Multimode coast station automatic watch method and system based on fusion signal analysis

Abstract

The application discloses a multi-mode coast radio station automatic watch method and system fusing signal analysis, and relates to the technical field of radio communication. The method comprises the following steps: acquiring a multi-mode baseband I / Q signal sequence, extracting a control channel incomplete bit stream and a voice service audio waveform; calculating a phase continuity parameter of the multi-mode baseband I / Q signal sequence to evaluate a channel fading degree; when the channel fading degree is lower than a threshold value, extracting high-confidence residual signaling features in the control channel incomplete bit stream through soft decision; mapping and comparing the high-confidence residual signaling features with a local maritime database to acquire ship identification information and dynamically construct a scenario customized acoustic micro-dictionary; using the acoustic micro-dictionary to perform acoustic space dimension reduction and matching constraint on the audio waveform, calculating an acoustic likelihood, and outputting a structured alarm text when an alarm condition is met. The application effectively solves the problem of missed alarms caused by bidirectional deterioration of communication signals in extreme sea conditions, and greatly improves the reliability of the coast radio station automatic watch system.

Description

Technical Field

[0001] This invention relates to the field of radio communication technology, and in particular to an automatic monitoring method and system for multi-standard coastal radio stations that integrates signal analysis. Background Technology

[0002] Coastal radio stations play a crucial role in receiving safety calls from ships in distress at sea. In the existing Global Maritime Distress and Safety System (GMDSS), coastal radio stations typically need to simultaneously monitor control channels used for data communication (such as the VHF CH70 band Digital Selective Calling (DSC)) and traffic channels used for voice communication (such as the VHF CH16 band Analog Voice).

[0003] Existing coastal radio station monitoring systems generally employ a data and voice processing architecture that isolates data and voice signals when handling multi-standard signals. Under extreme sea conditions such as typhoons and severe convective thunderstorms, radio channels experience severe multipath fading and impulse noise interference. In these extreme communication scenarios, the existing independent processing architecture reveals the following fatal flaws: On the one hand, when the digital control channel is interfered with, the error rate of burst data packets in the DSC surges. Traditional hardware devices will discard the data packet as invalid once the cyclic redundancy check (CRC) fails, causing the data alarm to completely fail. On the other hand, the distress calls of crew members in the voice service channel are masked and severely distorted by strong wind and wave noise. If this is directly input into a conventional large-vocabulary speech recognition (ASR) system, the extremely degraded signal-to-noise ratio will cause the ASR system to generate a large number of "illusions" and incorrect matches, making it impossible to extract effective distress information. When both data and voice fail to reach the decoding / recognition thresholds of their respective independent modules, the existing system will fall into a "double-blind" silent state, causing critical distress alarms to be completely missed. Summary of the Invention

[0004] To address the technical problem of missed alarms at coastal radio stations caused by the degradation of both data and voice signals under extreme sea conditions, the present invention aims to provide an automatic monitoring method and system for multi-standard coastal radio stations that integrates signal analysis.

[0005] The technical solution of this invention to solve the above-mentioned technical problems is as follows: A method for automatic monitoring of multi-standard coastal radio stations with fusion signal analysis, the method comprising: S10. Obtain the multi-standard baseband I / Q signal sequence received by the radio frequency front-end of the coastal radio station, and separate and extract the control channel residual bit stream and voice communication service channel audio characteristic waveform of the multi-standard baseband I / Q signal sequence. S20. Perform physical layer signal quality analysis on the multi-standard baseband I / Q signal sequence and calculate the phase continuity parameter used to characterize the degree of channel fading; S30. When the phase continuity parameter is lower than the set channel quality threshold, the feature extraction of the control channel residual bit stream is performed by a pre-configured soft decision decoder to obtain high-confidence residual signaling features. S40. The high-confidence residual signaling features are mapped and compared with a preset local maritime database that stores ship identification information and associated attributes. The successfully matched ship identification information is extracted, and the ship name and communication text associated with the ship identification information are extracted to dynamically construct a scenario-customized acoustic micro-dictionary. S50. In the constrained search space of the scenario-customized acoustic micro-dictionary, calculate the acoustic likelihood of the audio feature waveform of the voice communication service channel with the scenario-customized acoustic micro-dictionary. When the acoustic likelihood is greater than the set alarm threshold, generate and output a structured alarm text containing the ship identification information.

[0006] Further, the extraction of the audio feature waveform of the voice communication service channel in step S10 includes: S11. Use the set endpoint detection algorithm to locate the silent segment without speech in the baseband audio signal and extract the background noise spectrum envelope of the current environment. S12. Using the background noise spectrum envelope as a reference, the tap coefficient vector of the digital filter is iteratively updated using the set adaptive filtering formula. S13. The digital filter after state convergence is used to perform digital cancellation filtering on the continuous effective speech waveform to output an enhanced audio signal with background noise removed. S14. The enhanced audio signal is subjected to frame division and discrete cosine transform processing to reconstruct the audio feature waveform of the voice communication service channel.

[0007] Further, step S20 includes: S21. Window and frame the multi-standard baseband I / Q signal sequence, extract the in-phase component I(n) and quadrature component Q(n), and use the formula... Calculate the instantaneous phase angle of the nth sampling point; S22. Obtain the power delay spectrum of the multi-standard baseband I / Q signal sequence, and use the formula... Calculate the mean of multipath delay spread ,in For the multipath delay of the k-th path, For the corresponding power, This is the average delay; S23. Extract the total number of sampling points N and the set environmental attenuation coefficient. Using the formula Initial parameters are obtained by performing cascade operations; S24. Input the initial parameters into a set Kalman filter for state recursive estimation and smoothing, and output the phase continuity parameter after state convergence. .

[0008] Further, step S30 includes: S31. Obtain the original burst data packets that have not passed the cyclic redundancy check in the residual bit stream of the control channel, and construct the state transition mesh diagram of the soft decision decoder for the original burst data packets. S32. Based on the state transition mesh diagram, calculate the forward state metric value of each state node in the mesh. Branch metric and backward state metric ; S33, Using formulas Calculate input bits The log-likelihood ratio, where and These are the state transition sets for inputs equal to 1 and 0, respectively. S34. Extract multiple consecutive bit segments whose absolute values ​​of the log-likelihood ratios are all greater than the set single-bit likelihood threshold, and assemble the bit segments into the high-confidence residual signaling feature.

[0009] Further, step S40 includes: S41. Identify the maritime mobile communication service identification code fragment in the high-confidence residual signaling features, and use the maritime mobile communication service identification code fragment as the key value of the search engine. S42. In the preset local maritime database, the search engine key value is compared with the complete identification code of the registered vessel using regular expressions to filter and obtain a set of candidate vessel identifications and the associated vessel name and communication text. S43. Input the ship name and communication text into a set character-to-phoneme conversion neural network, and output the phoneme probability distribution matrix and hidden Markov model emission probability for each ship name and communication text. S44. Compile the phoneme probability distribution matrix and the emission probability of the hidden Markov model into a graph network to generate a state transition constraint topology graph, and load and update it into the scenario-customized acoustic micro-dictionary.

