LED lighting source distinguishing method based on wideband hardware fingerprint information

Abstract

The application discloses an LED lighting light source distinguishing method based on wideband hardware fingerprint information, comprising: numbering a plurality of lighting LEDs, and using a wideband photoelectric sensor to collect time domain waveforms of different numbered LED light sources respectively; in an offline stage, time-frequency feature extraction is performed on the time domain waveforms collected by the wideband photoelectric sensor, and LED numbers are labeled to form an LED hardware fingerprint library; in an online stage, a terminal is close to a certain in-library LED light source, the time domain waveform of the LED light source is collected, time-frequency features are extracted, the time-frequency features are compared with the LED hardware fingerprint library, and the number of the LED light source is determined. The application can distinguish lighting light sources only by collecting high-frequency light intensity fluctuations without changing the lighting LED hardware and without relying on imaging technology, has the advantages of low cost, stable features and long-term reliability, and can be widely applied to personnel positioning, robot navigation, smart home and other fields, and solves the positioning demand in low-cost, privacy-sensitive or electromagnetic denial scenarios.

Description

Technical Field

[0001] This invention belongs to the field of wireless positioning technology and relates to the optimization of visible light communication positioning technology, specifically to a method for distinguishing LED lighting sources based on broadband hardware fingerprint information. Background Technology

[0002] With the rapid development of the Internet of Things (IoT) and intelligent sensing technologies, indoor positioning technology has shown broad application prospects in areas such as personnel navigation, asset tracking, autonomous robot movement, and smart home control. Currently, mainstream indoor positioning solutions mainly include positioning technologies based on radio frequency signals (such as Wi-Fi, Bluetooth, and UWB) and positioning technologies based on wireless optical communication (VLC) or optical feature recognition.

[0003] Existing indoor positioning methods based on radio frequency signals, while possessing a certain coverage capability, generally suffer from low positioning accuracy and susceptibility to multipath interference. Particularly in complex indoor environments, metal obstacles and pedestrian movement cause signal attenuation and fluctuations, making it difficult to achieve sub-meter level high-precision positioning. Furthermore, to achieve positioning functionality, a large number of dedicated positioning base stations or beacon nodes are typically required to be deployed in the indoor environment. This not only increases the system's deployment cost and maintenance complexity but also limits its rapid promotion and application in existing buildings.

[0004] To overcome the limitations of radio frequency positioning, positioning technology based on visible light communication has emerged. However, most existing wireless optical positioning technologies employ active modulation, which requires hardware modifications to existing LED lighting sources to embed specific driving circuits to transmit modulated signals carrying location information (such as ID codes). This "light source modification" approach has significant drawbacks: firstly, the modification of large-scale existing lighting facilities is a massive undertaking with high costs; secondly, active modulation may introduce light flicker perceptible to the human eye, affecting lighting quality.

[0005] Another type of passive light positioning technology, which does not require modification of the light source, attempts to use the inherent physical characteristics of the light source (such as color temperature, center wavelength, and light intensity distribution) as a "fingerprint" for identification. However, the spectral characteristics of LED light sources are highly sensitive to temperature and time. As the operating time of the lamp increases, the junction temperature rises, causing a redshift in the emission wavelength and a drift in the color temperature. Furthermore, the inconsistent aging levels of different lamps lead to differences in light intensity decay. This instability makes fingerprints based on color temperature or wavelength easily confused, causing the system to struggle to distinguish between different light sources, ultimately resulting in positioning failure or a significant decrease in accuracy.

[0006] Therefore, it is evident that existing indoor positioning technologies using LED light sources have three core defects: First, positioning schemes based on radio frequency signals such as Wi-Fi, Bluetooth, and UWB are susceptible to interference from multipath propagation, metal obstructions, and pedestrian movement, resulting in poor positioning accuracy. Furthermore, they require a large number of new dedicated base stations, leading to high deployment and maintenance costs. Second, visible light communication active modulation positioning requires the modification of all LED lighting drive circuits, resulting in a large amount of engineering work for existing lighting fixtures. The additional modulation signals can easily produce visible flicker, disrupting the basic lighting experience. Third, existing passive optical fingerprint recognition relies on color temperature, emission wavelength, and static light intensity as distinguishing features. These parameters are significantly affected by LED junction temperature and the aging of the lighting fixtures. Temperature drift and light decay cause continuous fingerprint feature drift, making it highly susceptible to confusion and positioning failure in multi-light source scenarios.

