process. Analogous to tokenization in large-scale language models, the Deep Clustering Module 113 is optimized for high-dimensional CSI data. It employs contrast cluster mapping to transform raw or preprocessed CSI signals into a learned, discrete vocabulary of Wi-Fi capture primitives. The Deep Clustering Module 113 enables data-efficient learning of unlabeled CSI, improves cross-task generalization, and enhances robustness against CSI-specific noise. The resulting feature space serves as a powerful initialization for various Wi-Fi capture tasks, facilitating quick-to-adaptation and multi-task scalability.In some embodiments, the tokenization process performed by the Deep Clustering Module 113 is intended to operate across different hardware implementations, parameter configurations, and wireless standards. For example, the Deep Clustering Module 113 can perform the tokenization process regardless of the transmission scheme used to transmit the data stream containing the CSI. The Deep Clustering Module 113 can perform the tokenization process regardless of the hardware implementation or configuration (e.g., antenna panels, diversity, etc.) of the entity transmitting the data stream containing the CSI. The Deep Clustering Module 113 can perform the tokenization process regardless of the parameter configuration (e.g., beacon CSI or data CSI) of the CSI.In some embodiments, the Deep Clustering Module 113 may include abstraction mechanisms that are used to normalize inputs from different hardware configurations and wireless standards into a unified representation.

[0025] The Deep Clustering Module 113 can perform multiple view generation 301, generating several views (e.g., View1 303a, View2 303b, ViewN 303N) that preserve acquisition characteristics. These multiple views can help maintain physical signal properties of the CSI by providing a diversity of perspectives. The multiple views can be generated based on features identified via Feature Extraction 302. The Deep Clustering Module 113 implements a CSI-specific extension strategy T to generate diverse views of each CSI sample. For input x, extended views can be generated. x1t,x2t,…,xVt are created where t ∼ T. The extension 304 includes channel shuffling 305, which performs channel shuffling, time extension 306, which extends the timing of the CSI, or affine transformation 307, which performs random affine transformations.

[0026] The deep clustering module 113 includes a prototype mapping 308 that performs an iterative refinement of the CSI. The deep clustering module 113 can include scaling 309, mapping 310, and loss calculation 311, with the results of the loss calculation 311 being fed back into scaling 309 (feedback 312). For example, the prototype mapping 308 can use an algorithm (e.g., the Sinkhorn-Knopp algorithm) for entropy-controlled optimal transport that uses K prototype features. quote This approach ensures balanced clustering for heterogeneous CSI data. In some implementations, iterative refinement can be performed on Cij = |z t - cj| 2Based on and regularization ε, this provides robust, task-agnostic features. Combined with multi-view extension, this enables a powerful, unsupervised learning framework for various Wi-Fi capture tasks. Prototype mapping 308 can then generate learned embeddings 313 based on multi-view generation 301, extension 304, and prototype mapping 308.

[0027] In some embodiments, the exchanged prediction mechanism can be improved with entropy regularization to promote diverse and informative cluster assignments. For a pair of views (s, t), the loss function is: L(zs,zt)=−∑k[qs(k)log pt(k)+qt(k) log ps(k)]+λ[H(qs)+H(qt)] where q and p are the assigned and predicted probabilities, respectively, H(·) is the entropy function, and λ controls the entropy regularization strength. This formulation can promote consistency between different views while maintaining informative mappings.

[0028] Fig. Figure 4 is a block diagram 400 illustrating an exemplary basic model according to some embodiments of the present disclosure.

[0029] The basic model 114 can include input processing 401, state space layers 405, and multiscale integration 409. Input processing 401 can perform processing procedures (e.g., normalization 402, feature projection 403, and / or dimensioning 404) on the output from the deep clustering module 113. Normalization 402 can normalize the layers, feature projection 403 can determine which features associated with the task have been identified, and dimensioning 404 can adjust the dimension of the data received from the deep clustering module 113.

