Positioning method and system based on radio orientation

Abstract

The invention provides a positioning method and system based on radio orientation, and relates to the technical field of wireless communication and positioning, the system comprises a multi-band signal receiving and transmitting module, an adaptive beam forming and processing module, a cooperative positioning resolving module and a calibration and learning module, and all the modules work cooperatively to realize high-precision positioning in a complex electromagnetic environment. Compared with a traditional single-frequency-band positioning system (for example, the system only depends on UWB or Bluetooth), the system covers multiple frequency bands such as Sub-1GHz, 2.4 GHz, 5.8 GHz and millimeter waves, dynamic switching can be conducted according to the environment, high frequency bands such as millimeter waves are selected in the LoS (sight distance) environment, and centimeter-level angle measurement is achieved through the high angle resolution (capable of reaching the 0.1-degree level) of the high frequency bands; the system is switched to low-frequency bands such as Sub-1GHz in an NLoS (non-line-of-sight) environment, and penetrates through obstacles by means of the strong diffraction capability of the system, thereby avoiding signal interruption, and solving the problems of low precision, poor stability and weak environmental adaptability of a traditional positioning system in a complex electromagnetic environment.

Description

Technical Field

[0001] This invention relates to the fields of wireless communication and positioning technology, and in particular to a positioning method and system based on radio orientation. Background Technology

[0002] With the rapid development of the Internet of Things, intelligent manufacturing, and other fields, higher requirements are being placed on the accuracy, stability, and environmental adaptability of radio positioning technology. Traditional radio positioning systems rely on single-frequency signals for positioning, which has obvious limitations: GPS systems are easily blocked in indoor or underground environments, leading to a sharp drop in positioning accuracy or even failure; UWB (Ultra-Wideband) technology, while offering high positioning accuracy (centimeter-level), suffers from high power consumption, limited coverage (typically less than 100 meters), and susceptibility to metal obstacles; AoA (Angle of Arrival) / AoD (Angle of Departure) technologies based on Bluetooth or Wi-Fi are significantly affected by environmental interference (such as multipath reflection and interference from devices on the same frequency band), resulting in insufficient positioning stability and making it difficult to meet the continuous high-precision positioning requirements in industrial environments.

[0003] The common shortcomings of existing technologies are: 1) strong dependence on a single signal source or frequency band, lacking full utilization of the advantages of multi-frequency signals; 2) lack of adaptability to dynamic environments, unable to adjust positioning strategies according to changes in the electromagnetic environment; 3) lack of an effective multi-node collaborative mechanism, only transmitting simple measurement results and failing to fully exploit the information value of distributed nodes; 4) the use of static calibration methods, unable to adapt to long-term environmental changes, with positioning accuracy easily decaying over time. Therefore, there is an urgent need for a new scheme that can adapt to environmental changes, integrate the advantages of multi-frequency signals, and achieve high-precision, high-robust orientation positioning through distributed node collaboration. Summary of the Invention

[0004] This invention relates to a positioning method and system based on radio orientation, which solves the problems of low accuracy, poor stability and weak environmental adaptability of traditional positioning systems in complex electromagnetic environments.

[0005] This invention provides a radio-directed positioning system, specifically including: a multi-band signal transceiver module, an adaptive beamforming and processing module, a cooperative positioning calculation module, and a calibration and learning module. The modules work together to achieve high-precision positioning in complex electromagnetic environments. The multi-band signal transceiver module has a built-in signal preprocessing unit that performs signal filtering (supports low-pass and band-pass filtering, with configurable cutoff frequency), amplification (gain range 0-60dB), and noise suppression, reducing the complexity of subsequent signal processing steps. The adaptive beamforming and processing module is used to sense the electromagnetic environment in real time, dynamically formulate positioning strategies, and realize adaptive beamforming to improve the target signal reception quality and suppress interference. The collaborative positioning and calculation module also includes a data fusion and calculation unit. The data fusion and calculation unit adopts a fusion algorithm based on maximum likelihood estimation (MLE) or factor graph optimization to jointly calculate the multi-band measurement information of the master node and the collaborative information of the auxiliary node. The algorithm explicitly models the error characteristics of different frequency band signals in the LoS / NLoS environment, including the Gaussian distribution error model in the LoS environment and the Gaussian mixture distribution error model in the NLoS environment. The optimal position estimate of the target is obtained through iterative calculation. The calibration and learning module incorporates an online calibration model based on deep neural networks. It uses historical positioning data, environmental perception data, and final positioning error as training samples to continuously optimize the parameters of each part of the system and improve the system's adaptive capabilities.

[0006] Furthermore, the multi-band signal transceiver module includes at least two independently operating radio band transceiver units, with operating frequency bands covering at least two of the Sub-1GHz, 2.4GHz, 5.8GHz and millimeter wave bands, and each radio band transceiver unit is connected to a controllable multi-band antenna array; Furthermore, the radio transceiver unit adopts a software-defined radio (SDR) architecture, supporting signal generation, modulation / demodulation, and data acquisition in different frequency bands through software configuration. The sampling rate range is 1MHz-1GHz, meeting the bandwidth requirements of different frequency band signals. The multi-band antenna array adopts a reconfigurable array structure with 4-32 array elements, supporting switching between linear and planar array layouts. The element spacing can be dynamically adjusted according to the operating frequency band (the element spacing for low-frequency bands is λ / 2-λ, and the element spacing for high-frequency bands is λ / 4-λ / 2, where λ is the wavelength of the corresponding frequency band signal) to ensure array gain and angular resolution. The multi-band antenna array supports dynamic adjustment of array parameters according to the characteristics of different frequency band signals to adapt to the transmission and reception requirements of each frequency band signal.

