Bluetooth fingerprint database updating method based on voiceprint quality estimation

Abstract

The invention relates to the technical field of Bluetooth positioning, and particularly discloses a voiceprint quality estimation-based Bluetooth fingerprint database updating method, which comprises the following steps of: deploying a base station group with Bluetooth broadcasting and sound signal transmitting functions in a target area, and performing preliminary positioning by utilizing a Bluetooth path loss model; performing position estimation based on the received signal strength of each position node to establish a Bluetooth fingerprint database; meanwhile, sound signals of all the positions are synchronously received, quality evaluation is conducted on the received sound signals, effective sound signals are screened out to be subjected to TDOA positioning calculation, sound signal positioning coordinates of all the positions are determined, and the sound signal positioning accuracy is judged according to a position quality evaluation method; and the Bluetooth intensity of the current position is reversely deduced according to the accurate sound signal position information so as to update the Bluetooth fingerprint database in real time. According to the method, the direction sensitivity of the Bluetooth signal is compensated by using the propagation stability of the sound signal, the dynamic change of the indoor environment can be effectively dealt with, and the self-adaptive maintenance of the fingerprint data is realized.

Description

Technical Field

[0001] This application relates to the field of Bluetooth positioning technology, and more specifically, to a method for updating a Bluetooth fingerprint database based on voiceprint quality estimation. Background Art

[0002] With the rapid development of network communication technology and the widespread popularity of intelligent mobile terminals, location-based services have provided great convenience for people in outdoor scenarios. However, in indoor scenarios, such as finding a car in an underground parking lot, quickly locating trapped people, and finding someone in a large transportation hub, due to the limitations of the development of indoor positioning technology, providing high-precision location information services has always faced challenges. To address these challenges, various technologies and solutions have been proposed, including Wi-Fi, Bluetooth, Ultra-Wide Band (UWB), etc. Although these technologies have their own advantages, it is difficult to achieve a balance in terms of cost, accuracy, coverage, and compatibility with smartphones at the same time. For example, although the UWB-based positioning system can provide centimeter-level precise positioning, the cost of its receivers and transmitters is relatively high, and it is difficult to be compatible with smartphones, which greatly limits its practical applications.

[0003] In the prior art, the positioning method based on Bluetooth fingerprints has been favored because of its low cost, easy deployment, and extremely high device compatibility. Especially in an environment with high personnel mobility, relatively accurate positioning can be achieved without additional hardware. The flexible deployment characteristics of Bluetooth Low Energy (BLE) enable it to be easily installed in an indoor environment and use spread-spectrum communication technology to transmit positioning data. Even in a wall-blocked and complex environment, stable signal transmission can be maintained, making it a simple and economical indoor positioning solution.

[0004] However, the main defect of the positioning method based on Bluetooth fingerprints lies in the construction and update of the fingerprint database. Due to dynamic changes in the environment, such as factors like personnel movement, furniture movement, and wireless interference, the Received Signal Strength Indicator (RSSI) of Bluetooth signals will fluctuate significantly, making the initially constructed fingerprint database may not accurately reflect the real-time signal environment. In addition, the polarization effect of the device antenna and small changes in position will also cause changes in signal strength, further affecting the stability of the fingerprint database. Although regularly updating the fingerprint database can improve positioning accuracy, this process is both cumbersome and time-consuming, especially when deployed in a large area, the maintenance cost is very high.

[0005] Traditional Bluetooth fingerprint database update methods mainly include manual re - collection and regular batch update. However, manual re - collection requires technicians to manually measure and record new RSSI values at different locations. Although it can ensure a certain degree of accuracy, it is time - consuming and labor - intensive, especially in cases where the environment changes frequently, resulting in extremely high maintenance costs. Regular batch update updates the entire fingerprint database at set time intervals. Although it reduces some labor costs, it cannot cope with immediate environmental changes, resulting in the updated fingerprint database still being inaccurate. In addition, these methods cannot effectively handle signal instability caused by real - time signal fluctuations and multipath effects. Excessive update frequencies also consume computing and storage resources, making it difficult to maintain good positioning effects in dynamic and complex environments.

[0006] Therefore, a Bluetooth fingerprint database update method based on voiceprint quality estimation is expected. Summary of the Invention

[0007] To solve the above - mentioned technical problems, this application is proposed. Embodiments of this application provide a Bluetooth fingerprint database update method based on voiceprint quality estimation. It deploys a base station group with both Bluetooth broadcasting and acoustic signal transmitting functions in the target area, uses the Bluetooth path loss model for preliminary positioning, and estimates the position based on the received signal strength of each position node to establish a Bluetooth fingerprint database. At the same time, it synchronously receives acoustic signals from each position, evaluates the quality of the received acoustic signals, screens out effective acoustic signals for TDOA positioning calculation, determines the acoustic signal positioning coordinates of each position, and judges the accuracy of the acoustic signal positioning according to the position quality evaluation method. Furthermore, it inversely deduces the Bluetooth strength of the current position based on the accurate acoustic signal position information to update the Bluetooth fingerprint database in real - time. This method uses the stability of acoustic signal propagation to compensate for the direction sensitivity of Bluetooth signals, can effectively cope with the dynamic changes in the indoor environment, and realizes the adaptive maintenance of fingerprint data.

[0008] Correspondingly, according to one aspect of this application, a Bluetooth fingerprint database update method based on voiceprint quality estimation is provided, which includes: Construct a Bluetooth fingerprint database, where each Bluetooth fingerprint in the Bluetooth fingerprint database is {RSSI values of each position, two - dimensional coordinates of each position}; Receive the acoustic signals from each position; Judge whether the acoustic signals from each position meet the signal quality standard, and after the acoustic signals from each position meet the signal quality standard, estimate the position based on the acoustic signals from each position to obtain the acoustic signal positioning coordinates of each position; Calculate the hyperbola intersection deviation of the acoustic signal positioning coordinates based on a parameterized method to generate an acoustic signal position calculation quality evaluation value; Based on the comparison between the quality evaluation value calculated from the acoustic signal positions at each location and the acoustic positioning quality threshold, determine whether it is necessary to update the RSSI value at each location.

[0009] Correspondingly, according to another aspect of the present application, there is provided a method for updating a Bluetooth fingerprint database based on acoustic fingerprint quality estimation, which includes: Construct a Bluetooth fingerprint database, where each Bluetooth fingerprint in the Bluetooth fingerprint database is {RSSI values at each location, two-dimensional coordinates at each location}; Receive the acoustic signals at each location; Based on the RSSI values at each location and the acoustic signals at each location, determine whether the signal quality standard is met, and in response to meeting the signal quality standard, perform position estimation based on the acoustic signals at each location to obtain the acoustic signal positioning coordinates at each location; Calculate the hyperbola intersection deviation of the acoustic signal positioning coordinates based on a parametric method to generate an acoustic signal position calculation quality evaluation value; Based on the comparison between the acoustic signal position calculation quality evaluation value at each location and the acoustic positioning quality threshold, determine whether it is necessary to update the RSSI value at each location.

[0010] Compared with the prior art, the method for updating a Bluetooth fingerprint database based on acoustic fingerprint quality estimation provided by the present application deploys a base station group with both Bluetooth broadcasting and acoustic signal transmitting functions in the target area, uses the Bluetooth path loss model for preliminary positioning, and performs position estimation based on the received signal strength at each location node to establish a Bluetooth fingerprint database; at the same time, synchronously receive the acoustic signals at each location, perform quality evaluation on the received acoustic signals, screen out effective acoustic signals for TDOA positioning calculation, determine the acoustic signal positioning coordinates at each location, and judge the accuracy of the acoustic signal positioning according to the position quality evaluation method, and then inversely deduce the Bluetooth strength at the current location based on the accurate acoustic signal position information to update the Bluetooth fingerprint database in real time. This method uses the stability of acoustic signal propagation to compensate for the direction sensitivity of Bluetooth signals, can effectively cope with the dynamic changes in the indoor environment, and realizes the adaptive maintenance of fingerprint data. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] By describing the embodiments of the present application in more detail in conjunction with the drawings, the above and other objects, features, and advantages of the present application will become more obvious. The drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification, and are used to explain the present application together with the embodiments of the present application, and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0012] Figure 1Schematic diagram of the deployment of a Bluetooth and acoustic signal fusion-based broadcast positioning system according to an embodiment of the present application.

[0013] Figure 2 Polar coordinate schematic diagram of the directivity comparison between acoustic signals and Bluetooth RSSI signals according to an embodiment of the present application.

