Method for interference detection in a broadband satellite communication system

Abstract

The application designs a broadband satellite communication system interference detection method, comprising: acquiring system operation parameters, and based on the system operation parameters, segmented collection is performed on the broadband received signal to obtain segmented observation data; the segmented observation data is sequentially subjected to windowing processing and frequency domain transformation to obtain segmented power spectrum; the mean value of the segmented power spectrum in the preset period is counted to generate an average power spectrum; the average power spectrum is subjected to iterative deconvolution processing to obtain an estimated power spectrum; the noise floor estimate value is determined based on the estimated power spectrum, and the interference feature is extracted in the estimated power spectrum with the noise floor estimate value as the reference; and the interference feature is output as the detection result. The application alleviates the restriction of resolution and sidelobe in traditional spectrum estimation, and realizes high-precision, low-false-alarm frequency domain interference detection under limited data length.

Description

Technical Field

[0001] This invention relates to the field of satellite communication and signal processing technology, and in particular to an interference detection method for broadband satellite communication systems. Background Technology

[0002] In modern broadband satellite communication systems, the increasing complexity of the space electromagnetic environment and the scarcity of spectrum resources necessitate cognitive radio technology and dynamic spectrum access technology. Achieving the utilization and dynamic allocation of spectrum resources requires spectrum sensing of broadband received signals to identify frequency bands occupied by interference sources and available spectrum gaps. The accuracy of spectrum sensing directly determines the reliability of the communication link, system throughput, and the overall efficiency of spectrum resource utilization.

[0003] Currently, the most commonly used basic method for spectrum detection of broadband received signals is the periodogram estimation method based on Fast Fourier Transform (FFT). To reduce spectral leakage caused by finite data truncation and suppress sidelobes, windowing is commonly used in engineering, such as applying Hamming or Hanning windows, combined with periodic averaging of multiple signal segments to obtain the power spectral density. This architecture, due to its low computational complexity and ease of real-time hardware implementation, has been widely deployed in the signal processing modules of various spectrum monitoring and analysis instruments and receivers.

[0004] The technical system still needs further improvement to address the constraints between resolution and sidelobe suppression in related technologies. Summary of the Invention

[0005] The purpose of this invention is to construct an interference detection method for broadband satellite communication systems to solve at least one problem existing in the prior art.

[0006] According to one aspect of this application, a method for detecting interference in a broadband satellite communication system includes:

[0007] Obtain pre-configured system operating parameters, and collect broadband received signals in segments based on system operating parameters to obtain segmented observation data;

[0008] Windowing and frequency domain transformation are performed sequentially on the segmented observation data to obtain the segmented power spectrum;

[0009] The average power spectrum is generated by calculating the mean of the segmented power spectrum within a preset period.

[0010] Perform iterative deconvolution on the average power spectrum to obtain the estimated power spectrum;

[0011] The noise floor estimate is determined based on the estimated power spectrum, and the interference features are extracted from the estimated power spectrum with the noise floor estimate as a reference.

[0012] Output interference features as the detection result.

[0013] Beneficial effects: This invention overcomes the limitations of resolution and sidelobes in traditional spectral estimation, and achieves high-precision, low-false-alarm frequency domain interference detection with limited data length. Attached Figure Description

[0014] Figure 1 This is a flowchart of an interference detection method for a broadband satellite communication system provided in an embodiment of this application.

[0015] Figure 2 This is a flowchart illustrating how to obtain segmented observation data by collecting broadband received signals in segments based on system operating parameters, as provided in this embodiment of the application.

[0016] Figure 3 This is a flowchart of frequency response pre-equalization provided for an embodiment of this application.

[0017] Figure 4 The flowchart provided in this application describes the process of performing iterative deconvolution on the average power spectrum to obtain the estimated power spectrum.

[0018] Figure 5 This is a flowchart for determining whether the iteration termination condition is met, provided as an embodiment of this application.

[0019] Figure 6 This is a flowchart for determining the noise floor estimate based on the estimated power spectrum, provided as an embodiment of this application.

[0020] Figure 7 This is a flowchart illustrating the extraction of interference features from the estimated power spectrum using a noise floor estimate as a reference, provided for embodiments of this application. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0022] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a predetermined order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in sequences other than those illustrated or described herein. Furthermore, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0023] In view of the above-mentioned problems, the applicant conducted a search and analysis and found that:

[0024] Existing windowed spectrum estimation methods face inherent limitations when dealing with broadband complex electromagnetic environments. For example, the non-rectangular window used to suppress sidelobes leads to the widening of the main lobe, resulting in a decrease in frequency resolution. At the same time, constraints such as the uneven frequency response of the simulation front end also make it difficult for the system to accurately set a robust detection threshold.

[0025] Therefore, how to overcome the constraint between resolution and sidelobe suppression has become a technical challenge that urgently needs to be overcome in the field of broadband spectrum sensing.

[0026] To overcome the above shortcomings, combined with Figures 1 to 7 The present application will be further illustrated with reference to specific embodiments.

[0027] The hardware selection, parameter values, and process steps given in this application are merely exemplary embodiments and are not intended to limit the scope of protection of this invention.

[0028] Based on the core technical logic of this invention, the devices, parameters and implementation methods can be equivalently replaced and adjusted according to the actual application scenario; these will not be explained in detail below.

[0029] This embodiment provides a method for detecting interference in a broadband satellite communication system, which includes the following steps:

[0030] Step 101: Obtain the pre-configured system operating parameters, and collect the broadband received signal in segments based on the system operating parameters to obtain segmented observation data;

[0031] The system operating parameters may include the refresh cycle of the system interference detection scan, the main clock frequency of the device, and the sampling frequency of the analog-to-digital converter.

[0032] In this embodiment, the hardware environment can specifically use a Field-Programmable Gate Array (FPGA) as the processor, and a radio frequency agile transceiver for signal conversion. The broadband received signal is a baseband signal processed by the radio frequency front end and input to the digital domain.

[0033] Among them, segmented observation data refers to discrete data blocks formed after truncating a continuous baseband data stream.

[0034] In this embodiment, a non-continuous acquisition mode is used to read a signal segment of a predetermined length within a limited master clock cycle and store it in the local memory. After the current data segment enters the processing pipeline, the acquisition of the next signal segment is started.

[0035] This mode avoids buffering mechanisms, thus adapting to the real-time detection requirements of high-speed broadband signals.

[0036] Step 102: Perform windowing and frequency domain transformation on the segmented observation data in sequence to obtain the segmented power spectrum;

[0037] In this embodiment, directly performing a fast Fourier transform on a signal of finite length will introduce a spectrum leakage phenomenon caused by data truncation, causing the energy of a strong signal to leak to nearby idle frequency points.

[0038] Therefore, a window function with low sidelobe characteristics is selected to perform point-by-point multiplication on the segmented observation data.

[0039] In some cases, the coefficients processed by windowing can be in 16-bit unsigned fixed-point format.

