Digital event triggering method and system based on baseband data sequence edge detection

Abstract

The application discloses a digital event triggering method and system based on baseband data sequence edge detection, relates to the field of baseband data detection, and comprises the following steps: removing direct current bias, suppressing low-frequency drift and filtering noise from an original baseband data sequence to obtain a preprocessed baseband data sequence; determining a morphological structure element according to a signal-to-noise ratio and an expected event width of a target service scenario, and extracting a local mutation profile through an open operation and a top-hat transformation; generating a double threshold based on a morphological gradient, statistical characteristics and noise estimation to extract a candidate event interval; performing multi-dimensional feature matching on the candidate event segment in combination with a reference event feature template to determine a target effective event; and determining an event time center point and start and end boundaries according to a local event energy sequence and event positioning constraint information and outputting triggering information containing an event timestamp and event segment data. Thus, the stability, accuracy and time positioning reliability of digital event triggering in a complex interference environment are improved.

Description

Technical Field

[0001] This application relates to the field of baseband data detection technology, and in particular to a digital event triggering method and system based on baseband data sequence edge detection. Background Technology

[0002] In modern communication, radar signal analysis, and IoT data acquisition systems, baseband data detection is typically used to identify business-meaning information patterns, edge changes, and digital events from continuous or discrete baseband sequences, such as burst signals, synchronization headers, training sequences, or abrupt symbols. These detection results are often used to trigger sampling buffering, clock recovery, signal synchronization, demodulation decisions, or upper-layer processing flows; therefore, the stability of the detection and the timeliness of the triggering directly affect the reliability of subsequent links. With the widespread application of edge computing and low-power sensing devices, the baseband triggering process needs to be completed under limited computing power, storage, and communication resources, placing higher demands on real-time performance, anti-interference capabilities, and scenario adaptability.

[0003] Currently, common approaches for edge capture and event triggering of baseband signals include fixed threshold triggering, amplitude peak detection, linear filtering and shaping, differential edge detection, and synchronization detection based on clock recovery loops. These methods mostly rely on signal amplitude, zero-crossing changes, local slope, or energy characteristics within a window as judgment criteria. They are characterized by simple implementation, direct response, and low hardware deployment costs, and can meet basic usage requirements in scenarios with relatively stable signal characteristics, low noise levels, or well-defined triggering conditions.

[0004] However, in practical applications, baseband signals are often affected by factors such as front-end circuit noise, quantization errors, electromagnetic interference, propagation attenuation, low-frequency baseline drift, and random pulse disturbances, resulting in significant irregularities in their waveform edges and local structures. Relying solely on fixed thresholds, single amplitude characteristics, or simple linear variation features for judgment is susceptible to interference from background drift and occasional noise peaks, making the triggering results highly sensitive to threshold settings, signal amplitude, and local noise conditions. In environments with multi-morphological signals, weak signals, or strong interference, this approach also has relatively limited ability to express the true edge structure and sustained characteristics of effective events, thus making it difficult to maintain consistently stable event capture performance.

[0005] Furthermore, in IoT edge nodes, low-power wireless receivers, and distributed sensing terminals, front-end triggering results typically affect processor wake-up, RF link start / stop, buffer allocation, and data upload. If the triggering mechanism is insufficiently adaptable to complex noise and dynamic baselines, it may increase invalid triggers, duplicate acquisitions, and redundant transmissions, thereby increasing the burden on back-end processing and consuming limited energy and bandwidth resources. Summary of the Invention

[0006] This application provides a digital event triggering method, system, storage medium, computer program product, and electronic device based on baseband data sequence edge detection, which at least solves the problem of insufficient accuracy and stability of baseband digital event triggering under complex interference environments in the prior art.

[0007] In a first aspect, embodiments of this application provide a digital event triggering method based on baseband data sequence edge detection. The method includes: acquiring a raw baseband data sequence to be processed, and performing DC bias removal, low-frequency drift suppression, and noise filtering on the raw baseband data sequence to obtain a preprocessed baseband data sequence; determining morphological structuring elements based on the signal-to-noise ratio of the preprocessed baseband data sequence and the expected event width corresponding to the target business scenario, and performing morphological opening and top-hat transformation on the preprocessed baseband data sequence using the morphological structuring elements to obtain a top-hat feature sequence for characterizing the contour of local waveform mutations; performing dilation and erosion operations on the top-hat feature sequence using the morphological structuring elements, and obtaining a morphological gradient sequence based on the difference between the dilation and erosion results; generating a dual threshold based on the statistical characteristics and noise estimation results of the morphological gradient sequence, and extracting candidate events from the morphological gradient sequence using the dual threshold. The process involves: obtaining a baseline event feature template corresponding to the target business scenario; extracting candidate event fragments and corresponding multi-dimensional candidate event features from the candidate event interval; matching and verifying the multi-dimensional candidate event features with the baseline event feature template to determine the target valid event and its corresponding valid event interval; extracting valid event sampling fragments corresponding to the valid event interval from the preprocessed baseband data sequence; generating a local event energy sequence based on the amplitude representation value of the valid event sampling fragments; and determining the time center point, time start boundary, and time end boundary of the target valid event based on the local event energy sequence and event location constraint information; wherein the event location constraint information is generated based on the event duration corresponding to the baseline event feature template and the interval boundary of the valid event interval; and generating and outputting trigger information containing event timestamps and event fragment data based on the time center point, the time start boundary, and the time end boundary.

[0008] Secondly, embodiments of this application provide a digital event triggering system based on baseband data sequence edge detection. The system includes: a baseband data preprocessing unit, used to acquire the original baseband data sequence to be processed, and to perform DC bias removal, low-frequency drift suppression, and noise filtering on the original baseband data sequence to obtain a preprocessed baseband data sequence; a morphological feature extraction unit, used to determine morphological structural elements based on the signal-to-noise ratio of the preprocessed baseband data sequence and the expected event width corresponding to the target business scenario, and to perform morphological opening and top-hat transformation on the preprocessed baseband data sequence using the morphological structural elements to obtain a top-hat feature sequence for characterizing the contour of local waveform mutations; and a dynamic dual-threshold initial screening unit, used to perform dilation and erosion operations on the top-hat feature sequence using the morphological structural elements, and to obtain a morphological gradient sequence based on the difference between the dilation and erosion results, to generate dual thresholds based on the statistical characteristics and noise estimation results of the morphological gradient sequence, and to extract candidate event regions from the morphological gradient sequence using the dual thresholds. The system comprises: a multi-dimensional feature template matching unit, used to acquire a benchmark event feature template corresponding to the target business scenario, extract candidate event segments and corresponding multi-dimensional candidate event features from the candidate event interval, and match and verify the multi-dimensional candidate event features with the benchmark event feature template to determine the target valid event and its valid event interval; an event energy boundary positioning unit, used to extract valid event sampling segments corresponding to the valid event interval from the preprocessed baseband data sequence, generate a local event energy sequence based on the amplitude representation value of the valid event sampling segments, and determine the time center point, time start boundary, and time end boundary of the target valid event based on the local event energy sequence and event positioning constraint information; wherein, the event positioning constraint information is generated based on the event duration corresponding to the benchmark event feature template and the interval boundary of the valid event interval; and a digital event trigger output unit, used to generate and output trigger information containing event timestamps and event segment data based on the time center point, the time start boundary, and the time end boundary.

[0009] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the digital event triggering method based on baseband data sequence edge detection according to any embodiment of the present application.

[0010] Fourthly, embodiments of this application provide a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the digital event triggering method based on baseband data sequence edge detection according to any embodiment of this application.

[0011] Fifthly, embodiments of this application provide a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the digital event triggering method based on baseband data sequence edge detection according to any embodiment of this application.

[0012] The digital event triggering method and system based on baseband data sequence edge detection provided in this application can achieve at least the following technical effects:

[0013] By combining the preprocessed baseband data sequence with the expected event width and current signal-to-noise ratio corresponding to the target service scenario, the structural element used for morphological processing is determined. Local abrupt change contours and candidate event intervals are then extracted based on morphological opening operations, top-hat transformations, and morphological gradients. Since the scale of the structural element matches the persistent characteristics of the target event, morphological processing can suppress low-frequency background fluctuations and non-target gradual change components while highlighting edge abrupt change information that conforms to the event scale. Furthermore, by combining morphological gradients and dual-threshold screening, the formation of candidate intervals is simultaneously constrained by the intensity of local structural changes and the statistical level of noise. Therefore, the event triggering criterion is no longer limited to a single-point amplitude or a fixed threshold, but is transformed into a comprehensive judgment of waveform morphological changes at the target scale, which helps improve the stability, completeness, and anti-disturbance capability of candidate event capture.

[0014] After obtaining the candidate event intervals, a baseline event feature template corresponding to the target business scenario is further introduced to match and verify the multidimensional features of the candidate event fragments. After determining the target valid event, the time center point, time start boundary, and time end boundary are determined by combining the local event energy sequence and event location constraint information. This process ensures that the confirmation of a valid event depends not only on edge strength but also on the event morphology, duration, and local energy distribution. Simultaneously, the event location process does not use the boundaries of the initially screened candidate intervals but performs temporal correction based on the energy concentration location and the template's persistence pattern. This improves the correspondence of the target valid event identification, reduces triggering deviations caused by random pulses, local noise peaks, or boundary jitter, and makes the output event timestamps and event fragment data suitable for subsequent cache triggering, synchronization processing, demodulation decisions, or upper-layer business calls.

[0015] This technical solution constructs a digital event triggering link that connects scale-adaptive morphological mutation enhancement, gradient dual-threshold candidate extraction, template matching verification, and energy-constrained localization. This link transforms baseband event triggering from single-frame discrimination into a comprehensive judgment process that takes into account event scale, local morphology, noise status, template consistency, and energy localization. Thus, in scenarios where baseband data contains complex noise, dynamic baselines, and multi-morphological interference, it helps improve the stability, accuracy, and time-localization reliability of digital event triggering, and reduces the system resource consumption caused by invalid triggering and redundant acquisition and transmission. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating an example of a digital event triggering method based on baseband data sequence edge detection according to an embodiment of this application is shown;

[0018] Figure 2 A flowchart illustrating an example of determining morphological structural elements in a method according to an embodiment of this application is shown.

[0019] Figure 3 A flowchart illustrating an example of obtaining a morphological gradient sequence and extracting candidate event intervals from it according to an embodiment of this application is shown.

[0020] Figure 4 This paper illustrates the overall mechanism architecture of a digital event triggering method based on baseband data sequence edge detection provided in an embodiment of this application.

[0021] Figure 5 A schematic diagram of the simulation results of the comparative experiment on receiver operating characteristic curves of different methods in a low signal-to-noise ratio environment is shown.

[0022] Figure 6 A comparative experimental simulation diagram shows an example of the trigger time error distribution of different methods under a specific signal-to-noise ratio condition.

[0023] Figure 7 A comparative simulation diagram illustrating the engineering trade-off between system-level global power consumption and front-end computing overhead in an edge computing scenario using different methods is shown.

[0024] Figure 8A structural block diagram of an example of a digital event triggering system based on baseband data sequence edge detection according to an embodiment of this application is shown. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] It should be noted that in high-speed serial links and some baseband reception scenarios, current technologies often employ OS-CDR (Oversampling Clock and Data Recovery) or PLL-CDR (Phase-Locked Loop Clock and Data Recovery) schemes for clock recovery and edge capture. For example, in a filtered CDR or PLL-CDR structure, edge detectors can extract transition information from the NRZ (Non-Return-to-Zero) data sequence, and combine this with nonlinear processing, narrowband filtering, or PLL control loops to enhance clock-related components, thereby achieving phase adjustment of the local sampling clock. This type of scheme has certain applicability in high-speed serial data recovery, but its performance usually depends on well-defined data transitions and relatively stable clock components. When the input baseband sequence has waveform distortion, strong noise disturbances, or rapidly changing operating conditions, factors such as device response speed, loop phase jitter, duty cycle control, and lock-in state maintenance may affect the stability of edge capture.