[0007] Meanwhile, radio frequency (RF) devices cannot be used in electromagnetic exclusion and privacy-sensitive scenarios such as confidential spaces and medical computer rooms. Existing technologies are insufficient to meet the needs for low-cost, RF-free, lamp-free, and long-term stable light source differentiation and indoor positioning. Summary of the Invention

[0008] Purpose of the invention: In order to overcome the shortcomings of the existing technology, this invention provides an LED lighting source differentiation method based on broadband hardware fingerprint information. This method does not require modification of the original LED hardware or camera imaging. It relies on the inherent high-frequency light intensity fluctuations of the LED driving circuit to extract stable hardware fingerprints, thereby achieving accurate differentiation of multiple LED light sources and supporting high-precision indoor perception and positioning in various scenarios.

[0009] Technical Solution: To achieve the above objectives, this invention provides a method for distinguishing LED lighting sources based on broadband hardware fingerprint information, comprising the following steps: S1: Number multiple lighting LEDs and use a broadband photoelectric sensor to collect the time-domain waveforms of LEDs with different numbers; S2: In the offline stage, the time-frequency features of the time-domain waveforms acquired by the broadband photoelectric sensor are extracted and their corresponding LED numbers are labeled to form an LED hardware fingerprint database. S3: In the online phase, the terminal approaches a stock LED light source, collects the time-domain waveform of the LED light source, extracts the time-frequency features, compares the time-frequency features with the LED hardware fingerprint database, and determines the number of the LED light source.

[0010] Furthermore, the time-domain waveform in step S1 possesses two types of inherent stable fingerprint characteristics: the first type is the short-period high-frequency oscillation frequency of the waveform. f 1, f 1 itself undergoes slow, long-period changes over time; the period of this slow change is denoted as . f 2. Satisfy f 1> f 2; The second category is high frequency.f 1. The waveform is constant, but the amplitude and peak-to-peak position fluctuate periodically. The fluctuation period is denoted as . f 3.

[0011] Furthermore, in step S2, short-time Fourier transform is used as the time-frequency analysis method for time-frequency feature extraction, specifically including: A1: Windowed framing: Dividing a one-dimensional time-domain waveform into multiple short time frames; A2: Fourier Transform: Perform a Fast Fourier Transform on each frame of signal to obtain the spectrum of that frame; A3: Constructing a time-frequency diagram: Arrange the spectra of all frames in chronological order to form a two-dimensional time-frequency diagram.

[0012] Furthermore, in step S2, physical feature parameters are extracted based on the two-dimensional time-frequency diagram, and the physical feature parameters are combined to form a multi-dimensional feature parameter vector, which serves as the hardware fingerprint identifier of the LED hardware fingerprint library.

[0013] Furthermore, the physical characteristic parameters in step S2 include: Small-scale high frequency f 1: The main frequency where energy is most concentrated in the instantaneous frequency diagram, corresponding to the switching frequency of the LED driver circuit; Large-scale slow-varying frequency f 2: Main frequency f 1. Periodicity that changes over time; Time-domain amplitude fluctuation frequency f 3: That is, the period of change of waveform amplitude or peak-to-peak value over time; Harmonic power ratio: Calculate the ratio of the power of each harmonic to the power of the fundamental frequency; Harmonic phase difference: Calculate the phase difference between each harmonic; Effective harmonic count: The number of harmonics whose power exceeds a certain threshold; Harmonic time-varying trajectory: Records the trajectory characteristics of the frequency and power of each harmonic changing over time.

[0014] Furthermore, in step S2, the two-dimensional time-frequency graph is input into the machine learning module to automatically generate an abstract feature vector without a clear physical meaning, and the abstract feature vector is used as the hardware fingerprint identifier of the LED hardware fingerprint library.

[0015] Furthermore, the time-frequency feature extraction method in step S3 adopts the same extraction method as in step S2 to generate a real-time feature vector; the real-time feature vector is compared with the feature vectors of all samples in the LED hardware fingerprint database, and the number corresponding to the sample with the highest matching degree is the current LED light source number.

[0016] Furthermore, in step S3, cosine similarity is used as the matching metric for feature vectors; for two feature vectors X and Y, the cosine similarity calculation formula is: cosθ = (X·Y) / (||X|| × ||Y||), and the range of cosine similarity is [-1, 1].

