Interference signal separation model training method and related equipment
By performing dual-scale decomposition and training on Wi-Fi channel state information data, and using an interference signal separation model for adaptive differentiation and filtering, the problem of separating Wi-Fi breathing signals in complex environments was solved, achieving high-precision breathing detection.
Patent Information
- Application Number
- CN202511645872.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-11
- Publication Date
- 2026-02-27
AI Technical Summary
In complex environments, Wi-Fi breathing signals are susceptible to various interferences, which can lead to signal masking or distortion, resulting in poor detection stability and accuracy. In particular, when the frequencies of breathing signals and interference signals are close, traditional time-domain or frequency-domain analysis methods are difficult to achieve effective separation, leading to loss or distortion of the breathing rhythm.
By acquiring Wi-Fi channel state information data and performing dual-scale decomposition, the interference signal separation model is trained using training data to determine model parameters, including frequency mask parameters and linear layer parameters. Dual-scale decomposition and masking networks are then used to adaptively distinguish and filter signals from interference.
It achieves high-fidelity separation of weak respiratory signals under complex micro-motion interference, improving the reliability and accuracy of non-contact vital sign monitoring and avoiding the loss or distortion of respiratory rhythm.
Smart Images
Figure CN121585288A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] One or more embodiments of the present disclosure relate to the technical field of wireless communication sensing, and in particular to a training method of an interference signal separation model and related equipment. BACKGROUND
[0002] With the development of wireless communication and sensing integration, using wireless signals to monitor human vital signs has gradually become a research hotspot. Compared with traditional contact sensors (such as electrocardiogram, capacitive sensor, piezoelectric sensor, etc.), wireless sensing technology can obtain physiological information such as human respiration and heartbeat under non-contact conditions, has the advantages of high comfort and not disturbing users, and is particularly suitable for long-term health monitoring, elderly care, disaster rescue and other scenarios.
[0003] Among them, wireless sensing based on wireless network communication technology (Wi-Fi) has been widely researched and applied due to its low cost, flexible deployment and high compatibility with existing wireless networks. Wi-Fi channel state information (CSI) can reflect the amplitude and phase changes of wireless signals in the propagation process in a fine-grained manner, and contains information about environmental disturbances and human micro-movements. Respiratory activity will cause the chest to fluctuate periodically, thereby introducing low-frequency, weak but regular changes in CSI, so CSI is widely used for respiration detection. Compared with millimeter wave radar, ultrasonic and other sensors, Wi-Fi devices do not require additional hardware modification and can directly implement vital sign detection in the existing wireless network environment, which makes them have great application potential in the fields of smart home, medical monitoring and public safety.
[0004] However, Wi-Fi signals in complex environments are often subject to multiple interferences, including scattering of objects in the environment, activities of non-target humans, and even slight random disturbances. These interferences will be superimposed into the CSI, causing the amplitude of the respiratory signal to be masked or distorted, greatly reducing the stability and accuracy of detection. Especially in the low-frequency band, when the respiratory signal and the interference signal have close frequencies, it is difficult to separate them by relying solely on traditional analysis methods in the time or frequency domain, which poses higher requirements for respiratory detection algorithms.
[0005] It should be noted that the above introduction to the technical background is only to facilitate a clear and complete description of the technical solutions of the present disclosure, and to facilitate the understanding of those skilled in the art. The above technical solutions cannot be considered as known to those skilled in the art merely because they are described in the background section of the present disclosure. SUMMARY
[0006] In view of this, the purpose of one or more embodiments of this disclosure is to provide a training method and related equipment for an interference signal separation model, so as to solve the problems raised in the background art.
[0007] To achieve the above objectives, one or more embodiments of this disclosure provide a training method for an interference signal separation model, the method comprising: Acquire Wi-Fi channel status information data; The Wi-Fi channel state information data is decomposed into training data using a dual-scale decomposition method; wherein the dual-scale decomposition includes interference scale decomposition and signal scale decomposition. The interference signal separation model is trained using the training data to determine the model parameters of the interference signal separation model; wherein, the model parameters include frequency mask parameters and linear layer parameters.
