Continuous signal processing methods, systems, and storage media suitable for satellite communications

Abstract

This invention discloses a continuous signal processing method, system, and storage medium suitable for satellite communication. The method acquires a continuously received signal; performs coarse frequency offset estimation and compensation on the continuously received signal to obtain a frequency offset compensated signal; performs symbol timing recovery on the frequency offset compensated signal to obtain an optimal sampling sequence; performs modulation identification on the optimal sampling sequence to determine the modulation type of the optimal sampling sequence; combines the modulation type to perform fine frequency offset and carrier phase estimation compensation on the optimal sampling sequence to obtain a phase compensated sequence; performs multi-path candidate phase rotation and channel decoding decision on the phase compensated sequence, eliminates phase ambiguity using decoding convergence features, and obtains decoded data; and filters the decoded data based on a preset frame boundary flag to output valid service data. This invention can reduce physical layer frame structure overhead and improve spectrum utilization efficiency and link carrying capacity.

Description

Technical Field

[0001] This invention relates to the field of satellite communication signal processing technology, and in particular to a continuous signal processing method, system and storage medium suitable for satellite communication. Background Technology

[0002] Satellite communication plays a significant role in high-volume transmission scenarios such as emergency rescue, public safety, and large-scale data aggregation due to its wide coverage, strong independence, and high resilience.

[0003] In existing satellite communication systems, to achieve reliable carrier synchronization, symbol timing recovery, and channel parameter estimation at the receiver, auxiliary information such as pilot signals, synchronization words, or preamble sequences need to be periodically inserted into the physical layer frame structure at a fixed proportion. This processing mechanism provides a stable parameter estimation benchmark, but it introduces high physical layer protocol overhead, leading to a decrease in the proportion of effective data payload.

[0004] Therefore, it is necessary to study a data processing method that can reduce the consumption of auxiliary signal resources and improve demodulation stability under complex communication conditions. Summary of the Invention

[0005] Purpose of the invention: To propose a continuous signal processing method, system, and storage medium suitable for satellite communication, in order to solve the above-mentioned problems.

[0006] Technical solution: Firstly, a continuous signal processing method suitable for satellite communication is provided, including:

[0007] Acquire continuous received signals;

[0008] A coarse frequency offset estimate is performed on the continuously received signal to obtain a coarse frequency offset estimate value. Based on the coarse frequency offset estimate value, frequency offset compensation is performed on the continuously received signal to obtain a frequency offset compensated signal.

[0009] Symbol timing recovery is performed on the frequency offset compensation signal to obtain the optimal sampling sequence;

[0010] Modulation identification is performed on the optimal sampling sequence to determine the modulation type of the optimal sampling sequence;

[0011] The optimal sampling sequence is subjected to fine frequency offset estimation, and the optimal sampling sequence is subjected to phase estimation and compensation in combination with the modulation type to obtain the phase-compensated sequence;

[0012] The phase compensation sequence is subjected to multi-path candidate phase rotation and channel decoding decision, and the phase ambiguity is eliminated by using the decoding convergence feature to obtain the decoded data;

[0013] Data filtering based on frame boundary markers is performed to output valid business data.

[0014] In a second aspect, a continuous signal processing system suitable for satellite communication is provided. The system includes a processor coupled to a memory, the memory storing instructions, and the instructions being executed by the processor to implement the method in any possible implementation of the first aspect.

[0015] Thirdly, a computer-readable storage medium is provided, on which a computer program is stored, which, when run on a computer, implements the method in any possible implementation of the first aspect.

[0016] Beneficial effects: Reduces physical layer frame structure overhead, improves spectrum utilization efficiency and link carrying capacity. Attached Figure Description

[0017] Figure 1 This is a flowchart of the continuous signal processing method applicable to satellite communications according to this application.

[0018] Figure 2 This is a flowchart illustrating how the coarse frequency offset estimate is obtained in this application.

[0019] Figure 3 This is a flowchart for obtaining the optimal sampling sequence in this application.

[0020] Figure 4 This is a comparison diagram of the physical layer structure of a continuous signal processing method in another embodiment of this application.

[0021] Figure 5 This is a flowchart of continuous signal processing in another embodiment of this application.

[0022] Figure 6 This is a flowchart of sampling during symbol timing in another embodiment of this application.

[0023] Figure 7 This is a flowchart of symbol timing in another embodiment of this application.

[0024] Figure 8 This is a flowchart of a timed correction in another embodiment of this application. Detailed Implementation

[0025] 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.

[0026] Based on the technical background, existing satellite communication systems struggle to balance high spectral efficiency and synchronous demodulation under complex channel conditions when facing high-volume continuous transmission scenarios, especially sudden operational situations. There are limitations between control protocol overhead and maintaining tracking stability.

[0027] For example, in complex modulation scheme hybrid transmissions and long-term continuous transmissions, the continuous dynamic drift of the crystal oscillators at both the transmitting and receiving ends, as well as the accumulation of Doppler frequency shift, can lead to a decrease in the convergence accuracy of parameter tracking. Therefore, it is necessary to maintain or increase the insertion density of auxiliary signals. This operation limits the maximum spectral utilization of the satellite system and increases the overall link burden.

[0028] To solve these problems, combined with Figures 1 to 3 The present invention will be specifically described through the following embodiments.

[0029] like Figure 1 As shown, this embodiment proposes a pilot-free continuous signal processing method suitable for satellite communication, namely, a complete continuous signal processing and recovery process that does not rely on explicit physical pilots in high-volume burst transmission scenarios, specifically including:

[0030] Step 101: Acquire continuous received signals.

[0031] In this embodiment, the continuously received signal refers to the baseband or intermediate frequency digital signal after frequency conversion and analog-to-digital conversion at the receiver's radio frequency front-end. In applications such as emergency rescue or high-volume video backhaul, satellite communication systems employ a frame structure without physical pilots at the transmitting end to maximize spectrum utilization efficiency. This means the signal sequence only contains service data symbols and a small number of frame delimiters, without inserting periodic pilot sequences or synchronization words. However, due to Doppler frequency shift and local oscillator frequency deviation between the transmitting and receiving ends of the satellite channel, the acquired continuously received signal carries a significant carrier frequency offset.

[0032] Step 102: Perform coarse frequency offset estimation on the continuously received signal to obtain a coarse frequency offset estimate. Based on the coarse frequency offset estimate, perform frequency offset compensation on the continuously received signal to obtain a frequency offset compensated signal. Blind estimation is performed using the nonlinear characteristics of the signal, i.e., extracting the spectral peak positions reflecting the frequency offset and converting them into a coarse frequency offset estimate. Correspondingly, use this coarse frequency offset estimate to generate a reverse-rotated complex exponential sequence, multiply it by the original continuously received signal, and shift the signal's spectrum back to near the baseband to obtain the frequency offset compensated signal. In this embodiment, since symbol timing recovery and modulation identification are sensitive to frequency offset, the influence of large frequency offsets needs to be eliminated before the main processing flow begins.

[0033] Step 103: Perform symbol timing recovery on the frequency offset compensation signal to obtain the optimal sampling sequence.

[0034] The symbol timing recovery in this embodiment is used to find the sampling moment when the energy of each symbol pulse is most concentrated, and to extract the discrete symbol sequence, when there is a slight deviation in the transmit and receive clocks. During continuous signal transmission, small clock deviations accumulate over time, resulting in timing drift.

[0035] Step 104: Modulation identification is performed on the optimal sampling sequence to determine the modulation type of the optimal sampling sequence.

[0036] In satellite communication systems, the transmitter may dynamically change the modulation order based on channel conditions. Before demodulation, the receiver needs to determine the current modulation type. The system utilizes the statistical characteristics of the optimal sampling sequence, such as higher-order moment features or constellation diagram clustering features, to analyze its amplitude or phase distribution and blindly identify whether the current signal uses Phase Shift Keying (PSK), Quadrature Amplitude Modulation (QAM), or Combined Amplitude and Phase Shift Keying (APSK), outputting the corresponding modulation order as the modulation type.