[0040] Furthermore, the windowed time-domain data is then fed into the hardware frequency domain transformation core computing unit, converted into a frequency domain complex sequence, and further obtained by calculating the squares of the real and imaginary moduli separately and then summing them to obtain the segmented power spectrum characterizing the frequency distribution of signal energy within a single acquisition time period.

[0041] Step 103: Calculate the average value of the segmented power spectrum within a preset period to generate the average power spectrum;

[0042] Furthermore, the results of multiple observations need to be smoothed. Multiple consecutively generated segmented power spectra are continuously accumulated throughout the refresh cycle and stored in an accumulation register to prevent data overflow.

[0043] In some embodiments, the accumulator register can be configured to be 48 bits wide.

[0044] In some embodiments, after reaching a preset total number of acquisition segments, an arithmetic shift or division operation is performed to obtain the expected value of the accumulated result, thereby generating the average power spectrum.

[0045] This mechanism reduces noise variance and reveals the spectral profile of weak interference signals; however, it widens the main lobe of the signal spectrum, leading to a decrease in frequency resolution.

[0046] Step 104: Perform iterative deconvolution processing on the average power spectrum to obtain the estimated power spectrum;

[0047] Given the aforementioned issues, this step can treat the average power spectrum as the convolution result of the true unknown spectrum and the window function spectrum.

[0048] Accordingly, using a predetermined inverse algorithm, the power spectrum of the window function is used as the point spread function to perform multiple rounds of inverse calculation on the observed average power spectrum, forcing the dispersed energy to reconcentrate towards the center frequency.

[0049] After multiple iterations of compensation, the output estimated power spectrum has its main lobe width compressed, and the frequency resolution is restored to a high level close to that of the unwindowed version, while still maintaining low sidelobe interference.

[0050] Step 105: Determine the noise floor estimate based on the estimated power spectrum, and extract interference features from the estimated power spectrum with the noise floor estimate as a reference;

[0051] Accordingly, a power statistical benchmark is extracted within the preset effective detection bandwidth, and then the energy of each frequency point in the estimated power spectrum is compared with the noise baseline estimate.

[0052] When the ratio of the power of a certain local frequency band to the estimated noise floor exceeds a preset decision threshold, it is determined that there is external interference at that location.

[0053] For example, the noise floor estimate represents the average energy baseline level of pure background white noise that is not occupied by real signals in the current spectrum environment;

[0054] Interference characteristics can specifically include a set of multidimensional parameters that describe the physical properties of the external interference, including the frequency location of the interference and its peak intensity relative to the noise floor.

[0055] Step 106: Output the interference features as the detection result.

[0056] Accordingly, the extracted interference features, including location and intensity information, are formatted and encapsulated, and then sent to the host computer or the spectrum management and scheduling center of the satellite communication system.

[0057] In some optional implementations, a hardware limit can be preset for the number of reports submitted at one time, and only the interference features with the highest interference intensity can be selected for final output, in order to prevent the amount of data output from overflowing the system communication bus bandwidth in extremely dense interference scenarios.

[0058] In this application, an iterative deconvolution mechanism with the power spectrum of the window function as the point spread function is introduced. While maintaining the low sidelobe leakage property, the main lobe width of the signal is compressed, and the masked frequency domain detail resolution is restored.

[0059] The hardware environment described above, using FPGA as an example, is merely one specific deployment configuration of the method of this invention. Those skilled in the art will understand that the method of this invention is also applicable to software or firmware implementations on other computing platforms such as general-purpose digital signal processors, application-specific integrated circuits, or general-purpose computer processors; this invention does not impose any particular limitations on these implementations.

[0060] On the other hand, this embodiment further explains the automated determination process of segmented acquisition parameters and the hardware timing adaptation mechanism.

[0061] The system operating parameters include the pre-configured refresh cycle, master clock frequency, and sampling frequency;

[0062] The refresh cycle is the time interval between completing a full-bandwidth interference scan and outputting a detection report.

[0063] In some scenarios, such as the interference monitoring application of a broadband satellite earth station, the refresh cycle is usually set to 2 seconds in order to balance detection sensitivity and real-time reporting.

[0064] The master clock frequency is the reference clock frequency used inside the FPGA to drive the algorithm logic.

[0065] In this embodiment, the master clock frequency can be specifically set to 160MHz.

[0066] Sampling frequency, which is the rate at which the RF agile transceiver performs analog-to-digital conversion on the input broadband analog signal;

[0067] In this embodiment, 491.52MHz can be selected.

[0068] According to an embodiment of this application, the broadband received signal is segmented based on system operating parameters to obtain segmented observation data, specifically including:

[0069] Step 201: Calculate the segmented sampling interval based on the sampling frequency, the master clock frequency, and the pre-configured single-segment sampling length;

[0070] In this embodiment, the number of sampling points processed in a single fast Fourier transform calculation can be defined as the single-segment sampling length;

[0071] In some cases, it can be pre-configured to 4096 points.

[0072] Furthermore, the count value used to characterize the interval between two adjacent data acquisition actions in the master clock domain can be defined as the segmented sampling interval, and its calculation process follows the formula below:

[0073] T _s =ceil(Ψ×SCLK / F _s );

[0074] Above, T _s The segmented sampling interval has a value corresponding to the number of counts in the main clock cycle;

[0075] F _s The sampling frequency;

[0076] SCLK is the master clock frequency;

[0077] ceil is the floor function;

[0078] Ψ represents the single-segment sampling length.

[0079] This facilitates the relocation of the current data segment and the switching of memory states, thereby avoiding timing violations caused by high-speed data flow.

[0080] Step 202: Determine the logarithmic segment number exponent with reserved timing margin based on the refresh cycle, master clock frequency and segmented sampling interval;

[0081] Accordingly, it is necessary to ensure that the algorithm's computation time does not exceed the time limit, and the system needs to calculate the theoretically maximum number of segments that can be accommodated within a single refresh cycle. In this embodiment, to adapt to the complex scheduling of the hardware pipeline and reserve safety boundaries, the determination process of the logarithmic segment number exponent follows the following linear formula:

[0082] Num _log =floor(log2((T×SCLK) / T _s ))-1;

[0083] In the formula, Num _log The logarithmic segment number exponent;

[0084] T represents the refresh cycle;

[0085] SCLK is the master clock frequency;

[0086] T _s The sampling interval is segmented;

[0087] floor is the floor function;

[0088] log2 is the logarithmic function with base 2.

[0089] In the above formula, the subtraction of 1 at the end is to provide dimensionality reduction redundancy at the hardware level, that is, to reserve a timing margin of a complete segmented sampling interval based on the theoretical calculation results. This is used to offset the dynamic delays caused by FPGA internal state machine jumps, data bus arbitration, and algorithm pipeline clearing, so that the execution time of the entire statistical cycle is controlled within the refresh cycle.

[0090] Step 203: Determine the total number of segments using the logarithmic segment number exponent;

[0091] Accordingly, by using a constant of 2 as the base and the logarithmic segment number exponent as the power, the total number of segments to be processed within a statistical period can be obtained. The calculation process is as follows:

[0092] Num=2 Num_log ;

[0093] In the above formula, Num represents the total number of segments;

[0094] Num _log It is the logarithmic segment number exponent.