[0017] Furthermore, the LED light source number determined in step S3 is used for indoor personnel positioning.

[0018] Beneficial effects: Compared with the prior art, the present invention has the following advantages: 1. No need to modify existing lighting hardware, extremely low deployment cost: Fully reuse existing indoor LED lights and original switch current source drive circuits, no need to add modulation circuits or replace lights, and no need for large-scale construction in a large number of existing scenarios, greatly reducing implementation costs and construction period.

[0019] 2. No radio frequency radiation, suitable for special restricted scenarios: It only collects visible light intensity fluctuation signals throughout the process, without emitting or receiving radio frequency electromagnetic waves. It can be used in electromagnetic denial and privacy-sensitive scenarios such as confidential places, medical equipment rooms, and electromagnetic shielding laboratories, avoiding the problems of radio frequency signal leakage and interference with precision equipment.

[0020] 3. Strong fingerprint feature stability and high recognition accuracy: Hardware fingerprints are derived from the inductance, capacitance and inherent electrical characteristics of the chip in the driving circuit. Unlike traditional color temperature and wavelength fingerprints, they are not affected by LED junction temperature rise, long-term aging of lamps, light intensity decay and changes in ambient temperature. There is no feature drift after long-term use. The accuracy of differentiation is high in the environment where multiple LEDs coexist. There will be no problems of light source confusion and positioning failure.

[0021] 4. Simple hardware structure, lightweight terminal and easy to popularize: The identification terminal only requires a broadband photoelectric sensor, a simple analog-to-digital conversion and signal processing unit, without the need for a camera or imaging module. The device is small in size, has low power consumption and low hardware procurement cost, and can be integrated into various portable devices such as smart bracelets, mobile robots, and smart home sensing terminals.

[0022] 5. No flicker interference, no damage to lighting function: This invention only passively collects the naturally existing high-frequency driving fluctuations of LEDs, without applying additional modulation signals to the LEDs, and will not produce light flicker that can be perceived by the human eye, thus completely preserving the original lighting effect of the lamp and taking into account both perception function and lighting comfort.

[0023] 6. Wide application coverage, supporting multiple intelligent sensing services: Relying on the precise LED light source's ability to distinguish, it can realize high-precision positioning of indoor people, classification of indoor environmental areas, autonomous robot navigation, smart home zone control, asset tracking and other services, and is suitable for most indoor scenarios such as shopping malls, office buildings, homes, factories, and archives.

[0024] 7. Offline database creation once, online low-computing-power recognition: Fingerprint acquisition and modeling only need to be completed offline once. The online recognition stage only performs time-frequency transformation and simple vector comparison, with low computational overhead. Low-end embedded devices can complete the light source number recognition in real time, which is convenient for large-scale batch deployment. Attached Figure Description

[0025] Figure 1 This is a flowchart of the method of the present invention; Figure 2 Time-frequency domain plots for different LEDs; Figure 3 This is a time-domain plot for different LEDs. Detailed Implementation

[0026] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. After reading this invention, any modifications of the invention in various equivalent forms by those skilled in the art will fall within the scope defined by the appended claims.

[0027] Example 1:

[0028] like Figure 1 As shown, this embodiment provides a method for distinguishing LED lighting sources based on broadband hardware fingerprint information, including the following steps: S1: Number multiple lighting LEDs and use a broadband photoelectric sensor to collect the time-domain waveforms of LEDs with different numbers; S2: In the offline stage, the time-frequency features of the time-domain waveforms acquired by the broadband photoelectric sensor are extracted and their corresponding LED numbers are labeled to form an LED hardware fingerprint database. S3: In the online phase, the terminal approaches a stock LED light source, collects the time-domain waveform of the LED light source, extracts the time-frequency features, compares the time-frequency features with the LED hardware fingerprint database, and determines the number of the LED light source.

[0029] This embodiment applies the above method in a specific way, as follows: I. Hardware System Composition The hardware system of this embodiment includes the following components: (1) LED Lighting Source and Driving Circuit: Multiple sets of LED lamps are deployed in the indoor lighting scene, each set of LED lamps is driven by an independent switching current source. In this embodiment, the switching current source adopts a Buck-Boost topology and includes core components such as power inductors, filter capacitors, and driving control chips. Due to the discreteness of the manufacturing process, there are slight differences in parameters such as inductance value, capacitance value, and internal oscillator frequency in the driving circuits of different LED lamps. These differences result in the unique high-frequency fluctuation characteristics of the LED output light intensity, which constitutes the basis of the hardware fingerprint identification of this invention.