[0037] Step 105: Perform fine frequency offset estimation on the optimal sampling sequence, and combine it with the modulation type to perform phase estimation and compensation on the optimal sampling sequence to obtain the phase-compensated sequence.

[0038] Specifically, after coarse frequency offset compensation and symbol timing recovery, the sequence may still retain fine frequency offset and initial carrier phase deviation. The fine frequency offset can be calculated using the phase difference pattern between adjacent symbols. Using the modulation order corresponding to the obtained modulation type, interference from data modulation on the phase is eliminated, and the carrier phase deviation is extracted. Using the estimated fine frequency offset and carrier phase deviation, rotation compensation is applied to the optimal sampling sequence to re-aggregate the constellation points, outputting a phase-compensated sequence.

[0039] Step 106: Perform multi-path candidate phase rotation and channel decoding decision on the phase compensation sequence, and use the decoding convergence feature to eliminate phase ambiguity to obtain decoded data.

[0040] In this embodiment, the constellation points of the phase compensation sequence have been clustered. However, without pilots and using a nonlinear blind estimation algorithm, the system may face phase ambiguity issues. Specifically, the entire constellation may undergo equivalent periodic rotations, causing the decision result to differ from that of the transmitter. Based on the determined modulation type, all possible candidate phase rotation sequences are generated. These sequences are then individually or selectively fed into a channel decoder for processing, such as a low-density parity-check (LDPC) decoder. The decoder's parity-check matrix has strong constraints; the decoding iteration converges rapidly and satisfies the parity-check equation only when the input sequence has the correct phase. Utilizing this decoding convergence characteristic, the correct phase sequence can be locked, and the correct decoded data can be output.

[0041] Step 107: Filter the decoded data based on the frame boundary flag and output valid service data.

[0042] The channel decoding output is a continuous bit stream, containing valid service bits and flags for frame synchronization. The preset frame boundary flag is a pre-defined bit pattern with the beginning and end joined together, such as 01111110. By searching for this pattern in the decoded data, the start and end positions of each frame's payload are determined. After removing the frame boundary flag and performing link-layer recovery operations, the valid service data is extracted and output.

[0043] Based on the above embodiments, this application proposes a multi-order amplitude normalization adaptive frequency offset estimation method, specifically including:

[0044] Step 201: Obtain the signal amplitude value of the continuously received signal after matched filtering.

[0045] Continuously received signals require preprocessing before frequency offset feature extraction. Specifically, a raised cosine roll-off matched filter is used to filter the input baseband complex sequence to suppress out-of-band noise and restore symbol pulse shaping characteristics. The complex magnitude, i.e., the signal amplitude value, is then calculated for each discrete complex sampling point after filtering.

[0046] Step 202: Add the signal amplitude value to the preset regularization constant to obtain the corrected amplitude value.

[0047] The preset regularization constant is a positive real number, and its value can be set to 1% of the average signal amplitude. After shaping filtering and wireless channel fading, the signal amplitude of some sampling points in digital communication signals approaches zero. Directly dividing by a parameter approaching zero would amplify background noise exponentially, generating singularity interference. Therefore, after obtaining the signal amplitude value, the preset regularization constant is introduced and added to obtain the corrected amplitude value.

[0048] In some embodiments, the method includes: performing matched filtering and amplitude normalization on the continuously received signal to obtain a normalized received signal;

[0049] For the normalized received signal, perform at least one nonlinear phase doubling transformation of a predetermined order to obtain a phase doubling sequence;

[0050] The phase-doubling sequence is transformed to the frequency domain, and amplitude spectrum calculation and peak search are performed. The frequency offset features corresponding to the peak positions of the amplitude spectrum are extracted as coarse frequency offset estimates. Specifically, this includes:

[0051] Step 203: Perform complex division normalization on the continuously received signal after matched filtering using the corrected amplitude value to obtain a normalized received signal, so as to suppress noise amplification at low amplitude sampling points.

[0052] Specifically, the complex data output from the matched filter is divided by the correction amplitude value. This operation smooths out amplitude fluctuations in non-constant envelope modulated signals, mapping the signal envelope to an approximately constant envelope state. The complex division normalization is as follows:

[0053] x _norm (n)=x(n) / (|x(n)|+ε);

[0054] Where, x _norm x(n) is the amplitude-normalized sequence, i.e., the normalized received signal; x(n) is the continuous received signal sequence after matched filtering; |x(n)| is the complex modulus of the sequence, i.e. the signal amplitude; and ε is a preset regularization constant.

[0055] Step 204: Perform multiple nonlinear phase doubling transformations of different candidate orders in parallel to obtain multiple candidate phase doubling sequences.

[0056] Specifically, nonlinear frequency offset estimation using a fourth-order operation is difficult to be compatible with all high-order modulations in scenarios where the modulation scheme is unknown. For example, while a fourth-order operation can fold the phase of a phase shift keying signal, it can cause frequency offset energy splitting for high-order modulations, leading to errors in the extracted estimates. In this embodiment, the system sets candidate orders of 2, 4, and 8, and performs second, fourth, and eighth-order operations on the normalized received signal, respectively. The input signal already has a constant envelope, and the nonlinear phase doubling transformation is equivalent to performing a corresponding multiplication operation on the signal phase. Therefore, three independent data streams, i.e., multiple candidate phase doubling sequences, can be output in parallel.

[0057] Step 205: Convert each candidate phase doubling sequence to the frequency domain and calculate the peak-to-average power ratio of the amplitude spectrum corresponding to each candidate phase doubling sequence.

[0058] After obtaining multiple candidate phase-doubling sequences, a Fast Fourier Transform is performed on each sequence to calculate the corresponding spectral amplitude, resulting in multiple amplitude spectra. To quantitatively evaluate the folding effect of each candidate order on the modulation phase, in this embodiment, the ratio of the peak value to the average power of each amplitude spectrum, i.e., the peak-to-average power ratio (PAPR), is calculated using the following formula:

[0059] PAR _M =N×max(|Y _M (k)| 2 ) / ∑(|Y _M (k)| 2 );

[0060] Among them, PAR _M Y is the peak-to-average ratio (PAPR) metric for order M, N is the number of points in the Fast Fourier Transform, and Y... _M(k) is the complex value in the frequency domain at the k-th frequency point, max(|Y _M (k)| 2 ) represents the search for the maximum power spectral density value among all frequency points, ∑(|Y _M (k)| 2 ) represents the sum of the power spectral density of all frequency points within the entire observation window.

[0061] Step 206: According to the order from low to high, select the optimal order from the candidate orders where the peak-to-average ratio (PAPR) is greater than the preset detection threshold for the first time.

[0062] Decision logic is used to filter multiple peak-to-average ratio (PAR) metrics obtained through parallel computation. A preset detection threshold is used to distinguish between pure noise environments and environments with effective single-frequency components. Under pure white noise conditions, the expected PAR of the Fast Fourier Transform (FFT) output amplitude is approximately the natural logarithm of the point. In this embodiment, the preset detection threshold is set to 50 times the expected value of pure noise. When determining the optimal order, a search order from low to high is adopted. If the PAR of the current order is greater than the preset detection threshold, the search stops and the current order is locked as the optimal order.

[0063] In some optional implementations, if the current channel conditions are poor, causing the peak-to-average ratio (PAR) of all candidate orders to fail to exceed the preset detection threshold, the system will trigger an abnormal backoff mechanism, forcibly backoff to the optimal order of power 4, and hand over the accuracy compensation task to the time-domain fine frequency offset estimation module.

[0064] In other embodiments, if the communication protocol has indicated that the current transmitter uses a preset low-order modulation, the parallel computation and measurement comparison process can be skipped, and a fourth-order nonlinear transformation can be used as an alternative to reduce instantaneous computing power overhead.

