A method for detecting spectral holes based on LAR-CFAR

Abstract

This invention discloses a spectrum hole detection method based on LAR-CFAR, applicable to the field of wireless communication technology. The method includes the following steps: developing a frequency sweep plan based on the target frequency band and hardware bandwidth, acquiring IQ signals, and obtaining the power spectrum through Welch power spectrum estimation; performing Quantile-CFAR detection across the entire frequency band, generating an adaptive threshold, and achieving initial detection; identifying candidate regions and performing OS-CFAR refinement; using exponential moving average for cross-frame fusion, combining the hole decision threshold to extract continuous spectrum holes and applying a minimum bandwidth constraint; performing post-processing to output the hole frequency band, bandwidth, and confidence level; monitoring system operating indicators and adaptively adjusting parameters. This invention reduces computational complexity while maintaining boundary accuracy, significantly reduces the false alarm rate through cross-frame fusion, and can adaptively adjust parameters to match hardware capabilities and environmental characteristics, making it suitable for scenarios such as cognitive radio and dynamic spectrum access.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and more specifically to a spectrum hole detection method based on LAR-CFAR (Layered Adaptive Refinement - Constant False Alarm Rate). LAR-CFAR refers to a layered detection architecture that includes a full-band Quantile-CFAR coarse detection layer and a candidate region OS-CFAR fine detection layer, achieving a balance between detection accuracy and computational efficiency through layered refinement. Background Technology

[0002] With the rapid growth of wireless communication services, spectrum resources are becoming increasingly scarce. Spectrum hole detection technology, by sensing unoccupied spectrum segments, provides fundamental information for cognitive radio and dynamic spectrum access, and is crucial for improving spectrum utilization. Existing technologies mainly include the following categories: 1. Traditional Unit Average Constant False Alarm Rate (CA-CFAR) Detection: This method uses the arithmetic mean of the samples within the training window as the noise floor. However, when there is signal energy within the training window, the noise floor estimate will be raised, causing the detection threshold to drift and resulting in serious missed detections, especially in dense signal environments.

[0003] 2. Fixed threshold energy detection: This method determines the spectrum occupancy by setting a fixed threshold. Its drawback is that it cannot adapt to the time-varying nature of low ambient noise (such as noise fluctuations caused by temperature changes or changes in equipment status), which can easily lead to fluctuations in false alarm rate or a decrease in detection probability.

[0004] 3. Improved CFAR methods: Although methods such as OS-CFAR and CMLD can resist multi-target interference to a certain extent, the computational load is large when performing full-band operation, making it difficult to meet the requirements of wide-band real-time scanning, and often does not take into account the bandwidth limitations of actual hardware platforms.

[0005] 4. Insufficient spectrum fragmentation processing: In complex electromagnetic environments, spectrum occupancy is fragmented. Existing algorithms often fail to accurately identify continuously available spectrum holes, resulting in ambiguous boundary markings and affecting the reliability of dynamic access.

[0006] 5. Poor hardware compatibility: Many methods are based on ideal signal models and ignore the effective bandwidth, sampling rate constraints and processing power limitations of actual software radio devices (such as HackRF), resulting in poor actual deployment performance.

[0007] In summary, existing spectrum hole detection techniques suffer from several technical problems, including inaccurate noise floor estimation, threshold drift, a trade-off between real-time performance and accuracy, difficulties in handling spectrum fragmentation, and poor hardware adaptability. Therefore, providing a LAR-CFAR-based spectrum hole detection method is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0008] In view of this, the present invention provides a spectrum hole detection method based on LAR-CFAR, which solves the technical problems existing in the spectrum hole detection technology, such as inaccurate noise floor estimation, threshold drift caused by training window contamination, contradiction between real-time performance and accuracy, difficulty in spectrum fragmentation processing, and poor hardware adaptability.

[0009] To achieve the above objectives, the present invention provides the following technical solution: A spectral hole detection method based on LAR-CFAR, where LAR-CFAR is a hierarchical adaptive refined constant false alarm rate (CFAR) detection architecture, includes the following steps: S1. Based on the target frequency band bandwidth and the effective bandwidth of the hardware, formulate a frequency sweep plan for multiple hops, sequentially collect the IQ signal sequence at each center frequency, and preprocess the IQ signal, including DC offset removal, anti-aliasing filtering and amplitude normalization. S2. Divide the preprocessed IQ signal into multiple time-domain segments, set overlap between segments, perform fast Fourier transform on each segment after windowing, calculate the periodogram of each segment and perform Welch averaging to obtain the power spectral density estimate. S3. The Quantile-CFAR method is used to perform preliminary detection across the entire frequency band to obtain an adaptive detection threshold. The power spectral density estimate is compared with the adaptive detection threshold to obtain the single-frame decision result. S4. Based on the single-frame decision results and spectral gradient information, identify candidate regions that need refinement, perform OS-CFAR refinement on the candidate regions, generate refinement thresholds, and update the decision results. S5. The single-frame decision results of multiple consecutive frames are fused using exponential moving average to obtain the occupancy probability. The occupancy probability is compared with the preset hole decision threshold to determine the hole frequency. The connected components of consecutive hole frequency points are merged, and the minimum bandwidth constraint is applied to obtain the spectrum hole that meets the conditions. S6. Perform morphological processing on the detected spectral holes, calculate the confidence level of each spectral hole, and output a list of holes containing the start frequency, end frequency, bandwidth, and confidence level. S7 monitors saturation rate, occupancy rate, noise fluctuation index, and real-time pressure indicators. Based on the monitoring results, it dynamically adjusts the fusion parameters, FFT points and / or Welch average segment count, as well as the number of hops and / or hop dwell time, and feeds the adjusted parameters back to S1 to S5 for execution.