[0095] In this formula, the total number of segments is limited to an integer power of 2. When calculating the average power spectrum, the energy accumulation value at each frequency point can be calculated simply by a binary right shift operation, which improves the system's operating efficiency and saves hardware resources.

[0096] Step 204: Using the total number of segments and the segment sampling interval as constraints, perform discontinuous segmented acquisition on the broadband received signal to obtain segmented observation data.

[0097] Furthermore, the input broadband analog signal can be digitally captured according to the sampling frequency.

[0098] For example, whenever Ψ sampling points are accumulated, the acquisition module immediately stops sampling and stores the data segment in the local static random access memory, while marking that the data segment is ready;

[0099] At this point, the hardware counter, driven by the master clock, begins to count down the segmented sampling interval;

[0100] Once the count value reaches the value set for the segmented sampling interval, the acquisition module will start the next Ψ-point acquisition.

[0101] This process is repeated until all data collection tasks specified for the total number of segments are completed.

[0102] In this case, computational tasks, such as deconvolution iteration, can be performed using the timing interval between two samples to achieve a dynamic balance between sampling real-time performance and computational depth.

[0103] In one possible implementation, windowing is performed on the segmented observation data, specifically including:

[0104] A window function with real even symmetry is selected as the windowing operator to process the segmented observation data, resulting in windowed segmented data. This ensures that in subsequent iterative deconvolution processing, the Fourier transform result of the point spread function generated by the window function is a real number sequence, thus making frequency domain correlation operation equivalent to frequency domain convolution operation.

[0105] Furthermore, in the update iteration logic, the standard deconvolution algorithm needs to calculate the frequency correlation between the expected observation spectrum and the residual.

[0106] In digital signal processing, correlation operations in the frequency domain correspond to taking the conjugate complex number of the frequency domain representation of one of the variables and then performing multiplication.

[0107] In this embodiment, the analytical properties of the windowing operator are limited at the time-domain front end, forcing the use of a real even-symmetric window function. According to the mathematical properties of the Discrete Fourier Transform, the frequency domain response of a real even-symmetric sequence is also a real even-symmetric sequence, and the power spectrum formed by the square of its modulus is naturally a real number sequence. Since the conjugate of a real number is equal to itself, this allows for the direct use of multipliers to perform convolution operations instead of related calculations when performing frequency domain calculations within the FPGA.

[0108] As a specific implementation method, the windowing operator can be the Hanning window, which satisfies the real even symmetry condition in the time domain.

[0109] Based on this, the overhead of conjugate flip logic and complex multipliers is eliminated, which reduces computational latency and saves logic gate resources while ensuring the rigor of deconvolution theory.

[0110] According to embodiments of this application, before performing iterative deconvolution processing on the average power spectrum, a frequency response pre-equalization step is further included, namely:

[0111] Step 301: Apply moving median filtering to the average power spectrum to extract the noise floor envelope;

[0112] Correspondingly, the average power spectrum contains the global distribution characteristics of the broadband signal before processing.

[0113] In practical radio frequency systems, the receiving channel of the analog front-end typically exhibits in-band flatness fluctuations, and anti-aliasing filters introduce additional power roll-off at the frequency band edges, resulting in a non-flat, undulating background noise pattern. If this non-flat power spectrum is directly fed into the deconvolution iteration stage, the deconvolution algorithm will identify the slowly varying envelope of the background noise as a signal feature broadened by the point spread function and attempt to sharpen and compress it, thereby generating spurious interference peaks in the signal-free frequency band region.

[0114] To this end, a moving data window of predetermined width is constructed, which slides point by point on the frequency axis of the average power spectrum. At each sliding position, the median of all power values ​​within the window is calculated, and the series of output medians are connected to form the noise floor envelope.

[0115] Furthermore, the window width of the moving median filter needs to be larger than the main lobe span of a typical interference signal to prevent burst signal peaks from being included in the envelope estimation;

[0116] At the same time, the width of this window needs to be smaller than the characteristic scale of the frequency response fluctuation of the RF front end in order to track channel fluctuations.

[0117] In one possible system configuration, the width of the sliding window can be set to 128 frequency sampling points.

[0118] Step 302: Based on the noise floor envelope and its mean, determine the equalization coefficients used to compensate for system channel fluctuations;

[0119] Alternatively, calculate the mean of the noise floor envelope across all frequency points to obtain the envelope mean.

[0120] Based on the noise floor envelope and envelope mean, the equalization coefficient used to compensate for system channel fluctuations is determined;

[0121] In this embodiment, it is necessary to calculate a global scale transformation reference value to level the fluctuations at each frequency point to a unified baseline.

[0122] Furthermore, the arithmetic mean of the noise floor envelope is calculated over the entire effective detection bandwidth;

[0123] For each discrete frequency point, construct its corresponding equalization factor.

[0124] For example, G(k) = P _env_mean / (P _env (k)+δ _env );

[0125] In the formula, G(k) is the equalization coefficient corresponding to frequency index k, and P _env_mean P corresponds to the arithmetic mean of the noise floor envelope at all frequency points. _env (k) corresponds to the noise floor envelope value at frequency index k, δ _env This corresponds to a relatively small protective constant to prevent division by zero anomalies.

[0126] In the frequency range where the original noise floor is low, the calculated equalization coefficient is greater than 1, which has an amplification and compensation effect; while in the range where the original noise floor is high, the equalization coefficient is less than 1, which has an attenuation and suppression effect.

[0127] Step 303: Use the equalization coefficient to perform weighted correction on the average power spectrum to obtain the pre-equalized power spectrum;

[0128] Furthermore, the original amplitude of the average power spectrum is extracted point by point in the frequency domain, and multiplied with the corresponding equalization coefficient under the same frequency index.

[0129] After multiplicative weighted correction, the noise floor fluctuations in the broadband spectrum are flattened.

[0130] It should be understood that the output pre-equalized power spectrum visually and numerically presents an ideal signal model based on the assumption of flat white noise, which can be used to block the risk of false alarms caused by frequency response fluctuations.

[0131] Step 304: Use the pre-equalized power spectrum as input data for performing iterative deconvolution processing.

[0132] Accordingly, the pre-equalized power spectrum is fed into the subsequent deconvolution calculation. That is, the deconvolution state machine takes over the data control, using the flattened power distribution as the initial state, and performs multiple rounds of energy concentration calculation.

[0133] The process of obtaining the estimated power spectrum also includes a reverse restoration step:

[0134] Step ①: Obtain the power spectrum directly output after the end of the iterative deconvolution process, and use it as the preliminary estimated power spectrum;

[0135] In this process, after multiple rounds of iterative deconvolution processing meet the termination condition, the output data is the preliminary estimated power spectrum.

[0136] At this point, the main lobe width of the power spectrum has been compressed, improving the frequency resolution. Because pre-equalization was performed earlier, the overall power scale of the signal spectrum now deviates from the physical magnitude of the actual sampling by the RF front-end and needs correction.