[0065] Step 207: Calculate the coarse frequency offset estimate based on the peak position of the amplitude spectrum corresponding to the optimal order.

[0066] After determining the optimal order, the frequency index of the maximum peak value of the amplitude spectrum corresponding to the optimal order is extracted. Since the nonlinear phase doubling transformation amplifies the original frequency offset by a certain factor, it is necessary to divide by the optimal order when converting to the actual physical frequency offset. The calculation method for the coarse frequency offset estimate is as follows:

[0067] f _coarse =k _star ×f _s / (M _star ×N);

[0068] Among them, f _coarse k is the coarse frequency offset estimate of the target output. _starThis is the corrected index value for the amplitude spectrum peak position corresponding to the optimal order. The corrected index value is obtained by mapping the original spectrum peak index to zero frequency center on the frequency axis. _s Symbol sampling rate, M _star The optimal order is defined by N, where N is the number of points in the frequency domain transform. After the calculation is complete, the coarse frequency offset estimate is output to the frequency offset compensation module to generate a reverse rotation vector, which is used to correct the data stream.

[0069] Based on the above embodiments, this application also proposes a two-state predictive symbol timing tracking method, specifically including:

[0070] Step 301: Divide the frequency offset compensation signal into continuous group data.

[0071] In continuous data transmission scenarios, it is difficult for the receiving end to acquire all the data at once and process it offline. A fixed-length data buffer window needs to be maintained. The continuously input complex sampling sequence, after coarse frequency offset compensation, is truncated into multiple independent processing blocks, i.e., grouped data. The length of each group of data is set to include a fixed number of symbol periods. Dividing the data into groups provides the block processing algorithm with a set of observation data samples, allowing the estimation of timing bias to smooth additive noise within the integration time.

[0072] Step 302: Construct a dual-state tracking model that includes timing offset state and drift rate state.

[0073] Specifically, conventional symbol timing recovery typically estimates a single timing offset and directly uses the estimate of the current packet for compensation of the next packet. Due to physical frequency deviations in the hardware clock crystals at both the transmitting and receiving ends, the timing sampling point will drift linearly over time. This single-state open-loop compensation structure causes the compensation value to consistently lag behind the actual drift. To suppress this hysteresis effect, the state space dimension is expanded, and modeling of the drift rate is introduced to construct a two-state tracking model.

[0074] Step 303: Using a dual-state tracking model, perform feedforward prediction and feedback correction on the current grouped data, and update the timing offset state and drift rate state, specifically including:

[0075] Based on the timing offset and drift rate status of the previous group of data, calculate the predicted timing offset of the current group of data.

[0076] A dual-state tracking model is used to perform feedforward prediction. Specifically, the offset state and drift trend of the previous time step are linearly extrapolated to predict the theoretical sampling deviation when the current data block arrives. The formula for calculating the predicted timing offset is as follows:

[0077] tau _pred =tau _k +tau_dot_k ×DT;

[0078] Among them, tau _pred tau represents the predicted timing offset calculated from the current grouped data. _k Indicates the timing offset status of the previous group of data, tau _dot_k DT represents the drift rate state of the previous data group, and DT represents the time interval between two adjacent data groups. In this embodiment, DT is normalized to a value of 1. Through this prediction mechanism, the originally delayed post-compensation can be transformed into a delay-free pre-compensation.

[0079] Step 304: Use the predicted timing offset to pre-compensate the current grouped data, and estimate the timing bias of the pre-compensated grouped data to obtain the observation timing bias.

[0080] After obtaining the predicted values, an interpolation filter is used to resample the current grouped data based on the predicted timing offset, thus offsetting most of the known clock drift. After pre-compensation, a non-data-aided timing estimation algorithm is employed to blindly measure the remaining bias. Specifically, the classical square envelope nonlinear timing bias estimation module is invoked to calculate the instantaneous phase difference between the actual and ideal sampling points of the current data block on the pre-compensated data, outputting the observed timing bias. This estimation algorithm does not depend on specific modulation rules and can extract timing errors even at low to medium signal-to-noise ratios.

[0081] Step 305: Calculate the prediction residual based on the observation timing deviation and the prediction timing offset.

[0082] The system needs to quantify the deviation between the prediction model and the actual physical channel. Specifically, this is calculated by subtracting the prediction timing offset from the observation timing offset. The resulting difference is the prediction residual. The prediction residual reflects environmental random jitter or sudden channel changes that the current dual-state tracking model fails to fully capture.

[0083] In other words, since the predicted timing offset has been used to pre-compensate the current group data, the observation timing deviation has already reflected the amount of environmental random jitter or sudden channel changes that are difficult for the dual-state tracking model to capture.

[0084] Step 306: Obtain the modulation type generated in the subsequent steps.

[0085] In this embodiment, the symbol timing recovery module has difficulty determining the current signal modulation order. However, in the complete pipelined processing, the modulation identification module located downstream of it can output a category indication. The symbol timing recovery module receives and caches the modulation type returned by this module via an internal data bus or shared memory mechanism. In other words, different modulation schemes have different constellation distributions in the complex plane, and the degree of fluctuation in their envelopes determines the magnitude of the self-noise introduced in the timing deviation estimation process.

[0086] Step 307: In the first stage where the modulation type is not obtained, the state of the dual-state tracking model is updated using the preset default correction gain parameters.

[0087] During initial power-on or the reacquisition phase after signal loss, the modulation recognition module at the back end of the pipeline has not yet completed its first convergence and cannot provide feedback information, remaining in a blind start state, i.e., the first stage. The system loads a constant value with high fault tolerance as a preset default correction gain parameter. Using this default parameter, the prediction residual is weighted and superimposed onto the existing state, enabling the model to maintain its tracking stability even with unknown signal envelope characteristics, preventing loop divergence due to parameter mismatch.

[0088] Step 308: After the modulation type is first obtained, the second stage is triggered. The default correction gain parameter is immediately switched to a target correction gain parameter that matches the envelope characteristics of the current modulation type. The target correction gain parameter is then used for the state update of the packet data. In other words, based on the envelope characteristics of the modulation type, the preset correction gain parameter is adaptively switched, and the switched correction gain parameter is used for the state update of the packet data.

[0089] When the bus interface receives valid modulation type data, the blind start state is immediately terminated, and the system switches to normal tracking cycle. The system has a built-in lookup table that stores pre-optimized gain value combinations for different modulation envelope characteristics. For constant envelope modulation, its timing estimate has low internal self-noise; a smaller value is retrieved from the lookup table as the target correction gain parameter, enhancing the filtering and smoothing capability against external Gaussian white noise. For non-constant envelope modulation, its amplitude fluctuations easily generate significant jitter noise; a larger value can be retrieved to improve its response to actual drift. Accordingly, parameter switching is completed instantaneously at the boundaries of adjacent data packets, without requiring a smooth transition.

[0090] Step 309: Using the correction gain parameter, the prediction residual is applied to the timing offset state and drift rate state of the previous group of data respectively to obtain the updated timing offset state and drift rate state.

[0091] Based on the determined gain parameters, the closed-loop correction equation is executed to complete the state iteration, and the update formulas for calculating the offset and rate are as follows:

[0092] tau _k_plus_1= tau _pred +K _1× e _k_plus_1 ;

[0093] tau _dot_k_plus_1= tau _dot_k +K _2 ×e _k_plus_1 ;

[0094] Among them, tau _k_plus_1 For the updated timing offset state, tau _dot_k_plus_1 For the updated drift rate state, e _k_plus_1 To calculate the predicted residuals, K _1 K is the first correction gain parameter used to control the position response speed. _2 This is the second correction gain parameter that controls the smoothness of the rate. Assume the previous state tau measured at the current time... _k The drift rate is 0.5, tau _dot_k The value is 0.04, so the predicted offset tau is calculated. _pred =0.5 + 0.04 × 1 = 0.54. The residual e is obtained through blind testing. _k_plus_1 The value is 0.06. The system is in the second stage with a known modulation type, and the signal is identified as non-constant envelope modulation. K is adaptively allocated. _1= 0.25, K _2= 0.015. Perform the update and calculate tau. _k_plus_1= 0.54 + 0.25 × 0.06 = 0.555, tau _dot_k_plus_1= 0.04 + 0.015 × 0.06 = 0.0409.