[0010] Optionally, the frequency sweep plan for multiple hops can be specified in S1 as follows: Based on the effective bandwidth coefficient of the hardware Calculate the effective bandwidth: ; in, Sampling rate, The effective bandwidth coefficient of the hardware and Set overlapping bandwidth between adjacent hops. To ensure spectrum continuity, different hop dwell times are used for different frequency bands. .

[0011] Optionally, S3 specifically refers to: For each frequency point Define the training set and protection unit ; Calculate the training samples Quantiles as a local noise floor estimate : ; In the formula, Let be the empirical cumulative distribution function of the training samples. quantile and , For training sample vectors, The number of training units. express of Quantiles; generating adaptive thresholds : ; In the formula, The threshold ratio is used to perform initial decision-making and obtain the single-frame decision result. : ; In the formula, This is an indicator function that takes the value 1 when the condition is true and 0 otherwise. For the first Frame rate Power spectral density estimation, For the first Frame rate Adaptive threshold.

[0012] Optionally, the parameters of the Quantile-CFAR method are adaptively adjusted according to the frequency band characteristics, including the quantiles. The range of values ​​is Threshold ratio The target false alarm rate is determined through offline calibration. to .

[0013] Optionally, the candidate regions requiring refinement are identified in S4 as follows: Frequency points that meet any of the following conditions are identified as candidate regions: (1) Boundary points: In the formula, For frequency point The single-frame decision result, For frequency point The single-frame decision result; (2) High gradient: In the formula, For the first Frame rate Power spectral density estimation, For the first Frame rate Power spectral density estimation, This is the spectral gradient threshold; (3) Unstable points: In the formula, Indicates the recent The variance of the frame occupancy probability sequence is taken. This represents the occupancy probability of the previous frame. This is the threshold for the variance of the occupancy probability. Candidate region OS-CFAR refinement is performed only on the identified candidate points and their neighborhoods. The neighborhood is defined as the area before and after the candidate frequency point. To reduce computational complexity, several frequency points are used, including the spectral gradient threshold. Typical values ​​are taken as 0.3 to 0.5 times the median of the power spectral density of the current frequency band, occupying the probability variance threshold. The value range is 0.02 to 0.05.

[0014] Optionally, OS-CFAR refinement is performed on the candidate region as follows: Sort the training samples in ascending order: ; In the formula, They are respectively the 1st, 2nd, ... The order statistic is taken; the first one is taken. A sequence statistic is used as the OS noise floor estimate. : ; Generate OS-CFAR refinement threshold : ; In the formula, The OS-CFAR threshold ratio is used; the refinement decision is as follows: ; In the formula, For frequency point The single-frame judgment result.

[0015] Optionally, in S5, the exponential moving average is used to fuse the single-frame decision results of multiple consecutive frames, specifically as follows: ; In the formula, For the first Frame rate The probability of occupancy. For the first Frame rate The probability of occupancy. For the first Frame rate The single-frame decision result, The exponential moving average fusion coefficient and .

[0016] Optionally, the minimum bandwidth constraint in S5 is as follows: Map the frequency index to the actual frequency: ; In the formula, For actual frequency, For the center frequency, Sampling rate, Given the FFT points; calculate the hole bandwidth: ; In the formula, For hole bandwidth, This represents the lower limit of the empty interval. This represents the upper limit of the hole interval; only holes with sufficient bandwidth are retained. ; in, This represents the minimum hole bandwidth.

[0017] Optional, exponential moving average fusion coefficient The value range is 0.85 to 0.96, which is the hole decision threshold. The value range is 0.15 to 0.30, minimum bandwidth constraint. Pre-defined based on the channelization characteristics of the target frequency band.

[0018] Optionally, the monitoring and dynamic adjustment in S7 are as follows: Monitor the following indicators: saturation rate occupancy rate Noise fluctuation index Real-time pressure CPU utilization; parameters are dynamically adjusted based on monitoring results: when At that time, reduce the average number of Welch segments. or FFT points ;when When dB, increase or number of training units ;when At that time, increase the quantile and fusion coefficient ;when At that time, reduce and reduce .