[0007] Furthermore, the adaptive beamforming and processing module includes an environment sensing unit, a strategy selection unit, and an adaptive beamformer. The environment sensing unit scans the channels of each frequency band in real time, collects channel parameters, including multipath intensity, interference source distribution, signal attenuation characteristics, and line-of-sight (LoS) / non-line-of-sight (NLoS) propagation states, and generates an environmental electromagnetic characteristic assessment report. Based on the assessment report output by the environment sensing unit, the strategy selection unit dynamically selects the main frequency band, auxiliary frequency band, and corresponding operating mode for positioning. In the LoS environment, it prioritizes the high-frequency band to achieve high-precision angle measurement, and in the NLoS environment, it switches to the low-frequency band to utilize its strong diffraction capability for coarse positioning, while matching the corresponding signal sampling rate, modulation method, and data transmission protocol. The adaptive beamformer, based on the strategy determined by the strategy selection unit, iteratively calculates and optimizes the weight vector of the antenna array to form a high-gain main beam in the direction of arrival of the target signal, while forming nulls in the directions of arrival of interference signals and multipath signals, maximizing the signal-to-interference-plus-noise ratio (SINR) of the received signal, and supports real-time adjustment of the weight vector according to environmental changes.

[0008] Furthermore, the collaborative positioning solution module includes at least one primary positioning node and multiple auxiliary collaborative nodes. By fusing the measurement information of the primary positioning node with the collaborative information of the auxiliary collaborative nodes, joint positioning solution is performed to obtain the target position estimate.

[0009] Furthermore, the primary positioning node, based on the adaptive beamforming signal, completes the measurement of the target's angle of arrival (AoA) and / or time difference of arrival (TDoA), recording the signal's frequency band, sampling time, and corresponding environmental electromagnetic characteristics during the measurement process. After receiving the target signal, the auxiliary cooperative node extracts the original signal characteristics (including but not limited to channel impulse response (CIR) segments, signal amplitude spectrum, and phase spectrum), or performs local AoA and Received Signal Strength Indicator (RSSI) measurements, and sends the above information to the primary positioning node through a low-latency backhaul link (supporting LoRa, Wi-Fi, or dedicated industrial bus protocols).

[0010] Furthermore, the online calibration model adopts a deep neural network architecture, including an input layer, at least two hidden layers, and an output layer. The input layer receives environmental perception data (multipath intensity, interference intensity, frequency band channel parameters), master node measurement data (raw AoA / TDoA values, measurement variance), and auxiliary node collaborative data (RSSI sequence, CIR features). The output layer outputs correction factors for correcting the parameters of the positioning solution model, including the measurement error weight coefficients for each frequency band and the correction amount for the LoS / NLoS environmental discrimination threshold.

[0011] Furthermore, the online calibration model training process adopts a combination of supervised learning and reinforcement learning. When the system obtains known ground truth (such as the target entering a preset RFID trigger point or a manually calibrated location), supervised training is performed using the positioning error (Euclidean distance between the ground truth and the estimated value) as the loss function. When there is no ground truth, reinforcement learning is performed using the consistency of the positioning results (variance of multi-node measurement results) as the reward function to ensure that the model can continue to optimize in unsupervised scenarios.

[0012] Furthermore, the online calibration model supports incremental learning, with newly collected training samples added to the training set in real time. By periodically fine-tuning the network parameters, the model avoids overfitting to historical data and adapts to long-term environmental changes (such as warehouse shelf layout adjustments and seasonal changes in the outdoor environment).

[0013] This invention provides a positioning method and system based on radio direction finding, which has the following beneficial effects: Compared to traditional single-band positioning systems (such as those relying solely on UWB or Bluetooth), this invention covers multiple frequency bands including Sub-1GHz, 2.4GHz, 5.8GHz, and millimeter waves, and can dynamically switch according to the environment: in a LoS (line-of-sight) environment, it selects high-frequency bands such as millimeter waves, utilizing their high angular resolution (up to 0.1° level) to achieve centimeter-level angle measurement; in a NLoS (non-line-of-sight) environment, it switches to low-frequency bands such as Sub-1GHz, utilizing their strong diffraction ability to penetrate obstacles and avoid signal interruption.

[0014] When performing data fusion and computation, the system of this invention explicitly models the Gaussian distribution error under the LoS environment and the Gaussian mixture distribution error under the NLoS environment. Compared with the traditional one-size-fits-all error processing method, it can more accurately match the error characteristics of different propagation scenarios. At the same time, when there is strong interference, the algorithm automatically adds an interference suppression term and eliminates the influence of interference signals through matrix operations, further improving the signal-to-interference-plus-noise ratio (SINR) by 20-30dB and ensuring the reliability of measurement data.