[0095] Step 310: When the cumulative amount of the predicted timing offset reaches the integer sampling interval, trigger the integer offset jump and adjust the starting position of data extraction.

[0096] Specifically, due to oversampling during baseband processing, the timing offset is a sub-sampling precision floating-point number. As the drift accumulates unidirectionally, the absolute value of the predicted timing offset increases and spans a full sampling period. When the difference between the current predicted offset and the baseline value at the time of the last jump is ≥1, an out-of-bounds reset is performed. The internal data extraction window is moved forward or backward by one physical memory address, causing a hard jump of integer steps at the starting sampling point. The corresponding integer part of the floating-point state variables in the tracking model is subtracted, ensuring that the floating-point compensation calculation always remains within a high linearity region of ±0.5 sampling periods.

[0097] Step 311: Based on the updated timing offset state, perform timing compensation extraction on the grouped data to obtain the optimal sampling sequence.

[0098] After all predictions and corrections, the dual-state tracking model outputs the absolute timing deviation of the current group confirmation. Specifically, this state value is imported into the digital interpolation resampling module. The interpolation module reconstructs the amplitude value of the original analog waveform at the position deviating from the ideal grid by calculating a weighted polynomial of several adjacent discrete sampling points. After performing this extraction operation, the frequency reduction conversion from oversampling rate to symbol rate is completed, redundant sampling points are eliminated, and the optimal sampling sequence aligned with the peak of pulse energy is output to the next-level module.

[0099] In some alternative implementations, if the FPGA's internal multiplier resources are insufficient, the system can disable the drift rate state update logic. The two-state tracking model degenerates into a pure first-order open-loop feedforward structure. Forced setting of K... _2= 0, K _1 Set to 1, and update the timing offset status register by overwriting the observation timing deviation.

[0100] In the scenario of continuous transmission without pilot, this embodiment eliminates timing drift lag caused by crystal oscillator frequency deviation by constructing a two-dimensional state space; and dynamically and adaptively adjusts the control parameters of the tracking loop by utilizing the modulation characteristics of cross-module feedback.

[0101] In some embodiments, under continuous transmission conditions without physical pilots, the modulation order of the signal can be blindly identified through statistical characteristics and clustering algorithms, and the modulation order can be used to perform phase estimation and compensation with low hardware complexity, specifically including:

[0102] Modulation identification is performed on the optimal sampling sequence to determine its modulation type. Specifically, a complex sequence recovered by symbol timing is acquired. The statistical higher-order moment eigenvalues ​​of the optimal sampling sequence are calculated and compared with a preset classification threshold. If the statistical higher-order moment eigenvalues ​​are greater than the preset classification threshold, the current sequence is determined to be constant-envelope phase-shift keying modulation; otherwise, the current sequence is determined to be non-constant-envelope orthogonal amplitude modulation or amplitude-phase combined modulation.

[0103] After determining that the modulation is phase-shift keying (PSK), the system further identifies the specific modulation order. The system calculates the nonlinear envelope characteristics for different assumed orders. The calculation relationships are as follows:

[0104] R _M =∣(∑(r(n) M ))∣ / ∑(∣(r(n))∣ M );

[0105] Among them, R _MLet R be the nonlinear envelope characteristic quantity assumed to be of order M, r(n) be the nth complex symbol in the optimal sampling sequence, M be the assumed phase shift keying modulation order, whose value set includes 2, 4, and 8, |•| be the function for taking the complex modulus, and ∑() be the function for summing all symbols within the observation window. The system iterates through the value set and selects the value that makes R... _M The M value, which approaches 1, is used as the actual modulation order.

[0106] After determining that the modulation is non-constant envelope, the modulus values ​​of each complex symbol in the best sampling sequence are extracted, and an unsupervised clustering algorithm is used for differentiation. Based on the theoretical number of amplitude rings corresponding to the candidate non-constant envelope modulation order, the number of cluster centers K is determined, and clustering is performed for each candidate modulation format. The value of the number of cluster centers K can be determined according to the actual set of modulation formats supported by the system.

[0107] Accordingly, K initial center values ​​are randomly selected. The Euclidean distance between the magnitude of each symbol and each center value is calculated, and the symbols are assigned to the set containing the center value with the smallest distance. The mean of the magnitudes within each set is recalculated as the new center value, and the distance calculation and assignment are repeated until the center values ​​of each set converge. The converged set center values ​​are obtained, and the difference metric between adjacent center values ​​is calculated. When the difference metric satisfies the equidistant distribution constraint, the current sequence is determined to be orthogonal amplitude modulation; otherwise, it is determined to be amplitude-phase joint modulation.

[0108] The determination method for the equidistant distribution constraint can be set according to the set of modulation formats supported by the system. For example, for the QAM and APSK formats supported by the system, the decision threshold for the amplitude loop spacing ratio can be determined in advance through simulation and used as the implementation parameter for the equidistant distribution constraint. After determining the modulation method and modulation order, the corresponding phase symmetry order is determined. For M-order phase shift keying modulation, the phase symmetry order is equal to the modulation order M; for quadrature amplitude modulation, the phase symmetry order is 4; for combined amplitude and phase modulation, the phase symmetry order is determined according to the rotational symmetry of its constellation diagram. The determined modulation method, modulation order, and phase symmetry order are uniformly output as the modulation type.

[0109] The determined modulation scheme and its corresponding order are output as the modulation type. That is, a fine frequency offset estimate is performed on the optimal sampling sequence, and combined with the modulation type, phase estimation and compensation are performed on the optimal sampling sequence to obtain the phase-compensated sequence.

[0110] In this embodiment, a conjugate multiplication mechanism is used to construct a phase difference statistic to extract the residual frequency offset. The conjugate product of adjacent fixed-interval symbols in the optimal sampling sequence is calculated, and the time average of the product results within the observation window is calculated. The phase component of this time average is obtained and divided by the time interval parameter to obtain a precise time-domain frequency offset estimate. That is, this precise time-domain frequency offset estimate is used to perform reverse rotation compensation on the sequence to eliminate the residual linear frequency offset.

[0111] After fine-tuning frequency offset compensation, the system performs blind estimation of the initial carrier phase. Nonlinear phase estimation algorithms require performing complex multiplications on complex symbols an equal number of times as the modulation order, which consumes significant hardware multiplier resources under high-order modulation and long observation windows. This embodiment employs a calculation method based on phase angle scalar extraction. The instantaneous phase angle of the complex symbol is extracted, and then multiplied by the modulation order corresponding to the modulation type. Next, the scalar result of the multiplication is remapped to a unit vector on the complex plane, and vector accumulation is performed. The phase angle of the accumulated vector is extracted, divided by the modulation order, and the initial carrier phase estimate is obtained. The calculation formula is:

[0112] Φ _est =(1 / M _mod )×angle(∑(exp(j×M _mod ×angle(r _comp (n)))));

[0113] Where, Φ _est M is the calculated initial carrier phase estimate. _mod r is the phase symmetry order corresponding to the modulation type. _comp (n) is the complex symbol after fine frequency offset compensation, angle() is the complex phase angle extraction function, j is the imaginary unit, exp() is the complex exponential function with the natural constant as the base, and ∑() is the function to sum the symbols within the observation window.

[0114] Accordingly, after obtaining the initial carrier phase estimate, a static phase compensation factor is constructed, and all symbols within the observation window are subjected to inverse phase rotation to correct the constellation distribution of the symbols to the vicinity of the theoretical reference point, and a phase compensation sequence is output.