[0030] The state space layers 405 can comprise one or more state space layers (e.g., state space layer 1 405a, state space layer 2 405b, or state space layer N 405N). Each state space layer can include a state update 406, a selective mechanism 407, and dependencies 408. The state update 406 can maintain the space while the base model 114 is being trained. The selective mechanism 407 controls the flow of information and captures cross-carrier relationships. The dependencies 408 can identify global dependencies.

[0031] The multiscale integration 409 can include one or more scales (e.g., Scale1 410a, Scale2 410b, ScaleN 410N) that perform parallel processing, scale-specific convolutions, or adaptive pooling. In some embodiments where the multiscale integration 409 includes three scales (e.g., S = 3), the first scale can process the data to identify fine-grained motion (e.g., 20 ms timeframe), the second scale can process the data to identify medium-term patterns (e.g., 100 ms timeframe), and the third scale can process the data to identify long-term behavior (e.g., 500 ms timeframe). The results of the one or more scales can be used to generate a weighted sum 411.

[0032] In some embodiments, the basic model 114 utilizes its ability to capture far-reaching dependencies and influence temporal dynamics. The basic model 114, which can be based on a Mamba architecture tailored for Wi-Fi capture, can be defined by the following state-space equations: ddth(t)=A(x)h(t)+B(x)u(t)y(t)=C(x)h(t)+D(x)u(t) where h(t) is the hidden state, u(t) is the input, y(t) is the output, and A(x), B(x), C(x), D(x) are input-dependent parameters learned through a hypernetwork approach.

[0033] In some embodiments, a CSI-specific selective mechanism can be used for dynamic receptive field fitting: A(x) = diag(λ(x)) + low-rank(φ(x)), where λ(x) determines the state retention per feature dimension and φ(x) captures global dependencies via a low-rank update. This mechanism enables efficient modeling of CSI-specific temporal dynamics.

[0034] In some implementations, multiscale integration can capture various temporal patterns in CSI data: yt=∑s=1Sws(xt)yts where yts the output of the basic model at scale s is and w s (x t ) attention-learned input-dependent weights. This enables the present disclosure to model a spectrum of temporal dynamics from fast gesture patterns to slow gait patterns, thus improving the ability across various Wi-Fi detection tasks.

[0035] In some embodiments, the basic model 114 can be trained end-to-end using a combination of unsupervised and supervised targets. The total loss is a dynamically weighted combination: Ltotal=α(t)Lu+(1−α(t))Ls where α(t) is a curriculum for learning that gradually shifts the focus from unsupervised to supervised learning as training progresses. In some embodiments, sharpness-aware minimization (SAM) with layer-wise adaptive rate scaling (LARS) can be used to improve generalization and stabilize training across different Wi-Fi capture tasks. The end-to-end training approach can enable the basic Model 114 to learn general, transferable features from large amounts of unlabeled CSI data while also adapting to specific Wi-Fi capture tasks, enabling superior performance across a wide range of applications.

[0036] Fig. Figure 8 is a block diagram 800 illustrating an exemplary tokenization process according to some embodiments of the present disclosure. Block diagram 800 may include features or elements previously discussed herein, and such features or elements are not discussed here to minimize redundancy.

[0037] In some embodiments, the preprocessing module 112 can receive multi-standard CSI data 801. The multi-standard CSI data 801 can include CSI transmitted using various wireless standards, such as, but not limited to, 802.11b CSI 802, 802.11ac CSI 803, 802.11ax CSI 804, or 802.11be CSI 805. In some embodiments, the adaptive subcarrier selection 201 can select a subcarrier across different bandwidths. In some embodiments, the median normalization 203 can perform normalization of the multi-standard CSI data 801 based on an entity antenna configuration or a CSI source type (e.g., beacons, data). In some embodiments, the preprocessing module 112 can process the multi-standard CSI data 801 over different frequency bandwidths (e.g.20 MHz, 40 MHz) using a window function and an FFT transformation 806, enabling the extraction of high-precision, noise-resistant CSI features while maintaining adaptability across different hardware configurations and environmental conditions. In some embodiments, the window function may include a Chebyshev window function or the like.