[0015] The system of this invention adopts a main positioning node and a multi-auxiliary collaborative node architecture. The auxiliary nodes can collect the original signal characteristics (such as CIR segments, RSSI sequences) or local measurement data of areas not covered by the main node, and supplement the main node data through a low-latency backhaul link (latency <10ms).

[0016] This invention's adaptive beamforming and processing module uses multi-band channel scanning at 100ms-1s intervals to collect parameters such as multipath intensity, interference source distribution, and LosS / NLoS status in real time, generating an electromagnetic characteristic evaluation report. The strategy selection unit dynamically matches the positioning scheme based on the report: if sudden interference is detected, it immediately switches to a less affected frequency band and adjusts the sampling rate and modulation method (e.g., switching from QAM to the more robust FSK). When LosS switches to NLoS, it automatically switches from high-frequency primary positioning to low-frequency coarse positioning and multi-band joint correction. Compared to traditional static positioning systems, this system reduces positioning interruption time from seconds to milliseconds in electromagnetic environment change scenarios (such as interference caused by industrial equipment start-up and shutdown), improving stability by over 90%. The adaptive beamformer of this invention iteratively optimizes the antenna array weight vector to form a high-gain main beam (gain increase of 10-20dB) in the target signal direction and nulls (signal attenuation of 20-30dB) in the interference / multipath direction.

[0017] This invention's online calibration model based on deep neural networks employs a combined training approach of supervised learning, reinforcement learning, and incremental learning: when ground truth is available (e.g., RFID trigger points, manual calibration), the positioning error is used as the loss function to optimize parameters; when ground truth is unavailable, the consistency of multi-node results is used as the reward function for continuous iteration; new samples are added to the training set in real time, and network parameters are periodically fine-tuned to avoid overfitting. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments will be briefly described below.

[0019] The accompanying drawings described below are only related to some embodiments of the invention and are not intended to limit the invention.

[0020] In the attached diagram: Figure 1 A block diagram of the overall system composition of the present invention is shown; Figure 2 A block diagram of the multi-band signal transceiver module of the present invention is shown; Figure 3 A block diagram of the adaptive beamforming and processing module of the present invention is shown; Figure 4 A block diagram of the collaborative positioning solution module of the present invention is shown; Figure 5 A block diagram of the calibration and learning module of the present invention is shown; List of reference numerals 1. Multi-band signal transceiver module; 101. Signal preprocessing unit; 102. Radio band transceiver unit; 103. Multi-band antenna array; 2. Adaptive beamforming and processing module; 201. Environmental perception unit; 202. Strategy selection unit; 203. Adaptive beamformer; 3. Cooperative positioning and calculation module; 301. Data fusion and calculation unit; 302. Master positioning node; 303. Auxiliary cooperative node; 4. Calibration and learning module; 401. Online calibration model; Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the described embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Example: Please refer to Figures 1 to 5 : This invention proposes a radio-directed positioning method and system, comprising: a multi-band signal transceiver module 1, an adaptive beamforming and processing module 2, a cooperative positioning calculation module 3, and a calibration and learning module 4. These modules work collaboratively to achieve high-precision positioning in complex electromagnetic environments. The multi-band signal transceiver module 1 has a built-in signal preprocessing unit 101, which performs signal filtering (supporting low-pass and band-pass filtering, with configurable cutoff frequency), amplification (gain range 0-60dB), and noise suppression, reducing the complexity of subsequent signal processing stages. The adaptive beamforming and processing module 2 is used to sense the electromagnetic environment in real time, dynamically formulate positioning strategies, and implement adaptive beamforming to improve the target signal reception quality and suppress interference. The cooperative positioning calculation module 3 also... The system includes a data fusion and calculation unit 301, which employs a fusion algorithm based on maximum likelihood estimation (MLE) or factor graph optimization to jointly calculate the multi-band measurement information of the master node and the collaborative information of the auxiliary node. The algorithm explicitly models the error characteristics of different frequency band signals in the LoS / NLoS environment, including the Gaussian distribution error model in the LoS environment and the Gaussian mixture distribution error model in the NLoS environment. The optimal position estimate of the target is obtained through iterative calculation. The calibration and learning module 4 has a built-in online calibration model 401 based on deep neural network. It uses historical positioning data, environmental perception data and final positioning error as training samples to continuously optimize the parameters of each link of the system and improve the system's adaptive capability.

[0023] The multi-band signal transceiver module 1 includes at least two independently operating radio band transceiver units 102, covering at least two of the following frequency bands: Sub-1GHz, 2.4GHz, 5.8GHz, and millimeter wave. Each radio band transceiver unit 102 is connected to a controllable multi-band antenna array 103. The radio band transceiver unit 102 adopts a software-defined radio (SDR) architecture, supporting signal generation, modulation / demodulation, and data acquisition in different frequency bands through software configuration, with a sampling rate range of 1MHz-160MHz. The multi-band antenna array 103 meets the bandwidth requirements of different frequency band signals, and adopts a reconfigurable array structure with 4-32 array elements. It supports switching between linear array and planar array layouts. The element spacing can be dynamically adjusted according to the working frequency band (the element spacing for low frequency band is λ / 2-λ, and the element spacing for high frequency band is λ / 4-λ / 2, where λ is the wavelength of the corresponding frequency band signal) to ensure array gain and angular resolution. The multi-band antenna array 103 supports dynamic adjustment of array parameters according to the characteristics of different frequency band signals to adapt to the transmission and reception requirements of each frequency band signal.