[0137] Step 2: Perform inverse scaling on the initially estimated power spectrum using the equalization coefficients to restore the power scale;

[0138] Furthermore, the full-band equalization coefficients previously stored in the static random access memory are retrieved again to perform a reverse demodulation and repositioning operation on the initially estimated power spectrum;

[0139] This inverse operation is a division operation or an equivalent multiplication by the reciprocal, and its calculation process can be described as follows:

[0140] P _est (k)=P _pre (k) / G(k);

[0141] Above, P_est (k) represents the amplitude of the estimated power spectrum after restoring the true power scale, P _pre (k) represents the amplitude of the preliminary estimated power spectrum at frequency index k, and G(k) represents the equalization coefficient corresponding to frequency index k.

[0142] Based on this, the frequency band that is amplified and compensated is reduced by an equal amount, and the frequency band that is attenuated is amplified by an equal amount, which helps to restore the power density distribution pattern of the signal.

[0143] Step ③: Use the power spectrum after scale reduction as the final estimated power spectrum.

[0144] In other embodiments, the power spectrum after scale reduction can be defined as the scale-reduced power spectrum.

[0145] The power spectrum data stream after reverse demodulation is formally labeled as the estimated power spectrum and pushed into the next logic pipeline stage. The extracted power values ​​will then have electromagnetic physical meaning when performing signal-to-noise ratio calculations or interference threshold comparisons.

[0146] In this way, modeling errors caused by frequency response fluctuations can be smoothed out, and fake spikes induced by the deconvolution process in the no-signal frequency band can be avoided.

[0147] According to an embodiment of this application, an iterative deconvolution process is performed on the average power spectrum to obtain an estimated power spectrum, including:

[0148] Step 401: Obtain the point expansion function generated based on the pre-configured window function;

[0149] The point spread function is an important parameter used to characterize the frequency response diffusion characteristics of a system.

[0150] Accordingly, the frequency domain modulus square of the windowing operator is pre-calculated to form a power spectrum sequence, and the global energy sum of the power spectrum sequence is normalized to 1; this is used to conserve the energy of the deconvolution algorithm. Based on this, the entire iterative process can converge to the maximum likelihood estimate.

[0151] Step 402: Initialize the iterative power spectrum with the average power spectrum;

[0152] In other embodiments, the power spectrum input to the iterative deconvolution process may also be used to initialize the iterative power spectrum;

[0153] The iterative deconvolution algorithm requires the initial input sequence to have non-negativity. When the system does not perform the frequency response pre-equalization step, the average power spectrum is directly used as the initial input, which is generated by calculating the square of the complex modulus and accumulating it across segments, and has non-negativity.

[0154] When the system performs the frequency response pre-equalization step, the initial input here is the power spectrum after pre-equalization. Since both the equalization coefficient and the original power spectrum are non-negative, the pre-equalized power spectrum also satisfies the non-negative constraint.

[0155] Furthermore, the corresponding input power spectrum can be read and assigned to the iteration storage area as the starting point for the 0th iteration, thereby starting the entire iteration engine.

[0156] Step 403: In each iteration, the desired observation spectrum is calculated based on the convolution operation of the iterative power spectrum and the point spread function;

[0157] In the main iteration loop, the current iterative power spectrum is multiplied by the point spread function using a Fast Fourier Transform in the frequency domain. Then, an inverse Fast Fourier Transform is used to restore it to the original processing domain. Based on this, the theoretical expected value for the current iteration progress is generated.

[0158] Step 404: Generate an update factor based on the ratio characteristics of the average power spectrum and the expected observed spectrum;

[0159] In other words, an update factor is generated based on the ratio of the power spectrum input to the iterative deconvolution process to the expected observed spectrum.

[0160] Optionally, the division ratio of the average power spectrum to the desired observed spectrum is calculated for each frequency point, and this ratio sequence is then subjected to correlation operations with the point spread function. To prevent individual frequency bands from having a divisor of 0, which could trigger an overflow anomaly in the hardware divider, a preset zero-prevention constant is added when processing the desired observed spectrum.

[0161] It should be understood that the result output after frequency domain correlation operation is the update factor sequence that characterizes the current power deviation correction direction at each frequency point.

[0162] Step 405: Use the update factor to perform a weighted update on the current iterative power spectrum to obtain the updated iterative power spectrum;

[0163] In this step, when processing broadband radio frequency signals, if the noise floor frequency points where no signal exists continue to receive weighted updates, random white noise will be abnormally amplified and spikes will be generated. To solve this problem, a masking intervention mechanism is introduced, which uses the following sequence of steps to truncate and control the weighting behavior, specifically including:

[0164] Step 4051: Initialize the convergence mask indicating the non-converged state for each frequency point;

[0165] Optionally, a one-dimensional mask register array with the same length as the total number of frequency points of the Fast Fourier Transform can be allocated in the internal random access memory.

[0166] Before the first iteration starts, the mask flags corresponding to all frequency points are uniformly initialized to 1, where 1 indicates that the corresponding frequency point is in an active update state that has not converged.

[0167] Step 4052: After each generation of the update factor, calculate the deviation between the update factor at each frequency point and the preset convergence benchmark value.

[0168] For example, the set convergence baseline value can be constantly set to 1.0.

[0169] When the update factor calculated at a frequency point equals the convergence baseline value, it proves that the power estimate at that frequency point has reached a steady state and no further compensation is needed.

[0170] Furthermore, the absolute difference between the generated update factor and the convergence benchmark value is calculated each time to extract the deviation, which is used as a technical indicator to quantitatively evaluate the local convergence depth at each frequency point.

[0171] Step 4053: If the deviation at a certain frequency point is lower than the pre-configured single-point convergence threshold, then update the convergence mask corresponding to that frequency point to the converged state.

[0172] Accordingly, the deviation is read at each frequency point and a numerical determination is made.

[0173] For example, the pre-configured single-point convergence threshold can be set to 0.01.

[0174] Once the deviation at a certain frequency point is found to be lower than the single-point convergence threshold, it means that the power gain or attenuation brought by a single iteration is less than 1%, and the technical benefit of continuing to spend cycles to update it is low.

[0175] At this point, a register bit manipulation instruction is immediately triggered to overwrite the mask flag bit corresponding to the frequency point from 1 to 0. 0 indicates that the frequency point has entered a converged frozen state.

[0176] Step 4054: Based on the current convergence mask at each frequency point, perform mask allocation on the update factor to generate an effective update factor;

[0177] Among them, the effective update factor corresponding to the converged state is forcibly locked to the target value that does not change the original power spectrum distribution;

[0178] The physical process of mask allocation is equivalent to inserting a controlled dynamic multiplexer into the data flow.

[0179] Where, ρ _eff (k)=M[k]×ρ(k)+(1-M[k])×κ;

[0180] Above, ρ _eff(k) represents the effective update factor at frequency index k, M[k] represents the convergence mask at frequency index k, ρ(k) represents the original update factor at frequency index k, and κ represents the target value for forced locking; in some embodiments, it can be set to 1.0.

