Low-power-consumption Bluetooth chip signal receiving optimization method

Abstract

The invention relates to the technical field of low-power-consumption Bluetooth, and discloses a low-power-consumption Bluetooth chip signal receiving optimization method. The method comprises the following steps: acquiring signal strength, noise power and multipath interference data in an environment through a multichannel radio frequency front end, and generating an original environment feature set; performing space-time correlation analysis on the original environment feature set, and constructing a dynamic environment feature matrix; inputting the dynamic environment characteristic matrix into a filtering strategy selection model based on an attention mechanism, and outputting an optimal filtering strategy identifier; and calling a corresponding adaptive filtering algorithm according to the optimal filtering strategy identifier, and performing time-frequency joint processing on the received signal to generate a preliminary optimized signal. According to the method, adaptive adjustment of a filtering strategy is realized through multi-dimensional environment feature collection and space-time correlation analysis, the signal receiving performance is improved, hardware resource occupation is reduced, chip power consumption is reduced, the method is suitable for complex scenes such as the Internet of Things and wearable equipment, and the low-power-consumption Bluetooth technology is assisted to be applied to more fields.

Description

Technical Field

[0001] This invention relates to the field of low-power Bluetooth technology, specifically to a method for optimizing signal reception in a low-power Bluetooth chip. Background Technology

[0002] In fields such as the Internet of Things (IoT), wearable devices, and smart homes, Bluetooth Low Energy (BLE) chips, with their low power consumption and miniaturization, have become core components for wireless communication in various terminal devices. As the application scenarios of these devices continue to expand, their working environments are becoming increasingly complex, from enclosed indoor spaces to open outdoor areas, and from single-device deployments to densely interconnected multi-device scenarios. The interference factors encountered during signal transmission are constantly increasing, placing higher demands on the signal reception performance of chips. Currently, Bluetooth Low Energy (BLE) chips generally employ fixed filtering strategies to process received signals, meaning they use a pre-defined filtering algorithm for different environments. However, in real-world applications, environmental characteristics exhibit significant dynamic changes. For example, in indoor office scenarios, the movement of people causes real-time changes in signal propagation paths, resulting in fluctuating intensity and frequency of multipath interference. In smart home scenarios, the operating frequency bands of various wireless devices (such as Wi-Fi routers and ZigBee sensors) partially overlap with the BLE band, causing noise power to fluctuate dramatically depending on the device's operating status. Fixed filtering strategies cannot adapt to dynamic changes in environmental characteristics. When multipath interference in the environment increases, using filtering algorithms designed for low-interference scenarios will result in ineffective removal of interference components from the received signal, thus reducing the signal-to-noise ratio. Conversely, when noise power decreases, continuing to use strong filtering algorithms designed for high-noise scenarios will excessively suppress useful signal components, causing loss of signal details and affecting the accuracy of data transmission. Furthermore, Bluetooth Low Energy chips have strict requirements for power consumption control. Traditional solutions that run multiple filtering algorithms in parallel to cope with different environments will continuously consume a large amount of hardware resources, leading to a significant increase in chip power consumption and failing to meet the battery life requirements of wearable devices and other mobile terminals. Current technologies for collecting environmental characteristics often limit data to a single dimension or a fixed time point, lacking spatiotemporal correlation analysis of multi-dimensional data such as signal strength, noise power, and multipath interference. For example, collecting only the signal strength value at a certain moment cannot reflect the trend of that value over time or its correlation with the location of other interference sources in space, leading to a one-sided judgment of the environmental state and consequently affecting the rationality of filtering strategy selection. These problems collectively result in poor signal reception stability and high data transmission error rates for current Bluetooth Low Energy (BLE) chips in complex dynamic environments, and the difficulty in balancing signal reception performance and power consumption control, thus restricting the application of BLE technology in more demanding scenarios. Summary of the Invention

[0003] The purpose of this invention is to provide a method for optimizing signal reception in a low-power Bluetooth chip, so as to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides a method for optimizing signal reception in a low-power Bluetooth chip, the method comprising: Step 1: Collect signal strength, noise power, and multipath interference data in the environment through a multi-channel RF front-end to generate an original environmental feature set; Step 2: Perform spatiotemporal correlation analysis on the original environmental feature set to construct a dynamic environmental feature matrix; Step 3: Input the dynamic environment feature matrix into the attention-based filtering strategy selection model and output the optimal filtering strategy identifier; Step 4: Based on the optimal filtering strategy identifier, call the corresponding adaptive filtering algorithm to perform time-frequency joint processing on the received signal and generate a preliminary optimized signal.

[0005] Preferably, step 2 includes: Step 2.1: Perform sliding window segmentation on the original environmental feature set to form multiple time segment feature sets; Step 2.2: Calculate the correlation coefficient between feature sets of adjacent time segments and construct a spatiotemporal correlation matrix; Step 2.3: Dimensionality reduction of the spatiotemporal correlation matrix is ​​performed using principal component analysis to generate the dynamic environment feature matrix.

[0006] Preferably, step 3 includes: Step 3.1: Input the dynamic environment feature matrix into the attention-based filtering strategy selection model; Step 3.2: Calculate the dynamic weight distribution of each environmental feature using a multi-head attention mechanism; Step 3.3: Select the filter strategy combination with the highest matching degree from the preset strategy library according to the dynamic weight distribution, and generate the optimal filter strategy identifier.

[0007] Preferably, step 4 includes: Step 4.1: Parse the time-domain processing parameters and frequency-domain processing parameters in the optimal filtering strategy identifier; Step 4.2: Perform dynamic time-domain filtering on the received signal using a variable-length sliding window; Step 4.3: Perform adaptive frequency domain transformation on the time-domain processed signal and implement dynamic frequency domain filtering; Step 4.4: Reconstruct the signal from the time-frequency processing results to generate the preliminary optimized signal.

[0008] Preferably, the method further includes: Step 5: Input the preliminary optimized signal into a signal enhancement system based on a deep residual network, and output the enhanced target signal; Step 6: Perform a multi-dimensional quality assessment on the enhanced target signal to generate a signal quality assessment matrix.

[0009] Preferably, step 5 includes: Step 5.1: Perform segmented windowing processing on the preliminary optimized signal to generate a signal segment sequence; Step 5.2: Extract multi-scale signal features through the convolutional layers of the deep residual network; Step 5.3: Utilize the cross-layer connection structure to fuse feature representations of different depths, and generate the enhanced target signal through the feature reconstruction network.

