False alarm suppression method for mobile satellite terminal direction finding based on multi-dimensional feature joint screening

Abstract

The present application belongs to the technical field of wireless communication signal processing, and particularly relates to a mobile satellite terminal direction finding false alarm suppression method based on multi-dimensional feature joint screening, aiming to solve the technical problems in the prior art that a single power dimension is used for signal judgment, the uplink signal and the downlink signal on the mobile satellite system cannot be distinguished, and the judgment threshold is fixed and cannot adapt to complex electromagnetic environments. By constructing a multi-dimensional signal feature library, receiving and preprocessing the front-end signal, extracting multi-dimensional features in parallel, calculating the time domain, frequency domain, modulation domain and feature code domain confidence based on the multi-dimensional signal feature library, adaptively weighting and fusing based on information entropy, automatically adjusting the weight calculation and fusion score of each feature dimension according to the real-time electromagnetic environment complexity, and realizing the closed-loop optimization and feedback adjustment of the judgment threshold based on the dynamic threshold adaptive mechanism of the Q learning algorithm, the multi-dimensional joint screening signal is realized, the uplink and downlink signals are accurately distinguished by combining the uplink and downlink signal accurate distinguishing mechanism, and the direction finding false alarm is effectively suppressed.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication signal processing technology, and particularly relates to a method for suppressing false alarms in direction finding of mobile satellite terminals based on multi-dimensional feature joint screening. Background Technology

[0002] Mobile satellite systems operate in the L-band and play an irreplaceable role in areas such as communication support, ocean shipping monitoring, and IoT data transmission. This has created an urgent need for precise monitoring and direction finding of mobile satellite terminal signals, making this technology a key support for battlefield electromagnetic situational awareness, precision target engagement, and intelligence gathering.

[0003] However, the increasingly complex electromagnetic environment in which mobile satellite terminals operate presents unprecedented technical challenges to direction-finding systems. Within the aforementioned frequency bands, there are multiple adverse factors, including interference signals from other communication systems operating in the same band, adjacent-channel leakage signals, environmental thermal noise, and signal fading caused by multipath effects. These complex factors intertwine, severely weakening the accuracy of traditional direction-finding equipment. Even more problematic is the frequent occurrence of "false alarms" in actual combat, where the system triggers direction finding but the target does not exist. This high frequency of false alarms not only results in a significant waste of manpower and resources but also severely undermines the operational personnel's trust in the direction-finding system and its combat effectiveness.

[0004] From a technological evolution perspective, the earliest and most similar technical solutions in the industry today employ a relatively simple signal detection approach: In this approach, the receiver continuously scans the entire frequency band, performing real-time energy detection at each frequency point. When the detected signal power exceeds a preset threshold, that frequency point is identified as a suspected target signal, and the subsequent direction-finding process is immediately triggered. This single-dimensional decision-making method based on power thresholds has shown some effectiveness in early application scenarios where the electromagnetic environment was relatively clean. Its design principle is based on the fundamental assumption that real useful signals must exhibit power characteristics higher than background noise, and that effective separation of signal and noise can be achieved by setting a reasonable power threshold.

[0005] However, with the continuous expansion of application scenarios and the increasing complexity of the electromagnetic environment, the inherent limitations of the above-mentioned technical solutions at the principle level have gradually become apparent, and deep-seated technical defects have been exposed in multiple dimensions.

[0006] First, the scheme uses a single decision dimension, relying solely on the physical quantity of signal power for discrimination. It is completely unable to distinguish between the uplink signal of a mobile satellite terminal with similar power characteristics and other communication signals or man-made interference sources existing in the same frequency band. This essentially confuses the two completely different technical concepts of "signal presence" and "target presence".

[0007] Secondly, the scheme fails to effectively distinguish between uplink and downlink signals of a mobile satellite system. Since the transmission power of the satellite downlink is usually significantly higher than that of the terminal uplink, under the same power threshold setting, the downlink signal is very easy to be misjudged as the terminal target signal, which leads to the direction finding result pointing to the sky rather than the actual ground target. This directional error is both concealed and dangerous in actual combat.

[0008] Furthermore, this scheme uses a fixed decision threshold, which cannot adaptively adjust to real-time changes in the electromagnetic environment. In strong interference environments such as urban areas, the false alarm rate rises sharply, while in weak signal environments, the threshold may be too high, leading to missed detections. This dilemma fully exposes the fundamental flaws of the fixed threshold strategy in terms of environmental adaptability. In addition, this scheme does not utilize the unique feature codes of the signal for secondary verification, lacking technical means to confirm authenticity from the perspective of subtle signal features. It also lacks a multi-feature adaptive fusion mechanism, making it unable to dynamically adjust the decision weights of each dimension according to environmental changes.

[0009] Traditional solutions oversimplify the complex multivariate problem of "signal detection" into a single power threshold comparison problem, neglecting the rich signal characteristics inherent in mobile satellite systems. In reality, mobile satellite uplink signals exhibit specific burst pulse durations and time slot occupancy patterns in the time domain, clearly defined operating frequency bands and channel bandwidths in the frequency domain, and employ π / 4-DQPSK modulation with unique phase transition patterns in the modulation domain. More importantly, they contain uplink-specific frame synchronization words and pilot sequences in the feature code domain. These characteristics collectively form the physical basis for distinguishing genuine terminal signals from other interference signals. However, existing technologies ignore these inherent characteristics, relying solely on power as the determining factor. Logically, this is tantamount to acknowledging the obviously flawed assumption that "all signals above the threshold are target signals."

[0010] Furthermore, complex electromagnetic environments place higher demands on signal detection technology: the system needs to have the ability to jointly screen multi-dimensional features to fundamentally suppress false alarms; the system also needs to have the ability to distinguish between uplink and downlink signals to avoid misjudging satellite downlink signals as ground terminals; and the system needs to have environmental adaptability to maintain stable detection performance under different environmental conditions.