[0019] As can be seen from the above technical solution, compared with the prior art, the present invention provides a spectral hole detection method based on LAR-CFAR, which has the following beneficial effects: 1. Strong resistance to training window contamination: This invention uses quantized quantiles instead of the mean to estimate the noise floor. Even if there are some signal energies in the training window, as long as their proportion does not exceed the corresponding quantile, the noise floor estimation is almost unaffected, effectively avoiding threshold drift and reducing the false negative rate. 2. High detection accuracy: This invention refines the candidate region using OS-CFAR and re-estimates the occupancy status in areas with drastic changes or high gradients by using sorting statistics, which significantly improves the localization accuracy of hole boundaries and reduces boundary ambiguity. 3. Low false alarm rate: EMA cross-frame fusion suppresses single-frame random spikes, making it less likely for transient interference to trigger hole errors, while enhancing detection stability through inter-frame information complementarity; 4. Low computational complexity: The entire frequency band only performs lightweight Quantile-CFAR, resulting in low computational overhead; the complex OS-CFAR only operates on a few candidate region neighborhoods, significantly reducing the overall computational load compared to the entire frequency band OS-CFAR. 5. Highly adaptive: It can adaptively adjust the quantiles according to factors such as frequency band signal density, noise fluctuations, and real-time pressure. Fusion coefficient Its processing accuracy and frequency sweeping strategy make it suitable for complex and ever-changing electromagnetic environments. 6. Good hardware adaptability: The effective hardware bandwidth coefficient is explicitly introduced in the frequency sweep planning. By combining overlap bandwidth and dwell time settings, it can adapt to software radio platforms with different bandwidths and sampling rates; at the same time, it ensures stable operation under limited computing resources through an adaptive degradation strategy. 7. High spectrum utilization: It can accurately extract available holes of continuous width, avoid erroneous splicing of fragmented spectrum, provide reliable hole information for dynamic spectrum access, and improve spectrum resource utilization. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0021] Figure 1 This is a flowchart of the LAR-CFAR-based spectral hole detection method of the present invention; Figure 2 This is a system architecture block diagram of the present invention; Figure 3 This is a schematic diagram of the spectral states at each stage of the detection process of the present invention; Figure 4 This is the parameter adaptive adjustment logic diagram for the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] This invention discloses a spectral hole detection method based on LAR-CFAR, where LAR-CFAR is a hierarchical adaptive refined constant false alarm rate (CFAR) detection architecture, such as... Figure 1 and Figure 2 As shown, the detection method includes the following steps: S1. Based on the target frequency band bandwidth and the effective bandwidth of the hardware, formulate a frequency sweep plan for multiple hops, sequentially collect the IQ signal sequence at each center frequency, and preprocess the IQ signal, including DC offset removal, anti-aliasing filtering and amplitude normalization. S2. Divide the preprocessed IQ signal into multiple time-domain segments, set overlap between segments, perform fast Fourier transform on each segment after windowing, calculate the periodogram of each segment and perform Welch averaging to obtain the power spectral density estimate. S3. The Quantile-CFAR method is used to perform preliminary detection across the entire frequency band to obtain an adaptive detection threshold. The power spectral density estimate is compared with the adaptive detection threshold to obtain the single-frame decision result. S4. Based on the single-frame decision results and spectral gradient information, identify candidate regions that need refinement, perform OS-CFAR refinement on the candidate regions, generate refinement thresholds, and update the decision results. S5. The single-frame decision results of multiple consecutive frames are fused using exponential moving average to obtain the occupancy probability. The occupancy probability is compared with the preset hole decision threshold to determine the hole frequency. The connected components of consecutive hole frequency points are merged, and the minimum bandwidth constraint is applied to obtain the spectrum hole that meets the conditions. S6. Perform morphological processing on the detected spectral holes, calculate the confidence level of each spectral hole, and output a list of holes containing the start frequency, end frequency, bandwidth, and confidence level. S7 monitors saturation rate, occupancy rate, noise fluctuation index, and real-time pressure indicators. Based on the monitoring results, it dynamically adjusts the fusion parameters, FFT points and / or Welch average segment count, as well as the number of hops and / or hop dwell time, and feeds the adjusted parameters back to S1 to S5 for execution.

[0024] Spectral status at each stage of the detection process, as follows Figure 3 As shown, the architecture adopted in this embodiment of the invention includes a hardware and signal acquisition layer, a spectrum estimation layer, a detection layer, a fusion decision layer, a post-processing output layer, and an adaptive adjustment layer; The problem of spectral hole detection is modeled as a binary hypothesis test: ; in This is complex Gaussian white noise, meaning it has a mean of 0 and a variance of . It follows a complex normal distribution.

[0025] (Occupancy Assumption): Frequency Point Occupied by signal: ; in In order to transmit signals, For noise. The detection target is within a given false alarm rate. Maximize the detection probability under the constraints. : .

[0026] Furthermore, the frequency sweep plan for multiple hops in S1 is specifically defined as follows: Based on the effective bandwidth coefficient of the hardware Calculate the effective bandwidth: ; in, Sampling rate, The effective bandwidth coefficient of the hardware and Set overlapping bandwidth between adjacent hops. To ensure spectrum continuity, different hop dwell times are used for different frequency bands. .