[0181] In this formula, for active frequency points with a mask value of 1, their effective update factor directly inherits the original calculated value.

[0182] For frozen frequency points where the mask value has been overwritten as 0, the first product is cleared to zero, and its effective update factor is forcibly replaced and locked to κ.

[0183] Step 4055: Update the current iterative power spectrum by multiplying it point by point using the effective update factor to obtain the updated iterative power spectrum.

[0184] Furthermore, the current iterative power spectrum amplitude data in the buffer is read again and multiplied one by one with the effective update factor.

[0185] Since the multiplier for a frequency point determined to be in a frozen state is forced to be κ, the power estimate for that frequency point remains static throughout the current and all subsequent iterations. This mechanism allows the algorithm to concentrate its computing power on sharpening strong signal characteristics within the main frequency band, while suppressing links in signal-free regions from generating spoofed noise peaks due to excessive multiplication operations.

[0186] Step 406: The updated iterative power spectrum output when the preset iteration termination condition is met is used as the estimated power spectrum.

[0187] After the above control logic is applied, the final output steady-state data frame is designated as the estimated power spectrum, which means that the frequency domain energy enhancement calculation is terminated.

[0188] Based on the above-described inventive concept, an alternative implementation method is provided. In some scenarios, the convergence mask operation circuit and the range calculation logic can be separated, and a forced flow mode with a pre-configured fixed iteration period can be adopted.

[0189] For example, the loop constraint register is set to a constant Ω times, the internal pipeline does not perform any state probing, and exits directly after unconditionally traversing Ω times and saving the result data.

[0190] Accordingly, the possible value of Ω can be 30.

[0191] It should be understood that this solution reduces wiring complexity.

[0192] According to the embodiments of this application, the process of determining whether the iteration termination condition is met includes:

[0193] Optionally, extract the maximum and minimum values ​​of all update factors generated in a single iteration;

[0194] In some embodiments, when generating the raw update factor data for the entire frequency band, a multi-path comparison tree logic structure attached to the end of the pipeline can be optionally invoked to capture the numerical extreme value parameters in the global array in real time and synchronously without occupying additional storage scan clock cycles.

[0195] Optionally, the range between the maximum and minimum values ​​can be calculated;

[0196] In some embodiments, the maximum value is extracted and the minimum value is subtracted. The result is a technical indicator that characterizes the global fluctuation span of the iterative update amplitude throughout the entire frequency band.

[0197] Optionally, when the range is less than the pre-configured global convergence threshold, the iteration termination condition is determined to be met and an instruction to stop the iteration is triggered.

[0198] Among them, the single-point freezing mechanism handles local convergence, while the range parameter monitors the approximation state of the main peak component with slower energy concentration in the spectrum.

[0199] In some embodiments, the pre-configured global convergence threshold can be specifically set to 0.02.

[0200] When the range narrows to within the global convergence threshold, it can be determined that all characteristic components of the entire frequency band have been restored to the critical steady state. After the master control state machine captures this instruction, it can directly circuit break the nested iterative loop and execute the adaptive early termination process.

[0201] In dense multi-tone signal testing environments, this mechanism can reduce the actual number of executions.

[0202] According to embodiments of this application, the process of determining whether the iteration termination condition is met further includes a hardware-safe forced termination mechanism, which may be:

[0203] Optionally, the current iteration number of the iterative deconvolution process can be monitored synchronously;

[0204] Accordingly, a hardware accumulator counter is maintained independently within the iteration loop boundaries. The register value of this counter is incremented by 1 each time the top-level loop structure starts a new round of computation. This is used to prevent getting trapped in an unknown deadlock.

[0205] Optionally, if the current iteration count reaches the pre-configured upper limit threshold for iteration, the iteration termination condition will be forcibly determined to be met regardless of whether the features of the update factor meet the convergence requirements.

[0206] In low signal-to-noise ratio background detection scenarios, deconvolution algorithms may fall into a trap of continuous high-frequency oscillations that fail to approximate the optimal solution.

[0207] When the count value reaches the iteration upper limit threshold, the watchdog timing interception mechanism is triggered to force a circuit breaker, ignoring all status flags of the aforementioned mask and range. This establishes the worst-case time limit for handling latency, ensuring the stability of the overall refresh cycle.

[0208] For example, the pre-configured iteration upper limit threshold can be specifically set to 50.

[0209] Optionally, stop the iteration and directly output the current iterative power spectrum.

[0210] Accordingly, write access to the memory bus is immediately blocked, pipeline register remnants are cleaned up, and the last correctly held iteration data in the memory array is pushed into the output channel.

[0211] In this embodiment, a two-layer adaptive architecture based on local freezing of frequency point mask and global monitoring of range is constructed, which sharpens the target signal while blocking the iteration of white noise and releasing logic processing resources.

[0212] On the other hand, determining the noise basis estimate based on the estimated power spectrum includes:

[0213] Step 501: Extract the target spectral line sequence corresponding to the pre-configured effective detection bandwidth from the estimated power spectrum;

[0214] Furthermore, a center-symmetric flip operation is performed on the estimated power spectrum entering this processing stage, so that the DC component corresponds to the center index of the data array, and the negative frequency component and the positive frequency component are respectively arranged on both sides of the center index.

[0215] In this case, the frequency resolution ratio between the sampling frequency and the number of processing points is calculated in advance.

[0216] Next, based on the pre-configured effective detection bandwidth, index mapping is performed within the effective frequency range centered on the DC component as the axis of symmetry. Invalid frequency sidelobe components at the bandwidth edge are removed, and all effective frequency sampling point data falling within this closed interval are extracted to form the target spectral sequence.

[0217] For example, the effective detection bandwidth can be optionally pre-configured as a 400MHz detection bandwidth.

[0218] Step 502: Based on the power distribution characteristics of the target spectral line sequence, determine the reference power value representing the background white noise level;

[0219] In some basic alternative implementations, the target spectral line sequence can be traversed to search for and extract several frequency points with the smallest energy values, and then the arithmetic mean of several local minimum points can be calculated as the reference power value.

[0220] In some contexts, a few local minima may deviate from the background noise. A histogram mode estimation strategy can be used to replace the local mean extraction logic, which will be implemented in subsequent sequence steps.

[0221] Step 503: Use the reference power value as the noise floor estimate.

[0222] Furthermore, the extracted reference power value is uniformly solidified into a global noise floor estimate. This value can then be used as a level reference for calculating relative intensity during the subsequent decision-making phase.

[0223] According to the embodiments of this application, step 502 may specifically employ a histogram mode estimation strategy, namely:

[0224] Step 5021: Determine the upper power limit and the lower power limit with zero protection for the target spectral line sequence;

[0225] Accordingly, the target spectral line sequence is scanned to extract the global maximum value, which is then designated as the power upper limit.

[0226] In this case, to prevent illegal logarithmic zeros or negative infinity overflows in subsequent logarithmic operations, the upper limit of power is calculated by multiplying it by a preset smaller constant, and this product is used as the lower limit of power with zero protection, thus forcibly constraining the statistical space within the effective dynamic range.