[0011] Therefore, this invention aims to construct a technical solution that can jointly screen uplink signals of mobile satellite terminals from multiple dimensions such as time domain, frequency domain, modulation domain, and feature code domain, and on this basis, realize dynamic decision-making that is environmentally adaptive. Summary of the Invention

[0012] To address the shortcomings of existing technologies, such as using a single power dimension for signal decision-making, inability to distinguish between uplink and downlink signals in mobile satellite systems, fixed decision thresholds that cannot adapt to complex electromagnetic environments, and lack of multi-feature adaptive fusion mechanisms, this invention provides a method for suppressing false alarms in direction finding of mobile satellite terminals based on multi-dimensional feature joint screening. This method jointly extracts signal features from four dimensions: time domain, frequency domain, modulation domain, and feature code domain. It then uses an information entropy-driven adaptive weighted fusion algorithm for feature fusion and combines it with a dynamic threshold adaptive mechanism based on the Q-learning algorithm to achieve closed-loop optimization of the decision threshold. This fundamentally suppresses false alarms in direction finding and improves the reliability of the system in complex electromagnetic environments.

[0013] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: Firstly, a method for suppressing false alarms in direction finding of mobile satellite terminals based on joint screening of multi-dimensional features is provided, including the following steps: S1: Construct a multi-dimensional signal feature library: Construct a multi-dimensional signal feature library including time-domain features, frequency-domain features, modulation-domain features, and feature code-domain features. The template matching in the multi-dimensional signal feature library is used in S3 for parallel extraction of multi-dimensional features. The confidence scores in the time domain, frequency domain, modulation domain, and feature code domain are calculated respectively. S2: Front-end signal reception and preprocessing: The signal received by the antenna is amplified by a low-noise amplifier, filtered by a filter, and down-converted to intermediate frequency by a downconverter. It is then sampled by an ADC. The sampled digital signal is sent to the FPGA for preprocessing, including digital downconversion processing, frequency offset coarse compensation processing, and time slot synchronization and burst positioning processing, to obtain the preprocessed signal. S3: Parallel Extraction of Multidimensional Features and Preliminary Confidence Calculation: The preprocessed signal enters a parallel extractor of multidimensional features implemented in the FPGA PL domain. Template matching is performed based on a multidimensional signal feature library to calculate the original confidence vectors in four dimensions: time domain, frequency domain, modulation domain, and feature code domain. s ; Original confidence vector s =[ s t ,s f , s m ,s c ], s t For time-domain confidence; s f Frequency domain confidence level; s m For the modulation domain confidence level; s c , which represents the confidence level of the feature code field.

[0014] S4: Adaptive weighted fusion and dynamic threshold decision based on information entropy: Real-time assessment of electromagnetic environment complexity and generation of environmental state feature vectors. x =[ E,SNR est , Δ f doppler ] T ; Where E represents the real-time electromagnetic environment complexity; SNR est For real-time signal-to-noise ratio; Δ f doppler This is the Doppler frequency offset estimate.

[0015] Based on the original confidence vector s The adaptive weights of time-domain features, frequency-domain features, modulation-domain features, and feature code-domain features are calculated using information entropy theory, and then the weighted fusion score is calculated. Dynamic threshold adaptive adjustment is performed based on environmental state feature vectors using the Q-learning algorithm; The decision is obtained by comparing the weighted fusion score with the dynamic threshold: If the judgment result is a valid signal, proceed to direction finding, obtain the direction finding result, and make a reasonable judgment on the direction finding result; Otherwise, the judgment result is considered invalid and discarded.

[0016] Preferably, the specific process of digital down-conversion processing in step S2 is as follows: The intermediate frequency digital signal sampled by the ADC With the positive local oscillator signal generated by the local digitally controlled oscillator NCO and negative oscillator signal Multiplying yields the baseband in-phase component. and orthogonal components ,in The local oscillator frequency, The sampling period is For sampling point index; Output complex baseband signal ,in j It is the imaginary unit.

[0017] Preferably, the specific process of the frequency offset coarse compensation processing is as follows: Using the Fast Fourier Transform (FFT) method to analyze complex baseband signals Frequency offset estimation is performed by calculating the amplitude spectrum of the signal in the frequency domain and finding the frequency point corresponding to the amplitude peak. As a coarse estimate, the baseband complex signal after coarse frequency offset compensation is then obtained by compensating through the multiplier signal. .

[0018] Preferably, the specific synchronization process of time slot synchronization and burst location processing in step S2 is as follows: Local matched filtering: pre-stores the waveform template corresponding to the frame header synchronization word sequence of the uplink time slot. temperature ( m ), m =1,2,..., M 1; of which M 1 represents the length of the waveform template; m Index of sequence elements; Sliding correlation: Based on the sliding correlation function R ( k )right and temperature ( m Perform sliding cross-correlation calculations; Peak detection: finding | R ( k Peak position of )| k peak When the peak position k peak The corresponding peak value | R ( k peak Exceeding the preset synchronization threshold Th sync When this happens, it is determined that the frame header has been successfully captured, where | R ( k )| indicates R ( k The magnitude, i.e., the amplitude; Burst localization: Based on a fixed offset Δ between the known frame header and the starting position of the burst data. off set Determine the burst start position of the burst pulse signal that was successfully captured in the frame header. k start The burst termination position is determined based on the time slot length. k end .

[0019] Preferably, the original confidence vector in step S3 s The specific calculation process for the time-domain confidence component is as follows: Based on the burst initiation position obtained in step S2 k start and the location of the sudden end k end Calculate the measured burst pulse width W meas ; Measured burst pulse width W measCompared with the standard time slot pulse width in the feature template library W std By comparing the two, we obtain the absolute value of the difference, i.e., calculate | W meas -W std | The smaller the difference, the higher the time-domain feature matching degree, which is used for subsequent time-domain confidence calculation; Calculate the time-domain confidence level based on the absolute value of the difference between the two: ; in W tol This is the time error tolerance.

[0020] Preferably, the specific calculation process of the frequency domain confidence component in the original confidence vector s in step S3 is as follows: Perform an FFT on the extracted burst signal to obtain its power spectral density. P ( f ); Find the frequency point corresponding to the peak of the power spectrum f peak As a center frequency estimate ; Calculate the instantaneous bandwidth of the signal On the power spectrum, from the frequency point f peak Searching to the left and right respectively, when the descent distance is Δ P BW When points A and B are obtained, the difference between the frequency values ​​corresponding to points A and B is... ; Calculate frequency domain confidence s f : ; in The center value of the operating frequency band. For standard bandwidth, and These are the standard deviations of frequency matching and bandwidth matching, respectively.