[0027] In some digital receivers or signal processing front-ends, current technologies still employ peak detection and DC compensation mechanisms similar to those in IF-to-Baseband (Intermediate Frequency-to-Baseband) receivers. These mechanisms typically window the amplitude or absolute values ​​of samples in the I / Q data path, record peak values ​​or power levels, and compensate for DC bias through accumulation, averaging, or estimation updates. While this approach is useful for power measurement, gain control, and static bias suppression, its focus is primarily on amplitude levels or statistical energy changes, with limited ability to express waveform edge variations, event persistence characteristics, and structural correlations between consecutive samples. When the effective digital event and the peak value of occasional noise are close in amplitude, relying solely on amplitude statistics for triggering can easily lead to edge capture results being affected by local noise conditions.

[0028] Furthermore, in oscilloscope triggering systems and similar sampling devices, current technologies widely employ acquisition and control methods such as level triggering, slope triggering, and peak capture. These methods can display repetitive signals or capture some transient changes in conventional waveform observations, and are characterized by simple implementation and direct response. However, for bipolar, multi-level amplitude, or baseband sequences with large morphological changes, a single trigger level, fixed slope conditions, or local peak recordings are insufficient to fully reflect the continuous change relationship before and after the event. The triggering results are often quite sensitive to threshold settings, sampling windows, and transient noise.

[0029] In edge node intelligence applications, IoT nodes and low-power sensing terminals typically need to perform initial data interpretation and effective information filtering locally to reduce continuous reliance on backend processors, RF links, and cloud uploads. For baseband event triggering, the front-end detection results not only determine whether a single event is identified but also affect cache switching, link wake-up, and data upload strategies. Therefore, the relevant triggering mechanisms in engineering implementation usually need to balance detection stability, dynamic environment adaptability, and front-end computational overhead.

[0030] It should be understood that the above description of the relevant technologies is intended only to help the public better understand the inventive spirit and motivation of this application, and is not intended to limit this application. Furthermore, the technical solutions described in the above-mentioned relevant technologies are not prior art, and may also be undisclosed technical solutions, such as those under research or in the laboratory stage.

[0031] The technical solutions in this application, including the collection, storage, use, processing, transmission, provision, and disclosure of users' personal information, comply with relevant laws and regulations and do not violate public order and good morals.

[0032] Figure 1 A flowchart illustrating an example of a digital event triggering method based on baseband data sequence edge detection according to an embodiment of this application is shown.

[0033] Regarding the execution subject of the method in the embodiments of this application, it can be any controller or processor with computing or processing capabilities, such as a baseband signal processing controller, which implements the method in the embodiments of this application by running programs or instructions stored in a storage medium.

[0034] In some examples, it may be integrated into an electronic device or terminal through software, hardware, or a combination of both, and the type of terminal or electronic device may be diverse.

[0035] like Figure 1 As shown, in step S110, the original baseband data sequence to be processed is obtained, and the original baseband data sequence is subjected to DC bias removal, low frequency drift suppression and noise filtering to obtain a preprocessed baseband data sequence.

[0036] Here, the raw baseband data sequence can originate from a wireless communication receiver, a radar baseband processing link, an IoT edge sensing node, or other digital acquisition devices. It can be a discrete-time sequence sampled by an analog-to-digital converter, or a baseband sampling sequence composed of in-phase and quadrature components. For example, in a wireless communication receiving scenario, the raw baseband data sequence can correspond to the I / Q sampling stream output by the receiver front end; in a radar signal analysis scenario, the raw baseband data sequence can correspond to the echo baseband sequence after down-conversion and sampling; in an IoT edge node scenario, the raw baseband data sequence can correspond to the burst wake-up signal or short frame start segment acquired by the node in a low-power listening state.

[0037] In actual data acquisition, the raw baseband data sequence is often affected by low-frequency baseline drift caused by DC bias of the front-end analog circuit, temperature or device state changes, as well as broadband random noise or out-of-band interference. For example, during long-term standby listening of a low-power wireless node, temperature changes in the front-end devices may cause a slow shift in the sampling baseline; in the radar receiver link, analog gain fluctuations or spurious interference may cause non-target fluctuations to be superimposed on the echo baseband sequence; in weak signal or multipath environments, the wireless communication receiver may also experience both random noise and short-term spike disturbances. If edge detection or event triggering judgment is directly performed based on this raw sequence, subsequent threshold judgments may be pulled by the slowly changing baseline, or local high-frequency spikes may be mistaken for target events. Therefore, before entering morphological analysis, the system first performs front-end cleaning processing on the raw baseband data sequence to reduce the impact of non-target background components on subsequent event detection.

[0038] Specifically, the system can perform baseline estimation, bias correction, and noise filtering on the raw baseband data sequence using a digital signal processor, programmable logic device, microcontroller, or edge processing unit. This makes the output preprocessed baseband data sequence more suitable for subsequent structured feature extraction in terms of overall baseline stability and noise level. In one example, the IoT edge node can complete this preprocessing locally and only send segments that are subsequently confirmed to contain valid events to the protocol parsing or caching module, without having to continuously process the complete raw baseband stream.

[0039] In step S120, based on the signal-to-noise ratio of the preprocessed baseband data sequence and the expected event width corresponding to the target service scenario, morphological structural elements are determined, and morphological opening and top-hat transformation are performed on the preprocessed baseband data sequence using the morphological structural elements to obtain a top-hat feature sequence for characterizing the contour of local abrupt changes in the waveform.

[0040] Here, the signal-to-noise ratio (SNR) characterizes the distinguishability of the target signal relative to background noise within the current detection window, while the expected event width characterizes the typical duration of the digital event to be detected in the target service scenario on the discrete time axis. For example, in a wireless communication reception scenario, the target digital event can be a preamble, synchronization header, training sequence, or burst control signaling, and its expected event width can be determined by the number of symbols and sampling rate preset in the protocol. In a radar baseband analysis scenario, the target digital event can be abrupt changes in the echo envelope, a short-time response segment formed when the target enters the detection area, or edge changes corresponding to a specific modulation structure. In an IoT edge sensing scenario, the target digital event can be a wake-up signal received by a sensor node, a synchronization flag, or a short frame start segment. The system integrates the above two types of information to determine the morphological structural elements, ensuring that the scale of the structural elements matches the current detection environment and the time scale of the target event.

[0041] After determining the morphological structural element, the system uses this element to perform a morphological opening operation on the preprocessed baseband data sequence to obtain an opening result that reflects the local background contour. Subsequently, a top-hat transform is performed based on the difference between the preprocessed baseband data sequence and the opening result to extract waveform local abrupt changes that stand out relative to the local background contour. The resulting top-hat feature sequence can highlight short-term spikes, local transitions, or event edge contours, so that subsequent gradient calculations no longer rely solely on the original amplitude changes but are based on the morphologically enhanced local structural features. For example, when the absolute amplitude of the communication synchronization header is not significantly higher than the background noise, but its edge contour has a significant difference from the local background, the top-hat feature sequence can separate this type of local abrupt change from the slowly changing baseline background; when radar echo segments have short-term edge changes, the top-hat transform can also enhance the local difference of the edge relative to the surrounding background.

[0042] Therefore, the system can ensure that the morphological structural elements match the current detection environment and the time scale of the target event on an overall scale, thereby suppressing local non-target perturbations while preserving the local mutation contour of the target event as much as possible. In this way, the top-hat feature sequence can provide a clearer expression of local mutations for subsequent morphological gradient calculations, transforming the judgment basis of "whether there is a local business event" in the baseband sequence into a structured feature basis of "whether there is a morphological contour mutation that matches the scale of the target event," which is beneficial for subsequent candidate interval extraction.

[0043] In step S130, dilation and erosion operations are performed on the top-hat feature sequence using morphological structuring elements, and a morphological gradient sequence is obtained based on the difference between the dilation and erosion results. A dual threshold is generated based on the statistical characteristics and noise estimation results of the morphological gradient sequence, and candidate event intervals are extracted from the morphological gradient sequence using the dual threshold.

[0044] It should be noted that dilation can highlight strong responses in a local neighborhood, while erosion can reflect weaker responses in the same neighborhood. The difference between the two can characterize the intensity of local changes near the corresponding positions. The resulting morphological gradient sequence is used to reflect the significance of waveform abrupt changes in the top-hat feature sequence. For example, at the start of the synchronization header, the start boundary of the burst frame, or the edge of a radar echo abrupt change, local structural changes in the top-hat feature sequence usually produce a strong response in the morphological gradient sequence; while for slowly fluctuating backgrounds or isolated amplitude perturbations, their temporal continuity and structural consistency in the morphological gradient sequence are usually weaker.

[0045] Subsequently, the system generates dual thresholds based on the statistical characteristics of the morphological gradient sequence and the noise estimation results, and uses these dual thresholds to extract candidate event intervals from the morphological gradient sequence. Statistical characteristics can include the mean, dispersion, or local fluctuation level of the gradient sequence within the current detection window, while the noise estimation results reflect the impact of current background noise on the gradient response. The system can use the dual thresholds to perform tiered filtering of strong response locations and adjacent weak response locations in the morphological gradient sequence, enabling significant edge responses to serve as the core basis for candidate intervals, and reasonably preserving temporally continuous or adjacent weak response portions, thereby reducing the segmentation of true event intervals by local noise or fading.

[0046] In one example of wireless communication, the synchronization header or preamble may not have strong edge responses at all sampling points under weak signal conditions. If only a high threshold is used, it is easy to capture only a few edge peaks and miss the weak response parts in the same event; if only a low threshold is used, a large number of noise segments may be introduced. Therefore, a dual-threshold approach combined with temporal proximity can retain the weak response parts associated with significant edge responses, thus forming a more complete candidate event interval. In radar or IoT edge sensing scenarios, this candidate event interval can correspond to a local time domain range that may contain target echo abrupt changes, wake-up signals, or frame start segments.

[0047] Therefore, the system aggregates edge sampling points that meet the conditions according to the sampling time sequence to form candidate event intervals. These candidate event intervals represent the time range in which target digital events may exist in the baseband data sequence, but they have not yet been directly identified as final valid events. This allows the system to exclude a large number of non-event segments in the continuous baseband data stream and retain only segments with significant local mutations and temporal continuity for subsequent template matching verification, thereby reducing the computational burden of the subsequent identification process and improving the targeting of event screening.

[0048] In step S140, a baseline event feature template corresponding to the target business scenario is obtained, candidate event fragments and corresponding multi-dimensional candidate event features are extracted from the candidate event interval, and the multi-dimensional candidate event features are matched and verified with the baseline event feature template to determine the target valid event and the valid event interval in which it is located.

[0049] Here, the benchmark event feature template is used to characterize the typical waveform structure, duration characteristics, or edge change patterns of valid digital events in the target service scenario. For example, in wireless communication scenarios, the benchmark event feature template can correspond to the structural features of protocol preambles, synchronization headers, training sequences, or specific burst control signals; in narrowband IoT or low-power sensor networks, the benchmark event feature template can correspond to short-time synchronization signals used to wake up the receiving link; in radar or sensing scenarios, the benchmark event feature template can correspond to target echo edges, specific response segments, or preset abrupt waveforms. The benchmark event feature template enables the system to further introduce service scenario constraints on top of the initial screening at the physical edge, avoiding the direct determination of ordinary noise abrupt changes or non-target interference segments as valid events.