[0038] After preprocessing the multi-standard CSI data 801, the multi-standard CSI data 801 can be received by extension 304, which performs an extension process on the multi-standard CSI data 801. Extension 304 can further include data overlay 807 and enhanced views 808, which include CSI capture, as in the example shown in diagram 800. Contrasting cluster mapping 809 can receive the output from extension 304. Contrasting cluster mapping 809 includes feature extraction 810, algorithm 811, prototype mapping 812, entropy regularization 813, and prediction loss 814. Feature extraction 810 can be configured similarly to feature extraction 302. Algorithm 811 can use an algorithm (e.g., Sinkhorn-Knopp algorithm) for entropy-regulated optimal transport, which can be based on DTIM periods (DTIM = Delivery Traffic Indication Message) (e.g., 1, 3, or 10).Prototype Mapping 812 can perform iterative refinement of the multi-standard CSI data 801 in a similar manner to Prototype Mapping 308. Entropy Regularization 813 can perform various and informative cluster mappings, and Prediction Loss 814 can perform exchanged prediction loss based on the loss function described in conjunction with Loss Calculation 311.

[0039] Output 815 can result in unified CSI tokens 816 and / or task-agnostic embeddings 817. Output 815 can be hardware-agnostic and independent of the wireless standard used to transmit the multi-standard CSI data 801. The unified CSI tokens 816 and / or the task-agnostic embeddings can be obtained as the output of contrasting cluster mapping 809. The unified CSI tokens 816 and / or the task-agnostic embeddings can be associated with an activity prediction that can indicate a type of environmental change based on the multi-standard CSI data 801.

[0040] Fig. Figure 5 is a flowchart of a procedure 500 for various task captures using CSI according to some embodiments.

[0041] Method 500 can be performed by processing logic that may include hardware (e.g., a processing device), software (e.g., instructions running or being executed on a processing device), firmware (e.g., microcode), or a combination thereof. In some embodiments, at least part of Method 500 can be performed by Server 110 (shown in Fig. 1), the processing device 610 (shown in Fig. 6), the processing device 702 (shown in Fig. 7) or a combination thereof.

[0042] With reference to Fig. Figure 5 illustrates exemplary functions used by various embodiments of Method 500. Although specific functional blocks (“blocks”) are disclosed in Method 500, such blocks are examples. That is, embodiments are well suited to carrying out various other blocks or variations of the blocks mentioned in Method 500. It is understood that the blocks in Method 500 can be carried out in a different order than shown and that not all of the blocks can be carried out in Method 500.

[0043] With reference to Fig. Procedure 500 begins at block 510, after which the processing logic receives a wireless data stream containing CSI. The processing logic can receive the wireless data stream containing the CSI from an entity on a computer network.

[0044] In some embodiments, the processing logic at block 512 can determine whether the data stream containing the CSI should be preprocessed in preparation for the tokenization process. If, in some embodiments, the processing logic determines that the data stream containing the CSI should be preprocessed (e.g., a yes branch), the data stream containing the CSI is preprocessed at block 515 in preparation for the tokenization process. For example, the processing logic processes the wireless data stream containing the CSI to generate a dimensional vector compatible with the tokenization process. Transformations applied during the processing of the wireless data stream can improve the robustness of the tokenization process. If, in some embodiments, the processing logic determines that the data stream containing the CSI should not be preprocessed (e.g.,Without branching, the data stream containing the CSI can proceed to block 520. For example, the processing logic might determine that the data stream containing the CSI is compatible with the tokenization process, thus eliminating the need for preprocessing.

[0045] At block 520, the processing logic performs a tokenization process on the CSI to generate input embeddings associated with a task. In some embodiments, the tokenization process can operate independently of the entity's hardware configurations, parameter configurations, or wireless communication standards. For example, the tokenization process operates independently of the entity's hardware configurations or parameter configurations associated with a wireless data stream transmission, including one or more of the following: bandwidth configurations, antenna configurations, underlying hardware implementations, a CSI capture configuration (e.g., requested or unsolicited), a CSI source type (e.g., beacon CSI or data CSI), or DTIM periods (e.g., 1, 3, or 10).In some embodiments, the tokenization process projects CSI data within the input embeddings across different wireless communication standards into a consistent embedding representation. For example, the different wireless communication standards may include, but are not limited to, cellular communications (e.g., 4G, 5G, 6G, etc.), Institute of Electrical and Electronics Engineers (IEEE) standards (e.g., 802.11b, 802.11ac, 802.11ax, 802.11be, etc.). In some embodiments, the task comprises one or more specific tasks. For example, activity prediction determines the specific task based on the activity prediction. In some embodiments, a tokenized representation of the task can be consistent across different downstream capture configurations.