[0030] In this embodiment, the switching frequency of the LED driver circuit is... f 1. The frequency range is 100kHz to 500kHz, which is much higher than the flicker frequency perceptible to the human eye (usually below 200Hz), therefore it will not affect the lighting effect. The slow-changing period of the switching frequency. f 2. The range is 1Hz to 1kHz; if it is another type of switching circuit, f 2 may remain unchanged, peak position period f 3. The range is 1Hz to 1kHz.

[0031] (2) Broadband photoelectric sensor: used to acquire the time-domain waveform of the light intensity output from the LED light source. This embodiment uses a silicon-based photodiode with a response wavelength range covering the visible light band (400nm~700nm) and a bandwidth of not less than 10kHz, ensuring complete acquisition of the high-frequency light intensity fluctuation signal generated by the LED driving circuit. The photoelectric sensor is equipped with a focusing lens, which can improve the light signal acquisition efficiency and enhance the receiving sensitivity of the LED light source in a specific direction.

[0032] (3) Analog-to-digital converter (ADC): Used to convert the analog light intensity signal output by the photoelectric sensor into digital waveform data. This embodiment uses a 12-bit high-speed ADC with a sampling rate of 2MS / s (2 million samples per second). According to the Nyquist sampling theorem, this sampling rate is sufficient to accurately reproduce high-frequency signals within 1MHz. The ADC front end is equipped with a signal conditioning circuit, including a transimpedance amplifier and a low-pass filter, used to convert the current signal of the photodiode into a voltage signal and filter out out-of-band noise.

[0033] (4) Signal Processor: Used for time-frequency analysis and feature extraction of the acquired waveform data. In this embodiment, an ARM Cortex-M7 series microcontroller is used as the core processing unit with a main frequency of 400MHz. It is equipped with sufficient RAM and Flash storage space for running time-frequency analysis algorithms and storing fingerprint database data. The signal processor supports signal processing algorithms such as short-time Fourier transform and wavelet transform, and can complete feature extraction and fingerprint matching operations in real time.

[0034] (5) Terminal identification device: A portable device integrating the above-mentioned photoelectric sensor, analog-to-digital converter and signal processor, which can be integrated into various terminals such as smart bracelets, mobile robots, and smart home sensors. The terminal device also includes a power management module, a data storage module and a communication interface, which supports uploading the identification results to the upper-level system.

[0035] II. Offline fingerprint database construction phase The goal of the offline phase is to establish an LED hardware fingerprint database for all LED light sources to be identified indoors. The specific steps are as follows: 1) LED light source numbering and waveform acquisition First, all LED lighting sources to be identified indoors are uniquely numbered. In this embodiment, it is assumed that there are N LED light sources indoors, numbered sequentially as LED_1, LED_2, ..., LED_N. The numbering information corresponds one-to-one with the physical location of the LED and can be stored in a location mapping table for subsequent positioning services.

[0036] Then, the time-domain waveforms of each LED light source were acquired using a calibrated broadband photoelectric sensor. During acquisition, the photoelectric sensor was positioned directly in front of the LED light source, maintaining a distance of 0.5m to 2m to ensure sufficient light signal intensity was collected. Each acquisition session for each LED light source lasted 1 second, with a sampling rate of 2MS / s, thus obtaining data from 2 million sampling points per session. To improve the robustness of fingerprints, each LED light source was acquired 10 times, collecting waveform data from different time periods to cover characteristic variations under different operating temperatures and brightness levels.

[0037] The acquired analog light intensity signal is converted into digital waveform data by an analog-to-digital converter and then transmitted to a signal processor for storage. The raw waveform data can be preprocessed, including DC component removal, baseline correction, and bandpass filtering, to eliminate ambient light interference and DC offset in the circuit.

