Low-power-consumption Bluetooth weak link communication method and device based on adaptive symbol extension and DNN demodulation technology

Abstract

The invention discloses a low-power-consumption Bluetooth weak link communication method and device based on adaptive symbol extension and DNN demodulation technology, and the method comprises the steps: measuring a channel environment of a transmitting end, judging a current channel environment through transmitting different extension factors and data packet lengths, and determining an optimal symbol extension factor; sending data of a sending end, and after the optimal symbol expansion factor is determined, expanding a data packet and segmenting the data by a transmitter, and carrying out data de-whitening operation; a receiving end captures a BLE signal, captures the signal and preprocesses the signal; identifying a receiving end and positioning a BLE data packet, detecting a data packet preamble based on phase peak clustering, identifying the data packet and synchronizing symbols; and a receiving end demodulates the BLE signal to obtain final BLE data, and demodulates the wireless signal by adopting a demodulator based on a time-frequency domain combined DNN model. According to the invention, stable BLE communication can be maintained in a low signal-to-noise environment, the method is widely supported in commercial BLE chip equipment, the equipment does not need to be modified, and the communication capability of the BLE equipment in a weak link is improved.

Description

Technical Field

[0001] The present invention proposes a low-power Bluetooth weak link communication and device based on adaptive symbol extension and DNN demodulation technology. The invention proposes a new BLE communication system, which achieves maximum throughput communication under the premise of ensuring reliable communication through adaptive symbol extension and DNN demodulation technology. Background Art

[0002] In recent years, Bluetooth Low Energy (BLE) has become one of the first choices for wireless communication protocols due to its economy, low energy consumption and wide compatibility. According to statistics, by 2022, the annual shipments of BLE-compatible devices will exceed 4 billion. Bluetooth Low Energy technology (BLE) shows extremely high efficiency in connecting devices over short distances and provides a convenient way for devices to access the network. However, BLE is mainly used in indoor environments, where there are often many obstacles and interference from WiFi signals. In addition, in order to reduce energy consumption, BLE tends to use lower signal transmission power. Therefore, in the case of weak signals, BLE's performance is poor, which brings certain obstacles to the application of BLE.

[0003] It would be very promising if BLE could work properly in weak signal conditions. For example, in a health monitoring scenario, a smartwatch worn by an elderly person can report heart rate data to a gateway via BLE. If BLE supports communication over weak links, then even if the user and the gateway are in different rooms, the smartwatch can still transmit emergency information normally. Another example is that users can use a smartphone to remotely control the horn of a car in an underground garage via BLE, helping users find their cars. The numerous obstacles in a garage, such as vehicles and walls, can cause severe attenuation of the BLE signal. Users can remotely control the car horn at a greater distance, making it easier to find the car.

[0004] The existing solution is to use multi-hop Bluetooth networking. However, this method relies on infrastructure, which brings huge deployment and maintenance costs. There are two difficulties in communication under weak links: Bluetooth packet recognition under weak links and high demodulation bit error rate under weak links.

[0005] Regarding Bluetooth packet recognition under weak links, Bluetooth Low Energy (BLE) uses Gaussian Frequency Shift Keying (GFSK) modulation technology, and each Bluetooth packet begins with a 1-byte prefix code (0x10101010). Existing work uses short-time Fourier transform to identify repeated frequency peaks in order to identify prefix codes. However, when the short-time Fourier window is close to the symbol boundary, it contains two consecutive bit sampling points, and the symbol boundary of the short-time Fourier spectrum diagram will be difficult to identify. Therefore, it is difficult for current solutions to accurately identify the position of the prefix code in the weak signal. If the detected prefix code position and the actual prefix code position exceed half a physical layer symbol, the receiver will not be able to demodulate the subsequent physical layer symbols. In general, for protocols with very short symbols such as Bluetooth Low Energy, the impact of prefix code recognition errors based on short-time Fourier technology is huge.

[0006] In addition, in order to address the high demodulation bit error rate problem under weak links, under long-distance links, even if the receiver can accurately detect the position of each symbol, these symbols will still be damaged by high noise. Existing work has proposed using a deep neural network (DNN) model to filter noise on the spectrum graph to improve the LoRa demodulation performance. However, compared to LoRa signals, BLE signals have shorter symbol periods. For example, the symbol period of LoRa is 1 millisecond, while the symbol period of BLE is only 1 microsecond. This poses a serious challenge to BLE demodulation. Moreover, the frequency of BLE is constant within each symbol period, while the frequency of each LoRa symbol changes linearly over the entire bandwidth. Therefore, short-period BLE symbols carry the least information and can be effectively utilized by deep neural networks.