[0010] Preferably, step 6 includes: Step 6.1: Calculate the bit error rate, signal-to-noise ratio, and phase consistency index of the enhanced target signal; Step 6.2: Construct a three-dimensional evaluation matrix that includes time and indicator dimensions; Step 6.3: Generate the signal quality evaluation matrix using a moving weighted average algorithm.

[0011] Preferably, the method further includes: Step 7: Dynamically adjust the receiving parameter configuration according to the signal quality evaluation matrix to generate a new combination of receiving parameters.

[0012] Preferably, step 7 includes: Step 7.1: Analyze the distribution of abnormal data points in the signal quality assessment matrix; Step 7.2: Calculate the deviation between the current parameter configuration and the ideal configuration, and generate parameter adjustment instructions that include the adjustment direction and magnitude.

[0013] Preferably, the method further includes: Step 8: Establish a dynamic mapping relationship between environmental characteristics and receiving parameters, predict the optimal parameter configuration based on environmental change trends, and achieve parameter pre-adjustment.

[0014] Compared with the prior art, the beneficial effects of the present invention are: In step 1, signal strength, noise power, and multipath interference data are collected through a multi-channel RF front-end. Compared with the traditional single-dimensional environmental data collection method, this method can comprehensively capture the key factors affecting signal reception in the environment, avoid misjudgment of environmental conditions due to incomplete data collection, and provide more complete basic data support for the reasonable selection of subsequent filtering strategies. Step 2 involves performing spatiotemporal correlation analysis on the original environmental feature set and constructing a dynamic environmental feature matrix, overcoming the limitations of existing technologies that rely solely on fixed-time-point data or single-dimensional spatial data. Spatiotemporal correlation analysis can uncover the changing patterns of various environmental parameters at different time points, as well as the correlation between the distribution of interference sources and signal propagation characteristics in space. For example, it can identify the periodic trends of multipath interference within a specific time period, or the correspondence between noise power and the location of interference sources within a certain area, thus more accurately reflecting the dynamic characteristics of the environment. The construction of the dynamic environmental feature matrix presents multidimensional, spatiotemporally correlated environmental data in a structured form, making the quantitative description of the environmental state clearer and facilitating rapid identification and analysis of the environmental state by subsequent models. The attention-based filtering strategy selection model in step 3 can automatically focus on key environmental factors according to the influence weights of different environmental parameters on signal reception in the dynamic environmental feature matrix, thereby achieving accurate output of the optimal filtering strategy. The attention mechanism allows the model to assign higher weights to the main factors affecting signal reception (such as focusing on noise power parameters when noise power significantly increases at a certain moment) when analyzing environmental features. This avoids the limitation of traditional fixed filtering strategies that treat all environmental factors equally, thus selecting the filtering strategy that best matches the real-time environmental state and ensuring that the filtering process can specifically address the main interference problems in the current environment. In step 4, the corresponding adaptive filtering algorithm is invoked based on the optimal filtering strategy identifier for time-frequency joint processing. Compared to the traditional approach of running multiple filtering algorithms in parallel, this method only calls specific filtering algorithms when needed, significantly reducing hardware resource consumption. Time-frequency joint processing can simultaneously optimize the received signal in both the time and frequency domains. It can filter out burst interference in the time domain and suppress overlapping frequency band noise in the frequency domain, avoiding the problem of incomplete interference filtering caused by single-domain processing. The generated preliminary optimized signal retains useful signal components while minimizing the impact of interference and noise. From a power consumption control perspective, since it eliminates the need to continuously run multiple filtering algorithms and only dynamically calls the required algorithms based on environmental changes, it effectively reduces the chip's computing power consumption and lowers overall power consumption. This meets the battery life requirements of Bluetooth Low Energy (BLE) chips and is particularly suitable for power-sensitive terminal products such as wearable devices and portable sensors. Furthermore, this method does not require adding a large number of hardware modules; it improves signal reception performance solely through algorithm-level optimization, resulting in lower hardware costs. It is easy to integrate and promote on existing BLE chip architectures, enabling broad compatibility with different types of BLE terminal devices and expanding the application scenarios of BLE technology. Attached Figure Description

[0015] Figure 1This is a schematic diagram illustrating the working principle of the low-power Bluetooth chip signal reception optimization method described in this invention. Figure 2 A flowchart for selecting an attention-based filtering strategy; Figure 3 A flowchart for generating a preliminary optimized signal through time-frequency joint processing. Detailed Implementation

[0016] 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 embodiments of the present invention, and not all embodiments. Based on the 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.

[0017] Please see Figure 1 This invention provides a method for optimizing signal reception in a low-power Bluetooth chip. The method includes: acquiring signal strength, noise power, and multipath interference data from the environment via a multi-channel RF front-end to generate an original environmental feature set, which contains multi-dimensional real-time information about the signal reception environment; performing spatiotemporal correlation analysis on the original environmental feature set to construct a dynamic environmental feature matrix, which captures the changing patterns of environmental features in time and space; inputting the dynamic environmental feature matrix into an attention-based filtering strategy selection model, which evaluates the importance of different environmental features and outputs an optimal filtering strategy identifier; and calling the corresponding adaptive filtering algorithm based on the optimal filtering strategy identifier to perform time-frequency joint processing on the received signal, with time-domain processing suppressing burst noise and frequency-domain processing eliminating narrowband interference, ultimately generating a preliminary optimized signal.

[0018] Example 1: See Figure 2The generation of the original environmental feature set relies on the continuous operation of a multi-channel RF front-end. This front-end captures RF signals in the environment at a fixed sampling period and extracts signal strength, noise power, and multipath interference data in real time. These data are arranged according to timestamps to form the initial dataset. The density of data points depends on the sampling rate and the drastic nature of environmental changes. The sliding window segmentation process is implemented on the original environmental feature set. The length of the window is not fixed but dynamically adjusted based on the real-time evaluation of the signal change rate. When the signal statistical characteristics are stable, a longer window is used to obtain a more comprehensive temporal context. When rapid changes are detected, the window size is shortened to capture transient features. A certain overlap area can be set between windows to ensure temporal continuity and avoid the loss of effective information at the window boundaries. The calculation of correlation coefficients involves multiple dimensions. For signal strength sequences, the linear correlation is mainly examined. The Pearson correlation coefficient measures the similarity of amplitude changes between adjacent time segments. A value close to positive one indicates a strong positive correlation, close to negative one indicates a strong negative correlation, and close to zero indicates a weak correlation. For noise power sequences, since their statistical distribution may be non-Gaussian, mutual information entropy is used to quantify the generalized dependence between sequences. The larger the mutual information value, the stronger the predictive ability of the noise pattern of one segment on another segment. Multipath interference data is usually represented in matrix form. The off-diagonal elements of its covariance matrix are calculated to evaluate the spatial correlation of multipath components at different time segments. All these calculation results are summarized into a comprehensive matrix. The rows of this matrix correspond to different feature types and time offsets, and the columns correspond to different time segment pairs, thus forming a high-dimensional spatiotemporal correlation matrix. Its dimension expands rapidly with the number of features and windows.