[0115] In some alternative implementations, if the system hardware platform is equipped with a dedicated coordinate rotation digital computer coprocessor, the phase angle extraction and unit vector mapping operations can be directly offloaded to the coprocessor for execution, freeing up the computing resources of the main processor. For low-order modulation, it can also be configured to directly perform complex multiplication, avoiding the approximation errors caused by inverse trigonometric function operations.

[0116] In the absence of physical pilots, to reduce the computational power consumption of the high-order modulation blind decoding stage, this embodiment also proposes a phase ambiguity elimination mechanism based on constellation distance pre-screening. That is, multiple candidate phases are pre-screened using a low-complexity Euclidean distance metric. The specific process includes:

[0117] Step 401: Based on the theoretical phase ambiguity number corresponding to the modulation type, perform multiple different candidate phase rotations on the phase compensation sequence to generate a multi-path candidate rotation sequence.

[0118] Specifically, in a blind reception environment, it is difficult to obtain an absolute phase reference, and the phase compensation sequence has an inherent equivalent phase rotation deviation from the modulation rule. The theoretical phase ambiguity number is determined by the pre-modulation type. For example, for 8th-order phase shift keying modulation, its theoretical phase ambiguity number is 8. Following the principle of equally dividing the circumference, multiple preset complex rotation factors are generated sequentially. Each complex symbol of the phase compensation sequence is multiplied by its corresponding complex rotation factor to obtain multiple data branches output in parallel. The candidate phase rotation calculation formula is as follows:

[0119] r _m (n)=r(n)×exp(-j×2×PI×m / M _mod );

[0120] Where, r _m r(n) is the nth complex symbol in the m-th candidate rotation sequence, r(n) is the nth complex symbol in the phase compensation sequence, j is the imaginary unit, PI is the constant of pi, and M _mod The phase symmetry order corresponding to the modulation type, i.e., the theoretical phase ambiguity number m, is the index number of the rotation branch, which ranges from 0 to M. _mod -1. Exhaustively map the actual channel phase situation within a finite domain to independent data branches, i.e., multiple candidate rotation sequences.

[0121] Step 402: For each candidate rotation sequence, calculate the distance measure from each symbol to the corresponding ideal constellation point set to obtain the constellation distance metric for each candidate rotation sequence.

[0122] When the system is in a higher modulation order state, the simultaneous operation of multiple parallel decoder modules leads to an increase in overall computational load and power consumption. To avoid this increase in computational power, a low-complexity distance metric pre-calculation is performed before the data is fed into the decoder. Based on the currently determined modulation type, the standard reference template, i.e., the set of ideal constellation points, is read from local memory. Accordingly, the squared Euclidean distance between each complex symbol and the nearest ideal constellation point in each candidate rotation sequence is calculated one by one. The squared shortest Euclidean distances of all symbols within the sequence observation window are summed, and the average value is taken. This average value is then output as a scalar value as the constellation distance metric. The calculation formula is as follows:

[0123] D_m =(1 / N _s )×∑(min(|r _m (n)-s| 2 ));

[0124] Among them, D _m Let N be the constellation distance metric for the m-th candidate rotation sequence. _s r represents the total number of complex symbols contained in the currently processed sequence. _m (n) represents the nth complex number symbol in the current sequence, s represents any reference constellation point in the set of ideal constellation points, min() is the mathematical operation to obtain the square of the shortest Euclidean distance between two points, and ∑() is the operation to iterate over all N points within the observation window. _s The calculation results are summed.

[0125] In environments with a signal-to-noise ratio greater than 5 dB, if the sequence contains the correct initial phase, its symbols will cluster tightly around the ideal constellation point, and the constellation distance metric will be approximately equal to the additive noise variance of the physical channel. If the sequence contains the incorrect initial phase, its overall constellation distribution will deviate from the ideal reference point, and the constellation distance metric will be greater than the true noise variance.

[0126] Step 403: According to the constellation distance metric, pre-screen the multiple candidate rotation sequences, extract the set number of candidate rotation sequences with the smallest distance metric, and send them as the input subset to the channel decoder for channel decoding decision.

[0127] After obtaining the scalar measure values ​​corresponding to each sequence, all constellation distance measures are sorted in ascending order. An integer K is pre-configured as a retention threshold. The K candidate rotation sequences with the smallest distance values ​​(those at the top of the sorted results) are extracted, while redundant sequences with larger measure values ​​are discarded. The extracted and retained K sequences are combined into an input subset. The value of the quantity K is set to be less than the modulation order M. _mod For example, the quantity K can be set to 2 or 3. Since the metric value of the correct phase sequence is the smallest, this retention strategy discards most redundant branches while the statistical probability of missing a correct branch is less than 10. -6 It can make the M that originally needed to be done _mod The computational scale of the next iteration of decoding is compressed to K times, which can reduce the clock cycle usage and logic resource consumption of the backend.

[0128] Step 404: In the input subset, select the sequence that converges first as the correct sequence and output the corresponding decoded data, wherein the correct sequence has eliminated phase ambiguity.

[0129] In this embodiment, a very small number of candidate sequences from the input subset are assigned to independent low-density parity-check code decoder modules for parallel iterative demodulation. The parity-check matrix product result within each decoding channel is monitored in real time. When it is detected that an input sequence can output an all-zero vector that satisfies the parity constraint after a few iterations, the decoding process of that channel is considered to have converged, the calculation of the remaining redundant parallel channels is terminated, and the bit stream output by the register inside the converged channel is extracted as the decoded data with systematic phase deviation eliminated.

[0130] Based on the above embodiments, an optional implementation method is also provided. When the current modulation order is identified as 2 or 4, the total number of initially generated candidate rotation sequences is 2 or 4. When the hardware computing power margin is sufficient, the distance metric calculation and sorting pre-screening can be bypassed directly, and all initially generated candidate rotation sequences can be directly mapped as input subsets and sent to the back-end decoder, reducing the transmission handshake delay between modules. When the modulation order is greater than or equal to 8, the distance metric calculation and pruning extraction control logic is activated.

[0131] Based on the above implementation methods, this embodiment also provides dual verification of decoded frames and cross-layer parameter adjustment feedback, including:

[0132] Step 501: Verify the reliability of the decoded data to determine whether the current frame is a highly reliable decoded frame.

[0133] In the feedforward receiver architecture, the parameter estimation error of the physical layer propagates step-by-step along the processing chain. To perform closed-loop tracking of the parameters, the decoded data output from the channel decoder is extracted for reliability determination. The codeword length of the channel code matches the link layer frame length, ensuring that each decoded bit block corresponds to a complete link layer frame. The determination logic includes a dual verification mechanism: performing a parity check matrix constraint check on the decoded data using the channel code, and performing a frame boundary marker pattern matching check on the beginning and end positions of the decoded data. The current output bitstream vector is transposed and multiplied by the parity check matrix of the low-density parity check code. If the product is an all-zero vector, the parity check matrix constraint check is considered passed. When the start and end bytes of the detected data block equal a preset bit pattern, the pattern matching check is considered passed. When both verifications pass simultaneously, the current frame is determined to be a highly reliable decoded frame.

[0134] Step 502: When a frame is determined to be a highly reliable decoded frame, channel coding and modulation operations are re-executed on the decoded data to generate a virtual reference symbol sequence corresponding to the current frame.

[0135] After obtaining the judgment result, the data frames that passed the judgment are reconstructed. Since the original received signal does not contain physical pilots, the correct decoded bitstream can be used as the absolute reference. Following the same encoding as the transmitter, the generator polynomial and constellation mapping rules are used to perform forward error correction coding and baseband modulation mapping on the decoded data. After this mapping, a noise-free baseband complex sequence, i.e., a virtual reference symbol sequence, can be generated inside the receiver. This sequence can replace the explicit pilot sequence in traditional communication systems.