[0021] Preferably, the specific calculation process of the modulation domain confidence component in the original confidence vector s in step S3 is as follows: Perform on signal π / 4-DQPSK demodulation extracts a continuous phase difference sequence {Δ i meas ( l ), l =1,2,..., L -1};where L is the total length of the phase difference sequence; The measured phase difference sequence is compared with the standard phase transition template {Δ i ideal ( l Perform relevant matching; Calculate the confidence level of the modulation domain s m : ; in l The index is the phase difference sequence index; L is the total length of the phase difference sequence; Δ i meas ( l ) is the first l The actual phase difference obtained by demodulation between adjacent symbols; Δ i ideal ( l () represents the ideal standard phase jump value.

[0022] Preferably, the original confidence vector in step S3 s The specific calculation process for the confidence component in the feature code field is as follows: Pre-stored frame synchronization word pilot sequence for uplink of mobile satellite system C up : C up =[ c 1, c 2,..., c i ,..., c M ], where the sequence number is 1 i individual code element c i ∈{±1}, M for C up Total code length; The baseband complex signal after coarse compensation for frequency offset of the received signal s comp ( n Starting from the initial position of the burst, sliding correlation matching is performed to obtain the sliding cross-correlation output complex value. R c ( k ); Output complex values ​​based on sliding cross-correlation R c ( k Calculate the confidence level of the feature code field. s c : ; in,M for C up Total code length.

[0023] Preferably, the specific process of step S4 is as follows: S41: Define the real-time electromagnetic environment complexity E by analyzing background noise power. P n Signal density per unit time r and historical false alarm rate FAR hist Perform comprehensive characterization and generate environmental state feature vectors; S42: Based on the original confidence vector s Calculate the first i Normalized confidence entropy of dimensional features in the current environment H i ; Calculate the first i Dimensional adaptive weights w i ; S43: Based on adaptive weights w i Obtain weighted fusion score Score ; S44: Dynamic threshold adaptive adjustment based on environmental state feature vectors using the Q-learning algorithm; S45: Based on weighted fusion score Score The result is obtained by comparing with the dynamic threshold and then output: like Score≥Th If the result is valid, the direction finding process begins and the direction finding result is obtained. in Th As a threshold for judgment; Otherwise, the judgment result is considered an invalid signal and is discarded; S46: Feed the direction finding results back to the Q-learning algorithm module to form a closed-loop optimization.

[0024] Secondly, the mobile satellite terminal direction finding false alarm suppression method based on multi-dimensional feature joint screening feeds the direction finding results back to the Q-learning algorithm module, forming a closed-loop optimization mechanism as follows: The logic for determining a reasonable orientation and the method for awarding positive rewards: Judgment logic: The direction finding system will output the target's azimuth angle. and pitch angle ; Reasonableness Standard 1: For ground or low-altitude terminals, its pitch angle Below the upper limit That is, pointing towards low altitude; Reasonableness Criterion 2: If the rate of change of azimuth and the rate of change of elevation are less than 100° / s when the system continuously performs direction finding on the same target; Reasonableness Standard 3: Combining prior geographical information, i.e., the polygonal coordinates of the permitted area, starting from the location of the direction finding station, along the direction finding azimuth angle... i Draw a ray and calculate whether the ray intersects with the boundary of the permitted polygon area. If there is at least one intersection, and the distance from that intersection to the ground of the direction finding station is within a preset ranging range, the preset ranging range is […]. R min , R max ], R min ,R max These are the lower and upper limits for distance measurement, respectively. Positive reward is given when any of the above reasonableness criteria are met. The Q-learning algorithm module considers the current decision to be a valid signal and generates a positive reward. The logic for determining abnormal directions and the methods for penalty or threshold adjustment: Decision logic: Abnormal Standard 1: Continuous N abnormal The signals deemed valid in the second determination all point to high altitudes in their direction-finding results. ; Anomaly Criterion 2: The spatial spectrum energy of the direction finding output has no obvious single peak or main peak, that is, the ratio of the peak value to the secondary peak value of the spatial spectrum is less than 3dB, or the spatial spectrum energy corresponding to the peak value is less than 6dB lower than the average energy of the whole frequency band. Anomaly Criterion 3: For a signal believed to be from the same terminal, if the absolute value of the change in azimuth angle of the direction finding result exceeds 150° or the absolute value of the change in elevation angle exceeds 120° within a preset time, an abnormal jump occurs, and the abnormal jump does not show a monotonic trend in three consecutive direction finding results. Penalties and increased thresholds: If any of the above abnormal criteria are met, the system determines that a false alarm has occurred. Punishment: Generates negative rewards; Threshold adjustment method: Trigger: Calculate the false alarm rate within the sliding window. FAR short ,like FAR short If the target limit is exceeded, or if three false alarms occur consecutively, the threshold will be raised. Adjustment amount: Threshold adjustment action selection a =+ d Where δ is the threshold adjustment step size; Execution: Update judgment threshold Th for minutes ( Th + d , Th max ); in Th max This is the preset threshold upper limit.

[0025] This invention also includes a mechanism for accurate differentiation of uplink and downlink signals: confidence level of the feature code domain in step three. s c In the calculation, the frame synchronization word or pilot sequence unique to the uplink of the mobile satellite system is pre-stored, and the feature code template of the downlink is also pre-stored. The degree of matching between the received signal and the uplink feature code template is compared to confirm whether it is an uplink signal, and the degree of matching between the received signal and the downlink feature code template is compared to exclude whether it is a downlink signal. Only signals with a high degree of matching with the uplink feature code and a low degree of matching with the downlink feature code are judged as possible terminal uplink signals.

[0026] In step S2, the sampling rate of the ADC satisfies the Nyquist sampling theorem, and the sampling rate is at least twice the highest frequency of the signal.

[0027] In step S3, the multidimensional feature parallel extractor is specifically implemented in the FPGA using a parallel pipeline architecture, where feature extraction in four dimensions is performed simultaneously to ensure the real-time performance of signal processing.