[0027] In this embodiment of the invention, the effective hardware bandwidth coefficient is taken as: This reflects the ratio of the actual available bandwidth to the sampling rate of the SDR platform's analog front-end. Overlapping bandwidth is set between adjacent hops. To ensure spectral continuity, the overlap bandwidth is generally taken as... of Different hop dwell times are used for different frequency bands. Broadband uses shorter dwell times to speed up scanning, while narrowband uses longer dwell times to improve frequency resolution.

[0028] The preprocessing is as follows: Preprocessing includes three sub-steps: (1) DC offset removal: Calculate the mean of the IQ signal sequence and subtract it to eliminate the DC component introduced by the receiver local oscillator leakage: ; in is the length of the IQ sequence.

[0029] (2) Anti-aliasing filtering: Apply a low-pass FIR filter to the signal after DC removal, with the cutoff frequency set to 0. of This suppresses out-of-band components.

[0030] (3) Amplitude normalization: Normalize the amplitude of the filtered signal to a uniform range to eliminate the influence of receiver gain fluctuations. .

[0031] In this embodiment of the invention, a frame refers to the process of completing a full power spectrum estimation and detection decision for the IQ signal of a single hop. For multi-hop frequency sweeps, a complete sweep cycle contains the same number of frames as the number of hops. (The superscript in the following formulas...) Both refer to the frame number.

[0032] The specific mathematical model for Welch power spectrum estimation in S2 is as follows: Segmented windowing: [The windowing process is described in the original text, but the actual windowing is not included here.] IQ sequences are divided into paragraph, each paragraph Points, overlap rate . No. Segment signal is After adding a window, we get: ; in For the Hann window: ; Window normalization coefficient: ; FFT transform: Perform FFT on each signal segment. Point FFT: ; in The imaginary unit satisfies .

[0033] Periodic chart calculation: The periodic diagram of a segment is defined as follows: ; It represents the modulus of a complex number.

[0034] Welch average: for Averaging the periodic plots yields the Welch power spectral density estimate: ; Statistical characteristic analysis: In Assume that a single-segment periodic chart follows an exponential distribution: ; Welch average The mean of each of the three independent and identically distributed exponential variables follows a Gamma distribution: ; (Here, shape-scale parameterization is used, and the shape parameter is...) The scale parameter is ) Its mean and variance are: ; Increase It can reduce the variance of PSD estimation (variance and (Inversely proportional), thus reducing the false alarm rate, but increasing computational complexity and response latency.

[0035] Furthermore, S3 specifically refers to: For each frequency point Define the training set and protection unit The training sample set is: Calculate the training samples Quantiles as a local noise floor estimate : ; In the formula, In this embodiment, the empirical cumulative distribution function of the training samples is used. , quantile and , For training sample vectors, The number of training units. express of Quantiles; generating adaptive thresholds : ; In the formula, The threshold ratio is used to perform initial decision-making and obtain the single-frame decision result. : ; In the formula, This is an indicator function that takes the value 1 if the condition is true, and 0 otherwise. For the first Frame rate Power spectral density estimation, For the first Frame rate Adaptive threshold.

[0036] Furthermore, the parameters of the Quantile-CFAR method are adaptively adjusted according to the frequency band characteristics, and the quantiles... The range of values ​​is Threshold ratio The target false alarm rate is determined through offline calibration. to .

[0037] Furthermore, the specific steps for identifying candidate regions requiring refinement in S4 are as follows: Frequency points that meet any of the following conditions are identified as candidate regions: (1) Boundary points: In the formula, For frequency point The single-frame decision result, For frequency point The single-frame decision result; (2) High gradient: In the formula, For the first Frame rate Power spectral density estimation, For the first Frame rate Power spectral density estimation, This is the spectral gradient threshold; (3) Unstable points: In the formula, Indicates the recent Frame (in this embodiment, The variance of the occupancy probability sequence is taken. This represents the occupancy probability of the previous frame. This is the threshold for the variance of the occupancy probability. In this embodiment of the invention, when there is no historical frame data after system startup, condition (3) is skipped, and only conditions (1) and (2) are used to identify candidate regions. Candidate region OS-CFAR refinement is performed only on the identified candidate points and their neighborhoods. The neighborhood is defined as the frequency points before and after the candidate frequency point to reduce computational complexity.

[0038] Candidate region OS-CFAR refinement is performed only on the identified candidate points and their neighborhoods. The neighborhood is defined as the area before and after the candidate frequency point. To reduce computational complexity, several frequency points are used, including the spectral gradient threshold. Typical values ​​are taken as 0.3 to 0.5 times the median of the power spectral density of the current frequency band, occupying the probability variance threshold. The value range is 0.02 to 0.05.