[0136] Step 503: Compare the virtual reference symbol sequence with the physical layer received sequence buffered before decoding decision of the current frame, and extract the residual frequency offset feature and residual phase feature as residual physical layer parameters.

[0137] In this embodiment, due to the computational iteration time involved in the channel decoding process, a static random access memory (SRAM) needs to be configured at the output of the physical layer phase compensation module. A ping-pong buffer structure with a depth not exceeding 128 kilobits is used to temporarily store one frame of complex symbol data; this temporarily stored data is the physical layer received sequence. The physical layer received sequence and the virtual reference symbol sequence are read, and symbol-level correlation comparison is performed.

[0138] Specifically, the product of the conjugate values ​​of the physical layer received sequence and the virtual reference symbol sequence is calculated to obtain the symbol-level residual sequence. This product uses the conjugate phase to cancel the phase flip of the data modulation. Accordingly, the symbol-level residual sequence is subjected to phase angle extraction and time averaging to obtain the residual phase characteristics, and the corresponding calculation formula is as follows:

[0139] Φ _res =angle(∑(r _res (n)));

[0140] Where, Φ _res To extract the residual phase features, angle is the mathematical function for extracting complex phase angles, ∑ is the summation function within a valid one-frame data observation window, and r _res (n) is the nth complex element in the symbolic residual sequence.

[0141] Furthermore, differential complex multiplication is performed on adjacent symbols in the symbol-level residual sequence, and the average phase angle of the calculation result is extracted as the residual frequency offset feature. The calculation formula is as follows:

[0142] f _res =(1 / (2×PI×T _s ))×angle(∑(r _res (n+1)×conj(r _res (n))));

[0143] Among them, f _resTo extract the residual frequency offset features, PI is the constant of pi, and T _s The symbol period time is given by r, angle is the mathematical function for extracting the complex phase angle, ∑ is the summation function, and r is the symbol period time. _res (n+1) represents the (n+1)th complex element in the symbolic residual sequence, conj is the complex conjugate function, and r _res (n) is the nth complex element in the symbolic residual sequence.

[0144] Step 504: Using the residual physical layer parameters, the current states of the fine frequency offset estimation and phase estimation are updated through smooth feedback to obtain the estimated parameters. These are the adjusted fine frequency offset estimation values ​​and phase estimation values, which are used for subsequent frame signal processing.

[0145] After obtaining the residual physical layer parameters, which include residual frequency offset and residual phase features, a closed-loop parameter feedback process is performed from the data layer to the physical layer. To prevent residual noise in a single frame of data from causing tracking loop lock-out, a first-order exponential smoothing mechanism is used in this embodiment. Taking the frequency offset parameter as an example, the calculation formula for the smoothing feedback update is as follows:

[0146] f _new =f _old +β _f ×f _res ;

[0147] Among them, f _new To smooth the updated, precisely frequency offset estimated state value that will be applied to the next frame of the signal, f _old β represents the current fine-frequency offset estimation state value of the physical layer. _f f is a preset frequency offset smoothing update coefficient with a value between 0 and 1. _res This is the residual frequency offset feature extracted for this frame. The phase smoothing feedback update uses the same linear weighting structure, and the phase smoothing feedback update calculation formula is:

[0148] Φ _new =Φ _old +β _Φ ×Φ _res ;

[0149] Where, Φ _new To smooth the carrier phase estimation state value that will be applied to the next frame of the signal after the feedback update, Φ _old β is the estimated state value of the current carrier phase at the physical layer. _Φ Φ is a preset phase smoothing update coefficient with a value between 0 and 1. _res This represents the residual phase features extracted from this frame. β _ΦThe phase drift rate can be obtained through conventional engineering debugging based on the actual channel phase drift rate. In this embodiment, the ping-pong buffer and decoding introduce a physical time delay of one frame of data. Since the temperature drift of the crystal oscillators at both ends of the transmission and reception, as well as the Doppler change rate of the spatial channel, are slow-changing processes, the physical change of the channel parameters within this delay range is much smaller than the lower limit of the accuracy of the estimation algorithm.

[0150] As an optional implementation, during the receiver's cold start acquisition phase, or when the current channel experiences a sudden deep fading leading to a sharp drop in the signal-to-noise ratio, the bitstream output by the decoder may fail to meet dual verification, i.e., it may not meet either verification condition. In this case, the decoded data is determined to be a low-reliability frame, and the control path for feedback of residual physical layer parameters to the physical layer is immediately blocked. Accordingly, it degenerates into a pure open-loop feedforward operating mode, relying on the initial estimate extracted by an independent physical layer module for compensation demodulation, avoiding parameter divergence caused by erroneous reference symbols.

[0151] In a further embodiment, a variable-length frame delimiting mechanism based on bit-transparent transmission and sliding window is proposed, wherein a cooperative zero-bit insertion mechanism needs to be deployed at the sending end in order to enable the receiving end to delete zero bits.

[0152] Step 601: Perform zero-bit deletion on the decoded data to obtain the recovered bitstream.

[0153] The decoded data, which includes frame overhead and payload, is obtained from the output of the channel decoder. Before extracting the payload, the transparent transmission redundancy bits introduced by the transmitter are reverse-cleaned, i.e., zero-bit deletion is performed. The output bit sequence is restored to the original transmitter service data, which is used as the restored bit stream.

[0154] Step 602: Perform a continuous bit scan on the decoded data.

[0155] The baseband digital signal processor or FPGA at the receiving end starts the hardware shift register to perform bit-by-bit state monitoring on the decoded data. Through the built-in state machine, it continuously records the number of consecutive occurrences of preset polarity bits in the input bit stream.

[0156] Step 603: When a second state bit is detected immediately following a preset number of consecutive first state bits, the second state bit is automatically deleted to eliminate the probability of a preset frame boundary flag appearing within the valid service data.

[0157] Specifically, the first state bit is 1, the second state bit is 0, and the preset number of consecutive bits is 5. The state machine's counter is reset to zero each time it encounters a logic 0 and incremented when it encounters a logic 1. When the state machine records 5 consecutive logic 1s and then receives 1 logic 0 in the next clock cycle, it determines that the logic 0 is a redundant bit inserted by the transmitter, triggering a hardware enable signal to delete the redundant bit. This restores the continuous high-level characteristic of the original data.

[0158] Furthermore, this deletion operation is triggered only under preset state combinations, and the triggering frequency varies for different data sources. For example, in a completely randomly distributed data stream, the average bit expansion rate of this padding and deletion mechanism is approximately 1 / 32, requiring only 8 bits of delimiting markers at the beginning and end. The corresponding physical layer frame structure transmission efficiency is higher than that of conventional satellite communication systems, making it suitable for high-volume transmission requirements.

[0159] Step 604: The recovered bit stream is shifted in step by step through a sliding window, and the bit sequence in the current sliding window is pattern matched with the frame boundary flag.

[0160] In this embodiment, after zero-bit deletion, a variable number of redundant bits are removed, making the length of the received single-frame data no longer constant, but a floating variable related to the source data content. To address this variable length characteristic, a dynamic sliding matching mechanism is employed. Specifically, a sliding window consisting of 8-bit shift registers is maintained in memory. The recovered bit stream is shifted in according to the clock cycle. In each shift cycle, an XOR operation is performed to compare whether the current 8-bit sequence within the sliding window is equal to the reference sequence 01111110.

[0161] Step 605: When a match is successful, the current position of the sliding window is determined as a candidate point for the frame boundary.

[0162] When the XOR operation result indicates a full match, a valid delimiter is determined to have been scanned. The current bit address offset or timestamp counter value is extracted and stored as a candidate frame boundary point. In this embodiment, because the preceding deletion operation has cleared the data space inside the payload, the recovered bitstream will not generate false matching features between the two real delimiters. The output of the sliding window search is determined, and there is no intra-frame false triggering.

[0163] Step 606: Extract the continuous bit sequence between two adjacent frame boundary candidate points and output it as valid service data to adapt to frame delimitation of variable-length payloads.