[0010] Furthermore, after step S42, the method further includes: S421. Extract the spatial radio frequency data collected by the antenna array, and calculate the estimated spatial angle of arrival value using the set spatial spectrum estimation model. ; S422. Extract the area coordinate data bound to each candidate vessel in the candidate vessel identifier set, and calculate the theoretical azimuth of the k-th candidate vessel relative to the coastal radio station. ; S423, Using the formula Calculate the spatial absolute angle deviation parameter corresponding to the k-th candidate ship. ; S424, when the spatial absolute angle deviation parameter When the angle exceeds the preset multipath tolerance threshold, the corresponding vessel will be removed from the candidate vessel identifier set, and the search capacity of the scenario-customized acoustic micro-dictionary will be reduced accordingly.

[0011] Further, step S50 includes: S51. Obtain a feature sequence of length I corresponding to the audio feature waveform of the voice communication service channel. And the state reference sequence of length J corresponding to the scenario-customized acoustic micro-dictionary. ; S52. Extract the average log-likelihood ratio of the high-confidence residual signaling features. Using the formula Calculate the width w of the dynamically warped window, where and The set adjustment coefficient; S53. Within the restricted area of ​​the dynamically adjusted window width, use the formula Calculate the cumulative normalization cost matrix, where The local distance is used to extract the global endpoint regularization cost D(I,J); S54. Using the formula The global endpoint normalization cost D(I,J) is inverted and mapped to the acoustic likelihood S, where This is the set normalized bias constant.

[0012] Further, the output of the structured alarm text in step S50 includes: S55. Extract the first timestamp data of the demodulated high-confidence residual signaling features and the second timestamp data of the context-customized acoustic micro-dictionary; S56. Calculate the absolute value of the time difference between the first timestamp data and the second timestamp data. When the absolute value of the time difference is less than the set synchronization window threshold, determine that the current cross-modal alarm event satisfies the spatiotemporal consistency constraint. S57. For events that meet the spatiotemporal consistency constraints, extract the matching ship identification information and the parsed emergency alarm string, and encapsulate them into structured data fields according to the set maritime protocol specifications. S58. Use the set secure hash algorithm to digitally sign the structured data field, and append the signature to the end of the data packet to generate the structured alarm text to be sent.

[0013] Another object of the present invention is to provide an automatic watchkeeping system for multi-standard coastal radio stations that integrates signal analysis, the system comprising: The data acquisition module is used to acquire the multi-standard baseband I / Q signal sequence received by the radio frequency front-end of the coastal radio station, and to separate and extract the control channel residual bit stream and voice communication service channel audio characteristic waveform of the multi-standard baseband I / Q signal sequence. The signal analysis module is used to perform physical layer signal quality analysis on the multi-standard baseband I / Q signal sequence and calculate the phase continuity parameter used to characterize the degree of channel fading. The soft-decision feature module is used to extract features from the control channel residual bit stream through a pre-configured soft-decision decoder when the phase continuity parameter is lower than the set channel quality threshold, so as to obtain high-confidence residual signaling features. The micro-dictionary construction module is used to map and compare the high-confidence residual signaling features with a preset local maritime database that stores ship identification information and associated attributes, extract the successfully matched ship identification information, and extract the ship name and communication text associated with the ship identification information to dynamically construct a scenario-customized acoustic micro-dictionary. The dimension reduction constraint matching module is used to calculate the acoustic likelihood of the audio feature waveform of the voice communication service channel with the context-customized acoustic micro-dictionary in the constraint search space of the context-customized acoustic micro-dictionary. When the acoustic likelihood is greater than the set alarm threshold, a structured alarm text containing the ship identification information is generated and output.

[0014] Furthermore, the signal analysis module includes: The baseband framing unit is used to window and frame the multi-standard baseband I / Q signal sequence, extract the in-phase component I(n) and quadrature component Q(n), and use the formula... Calculate the instantaneous phase angle of the nth sampling point; The delay spread estimation unit is used to obtain the power delay spectrum of the multi-standard baseband I / Q signal sequence, using the formula... Calculate the mean of multipath delay spread ,in For the multipath delay of the k-th path, For the corresponding power, This is the average delay; The penalty cascaded integrator is used to extract the total number of sampling points N and the set environmental attenuation coefficient. Using the formula Initial parameters are obtained by performing cascade operations; The Kalman filter unit is used to input the initial parameters into a set Kalman filter for recursive state estimation and smoothing, and output the phase continuity parameter after state convergence. .

[0015] The beneficial effects of this invention are as follows: 1. When the channel is severely fading, this invention uses soft decision technology to extract high-confidence residual signaling from the DSC residual bit stream that should be discarded, and uses this as a benchmark to rescue distorted voice across modes, thus achieving effective alarm when all single-mode failures occur. 2. This invention utilizes extracted deterministic incomplete signaling to locate specific ships and dynamically constructs a context-customized acoustic micro-dictionary containing only the ship's name and key distress words; it forcibly reduces the search space of the general speech model's 100,000-level vocabulary to tens of levels, completely eliminating mismatches under strong noise. 3. This invention innovatively extracts phase continuity parameters directly from multi-standard baseband I / Q signals as a benchmark for system coordinated triggering. Compared with the traditional energy threshold detection method which is easily deceived by impulse noise, the phase characteristics of the physical layer can objectively reflect the degree of fading of electromagnetic waves being torn apart, avoiding the system computing power collapse caused by frequent false triggering of large voice models due to lightning background noise. Attached Figure Description

[0016] Figure 1 A flowchart illustrating the automatic watchkeeping method for multi-standard coastal radio stations using fusion signal analysis provided in an embodiment of the present invention; Figure 2 A schematic diagram of the architecture of a multi-standard coastal radio station automatic monitoring system for fusion signal analysis provided in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the dimensionality reduction principle of the cross-modal constrained acoustic micro-dictionary for control channel incomplete data in an embodiment of the present invention. Detailed Implementation

[0017] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0018] The example background of this application is that a cargo ship named "OceanExplorer" (MMSI code: 412888999, current coordinates: 30°N, 122°E) encountered a severe typhoon at sea. Its antenna was damaged and leaking, resulting in severe errors in its VHF CH70 (Digital Selective Calling, DSC) transmission, while its VHF CH16 (Voice of Distress) transmission was severely submerged by typhoon noise. A coastal radio station received this mixed and extremely weakened signal.

[0019] like Figure 1As shown, this application provides an automatic watchkeeping method for multi-standard coastal radio stations using fusion signal analysis. The method includes: Step S10: Obtain the multi-standard baseband I / Q signal sequence received by the radio frequency front-end of the coastal radio station, separate and extract the control channel residual bit stream and voice communication service channel audio characteristic waveform of the multi-standard baseband I / Q signal sequence.

