Building internal communication electric signal processing method

Abstract

The invention provides a building internal communication electric signal processing method, and relates to the technical field of signal processing, and the method specifically comprises the steps: making a building internal communication electric signal data set, and constructing a propagation response model; a curvature adjustment coefficient is introduced, a non-Euclidean manifold embedded structure is designed, a non-Euclidean convolution encoder and a non-Euclidean convolution decoder are trained, and encoding and reconstruction of signals are achieved; obtaining a disturbance response relation matrix through modal disturbance analysis, calculating a fusion weight in combination with a disturbance direction sensitivity matrix and a modal self-suppression factor, and carrying out fusion optimization on the multi-modal signal; meanwhile, a building internal communication electric signal processing model is provided, accurate expression and enhancement of communication signals under a complex structure are achieved, and the stability and the anti-interference capacity of the building internal communication electric signals are effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of signal processing, and in particular relates to a method for processing electric signals for internal communication in a building. Background Art

[0002] With the rapid development of emerging technologies such as smart buildings, IoT communications, and indoor positioning, wireless communication systems within buildings are increasingly responsible for real-time data transmission and high-reliability connections. In application scenarios such as multi-source perception fusion, intelligent security, and high-density terminal deployment, accurately acquiring and effectively processing communication electrical signals within buildings has become a key factor in improving system performance and service quality. However, the closed nature of the building environment, multipath effects, channel obstruction, and structural complexity pose severe challenges to traditional communication signal modeling and analysis methods.

[0003] Existing modeling and enhancement methods for internal communication electrical signals in buildings mostly rely on rule-based modeling, static path loss models, or traditional graph structure propagation mechanisms. These methods are difficult to adapt to practical problems such as complex building structures, multipath interference, and frequency drift, resulting in limited communication quality prediction and signal enhancement effects. For example, patent CN107370708A proposes a static propagation loss modeling method based on a building path graph, which achieves signal coverage prediction by presetting wall attenuation parameters. However, this method cannot dynamically adjust to environmental changes, and does not establish a coupling mechanism between path propagation and environmental parameters. This makes it difficult to meet the communication prediction accuracy requirements under changes in building structures or dynamic node deployment. Summary of the Invention

[0004] The present invention provides a method for processing internal communication electrical signals in a building, aiming to construct a building communication signal processing framework with dynamic modeling and multimodal fusion capabilities. The method obtains communication electrical signal data inside the building; establishes a path propagation loss strategy and a propagation power mechanism based on the communication electrical signal data inside the building, and extracts intermediate propagation state results reflecting the modulation characteristics of the building structure; utilizes a non-Euclidean convolution strategy to embed the communication electrical signals inside the building into a non-Euclidean convolution encoder and decoder structure, embeds and maps the communication electrical signals inside the building, and trains them to generate reconstructed modal signals with structural distribution expression capabilities, thereby improving the signal representation accuracy in a nonlinear propagation environment; in the modal perturbation modeling stage, by introducing perturbations to each modal signal and observing the response changes of other modes, a modal perturbation response relationship matrix and a perturbation direction sensitivity matrix are constructed, and a modal self-suppression factor is introduced to adjust the interference coupling between modals; multiple modal signals are weightedly fused according to the fusion weight coefficient to obtain the fused internal communication electrical signals of the building; and finally, through the internal communication electrical signal processing model of the building, the optimization and enhancement processing of the communication electrical signals in complex building scenes is completed.

[0005] In order to achieve the above-mentioned object, the present invention provides the following technical solution: a method for processing electrical signals for internal communication in a building, comprising the following specific steps: S1. Acquire communication electrical signal data in a building's internal communication environment, including transmitted signals, reflected signals, penetrated signals, and building structure-related information, and construct a dataset containing propagation paths, frequency band parameters, building material properties, and channel responses. S2. Designing building response parameters based on the communication electrical signal data in the building's internal communication environment and combining them with frequency modulation loss terms to establish a path propagation loss strategy, perform building propagation response modeling, and obtain intermediate propagation state results that describe signal modulation characteristics; S3. Based on the intermediate propagation state result and the communication electrical signal data, a curvature adjustment coefficient is introduced to perform non-Euclidean manifold embedding and construct a non-Euclidean convolution strategy; a non-Euclidean convolution encoder and decoder are trained based on the signal reconstruction error to obtain network weight parameters; the trained encoder and decoder are used to encode and reconstruct the embedded signal, respectively, to obtain multiple reconstructed modal signals; S4. Introduce disturbances to each reconstructed modal signal, measure the response changes of other modal signals, calculate the disturbance impact value between modal signals, and construct a modal disturbance response relationship matrix; S5. Constructing a disturbance direction sensitivity matrix based on the modal disturbance response relationship matrix, introducing a modal self-inhibition factor, calculating a fusion weight coefficient, and fusing multiple reconstructed modal signals to obtain a fused building internal communication electrical signal. S6. Construct a building internal communication electric signal processing model, input the building internal communication electric signal data set, go through steps S2 to S5 in sequence, and perform iterative training until convergence to complete the optimization processing of the building internal communication electric signal.

[0006] Preferably, in step S1, for the construction of the building internal communication electric signal data set, the original building internal communication electric signal data obtained is first preprocessed, including signal amplitude normalization, time series alignment, noise filtering and frequency domain conversion, to extract representative amplitude, phase and frequency short-term features, and construct a standardized spatiotemporal feature sequence; after completing the data preprocessing, further collect building structure information related to signal propagation, including building materials, wall thickness, floor layout, door and window distribution and space partition parameters, and extract spatial structure information with the help of architectural drawings, BIM models or laser scanning equipment; at the same time, combine the communication protocol parameters and hardware settings to extract frequency band information, including the transmission frequency The system obtains historical communication records and environmental change parameters, and extracts the channel state evolution characteristics. Based on multi-source information, the starting point, end point and intermediate traversal medium of the propagation path are marked according to the three-dimensional spatial position of the transmitting and receiving nodes, and the walls, boundaries and openings traversed in the path are identified, and a mapping relationship between the path topology and the building structure is established. Finally, the signal preprocessing characteristics, building structure information, frequency band parameters, historical communication data and the propagation path graph structure are multi-dimensionally encoded and integrated according to time, space and propagation attributes to generate a structured format of internal building communication electrical signal datasets, providing a unified standard data input for subsequent propagation modeling and multimodal processing.

[0007] Preferably, the communication electrical signals inside buildings have the characteristics of complex path structure, non-stationary channel state, and dynamic coupling of easy frequency response deviation during the propagation process. The signals are affected by spatial occlusion, multi-path reflection and building material properties, and show obvious time variability and structural sensitivity. Based on such characteristics, the present invention proposes a response modeling mechanism that integrates environmental parameters, frequency modulation and propagation path topology information to construct a propagation state expression model with structural perception ability and dynamic adaptability of propagation, which supports communication modeling, signal enhancement and modal fusion processing tasks in complex building interior scenarios.