[0050] In some implementations, the system extracts candidate event segments from the preprocessed baseband data sequence or related feature sequence based on candidate event intervals, and extracts multidimensional candidate event features corresponding to these segments. These multidimensional candidate event features can reflect information such as the duration, edge intensity, morphological contour, energy distribution, and similarity to the template of the candidate event segment. Subsequently, the system matches and verifies the multidimensional candidate event features with the baseline event feature template. When the matching result meets preset judgment conditions, the corresponding candidate event segment is determined as the target valid event, and its corresponding candidate event interval is determined as the valid event interval.

[0051] For example, if a candidate event interval exhibits a strong edge response in the morphological gradient sequence, but its duration, internal contour, or multidimensional features are inconsistent with the protocol synchronization header template, then this candidate event interval can be excluded. Conversely, if a candidate event segment matches the baseline event feature template in terms of edge changes, duration, and overall morphology, it can be confirmed as a valid target event. Thus, the system can further confirm the target event after the initial screening of candidate events, which helps distinguish between real business events and occasional noise, co-frequency interference, or non-target mutations with similar edge responses. This improves the matching degree between the trigger output and the target business scenario and reduces the processing of invalid event segments in subsequent links.

[0052] In step S150, valid event sampling segments corresponding to valid event intervals are extracted from the preprocessed baseband data sequence. A local event energy sequence is generated based on the amplitude characterization values ​​of the valid event sampling segments. Then, based on the local event energy sequence and event location constraint information, the time center point, time start boundary, and time end boundary of the target valid event are determined. The event location constraint information is generated based on the event duration corresponding to the baseline event feature template and the interval boundaries of the valid event interval.

[0053] It should be noted that, since the valid event interval has already been verified by template matching, the system can further perform time position refinement processing within this interval. Specifically, the system extracts valid event sampling segments corresponding to the valid event interval from the preprocessed baseband data sequence, and constructs a local event energy sequence based on the amplitude representation values ​​of each sampling point in the valid event sampling segment. The amplitude representation values ​​can be used to characterize the signal strength of the corresponding sampling point, and their specific form can be selected according to the data type of the original baseband data sequence. For example, it can use a representation quantity related to the sampling amplitude, envelope, or complex baseband strength. This local event energy sequence is used to describe the energy distribution state of the target valid event on the time axis.

[0054] After obtaining the local event energy sequence, the system can determine the time center point of the target valid event based on the energy distribution. Compared to using only the position where the threshold is first crossed as the trigger time, determining the time center point based on the local event energy sequence can better reflect the overall structure of the valid event segment, thereby reducing the impact of local instantaneous noise on time positioning. For example, when there are local high-frequency jitters or occasional spikes in the initial segment of the synchronization header, the trigger position of a single point may be premature or delayed, while the center position determined based on the overall energy distribution of the event segment can provide a more stable and representative time position.

[0055] Furthermore, the system combines event location constraint information to determine the start and end time boundaries, ensuring that the final boundaries conform to the event duration reflected by the baseline event feature template while remaining within the time range defined by the valid event interval. For example, in a communication preamble scenario, the event duration corresponding to the template can constrain the truncation length of the event segment; in a radar echo segment scenario, the interval boundary of the valid event interval can prevent the location result from crossing into adjacent non-target echo areas. Thus, the system can further transform the aforementioned candidate interval-level event judgment into a valid event location result with a clear time center and boundary position. This location result can be used to generate event timestamps and accurately truncate event segment data, thereby improving the time consistency of subsequent synchronization processing, protocol parsing, or signal analysis.

[0056] In step S160, trigger information containing event timestamps and event fragment data is generated and output based on the time center point, time start boundary, and time end boundary.

[0057] Specifically, the time center point can be used to represent the representative trigger position of the target valid event on the timeline, and the time start boundary and time end boundary can be used to determine the truncation range of event segment data. The system can generate an event timestamp based on the time center point, and truncate event segment data from the corresponding baseband data sequence or cached data based on the time start boundary and time end boundary.

[0058] The output trigger information can be sent to subsequent buffer control modules, synchronization processing modules, demodulation decision modules, protocol parsing modules, or upper-layer service processing modules. For example, in a wireless communication receiver, the trigger information can be used to notify the synchronization header parsing unit or demodulation link to start processing valid frame segments from the corresponding time position; in a low-power IoT node, the trigger information can be used to wake up the back-end processor or enable a higher-power protocol parsing link; in a radar signal processing scenario, the trigger information can be used to send target echo segments into subsequent parameter estimation or target recognition processes. Exemplarily, the trigger information may include an event timestamp, event segment data, and additional information related to the event boundaries, enabling subsequent processing links to perform targeted processing based on a determined event time position, without having to perform an indiscriminate search of the continuous raw bitstream.

[0059] It should be understood that since the trigger information is generated after preprocessing, morphological contour enhancement, gradient filtering, template verification, and energy localization, it helps reduce the consumption of subsequent processing resources by invalid fragments and improves the stability and business effectiveness of digital event triggering results, enabling the system to complete the conversion from raw baseband data stream to structured trigger output. In particular, in edge computing or low-power reception scenarios, the system can perform subsequent high-overhead processing only on confirmed target valid events, thereby reducing invalid wake-ups, redundant buffering, and meaningless data uploads.

[0060] Regarding the implementation details of the data preprocessing operation in step S110, in some examples of embodiments of this application, firstly, for the current discrete sampling point in the original baseband data sequence, based on a sliding analysis window ending at the current discrete sampling point, short-time local energy features of the current detection interval are extracted, and the noise spectrum distribution of the current detection interval is estimated. Since the digital event triggering process usually needs to progress in real time with the sampling data, using a sliding analysis window ending at the current discrete sampling point allows the current detection interval to depend only on the current sampling point and its historical sampling points, thereby meeting the causality requirements in real-time processing.

[0061] In one example, short-time local energy characteristics can be obtained through a weighted energy statistics method with a window function, specifically expressed as:

[0062] Equation (1)

[0063] In the formula, The current sampling time The corresponding short-time local energy characteristic quantity, The length of the sliding analysis window. For energy-weighted window functions, These are the original sampled values ​​within the sliding analysis window. The energy-weighted window function can be a rectangular window, a Hamming window, or other window functions that can adjust the contribution of different sampled points within the window. For complex I / Q baseband sampled values, It can represent the square of the modulus at the corresponding sampling point; for real baseband sample values, It can represent the square of the amplitude at the corresponding sampling point.

[0064] Meanwhile, the system can estimate the noise spectrum distribution of the current detection interval based on the frequency domain statistics, background data of the target event segment, or the recursive spectrum estimation results within the current detection interval. This noise spectrum distribution is used to characterize the distribution of noise energy with frequency within the current detection interval, and can serve as the basis for configuring noise filtering parameters, enabling the filtering process to suppress the main noise components in the current detection environment.

[0065] Then, an online update mechanism based on exponential decay is used to track and estimate the baseline offset component of the original baseband data sequence. The smoothing update weight factor of the online update mechanism is adjusted based on the short-time local energy characteristics, so that the smoothing update weight factor and the short-time local energy characteristics are negatively correlated, so as to reduce the degree of baseline estimation tracking of burst waveforms when the signal energy increases.

[0066] Specifically, the smoothing update weight factor is negatively correlated with the short-term local energy feature, meaning that when the signal energy within the current detection interval increases, the baseline bias estimation's tracking accuracy of the current sampled value decreases accordingly. This reduces the possibility of sudden target events being mistakenly included in the baseline estimation process.

[0067] In one example, the smooth update weighting factor can be determined based on the exponential decay relationship of short-time local energy features:

[0068] Equation (2)

[0069] In the formula, The current sampling time The corresponding smooth update weight factor, The preset base smooth update step size, This is a coefficient used to control the attenuation sensitivity. To give the exponential term a clear numerical meaning, the short-time local energy characteristic can be a normalized energy value; or, the coefficient... It can be configured according to the dimensions of short-time local energy characteristics, so that It is a dimensionless quantity.

[0070] Based on this, the current sampling time The corresponding current baseline bias component can be determined using an exponential moving average:

[0071] Equation (3)

[0072] In the formula, For the original baseband data sequence at the current sampling time The sampled values, The current sampling time The estimated current baseline bias components, The previous sampling time The baseline bias component is estimated.

[0073] As can be seen from the above update method, when the short-time local energy feature is low, the smooth update weight factor remains at a relatively high level, and the baseline bias component can be smoothly updated with the slowly changing background. When the short-time local energy feature increases due to sudden digital events, the smooth update weight factor decreases, so that the current baseline bias component maintains the historical baseline state more, reducing the tracking of sudden event waveforms. In this way, while suppressing low-frequency baseline drift, the risk of the effective event edges and envelope being weakened by the baseline estimation process can be reduced.

[0074] Then, based on the difference between the sampled value at the current sampling time and the corresponding baseline bias component, a current debiased sampled value is generated, and a debiased sequence is formed from multiple current debiased sampled values ​​according to the sampling time sequence. In one example, the current debiased sampled value can be represented as:

[0075] Equation (4)

[0076] In the formula, The current sampling time The corresponding current debiased sampled value, The current sampling time The sampled values, The current sampling time The current baseline bias component is estimated. By debiasing, the baseline walk caused by front-end device bias, temperature drift, or slowly varying channel conditions can be reduced, making the resulting debiased sequence more concentrated in reflecting short-term changing components.

[0077] Furthermore, based on the noise spectrum distribution and the preset group delay constraint, the filter tap order and filter coefficients of the digital finite impulse response filter are configured, and the filter coefficients are designed and optimized using the minimum phase criterion. The configured digital finite impulse response filter is then used to perform noise filtering on the debiased sequence to obtain the preprocessed baseband data sequence.

[0078] For example, the transfer function of a digital finite impulse response filter can be expressed as:

[0079] Equation (5)

[0080] In the formula, For transfer functions, For time delay operators, This represents the order of the filter taps. For filter tap index, For the first Each tap corresponds to a filter coefficient.

[0081] In some implementations, the system can determine the frequency range requiring focused suppression based on the estimated noise spectrum distribution, and limit the filter order and coefficient configuration according to a preset group delay constraint. This ensures that the filter meets noise suppression requirements while avoiding the introduction of excessive processing delay. When optimizing the filter coefficients using the minimum phase criterion, the impulse response energy of the filter can be concentrated at earlier time points, thereby reducing the delay accumulation caused by the filtering stage while maintaining the target amplitude-frequency response. Thus, the resulting preprocessed baseband data sequence can balance noise suppression, baseline stability, and real-time processing requirements.

[0082] Through the above processing, the original baseband data sequence can be sufficiently pre-processed. On the one hand, the short-time energy-driven smooth update weight adjustment method can reduce the possibility that the waveform of sudden events is weakened by the baseline estimation process; on the other hand, the minimum phase digital filtering processing based on the noise spectrum distribution and group delay constraint configuration can control the filtering delay while suppressing high-frequency noise. As a result, the preprocessed baseband data sequence has a more stable baseline, lower high-frequency noise interference, and better time sequence preservation capability, making it suitable as input data for baseband digital event detection.

[0083] Figure 2 A flowchart illustrating an example of determining morphological structural elements in a method according to an embodiment of this application is shown.

[0084] like Figure 2 As shown, in step S210, the average power of the preprocessed baseband data sequence and the variance of the background noise under the current detection window are estimated respectively, and the ratio between the average power of the signal and the variance of the background noise is calculated to obtain the linear signal-to-noise ratio parameter, which is used as the corresponding signal-to-noise ratio.