[0020] The purpose of this step is to achieve lossless acquisition of the underlying physical signals. The multi-standard baseband I / Q signal sequence is the original complex signal obtained by down-converting and quadrature sampling the spatial electromagnetic waves through the RF front-end of a software-defined radio (SDR) receiver. This signal contains all amplitude and phase information for communication. Traditional commercial hardware radios automatically discard garbled data, while this step directly isolates the frequency division multiplexing channel from the I / Q baseband, forcibly preserving the incomplete bit stream of the control channel that failed verification, and simultaneously segmenting the audio characteristic waveform of the voice communication service channel. Its technical effect lies in breaking the black-box limitation of traditional equipment's "silence upon packet loss," providing the most original physical data support for subsequent salvage decoding at extremely low signal-to-noise ratios.

[0021] In this embodiment, the SDR equipment of the coastal radio station intercepted a segment of incomplete burst data of DSC (i.e., incomplete bit stream of the control channel) that could not pass the conventional CRC check at the 156.525MHz frequency point, and at the same time recorded a segment of analog audio filled with the howling sound of wind and waves (i.e. audio characteristic waveform of voice communication service channel) at the 156.800MHz frequency point.

[0022] Furthermore, the extraction of audio feature waveforms of the voice communication service channel in step S10 includes: Step S11: Use the set endpoint detection algorithm to locate the silent segment without speech in the baseband audio signal and extract the background noise spectrum envelope of the current environment.

[0023] Endpoint detection algorithms aim to identify time periods of pure ambient noise in a communication channel. A Fourier transform of these periods yields the background noise spectral envelope. The benefit of this technique is providing a clean reference noise fingerprint sample for subsequent adaptive cancellation.

[0024] In this embodiment, before the crew member presses the talk button in the first 2 seconds of the radio recording, the system extracts the spectral envelope of the pure typhoon howling sound and the rain hitting the antenna during these 2 seconds.

[0025] Step S12: Using the background noise spectrum envelope as a reference, iteratively update the tap coefficient vector of the digital filter using the set adaptive filtering formula.

[0026] Explanation: The adaptive filtering formula uses a minimum mean square error algorithm to continuously adjust the internal weight matrix (i.e., the tap coefficient vector) of the digital filter, making its frequency response approximate the ambient background noise obtained above. Its technical principle is to use machine learning to establish an acoustic cancellation model that is the opposite of the ambient noise.

[0027] In this embodiment, the internal parameters of the digital filter have undergone hundreds of iterations, and its tap coefficient vector has completely simulated the waveform characteristics of the current typhoon howling sound.

[0028] Step S13: Use the digital filter after state convergence to perform digital cancellation filtering on the continuous effective speech waveform, and output an enhanced audio signal with background noise removed.

[0029] Explanation: Following the previous steps, the mixed audio including the crew's distress calls is input into a digital filter that has already achieved state convergence. Utilizing the physical acoustic interference principle of phase reversal and addition, environmental noise is canceled out, retaining only the human voice signal, resulting in an enhanced audio signal. The technical effect is a significant improvement in the signal-to-noise ratio for subsequent acoustic feature extraction.

[0030] In this embodiment, after the audio with a distress call passes through the filter, the background howling sound is greatly reduced, and the crew's hoarse voice (i.e., the enhanced audio signal) is clearly revealed.

[0031] Step S14: Perform frame segmentation and discrete cosine transform processing on the enhanced audio signal to reconstruct and generate the audio characteristic waveform of the voice communication service channel.

[0032] Explanation: The enhanced audio signal is segmented into short-time overlapping frames, and the Mel frequency cepstral coefficient features are extracted. The discrete cosine transform removes the correlation between features in different frequency bands and compresses redundant information. The final output feature sequence is the audio feature waveform of the voice communication service channel required in step S10.

[0033] In this embodiment, the denoised human voice waveform is converted into a matrix sequence containing 13-dimensional cepstral features and reconstructed into a standard feature waveform that can be directly used by subsequent DTW.

[0034] Step S20: Perform physical layer signal quality analysis on the multi-standard baseband I / Q signal sequence and calculate the phase continuity parameter used to characterize the degree of channel fading.

[0035] This step aims to abandon the amplitude threshold detection method, which is highly susceptible to interference from pulse energy such as typhoons and thunderstorms, and instead evaluate the signal from the perspective of electromagnetic wave propagation physics. By calculating the phase deflection stability of multi-mode baseband I / Q signal sequences on the time axis, a phase continuity parameter is obtained. This data accurately reflects the degree to which electromagnetic waves are physically torn apart during multipath reflection at the sea surface. The technical advantage lies in its ability to accurately determine whether the current channel has collapsed under severe weather conditions, avoiding meaningless computational waste in subsequent speech modeling.

[0036] In this embodiment, analysis of the above I / Q signal revealed that its phase dispersion is extremely high, and the calculated phase continuity parameter is only 0.15 (the maximum score is 1).

[0037] Furthermore, step S20 specifically includes the following processes: Step S21: Window and frame the multi-standard baseband I / Q signal sequence, extract the in-phase component I(n) and quadrature component Q(n), and use the formula... Calculate the instantaneous phase angle of the nth sampling point.

[0038] I(n) represents the real part of the radio frequency wave, and Q(n) represents the imaginary part. The arctangent of their ratio is the physical real angle (i.e., the instantaneous phase angle) of the electromagnetic wave at the instant it reaches the antenna.

[0039] In this embodiment, the data from the 100th sampling point is extracted, with an I value of 0.5 and a Q value of 0.5. The system calculates that its instantaneous phase angle is 45 degrees. radian).

[0040] Step S22: Obtain the power delay spectrum of the multi-standard baseband I / Q signal sequence, and use the formula... Calculate the mean of multipath delay spread ,in For the multipath delay of the k-th path, For the corresponding power, This represents the average latency.

[0041] The numerator of the formula represents the weighted sum of the squares of the arrival time differences of each reflected wave, and the denominator is the total power. Sea surface wave undulations cause multipath effects. The larger the value, the more severe the multipath fading and the greater the signal overlap and tearing.

[0042] In this embodiment, the typhoon generates huge waves that cause multiple reflections of radio waves, and the average multipath delay spread is calculated. The time was as high as 5 microseconds, far exceeding the normal and stable sea state index.

[0043] Step S23: Extract the total number of sampling points N and the set environmental attenuation coefficient. Using the formula The initial parameters are obtained by performing cascade operations.

[0044] In this formula, the summation of complex terms in the first term describes the smoothness of the phase change, while the exponential penalty term in the second term... The multipath damage effect of step S22 is introduced. An initial parameter for comprehensively evaluating signal fading is generated through a cascaded multiplication method.

[0045] In this embodiment, the initial parameter is 0.15 by combining the scattered phase (the former is evaluated as 0.3) and the large multipath delay penalty (the latter is evaluated as 0.5) and cascading multiplication.