[0008] Furthermore, in step S2, the specific process of building propagation response modeling includes: S21. Couple the environmental variables with the depth of the signal propagation path in the building structure, use the nonlinear mapping mechanism to model their joint effects, and generate building response parameters that can dynamically reflect the internal response characteristics of the building. The calculation formula is: Where E is the building response parameter, θ env is the environmental variable, D is the depth of the signal propagation path inside the building; S22. Define the frequency modulation loss term based on the quadratic function relationship between the frequency adjustment coefficient of the internal communication signal in the building and the frequency deviation of the center frequency relative to the actual frequency collected by the communication path node. The mathematical model is: Where F is the frequency modulation loss term, β F is the frequency adjustment coefficient of the communication signal inside the building, f0 is the center frequency, f c The actual frequency collected by each path node during the communication process; S23. Construct a path propagation loss strategy based on the frequency modulation loss term, combined with the signal propagation path depth inside the building and the relationship between the building response parameters; S24. Based on the path propagation loss strategy and taking into account the path complexity constraint, a propagation power mechanism is constructed: Where, P eff (t) is the propagation power mechanism, A(t) is the transmission power of the communication signal at time t, R is the path complexity factor, and L(D) is the path propagation loss strategy; S25. By constructing a dynamic mapping mechanism between the path loss weight adjustment factor and the propagation power mechanism, the intermediate propagation state result for the spatial node pair is obtained. The mathematical model is: w ij =exp(-λ·L(D ij )); S ij =P eff (t)·w ij ; Where D ij represents the propagation path depth between node i and node j, w ij is the path loss weight, λ is the path loss weight adjustment factor, L(D ij ) is the path propagation loss strategy between node i and node j, S ij is the state matrix value propagated between nodes.

[0009] Furthermore, in step S2, in the process of building propagation response modeling, first, the environmental variables are coupled with the depth of the propagation path of the internal communication electrical signal in the building structure to obtain the building response parameters. This parameter can dynamically reflect the degree of structural environmental influence on the internal communication electrical signal in the building at different path depths, providing an important basis for subsequent propagation modulation modeling. Secondly, considering the energy loss problem caused by frequency drift of the internal communication electrical signal in the building in different paths, the frequency adjustment coefficient is introduced, and the frequency modulation loss term is calculated by the deviation between the center frequency and the acquisition frequency of each path node. This loss term can accurately describe the impact of frequency band offset on the propagation quality of the internal communication electrical signal in the building, and the sensitivity is controlled by the frequency adjustment coefficient. Then, according to the relationship between the propagation path depth of the internal communication electrical signal in the building and the building structure, a node topology structure is constructed, and a path propagation loss strategy is established to represent the connection between each path. Furthermore, by combining the building response parameters, frequency modulation loss terms and propagation path depth, and combining the transmission power and path complexity factor of the internal communication electrical signals at different times in the building, a propagation power mechanism is established. This mechanism can dynamically adjust the propagation intensity of the internal communication electrical signals on different paths inside the building. Finally, based on the constructed propagation power mechanism and the path loss expression weight function, the inter-node propagation state matrix values ​​used to represent the propagation state are generated through the dynamic mapping mechanism of the propagation path depth and the path loss weight adjustment factor, thereby obtaining the intermediate propagation state expression result describing the propagation process of the internal communication signal in the building. The overall modeling strategy forms a dynamically adjustable building propagation modeling mechanism by combining the propagation path depth, frequency adjustment coefficient and building environment parameters, providing high-quality propagation state prior information for subsequent non-Euclidean embedding, non-Euclidean encoding and decoding processing and modal perturbation modeling.

[0010] Preferably, the internal communication electrical signals of a building have significant characteristics of non-uniform structural connections, spatial topological distortion, and multi-path interference during the propagation process. The distribution of signals in different building areas is limited by local geometric relationships and propagation path folding effects, making it difficult to achieve continuous and distinguishable structural representation in Euclidean space. Based on the non-Euclidean geometric characteristics of this propagation space, the present invention adopts a non-Euclidean convolutional coding structure to perform geometric embedding and structured modeling on the internal communication electrical signals of the building to support subsequent disturbance response extraction and multimodal collaborative modeling tasks.

[0011] Furthermore, in step S3, the specific steps of constructing a non-Euclidean convolution strategy and training a non-Euclidean encoder and decoder include: S31. Construct an embedding mapping function for mapping the input signal to a non-Euclidean space, and establish a geometric metric model based on the structural mapping relationship between the embedded samples, thereby obtaining the geodesic relationship between the samples in the embedded space. The mathematical model is: Among them, x i is the i-th input signal, z i is the embedding vector representation of the i-th input signal, φ(·) is the embedding mapping function, σ emb is the nonlinear embedding mapping function, W emb is the weight of the nonlinear embedding mapping function, b emb is the bias of the nonlinear embedding mapping function, n is the embedding dimension, is an n-dimensional Euclidean space, → is a function mapping symbol, is the target non-Euclidean manifold space; S32. By introducing a tensor quantization mechanism that reflects the logarithmic mapping relationship between embedded points and combining it with a geometric distance metric with a local curvature adjustment coefficient, we construct an adjacency weight function for describing the local structural relationship of the embedded space: Among them, K(z i , z j ) is the adjacency weight function, Indicates from z j to z i The logarithmic mapping tensor of κ i is point z i The curvature adjustment coefficient at , τ is the bandwidth; S33. Generate local perception graph structure based on adjacency weight function and construct non-Euclidean convolution strategy for embedding spatial structure. The mathematical model is: in, is the neighborhood of node i, σ is the differentiable processing module of nonlinear transformation, l is the lth layer in the non-Euclidean convolution strategy, W (l) is the weight matrix of the lth layer, Neighborhood The original feature map representation of the node at layer l, is the convolution output feature; S34. By introducing the convolution strategy of non-Euclidean space into the encoder and decoder structures, a non-Euclidean convolution encoder and decoder structure adapted to the embedded manifold space is constructed. The reconstruction error is used as the target loss function to train the non-Euclidean convolution encoder and decoder structure. The mathematical model is: Among them, N is the number of training samples, Enc(x i ) is the encoder function, Dec(·) is the decoder function, Dec(Enc(·)) is the non-Euclidean convolution encoder and decoder function, L recis the target loss function; S35. Encode and reconstruct the input signal based on the trained non-Euclidean convolutional encoder and decoder structure to obtain multiple reconstructed modal signals. The mathematical model is: z i =Enc(x i ); Where, is the reconstruction result of the modal signal, z m represents the mth reconstructed modal signal.