[0014] Figure 3 Flowchart of a method for updating a Bluetooth fingerprint database based on acoustic fingerprint quality estimation according to an embodiment of the present application.

[0015] Figure 4 Flowchart of step S1 in the method for updating a Bluetooth fingerprint database based on acoustic fingerprint quality estimation according to an embodiment of the present application.

[0016] Figure 5 Schematic diagram of TDOA positioning estimation results under the conditions of no error and the presence of noise error.

[0017] Figure 6 Flowchart for determining whether the signal quality standard is met based on the RSSI values at each position and the acoustic signals at each position in the method for updating a Bluetooth fingerprint database based on acoustic fingerprint quality estimation according to Embodiment 2 of the present application.

[0018] Figure 7 Flowchart of step S33 in the method for updating a Bluetooth fingerprint database based on acoustic fingerprint quality estimation according to Embodiment 2 of the present application. Detailed implementation manners

[0019] Next, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0020] Embodiment 1 In view of the deficiencies of the existing methods for updating Bluetooth fingerprint databases, the present application proposes a method for updating a Bluetooth fingerprint database based on acoustic fingerprint quality estimation. In this method, a group of base stations with both Bluetooth broadcast and acoustic signal transmission functions are deployed in the target area. The Bluetooth path loss model is used for preliminary positioning, and position estimation is performed based on the received signal strength of each position node to establish a Bluetooth fingerprint database. At the same time, acoustic signals at each position are synchronously received, and the quality of the received acoustic signals is evaluated. Effective acoustic signals are selected for TDOA positioning calculation to determine the positioning coordinates of the acoustic signals at each position. The accuracy of the acoustic signal positioning is judged according to the position quality evaluation method. Furthermore, the Bluetooth strength at the current position is inversely deduced based on the accurate acoustic signal position information to update the Bluetooth fingerprint database in real time. This method uses the stability of acoustic signal propagation to compensate for the direction sensitivity of Bluetooth signals, can effectively cope with the dynamic changes in the indoor environment, and realizes the adaptive maintenance of fingerprint data.

[0021] Specifically, Figure 1 FIG. is a deployment schematic diagram of a Bluetooth and acoustic signal fusion broadcast positioning system according to an embodiment of the present application. Figure 1 As shown, the Bluetooth and acoustic signal fusion broadcast positioning system mainly consists of intelligent positioning terminals, multi-modal positioning base stations with Bluetooth and acoustic signal broadcast functions, a central computing server, etc. N multi-modal positioning base stations are deployed at the four corners of the ceiling edge to emit acoustic signals with a certain delay for distinguishing the signal source, and at the same time, the built-in Bluetooth beacon module broadcasts BLE messages. The intelligent positioning terminal receives the acoustic signals from the acoustic signal base stations and at the same time receives the BLE messages, and communicates between the Bluetooth beacon module and the central positioning server for positioning. As Figure 2 shown, limited by the polarization effect of the Bluetooth on-board antenna, the RSSI values centered on the traditional Bluetooth beacon are not evenly distributed. If the polarization directions of the transmitting and receiving antennas do not match, the signal strength will decrease. At the same time, when the angle between the antenna and the receiving device changes greatly, the received signal strength will fluctuate, making it difficult to ensure the accuracy of the Bluetooth fingerprint database construction, and thus affecting the accuracy and stability of the overall positioning system. The sound pressure level distribution based on acoustic signal positioning has a stable directivity during the propagation process in the indoor environment, which can be used to make up for the inaccuracy of the fingerprint database data caused by the rapid fluctuation of Bluetooth RSSI with the environment.

[0022] Figure 3 FIG. is a flowchart of a method for updating a Bluetooth fingerprint database based on acoustic fingerprint quality estimation according to an embodiment of the present application. As Figure 3 shown, the method for updating a Bluetooth fingerprint database based on acoustic fingerprint quality estimation according to an embodiment of the present application includes the steps of: S1, constructing a Bluetooth fingerprint database, where each Bluetooth fingerprint in the Bluetooth fingerprint database is {RSSI values at each position, two-dimensional coordinates at each position}; S2, receiving acoustic signals at each position; S3, determining whether the acoustic signals at each position meet the signal quality standard, and after the acoustic signals at each position meet the signal quality standard, performing position estimation based on the acoustic signals at each position to obtain the acoustic signal positioning coordinates at each position; S4, calculating the hyperbola intersection deviation of the acoustic signal positioning coordinates based on a parameterization method to generate an acoustic signal position calculation quality evaluation value; S5, determining whether it is necessary to update the RSSI values at each position based on the comparison between the acoustic signal position calculation quality evaluation values at each position and the acoustic positioning quality threshold.

[0023] In the above Bluetooth fingerprint database update method based on voiceprint quality estimation, in step S1, a Bluetooth fingerprint database is constructed, and each Bluetooth fingerprint in the Bluetooth fingerprint database is {the RSSI value at each position, the two-dimensional coordinates at each position}. It should be understood that the Bluetooth fingerprint database is the core data foundation of the indoor positioning system. Its essence is to record the mapping relationship between the signal characteristics (RSSI value) and physical coordinates at different position points to achieve positioning inference. In this application, the initial database construction is carried out through the Bluetooth path loss model (FSPL), aiming to predict the signal attenuation law with a mathematical model, reduce the manual sampling cost, and provide a benchmark framework for the subsequent dynamic update of the Bluetooth fingerprint database.

[0024] Specifically, traditional Bluetooth fingerprint databases rely on static RSSI acquisition data in a fixed environment. However, in actual dynamic scenarios, factors such as personnel movement, obstacle movement, and electromagnetic interference will cause significant fluctuations in the signal strength (RSSI). If the uncalibrated original RSSI values are directly used to construct the fingerprint database, systematic biases will be introduced. For example, when the polarization direction of the device antenna does not match the environment, the RSSI may attenuate by 3 - 5 dB; when there are temporary obstacles (such as human body occlusion) in the signal path, the instantaneous RSSI fluctuation can reach more than 10 dB. These non-steady-state noises will cause the fingerprint database to mismatch the actual signal environment, resulting in positioning drift errors. Therefore, this application adopts the classical Friis free space path loss model to establish the physical mapping relationship between the signal strength and the distance, compensating for the device hardware differences and environmental baseline noises. For example, by referring to the calibrated signal strength ( at the reference distance and the path loss exponent , the mathematical relationship between the signal strength and the distance is established: where is the received signal strength value at the current distance , is the received signal strength value at the reference distance , and is the path loss exponent, which is used to reflect the attenuation characteristics of the signal in a specific environment.

[0025] Figure 4 FIG. is a flowchart of step S1 in the Bluetooth fingerprint database update method based on voiceprint quality estimation according to an embodiment of the present application. As Figure 4As shown in the figure, step S1 includes: S11, constructing a fingerprint matrix for each position; S12, calculating the distance between each reference position RSSI vector in the fingerprint matrix and the observed RSSI vector to obtain a distance vector; S13, selecting k reference positions corresponding to the k smallest distances from the distance vector; S14, calculating weights for the k smallest distances and performing weighted averaging on the k reference positions based on the weights to obtain the two-dimensional coordinates of the observed position.

[0026] Specifically, first, divide the target area into K grid nodes, and each position node corresponds to a unique two-dimensional coordinate. After the base station group is deployed, it periodically broadcasts Bluetooth signals, and the intelligent terminal receives RSSI measurement values from L beacon base stations at each position node , which are recorded as the reference position RSSI vectors of each position node, where is the index of the corresponding position node, is the beacon index of the transmitted signal. Combine the reference position RSSI vectors of the K position nodes to form a fingerprint matrix , and each row of it corresponds to the signal characteristics of a position node: Next, associate and store the fingerprint matrix with the two-dimensional coordinates of each position node to form an initial Bluetooth fingerprint library. On this basis, whenever a new Bluetooth signal is received, by comparing each reference position RSSI vector in the fingerprint matrix with the newly received RSSI observation vector, and using the similarity calculation method to evaluate the matching degree between the new signal and each fingerprint in the library, the position of the new signal can be estimated based on the position coordinates corresponding to the similar fingerprints.

[0027] Specifically, first, calculate the distance between each reference position RSSI vector in the fingerprint matrix and the observed RSSI vector to obtain a distance vector. It should be understood that the observed RSSI vector is a set of RSSI values scanned and recorded in real time by an intelligent terminal (such as a mobile phone) carried by the user, which reflects the actual signal strength distribution in the current environment. In order to find the position in the fingerprint library that is most similar to the currently observed Bluetooth signal for estimating the user's current position, the Euclidean distance is used as the metric standard in this application. By calculating the distance between each reference position RSSI vector in the fingerprint matrix and the observed RSSI vector, a distance vector containing K distance values is obtained, where each distance value corresponds to the degree of signal strength difference between a reference position in the fingerprint matrix and the user's current position.