[0046] In Block 530, the processing logic trains a base model based on the input embeddings. This base model can be trained to grasp the task. In some embodiments, the base model includes one or more state-space layers that maintain a state of the base model during training. In some embodiments, the base model includes multiscale integration, which involves parallel processing of different scales to obtain weights for the base model associated with grasping the task.

[0047] In block 540, the processing logic generates an activity prediction associated with the task. For example, the activity prediction might indicate a type of environmental change based on the CSI within the wireless data stream.

[0048] In some embodiments, the processing logic generates one or more views of the CSI corresponding to a feature of the task, with each of the one or more views corresponding to a capture characteristic associated with the feature. Each of the one or more views can be transformed based on at least one channel shuffling, time-slicing operation, or affine transformation. In some embodiments, the channel shuffling performs random subcarrier permutations on the CSI. In some embodiments, the time-slicing operation adjusts the timing of the CSI to preserve motion signatures. In some embodiments, the affine transformation scales or rotates the CSI. In some embodiments, each of the one or more views can be provided as input for adaptive learning associated with feature capture based on the CSI.

[0049] Fig. Figure 6 is a block diagram 600 illustrating an exemplary system for various task capture using CSI according to some embodiments of the present disclosure.

[0050] The computer system 601 comprises the processing device 610 and the memory 615. The memory 615 stores instructions 620 that are executed by the processing device 610. The processing device is operationally coupled to the memory for the following: receiving, from an entity 603 in a computer network 602, a wireless data stream 604 that includes CSI 605. The processing device is operationally coupled to the memory for the following: performing a tokenization process 630 on the CSI 605 to generate input embeds 632 that are associated with a task 631. The tokenization process 630 operates independently of the hardware configurations 603a, parameter configurations 603b, or wireless communication standards 603c of the entity 603.

[0051] The processing device is operationally coupled to the memory for the following: Training a basic model 640 based on the input embeddings 632, wherein the basic model 640 is trained to capture the task 631. The processing device is operationally coupled to the memory for the following: Generating an activity prediction 650 associated with the task 631.

[0052] Fig. Figure 7 illustrates a schematic representation of a machine in the exemplary form of a Computer System 700, containing a set of instructions to cause the machine to perform one or more of the methodologies discussed herein for acquiring various tasks using CSI.

[0053] In alternative embodiments, the machine can be connected (e.g., networked) to other machines in a local area network (LAN), an intranet, an extranet, or the internet. The machine can operate in the capacity of a server or client machine in a client-server network environment, or as a peer machine in a peer-to-peer (or distributed) network environment. The machine can be a personal computer (PC), a tablet PC, a set-top box (STB), a personal digital assistant (PDA), a mobile phone, a web application, a server, a network router, a switch or bridge, a hub, an access point, a network access control device, or any machine capable of executing a set of instructions (sequential or otherwise) that specify actions to be taken by that machine.Furthermore, although only a single machine is illustrated, the term "machine" is also to be understood as encompassing any collection of machines that, individually or collectively, execute a set (or sets) of instructions to carry out one or more of the methodologies discussed herein. In some embodiments, Computer System 700 may be representative of a server.

[0054] The exemplary computer system 700 comprises a processing device 702, a main memory 704 (e.g., read-only memory (ROM), flash memory, dynamic random-access memory (DRAM)), a static memory 706 (e.g., flash memory, static random-access memory (SRAM), etc.), and a data storage device 718, which communicate with each other via a bus 730. Any of the signals provided via different buses described herein can be time-division multiplexed with other signals and provided via one or more common buses. Additionally, the connection between circuit components or blocks can be shown as buses or as individual signal lines. Each of the buses can alternatively be one or more individual signal lines, and each of the individual signal lines can alternatively be a bus.