[0008] Furthermore, the acquisition of Wi-Fi channel status information data specifically includes: Collect Wi-Fi channel status information data, which includes phase information and amplitude information; The phase information and the amplitude information are analyzed respectively to obtain the phase timing sequence and the amplitude timing sequence; The phase timing sequence and the amplitude timing sequence are subjected to a reversible normalization process to obtain normalized phase timing sequences and amplitude timing sequences; the reversible normalization process specifically includes: Normalization:
[0009] Inverse normalization:
[0010] in, For the time series number The time series consists of one element, which is a phase time series or an amplitude time series; For the normalized time series sequence, the first... One element; and These are the maximum and minimum values of the time series, respectively.
[0011] Furthermore, the Wi-Fi channel state information data is decomposed into training data using a dual-scale method, specifically including: The normalized phase time series and amplitude time series are respectively subjected to interference scale decomposition to obtain the corresponding noise scale series. The noise scale sequence after interference scale decomposition is processed by signal scale decomposition to obtain the corresponding signal scale sequence. The noise scale sequence and the signal scale sequence are added together to obtain the training data.
[0012] Further, the step of performing interference scale decomposition on the normalized phase time series and amplitude time series respectively to obtain the corresponding noise scale sequence specifically includes: The normalized length is Time series Reshaping into dimensions matrix The normalized time series This includes the normalized phase time series and amplitude time series, specifically:
[0013] in, , The length of the noise scale decomposition. The number of noise decomposition scales. For matrix row index, For matrix Column index; For matrix Performing a discrete cosine transform yields the frequency domain features corresponding to each time series, specifically:
[0014] in, This is the result after the discrete cosine transform. Normalized time series A column of length short sequence fragments, For time indexing, The length of the short sequence fragment. For frequency index, These are the normalization coefficients; The frequency domain features The input data is fed into an interference scale masking network. The frequency domain signal of the input data is weighted using initially generated frequency domain masking parameters. Then, the weighted frequency domain signal is subjected to feature transformation using linear layer parameters, and the output is frequency domain information of the same length as the input data. ; Based on the inverse discrete cosine transform, the frequency domain information is transformed into a time domain sequence to obtain the noise scale sequence. Specifically:
[0015] in, This refers to the frequency domain information after filtering by the masking network. This is the noise scaling sequence after the inverse discrete cosine transform. For time indexing, The length of the short sequence fragment. For frequency index, This is the normalization coefficient.
[0016] Furthermore, the step of performing signal scale decomposition processing on the noise scale sequence after interference scale decomposition to obtain the corresponding signal scale sequence specifically includes: The lengths of the interference scale decomposed are respectively: Noise scale sequence Reshaping into dimensions matrix The noise scale sequence after interference scale decomposition. This includes the phase time series and amplitude time series after interference scale decomposition, specifically:
[0017] in, , The length of the signal scale decomposition. The number of signal decomposition scales. For matrix row index, For matrix Column index; For matrix Performing a discrete cosine transform yields the frequency domain features corresponding to each time series, specifically:
[0018] in, This is the result after the discrete cosine transform; The decomposed noise scale sequence A column of length short sequence fragments, For time indexing; The length of the short sequence fragment; Frequency index; These are the normalization coefficients; The frequency domain features The input signal is fed into a signal scaling masking network. The frequency domain signal of the input data is weighted using initially generated frequency domain masking parameters. Then, feature transformation is performed on the weighted frequency domain signal using linear layer parameters, outputting frequency domain information of the same length as the input data. ; Based on the inverse discrete cosine transform, the frequency domain information is... Transformed into a time-domain sequence, the signal scale sequence is obtained. Specifically:
[0019] in, This refers to the frequency domain information after filtering by the masking network. This is the signal scaling sequence after the inverse discrete cosine transform. For time indexing, The length of the short sequence fragment. For frequency index, This is the normalization coefficient.