[0227] For example, a smaller preset constant can be set to 0.001.

[0228] Step 5022: Based on the upper and lower power limits, construct a logarithmic domain power histogram containing a preset number of equal-width intervals;

[0229] Accordingly, a counter array of a specified length is pre-allocated in the static random access memory, the length of which is the preset number.

[0230] Perform logarithmic domain interpolation on the power boundaries to determine the logic width of each interval;

[0231] In some embodiments, the preset quantity can be specifically set to 64.

[0232] Accordingly, ΔQ = log2(P) _ceil / P _floor ) / L;

[0233] In the above, ΔQ is the logarithmic width of each interval, log2 is the logarithmic function with base 2, and P... _ceil That is, the upper limit of power, P _floor This refers to the lower limit of power with zero protection, where L is the preset number of intervals contained in the logarithmic domain power histogram.

[0234] Based on this, exponentially distributed power data can be transformed into a linear index space, which facilitates mapping addressing.

[0235] Step 5023: Map each frequency point in the target spectral line sequence to the corresponding interval of the logarithmic domain power histogram according to its power value and accumulate the frequency count.

[0236] Optionally, a loop state machine is started to read the power amplitude data of each frequency point in the target spectral line sequence one by one. For each read power amplitude, its logarithmic deviation relative to the lower power limit is calculated, and the corresponding array storage address is calculated. The index calculation process follows the following linear formula:

[0237] b=min(floor(log2(P _est / P _floor ) / ΔQ),L-1);

[0238] In the above, b is the interval index of the logarithmic domain power histogram obtained by mapping calculation, min is the minimum value function, floor is the floor function, log2 is the logarithmic function with base 2, and P... _est To estimate the power amplitude of the power spectrum at the current frequency, P _floor The lower limit of power with zero protection is ΔQ, where ΔQ is the logarithmic width of each interval, and L is the preset number of intervals.

[0239] After calculating the interval index, the corresponding array address unit is directly triggered to perform the accumulation operation. After traversing all target frequency points, the numerical sequence stored in the hardware array constitutes the discretized power probability density distribution map.

[0240] Step 5024: In the logarithmic power histogram after frequency accumulation, retrieve the mode interval with the highest accumulated frequency;

[0241] Furthermore, based on statistical principles in the detection scenario, interference signals typically occupy only a limited frequency band, while the vast signal-free region is dominated by flat white noise. Therefore, in the power probability density histogram, the power range containing white noise concentrates most of the frequency sampling points, manifesting as a peak in the statistical frequency.

[0242] Furthermore, all counter array cells can be scanned, and the address of the cell with the largest value can be extracted. This address corresponds to the mode interval with the highest accumulated frequency. This algorithm mechanism can be used to capture the interval where the noise floor is located.

[0243] Step 5025: Calculate the reference power value based on the center power value corresponding to the mode interval.

[0244] When reconstructing the actual power from the mode interval, a compensation bias of half the interval width is introduced. The calculation process for inversely reconstructing the power from the logarithmic domain to the linear domain is as follows:

[0245] P _base =P _floor ×2 (b_star+0.5)×ΔQ ;

[0246] Above, P _base For the calculated reference power value, P _floor b is the lower limit of power with zero protection. _star The index value of the retrieved mode interval is 0.5, the center point quantization compensation constant is 0.5, and ΔQ is the logarithmic width of each interval.

[0247] Optionally, the output P will be restored. _base Assign a reference power value to the register. The logarithmic histogram processing path decouples the noise distortion effect caused by the anti-convolution positive value iteration, ensuring the objectivity of the threshold setting.

[0248] Building upon this, optional implementation schemes for the average power spectrum generation step are provided to address high-speed, non-stationary transient interference sources, including pulse-sweep radar or frequency-hopping communication.

[0249] Optionally, for a pre-configured transient interference detection mode, the maximum power value in all segmented power spectra is extracted at the same frequency point to generate the maximum hold power spectrum;

[0250] In the conventional calculation of average power spectrum, if a transient interference exists only within an extremely short segmented sampling period, its energy will be diluted during the lengthy accumulation and averaging process, causing the output amplitude to be submerged under the overall white noise substrate and triggering a missed detection.

[0251] Therefore, a transient interference detection mode is preset internally.

[0252] When this mode is activated, the hardware control unit no longer performs cross-segment numerical accumulation, but is instead configured as a peak detector structure;

[0253] For each independent frequency channel, the absolute value of the currently received segmented power spectrum is compared with the value stored in the register.

[0254] If the current value is greater than the historical value, then overwrite it.

[0255] Otherwise, retain historical values.

[0256] The data array output after a complete refresh cycle constitutes the maximum hold power spectrum.

[0257] Alternatively, sort the power values ​​of all segmented power spectra at the same frequency point, extract the power values ​​corresponding to the preset percentiles, and generate percentile power spectra.

[0258] As an optional implementation, an internal sorting network unit can also be invoked. This unit collects the amplitude values ​​of all segmented spectra at a fixed frequency point within the same statistical period and sorts them in real-time in ascending order, directly extracting the amplitude values ​​occupying a specified proportion in the sorted sequence. Specifically, the set percentile can be 95%. The array sequence generated through this channel constitutes the percentile power spectrum.

[0259] Optionally, the maximum hold power spectrum or percentile power spectrum can be used as input data instead of the average power spectrum to perform subsequent iterative deconvolution processing.

[0260] Correspondingly, when the system is operating in the pre-configured transient interference detection mode, the underlying logic bus undergoes a path switch, and the maximum hold power spectrum or percentile power spectrum will replace the original data block obtained by arithmetic average and be routed to the input port of the deconvolution iterative operation module.

[0261] When facing extremely dense radio frequency interference environments, the mode extraction strategy of logarithmic domain power histogram is used to replace the minimum mean method, which suppresses the global extremum distortion problem and can promote a more robust anchoring to the real background noise.

[0262] In this embodiment, the specific implementation process of the sub-grid precision interpolation extraction mechanism for interference features, the continuous domain clustering merging strategy, and the capacity control output mechanism is further described.

[0263] Among them, the interference features are extracted from the estimated power spectrum with the noise basis estimate as a reference, including:

[0264] Step 601: In the estimated power spectrum, select the initial peak frequency points where the power value satisfies the preset threshold condition relative to the noise floor estimate.

[0265] Furthermore, a center-shift mapping is performed on the output power spectrum. This requires shifting the zero-frequency index containing the DC component to the center of the data array, so that the negative and positive frequency components are positioned on either side of the center point.

[0266] In some embodiments, for the extraction conditions, the ratio between the power amplitude at each frequency point and the pre-obtained noise floor estimate is taken as the logarithm to the base 10, and the product of the logarithm and 10 is the signal-to-noise ratio value in the logarithmic domain.

[0267] During frequency extraction, the ratio of the power amplitude of each frequency point to the noise floor estimate is converted into a logarithmic form of 10 times to obtain the logarithmic domain signal-to-noise ratio as the criterion.