[0085] In some implementations, the current detection window can be a time-domain window that slides with the sampling time. The system can estimate the average signal power based on the mean square statistics of the sampling points within the window, and estimate the background noise variance based on the sampling statistics of silent intervals, intervals without target events, or noise-dominated intervals. For scenarios where silent intervals cannot be clearly distinguished, the background noise variance can also be estimated based on the low-energy sampling segments or background spectrum statistics within the current window.

[0086] In some examples, the linear signal-to-noise ratio (SNR) parameter can be expressed as the ratio between the average signal power and the variance of the background noise. It should be noted that this embodiment uses a linear SNR parameter instead of a logarithmic SNR value so that it can be directly used as an input variable for the subsequent natural exponential decay mapping model. To reduce the impact of instantaneous sampling fluctuations on the SNR estimation results, the system can also smooth the linear SNR parameter obtained from multiple consecutive detection windows, making the parameter more stably characterize the signal discriminability in the current detection environment. This allows the system to obtain a channel state representation quantity used to determine the scale of morphological structural elements.

[0087] In step S220, based on the standard physical duration of the valid event to be detected in the target business scenario and the sampling frequency of the underlying analog-to-digital converter, the number of basic event sampling points mapped to the discrete time axis is calculated, and the number of basic event sampling points is determined as the expected event width.

[0088] Specifically, the standard physical duration can be derived from the target service protocol, system configuration parameters, or prior physical characteristics of the target event. For example, in wireless communication scenarios, the standard physical duration can correspond to the duration of a preamble, synchronization header, or training sequence; in radar baseband analysis scenarios, the standard physical duration can correspond to the duration of a target echo abrupt change segment or a preset response segment; and in IoT edge sensing scenarios, the standard physical duration can correspond to the duration of a wake-up signal or a short frame start signal.

[0089] In some examples, the system can map the duration of a target event in the continuous time domain to the width of a sampling point on a discrete time axis based on the correspondence between standard physical duration and sampling frequency. This mapping result can be rounded to obtain the desired event width suitable for discrete sequence processing. Thus, the scale of the morphological structuring element is no longer determined solely by the empirical window length, but corresponds to the actual time scale of the target event in the digital sampling sequence, enabling the structuring element to cover the main local contours of the target event on a scale.

[0090] In step S230, a natural exponential decay mapping model is constructed with the expected event width as the baseline scale and the linear signal-to-noise ratio parameter as the independent variable, and the adaptive structuring element length is calculated based on the natural exponential decay mapping model.

[0091] Here, the natural exponential decay mapping model is configured to include a scale compensation term that increases exponentially with decreasing linear signal-to-noise ratio (SNR) parameter, such that the length of the adaptive structuring element increases with decreasing SNR parameter and shrinks with increasing SNR parameter to approach the baseline structuring element length determined by the desired event width and a preset scale adjustment coefficient.

[0092] For example, adaptive structuring element length It can be calculated using the following formula:

[0093] Equation (6)

[0094] In the formula, The calculated adaptive structuring element length, The preset scaling factor. Let SNR be the desired event width, and SNR be the linear signal-to-noise ratio parameter. This indicates the rounding up operation. It is a natural constant. The preset scale gain coefficient, This is a preset channel sensitivity constant.

[0095] In the above formula (6), The scale gain coefficient is used to form the reference structuring element length related to the sampling width of the target event. Used to adjust the expansion of the structuring element length under low signal-to-noise ratio conditions, channel sensitivity constant This is used to adjust the convergence speed as the length of the structuring element changes with the signal-to-noise ratio (SNR). Since SNR uses a linear SNR parameter, The adaptive structuring element length decreases as the signal-to-noise ratio (SNR) increases, gradually approaching the baseline structuring element length determined by the desired event width and a preset scale adjustment coefficient. When the SNR is low, the scale compensation term is relatively large, causing the adaptive structuring element length to increase moderately relative to the baseline structuring element length. By rounding up, the resulting length is guaranteed to be an integer sampling point length suitable for discrete sequence morphological operations.

[0096] In step S240, based on the calculated adaptive structural element length, a one-dimensional symmetrical flat line segment structure is constructed, consisting of a continuous discrete numerical sequence with equal element values, and the one-dimensional symmetrical flat line segment structure is determined as the morphological structural element.

[0097] Here, the one-dimensional symmetric flat line segment structure can be understood as a flat structural element with a specified length on the discrete time axis, where all elements have the same value. It is suitable for erosion, dilation, opening, and top-hat transformation operations on one-dimensional baseband sequences. Due to its relatively simple structure, this structural element can be applied to digital signal processors, programmable logic devices, or edge processing units with low computational complexity.

[0098] Through the embodiments of this application, the scale of the morphological structuring element can be determined based on the linear signal-to-noise ratio (SNR) parameter within the current detection window and the expected event width of the target service event. When the SNR is low, a moderately increased structuring element length is beneficial for enhancing the smoothing ability against random spikes and local non-target disturbances; when the SNR is high, the structuring element length approaches a reference scale related to the target event width, which helps reduce the weakening of the local abrupt contour of the target event. Therefore, the determined morphological structuring element achieves a good balance between noise suppression and edge contour preservation, enabling the top-hat transform to obtain a more stable expression of local abrupt feature characteristics.

[0099] Regarding the implementation details of determining the top-cap feature sequence for characterizing the local abrupt change contour of the waveform in step S120, in some examples of embodiments of this application, firstly, one-dimensional morphological erosion and one-dimensional morphological dilation operations are sequentially performed on the preprocessed baseband data sequence using morphological structuring elements to generate a morphological opening operation sequence, and the difference between the sequence value of the preprocessed baseband data sequence and the sequence value of the morphological opening operation sequence is calculated point by point to obtain a positive top-cap feature sequence for characterizing the local rising abrupt change feature.

[0100] Specifically, morphological opening operations can be used to weaken local positive bulges in preprocessed baseband data sequences that are relatively short relative to the structuring element scale, and to obtain opening operation results that reflect the local background contours. In scenarios such as synchronization headers, preambles, or burst control signaling in wireless communication, and short-duration rising edges in radar echoes, the arrival of a target event may manifest as an increase in amplitude or envelope relative to the local background. By performing point-by-point difference calculations between the preprocessed baseband data sequence and the opening operation sequence, the rising abrupt components that are more prominent relative to the local background can be extracted, thus obtaining a positive top-hat feature sequence.

[0101] For example, the positive top-hat feature sequence is represented as:

[0102] Equation (7)

[0103] In the formula, For discrete location indexes, To preprocess the baseband data sequence, Indexing the positive top-hat feature sequence at discrete positions eigenvalues ​​at that location Indexing of baseband data sequences at discrete locations The sequence value at that location, Represents the morphological opening operator. These are morphological structural elements. Through the above processing, the local upward variations in the preprocessed baseband data sequence that stand out relative to the background contour of the opening operation can be mapped to a positive top-hat feature response.

[0104] Then, morphological structural elements are used to sequentially perform one-dimensional morphological dilation and one-dimensional morphological erosion operations on the preprocessed baseband data sequence to generate a morphological closing operation sequence. The difference between the sequence value of the morphological closing operation sequence and the sequence value of the preprocessed baseband data sequence is calculated point by point to obtain a negative bottom cap feature sequence used to characterize the local descent mutation features.

[0105] Specifically, morphological closing operations can be used to fill in local negative depressions in preprocessed baseband data sequences that are relatively short relative to the structuring element scale, and to obtain closing operation results that reflect the local envelope or compensation contour. In scenarios such as bipolar modulation, deep channel fading, echo envelope descent edges, or abrupt sign flips, target events may not only manifest as positive spikes, but also as local drops, depressions, or falling edges. By calculating the pointwise differences between the closing operation sequence and the preprocessed baseband data sequence, local descent abrupt changes can be extracted in the form of positive feature responses, thereby reducing the problem that a single positive top-hat transform is insufficient in expressing the features of negative abrupt changes.

[0106] For example, the negative bottom-hat feature sequence is represented as:

[0107] Equation (8)

[0108] In the formula, Indexing the negative bottom cap feature sequence at discrete positions eigenvalues ​​at that location This represents the morphological closing operator. Through the above processing, the local descent changes in the preprocessed baseband data sequence relative to the closing operation contour can be converted into a negative bottom-cap feature response.

[0109] Furthermore, a bidirectional feature fusion process is performed on the feature values ​​at the same discrete index of the positive top-hat feature sequence and the negative bottom-hat feature sequence to obtain a morphological contour feature sequence that incorporates both positive and negative top-hat features. This morphological contour feature sequence is then identified as the top-hat feature sequence. The bidirectional feature fusion process includes either point-by-point weighted summation or point-by-point maximum maximization. This fusion process allows for the simultaneous expression of local ascending and descending abrupt changes within the same feature sequence, making the top-hat feature sequence suitable for baseband digital event detection with different polarities or modulation schemes.

[0110] In one example, bidirectional feature fusion processing can employ point-by-point maximization, which can be represented as:

[0111] Equation (9)

[0112] In the formula, Indexing the fused morphological contour feature sequence at discrete locations eigenvalues ​​at that location This indicates a maximum value operation. This processing method can retain the stronger of the positive top-hat and negative bottom-hat features at each discrete location, and is suitable for scenarios where it is desirable to highlight the most significant local abrupt changes in the contour.

[0113] In another example, bidirectional feature fusion can be performed using point-by-point weighted summation, which can be expressed as:

[0114] Equation (10)

[0115] In the formula, and These are the preset positive top cap weighting coefficients and negative bottom cap weighting coefficients, respectively. By adjusting... and The system can set different feature contribution weights based on the degree of attention paid to rising or falling abrupt changes in the target business scenario. For example, when the target event is more likely to be represented by a rising edge, the weight of the positive top-hat feature can be increased; when the target event may be represented by a falling edge or a bipolar jump, the weight of the negative bottom-hat feature can be increased, or the two types of features can be made to have similar contribution weights.

[0116] Through the embodiments of this application, both local ascending and descending abrupt changes can be considered in the same morphological contour feature sequence. Compared with the processing method that only extracts abrupt change features in a single direction, this method can more completely express the local waveform contours in scenarios such as bipolar modulation, envelope concavity, and sudden signal edge changes, so that the generated top-hat feature sequence has better polarity adaptability and local structure expression ability.

[0117] Figure 3 A flowchart illustrating an example of obtaining a morphological gradient sequence and extracting candidate event intervals from it according to an embodiment of this application is shown.

[0118] like Figure 3 As shown, in step S310, one-dimensional morphological dilation and one-dimensional morphological erosion operations are performed on the top cap feature sequence using morphological structuring elements, and a morphological gradient sequence is generated based on the difference between the feature values ​​of the dilation operation and the feature values ​​of the erosion operation under the same discrete position index.

[0119] Specifically, the top-cap feature sequence obtained after the aforementioned processing is already able to highlight the local abrupt change contours in the preprocessed baseband data sequence. To further quantify the intensity of these local abrupt change contours along the time axis, the system performs morphological gradient calculations on the top-cap feature sequence. Specifically, morphological dilation is used to obtain larger local responses within the neighborhood of the structuring element, while morphological erosion is used to obtain smaller local responses within the neighborhood of the structuring element. The difference between these two operations at the same discrete index position reflects the local span of the feature values ​​near that position, thus characterizing the significance of local waveform changes.

[0120] For example, a morphological gradient sequence is represented as:

[0121] Equation (11)

[0122] In the formula, For top-hat feature sequences, Indexing morphological gradient sequences at discrete positions gradient value at, Indexing the top-hat feature sequence at discrete positions eigenvalues ​​at that location Represents the morphological dilation operator. Represents the morphological erosion operator. It is a morphological structural element.