[0055] The computer system 700 may further comprise a network interface device 708 that can communicate with a network 720. The computer system 700 may also comprise a video display unit 710 (e.g., a liquid crystal display (LCD) or a cathode ray tube (CRT)), an alphanumeric input device 712 (e.g., a keyboard), a cursor control device 714 (e.g., a mouse), and an acoustic signal generation device 716 (e.g., a loudspeaker). In some embodiments, the video display unit 710, the alphanumeric input device 712, and the cursor control device 714 may be combined into a single component or device (e.g., an LCD touchscreen).

[0056] The processing device 702 represents one or more general-purpose processing devices, such as a microprocessor, a central processing unit, or the like. Specifically, the processing device may be a complex instruction set (CISC) microprocessor, a reduced instruction set computer (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, or a processor implementing other instruction sets, or processors implementing a combination of instruction sets. The processing device 702 may also be one or more special-purpose processing devices, such as an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a digital signal processor (DSP), a network processor, or the like. The processing device 702 is configured to execute tokenization instructions 725 to perform the operations and steps discussed herein.

[0057] The data storage device 718 may comprise a machine-readable storage medium 728 on which one or more sets of tokenization instructions 725 (e.g., software) are stored, embodying one or more of the methodologies of functions described herein. The tokenization instructions 725 may, during their execution by the computer system 700, also reside wholly or at least partially within the main memory 704 or within the processing device 702; the main memory 704 and the processing device 702 also being machine-readable storage media. The tokenization instructions 725 may furthermore be transmitted or received over a network 720 via the network interface device 708.

[0058] The machine-readable storage medium 728 can also be used to store instructions for performing a method of intelligent container scheduling, as described herein. Although the machine-readable storage medium 728 is shown in an exemplary embodiment as a single medium, the term "machine-readable storage medium" should be understood to include a single medium or multiple media (e.g., a centralized or distributed database or associated caches and servers) that store the one or more sets of instructions. A machine-readable medium includes any mechanism for storing information in a form (e.g., software, processing application) that is readable by a machine (e.g., a computer). The machine-readable medium can be a magnetic storage medium (e.g., a floppy disk); an optical storage medium (e.g., a optical disc);CD-ROM); a magneto-optical storage medium; a read-only memory (ROM); a random-access memory (RAM); a erasable programmable memory (e.g., EPROM and EEPROM); a flash memory; or any other type of medium suitable for storing electronic instructions, but is not limited to.

[0059] Unless expressly stated otherwise, terms such as "receive," "perform," "train," "generate," "process," "transform," "shuffle," or the like refer to actions and processes performed or implemented by computing devices that manipulate and transform data represented as physical (electronic) quantities within the registers and memories of the computing device into other data similarly represented as physical quantities within the memories or registers of the computing device or other such information storage, transmission, or display devices. Furthermore, the terms "first," "second," "third," "fourth," etc., as used herein, are intended as markers to distinguish between different elements and need not necessarily have an ordinal meaning according to their numerical designation.

[0060] The examples described herein also refer to a device for performing the operations described herein. This device may be specially designed for specific purposes, or it may comprise a general-purpose computing device that is selectively programmed by a computer program stored in the computing device. Such a computer program may be stored on a computer-readable, non-volatile storage medium.

[0061] The methods and illustrative examples described herein do not inherently refer to any particular computer or other device. Various general-purpose systems may be used in accordance with the teachings described herein, or it may prove expedient to construct more specialized devices to carry out the method steps. The structure for a multitude of such systems appears as set forth in the preceding description.

[0062] The foregoing description is intended to be illustrative and not limiting. Although the present disclosure has been described with reference to specific illustrative examples, it is understood that the present disclosure is not limited to the examples described. The scope of the disclosure should be determined with reference to the following claims together with the full scope of equivalents to which the claims entitle the holder.