[0028] The beneficial effects of this invention include: A multi-dimensional feature joint screening mechanism is adopted to comprehensively judge whether the signal is an uplink signal of a mobile satellite terminal from four dimensions: time domain, frequency domain, modulation domain, and feature code domain, which effectively suppresses the false alarm problem caused by the decision based on a single power dimension. By utilizing the inherent differences in the signature codes of uplink and downlink signals from mobile satellite systems, an accurate distinguisher is constructed to avoid misidentifying downlink satellite signals as terminal uplink signals, thereby reducing the probability of misidentification. The information entropy-driven dynamic adaptive weighted fusion algorithm automatically adjusts the weights of each feature dimension according to the complexity of the real-time electromagnetic environment, enabling the system to maintain stable detection performance in complex environments such as low signal-to-noise ratio and strong interference. The dynamic threshold adaptive mechanism based on the Q-learning algorithm realizes the closed-loop online optimization of the decision threshold, enabling the system to have self-evolution capability and continuously learn and optimize its own parameters during long-term operation; By employing a closed-loop processing architecture that incorporates multi-dimensional front-end filtering, adaptive weighted fusion, and dynamic threshold decision-making, the computational load of invalid data is reduced, thereby improving the timeliness of direction finding and the reliability of the system. Attached Figure Description

[0029] Figure 1 This is a flowchart illustrating the method for suppressing false alarms in direction finding of mobile satellite terminals based on multi-dimensional feature joint screening according to the present invention.

[0030] Figure 2 This is another flowchart illustrating the mobile satellite terminal direction finding false alarm suppression method based on multi-dimensional feature joint screening according to the present invention. Detailed Implementation

[0031] The following is in conjunction with the appendix Figure 1~Figure 2 To further illustrate this invention in detail, for ease of understanding of the formulas, the following conventions are made for commonly used parameters: n : Discrete sampling point index, taking a non-negative integer; k : Slide the relevant position index and take a non-negative integer; m : Template or sequence element index, taken as a positive integer; i Feature dimension index i ∈{ t, f, m, c}, respectively corresponding to the time domain ( time ), frequency domain ( frequency ), modulation domain ( modulation ) and feature code field ( code ); j : Imaginary unit; *: Complex conjugate operator; T s ADC sampling period T s =1 / f s ,in f s The sampling rate is denoted by ; all confidence values ​​are dimensionless and range from [0, 1].

[0032] Example 1 See appendix Figure 1 and 2 As shown, the method for suppressing false alarms in direction finding of mobile satellite terminals based on multi-dimensional feature joint screening includes the following steps: S1: Constructing a multi-dimensional signal feature library: The multi-dimensional signal feature library includes time-domain features, frequency-domain features, modulation-domain features, and feature code-domain features. Time-domain features include burst pulse duration, time slot occupancy patterns, and frame period; frequency-domain features include a single-channel bandwidth of 41.667kHz and the Doppler frequency offset range; modulation-domain features include π / 4-DQPSK modulation scheme and its unique phase transition patterns; feature code-domain features include uplink-specific frame synchronization words, pilot sequences, and access prefix codes; each feature is stored in the FPGA's BRAM in the form of feature templates.

[0033] The multi-dimensional signal feature library constructed in step S1 is used in S3 for template matching during parallel extraction of multi-dimensional features, and the confidence scores in the time domain, frequency domain, modulation domain and feature code domain are calculated respectively.

[0034] In terms of temporal characteristics, the uplink signals of mobile satellite systems exhibit specific burst pulse durations, time slot occupancy patterns, and frame periods. These temporal characteristics are obtained through long-term monitoring and statistical analysis, forming standardized temporal characteristic templates. Specific temporal characteristic parameters include: standard time slot pulse width. W std Set to 8.28ms; Time error tolerance W tol The frame period is set to 0.5ms; the allowable burst pulse width range is set to 7.78ms~8.78ms. In terms of frame period, the mobile satellite system adopts the TDMA system, with each frame period being 90ms, divided into 8 time slots, each time slot lasting 11.25ms, of which burst pulse data occupies approximately 8.28ms.

[0035] In terms of frequency domain characteristics, the mobile satellite system uplink operates in the L-band, with a single-channel bandwidth of 41.667 kHz. Because the mobile satellite system operates in low Earth orbit, there is a significant Doppler frequency offset between the terminal and the satellite, ranging from -40 kHz to +40 kHz. In the feature library, the standard center frequency... f std Set to 1620.0MHz, standard bandwidth B std Set to 41.667kHz, frequency matching standard deviation s f Set to 5kHz, bandwidth matching standard deviation s B Set to 3kHz.

[0036] Regarding modulation domain characteristics, the mobile satellite system uplink employs π / 4-DQPSK modulation. This modulation scheme has unique phase transition patterns, with a phase transition value set of {±π / 4, ±3π / 4}. Standard phase transition templates are stored in the feature library for subsequent modulation domain confidence calculations.

[0037] In terms of feature code characteristics, the uplink of a mobile satellite system has unique frame synchronization words, pilot sequences, and access prefix codes. The frame synchronization word is a specific 32-bit binary sequence, the pilot sequence is a known PN sequence, and the access prefix code is used by the terminal to initiate an access request. The feature code sequences are pre-stored in a feature template library.

[0038] The above four types of features are stored in the FPGA's BRAM in the form of feature templates for real-time matching. The feature template library is implemented using a dual-port BRAM, supporting parallel read and write operations. The write end is used for updating and initializing feature templates, while the read end is used for real-time feature matching. The storage capacity of the BRAM is determined based on the number and length of the feature templates. Time-domain feature templates occupy approximately 1KB of storage space, frequency-domain feature templates occupy approximately 512 bytes, modulation-domain feature templates occupy approximately 2KB, and feature code-domain feature templates occupy approximately 4KB, totaling approximately 8KB of storage capacity, which is sufficient to meet the storage requirements of the feature templates.

[0039] S2: Front-end signal reception and preprocessing: The signal received by the antenna is amplified by a low-noise amplifier, filtered by a filter, and down-converted to an intermediate frequency by a downconverter. It is then sampled by an ADC, and the sampled digital signal is sent to the FPGA for preprocessing, including digital down-conversion processing, frequency offset coarse compensation processing, and time slot synchronization and burst positioning processing. The digital down-conversion process multiplies the intermediate frequency digital signal sampled by the ADC with the quadrature local oscillator signal generated by the local digitally controlled oscillator (NCO) to obtain the baseband in-phase component and quadrature component, and outputs a complex baseband signal. Coarse frequency offset compensation uses the FFT method to estimate and compensate for the frequency offset of the complex baseband signal, resulting in a coarsely compensated baseband complex signal. The time slot synchronization and burst positioning adopt a local matched filtering method. The compensated signal is subjected to sliding cross-correlation with the waveform template corresponding to the pre-stored frame header synchronization word sequence. When the correlation peak exceeds the preset synchronization threshold, it is determined that the frame header has been successfully captured. The burst start position is determined according to the fixed offset between the frame header and the burst data start position.