[0123] Equation (11) expresses a nonlinear gradient measure based on the difference between locally large and locally small responses. Through this processing, rising edges, falling edges, or abrupt change profiles with a certain sustained width in the top-hat feature sequence can be converted into a sequence representation with a more concentrated gradient response. Thus, the system can observe the strength changes of local abrupt changes within the gradient domain, providing a clearer basis for edge strength in the selection of candidate event intervals.

[0124] In step S320, the current high edge threshold and the current low edge threshold are generated based on the gradient mean, gradient standard deviation, and background noise estimation parameters of the morphological gradient sequence within the current detection window. Here, both the current high edge threshold and the current low edge threshold include a noise bias compensation term, and the gradient fluctuation scaling degree corresponding to the current low edge threshold is smaller than that corresponding to the current high edge threshold.

[0125] In some examples, the current high edge threshold and the current low edge threshold can be represented as follows:

[0126] Equation (12)

[0127] Equation (13)

[0128] In the formula, TH is the current high edge threshold, and TL is the current low edge threshold. The gradient mean, For the gradient standard deviation, To estimate parameters for background noise, The first scaling factor is... This is the second scaling factor. The preset noise bias weighting constant is used, and .

[0129] The threshold calculation method described above considers both the statistical fluctuations of the gradient sequence itself and the current background noise level. Specifically, the gradient mean provides a basic threshold reference, the gradient standard deviation is used to adjust the threshold width according to the degree of gradient fluctuation, and the noise bias term corresponding to the background noise estimation parameter is used to raise the threshold benchmark when the noise level increases, reducing the possibility of weak noise fluctuations being misjudged as valid edges. Since the second scaling factor is smaller than the first scaling factor, the current low edge threshold is lower than the current high edge threshold, allowing them to be used respectively for the identification of strong edge responses and the preservation of neighboring weak edge responses.

[0130] In step S330, the main edge sampling point is identified from the morphological gradient sequence based on the current high edge threshold, and the associated edge sampling point is identified from the neighborhood of the main edge sampling point based on the current low edge threshold.

[0131] In some implementations, the system can traverse the morphological gradient sequence and mark discrete sampling points with gradient values ​​greater than the current high edge threshold as primary edge sampling points. For discrete sampling points with gradient values ​​greater than the current low edge threshold but not greater than the current high edge threshold, the system further determines the time distance between the discrete sampling point and the identified primary edge sampling points; when the time distance meets a preset connection step size threshold, the discrete sampling point is marked as an associated edge sampling point.

[0132] For example, the discrete location index is Whether a sampling point belongs to an associated edge sampling point can be determined by the following constraints:

[0133] and Equation (14)

[0134] In the formula, This is the set of indices for the identified main edge sampling points. The location index of the main edge sampling point This is the preset connection step size threshold.

[0135] Specifically, the above constraints mean that if the gradient value of the sampling point to be judged is between the current low edge threshold and the current high edge threshold, and the time interval between the sampling point and at least one main edge sampling point does not exceed a preset connection step size threshold, then it can be determined as an edge sampling point associated with the main edge sampling point. Through this processing, isolated weak gradient responses far from the main edge sampling points are less likely to be included in the event interval, while weak response portions that are temporally adjacent to significant edge responses can be preserved, thereby reducing the possibility of the same event edge being segmented due to local fading or local perturbations.

[0136] In step S340, the main edge sampling points and associated edge sampling points are continuously aggregated according to the sampling time sequence to form candidate event intervals.

[0137] Specifically, the system can sequentially scan all marked sampling points along the time axis. When the time interval between adjacent marked sampling points is not greater than a preset connection step size threshold, they are grouped into the same candidate event interval. When the time interval between adjacent marked sampling points is greater than the preset connection step size threshold, the current candidate event interval ends, and a new candidate event interval is formed based on subsequent marked sampling points. Each candidate event interval can be represented by a corresponding start sampling index and end sampling index.

[0138] This embodiment enables the further conversion of local mutation contours in the top-hat feature sequence into morphological gradient responses, and the extraction of candidate event intervals through dual-threshold screening and temporal continuity constraints. The current high edge threshold is beneficial for screening edge positions with high mutation intensity, while the current low edge threshold is beneficial for retaining weak edge responses adjacent to strong edge positions. A preset connection step size threshold is used to limit the temporal correlation range between edge sampling points. Therefore, the obtained candidate event intervals can better reflect the continuous edge range of the same digital event on the time axis and reduce the influence of isolated noise points on the candidate interval extraction results.

[0139] Regarding the implementation details of obtaining the benchmark event feature template and determining the target valid event and its valid event interval in step S140, in some examples of the embodiments of this application, firstly, a set of historical benchmark event waveforms preset for the target business scenario is obtained, the historical benchmark event waveforms in the set of historical benchmark event waveforms are processed by duration standardization and amplitude normalization, and the dominant feature basis vector is extracted by principal component analysis to construct the benchmark event feature template.

[0140] Specifically, in scenarios such as preamble detection in wireless communication, synchronization header identification, radar specific pulse identification, or IoT short frame start signal detection, the same type of target digital event may exhibit slight duration shifts, amplitude fading, or local waveform perturbations at different sampling times or under different channel conditions. To ensure that historical reference event waveforms can enter a unified matching space, the system can first perform duration standardization processing on each historical reference event waveform, for example, by mapping them to the same sampling point length through interpolation, resampling, or sampling; simultaneously, amplitude normalization processing is performed on each historical reference event waveform to make samples with different amplitude levels comparable.

[0141] After standardization, the system can use principal component analysis to reduce the dimensionality of the historical benchmark event waveform set, extracting dominant feature basis vectors that reflect the main waveform structure of the target event. These dominant feature basis vectors can describe the relatively stable morphological distribution in the historical benchmark event waveform set, such as the main edge variations, envelope variations, or temporal structural features of event segments. Therefore, the constructed benchmark event feature template can express the main structural features of the target business event in a lower dimension, reducing the impact of random disturbances and redundant details on the matching calculation.

[0142] Then, candidate event segments are extracted from the preprocessed baseband data sequence according to the candidate event interval. The candidate event segments are then subjected to duration standardization and amplitude normalization. Multidimensional candidate event features are generated based on the processed candidate event segments.

[0143] Here, the standardization method for candidate event segments is consistent with that for the historical baseline event waveform set, ensuring that the candidate event segments and the baseline event feature templates are on the same duration and amplitude scales. Therefore, subsequent matching calculations can better reflect the structural similarity between the candidate event segments and the target event template, rather than being primarily affected by differences in sampling length or instantaneous amplitude variations.

[0144] Subsequently, using a weighted window function configured based on the relative position of the temporal boundary and the background noise estimation parameters, weighted constraint correlation calculations are performed on multiple dominant feature basis vectors in the multidimensional candidate event features and the baseline event feature template. The maximum weighted constraint correlation among the multiple weighted constraint correlation results is determined as the matching confidence. Here, the weighted window function is used to enhance the matching contribution of features in the middle of the candidate event segment and suppress the feature contribution affected by noise at the boundary of the candidate event segment. Moreover, the larger the background noise estimation parameter, the higher the degree of suppression of feature contribution at the boundary of the candidate event segment by the weighted window function.

[0145] For example, weighted constraint relevance can be expressed as:

[0146] Equation (15)

[0147] In the formula, To calculate the weighted constrained correlation, A set of time-series indexes representing candidate event intervals after duration standardization. For multidimensional candidate event features, Indexing multidimensional candidate event features at discrete locations The eigencomponent values ​​at that location, Index of dominant feature basis vectors The first in the baseline event feature template The dominant feature basis vectors are indexed at discrete locations. The baseline value at that location It is a weighted window function.

[0148] In equation (15), the middle position of a candidate event segment usually reflects the stable structure of the target event better, while the parts of the candidate event segment near the start and end boundaries may be affected by adjacent noise segments, interval aggregation errors, or channel tails. Therefore, the weighted window function... The system can be configured to exhibit smooth decay at both ends and remain flat in the middle to reduce the impact of boundary location features on correlation calculations. When the background noise estimation parameters increase, the system can increase the decay of the weighted window function at both ends, making the matching calculation more focused on relatively stable temporal regions within the candidate event segments. Therefore, the weighted constraint correlation can simultaneously reflect the structural similarity between the candidate event segment and the dominant feature basis vector, as well as the credibility of different positions within the candidate event segment.

[0149] After obtaining the weighted constrained relevances corresponding to each dominant feature basis vector, the system determines the maximum weighted constrained relevance as the matching confidence score for the candidate event segment. Subsequently, if the matching confidence score meets the preset confidence judgment conditions and the overall time span of the candidate event interval conforms to the preset event duration range corresponding to the target business scenario, the candidate event segment is determined as a target valid event, and the candidate event interval corresponding to the target valid event is determined as a valid event interval. Through this determination method, the system not only examines the structural similarity between the candidate event segment and the template, but also examines whether its duration conforms to the time scale requirements of the target business event, thereby reducing the possibility of interference segments with abnormal durations being mistakenly identified as target valid events.

[0150] Preferably, the system can also feed back the features of target valid events that have been continuously verified through matching to the feature space corresponding to the benchmark event feature template based on a weighted sliding time window configured with a forgetting update factor, so as to iteratively update the benchmark event feature template.

[0151] Here, for scenarios where underlying hardware parameters, temperature conditions, or channel environments change slowly, the baseline event feature template can absorb recently validated target event features while maintaining historically stable features.

[0152] For example, the following updates can be performed on the template feature center or matching template prototype corresponding to the target business event:

[0153] Equation (16)

[0154] In the formula, This is the index for the current iteration cycle. The index of the discrete position of the template feature center corresponding to the benchmark event feature template in the current iteration period. eigenvalues ​​at that location Indexing the updated template feature centers at discrete locations eigenvalues ​​at that location Indexing the multidimensional candidate event features at discrete locations corresponding to the currently valid target events that have passed matching verification. The eigencomponent values ​​at that location, The preset forgetting update factor, and In one example, It can be set to a value close to 1 to give historical template features a higher weight and newly verified event features a lower weight in the update, thereby smoothly filtering out instantaneous distortion interference of single atypical event waveforms while ensuring that the spatial manifold distribution of template features has high geometric stability.

[0155] In this way, through iterative updates to the template, the baseline event feature template can gradually adapt to the slowly changing waveform feature distribution during long-term operation. It should be noted that when the principal component analysis template uses multiple dominant feature basis vectors, the system can periodically update the dominant feature basis vectors or template weights based on the updated template feature centers or the recent set of effective event features, in order to maintain consistency between the baseline event feature template and the effective event waveforms in the current business scenario.

[0156] Through the embodiments of this application, standardization processing is used to eliminate scale differences between candidate event fragments and historical benchmark event waveforms, principal component analysis is used to extract the main structural features of the target event, weighted constraint correlation calculation is used to improve the stability of candidate fragment matching, and a forgetting update factor is used to smoothly update the template features. Therefore, the determination of the target valid event can reflect the structural similarity between the candidate fragment and the benchmark template, while also taking into account the influence of boundary noise and event duration constraints, making the confirmation result of the valid event interval more consistent with the event characteristics of the target business scenario.

[0157] Regarding the implementation details of determining the time center point, time start boundary, and time end boundary of the target valid event in step S150, in some examples of embodiments of this application, firstly, according to the time domain boundary of the valid event interval, the corresponding sampled data is extracted from the preprocessed baseband data sequence as a valid event sampling segment, and a local event energy sequence is constructed based on the amplitude characterization value of each discrete sampling point within the valid event sampling segment.