[0012] Furthermore, in step S3, for the embedding and reconstruction method of the non-Euclidean convolutional coding structure, first, based on the intermediate propagation state results and the internal communication electrical signal data of the building, an embedding mapping function is constructed to map the internal communication electrical signal data of the building from the original space to the non-Euclidean manifold space, so as to capture the structural distribution characteristics of the internal communication electrical signal data of the building in the geometric space; secondly, a tensor logarithmic mapping mechanism is introduced, and the curvature adjustment coefficient and the bandwidth parameter are combined to construct an adjacency weight function to measure the local geometric proximity relationship between signals, so as to effectively deal with the connection degradation problem under the occlusion of the building structure; then, a local perception map is constructed based on the adjacency weight, and a non-Euclidean convolution is designed. The strategy is used to extract and aggregate the features of the internal communication electrical signals of the building layer by layer, thereby enhancing the expression ability in irregular space. Then, the convolution strategy is embedded in the autoencoder structure to construct the encoder and decoder modules, and the reconstruction error is used as the training target to improve the encoding stability and restoration accuracy of the data features of the internal communication electrical signals of the building. Finally, the trained encoding structure is used to reconstruct the internal communication electrical signal data of the building to obtain a clear and separable embedding representation, which provides a stable structural prior for subsequent modal disturbance modeling and multimodal fusion, and can significantly improve the expression integrity and reconstruction reliability of the internal communication electrical signals of the building in the complex spatial environment inside the building.

[0013] Preferably, the internal communication electrical signals of a building exhibit the characteristics of asymmetric modal coupling structure, fluctuation of interference response with changes in spatial structure, and significant differences in frequency range and path penetration characteristics between different modes during the multimodal propagation process. The interaction between modes is driven by disturbance conditions and changes, showing unstable and nonlinear coupling characteristics. Based on the structural properties of this type of dynamic response relationship, the present invention adopts a disturbance-driven modal response mapping mechanism to model the internal communication electrical signals of a building, so as to realize the relationship expression between multimodal signals and support the weight calculation in the subsequent fusion strategy.

[0014] Furthermore, in step S4, the process of constructing the modal disturbance response relationship matrix includes: S41. Construct a multimodal signal set based on the reconstructed multiple modal signals, and add disturbance to each modal signal in the set to obtain a multimodal signal set after adding disturbance. The mathematical model is: z′ m =z m +δ m ; Where z m represents the original signal of the mth mode, δ m is the introduced disturbance, z′ m is the modal signal after disturbance; S42, through the modal input z before disturbance m and the perturbed modal input z′ m , a response difference mapping mechanism for the modal pair is established, and then the disturbance impact value between the modes is accurately quantified. The mathematical model is: ΔR m,n =R n (z′ m )-R n (z m ); Among them, R n (·) represents the response function of mode n, ΔR m，n Represents the disturbance effect of mode m on mode n; S43. The disturbance influence values ​​between the modes are matrix-integrated according to the index position to construct a unique modal disturbance response relationship matrix. The mathematical model is: R=[ΔR m,n ] M×M ; Where R represents the inter-modal disturbance response matrix, and M represents the total number of modes.

[0015] Furthermore, in step S4, a response difference mapping mechanism is constructed based on the reconstructed multiple modal building internal communication electrical signals; first, disturbances are injected into the reconstructed multiple modal building internal communication electrical signals respectively, and their influence on the reconstructed output of other modes is observed, so as to quantify the disturbance transmission relationship between modes and capture the mutual dependence of different modes under the changes of building structure; secondly, the disturbance response values ​​between all modes are matrixed and integrated in the index order to construct a modal disturbance response relationship matrix, which reflects the interference connection and dynamic linkage characteristics between different modal communication electrical signals inside the building, and solves the problem of unquantifiable interaction relationship between modalities in the fusion process; finally, the response relationship matrix is ​​used as the basis for calculating the modal directional sensitivity and modal self-inhibition factor, which is used to support the dynamic allocation strategy of fusion weights, improve the accuracy and adaptability of the fusion link, enhance the controllability and robustness of the fusion modeling of multi-modal building internal communication electrical signals in building scenes, and provide structured weight support for subsequent modal building internal communication electrical signal enhancement and perception tasks.

[0016] Preferably, the internal communication electrical signals of the building exhibit the characteristics of dynamic fluctuations in modal response stability with environmental changes, inconsistent influence weights of each mode on the fusion result under structural occlusion, frequency offset and path distribution during the multimodal fusion process, and the actual contribution of different modes to the fusion effect is time-varying and structurally dependent. Based on the characteristics of uneven modal weight response and dynamic adjustment driven by the environment, the present invention adopts a weight calculation mechanism based on the disturbance response relationship to perform multimodal fusion processing on the internal communication electrical signals of the building, so as to realize adaptive distribution of modal contributions and improve the stability and expression ability of fusion modeling.

[0017] Furthermore, in step S5, the specific steps of fusion processing of the building internal communication electrical signals include: S51. By normalizing and calculating the relative intensity ratio of the inter-modal disturbance response values ​​in a specific modal direction, a directional sensitivity matrix is ​​constructed to characterize the distribution characteristics of the inter-modal directional interference effect. The mathematical model is: Where S m，n Indicates the directional disturbance sensitivity coefficient of mode m to mode n, ΔR m，n is the disturbance impact value; S52. Perform a mean inverse measurement on the response strength of each mode in the directional sensitivity matrix and construct a modal self-suppression factor to measure the adaptability of the mode to external disturbances. The mathematical model is: Where, α m represents the self-inhibition factor of mode m; S53. Normalize the self-inhibition factors of each modality and construct a fusion weight coefficient to measure the relative contribution of multimodal information. The mathematical model is: Where w m Represents the fusion weight coefficient of the mth modality; S54. Perform weighted combination based on each reconstructed modal signal and its corresponding fusion weight factor to obtain a fused building internal communication electrical signal. The mathematical model is: Where Z represents the internal communication electrical signal of the building after fusion processing.

[0018] Furthermore, in step S5, the multimodal building internal communication electrical signals are weightedly fused according to the disturbance response relationship between the modes to obtain the fusion output result; first, according to the disturbance response value, a disturbance direction sensitivity matrix is ​​constructed to describe the disturbance direction and intensity between the modes, solving the problem that the influence path between the modes is difficult to model under the change of building structure; secondly, the modal self-inhibition factor is introduced to measure the response suppression ability of each mode to its own disturbance, which serves as a reference indicator for evaluating its stability, thereby addressing the problem of strong volatility and large noise interference of the building internal communication electrical signals in specific modes; then, the disturbance direction sensitivity matrix and the modal self-inhibition factor are combined to calculate the final fusion weight coefficient, which comprehensively reflects the contribution of each mode to the building internal communication electrical signals in the multimodal fusion process, and improves the accuracy and interpretability of the fusion strategy; finally, based on the fusion weight, the multiple modal signal representations are weightedly fused to obtain the fused building internal communication electrical signals. This fusion method effectively improves the integration capability and stability of internal communication electrical signals in buildings under multimodal conditions through a response structure-driven weight distribution mechanism, and enhances the signal expression consistency and fusion robustness in complex building scenes.