[0040] S3: Parallel Extraction of Multidimensional Features and Preliminary Confidence Calculation: The preprocessed signal enters a parallel extractor of multidimensional features implemented in the FPGA PL domain. Template matching is performed based on a multidimensional signal feature library to calculate the original confidence vector s, which includes four dimensions: time domain, frequency domain, modulation domain, and feature code domain. Original Confidence Vector s It is used for adaptive weighted fusion based on information entropy, to calculate the fusion score and compare it with a dynamic threshold for decision.

[0041] S4: Adaptive weighted fusion and dynamic threshold decision based on information entropy: Real-time assessment of electromagnetic environment complexity and generation of environmental state feature vectors x =[ E, SNR est , Δ f doppler ] T ; Where E represents the real-time electromagnetic environment complexity; SNR est For real-time signal-to-noise ratio; Δf doppler This is the Doppler frequency offset estimate.

[0042] Based on the original confidence vector s The adaptive weights of time-domain features, frequency-domain features, modulation-domain features, and feature code-domain features are calculated using information entropy theory, and then a weighted fusion score is calculated. Based on the environmental state feature vector, a dynamic threshold is adaptively adjusted using the Q-learning algorithm. The decision result is output based on the comparison between the fusion score and the dynamic threshold, and the rationality of the direction finding result is judged to form a feedback closed loop.

[0043] Example 2 Based on Example 1, the specific process of digital down-conversion processing in step S2 is as follows: The intermediate frequency digital signal sampled by the ADC With the positive local oscillator signal generated by the local digitally controlled oscillator NCO and negative oscillator signal Multiplying yields the baseband in-phase component. and orthogonal components ,in The local oscillator frequency, The sampling period is For sampling point index; Output complex baseband signal ,in j It is the imaginary unit.

[0044] The specific process of the frequency offset coarse compensation processing is as follows: Using the known maximum Doppler frequency offset range Δ of mobile satellite systems f d,max ∈[-40kHz,+40kHz] (typical value for low-Earth orbit satellites), and a coarse frequency offset estimation range is set by combining the receiver's own frequency error.

[0045] Using the FFT method Frequency offset estimation is performed by calculating the amplitude spectrum of the signal in the frequency domain and finding the frequency point corresponding to the amplitude peak. As a coarse estimate, the signal is then compensated using a multiplier to obtain the baseband complex signal after coarse frequency offset compensation: .

[0046] The specific synchronization process of time slot synchronization and burst location processing in step S2 is as follows: The mobile satellite system uplink signal uses a 90ms frame structure, with each time slot being 90ms long. N slot , N slotThis represents the number of time slots per frame, and can be set to 8.

[0047] Local matched filtering: Waveform template corresponding to the pre-stored frame header synchronization word sequence (known fixed bit pattern) of the uplink time slot. temperature ( m ), m =1,2,..., M 1; of which M 1 represents the length of the waveform template; m Index of sequence elements; Sliding correlation: Based on the sliding correlation function R ( k )right and temperature ( m Perform sliding cross-correlation calculations; correlation function R ( k )for: R ( k )=∑ M m=1 s comp ( k + m )⋅ temperature ( m )∗; Peak detection: finding | R ( k peak position of )| k peak When the peak position k peak The corresponding peak value | R ( k peak Exceeding the preset synchronization threshold Th sync When this happens, it is determined that the frame header has been successfully captured, where | R ( k )| indicates R ( k The magnitude of the amplitude; Burst localization: Based on a fixed offset Δ between the known frame header and the starting position of the burst data. off set Determine the burst start position of the burst pulse signal. k start : k start = k peak +Δ off set The burst termination position is determined based on the time slot length. k end.

[0048] The specific calculation process of the time-domain confidence component in the original confidence vector s in step S3 is as follows: Time domain confidence s t Used to evaluate the degree of matching between the burst structure of a signal and the uplink time slot structure of a mobile satellite.

[0049] Based on the burst initiation position obtained in step S2 k start and the location of the sudden end k end Calculate the measured burst pulse width W meas =( k end - k start )× T s ; Measured burst pulse width W meas Compared with the standard time slot pulse width in the feature template library W std By comparing the two, we obtain the absolute value of the difference, i.e., calculate | W meas -W std The smaller the difference, the higher the temporal feature matching degree; the temporal confidence degree is calculated based on the absolute value of the difference between the two. s t : ; in W std The standard burst pulse duration; W tol The time error tolerance can be set to 0.5ms to account for multipath effects and synchronization errors in real-world environments.

[0050] The specific calculation process of the frequency domain confidence component in the original confidence vector s in step S3 is as follows: Frequency domain confidence s f Used to assess whether the center frequency and bandwidth of a signal meet the mobile satellite uplink standards.

[0051] Perform an FFT on the extracted burst signal to obtain its power spectral density. P ( f ); Find the frequency point corresponding to the peak of the power spectrum f peak As a center frequency estimate ; Calculate the instantaneous bandwidth of the signal On the power spectrum, from the frequency point f peak Searching to the left and right respectively, when the descent distance is Δ P BW When points A and B are obtained, the difference between the frequency values ​​corresponding to points A and B is... ; Calculate the frequency domain confidence score: ; in f std The center value of the operating frequency band can be set. f std =1620.0 MHz; s f The frequency matching standard deviation can be set to... s f =5kHz; B std Standard bandwidth can be set. B std =41.667 kHz; s B To match the standard deviation of bandwidth, it can be set to 3kHz.

[0052] The specific calculation process of the modulation domain confidence component in the original confidence vector s in step S3 is as follows: Modulation domain confidence s m Used to evaluate whether the phase transition pattern of a signal conforms to the π / 4-DQPSK standard.

[0053] Perform on signal π / 4-DQPSK demodulation extracts a continuous phase difference sequence: {Δ i meas ( l ), l =1,2,..., L -1}; Ideally, Δ i ideal ∈{± π / 4,±3 π / 4}; The measured phase difference sequence is compared with the standard phase transition template {Δ i ideal ( l Relevant matching is performed, and the standard phase transition template can be obtained by locally demodulating a known pilot sequence; Calculate the confidence level of the modulation domain sm : ; in l The index is the phase difference sequence index; L is the total length of the phase difference sequence; Δ i meas ( l ) is the first l The actual phase difference obtained by demodulation between adjacent symbols; Δ i ideal ( l () represents the ideal standard phase jump value.