[0268] When the logarithmic domain signal-to-noise ratio of a certain frequency point is greater than or equal to a preset threshold, that point is determined as the initial peak frequency point.

[0269] In some cases, the preset threshold condition can be configured to 3dB.

[0270] Step 602: Extract the power amplitude of the initial peak frequency point and the power amplitude of the adjacent left and right frequency points;

[0271] Furthermore, based on the array index of the selected initial peak frequency point, an addressing operation is performed to access the random access memory, read the real amplitude at the index, and simultaneously decrement and increment the index by 1 respectively to read the two power amplitude data at the adjacent positions on the left and right sides of the frequency axis.

[0272] Step 603: Verify whether the power amplitude of the initial peak frequency point exhibits local maximum characteristics. If it does not meet the local maximum characteristics, discard the initial peak frequency point.

[0273] If the local maximum characteristic is satisfied, then parabolic interpolation calculation is performed based on the power amplitude of the initial peak frequency point and the frequency points to its left and right to obtain the sub-grid frequency offset.

[0274] In this embodiment, the geometric property of the main lobe of the Hanning window approximating a parabola in the frequency domain is utilized to perform extremum derivation in the continuous domain between discrete frequency grids.

[0275] Optionally, it can be determined whether the power amplitude at the initial peak frequency point is simultaneously greater than the power amplitude at the adjacent frequency point on the left and the power amplitude at the adjacent frequency point on the right.

[0276] After satisfying the local maximum verification condition, its calculation process can be expressed as the following formula:

[0277] δ=(γ-α) / (2×(2×β-α-γ));

[0278] In the above, δ is the subgrid frequency offset, which indicates that the peak value is located δ grid widths to the right of the current discrete grid center, γ is the power amplitude of the adjacent frequency point on the right, α is the power amplitude of the adjacent frequency point on the left, and β is the power amplitude of the initial peak frequency point.

[0279] Step 604: Compensate the initial peak frequency point that satisfies the local maximum characteristic using the sub-grid frequency offset to obtain the accurate interference frequency;

[0280] For example, the bandwidth constant corresponding to a single discrete frequency grid is calculated, and the calculated subgrid frequency offset is multiplied by the bandwidth constant to obtain the frequency offset compensation value.

[0281] Next, by adding the frequency offset compensation value to the center frequency of the discrete grid corresponding to the initial peak frequency point, the interference frequency in the continuous domain can be recovered.

[0282] Step 605: Information containing the accurate interference frequency is used as the interference feature.

[0283] After the compensation calculation is completed, the parameter structure carrying high-precision frequency coordinates is stored in the feature output queue for subsequent system positioning.

[0284] Based on the same inventive concept as the above embodiments, in some communication environments, broadband interference signals often occupy a certain bandwidth, so that the residual main lobe after deconvolution compression may still span multiple adjacent frequency sampling points.

[0285] To output events with bandwidth dimensions, this embodiment provides a continuous domain clustering extraction method for interference features. The method extracts interference features from the estimated power spectrum using the noise basis estimate as a reference, and includes the following steps:

[0286] In some embodiments, in the estimated power spectrum, frequency points where the power value satisfies a preset threshold condition relative to the noise floor estimate are extracted as detection frequency points;

[0287] This involves traversing all frequency points within the effective detection bandwidth and marking the coordinates of all frequency points whose relative power ratio in the logarithmic domain exceeds 3 dB, thus forming an unordered set of detection frequency points.

[0288] In some embodiments, all extracted detection frequency points are subjected to continuous domain merging processing, i.e.:

[0289] Adjacent detection frequency points whose frequency spacing does not exceed the pre-configured maximum allowable gap are grouped into the same interference event;

[0290] As an example, the maximum allowed gap can be set to 2 frequency index units.

[0291] Furthermore, the detection frequency points in the set are scanned in ascending order of frequency;

[0292] If the difference between the current detection frequency and the previous detection frequency on the array index is less than or equal to the maximum allowable gap, it is determined that the two frequency points belong to the energy residue of the same interference source and they are bound as the same interference event.

[0293] Conversely, if the difference is greater than the maximum allowable gap, the merging logic is cut off, and a new sequence of interference events is started.

[0294] In some embodiments, for each interference event generated by merging, the estimated bandwidth of the event is determined based on the bandwidth span of the first and last detection frequency points within the interference event;

[0295] For each independent interference event that has been clustered, read the index of the starting frequency point and the index of the ending frequency point to deduce the effective spectral width it occupies.

[0296] That is, B=(k _L -k _1 +1)×F _s / N _FFT ;

[0297] In the above, B can be the estimated bandwidth of the event, and k _L It can be the frequency index of the end detection frequency point within the interference event, k _1 It can be the frequency index of the starting detection frequency point within the interference event, F _s It can be the sampling frequency of the analog-to-digital converter, N _FFT It can be the total number of discrete points in the frequency domain transformation.

[0298] In some embodiments, a weighted centroid calculation is performed based on the power values ​​of each detection frequency point within the interference event to extract the event center frequency;

[0299] Accordingly, the frequency coordinates of each detection frequency point within the interference event are multiplied by their respective power amplitudes. After summing all the products, the result is divided by the sum of the power amplitudes of all detection frequency points within the event. This can shift the output event center frequency towards the frequency point with the most concentrated energy, thus making it closer to the radio frequency carrier frequency of the interference source.

[0300] In some embodiments, the maximum power value within an interference event is extracted as the event peak intensity;

[0301] Accordingly, the power amplitude of all detection frequency points contained in the current interference event is traversed, and the maximum value is locked and recorded as the event peak intensity.

[0302] In some embodiments, the set of estimated bandwidth, center frequency, and peak intensity of each interference event is used as the interference feature.

[0303] In this scheme, each output interference feature simultaneously carries parameters in three dimensions: bandwidth, center coordinates, and intensity, which meets the engineering data requirements of the host computer for spectrum planning and allocation.

[0304] Based on this, since harsh external environments may induce abnormally dense concurrent interference sources, in order to avoid bus congestion caused by overloading the output data volume, this embodiment also outputs interference characteristics as detection results, specifically including:

[0305] For example, all interference events are sorted in descending order of peak intensity.

[0306] Furthermore, before the end of the detection period, the hardware sorting network uses the event peak intensity attribute in the triplet as the comparison key to perform a fast descending sort operation on all extracted interference events.

[0307] For example, determine whether the total number of disorder events after sorting exceeds the pre-configured maximum number of reports;

[0308] Accordingly, the total number of individual interference events contained in the current array is counted and compared with the maximum number of reports pre-configured in the register.

[0309] In some optional schemes, the pre-configured maximum number of reports Θ can be set to 64.

[0310] For example, if the total number does not exceed Θ, then all interference events will be output as the detection result;

[0311] When the number of actual detected interference events is less than or equal to Θ, the enable signal of all data output channels is turned on, and the full amount of feature data is transmitted to the output interface.

[0312] For example, if the total number exceeds Θ, the interference events ranked within the maximum number of reports are extracted as the detection results and output in descending order.