[0007] Therefore, it is hoped that considerable Bluetooth packet recognition can be guaranteed under weak link communication, and the demodulation bit error rate can be maintained low to ensure practical weak link communication. Summary of the invention

[0008] The present invention aims to overcome the above-mentioned defects of the prior art and provide a low-power Bluetooth weak link communication technology method and device based on adaptive symbol extension and DNN demodulation technology. The present invention aims to provide a stable and effective low-power Bluetooth weak link communication technology, which can not only ensure stable communication under weak links, but also maintain considerable communication throughput.

[0009] To achieve the above objectives, the technical solution adopted by the present invention is as follows: a low-power Bluetooth weak link communication method based on adaptive symbol extension and DNN demodulation technology, the solution comprising the following steps:

[0010] (1) Transmitter channel environment measurement. Developers use commercial BLE chip devices to send different extension factors and data packet lengths to determine the current channel environment and determine the optimal symbol extension factor.

[0011] (2) Data transmission at the transmitter: After determining the optimal symbol extension factor, the transmitter expands the data packet, segments the data, and performs data de-whitening operations.

[0012] (3) The receiving end captures the BLE signal, captures the signal and performs preprocessing operations on the signal, including signal capture and data preprocessing;

[0013] (4) The receiving end identifies and locates the BLE data packet. The system uses data packet preamble detection technology based on phase peak clustering to achieve accurate data packet identification and symbol synchronization.

[0014] (5) The receiving end demodulates the BLE signal to obtain the final BLE data. The system uses a demodulator based on the time-frequency domain combined with DNN to demodulate the wireless signal, which includes three steps: symbol conversion, time-frequency domain feature extraction, and DNN demodulation.

[0015] Preferably, step (1) specifically includes:

[0016] 1.1) Evaluate the preamble detection rate: The transmitter first quickly transmits a probe packet to evaluate the preamble detection rate under different extension factors. These probe packets only contain mandatory fields but no payload, i.e., n max =0;

[0017] 1.2) Evaluating the bit error rate BER; the transmitter transmits a data packet with a known byte and a longest extension string for bit error rate BER evaluation;

[0018] 1.3) Determine the sign extension factor f ext The system determines the best symbol extension factor based on the signal-to-noise ratio (SNR)-bit error rate (BER) relationship of different symbol extension factors obtained in advance.

[0019] Preferably, step (2) specifically comprises:

[0020] Sign-extend factor f according to the specified ext Complete sign extension (i.e., each bit repeats the sign extension factor f ext times), the extended data is then segmented so that it can be encapsulated into AIPs and transmitted over the low-power Bluetooth physical layer; before sending, a data de-whitening operation is applied to offset the interference effect caused by the data whitening automatically performed by commercial BLE chips; this allows the commercial BLE chip to be controlled to transmit the extended symbols while ensuring compatibility with current hardware configurations.

[0021] Preferably, step (3) specifically comprises:

[0022] 3.1) Signal capture: Use software radio peripherals to continuously scan and capture wireless signals in the 2.4 GHz channel at a sampling rate of 8 MHz;

[0023] 3.2) Data preprocessing: normalize the original signal and filter it using a digital Butterworth low-pass filter to remove high-frequency noise in the baseband signal while retaining the original signal and the filtered signal.

[0024] Preferably, step (4) specifically includes:

[0025] 4.1) Using the phase difference between consecutive sampling points, the phase difference sequence is derived from the original signal, denoted as S pd ;

[0026] 4.2) For the phase difference sequence, the amplitude-weighted phase difference is calculated and recorded as Considering the short time interval between consecutive sampling points and the small phase difference involved, This is approximately expressed as follows:

[0027]

[0028]

[0029] in, is the sampling point interval; since the in-phase and quadrature signals have the same frequency, it is assumed that their attenuation is the same; therefore, the received in-phase and quadrature signals are expressed as:

[0030]

[0031] Among them, A is the attenuation weight of the signal, and the phase difference calculated at the receiving end is