[0020] Furthermore, the step of training the interference signal separation model using the training data to determine the model parameters of the interference signal separation model specifically includes: By restoring the loss The difference between the signal scale sequence after dual-scale decomposition and the normalized time series sequence is calculated using the following formula:
[0021] in, This is the normalized time series, i.e., the phase time series or the amplitude time series; It is a signal scale sequence. For time indexing; Through orthogonal loss The orthogonality value between the noise scale sequence and the signal scale sequence is calculated using the following formula:
[0022] in, It is a noise scale sequence. It is a signal scale sequence. For time indexing; The total loss function is constructed by weighted summing of the reduction loss and the orthogonality loss. The gradient descent algorithm is used to update the frequency mask parameters and the linear layer parameters according to the total loss function; The model parameters of the interference signal separation model are obtained through iterative training.
[0023] Based on the same inventive concept, one or more embodiments of this disclosure also provide a training system for an interference signal separation model, the system comprising: The acquisition module is configured to acquire Wi-Fi channel status information data. The decomposition module is configured to perform dual-scale decomposition on the Wi-Fi channel state information data to obtain training data; wherein the dual-scale decomposition includes interference scale decomposition and signal scale decomposition. The training module is configured to train the interference signal separation model using the training data to determine the model parameters of the interference signal separation model; wherein the model parameters include frequency mask parameters and linear layer parameters.
[0024] Based on the same inventive concept, one or more embodiments of this disclosure also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a training method for an interference signal separation model as described in any of the preceding claims.
[0025] Based on the same inventive concept, one or more embodiments of this disclosure also provide a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute any of the above-described methods for training an interference signal separation model.
[0026] Based on the same inventive concept, one or more embodiments of this disclosure also provide a computer program product, including one or more computer programs, which, when executed by one or more processors, implement the training method for any of the interference signal separation models described above.
[0027] As can be seen from the above, the interference signal separation model training method provided by one or more embodiments of this disclosure acquires Wi-Fi channel state information data, performs dual-scale decomposition on the Wi-Fi channel state information data to obtain training data, and uses the training data to train the interference signal separation model, thereby determining the model parameters of the interference signal separation model. This solves the technical problem that traditional methods struggle to separate weak respiratory signals from co-frequency or near-frequency interference with high fidelity under complex micro-motion interference. The technical solution of this disclosure achieves adaptive differentiation and filtering of signals and interference in the frequency domain through learnable intelligent masks, effectively solving the problem of poor reliability of non-contact vital sign monitoring caused by loss of respiratory rhythm or signal distortion.
[0028] The training system, electronic device, computer-readable storage medium, and computer program product for the interference signal separation model provided in this disclosure can all implement the steps of the training method for the interference signal separation model described above, and therefore also have the beneficial effects of the training method for the interference signal separation model described above. Attached Figure Description
[0029] To more clearly illustrate the technical solutions in one or more embodiments of this disclosure or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are only one or more embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0030] Figure 1 This is a flowchart illustrating a training method for an interference signal separation model according to one or more embodiments of this disclosure; Figure 2 This is a schematic diagram of the structure of a training system for an interference signal separation model according to one or more embodiments of this disclosure; Figure 3 This is a schematic diagram of the hardware structure of an electronic device according to one or more embodiments of this disclosure. Detailed Implementation
[0031] To make the objectives, technical solutions, and advantages of this disclosure clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.
[0032] It should be noted that, unless otherwise defined, the technical or scientific terms used in one or more embodiments of this disclosure should have the ordinary meaning understood by one of ordinary skill in the art to which this disclosure pertains. The terms "first," "second," and similar words used in one or more embodiments of this disclosure do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.
[0033] As described in the background section, the relevant technologies have the following problems: First, in complex environments, Wi-Fi breathing signals are easily affected by various interferences, resulting in signal masking or distortion, and poor detection stability and accuracy; Second, when the frequencies of breathing signals and interference signals are close, traditional time-domain or frequency-domain analysis methods are difficult to achieve effective separation, which can easily lead to loss or distortion of breathing rhythm.