[0024] The adaptive beamforming and processing module 2 includes an environment sensing unit 201, a strategy selection unit 202, and an adaptive beamformer 203. The environment sensing unit 201 scans channels in each frequency band in real time, collecting channel parameters including multipath intensity, interference source distribution, signal attenuation characteristics, and line-of-sight (LoS) / non-line-of-sight (NLoS) propagation status, and generates an environmental electromagnetic characteristic assessment report. Based on the assessment report output by the environment sensing unit 201, the strategy selection unit 202 dynamically selects the primary frequency band, auxiliary frequency band, and corresponding operating mode for positioning in the LoS environment. The system prioritizes high-frequency bands for high-precision angle measurement, and switches to low-frequency bands in NLoS environments to utilize their strong diffraction capabilities for coarse positioning. It also matches the corresponding signal sampling rate, modulation method, and data transmission protocol. The adaptive beamformer 203 optimizes the weight vector of the antenna array through iterative calculation based on the strategy determined by the strategy selection unit 202. It forms a high-gain main beam in the direction of arrival of the target signal, while forming nulls in the directions of arrival of interference signals and multipath signals, maximizing the signal-to-interference-plus-noise ratio (SINR) of the received signal. It also supports real-time adjustment of the weight vector according to environmental changes.

[0025] The collaborative positioning and calculation module 3 includes at least one main positioning node 302 and multiple auxiliary collaborative nodes 303. By fusing the measurement information of the main positioning node 302 and the collaborative information of the auxiliary collaborative nodes 303, joint positioning and calculation are performed to obtain the target position estimate. The main positioning node 302 completes the measurement of the angle of arrival (AoA) and / or time difference of arrival (TDoA) of the target based on the signal after adaptive beamforming. During the measurement, the frequency band, sampling time and corresponding environmental electromagnetic characteristics of the signal are recorded. After receiving the target signal, the auxiliary collaborative nodes 303 extract the original signal characteristics (including but not limited to channel impulse response CIR segments, signal amplitude spectrum and phase spectrum), or perform local angle of arrival (AoA) and received signal strength (RSSI) measurements, and send the above information to the main positioning node 302 through a low-latency backhaul link (supporting LoRa, Wi-Fi or dedicated industrial bus protocols).

[0026] The online calibration model 401 employs a deep neural network architecture, comprising an input layer, at least two hidden layers, and an output layer. The input layer receives environmental perception data (multipath intensity, interference intensity, frequency band channel parameters), master node measurement data (raw AoA / TDoA values, measurement variance), and auxiliary node collaborative data (RSSI sequence, CIR features). The output layer outputs correction factors used to adjust the positioning solution model parameters, including measurement error weight coefficients for each frequency band and correction amounts for the LoS / NLoS environmental discrimination threshold. The online calibration model 401 is trained using a combination of supervised learning and reinforcement learning. When the system obtains known true values ​​(such as when a target enters a preset RFID trigger point or a manually calibrated location), supervised training is performed using the positioning error (Euclidean distance between the true value and the estimated value) as the loss function. When there are no true values, reinforcement learning is performed using the consistency of the positioning results (variance of multi-node measurement results) as the reward function to ensure that the model can continue to optimize in unsupervised scenarios. The online calibration model 401 supports incremental learning, with newly collected training samples added to the training set in real time. By periodically fine-tuning the network parameters, the model avoids overfitting to historical data and adapts to long-term environmental changes (such as warehouse shelf layout adjustments and seasonal changes in the outdoor environment).

[0027] The working principle of this embodiment: Building the basic signal channel: The multi-band signal transceiver module 1 is the interface for the system to interact with the target signal. Its core function is to provide signal transmission and reception capabilities covering multiple frequency bands and to preprocess the signal to reduce the complexity of subsequent steps. The specific working logic is as follows: Multi-band signal coverage and independent control: The multi-band signal transceiver module 1 has at least two independently operating radio frequency band transceiver units 102, covering at least two frequency bands among Sub-1GHz, 2.4GHz, 5.8GHz and millimeter wave; each transceiver unit adopts a software-defined radio (SDR) architecture, which can flexibly configure signal generation, modulation and demodulation and data acquisition parameters (sampling rate 1MHz-1GHz) through software to adapt to the bandwidth requirements of different frequency bands (such as millimeter wave requires a high sampling rate to ensure broadband signal processing, while Sub-1GHz can meet narrowband transmission with a low sampling rate); Reconfigurable antenna array adaptation: Each transceiver unit connects to a controllable multi-band antenna array 103 (4-32 elements). The array supports switching between linear and planar layouts, and the element spacing can be dynamically adjusted according to the frequency band (λ / 2-λ for low-frequency bands, λ / 4-λ / 2 for high-frequency bands, where λ is the wavelength of the corresponding frequency band). For example, when using the 2.4GHz band (λ≈12.5cm), the element spacing is set to 6-12.5cm to ensure array gain; when switching to the millimeter-wave band (e.g., 28GHz, λ≈1.07cm), the spacing is adjusted to 0.27-0.53cm to avoid signal interference between elements and ensure angular resolution. Signal preprocessing optimization: The built-in signal preprocessing unit 101 first filters the received signal (configurable low-pass / band-pass cutoff frequency) to remove clutter outside the frequency band; then it enhances the weak signal through a 0-60dB adjustable gain amplifier; finally, it reduces background noise through a noise suppression algorithm (such as adaptive noise cancellation) and outputs a "clean" signal to the subsequent beamforming module, reducing the impact of invalid data on processing efficiency.