[0019] Preferably, in step S6, for the optimization processing of the building's internal communication electric signal, a building's internal communication electric signal processing model is constructed, and propagation response modeling, non-Euclidean coding reconstruction, modal disturbance analysis and multimodal fusion processing are performed in sequence; first, a building's internal communication electric signal data set is input, and in the propagation response modeling stage, a path propagation loss strategy and a propagation power mechanism are constructed according to the propagation path parameters and building structure information of the building's internal communication electric signal, and an intermediate propagation state result for describing the modulation characteristics of the building structure is obtained; then, in the non-Euclidean coding and decoding stage, an embedding mapping function and an adjacency weight function are constructed based on the intermediate propagation state result, a non-Euclidean convolution strategy is established and embedded into the non-Euclidean convolution encoder and decoder structure, and the encoder and decoder are trained with the signal reconstruction error as the optimization target to obtain multiple modal signals after reconstruction; then In the modal disturbance analysis stage, disturbances are introduced into the internal communication electrical signals of each modal building respectively, and their influence on the response changes of other modal responses is observed. The inter-modal disturbance influence value is calculated, and the modal disturbance response relationship matrix is ​​constructed. Then, in the fusion processing stage, according to the modal disturbance response relationship matrix, the disturbance direction sensitivity and the modal self-inhibition factor are calculated, and the fusion weight coefficient is generated. The weighted fusion of multiple modal internal communication electrical signals of the building is performed to obtain the fused internal communication electrical signal of the building. By integrating the above stages end-to-end and combining iterative training mechanism with structured optimization strategy, the model can realize dynamic adjustment of propagation state modeling, autoencoding reconstruction and multimodal fusion strategy according to changes in the building environment, and finally output the enhanced and optimized internal communication electrical signal of the building, effectively improving the stability and availability of communication signals in complex building environments.

[0020] Compared with the existing technology, the present invention constructs a communication electric signal modeling scheme that integrates building structure information, propagation path parameters and frequency band characteristics, proposes a path propagation loss strategy, introduces building response parameters and frequency modulation loss terms, and realizes dynamic modeling of the propagation state of internal communication electric signals in complex building structures, which significantly enhances the expression ability of the internal communication electric signal model of the building environment for the heterogeneity and multipath effects of the building; in the coding and reconstruction stage of the internal communication electric signal of the building, a non-Euclidean convolution strategy based on the curvature adjustment coefficient is constructed, and the embedding and reconstruction of the internal communication electric signal of the building are completed by embedding mapping function and non-Euclidean convolution encoder and decoder structure, which improves the modeling efficiency. The model has the ability to extract the spatial structural characteristics of internal communication electrical signals in buildings under nonlinear propagation conditions; in the modal disturbance modeling stage, by introducing disturbances to each modal signal and measuring the response changes of other modes, the modal disturbance response relationship matrix and the directional sensitivity matrix are constructed, and the modal self-suppression factor is introduced to adjust the response adaptability between modes, which effectively solves the problem of unstable fusion of multi-modal signals due to mutual interference; combined with the fusion weight coefficient, multiple modal signals are weightedly fused, and finally the accurate modeling and robust fusion of multi-source internal communication electrical signals in complex building communication environments are achieved, which significantly improves the expression accuracy and fusion robustness of the internal communication electrical signal processing system of buildings. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 The present invention provides a flow chart of a method for processing building internal communication signals.

[0022] Figure 2 It is a structural diagram of the building propagation response modeling provided by the present invention.

[0023] Figure 3 This is a structural diagram of the non-Euclidean embedding and non-Euclidean convolution encoder and decoder training provided by the present invention.

[0024] Figure 4 It is a structural diagram of the modeling of the modal disturbance response relationship provided by the present invention.

[0025] Figure 5 This is a structural diagram of the fusion processing of internal communication electrical signals in a building provided by the present invention.

[0026] Figure 6 This is a comparison diagram of the optimized building internal communication electrical signal provided by the present invention and the original building internal communication electrical signal. DETAILED DESCRIPTION

[0027] The present invention provides a method for processing building internal communication electrical signals, aiming to construct a building communication signal processing framework with dynamic modeling and multimodal fusion capabilities. The method obtains communication electrical signal data from within a building; based on this data, a path propagation loss strategy and a propagation power mechanism are established to extract intermediate propagation state results that reflect the modulation characteristics of the building structure; a non-Euclidean convolution strategy is embedded into a non-Euclidean convolution encoder and decoder structure to embed, map, and train the internal communication electrical signals, generating reconstructed modal signals with structural distribution expression capabilities to improve signal representation accuracy in nonlinear propagation environments; in the modal perturbation modeling stage, a modal perturbation response relationship matrix and a perturbation direction sensitivity matrix are constructed by introducing perturbations to each modal signal and observing the response changes of other modes; a modal self-suppression factor is introduced to regulate inter-modal interference coupling; multiple modal signals are weightedly fused according to fusion weight coefficients to obtain a fused internal communication electrical signal; and finally, through the internal communication electrical signal processing model, optimization and enhancement of communication electrical signals in complex building scenarios are completed.

[0028] See Figure 1 As shown, a method for processing electrical signals for internal communication in a building in an embodiment of the present application has the following specific steps.

[0029] S1. Acquire communication electrical signal data in a building's internal communication environment. The data includes transmitted signals, reflected signals, penetrated signals, and information related to the building structure. Construct a dataset containing propagation paths, frequency band parameters, building material properties, and channel responses.

[0030] Furthermore, in step S1, for the internal communication electrical signal dataset of the building, a multi-channel wireless signal acquisition device is used to perform the communication electrical signal data acquisition task in the actual building structure. The types of internal communication electrical signals collected in the building include transmitted signals, reflected signals and penetrated signals. The communication protocols include Wi-Fi, NB-IoT and 5G indoor communication systems. The signals are represented in complex form and saved in binary format. The building structure covers concrete walls, brick structures, glass curtain walls and metal partitions. The interference channel includes electromagnetic equipment interference, structural occlusion interference and background radiation interference. The acquisition area is the corridor area, underground space, wall mezzanine and closed room. The signal-to-noise ratio range is set to 10dB to 25dB. After the acquisition is completed, the original internal communication electrical signal of the building is subjected to data preprocessing operations, including time alignment, amplitude normalization, high-frequency noise filtering, frequency domain transformation and window function envelope processing, to improve the signal structure consistency and feature stability. The sampling rate is set to 20MHz, the quantization bit width is 16 bits, the sampling time of each signal segment is 2 milliseconds, and the number of sampling points is 4×10 4 , the total number of communication signal samples is 2×10 5Each sample is accompanied by annotated fields, including propagation path category, spatial structure type, and interference source type, ultimately constructing a building internal communication electrical signal dataset with structural integrity, clear labels, and parameter consistency.