[0034] Therefore, there is an urgent need to develop a technical solution for training an interference signal separation model to meet the pressing need for high-precision and robust separation of respiratory signals in complex interference scenarios and improve the reliability of non-contact vital sign monitoring.
[0035] Based on some implementations of this disclosure, a training method for an interference signal separation model is provided. In this method, Wi-Fi channel state information data is acquired, and the Wi-Fi channel state information data is decomposed into training data using a dual-scale method. The training data is then used to train the interference signal separation model, thereby determining the model parameters. This method enables the separation of noise and signal components in Wi-Fi CSI signals under micro-motion interference.
[0036] refer to Figure 1 This disclosure discloses a method for training an interference signal separation model according to one or more embodiments, comprising the following steps: S101. Obtain Wi-Fi channel state information data. The Wi-Fi channel state information data includes normalized phase timing sequences and amplitude timing sequences.
[0037] In this embodiment, the Wi-Fi channel state information is a technology that uses channel state information (CSI) in Wi-Fi signals to detect human activities (such as walking, breathing, etc.) and the state, position, and movement of other objects in the environment.
[0038] In this embodiment, obtaining Wi-Fi channel state information data specifically includes: S1011. Collect Wi-Fi channel status information data, wherein the Wi-Fi channel status information data includes phase information and amplitude information; In this embodiment, the present invention uses a CSI Tool built on an Intel 5300 network card to collect raw Wi-Fi Channel State Information (CSI) data through three receiving antennas. The collection time is 20 seconds, and the sampling rate is 100Hz, thereby constructing the raw dataset. The raw dataset contains a total of 19,600 time series sequences, including 17,000 normal breathing data and 2,600 breathing data containing micro-motion interference.
[0039] In this embodiment, the Wi-Fi channel status information data is collected: after the collection device is set up, the tester is positioned between the transmitting and receiving devices, swaying his upper body within a 5-centimeter range to the left and right while maintaining stable breathing.
[0040] S1012. The phase information and the amplitude information are analyzed respectively to obtain the phase timing sequence and the amplitude timing sequence, specifically as follows: Phase information: The phase information received by multiple antennas is subjected to conjugate multiplication to eliminate noise such as carrier frequency offset inherent in the device itself; and the processed phase information is then subjected to mean filtering and low-pass filtering in sequence to obtain a phase timing sequence with breathing characteristics. Amplitude information: The amplitude information is subjected to mean filtering and low-pass filtering in sequence to obtain an amplitude time series sequence with respiratory characteristics; S1013. Perform Reversible Normalization (RIN) processing on the phase time series and the amplitude time series to obtain normalized phase time series and amplitude time series, which are used for training and testing of the network model. The reversible normalization processing specifically includes: Normalization:
[0041] Inverse normalization:
[0042] in, For the time series number The time series consists of one element, which is either a phase time series or an amplitude time series. For the normalized time series sequence, the first... One element; and These are the maximum and minimum values of the time series, respectively, and need to be recorded after normalization.
[0043] In this embodiment, Reducible Normalization (RIN) is used to process the signal, ensuring that the network output signal can be completely restored to the input signal while normalizing it.
[0044] S102. Perform dual-scale decomposition on the Wi-Fi channel state information data to obtain training data.
[0045] In this embodiment, the dual-scale decomposition includes: Interference scale decomposition: used to capture low-frequency interference components with large amplitude fluctuations and frequencies close to the dominant respiratory frequency; Signal scaling: used to extract respiratory signal components with small amplitude but stable phase.
[0046] In this embodiment, the scale of interference scale decomposition is lower than that of signal scale decomposition, which is suitable for modeling co-channel interference with drastic amplitude fluctuations in the low-frequency band.
[0047] In this embodiment, the step of performing dual-scale decomposition on the Wi-Fi channel state information data to obtain training data specifically includes: S1021. Perform interference scale decomposition on the normalized phase time sequence and amplitude time sequence respectively to obtain the corresponding noise scale sequence.