[0054] This formula means that by calculating the cosine value of the phase difference error, which is sensitive to error and has an output range of [-1, 1], and then linearly mapping the result to the [0, 1] interval, the final result is obtained. s m Ideal match s m =1, when the phase is completely random s m ≈0.

[0055] The specific calculation process of the feature code domain confidence component in the original confidence vector s in step S3 is as follows: Confidence of feature code domain s c Used to verify whether a unique feature code sequence of the mobile satellite uplink exists in the signal.

[0056] Pre-stored frame synchronization word pilot sequence for uplink of mobile satellite system C up : C up =[ c 1, c 2,..., c i ,..., c M ], where the sequence number is 1 i individual code element c i ∈{±1}, M for C up Total code length; The baseband complex signal after coarse compensation for frequency offset of the received signal s comp ( n On the above, starting from the initial position of the outbreak, perform sliding correlation matching: ; in Rc ( k () represents the complex value output by the sliding cross-correlation; k For the sliding window offset index; i for C up Internal symbol index; C up ( i )for C up The i Each code element; n start This is the starting sampling point number for the burst signal segment to be detected, which is the starting point of a signal window; s comp (·) represents the baseband complex signal after coarse frequency offset compensation, and the parentheses contain the corresponding sampling point number; Output complex values ​​based on sliding cross-correlation R c ( k Calculate the confidence level of the feature code field. s c : ; This formula means: normalize the maximum correlation peak value to the [0, 1] interval; s c =1 indicates a complete match. s c =0 indicates complete uncorrelation, which is the basis for distinguishing between uplink and downlink signals and random noise.

[0057] Example 3 Based on Example 1 or Example 2, the specific process of step S4 is as follows: S41: Perform real-time assessment of electromagnetic environment complexity: Define the real-time electromagnetic environment complexity E by analyzing the background noise power. P n Signal density per unit time r and historical false alarm rate FAR hist Perform comprehensive characterization: ; in P n0 For interference-free reference noise power ，α 1 ,β 1 ,c 1 is the normalization coefficient, α1+β1+γ1=1; Generate environmental state feature vectors x : x =[ E, SNR est , Δ f doppler ] T ; in SNR est For real-time signal-to-noise ratio; Δ f doppler This is the Doppler frequency offset estimate; S42: Adaptive weight allocation based on information entropy: S421: Calculate the... i Normalized confidence entropy of dimensional features in the current environment H i : ; in The moving average of the confidence scores for this feature over the most recent N frames. ϵ =10 -6 To prevent the entropy value from being a very small constant of log0, H i ∈[0,1], the closer to 1, the less reliable the feature is in the current environment; S422: Calculate the... i Adaptive weights for dimension: ; Where: λ>0 is the entropy sensitivity coefficient, which can be taken as λ=2.5; ϕ i (x) and ϕ j (x) are respectively i peacekeeping j The feature-environment matching function of a dimension represents the theoretical validity of that feature dimension in the current environment. ϕ i The specific expression for (x) is as follows: ; Where parameters a i ,b i These are parameters pre-calibrated through offline learning or expert rules; S43: Calculate the weighted fusion score Score : ; S44: Dynamic threshold adaptive adjustment based on Q-learning algorithm: The state space represents the current environment state. x =[ E,SNRest ,FAR short ] T ; in FAR short For the most recent M =50 frames of short-term false alarm rate.

[0058] The motion space is for threshold adjustment actions: a ∈{- d ,0,+ d},in d =0.05.

[0059] The reward function is: ; Among them, false alarm indication function The value is 1 when a false alarm occurs (identified as a signal but subsequent direction finding is invalid or the pointing is abnormal), otherwise it is 0; Missed detection indication function. The value is 1 when a missed detection occurs (the real signal is not detected), otherwise it is 0. or System throughput efficiency (effective signal throughput or total processing capacity); can be set as false alarm penalty weight. w FA =2.0, penalty weight for missed detection w MD =1.0, positive weight for throughput efficiency w eff =0.5.

[0060] The Q-value update expression is as follows: ; Among them, Q( x,a ) represents the state-action value function, i.e., the Q value; x represents the current environmental state of the agent; a R represents the action to be performed by the agent in state x. a Afterwards, the immediate reward obtained from the environment in one step; learning rate α =0.1, discount factor c =0.9; x' is the new state obtained after the environment jumps after executing action a; a' All actions that can be performed in the new state x'; To find the optimal future Q value: In the new state x', traverse all possible actions. a' The term that yields the maximum Q value represents the optimal long-term return that can be obtained subsequently.

[0061] S45: Output the judgment result: if Score≥Th , ThIf the threshold is set, the decision result is a valid signal, and the direction finding process begins to obtain the direction finding result. Otherwise, the judgment result is considered an invalid signal and is discarded; S46: The direction finding results are fed back to the Q-learning algorithm module to form a closed-loop optimization. The specific mechanism is as follows: The logic for determining a reasonable orientation and the method for awarding positive rewards: Judgment logic: The direction finding system will output the target's azimuth angle. and pitch angle ; Reasonableness Standard 1 (Ground Target Constraint): For ground or low-altitude terminals, its pitch angle Below the upper limit That is, pointing towards the low altitude. It can be set to 30°; Reasonableness Criterion 2 (Spatial Consistency): The system will continuously perform direction finding on the same target. If the rate of change of the azimuth angle... and the rate of change of pitch angle It conforms to the physical motion laws of mobile terminals, that is, the rate of change is less than 100° / s; Reasonableness Standard 3 (Geographical / Situational Constraints): Combining prior geographic information, i.e., the polygonal coordinates of the permitted area, starting from the location of the direction finding station, along the direction finding azimuth... i Draw a ray and calculate whether the ray intersects with the boundary of the permitted polygon area. If there is at least one intersection, and the distance from that intersection to the ground of the direction finding station is within a preset ranging range, the preset ranging range is […]. R min ,R max ], R min ,R max These are the lower and upper limits for distance measurement, respectively. Positive reward is given when any of the above rationality criteria are met. The Q-learning algorithm module considers the current decision a valid signal and generates a positive reward. Specific value: Reward r =+1.0; Basis: This reward is used to encourage the algorithm to make a correct decision using the current threshold under the current environmental state; the reward will be directly used for Q-value updates. ; The logic for determining abnormal directions and the methods for penalty or threshold adjustment: Decision logic (conditions that trigger penalties): Anomaly Standard 1 (Continuous Upper-Altitude Direction): Continuous N abnormalThe signal for which the second decision is valid can be set as follows: N abnormal =5 times, and the direction finding results all pointed to the upper sky, that is This indicates that the system misinterpreted downlink satellite signals as uplink terminal signals; Anomaly Criterion 2 (Zodiac Energy Diffusion): The spatial spectrum energy of the direction finding output has no obvious single peak or main peak, that is, the ratio of the peak value to the secondary peak value of the spatial spectrum is less than 3dB, or the spatial spectrum energy corresponding to the peak value is less than 6dB below the average energy of the entire frequency band, indicating that the input may be noise or interference, rather than a real signal.