[0039] Furthermore, the OS-CFAR refinement of the candidate region is specifically performed as follows: Sort the training samples in ascending order: ; In the formula, They are respectively the 1st, 2nd, ... The order statistic is taken; the first one is taken. The order statistics are used as the OS noise floor estimate. : ; Generate OS-CFAR refinement threshold : ; In the formula, The OS-CFAR threshold ratio is used; the refinement decision is as follows: ; In the formula, For frequency point The single-frame decision result is obtained by replacing the original Quantile-CFAR decision result in the candidate region with the refined decision result.

[0040] Furthermore, S5 employs an exponential moving average to fuse the single-frame decision results of multiple consecutive frames, specifically as follows: ; In the formula, For the first Frame rate The probability of occupancy. For the first Frame rate The probability of occupancy. For the first Frame rate The single-frame decision result, The exponential moving average fusion coefficient and .

[0041] In an embodiment of the present invention, when When it is a stationary binary process, the occupancy probability The steady-state expectation is: ; in For frequency point Steady-state probability of being occupied. Equivalent time constant: ; For example, hour, Frames, that is, after about 25 frames, the historical information decays to .

[0042] Initial frame settings That is, the decision result of the first single frame is used as the initial value of the occupancy probability. This is before system startup. Within a frame (equivalent time constant), the occupancy probability has not yet converged, and the confidence value of the hole decision result is low.

[0043] Hole detection based on fusion results:

[0044] in The hole decision threshold is usually set to [value missing]. . Frequency point In the The frame was identified as a hole.

[0045] Consecutive hole frequencies are merged into intervals. One-dimensional connected component analysis is performed to merge consecutively adjacent frequency points in the frequency domain that all satisfy the hole decision condition into intervals, resulting in a set of hole intervals. .

[0046] Furthermore, the minimum bandwidth constraint in S5 is specifically as follows: Map the frequency index to the actual frequency: ; In the formula, For actual frequency, For the center frequency, Sampling rate, Given the FFT points; calculate the hole bandwidth: ; In the formula, For hole bandwidth, This represents the lower limit of the empty interval. This represents the upper limit of the hole interval; only holes with sufficient bandwidth are retained. ; in, This represents the minimum hole bandwidth.

[0047] For multi-hop frequency sweeping scenarios, after each hop independently executes the processing steps S2 to S5, the holes detected by each hop are spliced ​​together according to the actual frequency coordinates. For frequency points within the overlapping bandwidth area of ​​adjacent hops, the logical OR of the detection results of both sides is taken (i.e., if any hop is determined to be occupied, it is considered occupied) to ensure that no occupied signals are missed in the splicing area.

[0048] In this embodiment of the invention, the S6 morphological processing and confidence level are specifically as follows: The detected spectral holes were subjected to morphological processing using a one-dimensional opening operation (erosion followed by dilation), with length as the structuring element. Rectangular elements, used to remove rectangles with a width less than [the specified value]. The frequency points of burrs and isolated holes are then processed; followed by a closing operation (expansion followed by erosion), with the same structuring element length as above, used to fill the small gaps inside the holes.

[0049] Hole confidence is defined as:

[0050] in:

[0051] For the first The mean probability of occupancy within each empty interval. Confidence level. The closer the value is to 1, the higher the probability that the hole has been correctly identified. (First item) The second item reflects the degree of "vacancy" of the hole. It reflects the sufficiency of hole bandwidth. The larger the bandwidth, the closer this term is to 1, but it is not always 1.

[0052] The final output includes the starting frequency. Termination frequency ,bandwidth and confidence level A list of empty holes.

[0053] Furthermore, the exponential moving average fusion coefficient The value range is 0.85 to 0.96, which is the hole decision threshold. The value range is 0.15 to 0.30, minimum bandwidth constraint. Pre-defined based on the channelization characteristics of the target frequency band.

[0054] Furthermore, such as Figure 4 As shown, the monitoring and dynamic adjustment in S7 are specifically as follows: Monitor the following indicators: saturation rate occupancy rate Noise fluctuation index Real-time pressure CPU utilization; parameters are dynamically adjusted based on monitoring results: when At that time, reduce the average number of Welch segments. or FFT points ;when When dB, increase or number of training units ;when At that time, increase the quantile and fusion coefficient ;when At that time, reduce and reduce .

[0055] Saturation middle This is the power spectral density value corresponding to the upper limit of the receiver's dynamic range (related to the ADC bit depth and reference impedance). This represents the total number of frequency points across the entire frequency band (the sum of the number of frequency points in each hop in a multi-hop scenario). Saturation rate reflects the ADC overload ratio. Utilization rate. Reflects the signal density in the current frequency band. Noise fluctuation index. That is, noise floor estimation The standard deviation over time after conversion to decibels (dB) reflects the degree of noise floor fluctuation, and the unit is dB.