[0063] As used herein, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise. It is further understood that the terms "includes," "containing," "comprises," and / or "comprehensive," when used herein, indicate the presence of specified features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. Therefore, the terminology used herein serves only to describe certain embodiments and is not intended to be restrictive.

[0064] It should also be noted that in some alternative implementations, the noted functions / steps may occur out of the order shown in the figures. For example, two figures shown consecutively may actually be executed essentially simultaneously, or may sometimes be executed in reverse order, depending on the functionality / steps involved.

[0065] Although the procedural operations have been described in a specific order, it is understood that other operations can be performed between the described operations, the described operations can be adapted so that they occur at slightly different times, or the described operations can be distributed in a system that allows the processing operations to occur at different intervals associated with the processing.

[0066] Various units, circuits, or other components may be described or claimed to be "configured to" or "configurable to" perform a task or tasks. In such contexts, the expression "configured to" or "configurable to" is used to denote a structure by indicating that the units / circuits / components comprise a structure (e.g., a circuit) that performs the task or tasks during operation. Thus, the unit / circuit / component can be assumed to be configured to perform the task, or configurable to perform the task, even if the specified unit / circuit / component is not currently operational (e.g., not powered on).The units / circuits / components used with the language "configured to" or "configurable to" include hardware—for example, circuits, memory that stores program instructions executable to implement the operation, etc. The statement that a unit / circuit / component is "configured to" perform one or more tasks or "configurable to" perform one or more tasks is expressly not intended to invoke 35 USC § 112(f) for that unit / circuit / component. Additionally, "configured to" or "configurable to" may include a generic structure (e.g., a generic circuit) that is manipulated by software and / or firmware (e.g., an FPGA or a general-purpose processor running software) to operate in a manner capable of performing the task(s) in question. "Configured to" may also include the adaptation of a manufacturing process (e.g.,a semiconductor manufacturing plant) for the manufacture of devices (e.g., integrated circuits) that are adapted to implement or perform one or more tasks. "Configurable to" is expressly not to be applied to blank media, an unprogrammed processor or generic computer, or an unprogrammed programmable logic device, a programmable gate array, or any other unprogrammed device unless accompanied by programmed media that enable the unprogrammed device to be configured to perform the disclosed function(s).

[0067] The foregoing description has been provided for illustrative purposes with reference to specific embodiments. However, the foregoing illustrative discussions are not intended to be exhaustive or to limit the present disclosure to the exact forms disclosed. Many modifications and variations are possible in light of the foregoing teachings. The embodiments have been selected and described to best explain the principles of the embodiments and their practical applications, thereby enabling other skilled persons to best utilize the embodiments and various modifications as they may be suitable for the particular use under consideration.Accordingly, the present embodiments are to be regarded as illustrative and not limiting, and the present disclosure is not to be limited to the details given herein, but may be modified within the scope and equivalents of the attached claims. QUOTES INCLUDED IN THE DESCRIPTION

[0000] This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited patent literature

[0000] US 63 / 700,055

[0001]