[0031] S2. Based on the communication electrical signal data in the internal communication environment of the building, the building response parameters are designed and combined with the frequency modulation loss term to establish a path propagation loss strategy, perform building propagation response modeling, and obtain an intermediate propagation state result that describes the signal modulation characteristics.

[0032] Furthermore, in step S2, the building propagation response is modeled, and the process is as follows: Figure 2 As shown in Figure 2, the specific steps of building propagation response modeling are as follows.

[0033] S21. Couple the environmental variables with the depth of the signal propagation path in the building structure, use the nonlinear mapping mechanism to model their joint effects, and generate building response parameters that can dynamically reflect the internal response characteristics of the building. The calculation formula is: Among them, E is the building response parameter, which quantifies the response intensity of the building structure to the signal propagation process, θ env is the environmental variable, D is the depth of the signal propagation path inside the building, in meters (m); In this embodiment, the environment variable θ env The mathematical model is: θ env =θ0+γ1·T+γ2·H; Wherein, θ0 is the basic response constant, which sets the initial response value of the building under standard environmental conditions. Its value range is [0.5, 1.5]. In this embodiment, it is set to 0.8. γ1 is the temperature adjustment coefficient, which indicates the influence of each 1°C increase in temperature on the gain of the building response factor. Its unit is ( / °C). Its value range is [0.01, 0.05]. In this embodiment, it is set to 0.02. γ2 is the humidity adjustment coefficient, which indicates the influence of each 1% increase in humidity on the gain of the response factor. Its unit is ( / %). Its value range is [0.001, 0.01]. In this embodiment, it is set to 0.005. T is the real-time temperature inside the building. In this embodiment, T is set to 26. Its unit is Celsius (°C). h is the real-time relative humidity inside the building. In this embodiment, H is set to 60. Its unit is percentage (%). In this embodiment, the propagation path depth D inside the building is obtained by the three-dimensional space distance calculation formula: Among them, (x i ,y i , z i) is the spatial coordinate of the i-th communication node, (x j ,y j , z j ) is the spatial coordinate of the jth communication node.

[0034] S22. Define the frequency modulation loss term based on the quadratic function relationship between the frequency adjustment coefficient of the internal communication signal in the building and the frequency deviation of the center frequency relative to the actual frequency collected by the communication path node. The mathematical model is: Among them, F is the frequency modulation loss term, which indicates the degree of influence of frequency deviation on signal energy attenuation, β F is the frequency adjustment coefficient of the communication signal inside the building, and its value range is [0.01, 0.1]. In this embodiment, it is set to 0.05. f0 is the center frequency. In this embodiment, f0 = 2.4 GHz. c The actual frequency collected by each path node during the communication process, with a value range of [2.38, 2.42] and a unit of GHZ.

[0035] S23. Construct a path propagation loss strategy based on the frequency modulation loss term, combined with the signal propagation path depth inside the building and the relationship between the building response parameters; The mathematical model of the path propagation loss strategy L(D) is: L(D)=E·DF·D 2 ; Where L(D) is the path propagation loss strategy, which represents the attenuation trend of the signal in a specific building structure.

[0036] S24. Based on the path propagation loss strategy and taking into account the path complexity constraint, a propagation power mechanism is constructed: Where, P eff (t) is the propagation power mechanism, which represents the effective transmission capability of the signal under the path propagation conditions, A(t) is the transmission power of the communication signal at time t, and R is the path complexity factor; The mathematical model of transmit power A(t) is: A(t)=A0·(1+δ·sin(ωt)); Where A0 is the basic transmit power amplitude, which is set to 1.0W in this embodiment; δ is the power disturbance coefficient, which controls the dynamic power fluctuation amplitude and has a value range of [0, 0.5]. In this embodiment, it is set to 0.2; ω is the periodic angular frequency, which controls the speed of power change and is measured in revolutions per second (rad / s); The mathematical model of the path complexity factor R is: R=R0+ρ·Nwall ; Where R0 is the basic complexity factor, which indicates the minimum complexity under the simplest propagation path. Its value range is [1, 2]. In this embodiment, it is set to 1.5. ρ is the structural impedance adjustment coefficient, which controls the influence of each structural unit on the path complexity. Its value range is [0.1, 1.0]. In this embodiment, it is set to 0.5. N wall is the number of walls or structural units passed through in the propagation path, which is a non-negative integer determined by the building space layout. In this embodiment, N is set wall =4.

[0037] S25. By constructing a dynamic mapping mechanism between the path loss weight adjustment factor and the propagation power mechanism, the intermediate propagation state result for the spatial node pair is obtained. The mathematical model is: w ij =exp(-λ·L(D ij )); S ij =P eff (t)·w ij ; Where D ij represents the propagation path depth between node i and node j, w ij is the path loss weight, λ is the path loss weight adjustment factor, in this embodiment, λ=0.2, L(D ij ) is the path propagation loss strategy between node i and node j, S ij is the state matrix value propagated between nodes.

[0038] S3. According to the intermediate propagation state results and the communication electrical signal data, a curvature adjustment coefficient is introduced to perform non-Euclidean manifold embedding and construct a non-Euclidean convolution strategy; based on the signal reconstruction error, the non-Euclidean convolution encoder and decoder are trained to obtain network weight parameters; the trained encoder and decoder are used to encode and reconstruct the embedded signal, respectively, to obtain multiple reconstructed modal signals.

[0039] Furthermore, in step S3, a non-Euclidean convolution strategy is constructed and non-Euclidean encoder and decoder training is performed. The process is as follows: Figure 3 As shown in Figure 2, the specific steps for constructing a non-Euclidean convolution strategy and training a non-Euclidean convolution encoder and decoder are as follows.

[0040] S31. Construct an embedding mapping function for mapping the input signal to a non-Euclidean space, and establish a geometric metric model based on the structural mapping relationship between the embedded samples, thereby obtaining the geodesic relationship between the samples in the embedded space. The mathematical model is: Among them, xi is the i-th input signal, the initial dimension of the signal is 128, z i is the embedding vector representation of the i-th input signal, φ(·) is the embedding mapping function, σ emb is a nonlinear embedding mapping function used to embed signals from Euclidean space to non-Euclidean manifold, W emb is the weight of the nonlinear embedding mapping function, b emb is the bias of the nonlinear embedding mapping function, is the n-dimensional Euclidean space, that is, the original feature space of the input signal, the feature has a linear structure and fixed dimension, → is the function mapping representation is the target non-Euclidean manifold space, n is the embedding dimension, in this embodiment, n=64 is set, To complete the embedding mapping function φ(·) from arrive Mapping; Nonlinear mapping function σ emb The mathematical model is: Among them, α is the inhibition adjustment coefficient, which controls the strength of the nonlinear inhibition term in the activation function, and has a value range of [0.1, 2.0]. In this embodiment, α is set to 0.5. β is the scaling sensitivity factor, which adjusts the influence rate of the square term of the input value in the exponential function, and has a value range of [0.1, 1.0]. In this embodiment, β is set to 0.8.