[0038] 2) Time-frequency analysis and feature extraction Time-frequency analysis is performed on the time-domain waveform of each LED to extract its unique time-frequency features. This embodiment uses Short-Time Fourier Transform (STFT) as the time-frequency analysis method, and the specific implementation is as follows: (1) Windowing and framing: The one-dimensional time-domain waveform is divided into multiple short frames with a frame length of 1024 sampling points and a frame shift of 256 sampling points (i.e., 768 sampling points overlap between adjacent frames). The Hamming window is used as the window function to reduce spectral leakage.

[0039] (2) Fourier Transform: Perform a Fast Fourier Transform (FFT) on each frame of signal to obtain the spectrum of that frame. The frequency resolution of a 1024-point FFT is 2MS / s / 1024 ≈ 1953Hz, which can distinguish the switching frequency and its harmonics.

[0040] (3) Constructing a time-frequency diagram: Arrange the spectrum of all frames in chronological order to form a two-dimensional time-frequency diagram. The horizontal axis of the time-frequency diagram is time, and the vertical axis is frequency. The gray value or color value of each pixel represents the signal power at the corresponding time-frequency point. Figure 2 It demonstrates the typical characteristics of a two-dimensional time-frequency diagram, namely, the presence of harmonic components with multiple frequency periodic variations.

[0041] Based on the two-dimensional time-frequency diagram, the following interpretable physical feature parameters are extracted: (a) Small-scale high frequency f 1: The dominant frequency where energy is most concentrated in the time-frequency graph corresponds to the switching frequency of the LED driver circuit. This is determined by finding the location of the energy peak in the time-frequency graph. f The value of 1. Because f 1. If there is slow change, the average value over a period of time can be taken as the characteristic value, and its range of change should be recorded at the same time.

[0042] (b) Large-scale slowly varying frequency f 2: Main frequency f 1. The periodicity that changes over time. Through analysis of... f The time series of 1 was subjected to spectral analysis again to extract its fluctuation frequency as... f 2. f 2 reflects the oscillation characteristics of the control loop in the drive circuit.

[0043] (c) Frequency of time-domain amplitude fluctuation f 3: This refers to the period of change in waveform amplitude or peak-to-peak value over time. Extract 1 / 3 of each period. f The position of the amplitude or peak value within 1 is observed, and its recurrence after m cycles is used to obtain the amplitude fluctuation frequency. f 3= f 1 / m. Figure 3 It demonstrates how peak offsets repeat in time-domain waveforms.

[0044] (d) Harmonic Power Ratio: Calculate the ratio of the power of each harmonic (2nd, 3rd, 4th, etc.) to the fundamental power. Due to the nonlinear characteristics of the driving circuit, the harmonic distribution of different LEDs varies, and the harmonic power ratio constitutes an important fingerprint feature.

[0045] (e) Harmonic phase difference: Calculate the phase difference between each harmonic. The phase information also reflects the transmission characteristics of the driving circuit, and the phase characteristics of different LEDs are different.

[0046] (f) Number of effective harmonics: The number of harmonics whose power exceeds a certain threshold. Different drive circuits have different degrees of nonlinearity, resulting in different numbers of harmonics.

[0047] (g) Harmonic Time-Varying Trajectory: Records the trajectory characteristics of the frequency and power of each harmonic changing over time. Due to differences in component parameters, the harmonic variation trajectories of different LEDs are unique.

[0048] The seven types of feature parameters mentioned above are combined to form a multi-dimensional feature parameter vector, which serves as the hardware fingerprint of the LED. The dimension of the feature vector can be adjusted according to actual needs.

[0049] In addition to the interpretable physical features mentioned above, deep learning methods can also be used to automatically extract abstract features. Specifically, a two-dimensional time-frequency graph is used as input and fed into a pre-trained convolutional neural network (CNN) to extract feature vectors from the network's intermediate layers. The feature vectors extracted in this method may not have explicit physical meaning, but they can capture more subtle differences and may achieve higher recognition accuracy in complex scenarios.

[0050] 3) Fingerprint database construction and storage The original time-domain waveform, extracted feature vector, unique LED number, and location information of each LED are bound and stored to construct a standardized LED hardware fingerprint database. The fingerprint database can be stored using a relational database or a file system. Each record contains the following fields: LED number (unique identifier), feature vector (floating-point array), original waveform sample (multiple representative waveform data segments), acquisition time (time stamp of fingerprint acquisition), location information (physical location coordinates of the LED), and remarks information (lamp model, installation time, etc.).