[0032] 4.3) Identify the extended preamble. To detect the 8-bit preamble, a preamble kernel K is used and convolved with the phase difference sequence. The length of the kernel is 8×f ext ×n s , where n s represents the number of sampling points in each native low-power Bluetooth symbol; each item in the preamble kernel K corresponds to a sampling point in the extended preamble, and its value is the NRZ code of the data bit; specifically, the first f in the preamble kernel K ex t×n s The item is set to -1, and the next f ext ×n s The item is set to 1, and so on;

[0033] The result of the convolution of the preamble kernel and the phase difference sequence is a sequence representing the accumulated phase difference, denoted as S c , which contains multiple phase peaks; among all the phase peaks, some are caused by the preamble code and the rest are caused by channel noise; the repeated phase changes produce 7 phase peaks, with an extremely prominent peak in the middle and three smaller peaks on the left and right; from the center to the sides, the peaks indicate that the preamble code kernel is aligned with 8, 6, 4, and 2 in the phase difference sequence respectively; the number of sampling points between two adjacent peaks does not exceed 12×f ext ×n s In contrast, the peaks generated by noise and other data bits are scattered in time; in order to filter out these irrelevant peaks, c The peaks in the cluster are clustered; the peak clusters related to the preamble phase difference sequence have two key characteristics: the width is less than or equal to 12×f ext ×n s and consist of a maximum of 7 peaks; only the peak clusters that meet these conditions are identified as being caused by the Bluetooth Low Energy preamble, and the rest are discarded;

[0034] The central peak within the symbol pair cluster indicates that the preamble kernel is completely aligned with the preamble phase difference sequence, and the corresponding sampling point is identified as the starting sampling point of the preamble.

[0035] Preferably, the DNN model described in step (5) is a multi-input binary classification model. First, the input of the DNN model includes signal data in the time domain and the frequency domain. Then, these two types of data are respectively input into the time domain feature extractor and the frequency domain feature extractor. Both the time domain feature extractor and the frequency domain feature extractor adopt a six-layer CNN model, and each CNN layer includes a padding operation, a two-dimensional convolution, a batch normalization and a ReLU activation function. Then, the outputs of the time domain feature extractor and the frequency domain feature extractor are input into the demodulator in a channel merging manner. The demodulator consists of a layer of LSTM and ReLU activation function, three layers of fully connected layers including batch normalization and ReLU activation function, and a layer of fully connected layer with a Softmax activation function. Finally, the demodulator outputs the classification result.

[0036] Further, step (5) specifically includes:

[0037] 5.1) Symbol Conversion: In order to decode the data bits from the received Bluetooth Low Energy symbols, the dual-channel spectrogram and the time domain signal are taken as input and the detected data packet is cut into each symbol to be demodulated; the information required for demodulation exists in both the time domain and the frequency domain;

[0038] For the input of the time domain signal, the real and imaginary parts of the original signal and the filtered signal are connected and reshaped into a four-channel two-dimensional array as the input of the subsequent time domain DNN feature extraction model; then, the short-time Fourier transform spectrum of the filtered signal is performed, and the real and imaginary parts are connected, and finally a two-channel two-dimensional spectrum is obtained as the input of the subsequent frequency domain DNN feature extraction model;

[0039] 5.2) Time-frequency domain feature extraction: Two six-layer DNNs are used to extract feature information from the time and frequency domain inputs. Both feature extractors use two-dimensional convolutional layers to capture the time and frequency domain feature information. Finally, the outputs from the two feature extractors are concatenated and input into the demodulator.

[0040] 5.3) DNN demodulation: A DNN-based binary classifier is used to demodulate the data bits, and LSTM is used to further extract time domain features. A four-layer fully connected layer is used to demodulate the data. After the extracted feature information is input into the model, the final bit output is obtained to complete the data demodulation.

[0041] The second aspect of the present invention relates to a low-power Bluetooth weak-link communication device based on adaptive sign extension and DNN demodulation technology, comprising a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, they are used to implement the low-power Bluetooth weak-link communication method based on adaptive sign extension and DNN demodulation technology of the present invention.

[0042] The third aspect of the present invention relates to a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the low-power Bluetooth weak link communication method based on adaptive sign extension and DNN demodulation technology of the present invention.