[0048] The normalized time series (length is) Reshaping into dimensions matrix The normalized time series This includes the normalized phase time series and amplitude time series, specifically:
[0049] in, , The length of the noise scale decomposition. The number of noise decomposition scales. For matrix row index, For matrix Column indexes.
[0050] For matrix Perform Discrete Cosine Transform (DCT) to obtain the frequency domain features corresponding to each time series, specifically:
[0051] in, The result is after Discrete Cosine Transform (DCT). Normalized time series A column of length short sequence fragments, For time indexing, The length of the short sequence fragment. For frequency index, This is the normalization coefficient.
[0052] The frequency domain features The input data is fed into an interference scale masking network. The frequency domain signal of the input data is weighted using initially generated frequency domain masking parameters. Then, the weighted frequency domain signal is subjected to feature transformation using linear layer parameters, and the output is frequency domain information of the same length as the input data. .
[0053] In this embodiment, the masking network selectively filters out frequency domain features through a learnable frequency mask, suppressing interference components that are at or near the same frequency as breathing, while preserving respiratory rhythm features.
[0054] In this embodiment, the masking network performs masking calculations on the frequency domain features after Discrete Cosine Transform (DCT) using a masking layer and an earlier layer, thereby distinguishing between co-frequency interference and respiratory rhythm, and can perform iterative calculations.
[0055] Based on the inverse discrete cosine transform (IDCT), the frequency domain information is transformed into a time domain sequence to obtain the noise scale sequence. Specifically:
[0056] in, This refers to the frequency domain information after filtering by the masking network. This is the result after Inverse Discrete Cosine Transform (IDCT). For time indexing, The length of the short sequence fragment. For frequency index, This is the normalization coefficient.
[0057] S1022. Perform signal scale decomposition on the noise scale sequence after interference scale decomposition to obtain the corresponding signal scale sequence.
[0058] The lengths of the interference scale decomposed are respectively: Noise scale sequence Reshaping into dimensions matrix The noise scale sequence after interference scale decomposition. This includes the phase time series and amplitude time series after interference scale decomposition, specifically:
[0059] in, , The length of the signal scale decomposition. The number of signal decomposition scales. For matrix row index, For matrix Column indexes.
[0060] For matrix Perform Discrete Cosine Transform (DCT) to obtain the frequency domain features corresponding to each time series, specifically:
[0061] in, The result is after Discrete Cosine Transform (DCT). The decomposed noise scale sequence A column of length short sequence fragments, For time indexing; The length of the short sequence fragment; Frequency index; This is the normalization coefficient.
[0062] The frequency domain features The input signal is fed into a signal scaling masking network. The frequency domain signal of the input data is weighted using initially generated frequency domain masking parameters. Then, feature transformation is performed on the weighted frequency domain signal using linear layer parameters, outputting frequency domain information of the same length as the input data. .
[0063] Based on the Inverse Discrete Cosine Transform (IDCT), the frequency domain information is... Transformed into a time-domain sequence, the signal scale sequence is obtained. Specifically:
[0064] in, This refers to the frequency domain information after filtering by the masking network. This is the result after Inverse Discrete Cosine Transform (IDCT). For time indexing, The length of the short sequence fragment. For frequency index, This is the normalization coefficient.
[0065] S1023, the noise scale signal sequence and the signal scale sequence Add them together to get the training data.
[0066] In this embodiment, the lengths of the noise scale decomposition and the signal scale decomposition are changed. Different training effects can be obtained. After trying various parameter combinations, the best scale decomposition combination was found to be: noise scale decomposition length 20 and signal scale decomposition length 400. The two types of data interference data were respectively input into the system with the best separation effect parameter combination to obtain the restored signal and the separated respiratory signal, which effectively realized the separation of respiratory data in time series data under micro-motion interference.
[0067] S103. The interference signal separation model is trained using the training data to determine the model parameters of the interference signal separation model; wherein, the model parameters include frequency mask parameters and linear layer parameters.