[0051] To improve the robustness of the fingerprint database, each LED can store multiple sets of feature vectors, corresponding to features under different operating temperatures, brightness levels, and usage durations. Once the fingerprint database is built, it can be updated periodically to adapt to minor feature changes caused by lamp aging.

[0052] III. Online Light Source Identification Stage The goal of the online phase is to identify the current LED light source number based on the real-time waveform collected by the terminal. The specific steps are as follows: 1) Real-time waveform acquisition When the terminal device approaches the LED light source to be identified, it acquires the real-time light intensity time-domain waveform of the LED using a built-in broadband photoelectric sensor. The acquisition parameters remain consistent with the offline phase: a sampling rate of 2 MS / s and a single acquisition duration of 100 ms (i.e., 200,000 sampling points). The shorter acquisition duration improves the recognition speed and meets real-time requirements.

[0053] The acquired waveform data also underwent preprocessing, including DC removal, baseline correction, and bandpass filtering, to eliminate the effects of ambient light and noise.

[0054] 2) Real-time feature extraction The same time-frequency analysis algorithm used in the offline stage is reused to process real-time waveforms and generate real-time feature vectors. In this embodiment, the same short-time Fourier transform parameters and feature extraction methods as in the offline database construction are used to ensure feature consistency.

[0055] Due to the short online data acquisition time (100ms), for slowly changing features... f 2 and f 3. It may be impossible to observe the complete cycle. In this case, the following strategies can be adopted: (1) Extend the acquisition time: If the terminal stays under the LED for a long time, multiple waveform segments can be collected and spliced ​​together for analysis; (2) Use local features: Even if only part of the cycle is observed, it can be estimated by curve fitting and other methods. f 2 and f Approximate value of 3; (3) Weight adjustment: During feature matching, reduce the weight of slow-changing features and increase the weight of high-frequency features. f 1. Weights of harmonic features.

[0056] 3) Fingerprint matching and recognition The feature vector extracted in real time is compared with the feature vectors of all samples in the fingerprint database. The number corresponding to the sample with the highest matching degree is the number of the current LED light source.

[0057] This embodiment uses cosine similarity as the matching metric for feature vectors. For two feature vectors X and Y, the cosine similarity calculation formula is: cosθ = (X·Y) / (||X|| × ||Y||). The range of cosine similarity is [-1, 1], and the larger the value, the more similar the two feature vectors are.

[0058] The specific matching process is as follows: (1) Calculate the cosine similarity between the real-time feature vector and the feature vector of each LED in the fingerprint database; (2) Find the top K candidate LEDs with the highest similarity (K=3 in this embodiment); (3) If the highest similarity exceeds the preset threshold T (T=0.85 in this embodiment), the matching is successful and the LED number with the highest similarity is output; (4) If the highest similarity is lower than the threshold T, the LED is not in the fingerprint database and the "unknown light source" prompt is output.

[0059] To improve recognition accuracy, a multi-frame voting mechanism can be used: the terminal continuously acquires multiple waveform segments, each segment is recognized independently, and the LED number that appears most frequently is taken as the final recognition result. In this embodiment, 5-frame voting is used, that is, the number that appears most frequently in 5 consecutive recognitions is the final result.

[0060] Example 2:

[0061] This embodiment applies the method of Embodiment 1 to an indoor people positioning scenario, as follows: In a typical office setting, 50 LED downlights are deployed on the ceiling, each driven by an independent switching current source. First, an offline phase is used to establish a hardware fingerprint database for the 50 LEDs and record the physical coordinates of each light.

[0062] People wear smart bracelets integrated with broadband photoelectric sensors, which collect light intensity signals from the ceiling LED lights. As people walk indoors, the bracelets continuously collect the light intensity waveforms of the LED lights above, extract features in real time, match them with a fingerprint database, and identify the LED light number at the current location.

[0063] Since the location of the LED lights is known, the approximate location of a person can be determined by identifying the LED light number. If a person is simultaneously within the illumination range of multiple LED lights, weighted positioning can be performed based on the identification results of multiple lights and signal strength to further improve positioning accuracy. Alternatively, other RSS positioning algorithms can be executed based on signal strength.

[0064] This embodiment of the solution requires no modification to the existing lighting system, no deployment of radio frequency base stations, and poses no risk of privacy leakage. It is particularly suitable for special scenarios such as hospitals, archives, and confidential office areas.