[0043] The present invention sends different expansion factors and data packet lengths in commercial BLE chip devices to judge the current channel environment and determine the optimal symbol expansion factor. After determining the optimal symbol expansion factor, the transmitter expands the data packet, segments the data, and performs data de-whitening operations. At the receiving end, the system uses a data packet preamble detection technology based on phase peak clustering to achieve accurate data packet identification and symbol synchronization. The system uses a demodulator based on the time-frequency domain combined with DNN to demodulate the wireless signal, including three steps: symbol conversion, time-frequency domain feature extraction, and DNN demodulation. The present invention can maintain stable BLE communication in a low signal-to-noise environment, and is widely supported in commercial BLE chip devices. There is no need to modify the device and it will not cause interference to users. In addition, the model system proposed by the present invention can guarantee a considerable communication throughput, which greatly improves the communication capability of BLE devices under weak links.

[0044] Compared with the prior art, the beneficial effects of the present invention are: the proposed weak link communication technology can maintain stable BLE communication in a low signal-noise environment, and is widely supported in commercial BLE chip devices, does not require modification of the device, and does not cause interference to users. In addition, the model system proposed by the present invention can guarantee a considerable communication throughput, greatly improving the communication capability of BLE devices under weak links. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is an overall flow chart of weak link communication of BLE device of the method of the present invention.

[0046] Figure 2 It is a time-frequency domain combined DNN demodulator diagram of the present invention.

[0047] Figure 3 It is a schematic diagram of the device of the present invention. DETAILED DESCRIPTION

[0048] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0049] Example 1

[0050] This embodiment provides a low-power Bluetooth weak link communication method based on adaptive symbol extension and DNN demodulation technology. This method can achieve maximum throughput communication under the premise of ensuring reliable communication in a weak link communication environment with low signal-to-noise ratio through adaptive symbol extension and DNN demodulation technology. The overall process framework is as follows Figure 1 As shown, the specific implementation method is as follows:

[0051] (1) Transmitter channel environment measurement. Developers use commercial BLE chip devices to send different extension factors and data packet lengths to determine the current channel environment and determine the optimal symbol extension factor. The specific process includes the following steps:

[0052] 1.1) Evaluate the preamble detection rate. The transmitter first quickly transmits probe packets to evaluate the preamble detection rate under different extension factors. These probe packets only contain mandatory fields but no payload (i.e., n max =0).

[0053] 1.2) Evaluate the bit error rate BER. The transmitter transmits a data packet with known bytes and the longest extension string for bit error rate BER evaluation.

[0054] 1.3) Determine the sign extension factor f ext The system determines the best symbol extension factor based on the signal-to-noise ratio (SNR)-bit error rate (BER) relationship of different symbol extension factors obtained in advance.

[0055] (2) Data transmission at the transmitter. After determining the optimal symbol extension factor, the transmitter expands the data packet, segments the data, and performs data de-whitening operations. The specific implementation method is as follows:

[0056] Sign-extend factor f according to the specified ext Complete sign extension (i.e., each bit repeats the sign extension factor f ext times), the extended data is then segmented so that it can be encapsulated into AIPs and transmitted over the Bluetooth Low Energy physical layer. Before transmission, a data de-whitening operation is applied to counteract the interference effects of data whitening automatically performed by commercial BLE chips. This allows commercial BLE chips to be controlled to transmit extended symbols while ensuring compatibility with existing hardware configurations.

[0057] (3) The receiving end captures the BLE signal. The system captures the signal and performs preprocessing operations on the signal. The specific process includes the following two stages:

[0058] 3.1) Signal Capture: The system uses software radio peripherals such as USRP B210 devices to continuously scan and capture wireless signals in the 2.4 GHz channel at a sampling rate of 8 MHz.

[0059] 3.2) Data preprocessing. The system normalizes the original signal and uses a digital Butterworth low-pass filter to filter the original signal to remove high-frequency noise in the baseband signal. It should be noted that the digital filter (Butterworth filter in this article) may cause some distortion to the original signal, such as transient response and phase shift at the beginning, so the system needs to retain both the original signal and the filtered signal.

[0060] (4) The receiving end identifies and locates the BLE data packet. The system uses data packet preamble detection technology based on phase peak clustering to achieve accurate data packet identification and symbol synchronization. The specific implementation method is as follows:

[0061] 4.1) The system uses the phase difference between consecutive sampling points to deduce the phase difference sequence from the original signal, denoted as S pd .