[0158] Specifically, the valid event interval determined through matching verification is used to characterize the interval position of the target valid event on the time axis. To obtain a more accurate event time position within this interval, the system can extract the sampled data corresponding to the valid event interval from the preprocessed baseband data sequence, instead of directly using the feature sequence after morphological operations for time localization. This preserves the amplitude variation and temporal structure of the target valid event on the original sampling time axis while reducing the influence of DC bias, low-frequency drift, and high-frequency noise.

[0159] In one example, the amplitude characterization value can be determined based on the amplitude, envelope value, or complex baseband modulus of each discrete sampling point within the valid event sampling segment. The system can further calculate the corresponding energy value based on the amplitude characterization value and arrange the energy values ​​according to the sampling time sequence to construct a one-dimensional local event energy sequence. For real baseband sequences, the energy value can be obtained by squared sampling amplitude; for complex I / Q baseband sequences, the energy value can be obtained by squared complex modulus; for sequences that have undergone envelope detection, a corresponding energy characterization can also be formed based on the envelope strength. The resulting local event energy sequence can reflect the energy distribution state of the target valid event within the valid event interval.

[0160] Subsequently, based on the energy distribution of the local event energy sequence in the time domain, the energy centroid position of the target valid event is determined, and the energy centroid position is determined as the time center point.

[0161] For example, using local event energy sequences As a weighting factor on the discrete time axis, the energy centroid index of the target valid events is calculated, and the energy centroid index is determined as the time center point. :

[0162] Equation (17)

[0163] In the formula, Center point of time For local event energy sequences, Indexing local event energy sequences at discrete locations The energy value at that location, and These are the start and end sampling indices corresponding to the valid event sampling segments, respectively. For discrete location indexes.

[0164] In equation (17), the time index is weighted and averaged according to the energy proportion of each discrete position within the valid event sampling segment to obtain the position reflecting the center of the local event energy distribution. Compared with the method of determining the trigger time based solely on a single threshold crossing point, the energy centroid position makes better use of the overall energy distribution within the valid event sampling segment, thus reducing the impact of individual sampling point spikes, local high-frequency disturbances, or edge jitter on the time center point. It should be noted that in actual implementation, when the energy sum of the local event energy sequence meets the preset validity conditions, the system performs the above energy centroid calculation; when the energy sum does not meet the preset validity conditions, the system can use the center position of the valid event interval or the reference center position corresponding to the baseline event feature template as an alternative positioning result.

[0165] Then, the system obtains the standard event sampling one-sided span corresponding to the currently matched benchmark event feature template from the event location constraint information, and obtains the current global alignment compensation parameter updated based on historical decoding feedback. The system shifts the time center point to the start and end directions of the time axis by the standard event sampling one-sided span, and determines the initial time start boundary and the initial time end boundary by combining the current global alignment compensation parameter.

[0166] Here, the standard event sampling span is used to characterize the sampling point span of the target valid event relative to the time center point in the start and end directions, which can be calculated from the event duration corresponding to the baseline event feature template. The current global alignment compensation parameter is used to characterize the overall time offset between the baseband triggered positioning result and the upper layer decoding feedback result.

[0167] For example, the initial time start boundary and the initial time end boundary It can be represented as:

[0168] Equation (18)

[0169] Equation (19)

[0170] In the formula, This is the initial time starting boundary. This is the initial time end boundary. For standard events, sample one-sided span values. This is the current global alignment compensation parameter.

[0171] In equations (18) and (19) above, the system is centered on the time center point. Using the reference position and combining the standard event sampling unilateral span, the initial start boundary and initial end boundary of the target valid event on the time axis are determined, and the current global alignment compensation parameter is used to perform overall translation correction on the two. Because It acts on both the initial time start boundary and the initial time end boundary, so it is mainly used to correct the overall alignment deviation without changing the event time length determined by the single-sided span of the standard event sampling.

[0172] Subsequently, based on the sampling index range of the valid event interval, boundary constraint processing is performed on the initial time start boundary and the initial time end boundary to obtain the time start boundary and time end boundary of the target valid event.

[0173] Specifically, the system can limit or truncate the initial time start boundary and initial time end boundary based on the start and end sampling indices of the valid event interval. This ensures that the final time start and end boundaries fall within the sampling index range corresponding to the valid event interval, or within the sampling index range obtained by expanding the valid event interval by a preset protection width. This avoids boundary overflow issues caused by span calculations or alignment compensation, ensuring that the event segment truncation range remains consistent with the confirmed valid event interval.

[0174] Furthermore, when the business protocol frame corresponding to the target valid event is successfully decoded by the upper-layer decoding unit, the current global alignment compensation parameter is smoothly updated based on the alignment time deviation between the absolute midpoint of the real event synchronization header and the time center point fed back by the upper-layer decoding unit, and the updated global alignment compensation parameter is used for the boundary positioning of subsequent digital events.

[0175] Here, the absolute midpoint of the real event synchronization header can be obtained by the upper-layer decoding unit after completing the protocol frame parsing and verification, and this position is converted to the same sampling index reference or time reference as the time center point.

[0176] For example, the global alignment compensation parameter can be updated using an exponential moving average:

[0177] Equation (20)

[0178] In the formula, This is the updated global alignment compensation parameter. For the current global alignment compensation parameter, The upper-layer decoding unit feeds back and converts the data to the absolute midpoint of the real event synchronization header under the same time base. This is the time center point obtained in this calculation. The alignment deviation update factor is a preset value, and satisfies the following conditions: .

[0179] During the update process of the above combined formula (20), The alignment deviation update factor, used to characterize the time deviation between the current event localization result and the upper-level resolution result, controls the impact of new deviation information on the global alignment compensation parameter. Smoothing updates via exponential moving averages reduce the impact of single resolution errors or occasional noise on the compensation parameter, enabling the global alignment compensation parameter to reflect a relatively stable overall time offset trend over long-term operation. This compensation parameter participates in the overall translation of the initial time boundary during boundary localization, which helps reduce the impact of hardware clock skew, sampling link delay variations, or temperature drift on the consistency of event boundary localization.

[0180] Through the embodiments of this application, the time center point can be determined based on the energy distribution of valid event sampling segments, and the time start boundary and time end boundary can be determined by combining the single-sided span of standard event sampling, the boundary of valid event interval, and the global alignment compensation parameter. Energy centroid positioning helps reduce the impact of local spikes or instantaneous jitter on the center position, boundary constraint processing helps maintain the legality and consistency of the event segment range, and the smooth update of the global alignment compensation parameter helps correct the overall time offset formed in long-term operation, thus providing relatively stable time center and boundary positioning results for the target valid event.

[0181] Figure 4 This paper illustrates the overall mechanism architecture of a digital event triggering method based on baseband data sequence edge detection provided in an embodiment of this application.

[0182] like Figure 4 As shown, the system first acquires the underlying baseband I / Q data and performs front-end preprocessing on the baseband I / Q data through a debiasing and noise filtering module to reduce the impact of DC bias, low-frequency baseline drift, and high-frequency noise on subsequent event detection, resulting in a preprocessed signal. Subsequently, the adaptive morphological preprocessing module determines the morphological structural elements based on the current signal state and the event scale of the target service scenario, and performs morphological opening and top-hat transformation on the preprocessed signal to extract top-hat feature sequences that can characterize the contours of local waveform changes. This transforms baseband event detection from simple amplitude judgment to the analysis of local morphological structural changes.

[0183] After completing the top-hat feature extraction, the system enters the candidate event screening and valid event confirmation stage. The morphological gradient calculation module performs dilation and erosion operations based on the top-hat feature sequence, and obtains the morphological gradient sequence based on the difference between the two to characterize the significance of local waveform abrupt changes. The dynamic dual-threshold generation module generates dual thresholds based on the statistical characteristics of the morphological gradient sequence and the noise estimation results, and feeds the dual thresholds back to the morphological gradient calculation and screening process to extract candidate event intervals from the morphological gradient sequence. Subsequently, the multi-dimensional feature template matching module performs multi-dimensional feature matching verification on the candidate event segments corresponding to the candidate event intervals according to the benchmark event feature template corresponding to the target business scenario to determine the target valid events and the valid event intervals they belong to. The feature template dynamic clustering update module shown in the figure is used to update the benchmark event feature template based on the valid event features that have continuously passed the matching verification, so that the template features can adapt to the slow waveform changes that may occur during long-term operation.

[0184] After identifying the target valid event, the energy centroid localization module extracts a valid event sampling segment from the preprocessed signal that corresponds to the valid event interval. Based on the local event energy distribution of this valid event sampling segment, it determines the event's time center point and time boundary. Based on the time center point, time start boundary, and time end boundary, the system generates and outputs trigger information containing the event timestamp and event segment data. Through this architecture, the system can convert continuous underlying baseband data streams into structured trigger outputs with clear time positions and data ranges, providing an input basis for subsequent synchronization processing, protocol parsing, or event data analysis.

[0185] To further verify the effectiveness and engineering applicability of the method proposed in this application embodiment under complex baseband environments, a system-level simulation experimental environment can be constructed to comprehensively evaluate the triggering accuracy, detection probability, and system resource overhead of this method. Regarding experimental setup and synthetic signal generation, this application embodiment simulates real wireless baseband edge communication scenarios, such as LoRa-type low-power wide-area communication, narrowband sensor networks, or low-power wireless sensor node receiving scenarios. The test baseband signal uses QPSK modulation, with an oversampling rate of 8, and a synchronization header with a known waveform structure is periodically embedded in the continuous random data stream as the target digital event to be detected. To simulate complex interference in the RF front-end and wireless channel, AWGN (Additive White Gaussian Noise) is injected into the clean signal, causing the SNR to vary between -5 dB and 15 dB; simultaneously, the superimposed frequency is... The low-frequency disturbance component simulates baseline fluctuations caused by device temperature changes, local oscillator bias, or slow-varying factors in the link. The sampling frequency was set, and 10% random high-amplitude burst impulse noise was introduced to construct a test environment that includes background noise, baseline perturbation, and transient spike interference.

[0186] In the comparative verification and comprehensive evaluation, the proposed method, the classic constant false alarm rate (CFAR) detection algorithm, and the standard differential thresholding (SDT) edge detection algorithm were selected as benchmarks. CFAR estimates background noise through a sliding window to adjust the trigger level, while SDT combines first-order difference with a fixed threshold for edge detection. The evaluation system includes three dimensions: firstly, the detection probability... With false alarm probability The study evaluates the target event capture capabilities of different methods under false alarm rate constraints using ROC (Receiver Operating Characteristic) curves. Secondly, it measures the triggering timing error distribution by statistically analyzing the absolute deviation between the event timestamp output by the algorithm and the actual timestamp to assess triggering alignment accuracy. Finally, it examines processing latency and memory overhead by simulating a resource-constrained microcontroller environment, such as a 100 MHz processing platform, to statistically analyze the clock cycles and RAM resources used by each method, thus evaluating the feasibility of deploying digital event triggering methods in the underlying hardware.

[0187] Figure 5 The diagram shows a comparison of receiver operating characteristic curves for different methods in a low signal-to-noise ratio environment, based on experimental simulation results.

[0188] like Figure 5 As shown, this experiment simulates the detection performance of the method provided in this application embodiment compared with the classical constant false alarm rate (CFAR) detection method and the standard differential edge detection method (SDT) under complex backgrounds with low signal-to-noise ratio (SNR), such as -2 dB, accompanied by slow drift of the local oscillator baseline and random high-amplitude burst pulse interference. The horizontal axis in the figure represents the false alarm probability measured logarithmically. The vertical axis represents the detection probability using a linear metric. .