[0028] Dynamically optimized signal reception strategy: The adaptive beamforming and processing module 2 is the core decision-making and execution unit of the system in response to complex electromagnetic environments. By sensing the environment in real time, dynamically formulating strategies, and optimizing beamforming, it achieves accurate acquisition of target signals and interference suppression. The specific process is as follows: Environmental perception: Obtaining an electromagnetic environment "snapshot": The environmental perception unit 201 performs channel scanning on all frequency bands supported by the multi-band signal transceiver module 1 at intervals of 100ms-1s, and collects key channel parameters, including multipath intensity (such as the power ratio of multipath signals to direct signals), interference source distribution (such as the frequency band, power, and direction of arrival of interference signals), signal attenuation characteristics (propagation loss values ​​of signals in different frequency bands), and line-of-sight (LoS) / non-line-of-sight (NLoS) status (judged by the peak characteristics of the channel impulse response CIR: in the LoS environment, the CIR has a clear direct peak, and in the NLoS environment, the direct peak disappears and multipath peaks overlap), and finally generates an "electromagnetic characteristic assessment report" containing the environmental status of each frequency band; Strategy Selection: Matching the Optimal Positioning Solution: Based on the above evaluation report, strategy selection unit 202 formulates a positioning strategy using the "weighted scoring method": First, angular resolution, ranging error, and signal stability are defined as core performance indicators, and weights are assigned to these indicators according to the actual scenario (e.g., high stability is required in industrial workshops, and high resolution is required for outdoor navigation) (e.g., stability weight is 0.4, resolution weight is 0.3, and ranging error weight is 0.3 in workshop scenarios). Normalize each indicator for each frequency band (e.g., normalize the angular resolution of millimeter wave from 0.1° to 100 points, and normalize 1° of 2.4GHz to 50 points), and calculate the overall performance score; The frequency band with the highest score is selected as the primary positioning frequency band (e.g., millimeter wave scores the highest in the LoS environment and is prioritized for high-precision angle measurement). The frequency band with the second highest score and a difference of more than 1 GHz from the primary frequency band is selected as the auxiliary frequency band (e.g., 2.4 GHz, used to supplement positioning when the primary frequency band signal is missing). If the scores of all frequency bands are below the threshold (e.g., 60 points), the "frequency band joint mode" is activated, and three or more frequency bands are activated at the same time to improve reliability through signal redundancy. Finally, match the corresponding parameters: when the main frequency band is millimeter wave, configure a high sampling rate (500MHz-1GHz) and quadrature amplitude modulation (QAM) to increase the amount of data; when the auxiliary frequency band is Sub-1GHz, configure a low sampling rate (1-10MHz) and frequency shift keying (FSK) to reduce power consumption, and at the same time determine the information transmission protocol of the cooperative nodes (such as LoRa for long-distance low-latency backhaul).

[0029] Beamforming: Directional acquisition of target signals: The adaptive beamformer 203 adjusts the element weight vector of the multi-band antenna array 103 according to the strategy parameters through an "iterative weight optimization algorithm" (such as the minimum mean square error LMS algorithm). High weights are assigned to the direction of arrival of the target signal (determined by preliminary AoA estimation) to form a high-gain main beam (e.g., gain increase of 10-20dB) to enhance the reception strength of the target signal; Low weights (close to 0) are assigned to the direction of arrival of interference and multipath signals to form a "null" (signal attenuation of 20-30dB) and suppress the effects of interference and multipath. When the environment changes (such as the movement of interference sources or the switch from LoS to NLoS), the beamformer recalculates the weight vector in real time and dynamically adjusts the beam direction and null position to ensure that the signal-to-interference-plus-noise ratio (SINR) of the received signal is always maximized (e.g., from the initial 10dB to 30dB).

[0030] Accurate positioning is achieved by integrating multi-node data: The collaborative positioning solution module 3 is the system's "positioning calculation center." Through the collaboration of the main node and auxiliary nodes, combined with multi-source data and error models, it calculates the final position of the target. The specific principle is as follows: Node division of labor: collecting multi-dimensional measurement data: Main positioning node 302: Receives the optimized signal after beamforming, performs angle of arrival (AoA) measurement (calculates the beam direction by the phase difference of the array signals, such as estimating the angle using the MUSIC algorithm) and / or time difference of arrival (TDoA) measurement (calculates the distance by comparing the time difference of the signals received by different array elements), and records the frequency band, sampling time, and environmental electromagnetic characteristics (such as the interference intensity during measurement) during the measurement as "basic data" for subsequent calculations; Auxiliary Coordination Node 303: After receiving the target signal, it does not directly calculate the position, but extracts the original signal features (such as CIR segments, signal amplitude spectrum / phase spectrum) or performs simple local measurements (such as received signal strength RSSI, coarse AoA). It encapsulates the data into standard frames through a low-latency backhaul link (such as industrial bus protocol, latency <10ms) and sends them to the main positioning node 302 to supplement the "data blind spot" of the main node (such as the NLoS area not covered by the main node, where the auxiliary node can provide multipath feature data).