[0062] 4.2) The system calculates the amplitude-weighted phase difference for the phase difference sequence, which is recorded as Considering the short time interval between consecutive sampling points and the small phase difference involved, This approximation can be expressed as follows:

[0063]

[0064]

[0065] in, is the sampling point interval. Since the in-phase and quadrature signals have the same frequency, it can be assumed that their attenuation is the same. Therefore, the received in-phase and quadrature signals can be expressed as:

[0066]

[0067] Where A is the attenuation weight of the signal. The phase difference calculated at the receiving end is It is worth noting that the calculations proposed by the above two formulas only involve multiplication and subtraction operations, so the computational overhead is minimal.

[0068] 4.3) The system identifies the extended preamble. In order to detect the 8-bit preamble, the system uses a preamble kernel K and convolves it with the phase difference sequence. The length of the kernel is 8×f ext ×n s , where n s represents the number of sampling points in each native low-power Bluetooth symbol. Each item in the preamble kernel K corresponds to a sampling point in the extended preamble, and its value is the NRZ code of the data bit. Specifically, the first f in the preamble kernel K ext ×n s The item is set to -1, and the next f ext ×n s item is set to 1, and so on.

[0069] The result of the convolution of the preamble kernel and the phase difference sequence is a sequence representing the accumulated phase difference, denoted as S c , which contains multiple phase peaks. Among all the phase peaks, some are caused by the preamble, but others are caused by channel noise. The repeated phase changes will produce 7 phase peaks, with an extremely prominent peak in the middle and three smaller peaks on the left and right. From the center to the sides, the peaks indicate that the preamble kernel is aligned with 8, 6, 4, and 2 of the phase difference sequence respectively. The number of sampling points between two adjacent peaks does not exceed 12×f ext ×n s In contrast, the peaks generated by noise and other data bits are scattered in time. To filter out these irrelevant peaks, the system c The peak clusters associated with the preamble phase difference sequence have two key characteristics: their width is less than or equal to 12×f ext ×n s and they consist of a maximum of 7 peaks. Only the peak clusters that meet these conditions are identified as being caused by the Bluetooth LE preamble, and the rest are discarded.

[0070] The central peak within the symbol pair cluster indicates that the preamble kernel is perfectly aligned with the preamble phase difference sequence, and the corresponding sampling point can be identified as the start sampling point of the preamble. It is also possible to determine the acceptance windows of subsequent symbols because their lengths are known.

[0071] (5) The receiving end demodulates the BLE signal to obtain the final BLE data. The system uses a demodulator based on the time-frequency domain combined with DNN to demodulate the wireless signal, which includes three steps: symbol conversion, time-frequency domain feature extraction, and DNN demodulation. The DNN model is as follows Figure 2 As shown in the figure, the DNN model is a multi-input binary classification model. First, the input of the DNN model includes signal data in the time domain and frequency domain. Then, these two types of data are input into the time domain feature extractor and the frequency domain feature extractor respectively. Both feature extractors adopt a six-layer CNN model, and each CNN layer contains a padding operation, a two-dimensional convolution, a batch normalization, and a ReLU activation function. Then, the outputs of the two feature extractors are input into the demodulator in a channel-merging manner. The demodulator consists of a layer of LSTM and ReLU activation function, three layers of fully connected layers containing batch normalization and ReLU activation functions, and a layer of fully connected layers with a Softmax activation function. Finally, the demodulator outputs the classification result. The specific implementation method is as follows:

[0072] 5.1) Symbol Conversion. To decode the data bits from the received Bluetooth Low Energy symbols, the system takes the dual-channel spectrogram and the time domain signal as input and cuts the detected data packet into each symbol to be demodulated. The information required for demodulation exists in both the time domain (i.e., phase difference) and the frequency domain (i.e., frequency offset).

[0073] For the input of time domain signals, the system connects the real and imaginary parts of the original signal and the filtered signal, reshaping it into a four-channel two-dimensional array as the input of the subsequent time domain DNN feature extraction model. Subsequently, the filtered signal is subjected to a short-time Fourier transform spectrum, and the real and imaginary parts are connected, finally obtaining a two-channel two-dimensional spectrum as the input of the subsequent frequency domain DNN feature extraction model.

[0074] 5.2) Time-frequency domain feature extraction. Two six-layer DNNs are used to extract feature information from the time and frequency domain inputs. Both feature extractors use two-dimensional convolutional layers to capture the time and frequency domain feature information. Finally, the outputs from the two feature extractors are concatenated and input into the demodulator.