[0019] Principal component analysis (PCA) is applied to spatiotemporal correlation matrices to reduce data dimensionality while preserving most of the variation information. The calculation process begins with calculating the covariance matrix of the matrix, then solving for the eigenvalues ​​and eigenvectors of this covariance matrix. The magnitude of the eigenvalues ​​reflects the magnitude of the data variance in the direction pointed to by the corresponding eigenvector. Eigenvectors with eigenvalues ​​exceeding a preset threshold are selected as principal components. The threshold is usually set based on the cumulative variance contribution rate; for example, the top k components that achieve a cumulative contribution rate of 95% are retained. These principal components form a projection matrix. Multiplying the original spatiotemporal correlation matrix by the transpose of this projection matrix achieves a linear transformation of the data from a high-dimensional space to a low-dimensional space. The resulting dynamic environment feature matrix not only has a significantly reduced size but also eliminates redundant correlations between the original features, highlighting the main patterns of environmental change. The rows of this matrix correspond to principal components, and the columns correspond to time segments, facilitating subsequent model processing.

[0020] The dynamic environment feature matrix is ​​then fed into an attention-based filtering strategy selection model. This model first maps each feature vector of the input matrix to a higher-dimensional continuous vector space through an embedding layer. This mapping is achieved through a learnable linear transformation or lookup table, aiming to transform discrete or low-dimensional feature representations into distributed representations rich in semantic information. The multi-head attention mechanism is the core of the model. It performs multiple attention calculations in parallel, each called a head. Each head has an independent query, key, and value projection matrix, allowing the model to focus on different aspects of the input information simultaneously. For example, one attention head might focus on capturing long-term dependencies in the time dimension, calculating the association weights between features at the current time point and features at historical time points, while another head might focus on the interactions between different types of environmental features, such as the co-variation patterns of signal strength and noise power. The attention weights of each head are calculated by scaling the dot product, i.e., dividing the dot product of the query vector and the key vector by the square root of the vector dimension, and then normalizing using the Softmax function so that the sum of all weights is one. The normalized weights are then weighted and summed with the value vector to produce the output representation of that head. The output representations of all heads are concatenated and fused through a linear projection layer to form a comprehensive weighted environmental feature representation that incorporates attention information from different perspectives.

[0021] The fully connected output layer receives the weighted feature representations described above and maps them to a vector with a dimension equal to the number of policies in the preset policy library. The value in each dimension represents the matching score between the corresponding filtering policy and the current environment features. The matching score is typically generated by adding a bias to a linear transformation and then applying a sigmoid or linear activation function; a higher score indicates that the policy is more suitable for the current environment. The filter policy combination with the highest matching score is selected as the output, generating an optimal filter policy identifier. This identifier is usually a numeric code or bitmask that encodes the selected time-domain filtering method, frequency-domain filtering method, and their respective parameter settings. For example, some bits of the identifier indicate the type of time-domain filter, the middle bits indicate the cutoff frequency, and the last few bits indicate the number of points in the frequency-domain transformation. The entire selection process is dynamic, running periodically as the environment feature matrix is ​​updated to ensure that the filtering policy can track changes in the environment. The model parameters are obtained through training on historical data, and the training objective is to minimize the gap between the policy selection decision and the actual communication quality indicators.

[0022] Sliding window segmentation strategies need to consider the constraints of real-time computing resources. Adaptive window size adjustment algorithms are typically based on the rate of change of the first-order difference or variance of the signal, setting a rate-of-change threshold. When the rate of change exceeds the threshold for several consecutive time points, the window size is automatically reduced to improve temporal resolution. The size of the overlapping region affects the smoothness and latency of the calculation. Excessive overlap increases computation but improves continuity, while insufficient overlap may lead to insensitive detection of abrupt changes. The overlap ratio is typically set to 25% to 50% of the window length. In correlation calculations, the Pearson correlation coefficient is suitable for signal strength data with obvious linear relationships, while mutual information entropy can capture nonlinear relationships and is suitable for data such as noise power, which may exhibit complex statistical characteristics. The calculation of the covariance matrix is ​​for multipath interference data; each element represents the covariance of signal components on different delay paths at different time segments, revealing the time-varying characteristics of the multipath channel. Threshold selection in principal component analysis is a trade-off. A high threshold retains fewer principal components, resulting in high computational efficiency but potentially losing detailed information; a low threshold retains more components, increasing computational burden but preserving more complete information. In practical deployments, the threshold can be determined through cross-validation to strike a balance between compression ratio and information loss. Whether the generation of the dynamic environment feature matrix is ​​offline or online affects system design. For low-power Bluetooth chips, an online incremental PCA variant is typically used to reduce memory and computational requirements. The number of heads in the attention model is a hyperparameter; too many heads may lead to overfitting and increased computational costs, while too few heads may fail to adequately capture multiple dependencies. Typically, four to eight heads are set. The calculation of attention weights can introduce causal masks to ensure that features at the current time point only focus on historical time points, avoiding information leakage, which is suitable for real-time processing scenarios. Training the attention model requires a large amount of labeled data, i.e., various environment feature matrices and their corresponding optimal policy labels. Data can be collected through simulation or field measurements, and the model parameters are optimized using the gradient descent algorithm. The policy library is built offline, containing a series of validated and effective filtering policy combinations. Each policy is labeled with its applicable typical environmental conditions, such as high-noise environments, strong multipath environments, and static environments. The output of the strategy selection model needs a confidence mechanism. If the highest matching score falls below a certain threshold, a conservative default strategy or a re-evaluation request may be triggered to improve the system's robustness. The entire implementation of Example 1 is tightly integrated into the chip's firmware. The algorithm is optimized to meet low power consumption and real-time requirements, and computationally intensive operations such as matrix multiplication are performed using hardware accelerators.