[0056] Table 1 lists the symbols and their definitions used in this embodiment: Table 1 Symbols and Definitions

[0057] Table 2 shows the key technical parameters of this embodiment: Table 2 Key Technical Parameters

[0058] In one embodiment of the present invention, the test was conducted through an experiment, the experimental method of which is as follows: Test environment: Laboratory environment, HackRF One connected to an omnidirectional antenna, indoor office area, test distance from the signal source approximately 2-5 meters. Ground truth determination: The spectrum occupancy status is synchronously recorded using a spectrum analyzer (reference device) as the baseline truth value. Comparative experimental setup: CA-CFAR ( Fixed threshold energy detection (threshold set to noise floor + 6dB) was compared using the same IQ data. Evaluation metric: Detection probability False alarm rate Boundary positioning error (in frequency resolution units) (unit) processing delay Experiment 1: 5G n79 band spectrum hole detection System Configuration: Hardware platform: HackRF One + Python 3.8 Target frequency band: 4800~4960 MHz (sub-band within the 5G n79 band; the full range of n79 is 4400~5000MHz) Sampling rate: MS / s FFT points: (Quick Mode) / (Fine-grained mode) Welch average number of segments: (fast) / (fine) Effective bandwidth coefficient: , MHz Frequency sweep planning: The 4800~4960 MHz band is divided into 11 hops, with a center spacing of 15 MHz for each hop, an overlap bandwidth of 2 MHz, and a hop dwell time of 30 ms. Detection parameters: Quantile-CFAR: , , , OS-CFAR: , , EMA fusion: , Minimum hole bandwidth: MHz Experimental results: In actual testing on the 5G n79 band, the system took approximately 350 ms to complete a full-band scan of a 160 MHz bandwidth (11 hops × 30 ms + processing time). Three usable spectrum holes were detected in the test scenario, with hole bandwidths of 12 MHz, 8 MHz, and 6 MHz, and confidence levels of 0.92, 0.87, and 0.81, respectively. Compared to CA-CFAR, this invention, under conditions of a 35% signal content in the training window and an interference signal SNR of 10-15 dB, achieved a noise floor estimation deviation of less than 3 dB, while CA-CFAR exhibited a deviation of 8-12 dB under the same conditions (increasing with increasing interference SNR). The average boundary localization error was 1.5 frequency resolution units, an improvement of approximately 53% compared to the 3.2 units of coarse detection by Quantile-CFAR. After EMA fusion, the actual false alarm rate decreased from [previous frame value] to [previous frame value]. Down to OS-CFAR only applies to approximately 22% of the candidate regions across the entire frequency band, resulting in an overall computational cost reduction of approximately 72% compared to full-band OS-CFAR.

[0059] Experiment 2: Detection of 2.4GHz WiFi band

[0060] Frequency band characteristics: 2400~2483.5 MHz, dense WiFi signal, strong bursts.

[0061] Parameter configuration: Sampling rate: MS / s FFT points: Welch average number of segments: Effective bandwidth coefficient: , MHz Frequency sweep planning: The 83.5 MHz bandwidth is divided into 6 hops, with a center spacing of approximately 13.3 MHz between hops and an overlap bandwidth of approximately 3.7 MHz. hop stay time: ms Quantile-CFAR: , , OS-CFAR: EMA fusion: , Minimum hole bandwidth: MHz Experimental results: To address the bursty nature of WiFi signals, a smaller EMA coefficient is used. (Equivalent time constant) (Frames) to improve response speed, while appropriately reducing the hole decision threshold. To capture short-duration holes. In the test scenario, five spectral holes were detected, with bandwidths ranging from 1 to 4 MHz. Due to the strong burstiness of WiFi signals, cross-frame fusion effectively suppressed false alarms caused by instantaneous occupancy, reducing the false alarm rate from a single frame. Down to The detection latency is approximately 200 ms (6 hops × 30 ms + processing time), which meets the real-time requirements of dynamic spectrum access.

[0062] Experiment 3: Detection of 350MHz PDT Cluster Band

[0063] Frequency band characteristics: 351~356 MHz (downlink), 361~366 MHz (uplink), 12.5 kHz channel spacing

[0064] Parameter configuration: Sampling rate: MS / s FFT points: Welch average number of segments: Frequency sweep planning: The downlink band (351~356 MHz) and uplink band (361~366 MHz) are scanned independently, each requiring 4 hops. (8 hops, MHz, hop spacing approximately 1.2 MHz, overlap bandwidth approximately 0.5 MHz) hop stay time: ms (Narrowband frequencies use longer dwell times to improve frequency resolution) Quantile-CFAR: , , OS-CFAR: EMA fusion: , Minimum hole bandwidth: kHz Experimental results: To address the narrowband channel characteristics of PDT systems, a higher frequency resolution is employed. (Frequency resolution approximately 488 Hz) and more stringent parameter configurations ( , This enables single-channel level hole detection. Increase... To reduce the variance of narrowband PSD estimation, training units To cover sufficient channel samples. During testing, approximately 280 usable channels (12.5 kHz each) were identified within a 5 MHz bandwidth, with a mean confidence level of 0.88. Noise fluctuation index. The noise floor remained below 0.5 dB during long-term monitoring, indicating that the noise floor estimate was stable. Corresponding equivalent time constant For the slow-changing characteristics of PDT trunking systems (channel state change cycles are typically on the order of seconds), a response time of about 6 frames (about 50 ms for a single hop scan, about 300 ms for 6 frames) is sufficient to track channel state changes while maintaining sufficient cross-frame smoothness to suppress transient interference.