[0075] 5.3) DNN demodulation. The demodulator adopted by the system is a binary classifier based on DNN, which is used to demodulate data bits. It uses LSTM to further extract time domain features and uses four fully connected layers to demodulate data. After the extracted feature information is input into the model, the final bit output is obtained to complete data demodulation.

[0076] Example 2

[0077] like Figure 3 This embodiment relates to a low-power Bluetooth weak-link communication device based on adaptive sign extension and DNN demodulation technology, including a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, they are used to implement the low-power Bluetooth weak-link communication method based on adaptive sign extension and DNN demodulation technology of Example 1.

[0078] Example 3

[0079] The present embodiment relates to a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, the low-power Bluetooth weak link communication method based on adaptive sign extension and DNN demodulation technology of the present invention is implemented.

[0080] The contents described in the embodiments of this specification are merely an enumeration of the implementation forms of the inventive concept. The protection scope of the present invention should not be regarded as limited to the specific forms described in the embodiments. The protection scope of the present invention also extends to the equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A low-power Bluetooth weak link communication method based on adaptive symbol extension and DNN demodulation technology, comprising the following steps: (1) Measure the channel environment at the transmitting end, determine the current channel environment by sending different extension factors and data packet lengths, and determine the optimal symbol extension factor; (2) Sending the sender data, after determining the optimal symbol extension factor, the transmitter expands the data packet, segments the data, and performs data de-whitening operations; (3) The receiving end captures the BLE signal, captures the signal and performs preprocessing operations on the signal; (4) Identify the receiving end and locate the BLE data packet, detect the data packet preamble based on phase peak clustering, identify the data packet and synchronize the symbols; (5) The receiving end demodulates the BLE signal to obtain the final BLE data, and uses a demodulator based on the time-frequency domain combined with the DNN model to demodulate the wireless signal.

2. The low-power Bluetooth weak link communication method based on adaptive sign extension and DNN demodulation technology as claimed in claim 1, characterized in that: Step (1) specifically includes: 1.1) Evaluate the preamble detection rate: The transmitter first quickly transmits a probe packet to evaluate the preamble detection rate under different extension factors. These probe packets only contain mandatory fields but no payload, i.e., n max =0; 1.2) Evaluating the bit error rate BER; the transmitter transmits a data packet with a known byte and a longest extension string for bit error rate BER evaluation; 1.3) Determine the sign extension factor f ext The system determines the best symbol extension factor based on the signal-to-noise ratio (SNR)-bit error rate (BER) relationship of different symbol extension factors obtained in advance.

3. The low-power Bluetooth weak link communication method based on adaptive sign extension and DNN demodulation technology as claimed in claim 1, characterized in that: Step (2) specifically includes: Sign-extend factor f according to the specified ext Complete sign extension (i.e., each bit repeats the sign extension factor f ext times), the extended data is then segmented so that it can be encapsulated into AIPs and transmitted over the low-power Bluetooth physical layer; before sending, a data de-whitening operation is applied to offset the interference effect caused by the data whitening automatically performed by commercial BLE chips; this allows the commercial BLE chip to be controlled to transmit the extended symbols while ensuring compatibility with current hardware configurations.

4. The low-power Bluetooth weak link communication method based on adaptive sign extension and DNN demodulation technology as claimed in claim 1, characterized in that: Step (3) specifically includes: 3.1) Signal capture: Use software radio peripherals to continuously scan and capture wireless signals in the 2.4 GHz channel at a sampling rate of 8 MHz; 3.2) Data preprocessing: normalize the original signal and filter it using a digital Butterworth low-pass filter to remove high-frequency noise in the baseband signal while retaining the original signal and the filtered signal.