[0023] Example 2: See Figure 3The parsing of the optimal filtering strategy identifier is fundamental to subsequent processing. This identifier is typically an encoded digital sequence or bitmap structure, with different bit segments corresponding to processing parameters. The parsing process first decodes the identifier according to predefined protocol rules, identifying fields related to time-domain processing. These fields specify the type of time-domain filter to be used, the filter's cutoff frequency or order, and the variable length range of the sliding window. Simultaneously, fields related to frequency-domain processing are parsed, including the choice of frequency-domain transform method, the number of transform points, the type of window function, and the threshold setting rules for frequency-domain filtering. The parsing of time-domain processing parameters needs to be matched with the available filter library within the chip. Filter types may include mean filtering for smoothing, median filtering for suppressing impulse noise, or more complex linear filters based on finite-length and infinite-length unit impulse response theories. The cutoff frequency parameter determines the filter's passband width, and its value may be related to the channel estimation results or noise power. The sliding window length parameter is not a fixed value but a range that can be dynamically adjusted; its specific value will be determined in real-time based on signal characteristics during processing.

[0024] A variable-length sliding window is used for dynamic time-domain filtering of the received signal. The initial value of the window length is set based on the resolved parameter range. During processing, some local statistical characteristics of the signal are continuously monitored, such as the short-time energy change rate or zero-crossing rate. When a rapid change in signal characteristics is detected, such as strong transient interference or abrupt changes in the signal envelope, the length of the sliding window is automatically shortened to improve time resolution and better track transient changes in the signal. Conversely, when the signal is in a relatively stable phase, the window length is appropriately increased to more effectively suppress random noise in a statistical sense. The specific calculation of dynamic time-domain filtering is performed according to the selected filter type. If mean filtering is selected, the signal sampling points within the current sliding window are arithmetically averaged. If median filtering is selected, the sampling points within the window are sorted by amplitude and the median is taken. If a digital filter is selected, the resolved filter coefficients are convolved with the signal data within the window. The filter is designed so that its frequency response has the expected roll-off characteristics at the cutoff frequency, effectively filtering out out-of-band noise while retaining the frequency band where the main energy of the signal is concentrated.

[0025] An adaptive frequency domain transform is performed on the time-domain processed signal. The choice of transform method is determined by the parsed strategy identifier. Commonly used methods include the short-time Fourier transform, which segments the signal on the time axis and performs Fourier analysis segment by segment, or wavelet transform, which has variable time-frequency resolution. The number of transform points is a key parameter; too many points increase computational complexity and latency, while too few points result in insufficient frequency resolution. It is usually adaptively determined based on the signal's effective bandwidth and sampling rate to ensure that the main spectral components of the signal can be distinguished. Before the frequency domain transform, a window function is usually applied to the time-domain processed signal segment. The shape of the window function, such as the Hanning window, Hamming window, or Blackman window, affects the degree of spectral leakage. The window function is parsed from the strategy identifier and multiplied point by point with the signal segment before the transform. Dynamic frequency domain filtering is implemented, with the filtering threshold or transfer function shape set based on the analyzed parameters. The filtering operation may be performed directly in the transformed frequency domain, for example, by multiplying a frequency domain mask point-by-point with the signal spectrum. This mask is set to high gain in the expected passband of the signal spectrum and low gain or zero in the stopband, thereby suppressing narrowband interference at specific frequencies or attenuating frequency bands with high noise power. The filtering threshold can be fixed or dynamically adjusted based on real-time estimation of the noise spectrum, for example, by using the principle of spectral subtraction to subtract an estimated noise spectrum amplitude from the current spectrum amplitude.

[0026] The time-frequency processing results are then reconstructed using a method that strictly corresponds to the frequency domain transform method employed. If a short-time Fourier transform was used previously, an inverse short-time Fourier transform is applied to the filtered frequency domain data. This process involves performing an inverse Fourier transform on the spectrum of each time segment, resulting in a series of time domain segments. These segments are then merged into a continuous time domain signal using either a superposition and addition method or an overlap-preservation method. Careful handling of overlaps between segments is crucial to eliminate boundary effects during the merging process. If a wavelet transform is used, the corresponding wavelet basis functions are applied to the filtered wavelet coefficients using an inverse wavelet transform, gradually reconstructing the original time domain signal at the original scale. The signal reconstruction stage also requires consideration of phase information preservation to ensure that the reconstructed signal does not introduce additional phase distortion. For communication systems, phase continuity is critical for correct demodulation. The final pre-optimized signal should exhibit significantly improved quality compared to the original received signal, characterized by suppression of out-of-band noise and specific interference components, resulting in a clearer signal waveform and creating favorable conditions for subsequent deep signal enhancement. The entire time-frequency joint processing is completed within one processing cycle, with its parameters dynamically controlled by the optimal filtering strategy, achieving a high degree of adaptive matching between the processing strategy and the electromagnetic environment. The variable-length sliding window mechanism requires meticulous design; the window length adjustment algorithm needs to respond quickly to signal changes while avoiding frequent adjustments due to instantaneous fluctuations. A common strategy is to monitor the variance of the signal's first-order difference or short-time energy, triggering window length adjustment only after these statistics exceed a threshold several times consecutively. In dynamic time-domain filtering, filter coefficients may need to be recalculated or interpolated based on the current window length to ensure consistent filter performance. This increases computational complexity but can be optimized by pre-calculating the coefficient sets corresponding to different lengths and storing them in a lookup table.

[0027] In adaptive frequency domain transform, the adaptive selection algorithm for the number of transform points is usually linked to the estimation of the signal's effective bandwidth. The effective bandwidth can be obtained by calculating the bandwidth containing most of the energy in the signal's power spectral density. Then, according to the sampling theorem, the number of transform points should be set to at least twice the ratio of the sampling rate to the effective bandwidth to ensure that the spectrum does not alias. The choice of window function affects spectral leakage. The Hanning window has better sidelobe attenuation but a wider main lobe, while the Hamming window has a narrower main lobe but slightly worse sidelobe attenuation. The choice depends on the emphasis on frequency resolution and anti-interference capability. Threshold setting is crucial for dynamic frequency domain filtering. Fixed thresholds are suitable for environments with relatively stable noise spectra, while dynamic thresholds are more adaptable to changing environments. Dynamic thresholds can be obtained by estimating the noise floor, for example, by finding several frequency points with the lowest power in the signal spectrum, averaging them as an estimate of the noise power, and then setting an offset as the filtering threshold. Spectral subtraction needs to be used cautiously; excessive subtraction can cause musical noise. Therefore, a lower spectral limit is usually set to prevent excessive suppression of signal components. The overlap-addition method in signal reconstruction requires precise control of the size of the overlap region, which is typically half the window length. Weighted summation of the signals in the overlap region smooths the transition. The accuracy of the inverse transform directly affects the quality of the reconstructed signal. It is necessary to ensure strict matching of the kernel functions of the forward and inverse transforms, and the numerical accuracy during calculation must also meet requirements. In low-power chips with fixed-point processing, special attention must be paid to the impact of quantization errors.