[0062] Anomaly Criterion 3 (Source Location Conflict): For a signal believed to originate from the same terminal, if the absolute value of the azimuth change in the direction finding result exceeds 150° or the absolute value of the elevation change exceeds 120° within a preset time, an abnormal jump occurs, and the abnormal jump shows no monotonic trend in three consecutive direction finding results.

[0063] Penalty and threshold increase: Once any of the above abnormal criteria are met, the system determines that a false alarm has occurred.

[0064] Punishment: Resulting in negative rewards: r =-2.0; r According to the original reward function w FA =2.0 is set, and the Q value is updated negatively: Threshold adjustment method: Trigger: Within the sliding window (most recent M=50 decisions), calculate the false alarm rate. FAR short .like FAR short Exceeding the set target limit FAR targetmax or occur continuously K bad If a false alarm occurs, the threshold will be raised; this can be set. K bad =3, FAR targetmax =5%.

[0065] Adjustment amount: Threshold adjustment action selection a =+ d , can be set d =0.05, where δ is the threshold adjustment step size, i.e. the fixed step size for each threshold adjustment.

[0066] Execution: Update judgment threshold Th for minutes ( Th + d , Th max ); in Th max The preset threshold upper limit can be set. Th max =0.95, to prevent missed detections due to excessively high thresholds.

[0067] Effect: Raising the threshold makes it more difficult for subsequent signals to be judged as valid, thereby reducing the false alarm rate.

[0068] The Q-learning algorithm module remembers that in the current environment, taking the action of raising the threshold resulted in a negative reward, so it will tend not to take this action in similar environments in the future or adjust the strategy as needed.

Claims

1. A method for suppressing false alarms in direction finding of mobile satellite terminals based on multi-dimensional feature joint screening, characterized in that, Includes the following steps: S1: Construct a multi-dimensional signal feature library that includes time-domain features, frequency-domain features, modulation-domain features, and feature code-domain features; S2: The signal received by the antenna is amplified by a low-noise amplifier, filtered by a filter, and down-converted to intermediate frequency by a downconverter. It is then sampled by an ADC. The sampled digital signal is sent to an FPGA for preprocessing, including digital downconversion processing, frequency offset coarse compensation processing, and time slot synchronization and burst positioning processing, to obtain the preprocessed signal. S3: The preprocessed signal enters a multi-dimensional feature parallel extractor implemented in the FPGA PL domain. Template matching is performed based on a multi-dimensional signal feature library to calculate the original confidence vectors in four dimensions: time domain, frequency domain, modulation domain, and feature code domain. s ; S4: Real-time assessment of electromagnetic environment complexity and generation of environmental state feature vectors; Based on the original confidence vector s The adaptive weights of time-domain features, frequency-domain features, modulation-domain features, and feature code-domain features are calculated using information entropy theory, and then the weighted fusion score is calculated. Dynamic threshold adaptive adjustment is performed based on environmental state feature vectors using the Q-learning algorithm; The decision is obtained by comparing the weighted fusion score with the dynamic threshold: If the judgment result is a valid signal, proceed to direction finding, obtain the direction finding result, and determine the reasonableness of the direction finding result; otherwise, if the judgment result is an invalid signal, discard it. The specific process of step S4 is as follows: S41: Real-time assessment of electromagnetic environment complexity: Define real-time electromagnetic environment complexity E, and analyze background noise power. P n Signal density per unit time ρ and historical false alarm rate FAR hist Perform comprehensive characterization to generate environmental state feature vectors; S42: Adaptive weight allocation based on information entropy: S421: Calculate the... i Normalized confidence entropy of dimensional features in the current environment H i ; S422: Calculate the... i Dimensional adaptive weights w i ; S43: Based on adaptive weights w i Obtain weighted fusion score Score ; S44: Dynamic threshold adaptive adjustment based on Q-learning algorithm; S45: Output the judgment result: like Score≥Th , Th If the threshold is set, the decision result is a valid signal, and the direction finding process begins to obtain the direction finding result. Otherwise, the judgment result is considered an invalid signal and is discarded; S46: Feed the direction finding results back to the Q-learning algorithm module to form a closed-loop optimization.

2. The method for suppressing false alarms in direction finding of mobile satellite terminals based on multi-dimensional feature joint screening according to claim 1, characterized in that, The specific process of digital down-conversion processing in step S2 is as follows: The intermediate frequency digital signal sampled by the ADC With the positive local oscillator signal generated by the local digitally controlled oscillator NCO and negative oscillator signal Multiplying yields the baseband in-phase component. and orthogonal components ,in The local oscillator frequency, The sampling period is For sampling point index; Output complex baseband signal ,in j It is the imaginary unit.

3. The method for suppressing false alarms in direction finding of mobile satellite terminals based on multi-dimensional feature joint screening according to claim 2, characterized in that, The specific process of the frequency offset coarse compensation processing is as follows: Using Fast Fourier Transform for Complex Baseband Signals Frequency offset estimation is performed by calculating the amplitude spectrum of the signal in the frequency domain and finding the frequency point corresponding to the amplitude peak. As a coarse estimate, the baseband complex signal after coarse frequency offset compensation is then obtained by compensating through the multiplier signal. .