[0028] Next, select k reference positions corresponding to the k smallest distances from the distance vectors. It should be understood that considering that similar positions usually have similar signal characteristics, therefore, in this application, by selecting the k reference points with the smallest distances, the positioning search range is narrowed and the positioning accuracy is improved. In addition, through collaborative decision-making of multiple reference points, the local characteristics of the signal distribution around the current observation point can be captured, the problem of single nearest neighbor being sensitive to noise can be avoided, the positioning jump caused by single-point signal fluctuation can be reduced, and the robustness of position estimation can be enhanced.

[0029] Then, calculate weights for the k smallest distances, and perform weighted averaging on the k reference positions based on the weights to obtain the two-dimensional coordinates of the observation position. It should be understood that the k reference points are considered as candidate positions that best match the signal characteristics of the user's current position, but the credibility of each reference point is not the same. The closer the reference point is to the observation point and the more similar the signal environment is, the greater the contribution of its position information to the estimation of the user's current position. Therefore, this application designs a dynamic weight function, which comprehensively considers the Euclidean distance between the reference point and the observation point and the similarity of signal strength, and assigns higher weights to reference points with small distances and similar signal characteristics.

[0030] In particular, considering that in distance calculation, some base station signals may not be detected due to occlusion or excessive distance, resulting in not every message containing the RSSI values of all beacons. If equal weights are directly assigned to all RSSI values, then the beacon fingerprints with fewer RSSI values (i.e., the case of missing RSSI values) may obtain the same weight as the fingerprints with more RSSI values and similar distances in distance calculation, which will lead to inaccurate results. Therefore, this application adjusts the weight assignment strategy by adding a constant term to reduce the negative impact of beacons with missing RSSI values on the overall positioning result by giving them lower weights. This weighted function is expressed as: where, is the candidate reference point position, is the corresponding distance estimate value, represents the data integrity, represents the total number of beacons, represents the number of valid RSSI values.

[0031] In the above weighted function, a distance reciprocal term is introduced to strengthen the influence of the nearest neighbor, so that the smaller the distance , the greater the weight of the candidate reference point position. At the same time, by adding a constant Adjust the weight deviation caused by the missing RSSI value to ensure that reference points with higher signal integrity receive more attention in the positioning decision. At the same time, retain the value of reference points with some missing data through the compensation term to prevent them from being completely ignored due to incomplete data, so as to maintain global stability. Finally, after normalizing the weights of the obtained k candidate reference point positions, perform a weighted average calculation on the k reference positions to obtain the two-dimensional coordinates of the observed position.

[0032] In the above Bluetooth fingerprint database update method based on voiceprint quality estimation, in step S2, receive the acoustic signals at each position. It should be understood that Bluetooth signals are vulnerable to multipath effects and environmental interference, resulting in drastic fluctuations in RSSI values. While acoustic signals have a stable sound pressure level distribution and directivity when propagating indoors, which can make up for the deficiencies of Bluetooth signals. Therefore, in this application, by synchronously receiving acoustic signals at each position, that is, while the multimodal base station sends Bluetooth signals, it synchronously transmits acoustic wave signals (such as chirp pulses) with a specific delay, thereby providing an auxiliary data source independent of Bluetooth for high-precision positioning.

[0033] In the above Bluetooth fingerprint database update method based on voiceprint quality estimation, in step S3, determine whether the acoustic signals at each position meet the signal quality standard, and after the acoustic signals at each position meet the signal quality standard, perform position estimation based on the acoustic signals at each position to obtain the acoustic signal positioning coordinates at each position. It should be understood that since acoustic signals in the indoor environment may be affected by multipath reflection, noise interference, or device nonlinear distortion, resulting in different qualities of acoustic signals received at different positions within the indoor area, it is necessary to perform quality assessment between the acoustic signals received by the intelligent terminal and the reference signal to determine whether they meet the positioning signal quality standard, ensuring that only high-confidence data enters the positioning process. In a specific example of this application, perform channel impulse response analysis, signal peak distribution analysis, signal reverberation time analysis, and signal frequency distortion analysis on the acoustic signals at each position to determine whether the acoustic signals at each position meet the signal quality standard. The specific evaluation process is as follows: Channel impulse response analysis: Through the channel impulse response (CIR), the multipath effect and delay spread of the signal during transmission can be obtained to further evaluate the signal quality. In the indoor environment, the transmitted signal is , and the received output signal after propagation through the indoor channel is , and the two satisfy the following relationship: Among them, is the received output signal, is the impulse response of the channel, is the length of the channel impulse response. To simplify the convolution solution process, the input signal can be converted into a Toeplitz (T) matrix , and the output signal and the channel impulse response are represented in vector form: where is the length of the output signal.

[0034] Then, construct the Toeplitz matrix of the input signal At this time can be rewritten in matrix form: Directly solving the above equation may result in the matrix being singular or nearly singular. Therefore, this application uses the least squares method to solve, and the channel impulse response of the received signal can be obtained: where represents the transpose matrix of the matrix .

[0035] Analysis of signal peak distribution: When receiving acoustic signals, in a Non-Line-Of-Sight (NLOS) scenario, the signal is usually affected by strong multipath effects, manifested as multiple peaks in the signal peak distribution on the time axis. It is possible to determine whether it is in an NLOS environment by analyzing the peak distribution of the received signal. Extract the peak distribution of the signal through the channel impulse response , record the time delays of multiple main peaks and their corresponding amplitudes. In the NLOS scenario, multiple paths of the signal will cause the peaks to be densely distributed in time. Therefore, it is possible to determine whether there are strong multipath effects by calculating the time difference between adjacent peaks. For all the extracted peaks, calculate the time difference between every two adjacent peaks: Set the threshold of the time difference to determine the significance of the multipath effect. If the time difference , it is determined that this signal is affected by strong multipath effects and is regarded as a NLOS scenario; at the same time, the relative magnitude of the calculated peak amplitude is used to judge the NLOS scenario. Usually, in the Line-Of-Sight (LOS) scenario, the amplitude of the direct path signal is large, and the amplitude of subsequent reflection paths gradually decreases; while in the NLOS scenario, the amplitude of subsequent paths may be close to or even exceed the first peak. Calculate the first peak and subsequent peaks of the relative amplitude ratio : Here, set an amplitude ratio threshold , if a certain subsequent peak satisfies , it can be determined that there are strong multipath effects. Combining the time difference and the amplitude ratio of the two indicators, if the signal meets any one or all of the above conditions, it is determined that the signal is in the NLOS scenario, indicating that its quality may be low and it is not suitable for high-precision positioning.

[0036] Signal reverberation time analysis: In acoustic analysis, the reverberation time (RT) is a key indicator for evaluating the acoustic characteristics of the environment and the signal quality. The reverberation time RT60 is the time required for the sound signal to decay from the initial intensity to 60 dB. The reverberation time RT60 can be obtained from the energy decay curve. Usually, the energy decay of the impulse response can be used to estimate RT60: where represents the energy decay of the signal within time . The larger its value, the more sound wave reflections there are in the environment and the stronger the reverberation, which will have a greater impact on the signal transmission quality. Here, set an RT60 threshold , when the RT60 value exceeds the threshold , it means that the environmental reverberation is too severe and the signal quality cannot meet the requirements of high-precision acoustic signal positioning. By setting this threshold , it is possible to judge the impact of environmental reverberation and ensure that the signal quality is within an acceptable range to support high-precision acoustic signal positioning.

[0037] Signal frequency distortion analysis: Based on the frequency distortion of the received signal to evaluate the acoustic signal quality, mainly by comparing the differences in the frequency responses of the received signal and the reference signal to judge the signal fidelity. If the signal undergoes frequency distortion during transmission due to multipath effects, noise interference, or the nonlinear effects of the device, the amplitude and phase of the signal may deviate, resulting in a decrease in quality. Let and are the spectra of the reference signal and the received signal respectively, then the frequency response function is: wherein, the amplitude of represents the gain change of the frequency component, and the angle represents the phase change. The amplitude distortion is mainly reflected in the amplitude of deviating from 1. If the amplitude deviation is too large in some frequency bands, it indicates that the intensity of the received signal is distorted in these frequency bands. Define an amplitude deviation threshold , that is, when deviates from 1 by more than , it is regarded as amplitude distortion. At the same time, the phase response is detected. The phase distortion is mainly reflected in deviating from the reference phase. A large phase shift will cause signal distortion and affect the positioning accuracy. Define a phase deviation threshold , that is, when deviates from the reference phase by more than ° in a certain frequency band, it is regarded as phase distortion.