[0028] Example 3: The initially optimized signal is fed into a signal enhancement system based on a deep residual network. This system is carefully optimized for the computational and storage constraints of the low-power Bluetooth chip. The network structure strictly controls the number of layers and parameters while ensuring performance. The signal enhancement process begins with segmented windowing of the initially optimized signal. The window function is usually a Hanning window to reduce spectral leakage. The length and overlap rate of the window are balanced according to the expected stability of the signal and the computational load. The segmented signal segments constitute the basic units of the network input. Each signal segment is processed by an initial convolutional layer, which uses multiple convolutional kernels of different sizes to operate in parallel. Small-sized convolutional kernels focus on extracting local details and high-speed changes in the signal waveform, while large-sized convolutional kernels are used to capture the macroscopic trend and envelope features of the signal. Immediately after the convolution operation, a batch normalization layer is connected to standardize the output features, accelerating training convergence and improving generalization ability. Then, a nonlinear transformation is introduced through the ReLU activation function.

[0029] The core of deep residual networks lies in their cross-layer connection structure, namely residual blocks. Each residual block typically contains two to three convolutional layers, each followed by batch normalization and ReLU activation. The input to a residual block not only flows through these convolutional layers but also jumps directly to the block's output via an identity shortcut path, where it is element-wise added to the output of the convolutional layers. This addition operation requires the dimension of the shortcut path to match the dimension of the convolutional layer's output. When the dimensions do not match, a one-to-one convolutional operation is introduced on the shortcut path for linear projection to achieve dimension alignment. The residual learning mechanism allows the network to learn only the residual mapping between the input and output, i.e., the difference between the expected output and the input. This effectively alleviates the gradient vanishing or exploding problems that may occur as the network depth increases, allowing for the construction of deeper networks with stronger representation capabilities. Multiple residual blocks are stacked sequentially to form the main body of the network. The features extracted by shallow residual blocks are usually more local and detailed. The feature reconstruction network is responsible for mapping the high-level feature representations output from deep residual blocks back to the original signal space. It typically consists of a series of deconvolutional or upsampling layers, possibly interspersed with regular convolutional layers. Deconvolutional layers can be viewed as the inverse of convolution; by padding and sliding the convolutional kernels on the learned feature maps, they gradually increase the temporal resolution of the feature maps while reducing the number of feature channels, ultimately outputting single-channel data of the same length as the input signal segment. Upsampling layers may use methods such as nearest-neighbor interpolation or bilinear interpolation to directly expand the temporal dimension of the feature maps, and then refine the results through convolutional layers. During reconstruction, feature maps from residual blocks of different depths may be introduced into different stages of the reconstruction network through additional shortcut connections, achieving multi-scale feature fusion. For example, shallow features containing rich details and deep features containing semantic information can be concatenated or weighted before reconstruction, which helps to recover an enhanced target signal that retains both detail and structural clarity. The final output layer of the network typically uses a linear activation function to directly generate the enhanced signal sample values.

[0030] Multidimensional quality assessment of the enhanced target signal is a crucial step in closed-loop optimization. The assessment first focuses on calculating the bit error rate (BER). This is typically not done by directly analyzing large amounts of user data, but rather by using known preambles, access addresses, or cyclic redundancy check (CRC) fields in the data packets as reference sequences. The receiver demodulates and decodes the enhanced signal, comparing the recovered known bits (such as the preamble) bit by bit with the locally stored original sequence, counting the number of mismatched bits, and then dividing by the total number of compared bits to obtain an instantaneous BER estimate. Signal-to-noise ratio (SNR) assessment requires estimating both signal power and noise power. Signal power can be approximated by calculating the variance or mean square of the enhanced target signal over the effective data period. Noise power estimation is more challenging. One approach is to sample the signal during idle periods between data packets or known quiet periods and calculate the power of these noise samples. Another approach is to estimate the background noise floor level from the time period containing the signal using spectral analysis techniques. Phase consistency metrics are used to assess the stability of a signal phase, which is particularly important for systems employing phase modulation. Assessment methods may include analyzing the variance of the derivative of the instantaneous phase of the signal (i.e., the instantaneous frequency) to check whether it fluctuates excessively around the nominal carrier frequency, or using the variance of the error signal of the phase-locked loop to indirectly reflect the stability of phase tracking.

[0031] A three-dimensional evaluation matrix is ​​constructed, incorporating time and index dimensions. The time dimension consists of continuous and equally spaced evaluation time points, each corresponding to the center of an evaluation window. The index dimension includes three specific values ​​calculated within the evaluation window: bit error rate, signal-to-noise ratio, and phase consistency. Therefore, this matrix can be viewed as a series of two-dimensional slices arranged along the time axis, each slice being a vector containing three quality index values. A sliding weighted average algorithm is used to smooth the three-dimensional evaluation matrix to generate a signal quality evaluation matrix. Smoothing is performed individually for each quality index, applying a sliding window along the time dimension. Each historical data point within the window is assigned a weight, which typically varies with the distance of the data point from the current time, with more recent data points receiving higher weights and older data points receiving decreasing weights. For a given index at time... smoothing value Its calculation can be expressed as the calculation of each historical value within the window. Perform a weighted average: in: This represents the total number of time points contained in the sliding window. It is an index within the window. Is assigning an index Weighting coefficients for historical values That is at the moment The raw value of the calculated quality indicator. Weighting coefficients. A common choice is the exponentially decreasing form, i.e. ,in It is a decay factor between zero and one. A larger value indicates a more recent data point. Smoothing effectively suppresses drastic fluctuations that may occur during a single evaluation due to random factors, making the signal quality evaluation matrix more reflective of long-term quality trends and providing a more reliable and stable basis for subsequent dynamic parameter adjustments. The entire evaluation process runs periodically, continuously updating the evaluation matrix, thereby achieving continuous monitoring of the receiver link performance.