[0164] In this embodiment, the previously stored candidate boundary point of the previous frame is used as the start pointer for data extraction, and the latest candidate boundary point of the current frame is used as the end pointer for data extraction. The continuous bit range enclosed by these two pointer addresses is read. The data within this range is the source data after removing physical layer compensation interference and link layer encapsulation overhead. This continuous bit sequence is then used as valid service data and input to the higher-layer protocol stack for parsing and output.

[0165] As an alternative implementation, this application also proposes another continuous signal processing method.

[0166] The signal structure of conventional satellite communication consists of a preamble, a data segment, and a postamble. The data segment is not entirely composed of service symbols; discrete pilots still need to be inserted, resulting in a certain amount of open circuitry.

[0167] like Figure 4 As shown, the processed continuous signal structure contains only service symbols, omitting preambles, strobes, and discrete pilots. The service is carried by a basic unit in the form of 01111110 + service bit + 01111110, meaning the data bits begin or end with 01111110. The overall efficiency of the signal structure can reach (8192-8-8) / 8192 = 99.8%. Through an adaptive parameter estimation and modulation-independent processing mechanism for the received signal, robust identification and demodulation of various modulation methods are achieved.

[0168] Specifically, the signal processing flow is as follows: Figure 5 As shown, the received signal is divided into two paths: one for coarse frequency offset estimation and the other for subsequent main processing. In the coarse frequency offset estimation path, matched filtering is applied to the received signal to suppress out-of-band noise and restore symbol pulse shaping characteristics. Under additive white Gaussian noise (AWGN) conditions, matched filtering can achieve the theoretically optimal output signal-to-noise ratio. Frequency offset estimation is then performed on the data in the frequency domain. In this embodiment, the signal is sampled using 4 times the symbol rate, and the number of FFT points N is configurable.

[0169] Since the receiver may use multiple modulation schemes such as PSK, QAM, and APSK, this embodiment sends the fourth power of the matched-filtered data to the FFT module to find the location of the frequency peak, specifically including:

[0170] Perform a fourth power operation on the matched-filtered data r(n):

[0171] y(n)=r 4 (n);

[0172] The fourth power operation compresses the modulation phase distribution to a constant phase or a finite set, thus weakening the effect of modulation on the spectral structure.

[0173] Perform an N-point FFT on the fourth-order signal y(n) and calculate the amplitude spectrum, then search for the location of the maximum peak in the amplitude spectrum.

[0174] ;

[0175] The frequency offset value is calculated using the following formula:

[0176] ;

[0177] Where FFT{•} denotes Fast Fourier Transform, argmax(•) denotes the position of the maximum value in the sequence, and k maxIndicates the peak position of the amplitude spectrum, N represents the number of FFT points, and F... s The symbol rate, Δf 粗 This represents the coarse frequency offset estimate.

[0178] Accordingly, the coarse frequency offset estimate is used to compensate the frequency offset of the main path data. The compensated data is then processed by matched filtering and preprocessed, including power normalization, centering and standardization.

[0179] ;

[0180] Where E[•] denotes solving for the sequence mean, |•| 2 Let σ represent the squared magnitude of the sequence, σ represent the standard deviation of the sequence, x(n) represent the data after matched filtering, and x1(n) represent the data after power normalization. The mean E[x1] of the x1(n) data is calculated and centered to obtain x2(n). The x2(n) data is then standardized to obtain x3(n) so that the standard deviation is 1.

[0181] This process can unify different signal power scales, improve modulation recognition stability, avoid the impact of amplitude drift on higher-order moment calculations, and improve the convergence of symbol timing algorithms.

[0182] like Figure 6 As shown, the receiving end experiences offset during signal acquisition, or in other words, it is difficult to achieve optimal sampling. Therefore, symbol timing is used in this embodiment. Typically, the sampling offset is divided into integer multiples of the sampling offset (fluc) and fractional multiples of the offset (uu). Furthermore, since there are various modulation methods, this embodiment uses the OM symbol timing method, which is independent of the specific modulation method.

[0183] Because of the frequency deviation between the transmitting and receiving ends, in continuous signal scenarios, this deviation accumulates over time, causing symbol timing error drift and dynamic fluctuations, affecting the stability of the optimal sampling time. In this embodiment, a dynamic adjustment mechanism is also added, such as... Figure 7 As shown, it includes data storage, timing compensation, timing estimation, and timing correction of data storage and timing compensation based on timing estimation.

[0184] Specifically, the input data is grouped, with each group having a length of L. Timing deviation compensation and estimation are performed on each group. After compensation for the current group, a new deviation estimate is made to prepare for the next group's estimate. During data storage and retrieval, the stored data includes all data from the current group, the last two digits of the previous group, and the first two digits of the next group. The starting position for current data extraction is determined based on an integer multiple of the timing correction offset.

[0185] like Figure 8As shown, the timing correction includes converting continuous timing estimates into signed integers fluc (fluctuation) and fractional parts uu (residual deviation), and performing amplitude limiting correction on values ​​that are out of range, outputting two results.

[0186] In some specific embodiments, since different modulation schemes have different demodulation mapping rules and phase ambiguity, modulation identification is required before subsequent processing. Modulation schemes include PSK, QAM, and APSK, specifically including:

[0187] The formula for calculating the eigenvalue R is as follows:

[0188] ;

[0189] If R≈1 is satisfied, it is determined to be an MPSK. The order of the MPSK is identified, and T is calculated. M Specifically:

[0190] ;

[0191] The modulation order M can be 2, 4, or 8. If TM ≈ 1, it means that M matches the actual modulation order.

[0192] Calculate the modulus of each symbol point:

[0193] r m =∣x m |;

[0194] Where m = 1, 2, ..., N, |•| represents the modulo operation; x m This indicates the symbol value after the symbol has been timed.

[0195] K initial centers are randomly selected, including μ1, μ2, μ3, ..., μ k According to |r m -μ k |, divide the symbols into k sets C k It satisfies the following formula:

[0196] ∣r m -μ i |=argmin|r m -μ k |, r m ∈C k ;

[0197] That is, it belongs to C i The modulus of the sign point of each set is such that its distance to the i-th center value is minimized. The mean of each set is recalculated to obtain a new center value. This process is repeated until the mean of all sets converges. The formula is:

[0198] d i =μ i -μ i-1 ;

[0199] If d is satisfied i ≈ constant, i=2,3,…,k, is determined to be QAM, otherwise it is APSK.

[0200] In other embodiments, for continuous signal processing scenarios without pilots, there may be multiple modulation methods. This embodiment utilizes the characteristics of modulation symbols under high signal-to-noise ratio conditions, which are statistically independent and have zero mean. By performing conjugate multiplication on the received signal, a phase difference statistic is constructed, and accumulation and averaging are performed in the time dimension, so that the modulation components tend to cancel each other out in a statistical sense, thereby achieving parameter estimation that is independent of the specific modulation method.

[0201] Construct the conjugate product of equally spaced symbols and perform time averaging over a window of length N. The calculation formula is as follows:

[0202] ;

[0203] Among them, T s Let represent the symbol rate, m represent the symbol interval, ∠(•) represent the complex phase calculation, z(n) represent the conjugate multiplication result between the nth symbol and the nmth symbol after symbol timing, and r(n) represent the received signal samples; r * (nm) is the conjugate signal delayed by m samples; when the value of n slides N times, that is, takes the value n~(n+N-1), N sets of z(n) will be obtained, and the average of the N sets of z(n) data is denoted as A.

[0204] Using a smoothing filter, the calculation formula is as follows:

[0205] ;

[0206] in, Represents the frequency offset value of the i-th estimation. This represents the frequency offset value of the i-th estimate after smoothing. This represents the smoothing coefficient, which can be set to 0.05 during implementation.