[0038] After the acoustic signals at the respective positions meet the signal quality standard, the filtered high-quality acoustic signals are further converted into accurate position coordinates, thereby providing reliable data support for the update of the Bluetooth fingerprint database. In a specific example of the present application, position estimation is performed based on the acoustic signals at the respective positions to obtain the acoustic signal positioning coordinates at the respective positions, including: optimizing the position estimation of the time difference of arrival of the acoustic signals based on a parametric method to obtain the acoustic signal positioning coordinates at the respective positions.

[0039] Specifically, first, position estimation is performed based on the time difference of arrival of the acoustic signals to obtain the initial acoustic signal positioning coordinates at each position. It should be understood that since the signal propagation speed is known (for example, the propagation speed of sound in air is approximately 340 m / s), by comparing the time differences of arrival of signals from different base stations, the distance differences of this position relative to each base station can be calculated, and then hyperbola solving can be performed based on the distance difference information to determine the initial acoustic signal positioning coordinates at this position. The specific steps are as follows: (1) Calculate the arrival time: The acoustic signal base station emits an acoustic signal at time , and the arrival time of the positioning terminal receiving the signal is , and the propagation time between the two is expressed as: Assume that the sound wave propagates along a straight line, so the distance estimation can be simplified as: in, represents the speed of sound, To locate the terminal, is the known location of the acoustic signal base station.

[0040] (2) Calculate the arrival time difference: To eliminate the time it takes for the base station to send the acoustic signal The influence of and simplifies the positioning problem, the arrival time difference between different base stations can be calculated The specific formula is: in, and From the base station and base station The time when the signal reaches the intelligent positioning terminal. The base station can be calculated and Distance difference to the positioning terminal: TDOA between multiple base stations can establish multiple such equations, each equation representing the distance difference between the positioning terminal and two base stations.

[0041] (3) Solving the transmitter location: Assuming there are N base stations, multiple arrival time difference equations can be formed. By combining these equations, a mathematical optimization algorithm (such as the least squares method, nonlinear least squares method, etc.) can be used to solve the location coordinates of the positioning terminal, that is, to solve the initial acoustic signal positioning coordinates of each location.

[0042] The position estimation is achieved through the above-mentioned TDOA method based on acoustic signals. The positioning accuracy is high under noise-free conditions. However, the performance of this positioning method is easily affected by unfavorable environmental factors such as indoor non-line-of-sight (NLOS) and multipath propagation, which leads to deviations in the receiving timestamp and positioning inaccuracy. Figure 5 Figure 1 is a schematic diagram of TDOA positioning estimation results without error and with noise error. Figure 5 As shown, the sample geometry for N=4 is described, where For the acoustic signal base station, and They are the true and estimated location terminal positions respectively. From a geometric perspective, the noiseless TDOA equation defines a hyperbola, and the location terminal should be located on this hyperbola in a two-dimensional (2-D) space. By using at least two hyperbolas, the position of the location terminal can be obtained through the intersection point on the line. It can be observed that due to the multipath propagation effect, incorrect timestamps will be collected at the corresponding smart terminal, which may cause the hyperbola to be severely stretched, resulting in the hyperbola no longer intersecting at an exact point, greatly reducing the positioning performance.

[0043] To solve the TDOA positioning problem under adverse conditions, this application further optimizes the position estimation of the time difference of arrival of acoustic signals based on a parametric method to obtain the acoustic signal positioning coordinates of each position, that is, reshapes the relationship between the transmitter position and the TDOA measurement, and then optimizes the newly introduced parameters (each parameter is associated with the corresponding hyperbola), that is, it is necessary to search for a point on the hyperbola defined by the TDOA that is closest to all other hyperbolas. The specific steps are as follows: (1)Construct a hyperbola model: In this scenario, multiple base stations send acoustic signals, and the signals reach the location terminal. By calculating the time difference of arrival (TDOA) values between each base station through the signal arrival time received by the location terminal, the position of the location terminal can be obtained. The TDOA hyperbola equation can be written in the following form to describe the geometric relationship between the smart terminal and each base station: And the geometric relationship between each base station: Where, 、 represent the midpoint coordinates of base stations and respectively, represents the semi-major axis of the hyperbola, represents the semi-minor axis, is the rotation angle of the hyperbola. These hyperbolas define the time difference relationship between each pair of base stations and can be used to construct multiple hyperbola intersection points to infer the position of the smart terminal.

[0044] (2)Parametrize the hyperbola equation: The location terminal calculates the time difference of arrival by receiving the signals from the base stations, and based on these TDOA measurements, finds the point on the hyperbola that is closest to all base stations. To achieve this goal, the optimal point can be found through the parametric hyperbola equation: (3)Optimize the parameters: The goal is to find the points located on each hyperbola and minimize the Euclidean distance between them. Under the parametric model, the minimization problem can be defined as follows: This non - linear least - squares problem can be solved by optimizing algorithms such as gradient descent or least - squares method to obtain a set of parameters , thus obtaining the optimal points on each hyperbola.

[0045] (4) Calculate the final position estimate of the intelligent terminal: After finding the optimal points on the hyperbola , further apply the least - squares cost function to determine the final position of the intelligent terminal . At this time, the optimization goal is to minimize the error between the intersections of all hyperbolas, and the least - squares method is used to obtain the final position. The final position estimate can be solved by the following minimization problem: where, is the set of points calculated according to the parametric hyperbola equation in the previous steps, is the final estimated position of the intelligent terminal. This algorithm ensures the positioning accuracy by minimizing the error between the intersections of each hyperbola. Even in the presence of adverse environmental factors such as multipath propagation, the positioning performance of TDOA can be improved through the optimization algorithm.

[0046] In the above - mentioned method for updating the Bluetooth fingerprint database based on voiceprint quality estimation, in step S4, based on the parametric method, the deviation of the intersections of the hyperbolas of the acoustic signal positioning coordinates is calculated to generate an acoustic signal position calculation quality evaluation value. It should be understood that in the actual scenario, the position estimate of the intelligent terminal is updated in real - time, and each measurement result will be different. To avoid contaminating the Bluetooth fingerprint database with low - quality acoustic signal positioning data, this application evaluates the quality of the obtained positioning solution by the minimum value of the optimization objective function. The smaller the final value of the optimization objective function, the smaller the deviation of each point on the hyperbola, indicating that the calculated positioning points are more consistent and accurate. Specifically, the acoustic signal position calculation quality evaluation value of each position is calculated by the following formula: where, is the number of acoustic signal base stations, is the parameter of the hyperbola intersection optimized by the parametric method, , are the hyperbola indices, , are the coordinate components, is the square of the two - norm, is the acoustic signal position calculation quality evaluation value.

[0047] It should be understood that the essence of the Q value is the sum of the squares of the total distances between all hyperbola parameter points, which reflects the consistency of the hyperbola intersections. By adjusting the parameter t, all hyperbolas are made to converge at the same point as much as possible, that is, minimizing Q. When Q = 0, all hyperbolas converge at one point, and the positioning accuracy is extremely high (ideal noise-free scenario). When Q > 0, it indicates the existence of multipath or noise interference.

[0048] In the above method for updating the Bluetooth fingerprint database based on voiceprint quality estimation, in step S5, based on the comparison between the quality evaluation value calculated based on the voice signal positions at each location and the voice positioning quality threshold, it is determined whether the RSSI values at each location need to be updated. It should be understood that in actual positioning, the optimal objective function value can be used to construct a position-normalized quality score. According to the system accuracy requirements and test results, a voice positioning quality threshold is set. The threshold is set according to different application scenarios . If , it is considered that the calculated quality at this position meets the requirements, and the RSSI fingerprint database can be updated based on the voice signal positioning coordinates at this position; if Q ≥ Qth, it is considered that the calculated quality at this position does not meet the requirements, and at this time, the RSSI fingerprint database cannot be directly updated based on the voice signal positioning coordinates at this position. The positioning result of this time needs to be discarded and the voice signal position measurement needs to be redone to improve the data quality.

