Acoustic Positioning Service And Privacy Protection Algorithm In Satellite-denied Environment

Currently,there is no mature and stable universal solution for high-precision LBS(Location-Based Services)in satellite-denied environments,making it a largely untapped field.Near-ultrasonic positioning is an emerging short-range technology with low synchronization costs,high compatibility with smart devices,independence from image acquisition,and an ability to penetrate rooms.It is the preferred solution for safe,highprecision positioning.This dissertation focuses on complex and multi-occluded satellitedenied environments,using a portable intelligent terminal platform that leverages the complementary advantages of near-ultrasound and inertial navigation technology.It addresses theoretical and practical needs for positioning data and training privacy security,aiming to provide a novel,robust positioning solution in the untapped field of indoor LBS that differs from electromagnetic wave solutions.The main contributions of dissertation are as follows:Firstly,an implementation scheme based on near-ultrasonic positioning is proposed.The signal modulation is based on indoor complex multipath reverberation and data privacy security,and is combined with mobile phone compatibility.An 18-22 k Hz Chirp signal is used as the acoustic positioning carrier,which is combined with silent intervals to form a periodic continuous frequency modulation wave signal for continuous positioning.To solve the hyperbolic equation and consider the non-Gaussian nature of the NLo S(Non-Line-ofSight)error,a maximum likelihood search algorithm based on the point cloud search space is used.For a specific long and narrow corridor environment,a dual beacon positioning algorithm based on subspace division is proposed,following the idea of dimensionality reduction.To handle the redundant information of beacons caused by the uneven distribution of acoustic positioning beacons in large spaces,an acoustic array optimization algorithm based on Delaunay subdivision is proposed according to the near-ultrasound beam angle.The feasibility and accuracy improvement effect of the proposed algorithm are verified through simulation experiments and real-world experiments.Secondly,a multi-room NLo S identification algorithm based on personalized federated learning is proposed for identifing and filtering NLo S signals present in near-ultrasound measurements,and protecting user acoustic data privacy.The near-ultrasonic signal is captured and preprocessed to obtain an audio wav file containing only one Chirp signal in each segment.Subsequently,the audio file is converted into a two-dimensional complexvalued spectrum matrix,and a deep neural network architecture is utilized to train the spectrum matrix.The residual network pre-training model is then used to achieve singleroom NLo S recognition.To address multi-room non-uniformly distributed data and training security issues,the above training framework is incorporated into a distributed neural network based on personalized federated learning.The gradient distance of personalized room model and model aggregation on the server-side weighted by information volume are employed to achieve highly robust multi-room NLo S recognition accuracy.The scalability and malicious attack resistance of this algorithm are confirmed through strange room tests and malicious attack experiments.Thirdly,an auxiliary positioning method based on a nine-axis IMU(Inertial Measurement Unit)is proposed,and an improved solution for the positioning of pedestrians holding mobile phones is presented.Due to the cost-effectiveness of mobile phone sensor devices,the inherent system errors of the accelerometer and gyroscope are first measured using Allan variance.The idea of complementary filtering is used to achieve coordinate transformation and heading update through the quaternion differential algorithm based on nine-axis Mahony,thus solving the 3D heading angle.Additionally,the uncertainty law of human non-rigid body structure for the kinematic model is analyzed,and a gait estimation algorithm based on joint threshold judgment of acceleration and velocity is proposed.In actual experiments,the effectiveness of the system error correction algorithm and the step estimation algorithm is confirmed,and the stability of coordinate transformation and heading update is proved by continuous steering experiments.As the accurate estimation of the step size involves user privacy information,the fusion algorithm includes a step size update algorithm to protect user privacy.Fourthly,a fusion positioning algorithm named Pals(Pedestrians acoustics localization system)is proposed,which integrates information from near-ultrasound and IMU sensors using Kalman filtering,and two adaptive correction cumulative error schemes are introduced.The algorithm first proposes a method to align near-ultrasound and IMU sensors,and redistributes the weight of the state update matrix based on the likelihood value to give higher credibility to direct near-ultrasound information,considering the non-Gaussian noise generated by near-ultrasonic multipath reverberation.Subsequently,two feedback algorithms,KFBPE based on Bayesian estimation and KFLS based on least squares,are proposed to eliminate the cumulative error of inertial navigation data by adaptively adjusting the step size factor in the fusion algorithm.The positioning experiments conducted in general indoor environments demonstrate that the Pals algorithm outperforms near-ultrasonic and IMU positioning alone,in terms of error and fit with the ground truth.In severe occlusion experiments,the robustness and NLo S resistance of the KFBPE algorithm are demonstrated.Graphical analysis shows that even in an environment with 50% NLo S near-ultrasonic measurements,the KFBPE algorithm can still achieve a real-time positioning error of up to 35 cm.