[0065] Experiment 4: Effect of adaptive parameter adjustment

[0066] In the 5G n79 band scenario of Experiment 1, the system ran continuously for 30 minutes. The initial system parameters were the fine-mode configuration of Experiment 1 ( , , After 5 minutes of operation, the frequency band occupancy rate dropped to approximately 20% (low occupancy scenario), and the system followed the S7 adaptive strategy. Reduced from 0.96 to 0.92 to improve response speed, while also... The system was reduced from 8 to 4 to conserve computing resources. Subsequently, the system detected the following and triggered an adaptive adjustment: (1) When running for 8 minutes, the CPU utilization rate rose to 85% (high real-time pressure), and the system automatically adjusted the Welch average segment count. From 8 to 4, FFT points The CPU utilization dropped from 512 to 5048, and the CPU usage fell back to 58%. The changes in detection performance metrics (false alarm rate, detection probability) were within acceptable limits (false alarm rate from...). Rise to still satisfied constraint).

[0067] (2) At the 15th minute of operation, external interference caused noise fluctuation index The system automatically increases the number of training units as the value increases from 0.3 dB to 1.2 dB. From 64 to 96, the stability of the noise floor estimation recovered. It fell back to 0.5 dB. This is because the previous adjustment (1) had already... and The CPU utilization rate has ample margin to absorb the reduction. The additional sorting computation overhead due to the 50% increase (the computational complexity of the Quantile-CFAR approximate quantile is...) The overall computing load of the system remains within a safe range.

[0068] (3) At the 22nd minute of operation, the frequency band occupancy rate increased from 30% to 65% (signal density), and the system automatically increased the percentile. From 0.35 to 0.40, and the fusion coefficient The detection probability was increased from 0.92 in the low occupancy phase to 0.96 (i.e., restored to the initial configuration of Experiment 1) to reduce missed detections under dense signal conditions. The adjusted equivalent time constant increased from approximately 12 frames to approximately 25 frames, and the detection probability increased from 0.91 to 0.96.

[0069] Through the aforementioned adaptive mechanism, the system can achieve a dynamic balance between detection performance and computational load.

[0070] Hardware constraints: The HackRF One platform used in this embodiment employs an 8-bit ADC with a theoretical dynamic range of approximately 50 dB. The actual usable dynamic range is approximately 35-40 dB. When detecting weak signals near strong signals (such as base station transmission signals), ADC quantization noise may limit detection sensitivity. To address this constraint, this invention partially mitigates the impact of insufficient dynamic range through a Quantile-CFAR adaptive threshold mechanism—when a strong signal causes a rise in local noise, the quantile estimation automatically tracks and raises, avoiding misjudgment of weak signals. For applications requiring a larger dynamic range, a 12-bit or 14-bit ADC SDR platform (such as the USRP B210) can be used, in which case the effective bandwidth coefficient... The power spectral density value corresponding to the upper limit of the receiver's dynamic range The parameters need to be adjusted accordingly.

[0071] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0072] Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A spectral hole detection method based on LAR-CFAR, characterized in that, LAR-CFAR is a hierarchical adaptive refined constant false alarm rate (CFAR) detection architecture. The detection method includes the following steps: S1. Based on the target frequency band bandwidth and the effective bandwidth of the hardware, formulate a frequency sweep plan for multiple hops, sequentially collect the IQ signal sequence at each center frequency, and preprocess the IQ signal, including DC offset removal, anti-aliasing filtering and amplitude normalization. S2. Divide the preprocessed IQ signal into multiple time-domain segments, set overlap between segments, perform fast Fourier transform on each segment after windowing, calculate the periodogram of each segment and perform Welch averaging to obtain the power spectral density estimate. S3. The Quantile-CFAR method is used to perform preliminary detection across the entire frequency band to obtain an adaptive detection threshold. The power spectral density estimate is compared with the adaptive detection threshold to obtain the single-frame decision result. S4. Based on the single-frame decision results and spectral gradient information, identify candidate regions that need refinement, perform OS-CFAR refinement on the candidate regions, generate refinement thresholds, and update the decision results. S5. The single-frame decision results of multiple consecutive frames are fused using exponential moving average to obtain the occupancy probability. The occupancy probability is compared with the preset hole decision threshold to determine the hole frequency. The connected components of consecutive hole frequency points are merged, and the minimum bandwidth constraint is applied to obtain the spectrum hole that meets the conditions. S6. Perform morphological processing on the detected spectral holes, calculate the confidence level of each spectral hole, and output a list of holes containing the start frequency, end frequency, bandwidth, and confidence level. S7 monitors saturation rate, occupancy rate, noise fluctuation index, and real-time pressure indicators. Based on the monitoring results, it dynamically adjusts the fusion parameters, FFT points and / or Welch average segment count, as well as the number of hops and / or hop dwell time, and feeds the adjusted parameters back to S1 to S5 for execution.