[0049] After obtaining the accurate voice signal positioning coordinates , based on the voice positioning coordinates at the current position and the base station positions, the RSSI value at this position is recalculated to update the Bluetooth positioning fingerprint database, that is, based on the actual distances between the current position and each Bluetooth base station, the RSSI value at the current position is calculated based on the above Friis free space path loss model , and compared with the existing RSSI values in the fingerprint database . In the actual Bluetooth positioning fingerprint process, the RSSI signal is usually affected by various factors such as environmental noise, physical obstacles such as walls, and interference between devices, resulting in random fluctuations in the RSSI value. At the same time, the RSSI fluctuates rapidly over time. Therefore, it is necessary to establish a model based on Adaptive Kalman Filter (AKF) in the state space to smooth the RSSI value and accurately fuse the current position measurement value and historical data. The specific steps are as follows: Define the state space model: The state transition equation is set to describe the dynamic change of the RSSI value over time, that is, the slow change trend of the RSSI over time and the fluctuations caused by noise. It is assumed that the change of the RSSI is relatively stable, but a certain amount of random fluctuation is allowed: wherein, is the actual RSSI value at time , is the process noise with variance .

[0050] The observation equation is set to describe the RSSI value measured each time, taking into account measurement errors and environmental noise: wherein, is the measured RSSI value, is the measurement noise with variance .

[0051] Adaptive noise adjustment: To more accurately handle the rapid fluctuations of RSSI values and environmental noise, the process noise and the measurement noise can be adaptively adjusted to make the filter more flexible. After each new RSSI measurement value is obtained, the residual (i.e., the difference between the measured value and the predicted value) is calculated: The square of the residual is used to dynamically adjust the noise covariance to adapt to the rapid fluctuations of RSSI and random noise: wherein, and are smoothing coefficients that can ensure that the noise estimate is moderately smooth without being overly sensitive.

[0052] Adaptive Kalman filtering: First, based on the estimated value of the previous time step, the current RSSI value is predicted, and based on the covariance estimate of the previous step, the current state covariance is predicted: Second, based on the predicted covariance and the covariance adjusted by the measurement noise, the Kalman gain is calculated to determine the weights of the historical and current measurement values: The historical prediction and the current measured RSSI value are fused using the Kalman gain to obtain the updated RSSI value : The updated covariance It reflects the uncertainty of the filter for the current RSSI estimation, and the next prediction will be based on this covariance: Finally obtained As the fusion result of the current location and historical measurements, it reflects the best estimate of the RSSI signal in the current state. Store it in the Bluetooth fingerprint database to replace the original historical value. This updated RSSI value is more stable and can better adapt to environmental changes and rapid fluctuations of RSSI. In this way, the Bluetooth fingerprint database can be updated in real time to ensure that the RSSI value more accurately reflects the signal propagation characteristics in the environment, especially to cope with problems such as multipath effects and signal attenuation in complex indoor environments. By updating the fingerprint database in real time, the Bluetooth positioning system can maintain a high positioning accuracy in different environments, especially in complex indoor environments, to achieve a more reliable positioning service.

[0053] Embodiment 2 In particular, considering the traditional methods for evaluating the quality of acoustic signals based on channel impulse response, peak distribution, reverberation time, and frequency distortion analysis, although they can effectively detect multipath effects, environmental reverberation, and signal distortion, fixed thresholds (such as time difference threshold, amplitude ratio threshold, reverberation time threshold) need to be preset for each analysis. In an actual complex environment, it is difficult to dynamically adjust the thresholds, resulting in an increase in the misjudgment rate. For example, sudden noise in a shopping mall may instantaneously change the reverberation time, but the fixed threshold cannot adaptively filter such interference. In addition, traditional methods only focus on local features in the time domain (peak time difference) and frequency domain (spectrum distortion), do not fully utilize the time-frequency joint characteristics of the signal (such as the time-frequency distribution pattern of transient noise), are difficult to capture the global characteristics of complex interference, and are highly sensitive to non-stationary signals (such as instantaneous multipath changes caused by people walking), and rely on multiple measurements to take the average, lacking real-time performance. Based on this, this application proposes an optimized method for jointly evaluating the quality of acoustic signals and Bluetooth signals. By introducing a deep learning algorithm to extract and classify the joint time-frequency feature patterns of acoustic signals and Bluetooth signals, and combining the characteristics of acoustic signals and Bluetooth signals, dynamic evaluation of the signal quality of both is achieved. This method can not only fully consider the time-frequency distribution characteristics of acoustic signals, but also comprehensively utilize information such as the RSSI value of Bluetooth signals. With the powerful feature learning ability and generalization ability of the deep learning model, it overcomes the problem that the fixed threshold in traditional methods cannot adapt to complex environmental interference. By jointly processing acoustic signals and Bluetooth signals, it can more accurately capture signal quality changes and respond in real time to multipath effects, noise interference, and non-stationary signal changes under different environmental conditions, significantly improving the accuracy and real-time performance of signal quality evaluation.

[0054] In Embodiment 2, a method for updating a Bluetooth fingerprint database based on voiceprint quality estimation includes: constructing a Bluetooth fingerprint database, where each Bluetooth fingerprint in the Bluetooth fingerprint database is {the RSSI value at each location, the two-dimensional coordinates at each location}; receiving the voice signals at each location; based on the RSSI value at each location and the voice signals at each location, determining whether the signal quality standard is met, and in response to meeting the signal quality standard, performing position estimation based on the voice signals at each location to obtain the voice signal positioning coordinates at each location; calculating the hyperbola intersection deviation of the voice signal positioning coordinates based on a parameterization method to generate a voice signal position calculation quality evaluation value; based on the comparison between the voice signal position calculation quality evaluation value at each location and the voice positioning quality threshold, determining whether it is necessary to update the RSSI value at each location.

[0055] Figure 6 FIG. is a flowchart for determining whether the signal quality standard is met based on the RSSI value at each location and the voice signals at each location in the method for updating a Bluetooth fingerprint database based on voiceprint quality estimation according to Embodiment 2 of the present application. As Figure 6 shown, determining whether the signal quality standard is met based on the RSSI value at each location and the voice signals at each location includes: S31, converting the voice signal into a time-frequency diagram through wavelet analysis to obtain a voice signal time-frequency diagram; S32, performing joint time-frequency feature extraction on the voice signal time-frequency diagram based on dilated convolutional coding to obtain a voice signal time-frequency feature coding diagram; S33, performing feature enhancement on the voice signal time-frequency feature coding diagram to obtain a voice signal time-frequency feature enhanced coding diagram; S34, normalizing the RSSI value of the Bluetooth signal at each location to obtain a normalized Bluetooth signal RSSI feature vector; S35, fusing the voice signal time-frequency feature enhanced coding diagram with the Bluetooth signal RSSI feature vector to form a joint time-frequency feature coding diagram; S36, based on the joint time-frequency feature coding diagram, determining whether the joint signal meets the signal quality standard.

[0056] Specifically, in step S31, the voice signal is converted into a time-frequency diagram through wavelet analysis to obtain a voice signal time-frequency diagram. It should be understood that the voice signal contains rich feature information in the time domain and the frequency domain. Traditional methods of analyzing the time domain or the frequency domain alone cannot capture the time and frequency local characteristics of the signal simultaneously. However, wavelet transform can accurately represent the transient characteristics of the signal (such as impulse noise, multipath reflection) in the time-frequency domain through multi-scale decomposition, convert the voice signal into a time-frequency diagram, and intuitively present the energy distribution of the signal at different times and frequencies, so as to clearly distinguish the effective components (such as direct sound waves) from the interference (such as multipath reflection, burst noise) in the signal. For example, the direct sound wave appears as a continuous high-energy band region in the time-frequency diagram, while the instantaneous noise presents isolated spikes.

[0057] Specifically, in step S32, time-frequency feature extraction of the acoustic signal time-frequency map based on dilated convolution coding is performed to obtain an acoustic signal time-frequency feature coding map. That is, in order to further extract representative time-frequency features from the acoustic signal time-frequency map, so as to provide more discriminative feature information for subsequent signal quality evaluation. In this application, a dilated convolutional neural network is used to perform time-frequency feature extraction on the acoustic signal time-frequency map. It should be understood that the receptive field of the traditional convolutional neural network (CNN) is limited, and it is difficult to capture long-range time-frequency dependencies (such as periodic interference). Dilated convolution can expand the receptive field without increasing the number of parameters by introducing spaced sampling, effectively extract time-frequency features in a larger context, and thus capture long-range dependencies across time-frequency (such as periodic reverberation) while retaining local details (such as signal mutation edges), and encode them into an acoustic signal time-frequency feature coding map, providing a high-information-density input for subsequent signal quality evaluation.