[0031] Data fusion and solution: Eliminating errors and outputting the optimal position: The data fusion and solution unit 301 completes the positioning in three steps: "verification - modeling - iterative solution". Data validity verification: First, filter out abnormal data (such as RSSI values ​​transmitted by auxiliary nodes that exceed the normal range of ±3σ, which are determined to be invalid and discarded) to ensure the reliability of input data; Error model loading: Load the corresponding error model according to the environmental conditions. Under the Loss of Space (LoS) environment, the measurement error conforms to a Gaussian distribution (e.g., AoA error mean 0°, variance 0.05°). 2 The Gaussian error model is used; under NLoS conditions, the error is caused by multipath superposition and conforms to a Gaussian mixture distribution (e.g., a mean of 1° and a variance of 0.2° with a probability of 0.5). 2 Distribution, with a probability of 0.5, mean 2°, variance 0.5°. 2 (Distribution), using a Gaussian mixture error model; Joint calculation: Maximum likelihood estimation (MLE) or factor graph optimization algorithm is used to jointly calculate the AoA / TDoA data of the master node and the RSSI / CIR data of the auxiliary node. For example, the MLE algorithm maximizes the "match probability between the measured data and the theoretical position" and iterates 10-50 times (until the difference between two iterations is <0.1m) to output the optimal position estimate of the target. If there is strong interference, the algorithm automatically adds an "interference suppression term" (such as eliminating the influence of interference signals on the measurement matrix by matrix inversion) to further reduce the error. Confidence verification: After the solution is completed, calculate the confidence level of the positioning result (e.g., the consistency variance of multi-node data; the smaller the variance, the higher the confidence level). If the confidence level is lower than the threshold (e.g., variance > 0.5m), the confidence level is verified. 2 This triggers the system to re-execute the "environmental perception-policy selection-measurement-solution" process until the confidence level is met.

[0032] Continuous optimization of system parameters: The calibration and learning module 4 continuously corrects system parameters and adapts to environmental changes through an online calibration model 401 based on a deep neural network. The specific principle is as follows: The 401 architecture of the online calibration model: Constructing a parameter correction "mapping relationship": The model adopts a deep neural network structure. The input layer receives three types of data: environmental perception data (multipath intensity, interference intensity, channel parameters), master node measurement data (raw values ​​of AoA / TDoA, measurement variance), and auxiliary node collaborative data (RSSI sequence, CIR features). At least two hidden layers are set in the middle (e.g., 64 neurons in the first layer and 32 neurons in the second layer), and data features are extracted through the ReLU activation function. The output layer outputs "correction factors", including the measurement error weight coefficients of each frequency band (e.g., the error weight of millimeter wave is adjusted from 0.2 to 0.15 to reduce its error impact in the NLoS environment) and the correction amount of the LoS / NLoS environment discrimination threshold (e.g., the original discrimination threshold is adjusted from the CIR peak power ratio of 0.8 to 0.7 to improve the NLoS recognition accuracy).

[0033] Model training: Combining supervised and reinforcement learning to ensure optimization results. Supervised learning (with truth value scenarios): When the system obtains known truth values ​​(such as the target entering a preset RFID trigger point with known location coordinates; or the target location being manually calibrated), the network parameters are updated using the "positioning error" (the Euclidean distance between the truth value and the estimated value) as the loss function (such as mean square error MSE) through the gradient descent algorithm, so that the correction factor output by the model can minimize the positioning error (such as reducing the error from 1m to 0.5m). Reinforcement learning (no truth reference): When there is no truth reference, the reward function is "consistency of positioning results" (e.g., the smaller the variance of multi-node measurement results, the higher the reward value). The parameters are optimized through the Q-learning algorithm to ensure that the model can continue to adjust the correction factor in unsupervised scenarios (e.g., outdoor dynamic environment) to avoid error accumulation. Incremental Learning: Adapting to Long-Term Environmental Changes: The model supports incremental learning. Newly collected training samples (environmental, measurement, and solution data generated in each positioning process) are added to the training set in real time. After the system completes 100-1000 positioning processes, the model is "fine-tuned" once (only some network parameters are updated, such as the output layer weights). This avoids overfitting of the model to historical data (such as early indoor data causing the model to fail in outdoor scenarios) and adapts to long-term environmental changes (such as changes in multipath distribution caused by warehouse shelf adjustments and signal attenuation caused by seasonal changes), ensuring the long-term stability of the system's positioning accuracy.