[0046] Step S24: Input the initial parameters into the set Kalman filter for state recursive estimation and smoothing, and output the phase continuity parameter after state convergence. .

[0047] Initial parameters calculated from isolated sampling points are highly susceptible to numerical jumps caused by accidental lightning pulses. The Kalman filter uses the optimal estimate from the previous moment to predict the current state, thereby filtering out random abrupt changes and obtaining extremely stable phase continuity parameters. .

[0048] In this embodiment, the parameter jumps to 0.8 due to lightning within a certain millisecond, but the Kalman filter smoothly removes this outlier based on the historical state, and finally outputs the true degradation evaluation parameter of 0.15.

[0049] Step S30: When the phase continuity parameter is lower than the set channel quality threshold, the feature extraction of the control channel residual bit stream is performed by a pre-configured soft decision decoder to obtain high-confidence residual signaling features.

[0050] This step connects to the preceding step S20. When the phase continuity parameter proves that the channel has severely faded, conventional hard decision methods completely fail. At this point, the receiver's underlying soft decision decoder outputs soft bits with a log-likelihood ratio (probability attribute). The system isolates the consecutive bit segments that are most likely correct as high-confidence residual signaling features. The technical effect is that absolutely objective digital prior information is extracted from the garbled data that was judged as invalid packets.

[0051] In this embodiment, a soft decision is triggered because the parameter 0.15 is lower than the set channel quality threshold of 0.4. From the DSC scrambled data, a residual bit sequence with a confidence level of 99% is extracted and decoded into the digital fragment "412888".

[0052] Furthermore, step S30 specifically includes the following processes: Step S31: Obtain the original burst data packets that have not passed the cyclic redundancy check in the control channel residual bit stream, and construct the state transition mesh diagram of the soft decision decoder for the original burst data packets.

[0053] Explanation: Messages that should have been discarded due to CRC check failures are recycled, and a Markov chain state probability graph (i.e., a state transition mesh graph) based on the time axis is constructed for the soft decision decoder, thus building a topology framework for subsequent probability tracing.

[0054] In this embodiment, a standard DSC packet fails the CRC check, so the system captures it and expands it into a grid diagram containing 256 time steps in memory.

[0055] Step S32: Based on the state transition mesh diagram, calculate the forward state metric value of each state node in the mesh. Branch metric and backward state metric .

[0056] This is the feature expansion of the BCJR decoding algorithm. Represents the forward cumulative probability. This represents the cumulative probability calculated backwards. This represents the single-step jump probability. This step completes the global probability traversal calculation of the incomplete signaling within the grid.

[0057] In this embodiment, for the 50th bit, the system simultaneously calculates the values ​​derived from the 1st bit. The value, and the value derived from the last digit back to this point. value.

[0058] Step S33: Using the formula Calculate input bits The log-likelihood ratio, where and These are the state transition sets for inputs equal to 1 and 0, respectively.

[0059] The numerator of the formula represents the overall probability that the true value of the bit is 1, and the denominator represents the overall probability that the true value is 0. The logarithm of their ratio is the log-likelihood ratio (LLR). The absolute value of the LLR represents the confidence level of the decision (the higher the value, the greater the confidence).

[0060] In this embodiment, the probability of a certain bit being 1 is calculated to be extremely high. After taking the logarithm, the LLR is +8.5, indicating that the bit is definitely a valid bit "1".

[0061] Step S34: Extract bit segments where the absolute value of multiple consecutive log-likelihood ratios is greater than the set single-bit likelihood threshold, and assemble the bit segments into a high-confidence residual signaling feature.

[0062] Explanation: The reliable bits with extremely high LLR absolute values ​​from step S33 are extracted. These consecutive correct fragments are pieced together to form the prior information used for dictionary constraints, namely, the high-confidence residual signaling features.

[0063] In this embodiment, the system scans the entire LLR sequence of the discarded packet and extracts a 30-bit segment where the absolute value of the LLR is 6 or higher, which is then restored to the decimal number "412888".

[0064] Step S40: Map and compare the high-confidence residual signaling features with the preset local maritime database that stores ship identification information and associated attributes, extract the successfully matched ship identification information, and extract the ship name and communication text associated with the ship identification information to dynamically construct a scenario-customized acoustic micro-dictionary.

[0065] This step utilizes the objective digital fragments output from step S30 for cross-modal search and dimensionality reduction. A pre-set local maritime database contains features such as the MMSI and ship names of globally registered vessels. By searching the database for the digital fragments, a few specific vessels (ship identification information) can be identified. Then, the English names of these vessels and commonly used distress phrases (ship names and communication text) are extracted and packaged into a context-customized acoustic micro-dictionary containing only a few dozen words. The technical effect is that it forcibly compresses the originally massive decoding space of a general language model into an extremely simplified micro-dictionary space, physically and logically eliminating the illusion of recognition by the large model under strong noise.

[0066] In this embodiment, the digital fragment "412888" is matched using regular expressions in a preset local maritime database, instantly identifying the unique matching vessel as "Ocean Explorer". The system then constructs a tiny micro-dictionary from specific words such as "OceanExplorer" and "Mayday".

[0067] Furthermore, step S40 specifically includes the following processes: Step S41: Identify the maritime mobile communication service identification code fragments in the high-confidence residual signaling features, and use the maritime mobile communication service identification code fragments as the key values ​​of the search engine.

[0068] Explanation: The Maritime Mobile Service Identifier (MMSI) is a unique identification code for ships. This MMSI fragment is extracted from the aforementioned pieced-together feature segments and used as a keyword (i.e., a search engine key value) for subsequent database comparison.

[0069] In this embodiment, the system extracts that the above fragment happens to be in MMSI format, and then uses the extracted "412888" as the key value of the search engine.

[0070] Step S42: In the preset local maritime database, the search engine key value is compared with the complete identification code of the registered vessel using regular expressions to filter and obtain the candidate vessel identification set and associated vessel name and communication text.

[0071] Explanation: The extracted short string is compared with the database using regular expressions to find all valid MMSI codes containing the fragment, thereby drastically narrowing down the massive target to a few specific ships (i.e., a set of candidate ship identifiers), and then its text data is retrieved.

[0072] In this embodiment, three ships matched the fragment in the database. The system extracted the text data of these three ships, including the English ship name "OceanExplorer" and the standard communication text "Mayday".

[0073] Step S43: Input the ship name and communication text into the set character-to-phoneme conversion neural network, and output the phoneme probability distribution matrix and hidden Markov model emission probability for each ship name and communication text.

[0074] Explanation: The character-to-phoneme (G2P) model includes an attention encoder that receives character sequence input, captures letter pronunciation patterns through pre-trained network weights, and finally outputs the probability transition matrix of each phoneme in the standard phoneme dictionary. This model enables plain text to be automatically and accurately converted into the standard pronunciation state at the underlying acoustic level.