[0041] S32. By introducing a tensor quantization mechanism that reflects the logarithmic mapping relationship between embedded points and combining it with a geometric distance metric with a local curvature adjustment coefficient, we construct an adjacency weight function for describing the local structural relationship of the embedded space: Among them, K(z i , z j ) is the adjacency weight function, Indicates from z j to z i The logarithmic mapping tensor of κ i is point z i The curvature adjustment coefficient at , τ is the bandwidth, and the value range is [0.5, 2.0]. In this embodiment, τ = 1.0; Curvature adjustment coefficient κ i The mathematical model is: Wherein, κ0 is the basic bandwidth constant, and its value range is [0.5, 1.0]. In this embodiment, it is set to 0.8. γ is the disturbance amplification factor, which controls the degree of influence of neighborhood disturbance on bandwidth. Its value range is [0.1, 2.0]. In this embodiment, it is set to 0.5. From point z i To its neighboring point z j The logarithmic mapping tensor of , is the neighborhood of point i, is the size of the neighborhood, that is, the number of neighbor nodes; Adjacent point set The mathematical model is: Among them, ||z i -z j′ ||2 is the Euclidean distance between node i and node j′ in the embedding space, j′≠i, argmin (k) Select the nearest k points, where k is the number of neighbor nodes. In this embodiment, k=8 is set.

[0042] S33. Generate local perception graph structure based on adjacency weight function and construct non-Euclidean convolution strategy for embedding spatial structure. The mathematical model is: in, is the neighborhood of node i; σ is the differentiable processing module of nonlinear transformation; l is the lth layer in the non-Euclidean convolution strategy. In this embodiment, the graph structure contains 3 layers of non-Euclidean convolution units, i.e., l = 0, 1, 2, and the feature dimension of each layer is 64; W (l) is the weight matrix of the first layer, which is used to perform weighted transformation on the features of adjacent nodes in the first layer. As a learnable parameter in the neural network structure, it is dynamically updated during the training process through the error back propagation algorithm. In this embodiment, the weight matrix Neighborhood The original feature map of the node at layer l is represented. In this embodiment, the input feature map of layer l=0 is Directly take the embedding vector z i As initial feature input; is the convolution output feature; the mathematical model of the differentiable processing module σ of the nonlinear transformation is: Among them, η is the feature enhancement adjustment coefficient, which controls the nonlinear stretching degree of the activation function. In this embodiment, it is set to η = 0.5. ζ is the feature suppression adjustment factor, which adjusts the suppression effect when the input amplitude is relatively small. In this embodiment, it is set to ζ = 0.1.

[0043] S34. By introducing the convolution strategy of non-Euclidean space into the encoder and decoder structures, a non-Euclidean convolution encoder and decoder structure adapted to the embedded manifold space is constructed. The reconstruction error is used as the target loss function to train the non-Euclidean convolution encoder and decoder structure. The mathematical model is: Wherein, N is the number of training samples, which is set to N=512 in this embodiment, Enc(x i ) is the encoder function, Dec(·) is the decoder function, Dec(Enc(·)) is the non-Euclidean convolution encoder and decoder function, L rec is the target loss function; Encoder function Enc(x i ) is: Enc(x i )=σ(W (2) ·σ(W (1) ·σ(W (0) ·x i ))); Among them, W (0) is the input signal x i The linear transformation weight matrix mapped to the first layer embedding, in this embodiment, W (1) is the linear transformation weight matrix of the second layer embedding mapping. In this embodiment, W (2) is the linear transformation weight matrix of the third layer embedding mapping. In this embodiment, The mathematical model of the decoder function Dec(·) is: The mathematical model of the non-Euclidean convolutional encoder and decoder function Dec(Enc(·)) is:

[0044] S35. Encode and reconstruct the input signal based on the trained non-Euclidean convolutional encoder and decoder structure to obtain multiple reconstructed modal signals. The mathematical model is: z i =Enc(x i ); Where, is the reconstruction result of the modal signal, z m represents the mth reconstructed modal signal.

[0045] S4. Introduce disturbances to each reconstructed modal signal, measure the response changes of other modal signals, calculate the disturbance impact values ​​between modal signals, and construct the modal disturbance response relationship matrix.

[0046] Furthermore, in step S4, the modal disturbance response relationship matrix is ​​constructed, and the process is as follows: Figure 4 As shown in Figure 2, the specific steps for constructing the modal disturbance response relationship matrix are:

[0047] S41. Construct a multimodal signal set based on the reconstructed multiple modal signals, and add disturbance to each modal signal in the set to obtain a multimodal signal set after adding disturbance. The mathematical model is: z′ m =z m +δ m ; Where z m represents the original signal of the mth mode, δ m In this embodiment, δ is set to be the disturbance introduced. m It obeys a Gaussian distribution with zero mean and standard deviation of 0.01, that is The disturbance amplitude is controlled between [-0.05, 0.05], z′ m is the modal signal after disturbance.

[0048] S42, through the modal input z before disturbance m and the perturbed modal input z′ m , a response difference mapping mechanism for the modal pair is established, and then the disturbance impact value between the modes is accurately quantified. The mathematical model is: ΔR m,n =R n (z′ m )-R n (z m ); Among them, R n (·) represents the response function of mode n, ΔR m,n Represents the disturbance effect of mode m on mode n; Response function R n The mathematical model of (·) is: Among them, v n is the response direction vector of the nth mode. In this embodiment, Set to a normalized vector; normalized vector v n The mathematical model is: Wherein, d is the modal input feature dimension, which is set to d=64 in this embodiment.

[0049] S43. The disturbance influence values ​​between the modes are matrix-integrated according to the index position to construct a unique modal disturbance response relationship matrix. The mathematical model is: R=[ΔR m,n ] M×M Wherein, R represents the inter-modal disturbance response matrix, M represents the total number of modes, which is set to M=8 in this embodiment. The disturbance response values ​​are combined according to the index position to form a disturbance response relationship matrix R with a dimension of M×M.

[0050] S5. Based on the modal disturbance response relationship matrix, a disturbance direction sensitivity matrix is ​​constructed, a modal self-suppression factor is introduced, a fusion weight coefficient is calculated, and a fusion process is performed on the multiple reconstructed modal signals to obtain a fused internal building communication electrical signal.

[0051] Furthermore, in step S5, the communication signal inside the building is fused and processed, and the process is as follows: Figure 5 As shown in FIG, the specific steps for fusion processing of internal communication electrical signals of a building are as follows.

[0052] S51. By normalizing and calculating the relative intensity ratio of the inter-modal disturbance response values ​​in a specific modal direction, a directional sensitivity matrix is ​​constructed to characterize the distribution characteristics of the inter-modal directional interference effect. The mathematical model is: Where S m，n Indicates the directional disturbance sensitivity coefficient of mode m to mode n, ΔR m，n is the disturbance impact value.