[0032] The specific configuration of deep residual networks requires careful consideration. A shallow network may lack sufficient learning capacity, while an excessively deep one increases computational complexity and latency. For Bluetooth signals, stacking several to a dozen layers of residual blocks is common. Batch normalization layers behave differently during training and inference; in inference, they use global statistics estimated from training data rather than statistics from the current batch. The number of filters (i.e., output channels) in the convolutional layers of residual blocks typically increases gradually with network depth to compensate for information loss due to reduced temporal resolution. The upsampling factor of the feature reconstruction network needs precise calculation to ensure the final output signal length matches the input segment. During multi-scale feature fusion, feature maps at different scales may need to be cropped or interpolated to align dimensions. Network training requires a large amount of labeled data, i.e., noisy input signals and their corresponding ideal clean signals. This data can be generated through simulation or collected in a controlled environment. The accuracy of bit error rate estimation depends on the length of the reference sequence; a short sequence results in large estimation variance, while a long sequence results in large estimation latency, requiring a trade-off. In signal-to-noise ratio (SNR) estimation, idle period noise estimation methods rely on accurate quiet period detection. In phase consistency assessment, calculating instantaneous frequencies requires phase differentiation, and the issue of phase entanglement must be considered. The time resolution of the three-dimensional assessment matrix (i.e., the interval between assessment time points) needs to be set; too short an interval results in frequent calculations, while too long an interval may miss important changes. The window size for the moving weighted average also needs to be considered. and weighting coefficients The choice of weights affects the smoothing effect. Weights with large windows and rapid decay can quickly forget the past but are more resistant to fluctuations, while weights with small windows and slow decay track trend changes more closely but are more susceptible to instantaneous fluctuations. Exponentially weighted moving average is another commonly used smoothing method. The output of the entire quality assessment module, namely the signal quality assessment matrix, is a key input for the system to perform self-diagnosis and parameter optimization.

[0033] Example 4: The signal quality assessment matrix serves as input. Its structure is a multi-dimensional array arranged in a time series, with each row corresponding to an assessment time point. The columns contain the specific values ​​of indicators such as bit error rate (BER), signal-to-noise ratio (SNR), and phase consistency calculated at that time. The analysis process first requires identifying outlier data points in this matrix. The definition of an anomaly is based on a pre-defined normal operating range for each quality indicator. For example, the BER might have an upper limit threshold, the SNR a lower limit threshold, and phase consistency a minimum allowable value. Identifying anomalies focuses not only on whether an indicator exceeds its limits at a single time point but also on its distribution pattern. For instance, is it an isolated, sudden anomaly or a continuous period of low quality? A clustering of anomalies on the time axis may indicate a persistent source of interference or equipment degradation, while a scattered distribution may stem from random external disturbances. The analysis also needs to examine the correlation between anomalies in different indicators. For example, does a sharp increase in BER accompany a sharp decrease in SNR, or does the deterioration of phase consistency precede other indicators? This correlation provides clues for locating the root cause of the problem.

[0034] Calculating the deviation between the current parameter configuration and the ideal configuration is a crucial step in quantifying adjustment needs. The current parameter configuration refers to the values ​​of various parameters currently being used by the RF front-end and baseband processing unit, such as low-noise amplifier gain, automatic gain control target power, channel filter bandwidth, and analog-to-digital converter sampling rate. The ideal configuration is not a fixed value but a dynamic concept, potentially derived from several sources: first, a set of parameters recorded in a historical database that produce the best signal quality assessment results under similar environmental characteristics; second, the theoretically optimal value calculated based on the current channel estimation using a theoretical model; and third, a set of baseline parameters established during system initialization or calibration. Deviation calculation is performed independently for each parameter requiring adjustment. Essentially, it quantifies the difference between the current parameter value and its corresponding ideal value. This difference can be represented by absolute difference, relative difference, or normalized distance. For example, for parameters like gain, the absolute difference between the current and ideal values ​​might be calculated; for bandwidth or sampling rate, the relative percentage change might be calculated. Ultimately, each parameter will receive a quantified deviation value. A larger value indicates a greater deviation from the ideal state, and a more urgent need for adjustment.

[0035] The process generates parameter adjustment instructions containing both direction and magnitude. The direction indicates whether the parameter needs to be increased or decreased, directly determined by the relationship between the current parameter value and the ideal value. If the current value is less than the ideal value, the adjustment direction is to increase; conversely, it is to decrease. Determining the adjustment magnitude is more complex, requiring consideration of the parameter's deviation, its sensitivity to signal quality, and the severity of current signal quality degradation. A typical approach is to use an algorithm based on proportional-integral-derivative (PID) control. The proportional term is directly proportional to the calculated deviation; the larger the deviation, the larger the initial adjustment magnitude. The integral term considers the accumulation of historical deviations; if a parameter continuously deviates from the ideal value, even a small individual deviation will lead to a gradual increase in the adjustment magnitude due to the cumulative effect. The derivative term focuses on the trend of deviation changes; if the deviation is rapidly increasing, an additional adjustment is introduced to suppress the deterioration. The adjustment instructions are ultimately assembled into a structured data packet, explicitly listing the identifier of each parameter to be adjusted, the specified adjustment direction, and the calculated adjustment magnitude. Parameter adjustment commands are sent to the corresponding hardware control registers of the chip. The gain of the low-noise amplifier is adjusted by changing its bias voltage or current. Gain adjustment directly affects the receiver's sensitivity and dynamic range. Excessive gain may cause saturation of subsequent amplifiers, introducing nonlinear distortion; insufficient gain will fail to effectively amplify weak signals. The response time parameter of the automatic gain control loop determines how quickly the loop tracks changes in signal strength. A response time that is too fast will make the loop overly sensitive to noise fluctuations, causing jitter; a response time that is too slow will fail to effectively handle rapid signal fading. Adjustment is typically achieved by changing the bandwidth of the loop filter. The bandwidth adjustment of the channel selection filter needs to match the signal bandwidth of the current channel. Excessive bandwidth will introduce more adjacent channel interference; insufficient bandwidth will truncate useful spectral components of the signal, leading to distortion. Adjustment is typically achieved by switching the filter array or modifying the filter coefficients through programming. Optimizing the demodulator's decision threshold is crucial for reducing the bit error rate, especially at low signal-to-noise ratios, where even small deviations in the threshold can significantly increase the false positive rate. After each parameter adjustment, the system does not immediately make the next adjustment, but enters a short observation period to continuously monitor the changes in the signal quality evaluation matrix in order to evaluate the effect of the adjustment. If the quality indicators tend to improve, the subsequent adjustments may continue in the current direction. If the indicators deteriorate or do not improve, the adjustments may be reversed or the opposite direction may be tried. Through this iterative trial method, the system gradually approaches the optimal parameter combination under the current environment. See Table 1.