[0313] When the total number exceeds Θ, the read access to subsequent storage blocks is cut off through the address control mechanism. Only the top Θ interference events with the highest energy threat level are extracted, encapsulated, and sent as the final detection result. The remaining low-intensity interference entities are directly discarded.

[0314] As an optional implementation, the continuous domain merging and weighted centroid calculation logic units are disabled. When a single detection frequency point with a relative power greater than a preset threshold is detected, its index coordinates are directly converted into frequency values ​​along with its intensity and reported in isolation, skipping some stages. This scheme is suitable for situations where sensing bandwidth is not required and logic resources are limited.

[0315] In this application, based on local maximum parabolic interpolation, subgrid frequency offset compensation is performed to improve the center frequency extraction accuracy to the subgrid level; and in conjunction with a continuous domain clustering and merging strategy, scattered over-threshold frequency points are reconstructed into entity events with bandwidth, weighted centroid and peak intensity, which meets the resource allocation requirements of the spectrum management system.

[0316] According to one aspect of this application, the specific values ​​of the single-point convergence threshold, the global convergence threshold, and the iteration upper limit threshold can be adaptively adjusted according to the balance requirements of frequency resolution accuracy and hardware real-time processing latency in actual application scenarios.

[0317] In some scenarios, it is assumed that there are two narrowband single-tone interference signals within the effective detection bandwidth, with a frequency interval of only two unwindowed FFT frequency resolution intervals. Under the traditional Hanning windowed FFT method, since the main lobe width of the Hanning window is approximately four frequency resolution intervals, the main lobes of the two signals overlap, appearing as a merged single peak on the average power spectrum, making it impossible to distinguish the existence of two independent signals.

[0318] After the iterative deconvolution processing of this invention, the energy diffusion effect introduced by the window function is reversed and compensated. The main lobe width is compressed to a level close to that of the unwindowed state. The spectral peaks of the two signals are separated into two independent sharp peaks in the estimated power spectrum, and the sidelobe energy remains at a low level, thus achieving the effect of restoring high frequency resolution while maintaining low sidelobes.

[0319] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. 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 this application. Therefore, this application 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 method for detecting interference in a broadband satellite communication system, characterized in that, include: Obtain pre-configured system operating parameters, and collect broadband received signals in segments based on system operating parameters to obtain segmented observation data; Windowing and frequency domain transformation are performed sequentially on the segmented observation data to obtain the segmented power spectrum; The average power spectrum is generated by calculating the mean of the segmented power spectrum within a preset period. Perform iterative deconvolution on the average power spectrum to obtain the estimated power spectrum; The noise floor estimate is determined based on the estimated power spectrum, and the interference features are extracted from the estimated power spectrum with the noise floor estimate as a reference. Output interference features as the detection result.

2. The method according to claim 1, characterized in that, The system operating parameters include the pre-configured refresh cycle, master clock frequency, and sampling frequency.

3. The method according to claim 2, characterized in that, Based on system operating parameters, the broadband received signal is acquired in segments to obtain segmented observation data, which specifically includes: The segmented sampling interval is calculated based on the sampling frequency, the master clock frequency, and the pre-configured single-segment sampling length. Based on the refresh cycle, master clock frequency, and segmented sampling interval, determine the logarithmic segment number exponent with reserved timing margin; The total number of segments is determined using the logarithmic segment number exponent; Constrained by the total number of segments and the segment sampling interval, discontinuous segmented acquisition is performed on the broadband received signal to obtain segmented observation data.

4. The method according to claim 1, characterized in that, Before performing iterative deconvolution on the average power spectrum, a frequency response pre-equalization step is also included: The average power spectrum is processed by a moving median filter to extract the noise floor envelope; Based on the noise floor envelope and its mean, the equalization coefficients used to compensate for system channel fluctuations are determined. The average power spectrum is weighted and corrected using the equalization coefficient to obtain the pre-equalized power spectrum; The pre-equalized power spectrum is used as input data for performing iterative deconvolution processing.

5. The method according to claim 1, characterized in that, Perform iterative deconvolution on the average power spectrum to obtain the estimated power spectrum, including: Obtain the point expansion function generated based on the pre-configured window function; The iterative power spectrum is initialized with the average power spectrum; In each iteration, the desired observation spectrum is calculated based on the convolution operation between the iterative power spectrum and the point spread function; An update factor is generated based on the ratio of the average power spectrum to the expected observed spectrum. The current iterative power spectrum is updated using a weighted update factor; The iterative power spectrum output when the iteration termination condition is met is used as the estimated power spectrum.

6. The method according to claim 5, characterized in that, The process of determining whether the iteration termination condition is met includes: Extract the maximum and minimum values ​​of all update factors generated in a single iteration; Calculate the range between the maximum and minimum values; When the range is less than the pre-configured global convergence threshold, the iteration termination condition is met and an instruction to stop the iteration is triggered.

7. The method according to claim 1, characterized in that, Determining the noise floor estimate based on the estimated power spectrum includes: In the estimated power spectrum, the target spectral line sequence corresponding to the pre-configured effective detection bandwidth is extracted; Based on the power distribution characteristics of the target spectral line sequence, a reference power value representing the background white noise level is determined; The reference power value is used as the noise floor estimate.

8. The method according to claim 1, characterized in that, Using the noise floor estimate as a reference, interference features are extracted from the estimated power spectrum, including: In the estimated power spectrum, the initial peak frequency points where the power value relative to the noise floor estimate satisfies the preset threshold condition are selected; Extract the power amplitude at the initial peak frequency point, as well as the power amplitude at the adjacent left and right frequency points; After verifying that the power amplitude at the initial peak frequency point exhibits local maximum characteristics, parabolic interpolation calculation is performed based on the power amplitude at the initial peak frequency point and the frequency points to its left and right to obtain the sub-grid frequency offset. The initial peak frequency point is compensated by using the sub-grid frequency offset to obtain the accurate interference frequency; Information containing interference frequencies is used as interference characteristics.

9. The method according to claim 1, characterized in that, Using the noise floor estimate as a reference, interference features are extracted from the estimated power spectrum, including: In the estimated power spectrum, frequency points where the power value relative to the noise floor estimate satisfies a preset threshold condition are extracted and used as detection frequency points. Perform continuous domain merging processing on all extracted detection frequency points: group adjacent detection frequency points whose frequency interval does not exceed the pre-configured maximum allowable gap into the same interference event; For each interference event generated by merging, the estimated bandwidth of the event is determined based on the bandwidth span of the first and last detection frequency points within the interference event; Weighted centroid calculation is performed based on the power values ​​of each detection frequency point within the interference event to extract the event center frequency; The maximum power value within the interference event is extracted as the event peak intensity; The set of estimated bandwidth, center frequency, and peak intensity of each interference event is used as the interference feature.

10. The method according to claim 1, characterized in that, Windowing is applied to the segmented observation data, specifically including: A window function with real even symmetry is selected as the windowing operator to process the segmented observation data, so that in the subsequent iterative deconvolution process, the Fourier transform result of the point spread function generated based on the window function is a real number sequence.

Images