[0053] S52. Perform a mean inverse measurement on the response strength of each mode in the directional sensitivity matrix and construct a modal self-suppression factor to measure the adaptability of the mode to external disturbances. The mathematical model is: Where, α m represents the self-suppression factor of mode m.

[0054] S53. Normalize the self-inhibition factors of each modality and construct a fusion weight coefficient to measure the relative contribution of multimodal information. The mathematical model is: Where w m Represents the fusion weight coefficient of the m-th modality.

[0055] S54. Perform weighted combination based on each reconstructed modal signal and its corresponding fusion weight factor to obtain a fused building internal communication electrical signal. The mathematical model is: Where Z represents the internal communication electrical signal of the building after fusion processing.

[0056] S6. Construct a building internal communication electric signal processing model, input the building internal communication electric signal data set, go through steps S2 to S5 in sequence, and perform iterative training until convergence to complete the optimization processing of the building internal communication electric signal.

[0057] In step S6, for the building internal communication electrical signal processing model, first input the data set containing building structure information, propagation path information and multimodal communication characteristics, and execute the processing processes defined in steps S2 to S5 in sequence; specifically, in the propagation response modeling stage, according to the propagation path characteristics and structural parameters of the communication signal in the building environment, construct a propagation model reflecting the modulation capability of the building structure, extract the response characteristics of the communication path, and generate the intermediate propagation state for subsequent modeling; then, in the non-Euclidean geometric encoding stage, the structural relationship between the signals is used to construct a mapping mechanism, and a graph convolution strategy with local curvature adjustment capability is introduced to construct a non-Euclidean convolution encoder. The proposed method combines a decoder structure with an encoder to achieve low-dimensional embedding and signal reconstruction while maintaining the geometric structure of the signal. In the modal perturbation analysis phase, by introducing small perturbations to each modal signal and observing the response changes of other modes, the interference relationship between the modes is quantified, and a modal perturbation response relationship matrix is ​​constructed to characterize the mutual influence characteristics between the modes. Then, in the signal fusion phase, the directional sensitivity matrix between the modes is constructed by normalizing the perturbation response intensity, and the self-suppression ability of each mode is calculated based on its average distribution. The fusion weight coefficient is further constructed to achieve weighted fusion processing of multimodal signals, resulting in a stable and enhanced building communication electrical signal. This processing flow is integrated and trained in an end-to-end manner, and combined with a structured loss mechanism and optimization algorithm, enabling the model to dynamically adjust the propagation modeling strategy and modal fusion strategy in different building environments, thereby improving the adaptability, stability and availability of communication signals in complex indoor scenarios.

[0058] Furthermore, in step S6, the building internal communication signal processing model is developed based on the Python programming language and implemented using the PyTorch framework. The input is the communication signal data with a signal sample dimension of 128. The model is trained using the RMSprop optimizer, the learning rate is set to 0.0005, the batch size is set to 32, and the number of training rounds is 150. The model uses the mean square error (MSE) as the loss function for training, and is iteratively trained through a non-Euclidean encoder structure. During the training process, the loss function converges to a stable range [10 -3 , 10 -4 ], indicating that the model can well restore the amplitude structure and change trend of the internal communication electrical signals of the building, and complete the mapping and reconstruction of the internal communication electrical signals of the building.

[0059] Furthermore, in step S6, the original building internal communication electric signal is input into the constructed communication electric signal processing model for processing, and the processing effect is as follows: Figure 6 As shown in the figure, the horizontal axis is time (in seconds), and the vertical axis is the signal amplitude (without units). As can be seen from the figure, the original communication electrical signal has obvious time-domain fluctuations and noise interference, which affects the signal stability and clarity. However, after optimization processing using the proposed internal building communication electrical signal processing model, the signal fluctuation amplitude is significantly reduced, the overall curve is smoother, noise interference is effectively suppressed, and signal clarity and continuity are greatly improved, verifying the effectiveness and robustness of the model in improving the quality of communication electrical signals in complex building propagation environments.

[0060] The above are only preferred embodiments of the present invention. It should be pointed out that those skilled in the art can make several modifications and improvements without departing from the inventive concept of the present invention, and these all fall within the scope of protection of the present invention.

Claims

1. A method for processing electrical signals for internal communication in a building, characterized in that: The specific steps include: S1. Acquire communication electrical signal data in a building's internal communication environment, including transmitted signals, reflected signals, penetrated signals, and building structure-related information, and construct a dataset containing propagation paths, frequency band parameters, building material properties, and channel responses. S2. Designing building response parameters based on the communication electrical signal data in the building's internal communication environment and combining them with frequency modulation loss terms to establish a path propagation loss strategy, perform building propagation response modeling, and obtain intermediate propagation state results that describe signal modulation characteristics; S3. According to the intermediate propagation state result and the communication electrical signal data, a curvature adjustment coefficient is introduced to perform non-Euclidean manifold embedding and construct a non-Euclidean convolution strategy; The non-Euclidean convolutional encoder and decoder are trained based on the signal reconstruction error to obtain network weight parameters; the trained encoder and decoder are used to encode and reconstruct the embedded signal, respectively, to obtain reconstructed multiple modal signals; S4. Introduce disturbances to each reconstructed modal signal, measure the response changes of other modal signals, calculate the disturbance impact value between modal signals, and construct a modal disturbance response relationship matrix; S5. Constructing a disturbance direction sensitivity matrix based on the modal disturbance response relationship matrix, introducing a modal self-inhibition factor, calculating a fusion weight coefficient, and fusing multiple reconstructed modal signals to obtain a fused building internal communication electrical signal. S6. Construct a building internal communication electric signal processing model, input the building internal communication electric signal data set, go through steps S2 to S5 in sequence, and perform iterative training until convergence to complete the optimization processing of the building internal communication electric signal.

2. A method for processing electrical signals for internal communication in a building according to claim 1, characterized in that: The specific method for constructing the data set including propagation path, frequency band parameters, building material properties and channel response is as follows: Acquire raw data of communication electrical signals within a building's interior environment, including transmitted, reflected, and penetrated signals, and obtain signal sequences through time-sharing sampling. Simultaneously, acquire building structural information, including building material type, wall thickness, floor structure, door and window distribution, and spatial partitioning. Further acquire frequency band parameters, including transmit frequency, carrier frequency, and channel bandwidth, as well as historical communication data and building environment parameters. Marking the starting point, end point, and intermediate transmission medium of the propagation path based on the original data of the communication electrical signal and the building structure information, and constructing a propagation path topology structure; A communication electric signal dataset including propagation path, frequency band parameters, building material properties and channel response information is created based on the communication electric signal original data, building structure information, frequency band parameters, historical communication data and propagation path topology.