[0036] Table 1: Example of Signal Quality Assessment Matrix and Parameter Adjustment Analysis Timestamp (ms) Bit Error Rate (BER) Signal-to-noise ratio (SNR, dB) Phase consistency (Index) Exception type identifier 100 0.001 15.2 0.92 none 200 0.002 14.8 0.91 none 300 0.015 9.5 0.75 Joint anomalies in bit error rate, signal-to-noise ratio, and phase consistency 400 0.012 10.1 0.78 Joint anomalies in bit error rate, signal-to-noise ratio, and phase consistency 500 0.003 13.5 0.88 Slightly abnormal signal-to-noise ratio Analysis based on the signal quality assessment matrix segment shown in the table above reveals joint anomalies in bit error rate, signal-to-noise ratio (SNR), and phase consistency at timestamps of 300ms and 400ms, persisting for two cycles. This suggests a potentially strong and persistent interference or significant deterioration in channel conditions. Assuming that in the historically optimal configuration, for a similar interference environment, the ideal low-noise amplifier gain is 20dB and the ideal channel filter bandwidth is 1.5MHz, while the current configurations are 25dB and 1.2MHz respectively, the deviation is calculated as |25-20|=5dB for gain and |1.2-1.5|=0.3MHz for bandwidth. Due to the severe and persistent SNR deterioration, the adjustment range will be large. Adjustment commands are generated: Low-noise amplifier gain, decreasing (because the current value of 25dB is greater than the ideal value of 20dB), with an initial decrease of 3dB; Channel filter bandwidth, increasing (the current 1.2MHz is less than the ideal 1.5MHz), with an initial increase of 0.4MHz. Simultaneously, the deterioration in phase consistency may affect the carrier recovery loop bandwidth of the tuner. After applying these adjustments, the signal-to-noise ratio (SNR) recovered to 13.5 dB and the bit error rate (BER) decreased to 0.003 at a 500 ms time point, indicating that the adjustment direction was correct but may not be fully implemented. In the next adjustment cycle, further minor adjustments to the gain and bandwidth may be made in the same direction until the indicators stabilize within the normal range. The entire dynamic adjustment process constitutes a closed-loop control system, the goal of which is to maintain stable receiver performance in a changing wireless environment.

[0037] Example 5: The dynamic mapping relationship between environmental features and receiving parameters is established on a continuously updated data structure, which can be viewed as a constantly growing historical experience database. Each record in the database contains two main parts: one part is the environmental feature vector at the time of record generation, which comes from the principal component values ​​of the dynamic environmental feature matrix after dimensionality reduction by principal component analysis. These components condense the key spatiotemporal patterns of signal strength, noise power, and multipath interference in the environment at that time. The other part is the set of receiving parameter configurations that have been verified as effective through signal quality assessment or have finally stabilized through parameter adjustment processes, such as low-noise amplifier gain, automatic gain control setpoint, and channel filter bandwidth. The mapping relationship can be stored in a lookup table, where the environmental feature vector is quantized into intervals or clusters, with each interval pointing to a set of recommended parameter configurations. Alternatively, a more flexible regression model can be used, such as a lightweight neural network or decision tree model, which can accept continuous environmental feature vectors as input and directly output the predicted parameter configuration vector. During system operation, whenever a round of parameter adjustment is completed and the signal quality evaluation matrix shows that the performance has reached a satisfactory state, a recording operation is triggered. The current environmental feature vector and the parameter configuration that is currently in effect are combined as a new data point and added to the historical mapping dataset. As the running time accumulates, this dataset will cover more and more diverse environmental conditions, making the mapping relationship more and more accurate and comprehensive.

[0038] Predicting optimal parameter configuration based on environmental change trends is the core of pre-adjustment. Trend analysis relies on observing a series of recently collected environmental feature vector sequences; this sequence is arranged chronologically, with each vector representing a snapshot of the environment at a certain moment. The analysis algorithm needs to extract the patterns and directions of environmental change from this time series. For example, linear prediction methods can be used to fit a trend line to each major environmental feature component, estimating its potential value at a short future time point. For more complex nonlinear changes, state estimation methods such as Kalman filtering may be used, treating the environmental features as the state of a dynamic system, estimating its future state through model prediction and measurement updates. The accuracy of the prediction is highly dependent on the inherent regularity of environmental change and the quality of the observation data. In relatively stable environments, linear prediction may be effective enough, but in rapidly changing environments full of sudden changes, the prediction error will increase. The prediction time span is also a factor that needs to be weighed. Predictions that are too far ahead have high uncertainty and low accuracy; predictions that are too far behind lose the meaning of pre-adjustment. Typically, the prediction period is matched with the system processing frame length or parameter adjustment period.

[0039] After predicting the environmental feature vectors for future moments, the k nearest historical environmental feature vectors are found from the historical mapping dataset. Proximity is typically measured using Euclidean distance or Manhattan distance. The distance between the predicted vector and each vector in the historical dataset is calculated, and the k historical vectors with the smallest distances are selected. In the parameter prediction stage, a weighted voting method or locally weighted regression is used to synthesize the information from these k nearest neighbors. If a weighted voting method is used, the parameter configuration corresponding to each nearest historical vector is considered a "candidate," and its weight is determined based on the distance between the historical vector and the predicted vector; the closer the distance, the higher the weight. Then, a weighted average is taken of all candidate values ​​for each parameter (e.g., gain value) to obtain the predicted value of that parameter. The locally weighted regression method is more refined. Within a local region near the predicted vector, a simple regression model (e.g., linear regression) is built for each parameter to be predicted. This model is fitted using the environmental features of the k nearest neighbors and their corresponding parameter values. The fitted model is then used to calculate the parameter value corresponding to the predicted vector. This method better captures the local dependencies between parameters and environmental features.