[0207] Accordingly, this embodiment uses an M-power modulation method and performs phase estimation for high signal-to-noise ratio environments. The calculation method used is as follows:

[0208] ;

[0209] Here, the phase value of the symbol r(n) is [θ(n)]rad, where rad represents radians, and the complex number e is constructed based on θ(n). 1i•M•θ(n)This process involves sequentially obtaining N sets of constructed complex numbers, summing these N sets, calculating the phase value, and dividing by M. The phase value calculation differs from conventional methods by omitting M complex multiplications.

[0210] After frequency offset estimation, symbol timing recovery, and carrier phase estimation are completed, the constellation points of the received signal will cluster around the ideal reference position, thus meeting the conditions for symbol decision. The scenario proposed in this embodiment does not rely on pilots or reference symbols, but inherent phase ambiguity issues may still exist. Different phase rotations can cause an equivalent rotation of the entire constellation. For example, the 2π / M periodic phase uncertainty under PSK modulation results in a systematic shift in the clustered symbol position relative to the true reference point.

[0211] In this embodiment, the maximum number of iterations of the decoding module is set as the decision threshold. Under high signal-to-noise ratio conditions, if the input symbol phase is correct, the decoder can converge within a few iterations and output data results that meet the verification constraints; if phase ambiguity exists, the decoding process is difficult to converge, and even after reaching the preset maximum number of iterations, it still cannot pass the verification constraints. That is, a joint decision is made with the channel decoding module to correct phase ambiguity.

[0212] Based on this characteristic, the signal data after phase estimation undergoes various possible initial phase offset processing. For example, according to the modulation scheme, several equivalent phase rotation angles are pre-rotated to form multiple candidate data streams, which are then sent to the decoding module for parallel or time-division decoding. By comparing the convergence speed or verification results of each decoder, the decoding output that converges fastest or first satisfies the parity check matrix constraints is selected as the correct phase correspondence result.

[0213] This application also provides a continuous signal processing system suitable for satellite communication. The system includes a processor coupled to a memory. The memory stores instructions, and when the instructions are executed by the processor, a method for continuous signal processing suitable for satellite communication is implemented.

[0214] This application also provides a computer-readable storage medium storing computer instructions that, when executed on a computer, implement a method for continuous signal processing suitable for satellite communications.

[0215] This application also provides a computer program product comprising instructions which, when executed by a computer, implement the methods performed in the above-described method embodiments.

[0216] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A continuous signal processing method suitable for satellite communication, characterized in that, include: Acquire continuous received signals; A coarse frequency offset estimate is performed on the continuously received signal to obtain a coarse frequency offset estimate value. Based on the coarse frequency offset estimate value, frequency offset compensation is performed on the continuously received signal to obtain a frequency offset compensated signal. Symbol timing recovery is performed on the frequency offset compensation signal to obtain the optimal sampling sequence; Modulation identification is performed on the optimal sampling sequence to determine the modulation type of the optimal sampling sequence; The optimal sampling sequence is subjected to fine frequency offset estimation, and the optimal sampling sequence is subjected to phase estimation and compensation in combination with the modulation type to obtain the phase-compensated sequence; The phase compensation sequence is subjected to multi-path candidate phase rotation and channel decoding decision, and the phase ambiguity is eliminated by using the decoding convergence feature to obtain the decoded data; Data filtering based on frame boundary markers is performed to output valid business data.

2. The method of claim 1, wherein, Coarse frequency offset estimation is performed on the continuously received signal to obtain the coarse frequency offset estimate, which specifically includes: The continuously received signal is subjected to matched filtering and amplitude normalization to obtain a normalized received signal; For the normalized received signal, perform at least one nonlinear phase doubling transformation of a predetermined order to obtain a phase doubling sequence; The phase-doubling sequence is converted to the frequency domain, and amplitude spectrum calculation and peak search are performed. The frequency offset features corresponding to the peak positions of the amplitude spectrum are extracted as coarse frequency offset estimates.

3. The method of claim 2, wherein, Perform at least one nonlinear phase-doubling transform of a predetermined order on the normalized received signal to obtain a phase-doubling sequence, and extract frequency offset features as coarse frequency offset estimates, including: Multiple nonlinear phase doubling transformations of different candidate orders are executed in parallel to obtain multiple candidate phase doubling sequences. Each candidate phase doubling sequence is converted to the frequency domain, and the peak-to-average power ratio of the amplitude spectrum corresponding to each candidate phase doubling sequence is calculated. According to the order from low to high, the optimal order is selected from the candidate orders when the corresponding peak-to-average ratio metric first exceeds the preset detection threshold. Based on the peak position of the amplitude spectrum corresponding to the optimal order, the coarse frequency offset estimate is calculated.

4. The method of claim 1, wherein, The optimal sampling sequence is obtained by performing symbol timing recovery on the frequency offset compensation signal, specifically including: The frequency offset compensation signal is divided into continuous grouped data; Construct a two-state tracking model that includes timing offset state and drift rate state; Using a dual-state tracking model, feedforward prediction and feedback correction are performed on the current grouped data to update the timing offset state and drift rate state. Based on the updated timing offset state, timing compensation is performed on the grouped data to obtain the optimal sampling sequence.

5. The method of claim 1, wherein, The phase compensation sequence undergoes multi-path candidate phase rotation and channel decoding decision-making. Phase ambiguity is eliminated using decoding convergence features to obtain decoded data, specifically including: Based on the theoretical phase ambiguity number corresponding to the modulation type, multiple different candidate phase rotations are performed on the phase compensation sequence to generate a multi-path candidate rotation sequence. For each candidate rotation sequence, calculate the distance measure from each symbol to the corresponding ideal constellation point set to obtain the constellation distance metric for each candidate rotation sequence; Based on the constellation distance metric, the multi-path candidate rotation sequences are pre-screened, and the candidate rotation sequences with the smallest distance metric are extracted as a set number of input subsets and sent to the channel decoder for channel decoding decision. In the input subset, the sequence that converges first in the decoding is selected as the correct sequence, and the corresponding decoded data is output.

6. The method of claim 1, wherein, After performing multi-path candidate phase rotation and channel decoding decision on the phase compensation sequence, and eliminating phase ambiguity using decoding convergence characteristics to obtain the decoded data, cross-layer parameter adjustment is also included, including: The decoded data is verified for reliability to determine whether the current frame is a highly reliable decoded frame. When a frame is determined to be a highly reliable decoded frame, channel coding and modulation operations are re-executed on the decoded data to generate a virtual reference symbol sequence corresponding to the current frame. The virtual reference symbol sequence is compared with the physical layer received sequence buffered before the decoding decision of the current frame, and the residual frequency offset feature and residual phase feature are extracted as residual physical layer parameters. Using the residual physical layer parameters, the current states of the fine frequency offset estimation and phase estimation are updated smoothly via feedback.

7. The method of claim 1, wherein, The process of filtering decoded data based on frame boundary markers to output valid business data specifically includes: Zero-bit deletion is performed on the decoded data to obtain the recovered bitstream; The recovered bit stream is shifted in one by one through a sliding window, and the bit sequence in the current sliding window is pattern matched with the frame boundary flag. When a match is successful, the current position of the sliding window is determined as a candidate point for the frame boundary; Extract the continuous bit sequence between two adjacent frame boundary candidate points and output it as valid service data.

8. The method of claim 2, wherein, The continuously received signal is subjected to matched filtering and amplitude normalization to obtain a normalized received signal, specifically including: Obtain the amplitude value of the continuously received signal after matched filtering; The signal amplitude value is added to a preset regularization constant to obtain the corrected amplitude value; The normalized received signal is obtained by performing a complex division normalization operation on the continuously received signal after matched filtering using the corrected amplitude value.

9. A continuous signal processing system suitable for satellite communications, characterized by The system includes a processor coupled to a memory storing instructions which, when executed by the processor, implement the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed on a computer, implement the method as described in any one of claims 1 to 8.

Images