Claims

[1] A procedure that includes the following: Receiving, from an entity in a computer network, a wireless data stream that includes channel state information (CSI); Performing, through a processing device, a tokenization process at the CSI to generate input embeddings associated with a task, wherein the tokenization process operates independently of the entity's hardware configurations, parameter configurations, or wireless communication standards; Training a base model based on the input embeddings, whereby the base model is trained to grasp the task; and Generating an activity prediction associated with the task. [2] The method according to claim 1, further comprising: Processing the wireless data stream encompassing the CSI to generate a dimensional vector compatible with the tokenization process, with transformations applied during the processing of the wireless data stream improving the robustness of the tokenization process. [3] Method according to claim 1, wherein the tokenization process further comprises: Generating one or more views of the CSI that correspond to a feature of the task, wherein each of the one or more views corresponds to a capture characteristic associated with the feature. [4] Method according to claim 3, wherein each of the one or more views is transformed based on at least one channel shuffling, time step or affine transformation. [5] Method according to claim 4, wherein the channel shuffling performs random subcarrier permutations on the CSI, wherein the time path adjusts a timing control of the CSI to preserve motion signatures, wherein the affine transformation scales or rotates the CSI. [6] Method according to claim 4, wherein each of the one or more views is provided as input for adaptive learning associated with the detection of the feature based on the CSI. [7] Method according to claim 1, wherein the basic model comprises one or more state space layers that maintain a state of the basic model during the training of the basic model. [8] Method according to claim 1, wherein the basic model comprises a multiscale integration which includes parallel processing of different scales to obtain weights for the basic model associated with capturing the task. [9] Method according to claim 1, wherein the tokenization process operates independently of the hardware configurations or parameter configurations of the entity associated with a transmission of the wireless data stream, wherein the hardware configurations or parameter configurations of the entity comprise at least one or more of the following: Bandwidth configurations, Antenna configurations, underlying hardware implementations, a CSI capture configuration, a CSI source type or DTIM periods (DTIM = Delivery Traffic Indication Message). [10] Method according to claim 1, wherein the task comprises one or more specific tasks, wherein the activity prediction determines a specific task based on the activity prediction, wherein a tokenized representation of the task is consistent across different downstream sensing configurations. [11] Method according to claim 1, wherein the tokenization process projects CSI data within the input embeddings across different wireless communication standards into a consistent embedding representation. [12] A system that includes the following: a storage facility; and a processing device that is operationally coupled to the memory and configured for the following: Receiving, from an entity in a computer network, a wireless data stream that includes channel state information (CSI); Performing, through the processing device, a tokenization process at the CSI to generate input embeddings associated with a task, the tokenization process operating independently of the entity's hardware configurations, parameter configurations, or wireless communication standards; Training a base model based on the input embeddings, wherein the base model is trained to grasp the task; and Generating an activity prediction associated with the task. [13] System according to claim 12, wherein the processing device is configured for the following: Processing the wireless data stream encompassing the CSI to generate a dimensional vector compatible with the tokenization process, with transformations applied during the processing of the wireless data stream improving the robustness of the tokenization process. [14] System according to claim 12, wherein the processing device for performing the tokenization process is configured for the following: Generating one or more views of the CSI corresponding to a feature of the task, wherein each of the one or more views corresponds to a capture characteristic associated with the feature, and wherein each of the one or more views is transformed based on at least one channel shuffling, time span, or affine transformation. [15] System according to claim 14, wherein the channel shuffling performs random subcarrier permutations on the CSI, wherein the time path adjusts a timing control of the CSI to preserve motion signatures, wherein the affine transformation scales or rotates the CSI, wherein each of the one or more views is provided as input for adaptive learning associated with feature capture based on the CSI. [16] System according to claim 12, wherein the basic model comprises one or more state space layers to maintain a state of the basic model during the training of the basic model, wherein the basic model comprises multiscale integration which includes parallel processing of different scales to obtain weights for the basic model associated with capturing the task. [17] System according to claim 12, wherein the tokenization process operates independently of the hardware configurations or parameter configurations of the entity associated with a transmission of the wireless data stream, wherein the hardware configurations or parameter configurations of the entity comprise at least one or more of the following: Bandwidth configurations, Antenna configurations, underlying hardware implementations, a CSI capture configuration, a CSI source type or DTIM periods (DTIM = Delivery Traffic Indication Message). [18] System according to claim 12, wherein the task comprises one or more specific tasks, wherein the activity prediction determines the specific task based on the activity prediction, wherein a tokenized representation of the task is consistent across different downstream sensing configurations. [19] System according to claim 12, wherein the tokenization process serves to project CSI data within the input embeddings across different wireless communication standards into a consistent embedding representation. [20] A non-volatile, computer-readable storage medium comprising instructions which, when executed by a processing device, cause the processing device to: Receiving, from an entity in a computer network, a wireless data stream that includes channel state information (CSI); Performing a tokenization process on the CSI to generate input embeddings associated with a task, with the tokenization process operating independently of the entity's hardware configurations, parameter configurations, or wireless communication standards; Training a base model based on the input embeddings, wherein the base model is trained to grasp the task; and Generating an activity prediction associated with the task.

Images