[0034] The above four modules form a complete closed loop through "data flow and parameter feedback," and the specific collaborative logic is as follows: Multi-band signal transceiver module 1 outputs pre-processed signal → Adaptive beamforming module senses the environment and formulates strategies, optimizes the beam and then transmits parameters back to the transceiver module (such as adjusting the antenna array layout) → The beamformed signal is sent to the main / auxiliary nodes respectively. The main positioning node 302 and the auxiliary cooperative node 303 collect measurement data → the cooperative positioning solution module 3 fuses the data and solves the position, and at the same time sends the solution error and environmental data to the calibration and learning module 4; The calibration and learning module 4 generates correction factors through the online model, which are then fed back to the adaptive beamforming module (to adjust the weight coefficients of the strategy selection) and the cooperative positioning solution module 3 (to correct the error model parameters). The corrected module re-executes the next round of the positioning process, realizing continuous iteration of "perception-decision-execution-optimization", and ultimately maintaining meter-level or even centimeter-level positioning accuracy even in complex electromagnetic environments (such as multiple interferences, NLoS, frequency band blockage).

Claims

1. A positioning system based on radio direction finding, characterized in that, include: The multi-band signal transceiver module (1), the adaptive beamforming and processing module (2), the collaborative positioning solution module (3) and the calibration and learning module (4) work together to achieve high-precision positioning in complex electromagnetic environments. The multi-band signal transceiver module (1) has a built-in signal preprocessing unit (101) to complete signal filtering, amplification and noise suppression, reducing the complexity of subsequent signal processing steps; The adaptive beamforming and processing module (2) is used to sense the electromagnetic environment in real time, dynamically formulate positioning strategies and realize adaptive beamforming to improve the target signal reception quality and suppress interference. The collaborative positioning solution module (3) also includes a data fusion and solution unit (301). The data fusion and solution unit (301) adopts a fusion algorithm based on maximum likelihood estimation or factor graph optimization to jointly solve the multi-band measurement information of the master node and the collaborative information of the auxiliary node. The algorithm explicitly models the error characteristics of different frequency band signals in the LoS / NLoS environment, including the Gaussian distribution error model in the LoS environment and the Gaussian mixture distribution error model in the NLoS environment. The optimal position estimate of the target is obtained through iterative calculation. The calibration and learning module (4) has a built-in online calibration model (401) based on deep neural networks. It uses historical positioning data, environmental perception data and final positioning error as training samples to continuously optimize the parameters of each link of the system and improve the system's adaptive capability.

2. The positioning system based on radio direction finding according to claim 1, characterized in that, The multi-band signal transceiver module (1) includes at least two independently operating radio band transceiver units (102), with operating frequency bands covering at least two of the Sub-1GHz, 2.4GHz, 5.8GHz and millimeter wave bands. Each radio band transceiver unit (102) is connected to a controllable multi-band antenna array (103).

3. The positioning system based on radio orientation according to claim 2, characterized in that, The radio frequency band transceiver unit (102) adopts a software-defined radio architecture, which supports the generation, modulation and demodulation and data acquisition of signals in different frequency bands through software configuration. The sampling rate range is 1MHz-1GHz, which meets the bandwidth requirements of different frequency band signals. The multi-band antenna array (103) adopts a reconfigurable array structure with 4-32 array elements. It supports switching between linear array and planar array layouts. The spacing between array elements can be dynamically adjusted according to the working frequency band to ensure array gain and angular resolution. The multi-band antenna array (103) supports dynamic adjustment of array parameters according to the characteristics of different frequency band signals to adapt to the transmission and reception requirements of each frequency band signal.

4. The positioning system based on radio direction finding according to claim 3, characterized in that, The adaptive beamforming and processing module (2) includes an environment sensing unit (201), a strategy selection unit (202), and an adaptive beamformer (203). The environment sensing unit (201) scans the channels of each frequency band in real time, collects channel parameters, including multipath intensity, interference source distribution, signal attenuation characteristics, and line-of-sight / non-line-of-sight propagation status, and generates an environmental electromagnetic characteristic assessment report. The strategy selection unit (202) dynamically selects the main frequency band, auxiliary frequency band and corresponding working mode based on the evaluation report output by the environment perception unit (201). In the LoS environment, it prioritizes the high frequency band to achieve high-precision angle measurement, and switches to the low frequency band in the NLoS environment to utilize its strong diffraction capability for coarse positioning. At the same time, it matches the corresponding signal sampling rate, modulation method and data transmission protocol. The adaptive beamformer (203) optimizes the weight vector of the antenna array by iteratively calculating according to the strategy determined by the strategy selection unit (202). It forms a high-gain main beam in the direction of arrival of the target signal, and forms nulls in the direction of arrival of interference signals and multipath signals to maximize the signal-to-interference-plus-noise ratio of the received signal. It also supports real-time adjustment of the weight vector according to environmental changes.

5. The positioning system based on radio direction finding according to claim 4, characterized in that, The collaborative positioning solution module (3) includes at least one main positioning node (302) and multiple auxiliary collaborative nodes (303). By fusing the measurement information of the main positioning node (302) and the collaborative information of the auxiliary collaborative nodes (303), joint positioning solution is performed to obtain the target position estimate.