[0189] from Figure 5It can be seen that in complex interference environments, the SDT method, which mainly relies on first-order difference and fixed threshold for edge extraction, is easily affected by high-frequency noise and sudden spikes. Its detection probability decreases significantly as the false alarm constraint tightens. The CFAR method can adjust the trigger threshold according to the background noise level, thus showing some improvement over the SDT method; however, in the presence of sudden impulse interference and baseline perturbation, its background statistics may still be affected by abnormal interference segments, thus limiting the detection probability. For example, in cases of false alarm probability... At that time, the detection probability corresponding to the CFAR method is approximately 72%.

[0190] In contrast, the method provided in this application has a higher detection probability under the same false alarm probability constraint. For example... Figure 5 As shown in the annotation, when At that time, the detection probability corresponding to the method provided in this application embodiment is approximately 94%. The reason is that this application embodiment reduces baseline drift and noise interference through front-end debiasing and noise filtering, and uses adaptive morphological processing, morphological gradient extraction and multi-dimensional feature template matching to jointly determine the local mutation contour and business features of the target digital event, thereby improving the event capture stability under low signal-to-noise ratio and sudden interference conditions.

[0191] therefore, Figure 5 Simulation results show that, under the same false alarm probability constraint, the method provided in this application embodiment can achieve a higher detection probability than the CFAR method and the SDT method, indicating that it has good event detection robustness in complex baseband interference environment.

[0192] Figure 6 A schematic diagram of the comparative experimental simulation results of an example of the trigger time error distribution of different methods under a specific signal-to-noise ratio condition is shown.

[0193] like Figure 6 As shown, the experiment simulated and statistically analyzed the trigger time error distribution of the method provided in this application embodiment compared to the classic constant false alarm rate (CFAR) detection method and the standard differential edge detection (SDT) method under specific signal-to-noise ratio (SNR) conditions, such as an SNR of 2 dB. The figure uses a violin plot combined with an internal box plot to illustrate the differences. The horizontal axis represents different detection methods, and the vertical axis represents the trigger time error, with units of discrete sampling points. The trigger time error characterizes the degree of deviation of the event timestamp output by each method relative to the actual event timestamp. The more concentrated the distribution, the higher the temporal stability of the trigger positioning result.

[0194] from Figure 6It can be seen that the trigger time error distributions of the SDT and CFAR methods are relatively wide and exhibit certain long-tail characteristics. Specifically, the SDT method mainly relies on first-order difference and a fixed threshold for edge extraction, making it susceptible to the influence of local high-frequency noise spikes. While the CFAR method can adjust the threshold according to the background noise level, it may still experience premature or delayed triggering when there is sudden interference or strong local noise fluctuations. Therefore, under the same signal-to-noise ratio conditions, the trigger time error range of these two methods is relatively large, and the stability of time positioning is somewhat limited.

[0195] In contrast, the trigger time error distribution of the method provided in this application is more concentrated, mainly distributed near zero error. This is because, after determining the target valid event, this application determines the time center point based on the local event energy distribution of the valid event sampling segment, rather than relying solely on the threshold crossing position of a single sampling point for trigger positioning. This process can utilize the overall energy distribution of the valid event segment to smooth local noise spikes and instantaneous jitter, thereby reducing the random offset of the trigger timestamp. Figure 6 As can be seen, the error distribution of the method provided in this application embodiment is mainly concentrated within a range of approximately ±1.5 sampling points, indicating that it can obtain more stable event time localization results.

[0196] therefore, Figure 6 Simulation results show that, under the same signal-to-noise ratio, the method provided in this application embodiment has a more concentrated trigger time error distribution than the CFAR method and the SDT method, which can improve the alignment stability of event timestamps.

[0197] Figure 7 This diagram illustrates a comparative simulation of different methods' engineering trade-offs between system-level global power consumption and front-end computational overhead in edge computing scenarios. The bar chart shows a comparison of the energy consumption of the method provided in this application embodiment with the classic Constant False Alarm Rate (CFAR) detection method and the Standard Differential Edge Detection (SDT) method, under a normalized expected total energy consumption index for a single detection cycle. The overall height of the bars in the chart represents the normalized expected total energy consumption; the lower area represents the processing energy consumption generated by the front-end microcontroller (MCU) executing the detection algorithm; and the upper area represents the additional energy consumption caused by invalid wake-ups of the main processor and RF link due to false alarms.

[0198] like Figure 7As shown, the CFAR and SDT methods have relatively low front-end MCU processing power consumption. However, due to their susceptibility to false triggers in complex interference environments, which can lead to invalid wake-ups of the main processor, cache link, or RF communication link, the power consumption from invalid wake-ups accounts for a high proportion of the total power consumption. In other words, the low computational overhead of a single front-end detection algorithm does not necessarily mean low system-level power consumption; when false alarms cause frequent wake-ups of the back-end link, the overall system power consumption may still remain at a high level.

[0199] In comparison, the method provided in this application, due to the introduction of adaptive morphological processing, morphological gradient filtering, and multi-dimensional feature template matching, results in increased front-end MCU processing energy consumption compared to the CFAR and SDT methods. The figure also illustrates the increased front-end computational overhead caused by the complexity of morphological operations. However, because the method provided in this application can reduce the probability of false alarms, the energy consumption of invalid wake-ups caused by false alarms is significantly reduced, resulting in a lower expected total energy consumption per normalized detection cycle compared to the comparative methods.

[0200] therefore, Figure 7 Simulation results show that, in edge computing or low-power IoT node scenarios, although the method provided in this application increases the front-end computing overhead to a certain extent, it can improve the system-level energy consumption performance by reducing the energy consumption of invalid wake-up caused by false alarms, demonstrating a good trade-off effect in engineering deployment.

[0201] In summary, this application compared and evaluated its detection performance, trigger time stability, and system-level energy consumption through system-level simulation experiments. Experimental results show that, under conditions of low signal-to-noise ratio, baseline perturbation, and random burst impulse noise, the method provided in this application's embodiments achieves a higher detection probability than the classic constant false alarm rate (CFAR) detection method and the standard differential edge detection method (SDT) under the same false alarm probability constraint. For example, when the false alarm probability is... Under the given conditions, the detection probability of the method provided in this application embodiment is higher than that of the comparative method, indicating that it can better improve the stability of target event capture under complex interference environment by means of bias removal and noise filtering, adaptive morphological processing, morphological gradient screening and multi-dimensional feature template matching.

[0202] Furthermore, the trigger time error distribution results show that the trigger time error of the method provided in this application embodiment is mainly concentrated near zero error, and the error distribution is more concentrated than that of the CFAR and SDT methods. This indicates that the method of determining the time center point based on the local event energy distribution can reduce the impact of local noise spikes and instantaneous jitter on the trigger timestamp. The system-level energy consumption comparison results further show that although the method provided in this application embodiment increases the front-end computational overhead due to the introduction of morphological operations and template matching, it can reduce the energy consumption of invalid wake-ups caused by false alarms, making the normalized expected total energy consumption per detection cycle lower than that of the comparative methods. Therefore, the method provided in this application embodiment has a good comprehensive trade-off effect between detection reliability, time positioning stability, and edge device deployment overhead.

[0203] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of combined actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Secondly, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application. In the above embodiments, the descriptions of each embodiment have their own emphasis; for parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0204] Figure 8 A structural block diagram of an example of a digital event triggering system based on baseband data sequence edge detection according to an embodiment of this application is shown.

[0205] like Figure 8 As shown, the digital event triggering system 800 based on baseband data sequence edge detection includes a baseband data preprocessing unit 810, a morphological feature extraction unit 820, a dynamic dual-threshold initial screening unit 830, a multi-dimensional feature template matching unit 840, an event energy boundary localization unit 850, and a digital event triggering output unit 860.

[0206] The baseband data preprocessing unit 810 is used to acquire the original baseband data sequence to be processed, and to perform DC bias removal, low-frequency drift suppression and noise filtering on the original baseband data sequence to obtain a preprocessed baseband data sequence.

[0207] The morphological feature extraction unit 820 is used to determine morphological structural elements based on the signal-to-noise ratio of the preprocessed baseband data sequence and the expected event width corresponding to the target business scenario, and to perform morphological opening and top-hat transformation on the preprocessed baseband data sequence using the morphological structural elements to obtain a top-hat feature sequence for characterizing the contour of local abrupt changes in the waveform.

[0208] The dynamic dual-threshold screening unit 830 is used to perform dilation and erosion operations on the top-hat feature sequence using the morphological structuring elements, and obtain a morphological gradient sequence based on the difference between the dilation and erosion operation results. Based on the statistical characteristics and noise estimation results of the morphological gradient sequence, a dual threshold is generated, and candidate event intervals are extracted from the morphological gradient sequence using the dual threshold.

[0209] The multidimensional feature template matching unit 840 is used to obtain a benchmark event feature template corresponding to the target business scenario, extract candidate event fragments and corresponding multidimensional candidate event features from the candidate event interval, and match and verify the multidimensional candidate event features with the benchmark event feature template to determine the target valid event and the valid event interval in which it is located.

[0210] The event energy boundary localization unit 850 is used to extract valid event sampling segments corresponding to the valid event interval from the preprocessed baseband data sequence, generate a local event energy sequence based on the amplitude characterization value of the valid event sampling segments, and determine the time center point, time start boundary, and time end boundary of the target valid event according to the local event energy sequence and event localization constraint information; wherein, the event localization constraint information is generated based on the event duration corresponding to the reference event feature template and the interval boundary of the valid event interval.

[0211] The digital event trigger output unit 860 is used to generate and output trigger information containing event timestamps and event fragment data based on the time center point, the time start boundary, and the time end boundary.

[0212] In some embodiments, this application provides a non-volatile computer-readable storage medium storing one or more programs including execution instructions. The execution instructions can be read and executed by an electronic device (including but not limited to a computer, server, or network device) to perform the steps of any of the digital event triggering methods based on baseband data sequence edge detection described above.

[0213] In some embodiments, this application also provides a computer program product, the computer program product including a computer program stored on a non-volatile computer-readable storage medium, the computer program including program instructions, which, when executed by a computer, cause the computer to perform the steps of any of the above-described digital event triggering methods based on baseband data sequence edge detection.

[0214] In some embodiments, this application also provides an electronic device comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform steps of a digital event triggering method based on baseband data sequence edge detection.

[0215] The above-described product can perform the methods provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects for performing the methods. Technical details not described in detail in this embodiment can be found in the methods provided in the embodiments of this application.

[0216] The electronic devices in this application can exist in various forms, including but not limited to: mobile communication devices, ultra-mobile personal computer devices, portable entertainment devices, or other airborne electronic devices with data interaction functions.

[0217] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0218] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0219] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A digital event triggering method based on baseband data sequence edge detection, characterized in that, The method includes: The raw baseband data sequence to be processed is obtained, and the raw baseband data sequence is subjected to DC bias removal, low frequency drift suppression and noise filtering to obtain a preprocessed baseband data sequence. Based on the signal-to-noise ratio of the preprocessed baseband data sequence and the expected event width corresponding to the target service scenario, morphological structural elements are determined, and morphological opening and top-hat transformation are performed on the preprocessed baseband data sequence using the morphological structural elements to obtain a top-hat feature sequence for characterizing the contour of local abrupt changes in the waveform. The top-hat feature sequence is subjected to dilation and erosion operations using the morphological structuring elements. A morphological gradient sequence is obtained based on the difference between the dilation and erosion results. A dual threshold is generated based on the statistical characteristics and noise estimation results of the morphological gradient sequence. Candidate event intervals are extracted from the morphological gradient sequence using the dual threshold. Obtain a baseline event feature template corresponding to the target business scenario, extract candidate event fragments and corresponding multi-dimensional candidate event features from the candidate event interval, and match and verify the multi-dimensional candidate event features with the baseline event feature template to determine the target valid event and the valid event interval in which it is located; Effective event sampling segments corresponding to the effective event interval are extracted from the preprocessed baseband data sequence. A local event energy sequence is generated based on the amplitude characterization value of the effective event sampling segments. The time center point, time start boundary, and time end boundary of the target effective event are determined according to the local event energy sequence and event location constraint information. The event location constraint information is generated based on the event duration corresponding to the reference event feature template and the interval boundary of the effective event interval. Based on the time center point, the time start boundary, and the time end boundary, trigger information containing event timestamps and event fragment data is generated and output.