3. A method for processing electrical signals for internal communication in a building according to claim 2, characterized in that: The specific process of establishing the path propagation loss strategy includes: S21. Couple the environmental variables with the depth of the signal propagation path in the building structure, use the nonlinear mapping mechanism to model their joint effects, and generate building response parameters that can dynamically reflect the internal response characteristics of the building. The calculation formula is: Where E is the building response parameter, θ env is the environmental variable, D is the depth of the signal propagation path inside the building; S22. Define the frequency modulation loss term based on the quadratic function relationship between the frequency adjustment coefficient of the internal communication signal in the building and the frequency deviation of the center frequency relative to the actual frequency collected by the communication path node. The mathematical model is: Where F is the frequency modulation loss term, β F is the frequency adjustment coefficient of the communication signal inside the building, f0 is the center frequency, f c The actual frequency collected by each path node during the communication process; S23. Based on the frequency modulation loss term, combined with the signal propagation path depth inside the building and the relationship between the building response parameters, a path propagation loss strategy is constructed.

4. A method for processing electrical signals for internal communication in a building according to claim 3, characterized in that: The intermediate propagation state result describing the signal modulation characteristics is obtained by: S201: Based on the path propagation loss strategy and taking into account the path complexity constraint, a propagation power mechanism is constructed: Where, P eff (t) is the propagation power mechanism, A(t) is the transmission power of the communication signal at time t, R is the path complexity factor, and L(D) is the path propagation loss strategy; S202. By constructing a dynamic mapping mechanism that integrates the path loss weight adjustment factor and the propagation power mechanism, an intermediate propagation state result for the spatial node pair is obtained. The mathematical model is: w ij =exp(-λ·L(D ij )); S ij =P eff (t)·w ij ; Where D ij represents the propagation path depth between node i and node j, w ij is the path loss weight, λ is the path loss weight adjustment factor, L(D ij ) is the path propagation loss strategy between node i and node j, S ij is the state matrix value propagated between nodes.

5. A method for processing electrical signals for internal communication in a building according to claim 4, characterized in that: The specific steps to build a non-Euclidean convolution strategy include: S31. Construct an embedding mapping function for mapping the input signal to a non-Euclidean space, and establish a geometric metric model based on the structural mapping relationship between the embedded samples, thereby obtaining the geodesic relationship between the samples in the embedded space. The mathematical model is: z i =φ(x i )=σ emb (W emb x i +b emb ), Among them, x i is the i-th input signal, z i is the embedding vector representation of the i-th input signal, φ(·) is the embedding mapping function, σ emb is the nonlinear embedding mapping function, W emb is the weight of the nonlinear embedding mapping function, b emb is the bias of the nonlinear embedding mapping function, n is the embedding dimension, is an n-dimensional Euclidean space, → is a function mapping symbol, is the target non-Euclidean manifold space; S32. By introducing a tensor quantization mechanism that reflects the logarithmic mapping relationship between embedded points and combining it with a geometric distance metric with a local curvature adjustment coefficient, we construct an adjacency weight function for describing the local structural relationship of the embedded space: Among them, K(z i , z j ) is the adjacency weight function, Indicates from z j to z i The logarithmic mapping tensor of κ i is point z i The curvature adjustment coefficient at , τ is the bandwidth; S33. Generate local perception graph structure based on adjacency weight function and construct non-Euclidean convolution strategy for embedding spatial structure. The mathematical model is: in, is the neighborhood of node i, σ is the differentiable processing module of nonlinear transformation, l is the lth layer in the non-Euclidean convolution strategy, W (l) is the weight matrix of the lth layer, Neighborhood The original feature map representation of the node at layer l, is the convolution output feature.

6. A method for processing electrical signals for internal communication in a building according to claim 5, characterized in that: The specific steps of obtaining the reconstructed multiple modal signals include: S301. By introducing the convolution strategy of non-Euclidean space into the encoder and decoder structures, a non-Euclidean convolution encoder and decoder structure adapted to the embedded manifold space is constructed, and the reconstruction error is used as the target loss function to train the non-Euclidean convolution encoder and decoder structure. The mathematical model is: Among them, N is the number of training samples, Enc(x i ) is the encoder function, Dec(·) is the decoder function, Dec(Enc(·)) is the non-Euclidean convolution encoder and decoder function, L rec is the target loss function; S302: Encode and reconstruct the input signal based on the trained non-Euclidean convolutional encoder and decoder structure to obtain multiple reconstructed modal signals. The mathematical model is: z i =Enc(x i ); Where, is the reconstruction result of the modal signal, z m represents the mth reconstructed modal signal.

7. A method for processing electrical signals for internal communication in a building according to claim 6, characterized in that: The process of constructing the modal disturbance response relationship matrix includes: S41. Construct a multimodal signal set based on the reconstructed multiple modal signals, and add disturbance to each modal signal in the set to obtain a multimodal signal set after adding disturbance. The mathematical model is: With' m =z m +δ m ; Where z m represents the original signal of the mth mode, δ m is the introduced disturbance, z′ m is the modal signal after disturbance; S42, through the modal input z before disturbance m and the perturbed modal input z′ m , a response difference mapping mechanism for the modal pair is established, and then the disturbance impact value between the modes is accurately quantified. The mathematical model is: ΔR m，n =R n (from' m )-R n (from m ); Among them, R n (·) represents the response function of mode n, ΔR m，n Represents the disturbance effect of mode m on mode n; S43. The disturbance influence values ​​between the modes are matrix-integrated according to the index position to construct a unique modal disturbance response relationship matrix. The mathematical model is: R=[ΔR m，n ] m×M ; Where R represents the inter-modal disturbance response matrix, and M represents the total number of modes.

8. A method for processing electrical signals for internal communication in a building according to claim 7, characterized in that: The specific steps for fusion processing of internal communication electrical signals in buildings include: S51. By normalizing and calculating the relative intensity ratio of the inter-modal disturbance response values ​​in a specific modal direction, a directional sensitivity matrix is ​​constructed to characterize the distribution characteristics of the inter-modal directional interference effect. The mathematical model is: Where S m,n Indicates the directional disturbance sensitivity coefficient of mode m to mode n, ΔR m，n is the disturbance impact value; S52. Perform a mean inverse measurement on the response strength of each mode in the directional sensitivity matrix and construct a modal self-suppression factor to measure the adaptability of the mode to external disturbances. The mathematical model is: Where, α m represents the self-inhibition factor of mode m; S53. Normalize the self-inhibition factors of each modality and construct a fusion weight coefficient to measure the relative contribution of multimodal information. The mathematical model is: Where w m Represents the fusion weight coefficient of the mth modality; S54. Perform weighted combination based on each reconstructed modal signal and its corresponding fusion weight factor to obtain a fused building internal communication electrical signal. The mathematical model is: Where Z represents the internal communication electrical signal of the building after fusion processing.

Images