6. The positioning system based on radio orientation according to claim 5, characterized in that, The main positioning node (302) completes the measurement of the angle of arrival and / or time difference of arrival of the target based on the signal after adaptive beamforming. During the measurement, the frequency band, sampling time and corresponding environmental electromagnetic characteristics of the signal are recorded. After receiving the target signal, the auxiliary cooperative node (303) extracts the original signal characteristics or performs local angle of arrival and received signal strength measurement, and sends the above information to the main positioning node (302) through a low-latency backhaul link.

7. The positioning system based on radio orientation according to claim 6, characterized in that, The online calibration model (401) adopts a deep neural network architecture, which includes an input layer, at least two hidden layers and an output layer. The input layer receives environmental perception data, master node measurement data and auxiliary node collaborative data. The output layer outputs correction factors for correcting the parameters of the positioning solution model, including the measurement error weight coefficients of each frequency band and the correction amount of the LoS / NLoS environmental discrimination threshold.

8. The positioning system based on radio orientation according to claim 7, characterized in that, The training process of the online calibration model (401) adopts a combination of supervised learning and reinforcement learning. When the system obtains the known true value, it performs supervised training with the positioning error as the loss function; when there is no true value, it performs reinforcement learning with the consistency of the positioning result as the reward function, so as to ensure that the model can continue to optimize in unsupervised scenarios.

9. The positioning system based on radio orientation according to claim 8, characterized in that, The online calibration model (401) supports incremental learning. Newly collected training samples are added to the training set in real time. By periodically fine-tuning the network parameters, the model avoids overfitting to historical data and adapts to long-term environmental changes.

10. The positioning method based on a radio orientation positioning system as described in any one of claims 1-9, characterized in that, Includes the following steps: Step 1. Environmental perception and feature acquisition: After the system is powered on, the environmental perception unit (201) in the adaptive beamforming and processing module (2) scans all frequency bands supported by the multi-band signal transceiver module (1) with a scanning time interval of 100ms-1s, collects the channel parameters and propagation status of each frequency band, generates an environmental electromagnetic feature matrix, and stores it in the system's local cache. Step 2. Dynamic formulation of positioning strategy. The strategy selection unit (202) reads the environmental electromagnetic feature matrix generated in Step 1, evaluates the positioning performance index of each frequency band through the preset decision algorithm, determines the main positioning frequency band, auxiliary frequency band and corresponding working mode, and outputs the initial weight vector configuration parameters of the adaptive beamformer (203) and the information transmission strategy of the cooperative node. Define positioning performance evaluation indicators, including angular resolution, ranging error, and signal stability, and assign weight coefficients to each indicator; Normalize each performance index for each frequency band and calculate the overall performance score for each frequency band; The frequency band with the highest overall performance score is selected as the primary positioning frequency band, and the frequency band with the second highest score and a frequency difference greater than 1 GHz from the primary frequency band is selected as the auxiliary frequency band. If the overall performance score of all frequency bands is lower than the preset threshold, the frequency band joint mode is activated, and three or more frequency bands are used for positioning to improve signal redundancy. Step 3. Beamforming and signal measurement: The adaptive beamformer (203) adjusts the element weights of the multi-band antenna array (103) according to the configuration parameters output in Step 2 to form a directional beam. The main positioning node (302) performs AoA / TDoA measurements based on the shaped signal, with 5-20 measurements taken and the average of the multiple measurements taken as the preliminary measurement result. At the same time, the auxiliary coordination node (303) collects the target signal characteristics or performs local measurements according to the information transmission strategy, encapsulates the data into standard data frames, and sends them to the main positioning node (302) through the backhaul link. Step 4. Multi-source data fusion and calculation, collaborative positioning and calculation module (3) The data fusion and calculation unit (301) receives the preliminary measurement results of the master node and the collaborative data of the auxiliary node. First, it verifies the validity of the data, then loads the pre-modeled error model, and uses the maximum likelihood estimation or factor graph optimization algorithm for joint calculation. The number of iterations is 10-50 times until the calculation results converge, and the final position estimate of the target is output. When the auxiliary node transmits the original signal features, the data fusion and solution unit (301) first performs time delay estimation on the CIR segment, extracts the propagation time delay information of the target signal, and calculates additional TDoA measurement values ​​by combining the known positional relationship between nodes and adding them to the fusion solution; When strong interference exists in the environment, the algorithm automatically adds an interference suppression term and eliminates the influence of interference signals on the measurement results through matrix operations; After the calculation is completed, the confidence level of the positioning result is output. When the confidence level is lower than the preset threshold, the system is triggered to re-execute steps one to four until the confidence level meets the requirements. Step 5. Model calibration and parameter optimization. The calibration and learning module (4) reads the environmental perception data from Step 1, the original measurement data from Step 3, and the final positioning results from Step 4, and inputs them as training samples into the online calibration model (401). The model parameters are updated through the gradient descent algorithm to generate new correction factors. The correction factors are sent to the adaptive beamforming and processing module (2) and the collaborative positioning solution module (3) in real time to adjust the beamforming weight vector calculation parameters and the error model parameters of the fusion algorithm, and complete a closed-loop optimization. After the system completes 100-1000 positioning processes, it performs an incremental training on the online calibration model (401) to further improve the model accuracy.

Images