[0058] Specifically, in step S33, feature enhancement is performed on the acoustic signal time-frequency feature coding map to obtain an acoustic signal time-frequency feature enhanced coding map. It should be understood that considering that the acoustic signal time-frequency feature coding map may contain certain redundant information (such as random patterns of background noise) and is difficult to be directly used for accurate signal quality evaluation. Therefore, in this application, feature enhancement is further performed on the acoustic signal time-frequency feature coding map to jointly model the spatial distribution characteristics and semantic association information of the acoustic signal time-frequency feature coding map, enhance the expressiveness of key features based on cross-channel feature dependencies, and at the same time suppress the interference of non-key information to obtain a more discriminative acoustic signal time-frequency feature enhanced coding map.

[0059] Figure 7 The flowchart of step S33 in the Bluetooth fingerprint database update method based on voiceprint quality estimation according to Embodiment 2 of this application is as follows. As Figure 7 shown, step S33 includes: S331, extracting the channel feature vector at the (i, j) pixel position from the acoustic signal time-frequency feature coding map as the acoustic signal time-frequency channel feature vector to be enhanced; S332, based on the acoustic signal time-frequency feature coding map, performing information compensation coding on the acoustic signal time-frequency channel feature vector to be enhanced guided by spatial-semantic dual-domain context to obtain an implicit coding vector of the enhanced component of the acoustic signal time-frequency channel to be enhanced; S333, fusing the implicit coding vector of the enhanced component of the acoustic signal time-frequency channel to be enhanced and the acoustic signal time-frequency channel feature vector to be enhanced to obtain an enhanced acoustic signal time-frequency channel feature vector, where the enhanced acoustic signal time-frequency channel feature vector is the channel feature vector at the pixel position (i, j) of the acoustic signal time-frequency feature enhanced coding map.

[0060] In a specific example of the present application, the step S331 can be expressed by the formula: Wherein, 、 and respectively represent the height, width and number of channels of the time-frequency feature encoding map of the acoustic signal, represents the time-frequency feature encoding map of the acoustic signal, represents the channel feature vector at the (i, j) pixel position of the time-frequency feature encoding map of the acoustic signal, represents the time-frequency channel feature vector to be enhanced of the acoustic signal.

[0061] It should be understood that each pixel position of the time-frequency feature encoding map of the acoustic signal corresponds to a multi-dimensional channel feature vector, which carries local features such as the signal energy distribution and frequency domain pattern at this spatio-temporal point. However, due to sensor noise, multipath effects or environmental interference, the local features may have missing or distorted information and are difficult to accurately represent the complex time-frequency characteristics of the acoustic signal. Therefore, the present application extracts the channel feature vector at the (i, j) pixel position in the time-frequency feature encoding map of the acoustic signal as the time-frequency channel feature vector to be enhanced of the acoustic signal, so as to focus on the refined processing of local features, thereby enhancing the features of the entire time-frequency feature encoding map of the acoustic signal in a point-to-face manner.

[0062] In a specific example of the present application, the step S332 includes: First, perform n random scans on the time-frequency feature encoding map of the acoustic signal to obtain a sparse set of n channel feature vectors as the time-frequency reference feature vectors of the acoustic signal , wherein, 、 、 and respectively represent the 1st, 2nd, th and th time-frequency reference feature vectors in the sparse set of time-frequency reference feature vectors of the acoustic signal. It should be understood that the fixed receptive field of the traditional convolutional network limits its ability to capture global context information, especially in the face of complex and changing acoustic environments, and calculating the feature correlations at all positions globally will bring a high computational complexity. Therefore, the present application performs random scans on the time-frequency feature encoding map of the acoustic signal to generate a sparse set of time-frequency reference feature vectors, which can introduce spatial diversity, avoid overfitting local noise, and at the same time construct a time-frequency feature context information library at a low cost to provide rich context information guidance for subsequent feature enhancement processing.

[0063] Next, calculate the Poincaré distance between the feature vector of the channel to be enhanced in the time-frequency domain of the acoustic signal and each reference feature vector of the acoustic signal in the sparse set of reference feature vectors in the time-frequency domain of the acoustic signal to obtain a modulation matrix of the feature space of the channel to be enhanced in the time-frequency domain of the acoustic signal composed of multiple Poincaré distances, which is expressed by the formula: Among them, represents the square of the norm of the vector, represents the inverse hyperbolic cosine function, represents and the Poincaré distance between them, represents and the Poincaré distance between them, represents the modulation matrix of the feature space of the channel to be enhanced in the time-frequency domain of the acoustic signal.

[0064] It should be understood that the traditional Euclidean distance is difficult to effectively represent the hierarchical relationship in the high-dimensional feature space, while the Poincaré distance is applicable to the hyperbolic space and can better model the tree-like or hierarchical structure (such as the multipath reflection path of the acoustic signal). By quantifying the Poincaré distance between the feature vector of the channel to be enhanced in the time-frequency domain of the acoustic signal and each reference feature vector of the acoustic signal, this application constructs a modulation matrix of the feature space of the channel to be enhanced in the time-frequency domain of the acoustic signal, which can more finely depict the spatial correlation and hierarchical structure difference between the feature of the channel to be enhanced in the time-frequency domain of the acoustic signal and each reference feature. For example, the distance between the direct wave feature and the multipath reflection feature is large, while the distance between different reflection paths is small, thereby providing accurate spatial correlation information guidance for subsequent feature enhancement processing.

[0065] Then, calculate the implicit semantic association between the feature vector of the channel to be enhanced in the time-frequency domain of the acoustic signal and each reference feature vector of the acoustic signal in the sparse set of reference feature vectors in the time-frequency domain of the acoustic signal to obtain a set of semantic association coding matrices of the channel to be enhanced in the time-frequency domain of the acoustic signal, which is expressed by the formula: Among them, represents the weight matrix, represents the transpose of the vector, represents the vector multiplication, represents the normalized exponential function, represents and the semantic association coding matrix of the channel to be enhanced in the time-frequency domain of the acoustic signal between them.

[0066] Here, since the surface similarity of the time-frequency features of the acoustic signal (such as energy distribution) is difficult to reflect the deep semantic associations (such as signal category, interference type), therefore, in order to establish the implicit semantic association between the time-frequency feature vector to be enhanced of the acoustic signal and each time-frequency reference feature vector of the acoustic signal at a higher level, the present application introduces a cross-attention mechanism. By calculating the attention weights between the time-frequency feature vector to be enhanced of the acoustic signal and each time-frequency reference feature vector in the sparse set, the association degree between the two at the semantic level is revealed, so as to provide rich context semantic information guidance for subsequent feature enhancement.

[0067] Secondly, calculate the information compensation coding vectors between each time-frequency reference feature vector in the sparse set of the time-frequency reference feature vectors of the acoustic signal and the time-frequency feature vector to be enhanced of the acoustic signal to obtain a set of time-frequency feature vector to be enhanced channel information compensation coding vectors, which is expressed by the formula: , where represents the time-frequency feature vector to be enhanced channel information compensation coding vector relative to . Here, by calculating the complementary information between each time-frequency reference feature vector and the time-frequency feature vector to be enhanced of the acoustic signal, a set of time-frequency feature vector to be enhanced channel information compensation coding vectors is obtained, which helps the subsequent feature enhancement process to focus on the missing or weakened feature components at the position to be enhanced. For example, when the energy of a frequency band is missing due to occlusion, complementary information can be extracted from the reference features to ensure the pertinence of the feature enhancement process and avoid the problems of information redundancy and mismatch caused by directly using the reference features.

[0068] Furthermore, based on the time-frequency feature space modulation matrix of the time-frequency feature vector to be enhanced of the acoustic signal and the set of time-frequency feature semantic association coding matrices of the time-frequency feature vector to be enhanced of the acoustic signal, perform explicit modeling modulation aggregation coding on the set of time-frequency feature vector to be enhanced channel information compensation coding vectors to obtain the time-frequency feature vector to be enhanced channel enhancement component implicit coding vector. In a preferred example of the present application, first, based on the time-frequency feature space modulation matrix of the time-frequency feature vector to be enhanced of the acoustic signal and the set of time-frequency feature semantic association coding matrices of the time-frequency feature vector to be enhanced of the acoustic signal, perform double-domain nested constraint optimization on each time-frequency feature vector to be enhanced channel information compensation coding vector in the set of time-frequency feature vector to be enhanced channel information compensation coding vectors to obtain a set of optimized time-frequency feature vector to be enhanced channel information compensation coding vectors, which is expressed by the formula: where represents the corresponding semantic nested optimized time-frequency feature vector to be enhanced channel information compensation coding vector, representation The corresponding compensated coding vector of the time-frequency channel information of the optimized acoustic signal to be enhanced.