[0068] In this embodiment, the initially generated frequency domain mask parameters and linear layer parameters do not have the ability to separate signals and require loss constraints for training. Therefore, this invention uses restoration loss and orthogonal loss to constrain network training.
[0069] In this embodiment, the step of training the restored signal to determine the model parameters specifically includes: By restoring the loss The difference between the signal scale sequence after dual-scale decomposition and the normalized time sequence is calculated to constrain the fidelity of the signal decomposition and reconstruction process. The calculation formula is as follows:
[0070] in, This is the normalized time series, i.e., the phase time series or the amplitude time series; It is a signal scale sequence. For time indexing.
[0071] Through orthogonal loss Calculate the noise scale sequence With signal scale sequence The orthogonality value is used to ensure that the two reconstructed sub-signals are as orthogonal as possible. Its calculation formula is:
[0072] in, It is a noise scale sequence. It is a signal scale sequence. For time indexing.
[0073] The reduction loss and orthogonality loss are weighted and summed to construct the total loss function. The coefficients before the two losses are the parameters that performed best in the test; The gradient descent algorithm is used to update the frequency mask parameters and the linear layer parameters according to the total loss function; through iterative training, the total loss is minimized to obtain the optimal model parameters of the interference signal separation model.
[0074] In this embodiment, the present invention, through dual-scale decomposition combined with a masking network, can achieve high-precision separation even when the frequency of micro-motion interference is close to that of the respiratory signal, thus avoiding loss or distortion of the respiratory rhythm.
[0075] It is understandable that this method can be executed by any device, equipment, platform, or cluster of devices with computing and processing capabilities.
[0076] In some embodiments, after training the interference signal separation model, the interference signal separation model can be used to separate the interference signal from the Wi-Fi channel state information data, thereby obtaining the Wi-Fi channel state information data after removing the interference signal for breathing detection, thus improving the accuracy of breathing detection.
[0077] It should be noted that the methods of one or more embodiments of this disclosure can be executed by a single device, such as a computer or server. The methods of this embodiment can also be applied in a distributed scenario, where multiple devices cooperate to complete the process. In such a distributed scenario, one of these devices may execute only one or more steps of the methods of one or more embodiments of this disclosure, and the multiple devices will interact with each other to complete the method described.
[0078] It should be noted that the above description pertains to specific embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than those shown in the embodiments and may still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.
[0079] Based on the same inventive concept, corresponding to any of the above embodiments, this disclosure also provides a training system for an interference signal separation model, such as... Figure 2 As shown, the above system includes: The acquisition module 201 is configured to acquire Wi-Fi channel status information data. The decomposition module 202 is configured to perform dual-scale decomposition on the Wi-Fi channel state information data to obtain training data; wherein, the dual-scale decomposition includes interference scale decomposition and signal scale decomposition. The training module 203 is configured to train the interference signal separation model using the training data to determine the model parameters of the interference signal separation model; wherein the model parameters include frequency mask parameters and linear layer parameters.
[0080] For ease of description, the above system is described by dividing it into various modules based on their functions. Of course, when implementing one or more embodiments of this disclosure, the functions of each module can be implemented in one or more software and / or hardware.
[0081] The system described above is used to implement the corresponding methods in the foregoing embodiments and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0082] Figure 3This embodiment illustrates a more specific hardware structure of an electronic device, which may include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, memory 1020, input / output interface 1030, and communication interface 1040 are interconnected internally via the bus 1050.
[0083] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this disclosure.
[0084] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 can store the operating system and other applications. When the technical solutions provided in the embodiments of this disclosure are implemented by software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.
[0085] The input / output interface 1030 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components within the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touchscreens, microphones, various sensors, etc., while output devices may include displays, speakers, vibrators, indicator lights, etc.
[0086] The communication interface 1040 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, Wi-Fi, Bluetooth, etc.).
[0087] Bus 1050 includes a pathway for transmitting information between various components of the device, such as processor 1010, memory 1020, input / output interface 1030, and communication interface 1040.