2. The method according to claim 1, characterized in that, The process of performing DC bias removal, low-frequency drift suppression, and noise filtering on the original baseband data sequence to obtain a preprocessed baseband data sequence includes: For the current discrete sampling point in the original baseband data sequence, based on the sliding analysis window ending at the current discrete sampling point, the short-time local energy feature of the current detection interval is extracted, and the noise spectrum distribution of the current detection interval is estimated. An online update mechanism based on exponential decay is used to track and estimate the baseline offset component of the original baseband data sequence. The smoothing update weight factor of the online update mechanism is adjusted based on the short-time local energy feature, so that the smoothing update weight factor and the short-time local energy feature have a negative correlation, so as to reduce the degree of baseline estimation tracking of burst waveforms when the signal energy increases. Based on the difference between the sampled value at the current sampling time and the corresponding baseline bias component, a current debiased sampled value is generated, and a debiased sequence is formed from multiple current debiased sampled values ​​according to the sampling time sequence. Based on the noise spectrum distribution and the preset group delay constraint, the filter tap order and filter coefficients of the digital finite impulse response filter are configured, and the filter coefficients are designed and optimized using the minimum phase criterion. The configured digital finite impulse response filter is then used to perform noise filtering on the debiased sequence to obtain the preprocessed baseband data sequence.

3. The method according to claim 1, characterized in that, The step of determining the morphological structural elements based on the signal-to-noise ratio of the preprocessed baseband data sequence and the expected event width corresponding to the target business scenario includes: The average signal power and background noise variance of the preprocessed baseband data sequence under the current detection window are estimated respectively, and the ratio between the average signal power and the background noise variance is calculated to obtain the linear signal-to-noise ratio parameter, which is used as the corresponding signal-to-noise ratio. Based on the standard physical duration of the effective event to be detected in the target business scenario and the sampling frequency of the underlying analog-to-digital converter, the number of basic event sampling points mapped to the discrete time axis is calculated, and the number of basic event sampling points is determined as the expected event width; A natural exponential decay mapping model is constructed with the expected event width as the baseline scale and the linear signal-to-noise ratio parameter as the independent variable. The adaptive structuring element length is calculated based on the natural exponential decay mapping model. The natural exponential decay mapping model is configured to include a scale compensation term that increases exponentially with the decrease of the linear signal-to-noise ratio parameter, such that the adaptive structuring element length increases with the decrease of the linear signal-to-noise ratio parameter and shrinks to approach the baseline structuring element length determined by the expected event width and a preset scale adjustment coefficient as the linear signal-to-noise ratio parameter increases. Based on the calculated adaptive structural element length, a one-dimensional symmetrical flat line segment structure is constructed, consisting of a continuous discrete numerical sequence with equal element values, and the one-dimensional symmetrical flat line segment structure is determined as the morphological structural element.

4. The method according to claim 3, characterized in that, The step of performing morphological opening and top-hat transformation on the preprocessed baseband data sequence using the morphological structuring elements to obtain a top-hat feature sequence for characterizing the contours of local waveform abrupt changes includes: The morphological structuring elements are used to sequentially perform one-dimensional morphological erosion and one-dimensional morphological dilation operations on the preprocessed baseband data sequence to generate a morphological opening operation sequence. The difference between the sequence value of the preprocessed baseband data sequence and the sequence value of the morphological opening operation sequence is calculated point by point to obtain a positive top-cap feature sequence for characterizing local rising mutation features. The morphological structural elements are used to sequentially perform one-dimensional morphological dilation and one-dimensional morphological erosion operations on the preprocessed baseband data sequence to generate a morphological closing operation sequence. The difference between the sequence value of the morphological closing operation sequence and the sequence value of the preprocessed baseband data sequence is calculated point by point to obtain a negative bottom cap feature sequence for characterizing local descent abrupt change features. A bidirectional feature fusion process is performed on the feature values ​​of the positive top-hat feature sequence and the negative bottom-hat feature sequence at the same discrete position index to obtain a morphological contour feature sequence that integrates the positive top-hat feature and the negative bottom-hat feature, and the morphological contour feature sequence is determined as the top-hat feature sequence; wherein, the bidirectional feature fusion process includes point-by-point weighted summation or point-by-point maximum value taking.

5. The method according to claim 1, characterized in that, The process involves performing dilation and erosion operations on the top-hat feature sequence using the morphological structuring elements, obtaining a morphological gradient sequence based on the difference between the dilation and erosion results, generating a dual threshold based on the statistical characteristics and noise estimation results of the morphological gradient sequence, and extracting candidate event intervals from the morphological gradient sequence using the dual threshold, including: The morphological structural elements are used to perform one-dimensional morphological dilation and one-dimensional morphological erosion operations on the top cap feature sequence, and a morphological gradient sequence is generated based on the difference between the feature values ​​of the dilation operation and the feature values ​​of the erosion operation under the same discrete position index. Based on the gradient mean, gradient standard deviation, and background noise estimation parameters of the morphological gradient sequence within the current detection window, a current high edge threshold and a current low edge threshold are generated respectively; wherein, both the current high edge threshold and the current low edge threshold include a noise bias compensation term, and the gradient fluctuation scaling degree corresponding to the current low edge threshold is less than the gradient fluctuation scaling degree corresponding to the current high edge threshold; Identify the main edge sampling point from the morphological gradient sequence based on the current high edge threshold, and identify associated edge sampling points from the vicinity of the main edge sampling point based on the current low edge threshold; The main edge sampling points and the associated edge sampling points are continuously aggregated according to the sampling time sequence to form candidate event intervals.

6. The method according to claim 5, characterized in that, The steps of obtaining a baseline event feature template corresponding to the target business scenario, extracting candidate event fragments and corresponding multi-dimensional candidate event features from the candidate event interval, and matching and verifying the multi-dimensional candidate event features with the baseline event feature template to determine the target valid event and its corresponding valid event interval include: Obtain a set of historical benchmark event waveforms preset for the target business scenario, perform duration standardization and amplitude normalization on the historical benchmark event waveforms in the set, and extract the dominant feature basis vectors through principal component analysis to construct a benchmark event feature template. Candidate event segments are extracted from the preprocessed baseband data sequence according to the candidate event interval, and the candidate event segments are subjected to duration standardization and amplitude normalization. Multidimensional candidate event features are generated based on the processed candidate event segments. By combining a weighted window function configured based on the relative position of the temporal boundary and the background noise estimation parameters, weighted constraint correlation calculations are performed on the multidimensional candidate event features and multiple dominant feature basis vectors in the baseline event feature template, and the maximum weighted constraint correlation among the multiple weighted constraint correlation results is determined as the matching confidence. The weighted window function is used to enhance the matching contribution of features in the middle of the candidate event segment and suppress the feature contribution affected by noise at the boundary of the candidate event segment. Furthermore, the larger the background noise estimation parameter, the higher the degree of suppression of feature contribution at the boundary of the candidate event segment by the weighted window function. If the matching confidence meets the preset confidence judgment conditions and the overall time span of the candidate event interval conforms to the preset event duration range corresponding to the target business scenario, the candidate event segment is determined as the target valid event, and the candidate event interval corresponding to the target valid event is determined as the valid event interval. Based on a weighted sliding time window configured with a forgetting update factor, the features of target valid events that continuously pass the matching verification are fed back to the feature space corresponding to the benchmark event feature template, so as to iteratively update the benchmark event feature template.

7. The method according to claim 6, characterized in that, The step of extracting valid event sampling segments corresponding to the valid event interval from the preprocessed baseband data sequence, generating a local event energy sequence based on the amplitude characterization value of the valid event sampling segments, and determining the time center point, time start boundary, and time end boundary of the target valid event according to the local event energy sequence and event location constraint information includes: Based on the time domain boundary of the effective event interval, the corresponding sampled data is extracted from the preprocessed baseband data sequence as an effective event sampling segment, and a local event energy sequence is constructed based on the amplitude characterization value of each discrete sampling point within the effective event sampling segment. Based on the energy distribution of the local event energy sequence in the time domain, the energy centroid position of the target valid event is determined, and the energy centroid position is determined as the time center point; Obtain the standard event sampling one-sided span corresponding to the benchmark event feature template that has been matched and passed from the event location constraint information, and obtain the current global alignment compensation parameter updated based on historical decoding feedback. Offset the time center point to the start and end directions of the time axis by the standard event sampling one-sided span, and combine it with the current global alignment compensation parameter to determine the initial time start boundary and the initial time end boundary. Based on the sampling index range of the effective event interval, boundary constraint processing is performed on the initial time start boundary and the initial time end boundary to obtain the time start boundary and time end boundary of the target effective event; When the service protocol frame corresponding to the target valid event is successfully decoded by the upper-layer decoding unit, the current global alignment compensation parameter is smoothly updated based on the alignment time deviation between the absolute midpoint of the real event synchronization header fed back by the upper-layer decoding unit and the time center point, and the updated global alignment compensation parameter is used for the boundary positioning of subsequent digital events.

8. A digital event triggering system based on baseband data sequence edge detection, characterized in that, The system includes: The baseband data preprocessing unit is used to acquire the original baseband data sequence to be processed, and to perform DC bias removal, low-frequency drift suppression and noise filtering on the original baseband data sequence to obtain the preprocessed baseband data sequence. The morphological feature extraction unit is used to determine the morphological structural elements based on the signal-to-noise ratio of the preprocessed baseband data sequence and the expected event width corresponding to the target business scenario, and to perform morphological opening and top-hat transformation on the preprocessed baseband data sequence using the morphological structural elements to obtain a top-hat feature sequence for characterizing the contour of local abrupt changes in the waveform. The dynamic dual-threshold initial screening unit is used to perform dilation and erosion operations on the top-hat feature sequence using the morphological structuring elements, and obtain a morphological gradient sequence based on the difference between the dilation and erosion results. It generates dual thresholds based on the statistical characteristics and noise estimation results of the morphological gradient sequence, and extracts candidate event intervals from the morphological gradient sequence using the dual thresholds. A multi-dimensional feature template matching unit is used to obtain a benchmark event feature template corresponding to the target business scenario, extract candidate event fragments and corresponding multi-dimensional candidate event features from the candidate event interval, and match and verify the multi-dimensional candidate event features with the benchmark event feature template to determine the target valid event and the valid event interval in which it is located. An event energy boundary localization unit is used to extract valid event sampling segments corresponding to the valid event interval from the preprocessed baseband data sequence, generate a local event energy sequence based on the amplitude characterization value of the valid event sampling segments, and determine the time center point, time start boundary, and time end boundary of the target valid event according to the local event energy sequence and event localization constraint information; wherein, the event localization constraint information is generated based on the event duration corresponding to the reference event feature template and the interval boundary of the valid event interval; The digital event trigger output unit is used to generate and output trigger information containing event timestamps and event fragment data based on the time center point, the time start boundary, and the time end boundary.

9. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the method as described in any one of claims 1-7.

10. A storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1-7.

Images