[0069] It should be understood that the compensated coding vector of the time-frequency channel information of the acoustic signal to be enhanced contains the compensation information of each reference feature relative to the feature of the channel to be enhanced. However, directly using this information may lead to information redundancy or inconsistency. Therefore, this application combines the set of the space modulation matrix of the time-frequency channel features of the acoustic signal to be enhanced and the semantic association coding matrix of the time-frequency channel features of the acoustic signal to be enhanced (that is, using the spatial correlation and semantic correlation between each reference feature and the feature of the channel to be enhanced) to perform double-domain nested constraint optimization on the corresponding compensated coding vector of the time-frequency channel information of the acoustic signal to be enhanced. Here, the semantic association coding matrix of the time-frequency channel features of the acoustic signal to be enhanced can be regarded as a gauge field, and the space modulation matrix of the time-frequency channel features of the acoustic signal to be enhanced can be regarded as a covariant field. Thus, this application uses the coupling effect of the gauge field and the covariant field to force the compensated coding vector of the time-frequency channel information of the acoustic signal to be enhanced to maintain a smooth transition of local neighborhood features under the covariant guidance of the space modulation matrix of the time-frequency channel features of the acoustic signal to be enhanced, and at the same time strengthen the consistency of semantic association under the gauge constraint of the semantic association coding matrix of the time-frequency channel features of the acoustic signal to be enhanced, and finally realize the optimized expression of the compensated coding vector of the time-frequency channel information of the acoustic signal to be enhanced, and enhance the robustness and accuracy of the time-frequency features of the acoustic signal.

[0070] Then, taking each semantic association coding matrix of the time-frequency channel features of the acoustic signal to be enhanced in the set of the semantic association coding matrices of the time-frequency channel features of the acoustic signal to be enhanced as a first-level mask modulation unit, and taking the space modulation matrix of the time-frequency channel features of the acoustic signal to be enhanced as a second-level mask modulation unit, perform explicit modeling modulation aggregation coding on the set of the compensated coding vectors of the time-frequency channel information of the optimized acoustic signal to obtain the hidden coding vector of the enhanced component of the time-frequency channel of the acoustic signal, which is expressed by the formula: where represents the hidden coding vector of the enhanced component of the time-frequency channel features of the acoustic signal to be enhanced.

[0071] Here, in order to balance the contributions of spatial and semantic information, the present application explicitly controls the information fusion process through hierarchical modulation of a first-level semantic mask and a second-level spatial mask. Specifically, the semantic association encoding matrix of the acoustic signal time-frequency channel features to be enhanced serves as the first-level mask modulation unit to preliminarily modulate the information compensation coding vector of the acoustic signal time-frequency channel to be enhanced, aiming to strengthen the contribution of highly correlated compensation vectors through the semantic mask. And the spatial modulation degree matrix of the acoustic signal time-frequency channel features to be enhanced serves as the second-level mask modulation unit, which is further weighted through the spatial mask, aiming to highlight the compensation effect of spatially adjacent features, thereby effectively strengthening the aggregation of compensation information beneficial to the acoustic signal time-frequency channel feature vector and suppressing the interference of irrelevant or redundant information.

[0072] In a specific example of the present application, the step S333 can be expressed by the formula: Wherein, and represent different fusion weight parameters, represents the enhanced acoustic signal time-frequency channel feature vector.

[0073] Here, by fusing the implicit coding vector of the enhanced component of the acoustic signal time-frequency channel to be enhanced and the acoustic signal time-frequency channel feature vector, on the basis of retaining the effective information of the original features, the feature expression of the channel to be enhanced can be specifically enhanced, so that the generated enhanced acoustic signal time-frequency channel feature vector can more accurately reflect the physical characteristics and semantic information of the target sound field environment, thereby improving the accuracy and robustness of acoustic signal processing.

[0074] Specifically, in step S34, the RSSI values of the Bluetooth signals at each position are normalized to obtain a normalized Bluetooth signal RSSI feature vector. In a specific example of the present application, first, the RSSI values of the Bluetooth signals at each position are collected, and these RSSI values are the signal strength measurements between the Bluetooth receiver and the device. Then, the RSSI values are normalized by the min-max normalization or standardization method to convert the RSSI values at different positions or times into feature vectors of a unified scale. Specifically, the normalized Bluetooth signal RSSI feature vector can eliminate the amplitude difference of the signal strength, enabling the Bluetooth signal features in different positions or environments to be directly compared. In this way, by normalizing the RSSI signal, the stability and comparability of the signal are ensured, which is helpful for subsequent feature fusion and quality assessment.

[0075] Specifically, in step S35, the enhanced time-frequency feature encoded map of the acoustic signal is fused with the RSSI value information of the Bluetooth signal to form a joint time-frequency feature encoded map. In a specific example of the present application, first, the enhanced time-frequency feature encoded map of the acoustic signal represents the multi-dimensional features of the acoustic signal in the time-frequency domain, including the spectral characteristics and time-varying information of the acoustic signal; while the RSSI value information of the Bluetooth signal represents the intensity information of the Bluetooth signal. To achieve the joint feature representation of the acoustic signal and the Bluetooth signal, this step fuses these two different types of signal features. Specifically, through the method of feature splicing, the enhanced time-frequency feature encoded map of the acoustic signal is spliced with the RSSI feature vector of the Bluetooth signal in the feature dimension to form a new joint feature vector. This joint feature vector simultaneously contains the time-frequency information and intensity information of the acoustic signal and the Bluetooth signal, and can provide comprehensive information for subsequent signal quality assessment. In some applications, if the weights of the two signals are different, the fusion can also be performed by the method of weighted average, making the influence of a certain signal more significant in the joint feature. In this way, by fusing the features of the acoustic signal and the Bluetooth signal to form a joint time-frequency feature encoded map, the system can utilize the information of both signals simultaneously, thereby improving the accuracy and reliability of signal quality assessment.

[0076] Specifically, in step S36, based on the joint time-frequency feature encoded map, it is determined whether the joint signal meets the signal quality standard. In a specific example of the present application, first, the joint time-frequency feature encoded map represents the fused time-frequency features of the acoustic signal and the RSSI features of the Bluetooth signal, including the multi-dimensional features of the acoustic signal and the Bluetooth signal in the time-frequency domain and the intensity information of the Bluetooth signal. To determine whether the joint signal meets the signal quality standard, this step introduces an automated quality assessment method based on a deep learning model. Specifically, the joint time-frequency feature encoded map is input into a pre-trained deep learning classifier for quality assessment.

[0077] In a specific example of the present application, first, the deep learning model is trained with a large number of signal samples of known quality to learn the feature patterns of signals of different qualities. These training samples include the joint time-frequency features of the acoustic signal and the Bluetooth signal and their corresponding quality labels (for example, high quality or low quality). During the training process, the model can automatically learn the complex non-linear relationship between the signal features and the quality standard.

[0078] When the combined time-frequency feature encoding map is input into the deep learning classifier, the classifier classifies it according to the learned feature patterns to determine whether the combined signal belongs to a high-quality signal or a low-quality signal. Specifically, the classifier outputs a probability value indicating the probability that the signal meets the predetermined quality standard. When the output value exceeds the set threshold, the combined signal is determined to be "high-quality"; otherwise, it is determined to be "low-quality". This automated process based on supervised learning helps to avoid the subjectivity and environmental dependence of relying on fixed thresholds in traditional methods. By adopting a deep learning model, the system can automatically extract complex patterns from the time-frequency features of the combined signal, evaluate the signal quality in real time, and has strong adaptability, enabling accurate evaluation of signals in different environments. In addition, the deep learning model can continuously optimize the quality evaluation criteria according to environmental changes, further improving the accuracy and reliability of the evaluation.

[0079] In summary, the method for updating the Bluetooth fingerprint database based on voiceprint quality estimation described in this application is elucidated. A base station group with both Bluetooth broadcast and acoustic signal transmission functions is deployed in the target area. The Bluetooth path loss model is used for preliminary positioning, and position estimation is performed based on the received signal strength of each location node to establish a Bluetooth fingerprint database. At the same time, acoustic signals at each location are received synchronously, and the quality of the received acoustic signals is evaluated. Effective acoustic signals are selected for TDOA positioning calculation to determine the positioning coordinates of the acoustic signals at each location, and the positioning accuracy of the acoustic signals is judged according to the position quality evaluation method. Furthermore, the Bluetooth strength at the current location is inferred based on the accurate acoustic signal position information to update the Bluetooth fingerprint database in real time. This method uses the propagation stability of acoustic signals to compensate for the direction sensitivity of Bluetooth signals, can effectively cope with the dynamic changes in the indoor environment, and realizes the adaptive maintenance of fingerprint data.