4. The method for suppressing false alarms in direction finding of mobile satellite terminals based on multi-dimensional feature joint screening according to claim 3, characterized in that, The specific synchronization process of time slot synchronization and burst location processing in step S2 is as follows: Waveform template corresponding to the frame header synchronization word sequence of the pre-stored uplink time slot temp ( m ), m =1,2,..., M 1; of which M 1 represents the length of the waveform template; m Index of sequence elements; Based on sliding correlation function R ( k )right and temp ( m Perform sliding cross-correlation calculations; Searching for | R ( k peak position of )| k peak When the peak position k peak The corresponding peak value | R ( k peak Exceeding the preset synchronization threshold Th sync When this happens, it is determined that the frame header has been successfully captured, where | R ( k )| indicates R ( k The magnitude of the amplitude; Based on the known fixed offset Δ between the frame header and the start position of the burst data off set Determine the burst start position of the burst pulse signal that was successfully captured in the frame header. k start The burst termination position is determined based on the time slot length. k end .

5. The method for suppressing false alarms in direction finding of mobile satellite terminals based on multi-dimensional feature joint screening according to claim 4, characterized in that, The original confidence vector in step S3 s The specific calculation process for the time-domain confidence component is as follows: Based on the burst initiation position obtained in step S2 k start and the location of the sudden end k end Calculate the measured burst pulse width W meas ; Measured burst pulse width W meas Compared with the standard time slot pulse width in the feature template library W std By comparing the two, we obtain the absolute value of the difference, i.e., calculate | W meas -W std | The smaller the difference, the higher the matching degree of the time domain features; Calculate the time-domain confidence level based on the absolute value of the difference between the two. s t : ； in W tol This is the time error tolerance.

6. The method for suppressing false alarms in direction finding of mobile satellite terminals based on multi-dimensional feature joint screening according to claim 4, characterized in that, The original confidence vector in step S3 s The specific calculation process for the mid-frequency domain confidence component is as follows: Perform a Fast Fourier Transform on the burst data to obtain its power spectral density. P ( f ); Find the frequency point corresponding to the peak of the power spectrum f peak As a center frequency estimate ; Calculate the instantaneous bandwidth of the signal On the power spectrum, from the frequency point f peak Searching to the left and right respectively, when the descent distance is Δ P BW When points A and B are obtained, the difference between the frequency values ​​corresponding to points A and B is... ; Calculate frequency domain confidence s f : ； in The center value of the operating frequency band. For standard bandwidth, and These are the standard deviations of frequency matching and bandwidth matching, respectively.

7. The method for suppressing false alarms in direction finding of mobile satellite terminals based on multi-dimensional feature joint screening according to claim 1, characterized in that, The specific calculation process of the modulation domain confidence component in the original confidence vector s in step S3 is as follows: Perform on signal π / 4-DQPSK demodulation extracts a continuous phase difference sequence {Δ θ meas ( l ), l =1,2,..., L -1}; where L is the total length of the phase difference sequence. l For phase difference sequence index; The measured phase difference sequence is compared with the standard phase transition template {Δ θ ideal ( l Perform relevant matching; Calculate the confidence level of the modulation domain s m : ； in l The index is the phase difference sequence index; L is the total length of the phase difference sequence; Δ θ meas ( l ) is the first l The actual phase difference obtained by demodulation between adjacent symbols; Δ θ ideal ( l () represents the ideal standard phase jump value.

8. The method for suppressing false alarms in direction finding of mobile satellite terminals based on multi-dimensional feature joint screening according to claim 1, characterized in that, The original confidence vector in step S3 s The specific calculation process for the confidence component in the feature code field is as follows: Pre-stored frame synchronization word pilot sequence for uplink of mobile satellite system C up : C up =[ c 1, c 2,..., c i ,..., c M ], where the sequence number is 1 i individual code element c i ∈{±1}, M for C up Total code length; The baseband complex signal after coarse compensation for frequency offset of the received signal s comp ( n Starting from the initial position of the burst, sliding correlation matching is performed to obtain the sliding cross-correlation output complex value. R c ( k ); Output complex values ​​based on sliding cross-correlation R c ( k Calculate the confidence level of the feature code field. s c : ； in, M for C up Total code length.

9. The method for suppressing false alarms in direction finding of mobile satellite terminals based on multi-dimensional feature joint screening according to claim 1, characterized in that, The direction-finding results are fed back to the Q-learning algorithm module, forming a closed-loop optimization mechanism as follows: The logic for determining a reasonable orientation and the method for awarding positive rewards: Judgment logic: The direction finding system will output the target's azimuth angle. and pitch angle ; Reasonableness Standard 1: For ground or low-altitude terminals, its pitch angle Below the upper limit That is, pointing towards low altitude; Reasonableness Standard 2: When performing continuous direction finding on the same target, the rate of change of azimuth and elevation angle is less than 100° / s; Reasonableness Standard 3: Combining prior geographical information, i.e., the polygonal coordinates of the permitted area, starting from the location of the direction finding station, along the direction finding azimuth angle... θ Draw a ray and calculate whether the ray intersects with the boundary of the permitted polygon area. There must be at least one intersection point, and the ground distance from this intersection point to the direction-finding station must be within a preset ranging range. The preset ranging range is […]. R min , R max ], R min ,R max These are the lower and upper limits for distance measurement, respectively. Positive reward is given when any of the above reasonableness criteria are met. The Q-learning algorithm module considers the current decision to be a valid signal and generates a positive reward. The logic for determining abnormal directions and the methods for penalty or threshold adjustment: Decision logic: Abnormal Standard 1: Continuous N abnormal The signals deemed valid in the second determination all point to high altitudes in their direction-finding results. ; Anomaly Criterion 2: The spatial spectrum energy of the direction finding output has no obvious single peak or main peak, that is, the ratio of the peak value to the secondary peak value of the spatial spectrum is less than 3dB, or the spatial spectrum energy corresponding to the peak value is less than 6dB lower than the average energy of the whole frequency band. Anomaly Criterion 3: For a signal believed to be from the same terminal, if the absolute value of the change in azimuth angle of the direction finding result exceeds 150° or the absolute value of the change in elevation angle exceeds 120° within a preset time, an abnormal jump occurs, and the abnormal jump does not show a monotonic trend in three consecutive direction finding results. Penalties and increased thresholds: If any of the above abnormal criteria are met, the system determines that a false alarm has occurred. Punishment: Generates negative rewards; Threshold adjustment method: Trigger: Calculate the false alarm rate within the sliding window. FAR short ,like FAR short If the target limit is exceeded, or if three false alarms occur consecutively, the threshold will be raised. Adjustment amount: Threshold adjustment action selection a =+ δ Where δ is the threshold adjustment step size; Execution: Update judgment threshold Th for min ( Th + δ , Th max ); in Th max This is the preset threshold upper limit.

Images