[0088] It should be noted that although the above-described device only shows the processor 1010, memory 1020, input / output interface 1030, communication interface 1040, and bus 1050, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this disclosure, and not necessarily all the components shown in the figures.
[0089] The electronic devices described above are used to implement the corresponding methods in the foregoing embodiments and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0090] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.
[0091] Based on the same inventive concept, corresponding to the training method for an interference signal separation model in any of the above embodiments, this disclosure also provides a computer program product, which includes one or more computer programs. In some embodiments, the one or more computer programs are executable by one or more processors to cause the one or more processors to execute the training method for an interference signal separation model. Corresponding to the execution entity for each step in each embodiment of the training method for an interference signal separation model, the processor executing the corresponding step may belong to the corresponding execution entity. The computer program product of the above embodiments is used to cause the processor to execute the training method for an interference signal separation model as described in any of the above embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0092] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this disclosure (including the claims) is limited to these examples; within the framework of this disclosure, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of one or more embodiments of this disclosure as described above, which are not provided in detail for the sake of brevity.
[0093] Additionally, to simplify the description and discussion, and to avoid obscuring one or more embodiments of this disclosure, the provided drawings may or may not show well-known power / ground connections to integrated circuit (IC) chips and other components. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring one or more embodiments of this disclosure, and this also takes into account the fact that the details of implementation of these block diagram apparatuses are highly dependent on the platform on which one or more embodiments of this disclosure will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuitry) are set forth to describe exemplary embodiments of this disclosure, it will be apparent to those skilled in the art that one or more embodiments of this disclosure may be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.
[0094] Although this disclosure has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.
[0095] This disclosure includes one or more embodiments intended to cover all such substitutions, modifications, and variations falling within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of this disclosure should be included within the scope of protection of this disclosure.
Claims
1. A method for training an interference signal separation model, characterized in that, The method comprises: obtaining wireless network communication technology Wi-Fi channel state information data; performing bi-scale decomposition on the Wi-Fi channel state information data to obtain training data; wherein the bi-scale decomposition comprises interference scale decomposition and signal scale decomposition; training an interference signal separation model using the training data to determine model parameters of the interference signal separation model; wherein the model parameters comprise frequency mask parameters and linear layer parameters.
2. The method of claim 1, wherein, The obtaining wireless network communication technology Wi-Fi channel state information data specifically comprises: collecting Wi-Fi channel state information data, which includes phase information and amplitude information; analyzing the phase information and the amplitude information respectively to obtain phase time sequence and amplitude time sequence; performing reversible normalization processing on the phase time sequence and the amplitude time sequence to obtain normalized phase time sequence and amplitude time sequence; the reversible normalization processing specifically comprises: normalization: inverse normalization: wherein, is the i-th element of the normalized timing sequence; is the i-th element of the timing sequence, which is either a phase timing sequence or an amplitude timing sequence; is the i-th element of the normalized timing sequence; is the i-th element of the normalized timing sequence; and are the maximum and minimum values of the timing sequence, respectively.
3. The method of claim 2, wherein, performing bi-scale decomposition on the Wi-Fi channel state information data to obtain training data, specifically comprising: performing interference scale decomposition processing on the normalized phase time sequence and amplitude time sequence respectively to obtain corresponding noise scale sequence; performing signal scale decomposition processing on the noise scale sequence after interference scale decomposition respectively to obtain corresponding signal scale sequence; adding the noise scale sequence and the signal scale sequence to obtain training data.