[0075] In this embodiment, the text "OceanExplorer" is input into the G2P neural network, and the model outputs the corresponding standard phoneme sequence probability and its emission probability matrix in the hidden Markov model.

[0076] Step S44: Compile the phoneme probability distribution matrix and the emission probability of the hidden Markov model into a graph network to generate a state transition constraint topology graph, load and update it into a context-customized acoustic micro-dictionary.

[0077] Explanation: Using the weighted finite state transformer (WFST) technique, the above probability matrix is ​​compiled into a grid map (i.e., a state transition constraint topology map) that expands from the word level down to the phoneme level and the state level, which constitutes the bottom-level map boundary of the dimension reduction and normalization in step S50.

[0078] In this embodiment, the system packages the generated phoneme matrix and constructs a probabilistic topology map in memory that only allows the pronunciation to go to specific word states such as "OceanExplorer", which is a context-customized acoustic micro-dictionary.

[0079] Furthermore, to further improve matching accuracy, step S42 includes the following: Step S421: Extract the spatial radio frequency data collected by the antenna array, and calculate the estimated spatial angle of arrival value using the set spatial spectrum estimation model. .

[0080] This step incorporates the spatial phase difference characteristics of the antenna physical array and uses a high-resolution spatial spectrum estimation model to calculate the true physical direction of arrival (i.e., the estimated spatial angle of arrival) of the radio frequency signal. ).

[0081] In this embodiment, the antenna array of the coastal radio station calculates the estimated actual spatial angle of arrival of the distress call wave as 105 degrees based on the time difference of the received radio waves.

[0082] Step S422: Extract the area coordinate data bound to each candidate vessel in the candidate vessel identifier set, and calculate the theoretical azimuth of the k-th candidate vessel relative to the coastal radio station. .

[0083] This step, combined with the ship's most recent AIS coordinates obtained in step S42, uses spherical geometry calculation formulas to calculate the absolute physical angle (i.e., theoretical azimuth angle) that the ship should have been at relative to the receiving base station. ).

[0084] In this embodiment, the database shows that the closest coordinates of the "Ocean Exploration" vessel are 30°N, 122°E, and its theoretical azimuth angle relative to the base station is calculated. It is 104 degrees.

[0085] Step S423: Using the formula Calculate the spatial absolute angle deviation parameter corresponding to the k-th candidate ship. .

[0086] The above formula is used to calculate the absolute value of the shortest angle between the actual incoming wave direction and the theoretical direction (considering the 360-degree circular reflection characteristic), and to obtain the spatial deviation between the physical actual incoming wave value and the theoretical coordinate value.

[0087] In this embodiment, the actual arrival angle of 105 degrees is compared with the theoretical position of 104 degrees to calculate the absolute angle deviation parameter. It is 1 degree.

[0088] Step S424, when the spatial absolute angle deviation parameter When the angle exceeds the preset multipath tolerance threshold, the corresponding vessel will be removed from the candidate vessel identifier set, and the search capacity of the scenario-customized acoustic micro-dictionary will be reduced accordingly.

[0089] Explanation: If there is an inexplicable and significant deviation between the theoretical position of a candidate ship and the actual direction of the incoming radio waves, it indicates that the matched object is a mismatch (i.e., a coincidental identical signaling fragment). Removing it can prevent misidentification in subsequent voice matching.

[0090] In this embodiment, another candidate ship is located directly north of the base station (with a deviation of 80 degrees), and the system directly eliminates it, further reducing the capacity of the micro-dictionary and ensuring extremely high-precision acoustic matching in the future.

[0091] Step S50: In the constrained search space of the context-customized acoustic micro-dictionary, calculate the acoustic likelihood of the audio characteristic waveform of the voice communication service channel with the context-customized acoustic micro-dictionary. When the acoustic likelihood is greater than the set alarm threshold, generate and output a structured alarm text containing ship identification information.

[0092] This step inputs the substandard speech from step S10 into the simplified space constructed in step S40 for fallback comparison. Instead of requiring the system to perform open full speech-to-text transcription, it judges the similarity (i.e., acoustic likelihood) between the audio waveform and high-risk words in the micro-dictionary. Once this likelihood exceeds the safety threshold, a structured alarm text is immediately triggered and generated. The technical effect is that, through cross-standard hard coupling, it successfully salvages damaged speech using incomplete data, achieving a highly sensitive closed-loop automatic alarm in extreme environments.

[0093] In this embodiment, a forced similarity comparison was performed between the blurred speech in the typhoon noise and the micro-dictionary. It was found that the audio trajectory highly overlapped with "OceanExplorer Mayday", with a likelihood of 85%. The system immediately output a structured alarm text containing "OceanExplorer, 412888999" to the maritime command center.

[0094] Furthermore, step S50 specifically includes the following processes: Step S51: Obtain the feature sequence of length I corresponding to the audio feature waveform of the voice communication service channel. And the context-customized acoustic micro-dictionary corresponding to the state reference sequence of length J. .

[0095] Explanation: Prepare two-dimensional input sequences for the dynamic time warping algorithm. Sequence X is an unknown sequence of damaged speech features; sequence Y is a standard micro-dictionary pronunciation sequence pre-compiled in step S40.

[0096] In this embodiment, the extracted damaged speech feature sequence X has a length of 500 frames; the micro-dictionary standard sequence Y has a length of 450 frames.

[0097] Step S52: Extract the average log-likelihood ratio of high-confidence residual signaling features. Using the formula Calculate the width w of the dynamically warped window, where and The set adjustment coefficient.

[0098] Explanation: This relates to the core innovative algorithm. It involves linking the confidence level of the signaling data dimension (…). This is used as a hard penalty parameter. The latter part of the formula... The window width is constrained: if the underlying signaling confidence is extremely high, this term approaches 0, and the dynamically warped window width w is forcibly narrowed. The technical effect is to establish a deterministic mathematical mapping between the digital signal-to-noise ratio and the acoustic matching tolerance space.

[0099] In this embodiment, the signaling extraction of fragments has a very high degree of certainty ( (Very large), the system calculates that the dynamic regularization window width w for acoustic matching is extremely narrow, allowing features to be distorted and aligned only within a very small time deviation.

[0100] Step S53: Within the restricted area of ​​the dynamically adjusted window width, use the formula... Calculate the cumulative normalization cost matrix, where The local distance is used to extract the global endpoint regularization cost D(I,J).

[0101] This is the DTW matrix recursion process under bandwidth constraints. Limiting execution to a width w significantly reduces system computational power consumption. The final output D(I,J) represents the total physical difference between two unequal-length acoustic waveforms.

[0102] In this embodiment, under the constraint of an extremely narrow window, the system aligns 500 frames of unknown audio with 450 frames of standard dictionary audio, and calculates their final physical acoustic difference cost D(I,J) to be 12.5.