5. The low-power Bluetooth weak link communication method based on adaptive sign extension and DNN demodulation technology as claimed in claim 1, characterized in that: Step (4) specifically includes: 4.1) Using the phase difference between consecutive sampling points, the phase difference sequence is derived from the original signal, denoted as S pd ; 4.2) For the phase difference sequence, the amplitude-weighted phase difference is calculated and recorded as Considering the short time interval between consecutive sampling points and the small phase difference involved, This is approximately expressed as follows: in, is the sampling point interval; since the in-phase and quadrature signals have the same frequency, it is assumed that their attenuation is the same; therefore, the received in-phase and quadrature signals are expressed as: Among them, A is the attenuation weight of the signal, and the phase difference calculated at the receiving end is 4.3) Identify the extended preamble. To detect the 8-bit preamble, a preamble kernel K is used and convolved with the phase difference sequence. The length of the kernel is 8×f ext ×n s , where n s represents the number of sampling points in each native low-power Bluetooth symbol; each item in the preamble kernel K corresponds to a sampling point in the extended preamble, and its value is the NRZ code of the data bit; specifically, the first f in the preamble kernel K ext ×n s The item is set to -1, and the next f ext ×n s The item is set to 1, and so on; The result of the convolution of the preamble kernel and the phase difference sequence is a sequence representing the accumulated phase difference, denoted as S c , which contains multiple phase peaks; among all the phase peaks, some are caused by the preamble code and the rest are caused by channel noise; the repeated phase changes produce 7 phase peaks, with an extremely prominent peak in the middle and three smaller peaks on the left and right; from the center to the sides, the peaks indicate that the preamble code kernel is aligned with 8, 6, 4, and 2 in the phase difference sequence respectively; the number of sampling points between two adjacent peaks does not exceed 12×f ext ×n s ; In contrast, the peaks generated by noise and other data bits are scattered in time; in order to filter out these irrelevant peaks, c The peaks in the cluster are clustered; the peak clusters related to the preamble phase difference sequence have two key characteristics: the width is less than or equal to 12×f ext ×n s and consist of a maximum of 7 peaks; only the peak clusters that meet these conditions are identified as being caused by the Bluetooth Low Energy preamble, and the rest are discarded; The central peak within the symbol pair cluster indicates that the preamble kernel is completely aligned with the preamble phase difference sequence, and the corresponding sampling point is identified as the starting sampling point of the preamble.

6. The low-power Bluetooth weak link communication method based on adaptive sign extension and DNN demodulation technology as claimed in claim 1, characterized in that: The DNN model described in step (5) is a multi-input binary classification model. First, the input of the DNN model includes signal data in the time domain and the frequency domain. Then, these two types of data are input into the time domain feature extractor and the frequency domain feature extractor, respectively. Both the time domain feature extractor and the frequency domain feature extractor adopt a six-layer CNN model, and each CNN layer includes a padding operation, a two-dimensional convolution, a batch normalization, and a ReLU activation function. Then, the outputs of the time domain feature extractor and the frequency domain feature extractor are input into the demodulator in a channel-merging manner. The demodulator consists of a layer of LSTM and ReLU activation function, three layers of fully connected layers including batch normalization and ReLU activation function, and a layer of fully connected layer with a Softmax activation function. Finally, the demodulator outputs the classification result.

7. The low-power Bluetooth weak link communication method based on adaptive sign extension and DNN demodulation technology as claimed in claim 6, characterized in that: Step (5) specifically includes: 5.1) Symbol Conversion: In order to decode the data bits from the received Bluetooth Low Energy symbols, the dual-channel spectrogram and the time domain signal are taken as input and the detected data packet is cut into each symbol to be demodulated; the information required for demodulation exists in both the time domain and the frequency domain; For the input of the time domain signal, the real and imaginary parts of the original signal and the filtered signal are connected and reshaped into a four-channel two-dimensional array as the input of the subsequent time domain DNN feature extraction model; then, the short-time Fourier transform spectrum of the filtered signal is performed, and the real and imaginary parts are connected, and finally a two-channel two-dimensional spectrum is obtained as the input of the subsequent frequency domain DNN feature extraction model; 5.2) Time-frequency domain feature extraction: Two six-layer DNNs are used to extract feature information from the time and frequency domain inputs. Both feature extractors use two-dimensional convolutional layers to capture the time and frequency domain feature information. Finally, the outputs from the two feature extractors are concatenated and input into the demodulator. 5.3) DNN demodulation: A DNN-based binary classifier is used to demodulate the data bits, and LSTM is used to further extract time domain features. A four-layer fully connected layer is used to demodulate the data. After the extracted feature information is input into the model, the final bit output is obtained to complete the data demodulation.

8. A low-power Bluetooth weak link communication device based on adaptive symbol extension and DNN demodulation technology, characterized in that: It includes a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, it is used to implement the low-power Bluetooth weak link communication method based on adaptive sign extension and DNN demodulation technology as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that: A program is stored thereon, and when the program is executed by a processor, the low-power Bluetooth weak link communication method based on adaptive sign extension and DNN demodulation technology described in any one of claims 1-7 is implemented.

Images