4. The method of claim 3, wherein, The performing interference scale decomposition processing on the normalized phase time sequence and amplitude time sequence respectively to obtain corresponding noise scale sequence specifically comprises: The normalized length is Time series Reshaping into dimensions matrix The normalized time series This includes the normalized phase time series and amplitude time series, specifically: wherein , is a noise scale decomposition length, is a number of noise decomposition scales, is a row index of the matrix , is a column index of the matrix ; On the matrix Discrete cosine transform is performed to obtain the frequency domain features corresponding to each time sequence, specifically: wherein, is the result of a discrete cosine transform, is a normalized time series of a certain column, i.e. a short sequence segment of length is a time index, is the length of the short sequence segment, is a frequency index, is a normalization coefficient; The frequency domain features are input to an interference scale mask network, the frequency domain signals of the input data are weighted by the initial generated frequency domain mask parameters, and the weighted frequency domain signals are feature transformed by linear layer parameters, and frequency domain information with the same length as the input data is output ; based on a discrete cosine inverse transform, transform the frequency domain information into a time domain sequence to obtain a noise scale sequence , specifically: wherein, is the frequency domain information filtered by the mask network, is the noise scale sequence after inverse discrete cosine transform, is the time index, is the length of the short sequence segment, is the frequency index, is the normalization coefficient.
5. The method of claim 3, wherein, The performing signal scale decomposition processing on the noise scale sequence after interference scale decomposition respectively to obtain corresponding signal scale sequence specifically comprises: The lengths of the interference scale decomposed are respectively: Noise scale sequence Reshaping into dimensions matrix The noise scale sequence after interference scale decomposition. This includes the phase time series and amplitude time series after interference scale decomposition, specifically: wherein , is a signal decomposition scale length, is a signal decomposition scale number, is a row index of the matrix , is a column index of the matrix ; On the matrix Discrete cosine transform is performed to obtain the frequency domain features corresponding to each time sequence, specifically: wherein is the result of the discrete cosine transform; is the decomposed noise scale sequence is a certain column of the decomposed noise scale sequence, i.e. a short sequence segment of length is the time index; is the length of the short sequence segment; is the frequency index; is the normalization coefficient; The frequency domain features are input into a signal scale mask network, the frequency domain signals of the input data are weighted by the initial generated frequency domain mask parameters, the weighted frequency domain signals are feature transformed by linear layer parameters, and frequency domain information with the same length as the input data is output ; based on an inverse discrete cosine transform, the frequency domain information is converted into a time domain sequence to obtain a signal dimension sequence , specifically: wherein, is the frequency domain information filtered by the mask network, is the signal scale sequence after inverse discrete cosine transform, is the time index, is the length of the short sequence segment, is the frequency index, is the normalization coefficient.
6. The method of claim 1, wherein, The training an interference signal separation model using the training data to determine model parameters of the interference signal separation model specifically comprises: by reducing the loss The difference value between the signal scale sequence after the two-scale decomposition and the normalized time sequence is calculated, and the calculation formula is: wherein is the normalized time series, i.e. the phase time series or the amplitude time series; is the signal dimension sequence, is the time index; By orthogonal loss The orthogonal value of the noise scale sequence and the signal scale sequence is calculated, and the calculation formula is: wherein is a sequence of noise scales, is a sequence of signal scales, is a time index; weighting and summing the restoration loss and the orthogonal loss to construct a total loss function; updating the frequency mask parameters and the linear layer parameters according to the total loss function by using a gradient descent algorithm; obtaining model parameters of the interference signal separation model through iterative training.
7. A training system of an interference signal separation model, characterized by, The system comprises: an acquisition module configured to obtain wireless network communication technology Wi-Fi channel state information data; a decomposition module configured to perform bi-scale decomposition on the Wi-Fi channel state information data to obtain training data; wherein the bi-scale decomposition comprises interference scale decomposition and signal scale decomposition; a training module configured to train an interference signal separation model using the training data to determine model parameters of the interference signal separation model; wherein the model parameters comprise frequency mask parameters and linear layer parameters.
8. An electronic device comprising a memory, a processor, and a computer program stored on the memory and run by the processor, characterized in that, The processor implements the method of any one of claims 1 to 6 when executing the computer program.
9. A non-transitory computer-readable storage medium, comprising: The non-transitory computer readable storage medium stores computer instructions for causing the computer to execute the method of any one of claims 1 to 6.
10. A computer program product, characterised in that, comprising one or more computer programs that, when executed by one or more processors, implement the method of any one of claims 1 to 6.