[0040] The predicted optimal parameter configuration is pre-loaded into the relevant control registers of the RF front-end and baseband processing unit. This process occurs before actual environmental changes and is a proactive configuration update. The effectiveness of parameter pre-tuning needs to be verified by the actual generated signal quality evaluation matrix. The system records the signal quality actually observed under the pre-tuned parameters and compares it with the quality that could be achieved without pre-tuning (the latter is usually estimated based on the parameter configuration at the time of prediction and the actual environment). If the signal quality after pre-tuning is better than or equal to the expected quality, the pre-tuning is considered successful, and this experience is reinforced and recorded in the historical mapping dataset. If the quality decreases after pre-tuning, the system analyzes the reason, whether it is due to inaccurate prediction or unreliable historical mapping in that region, and may trigger an emergency parameter readjustment. At the same time, this failed pre-tuning experience is also recorded to avoid similar erroneous predictions in the future. This continuous feedback and learning mechanism enables the pre-tuning system to continuously optimize itself. Over time, in frequently experienced environmental change patterns, the accuracy and effectiveness of pre-tuning will become increasingly higher.

[0041] Suppose a Bluetooth device is moving from an open indoor environment to a warehouse area filled with metal shelves. The historical mapping dataset may contain data previously recorded at similar warehouse entrances. For example, environmental feature vectors might show a significant increase in multipath interference and a slow decrease in signal strength. The corresponding successful parameter configuration would be to moderately increase the roll-off factor of the channel filter to enhance adjacent channel rejection and slightly reduce the target power of automatic gain control to avoid saturation caused by deep fading due to multipath interference. When the device approaches such an environment again, the real-time monitored environmental feature vector sequence shows that the multipath interference component is continuously increasing. The trend prediction algorithm determines that this component will reach a level similar to the historical warehouse entrance data in the next 100 milliseconds. The system then identifies k historical feature vectors with similar upward trends and final levels from the historical data. The parameter configurations corresponding to these vectors all include increasing the filter roll-off factor and decreasing the gain. By weighted averaging these historical parameters, a set of predictive configurations is obtained. Therefore, tens of milliseconds before the device actually enters the warehouse area, this set of parameters has been pre-configured into the chip. When the device actually enters the environment with severe multipath, the parameters of its receiving link are already in a relatively optimized state, thus avoiding possible brief communication interruptions or sharp performance degradation due to parameter mismatch during sudden environmental changes, and achieving a smooth transition.

[0042] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for optimizing signal reception in a low-power Bluetooth chip, characterized in that, Includes the following steps: Step 1: Collect signal strength, noise power, and multipath interference data in the environment through a multi-channel RF front-end to generate an original environmental feature set; Step 2: Perform spatiotemporal correlation analysis on the original environmental feature set to construct a dynamic environmental feature matrix; Step 3: Input the dynamic environment feature matrix into the attention-based filtering strategy selection model and output the optimal filtering strategy identifier; Step 4: Based on the optimal filtering strategy identifier, call the corresponding adaptive filtering algorithm to perform time-frequency joint processing on the received signal and generate a preliminary optimized signal.

2. The low-power Bluetooth chip signal reception optimization method according to claim 1, characterized in that, Step 2 includes: Step 2.1: Perform sliding window segmentation on the original environmental feature set to form multiple time segment feature sets; Step 2.2: Calculate the correlation coefficient between feature sets of adjacent time segments and construct a spatiotemporal correlation matrix; Step 2.3: Dimensionality reduction of the spatiotemporal correlation matrix is ​​performed using principal component analysis to generate the dynamic environment feature matrix.

3. The low-power Bluetooth chip signal reception optimization method according to claim 2, characterized in that, Step 3 includes: Step 3.1: Input the dynamic environment feature matrix into the attention-based filtering strategy selection model; Step 3.2: Calculate the dynamic weight distribution of each environmental feature using a multi-head attention mechanism; Step 3.3: Select the filter strategy combination with the highest matching degree from the preset strategy library according to the dynamic weight distribution, and generate the optimal filter strategy identifier.

4. The low-power Bluetooth chip signal reception optimization method according to claim 3, characterized in that, Step 4 includes: Step 4.1: Parse the time-domain processing parameters and frequency-domain processing parameters in the optimal filtering strategy identifier; Step 4.2: Perform dynamic time-domain filtering on the received signal using a variable-length sliding window; Step 4.3: Perform adaptive frequency domain transformation on the time-domain processed signal and implement dynamic frequency domain filtering; Step 4.4: Reconstruct the signal from the time-frequency processing results to generate the preliminary optimized signal.

5. The low-power Bluetooth chip signal reception optimization method according to claim 4, characterized in that, Also includes: Step 5: Input the preliminary optimized signal into a signal enhancement system based on a deep residual network, and output the enhanced target signal; Step 6: Perform a multi-dimensional quality assessment on the enhanced target signal to generate a signal quality assessment matrix.

6. The low-power Bluetooth chip signal reception optimization method according to claim 5, characterized in that, Step 5 includes: Step 5.1: Perform segmented windowing processing on the preliminary optimized signal to generate a signal segment sequence; Step 5.2: Extract multi-scale signal features through the convolutional layers of the deep residual network; Step 5.3: Utilize the cross-layer connection structure to fuse feature representations of different depths, and generate the enhanced target signal through the feature reconstruction network.

7. The low-power Bluetooth chip signal reception optimization method according to claim 6, characterized in that, Step 6 includes: Step 6.1: Calculate the bit error rate, signal-to-noise ratio, and phase consistency index of the enhanced target signal; Step 6.2: Construct a three-dimensional evaluation matrix that includes time and indicator dimensions; Step 6.3: Generate the signal quality evaluation matrix using a moving weighted average algorithm.

8. The low-power Bluetooth chip signal reception optimization method according to claim 7, characterized in that, Also includes: Step 7: Dynamically adjust the receiving parameter configuration according to the signal quality evaluation matrix to generate a new combination of receiving parameters.

9. The low-power Bluetooth chip signal reception optimization method according to claim 8, characterized in that, Step 7 includes: Step 7.1: Analyze the distribution of abnormal data points in the signal quality assessment matrix; Step 7.2: Calculate the deviation between the current parameter configuration and the ideal configuration, and generate parameter adjustment instructions that include the adjustment direction and magnitude.

10. The low-power Bluetooth chip signal reception optimization method according to claim 9, characterized in that, Also includes: Step 8: Establish a dynamic mapping relationship between environmental characteristics and receiving parameters, predict the optimal parameter configuration based on environmental change trends, and achieve parameter pre-adjustment.

Images