Claims

1. A method for distinguishing LED lighting sources based on broadband hardware fingerprint information, characterized in that, Includes the following steps: S1: Number multiple lighting LEDs and use a broadband photoelectric sensor to collect the time-domain waveforms of LEDs with different numbers; S2: In the offline stage, the time-frequency features of the time-domain waveforms acquired by the broadband photoelectric sensor are extracted and their corresponding LED numbers are labeled to form an LED hardware fingerprint database. S3: In the online phase, the terminal approaches a stock LED light source, collects the time-domain waveform of the LED light source, extracts the time-frequency features, compares the time-frequency features with the LED hardware fingerprint database, and determines the number of the LED light source.

2. The method for distinguishing LED lighting sources based on broadband hardware fingerprint information according to claim 1, characterized in that, The time-domain waveform in step S1 possesses two types of inherent stable fingerprint characteristics: the first type is the short-period high-frequency oscillation frequency of the waveform. f 1, f 1 itself undergoes slow, long-period changes over time; the period of this slow change is denoted as . f 2. Satisfy f 1> f 2; The second category is high frequency. f 1. The waveform is constant, but the amplitude and peak-to-peak position fluctuate periodically. The fluctuation period is denoted as . f 3.

3. The method for distinguishing LED lighting sources based on broadband hardware fingerprint information according to claim 2, characterized in that, Step S2 employs the short-time Fourier transform as a time-frequency analysis method for time-frequency feature extraction, specifically including: A1: Windowed framing: Dividing a one-dimensional time-domain waveform into multiple short time frames; A2: Fourier Transform: Perform a Fast Fourier Transform on each frame of signal to obtain the spectrum of that frame; A3: Constructing a time-frequency diagram: Arrange the spectra of all frames in chronological order to form a two-dimensional time-frequency diagram.

4. The method for distinguishing LED lighting sources based on broadband hardware fingerprint information according to claim 3, characterized in that, In step S2, physical feature parameters are extracted based on the two-dimensional time-frequency diagram, and the physical feature parameters are combined to form a multi-dimensional feature parameter vector, which serves as the hardware fingerprint identifier of the LED hardware fingerprint library.

5. The method for distinguishing LED lighting sources based on broadband hardware fingerprint information according to claim 4, characterized in that, The physical characteristic parameters in step S2 include: Small-scale high frequency f 1: The main frequency where energy is most concentrated in the instantaneous frequency diagram, corresponding to the switching frequency of the LED driver circuit; Large-scale slow-varying frequency f 2: Main frequency f 1. Periodicity that changes over time; Time-domain amplitude fluctuation frequency f 3: That is, the period of change of waveform amplitude or peak-to-peak value over time; Harmonic power ratio: Calculate the ratio of the power of each harmonic to the power of the fundamental frequency; Harmonic phase difference: Calculate the phase difference between each harmonic; Effective harmonic count: The number of harmonics whose power exceeds a certain threshold; Harmonic time-varying trajectory: Records the trajectory characteristics of the frequency and power of each harmonic changing over time.

6. The method for distinguishing LED lighting sources based on broadband hardware fingerprint information according to claim 2, characterized in that, In step S2, the two-dimensional time-frequency graph is input into the machine learning module to automatically generate an abstract feature vector without a clear physical meaning. The abstract feature vector is then used as the hardware fingerprint identifier of the LED hardware fingerprint library.

7. The method for distinguishing LED lighting sources based on broadband hardware fingerprint information according to claim 1, characterized in that, The time-frequency feature extraction method in step S3 is the same as that in step S2, generating a real-time feature vector. The real-time feature vector is then compared with the feature vectors of all samples in the LED hardware fingerprint database. The number corresponding to the sample with the highest matching degree is the current LED light source number.

8. The method for distinguishing LED lighting sources based on broadband hardware fingerprint information according to claim 7, characterized in that, In step S3, cosine similarity is used as the matching metric for feature vectors. For two feature vectors X and Y, the cosine similarity calculation formula is: cosθ = (X·Y) / (||X|| × ||Y||), and the range of cosine similarity is [-1, 1].

9. The method for distinguishing LED lighting sources based on broadband hardware fingerprint information according to claim 1, characterized in that, The LED light source number determined in step S3 is used for indoor personnel positioning.

Images