Claims

1. A method for updating a Bluetooth fingerprint database based on voiceprint quality estimation, characterized in that including: constructing a Bluetooth fingerprint database, where each Bluetooth fingerprint in the Bluetooth fingerprint database is {the RSSI values at various positions, the two-dimensional coordinates at various positions}; receiving the acoustic signals at the various positions; judging whether the acoustic signals at the various positions meet the signal quality standard, and after the acoustic signals at the various positions meet the signal quality standard, performing position estimation based on the acoustic signals at the various positions to obtain the acoustic signal positioning coordinates at the various positions; calculating the hyperbola intersection deviation of the acoustic signal positioning coordinates based on a parameterization method to generate an acoustic signal position calculation quality evaluation value; determining whether to update the RSSI values at the various positions based on the comparison between the acoustic signal position calculation quality evaluation values at the various positions and the acoustic positioning quality threshold; 2. The method for updating a Bluetooth fingerprint database based on voiceprint quality estimation according to claim 1, wherein constructing a Bluetooth fingerprint database, including: constructing a fingerprint matrix at each position; calculating the distance between each reference position RSSI vector and the observed RSSI vector in the fingerprint matrix to obtain a distance vector; selecting k reference positions corresponding to the k smallest distances from the distance vector; calculating weights for the k smallest distances and performing weighted averaging on the k reference positions based on the weights to obtain the two-dimensional coordinates of the observed position; 3. The method for updating a Bluetooth fingerprint database based on voiceprint quality estimation according to claim 1, characterized in that, judging whether the acoustic signals at the various positions meet the signal quality standard, including: performing channel impulse response analysis, signal peak distribution analysis, signal reverberation time analysis, and signal frequency distortion analysis on the acoustic signals at the various positions to determine whether the acoustic signals at the various positions meet the signal quality standard; 4. The method for updating a Bluetooth fingerprint database based on voiceprint quality estimation according to claim 1, wherein after the acoustic signals at the various positions meet the signal quality standard, performing position estimation based on the acoustic signals at the various positions to obtain the acoustic signal positioning coordinates at the various positions, including: optimizing the position estimation of the time difference of arrival of acoustic signals based on a parameterization method to obtain the acoustic signal positioning coordinates at the various positions; 5. The method for updating a Bluetooth fingerprint database based on voiceprint quality estimation according to claim 1, wherein calculating the hyperbola intersection deviation of the acoustic signal positioning coordinates based on a parameterization method to generate an acoustic signal position calculation quality evaluation value, including: calculating the acoustic signal position calculation quality evaluation values at the various positions with the following formula, and the formula is: Among them, is the number of acoustic signal base stations, is the hyperbola intersection parameter optimized by the parameterization method, , is the hyperbola index, , are coordinate components, is the square of the two-norm, is the quality evaluation value of the acoustic signal position calculation.

6. A method for updating a Bluetooth fingerprint database based on voiceprint quality estimation, characterized in that, including: constructing a Bluetooth fingerprint database, where each Bluetooth fingerprint in the Bluetooth fingerprint database is {the RSSI values at various positions, the two-dimensional coordinates at various positions}; receiving the acoustic signals at the various positions; judging whether it meets the signal quality standard based on the RSSI values at the various positions and the acoustic signals at the various positions, and in response to meeting the signal quality standard, performing position estimation based on the acoustic signals at the various positions to obtain the acoustic signal positioning coordinates at the various positions; calculating the hyperbola intersection deviation of the acoustic signal positioning coordinates based on a parameterization method to generate an acoustic signal position calculation quality evaluation value; determining whether to update the RSSI values at the various positions based on the comparison between the acoustic signal position calculation quality evaluation values at the various positions and the acoustic positioning quality threshold; 7. The method for updating a Bluetooth fingerprint database based on voiceprint quality estimation according to claim 6, wherein judging whether it meets the signal quality standard based on the RSSI values at the various positions and the acoustic signals at the various positions, including: transforming the acoustic signal into a time-frequency diagram through wavelet analysis to obtain an acoustic signal time-frequency diagram; Perform time-frequency feature extraction of the acoustic signal time-frequency diagram based on dilated convolution coding to obtain an acoustic signal time-frequency feature coding diagram; Perform feature enhancement on the acoustic signal time-frequency feature coding diagram to obtain an enhanced acoustic signal time-frequency feature coding diagram; Normalize the RSSI values of the Bluetooth signals at each position to obtain a normalized Bluetooth signal RSSI feature vector; Fuse the enhanced acoustic signal time-frequency coding diagram and the Bluetooth signal RSSI feature vector to form a joint time-frequency feature coding diagram; Based on the joint time-frequency feature coding diagram, determine whether the joint signal meets the signal quality standard.

8. The method for updating a Bluetooth fingerprint database based on voiceprint quality estimation according to claim 7, characterized in that Perform feature enhancement on the joint time-frequency feature coding diagram of the acoustic signal and the Bluetooth signal to obtain an enhanced joint time-frequency feature coding diagram of the acoustic signal and the Bluetooth signal, including: Extract the channel feature vector at the (i,j) pixel position from the joint time-frequency feature coding diagram as the joint time-frequency channel feature vector to be enhanced for the acoustic signal and the Bluetooth signal; Based on the joint time-frequency feature coding diagram, perform information compensation coding guided by spatial-semantic dual-domain context on the joint time-frequency channel feature vector to be enhanced for the acoustic signal and the Bluetooth signal to obtain a latent coding vector of the enhanced component of the joint time-frequency channel to be enhanced for the acoustic signal and the Bluetooth signal; Fuse the latent coding vector of the enhanced component of the joint time-frequency channel to be enhanced for the acoustic signal and the Bluetooth signal and the joint time-frequency channel feature vector to be enhanced for the acoustic signal and the Bluetooth signal to obtain an enhanced joint time-frequency channel feature vector for the acoustic signal and the Bluetooth signal, where the enhanced joint time-frequency channel feature vector for the acoustic signal and the Bluetooth signal is the channel feature vector at the pixel position (i,j) of the enhanced joint time-frequency feature coding diagram.

9. The method for updating a Bluetooth fingerprint database based on voiceprint quality estimation according to claim 8, wherein Based on the joint time-frequency feature coding diagram of the acoustic signal and the Bluetooth signal, perform information compensation coding guided by spatial-semantic dual-domain context on the joint time-frequency channel feature vector to be enhanced to obtain a latent coding vector of the enhanced component of the joint time-frequency channel to be enhanced, including: Randomly scan the joint time-frequency feature coding diagram and extract feature vectors of different channels; Calculate the Poincaré distance between the feature vectors to measure the similarity of each vector and generate a feature space scheduling matrix; Based on the semantic association of the feature vectors, optimize the features using information compensation coding; Perform explicit modeling modulation and aggregation coding on the optimized feature vectors to obtain the final enhanced channel feature vector.

10. The method for updating a Bluetooth fingerprint database based on voiceprint quality estimation according to claim 9, wherein Based on the set of the joint time-frequency channel feature space modulation matrix to be enhanced for the acoustic signal and the Bluetooth signal and the joint time-frequency feature semantic association coding matrix, perform explicit modeling modulation aggregation coding on the set of joint time-frequency channel information compensation coding vectors to obtain a latent coding vector of the enhanced component of the joint time-frequency channel, including: According to the above-mentioned space modulation matrix and semantic association coding matrix, impose dual-domain nested constraint optimization on each information compensation coding vector to generate an optimized set of information compensation coding vectors; Using the joint semantic association coding matrix as the primary mask modulation unit and the spatial modulation matrix as the secondary mask modulation unit, perform explicit modeling and aggregation processing on the optimized coding vector set, and finally obtain the enhanced component implicit coding vector of the joint time-frequency channel to be enhanced.

11. The method for updating a Bluetooth fingerprint database based on voiceprint quality estimation according to claim 10, wherein Based on the joint time-frequency feature enhanced coding map of the acoustic signal and the Bluetooth signal, determining whether the joint signal meets the signal quality standard includes: Inputting the joint time-frequency feature enhanced coding map of the acoustic signal and the Bluetooth signal into the classifier-based signal quality module, and through classifier analysis, determining whether the joint signal meets the preset signal quality standard.

Images