[0103] Step S54: Using the formula The global endpoint normalization cost D(I,J) is inverted and mapped to the acoustic likelihood S, where This is the set normalization bias constant. The absolute value of the physical distance D(I,J) is smoothly mapped to a probability assessment value (i.e., acoustic likelihood S) between 0 and 1 using a nonlinear function, so that the main process can perform threshold comparison to trigger an alarm.

[0104] In this embodiment, after substituting the cost difference of 12.5 into the formula for inversion, the acoustic likelihood S is obtained as 0.85 (that is, an 85% confidence level to determine that the distorted audio hits a high-risk word in the dictionary).

[0105] Furthermore, the structured alert text output in step S50 includes: Step S55: Extract the first timestamp data of the demodulated high-confidence residual signaling features, and the second timestamp data of the context-customized acoustic micro-dictionary. Obtain the occurrence time of the physical reception event from the data signaling subsystem and the analog voice subsystem, respectively. This is the cornerstone of cross-standard data mutual verification and fusion triggering.

[0106] In this embodiment, the precise time when the incomplete DSC message arrived at the system (i.e., the first timestamp data is 12:00:01.005) and the voice segment start time (i.e., the second timestamp data is 12:00:01.350) are extracted.

[0107] Step S56: Calculate the absolute value of the time difference between the first timestamp data and the second timestamp data. When the absolute value of the time difference is less than the set synchronization window threshold, determine that the current cross-modal alarm event meets the spatiotemporal consistency constraint.

[0108] Explanation: This determines whether the voice distress call and signaling data originate within the same physical timeframe. Only when the times are extremely close can it be proven that they were triggered by the same event. The technical effect is to prevent false alarms caused by accidental combinations of random random data and unrelated noise.

[0109] In this embodiment, the difference between the two is 0.345 seconds, which is less than the set maximum delay synchronization window threshold (2.0 seconds), and the system confirms that the event meets the spatiotemporal consistency constraint.

[0110] Step S57: For events that meet the spatiotemporal consistency constraints, extract the matching ship identification information and the parsed emergency alarm string, and encapsulate them into structured data fields according to the set maritime protocol specifications.

[0111] The final conclusions derived from the above cross-modal physical and mathematical operations (including MMSI, vessel name, and alarm category) are reorganized and assembled into computer-readable code (structured data fields) that conform to current international maritime network standards.

[0112] In this embodiment, the ship name and coordinates are converted into a standard data field encapsulation structure such as "PCDSC,412888999,DISTRESS...".

[0113] Step S58: Use the set secure hash algorithm to digitally sign the structured data fields, and append the signature to the end of the data packet to generate the structured alarm text. Explanation: A hash algorithm is used to generate a tamper-proof verification code for the entire alarm text, ensuring the evidentiary validity of the rescue alarm in subsequent transmission over the maritime private network, and ultimately completing the closed-loop output of the alarm service.

[0114] In this embodiment, the system performs a hash operation on the assembled fields to generate an encrypted signature string, which is then appended to the end. The coastal radio station officially announces the successful triggering of the fusion alarm service under the extreme environment of this typhoon.

[0115] like Figure 2 and Figure 3 As shown, this invention provides an automatic monitoring system for multi-standard coastal radio stations that integrates signal analysis. The system includes: The data acquisition module is used to acquire the multi-standard baseband I / Q signal sequence received by the radio frequency front-end of the coastal radio station, separate and extract the control channel residual bit stream of the multi-standard baseband I / Q signal sequence and the audio characteristic waveform of the voice communication service channel in the intermediate frequency division multiplexing; The signal analysis module is used to perform physical layer signal quality analysis on multi-standard baseband I / Q signal sequences and calculate phase continuity parameters used to characterize the degree of channel fading. The soft-decision feature module is used to extract features from the control channel residual bit stream through a pre-configured soft-decision decoder when the phase continuity parameter is lower than the set channel quality threshold, so as to obtain high-confidence residual signaling features. The micro-dictionary construction module is used to map and compare high-confidence residual signaling features with a preset local maritime database that stores ship identification information and associated attributes, extract successfully matched ship identification information, and extract the ship name and communication text associated with the ship identification information to dynamically construct a scenario-customized acoustic micro-dictionary. The dimension reduction constraint matching module is used to calculate the acoustic likelihood of the audio feature waveform of the voice communication service channel with the context-customized acoustic micro-dictionary in the constraint search space of the context-customized acoustic micro-dictionary. When the acoustic likelihood is greater than the set alarm threshold, a structured alarm text containing ship identification information is generated and output.

[0116] Furthermore, the signal analysis module includes: The baseband framing unit is used to window and frame multi-standard baseband I / Q signal sequences, extracting the in-phase component I(n) and quadrature component Q(n) using the formula... Calculate the instantaneous phase angle at the nth sampling point; the time delay spread estimation unit is used to obtain the power delay spectrum of multi-standard baseband I / Q signal sequences, using the formula Calculate the mean of multipath delay spread ,in For the multipath delay of the k-th path, For the corresponding power, The average time delay is calculated using a penalized cascaded integrator unit, which is used to extract the total number of sampling points N and the set environmental attenuation coefficient. Using the formula Initial parameters are obtained through cascaded operations; the Kalman filter unit is used to input the initial parameters into the set Kalman filter for state recursive estimation and smoothing, and outputs the phase continuity parameter after state convergence. .

[0117] To further illustrate the above embodiments, the following description is provided in conjunction with the accompanying drawings.

[0118] like Figure 2 As shown, in practical engineering applications, the system also includes a low-level hardware execution environment, which mainly includes: a data acquisition module 210, a signal analysis module 220, a soft decision feature module 230, a micro-dictionary construction module 240, and a dimension reduction constraint matching module 250.

[0119] Specifically, the signal analysis module 220 is further refined to meet the underlying computational requirements, including: a baseband framing unit 221, a delay spread estimation unit 222, a penalty cascade integration unit 223, and a Kalman filter unit 224.

[0120] Meanwhile, in order to enable those skilled in the art to clearly understand the physical implementation of the aforementioned software algorithm model, such as... Figure 3 As shown, this embodiment also relies on a low-level hardware execution environment. This low-level hardware execution environment mainly includes: a radio frequency receiving front-end 310, a processor, a memory, and a communication output interface 320. In practical engineering applications, the various functional modules (210-250) and sub-units of the coastal radio station automatic monitoring system are all permanently stored in the memory as computer program instructions or AI model weights, and are called and executed by the processor in real time to realize the scheduling of the underlying physical radio frequency hardware and cross-modal data processing.

[0121] Combination Figure 1 The flowchart shown illustrates the automatic monitoring method for multi-standard coastal radio stations using fusion signal analysis in this embodiment. This method is executed collaboratively by the processor in the underlying hardware execution environment, which calls various functional modules of the automatic monitoring system for coastal radio stations.