2. The spectral hole detection method based on LAR-CFAR according to claim 1, characterized in that, The specific sweep frequency planning for multiple hops in S1 is as follows: Based on the effective bandwidth coefficient of the hardware Calculate the effective bandwidth: ； in, Sampling rate, The effective bandwidth coefficient of the hardware and Set overlapping bandwidth between adjacent hops. To ensure spectrum continuity, different hop dwell times are used for different frequency bands. .

3. The method for detecting spectral holes based on LAR-CFAR according to claim 1, characterized in that, S3 specifically refers to: For each frequency point Define the training set and protection unit ; Calculate the training samples Quantiles as a local noise floor estimate : ； In the formula, Let be the empirical cumulative distribution function of the training samples. quantile and , For training sample vectors, The number of training units. express of Quantiles; generating adaptive thresholds : ； In the formula, The threshold ratio is used to perform initial decision-making and obtain the single-frame decision result. : ； In the formula, This is an indicator function that takes the value 1 when the condition is true and 0 otherwise. For the first Frame rate Power spectral density estimation, For the first Frame rate Adaptive threshold.

4. The method for detecting spectral holes based on LAR-CFAR according to claim 3, characterized in that, The parameters of the Quantile-CFAR method are adaptively adjusted according to the frequency band characteristics, quantiles The range of values ​​is Threshold ratio The target false alarm rate is determined through offline calibration. to .

5. The spectral hole detection method based on LAR-CFAR according to claim 1, characterized in that, The specific steps for identifying candidate regions requiring refinement in S4 are as follows: Frequency points that meet any of the following conditions are identified as candidate regions: (1) Boundary points: In the formula, For frequency point The single-frame decision result, For frequency point The single-frame decision result; (2) High gradient: In the formula, For the first Frame rate Power spectral density estimation, For the first Frame rate Power spectral density estimation, This is the spectral gradient threshold; (3) Unstable points: In the formula, Indicates the recent The variance of the frame occupancy probability sequence is taken. This represents the occupancy probability of the previous frame. This is the threshold for the variance of the occupancy probability. Candidate region OS-CFAR refinement is performed only on the identified candidate points and their neighborhoods. The neighborhood is defined as the area before and after the candidate frequency point. To reduce computational complexity, several frequency points are used, including the spectral gradient threshold. Typical values ​​are taken as 0.3 to 0.5 times the median of the power spectral density of the current frequency band, occupying the probability variance threshold. The value range is 0.02 to 0.

05.

6. The spectral hole detection method based on LAR-CFAR according to claim 1, characterized in that, The specific steps for performing OS-CFAR refinement on candidate regions are as follows: Sort the training samples in ascending order: ； In the formula, They are respectively the 1st, 2nd, ... The order statistic is taken; the first one is taken. A sequence statistic is used as the OS noise floor estimate. : ； Generate OS-CFAR refinement threshold : ； In the formula, The OS-CFAR threshold ratio is used; the refinement decision is as follows: ； In the formula, For frequency point The single-frame judgment result.

7. The method for detecting spectral holes based on LAR-CFAR according to claim 1, characterized in that, In S5, the exponential moving average is used to fuse the single-frame decision results of multiple consecutive frames, specifically as follows: ； In the formula, For the first Frame rate The probability of occupancy. For the first Frame rate The probability of occupancy. For the first Frame rate The single-frame decision result, The exponential moving average fusion coefficient and .

8. The method for detecting spectral holes based on LAR-CFAR according to claim 7, characterized in that, The minimum bandwidth constraint in S5 is as follows: Map the frequency index to the actual frequency: ； In the formula, For actual frequency, For the center frequency, Sampling rate, Given the FFT points; calculate the hole bandwidth: ； In the formula, For hole bandwidth, This represents the lower limit of the empty interval. This represents the upper limit of the hole interval; only holes with sufficient bandwidth are retained. ； in, This represents the minimum hole bandwidth.

9. The method for detecting spectral holes based on LAR-CFAR according to claim 8, characterized in that, Exponential moving average fusion coefficient The value range is 0.85 to 0.96, which is the hole decision threshold. The value range is 0.15 to 0.30, minimum bandwidth constraint. Pre-defined based on the channelization characteristics of the target frequency band.

10. The method for detecting spectral holes based on LAR-CFAR according to claim 1, characterized in that, The monitoring and dynamic adjustment in S7 are specifically as follows: Monitor the following indicators: saturation rate occupancy rate Noise fluctuation index Real-time pressure CPU utilization; parameters are dynamically adjusted based on monitoring results: when At that time, reduce the average number of Welch segments. or FFT points ;when When dB, increase or number of training units ;when At that time, increase the quantile and fusion coefficient ;when At that time, reduce and reduce .

Images