[0122] Specifically, the data acquisition module 210, in coordination with the radio frequency receiving front-end 310, executes step S10; the signal analysis module 220 and its sub-units (baseband framing unit 221, delay spread estimation unit 222, penalty cascade integration unit 223, and Kalman filter unit 224) execute step S20; the soft decision feature module 230 executes step S30; the micro-dictionary construction module 240, in coordination with the memory, executes step S40; and the dimensionality reduction constraint matching module 250, in coordination with the communication output interface 320, executes step S50.

[0123] like Figure 3As shown, the core technology of this invention lies in establishing a cross-modal collaborative monitoring data flow link: starting from the incomplete bit stream of the control channel, high-confidence residual signaling features are obtained through soft-decision decoding feature extraction, and matched ship identification information is obtained by mapping and comparison in a preset local maritime database. The associated ship name and communication text are extracted, and a scenario-customized acoustic micro-dictionary is constructed through a character-to-phoneme neural network compilation. This micro-dictionary serves as a minimal constraint search space for dynamic time warping of the audio feature waveform of the voice communication service channel, outputting acoustic likelihood. When it exceeds the set alarm threshold, a structured alarm text containing ship identification information is output.

[0124] To more intuitively and coherently illustrate the technical logic and implementation effect of the cross-modal collaborative monitoring mechanism in this invention, the following is an example of a specific application scenario under extreme sea conditions: Scenario: A cargo ship named "OceanExplorer" (MMSI code: 412888999, current coordinates: 30°N, 122°E) encounters a severe typhoon in international waters. The ship's antenna is damaged by water ingress, resulting in severe errors in its VHF CH70 Digital Selective Call (DSC) control signal. Simultaneously, its VHF CH16 voice communication signal is severely drowned out by the typhoon's howling. The conventional coastal radio station is in a "double-blind" state of missed alarms due to CRC check failure and a zero voice ASR recognition rate.

[0125] Implementation process (in combination) Figure 3 (as shown) 1. Execution step S10: After intercepting the electromagnetic waves in the airspace, the radio frequency receiving front-end 310 in the underlying hardware execution environment of the coastal radio station down-converts the frequency to generate a multi-standard baseband I / Q signal sequence. The data acquisition module 210 no longer discards erroneous codes, but forcibly captures the DSC incomplete burst data (i.e., the incomplete bit stream of the control channel) that has not passed the CRC check, and simultaneously segments out the analog audio (i.e., the audio characteristic waveform of the voice communication service channel) that is completely covered by wind and wave noise.

[0126] 2. Step S20: The signal analysis module 220 continuously analyzes the real and imaginary parts of the baseband sequence and calculates that the current multipath delay is extremely large and the phase is scattered. After smoothing by the underlying algorithm, the final output phase continuity parameter is only 0.15 (the preset channel quality threshold is 0.4), and the system accurately determines that the physical channel is on the verge of collapse.

[0127] 3. Step S30: Because the parameter 0.15 is significantly lower than the threshold 0.4, the soft-decision feature module 230 is immediately activated. This module calls the underlying Viterbi decoder to calculate the soft-decision log-likelihood ratio (LLR) of the aforementioned DSC scrambled code. The system successfully extracted a residual bit sequence with a confidence level of 99% from the massive amount of erroneous codes, restoring it to the decimal digital fragment "412888" (i.e., the high-confidence residual signaling feature).

[0128] 4. Step S40: The micro-dictionary construction module 240 uses the digital fragment "412888" as the key value of the search engine and inputs it into the preset local maritime database in the memory for regular expression matching. Instantly, it locks down the unique matching vessel "Ocean Explorer" from tens of thousands of ship records. Subsequently, the system retrieves the ship name "OceanExplorer" and the maritime standard distress term "Mayday", and compiles them through a character-to-phoneme (G2P) neural network to dynamically generate a very small-capacity scenario-customized acoustic micro-dictionary in memory (completing the physical dimensionality reduction of the search space).

[0129] 5. Step S50: The dimensionality reduction constraint matching module 250 inputs the severely distorted voice communication service channel audio feature waveform from step S10 into the constraint search space of the micro-dictionary for dynamic time warping (DTW) calculation. After hard constraint calculation, the acoustic trajectory of the damaged speech has an acoustic likelihood of 85% with "OceanExplorer Mayday", which is much greater than the set alarm threshold (e.g., 60%). Finally, the system sends a structured alarm text containing "OceanExplorer Mayday, 412888999, in distress" to the maritime command center through the communication output interface 320.

[0130] As can be seen from the above specific application examples, this invention successfully constrained and rescued distorted voice in a communication deadlock where both data and voice were degraded, triggering a high-confidence and effective rescue alarm.

[0131] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for automatic monitoring of multi-standard coastal radio stations using integrated signal analysis, characterized in that, The method includes: S10. Obtain the multi-standard baseband I / Q signal sequence received by the radio frequency front-end of the coastal radio station, and separate and extract the control channel residual bit stream and voice communication service channel audio characteristic waveform of the multi-standard baseband I / Q signal sequence. S20. Perform physical layer signal quality analysis on the multi-standard baseband I / Q signal sequence and calculate the phase continuity parameter used to characterize the degree of channel fading; S30. When the phase continuity parameter is lower than the set channel quality threshold, the feature extraction of the control channel residual bit stream is performed by a pre-configured soft decision decoder to obtain high-confidence residual signaling features. S40. The high-confidence residual signaling features are mapped and compared with a preset local maritime database that stores ship identification information and associated attributes. The successfully matched ship identification information is extracted, and the ship name and communication text associated with the ship identification information are extracted to dynamically construct a scenario-customized acoustic micro-dictionary. S50. In the constrained search space of the scenario-customized acoustic micro-dictionary, calculate the acoustic likelihood of the audio feature waveform of the voice communication service channel with the scenario-customized acoustic micro-dictionary. When the acoustic likelihood is greater than the set alarm threshold, generate and output a structured alarm text containing the ship identification information.

2. The automatic monitoring method for multi-standard coastal radio stations based on fusion signal analysis according to claim 1, characterized in that, The step S10 of extracting the audio feature waveform of the voice communication service channel includes: S11. Use the set endpoint detection algorithm to locate the silent segment without speech in the baseband audio signal and extract the background noise spectrum envelope of the current environment. S12. Using the background noise spectrum envelope as a reference, the tap coefficient vector of the digital filter is iteratively updated using the set adaptive filtering formula. S13. The digital filter after state convergence is used to perform digital cancellation filtering on the continuous effective speech waveform to output an enhanced audio signal with background noise removed. S14. The enhanced audio signal is subjected to frame division and discrete cosine transform processing to reconstruct the audio feature waveform of the voice communication service channel.

3. The automatic monitoring method for multi-standard coastal radio stations based on fusion signal analysis according to claim 1, characterized in that, Step S20 includes: S21. Window and frame the multi-standard baseband I / Q signal sequence, extract the in-phase component I(n) and quadrature component Q(n), and use the formula... Calculate the instantaneous phase angle of the nth sampling point; S22. Obtain the power delay spectrum of the multi-standard baseband I / Q signal sequence, and use the formula... Calculate the mean of multipath delay spread ,in For the multipath delay of the k-th path, For the corresponding power, This is the average delay; S23. Extract the total number of sampling points N and the set environmental attenuation coefficient. Using the formula Initial parameters are obtained by performing cascade operations; S24. Input the initial parameters into a set Kalman filter for state recursive estimation and smoothing, and output the phase continuity parameter after state convergence. .

4. The automatic monitoring method for multi-standard coastal radio stations based on fusion signal analysis according to claim 1, characterized in that, Step S30 includes: S31. Obtain the original burst data packets that have not passed the cyclic redundancy check in the residual bit stream of the control channel, and construct the state transition mesh diagram of the soft decision decoder for the original burst data packets. S32. Based on the state transition mesh diagram, calculate the forward state metric value of each state node in the mesh. Branch metric and backward state metric ; S33, Using formulas Calculate input bits The log-likelihood ratio, where and These are the state transition sets for inputs equal to 1 and 0, respectively. S34. Extract multiple consecutive bit segments whose absolute values ​​of the log-likelihood ratios are all greater than the set single-bit likelihood threshold, and assemble the bit segments into the high-confidence residual signaling feature.

5. The automatic monitoring method for multi-standard coastal radio stations based on fusion signal analysis according to claim 1, characterized in that, Step S40 includes: S41. Identify the maritime mobile communication service identification code fragment in the high-confidence residual signaling features, and use the maritime mobile communication service identification code fragment as the key value of the search engine. S42. In the preset local maritime database, the search engine key value is compared with the complete identification code of the registered vessel using regular expressions to filter and obtain a set of candidate vessel identifications and the associated vessel name and communication text. S43. Input the ship name and communication text into a set character-to-phoneme conversion neural network, and output the phoneme probability distribution matrix and hidden Markov model emission probability for each ship name and communication text. S44. Compile the phoneme probability distribution matrix and the emission probability of the hidden Markov model into a graph network to generate a state transition constraint topology graph, and load and update it into the scenario-customized acoustic micro-dictionary.

6. The automatic monitoring method for multi-standard coastal radio stations based on fusion signal analysis according to claim 5, characterized in that, Following step S42, the following is also included: S421. Extract the spatial radio frequency data collected by the antenna array, and calculate the estimated spatial angle of arrival value using the set spatial spectrum estimation model. ; S422. Extract the area coordinate data bound to each candidate vessel in the candidate vessel identifier set, and calculate the theoretical azimuth of the k-th candidate vessel relative to the coastal radio station. ; S423, Using the formula Calculate the spatial absolute angle deviation parameter corresponding to the k-th candidate ship. ; S424, when the spatial absolute angle deviation parameter When the angle exceeds the preset multipath tolerance threshold, the corresponding vessel will be removed from the candidate vessel identifier set, and the search capacity of the scenario-customized acoustic micro-dictionary will be reduced accordingly.

7. The automatic monitoring method for multi-standard coastal radio stations based on fusion signal analysis according to claim 1, characterized in that, Step S50 includes: S51. Obtain a feature sequence of length I corresponding to the audio feature waveform of the voice communication service channel. And the state reference sequence of length J corresponding to the scenario-customized acoustic micro-dictionary. ; S52. Extract the average log-likelihood ratio of the high-confidence residual signaling features. Using the formula Calculate the width w of the dynamically warped window, where and The set adjustment coefficient; S53. Within the restricted area of ​​the dynamically adjusted window width, use the formula Calculate the cumulative normalization cost matrix, where The local distance is used to extract the global endpoint regularization cost D(I,J); S54. Using the formula The global endpoint normalization cost D(I,J) is inverted and mapped to the acoustic likelihood S, where This is the set normalized bias constant.

8. The automatic monitoring method for multi-standard coastal radio stations based on fusion signal analysis according to claim 1, characterized in that, The structured alert text output in step S50 includes: S55. Extract the first timestamp data of the demodulated high-confidence residual signaling features and the second timestamp data of the context-customized acoustic micro-dictionary; S56. Calculate the absolute value of the time difference between the first timestamp data and the second timestamp data. When the absolute value of the time difference is less than the set synchronization window threshold, determine that the current cross-modal alarm event satisfies the spatiotemporal consistency constraint. S57. For events that meet the spatiotemporal consistency constraints, extract the matching ship identification information and the parsed emergency alarm string, and encapsulate them into structured data fields according to the set maritime protocol specifications. S58. Use the set secure hash algorithm to digitally sign the structured data field, and append the signature to the end of the data packet to generate the structured alarm text to be sent.

9. A system for an automatic watchkeeping method for multi-standard coastal radio stations applied to the fusion signal analysis described in any one of claims 1 to 8, characterized in that, The system includes: The data acquisition module is used to acquire the multi-standard baseband I / Q signal sequence received by the radio frequency front-end of the coastal radio station, and to separate and extract the control channel residual bit stream and voice communication service channel audio characteristic waveform of the multi-standard baseband I / Q signal sequence. The signal analysis module is used to perform physical layer signal quality analysis on the multi-standard baseband I / Q signal sequence and calculate the phase continuity parameter used to characterize the degree of channel fading. The soft-decision feature module is used to extract features from the control channel residual bit stream through a pre-configured soft-decision decoder when the phase continuity parameter is lower than the set channel quality threshold, so as to obtain high-confidence residual signaling features. The micro-dictionary construction module is used to map and compare the high-confidence residual signaling features with a preset local maritime database that stores ship identification information and associated attributes, extract the successfully matched ship identification information, and extract the ship name and communication text associated with the ship identification information to dynamically construct a scenario-customized acoustic micro-dictionary. The dimension reduction constraint matching module is used to calculate the acoustic likelihood of the audio feature waveform of the voice communication service channel with the context-customized acoustic micro-dictionary in the constraint search space of the context-customized acoustic micro-dictionary. When the acoustic likelihood is greater than the set alarm threshold, a structured alarm text containing the ship identification information is generated and output.

10. The multi-standard coastal radio station automatic monitoring system with fused signal analysis according to claim 9, characterized in that, The signal analysis module includes: The baseband framing unit is used to window and frame the multi-standard baseband I / Q signal sequence, extract the in-phase component I(n) and quadrature component Q(n), and use the formula... Calculate the instantaneous phase angle of the nth sampling point; The delay spread estimation unit is used to obtain the power delay spectrum of the multi-standard baseband I / Q signal sequence, using the formula... Calculate the mean of multipath delay spread ,in For the multipath delay of the k-th path, For the corresponding power, This is the average delay; The penalty cascaded integrator is used to extract the total number of sampling points N and the set environmental attenuation coefficient. Using the formula Initial parameters are obtained by performing cascade operations; The Kalman filter unit is used to input the initial parameters into a set Kalman filter for recursive state estimation and smoothing, and output the phase continuity parameter after state convergence. .

Images