A low signal-to-noise ratio direct spread signal detection method based on noise cancellation

Abstract

The application discloses a low signal-to-noise ratio direct spread signal detection method based on noise cancellation, and belongs to the field of communication spectrum sensing. The application adopts a minimum mean square error-based adaptive noise cancellation method, can track and eliminate non-stationary noise interference in real time, automatically optimizes filter parameters in an unknown channel environment, and enables a signal detection system to adapt to different background noise conditions. By using a cyclic spectrum analysis method, the cyclic stationary characteristics of the direct spread signal are extracted, and robust signal detection is realized in a low signal-to-noise ratio environment. The existence of the signal is determined through a characteristic spectrum peak on a non-zero cycle frequency, and the carrier frequency and pseudo code rate are further estimated, so that more accurate signal identification and parameter extraction are realized, the influence of noise uncertainty on the detection performance is avoided, and the robustness and reliability of detection are improved. Welch smoothing processing and short-time Fourier transform are combined, the variance of spectrum estimation is reduced when the cyclic spectrum is calculated, and the robustness of detection is improved.

Description

Technical Field

[0001] This invention belongs to the field of communication spectrum sensing, and relates to a signal detection method for non-cooperative conditions, and more particularly to a method for detecting direct-sequence spread spectrum signals and estimating parameters under extremely low signal-to-noise ratio conditions. Background Technology

[0002] Signal detection technology is a method for determining channel occupancy in non-cooperative communication and is a key technology in the field of spectrum sensing. Direct-sequence spread spectrum (DSSS) technology has been widely used in both civilian and military communications due to its strong anti-interference and multipath capabilities, signal concealment, low interception rate, and ease of implementing code division multiple access. Therefore, spread spectrum technology is crucial for electronic countermeasures and civilian radio resource management, and DSSS signal detection is vital for gaining control of the electromagnetic spectrum on the future battlefield.

[0003] Currently, there are some methods for the detection and parameter estimation of direct sequence spread spectrum signals. These methods have good detection results for a certain parameter, but their performance deteriorates at low signal-to-noise ratios. Therefore, the effective detection and estimation of direct sequence spread spectrum signals without prior knowledge is a field worthy of in-depth research, and it also has corresponding theoretical significance for communication countermeasures, military reconnaissance, electronic spectrum management, and interference identification.

[0004] Different methods can effectively estimate different parameters of direct-sequence spread spectrum (DSSS) signals. Some methods are very effective at estimating a certain parameter of a DSSS signal, but may have high computational complexity and cost. Moreover, these methods still face challenges such as how to achieve a lower signal-to-noise ratio to accommodate the estimation of signal parameters submerged in noise, which requires further in-depth research.

[0005] Due to the signal concealment inherent in direct-sequence spread spectrum (DSSS) technology, non-cooperative detection faces significant challenges. This is especially true in low signal-to-noise ratio (SNR) environments, where the signal is often overwhelmed by background noise, making detection and identification extremely difficult.

[0006] In recent years, with the rapid development of computing power, direct-sequence spread spectrum (DSS) signal detection and parameter estimation techniques have advanced rapidly. Energy detection is the simplest, but its model does not incorporate the available features of the DSS signal, making it relatively crude and susceptible to noise and interference. Correlation methods utilize the correlation delay of the DSS signal for detection, but they are also susceptible to noise, converge slowly under low signal-to-noise ratio (SNR) conditions, and require prior conditions. Higher-order spectra, while revealing more information about the signal, still need improvement to achieve lower SNR tolerances. Cyclic spectra offer good detection performance, but require long-term data accumulation.

[0007] Due to the uncertainty of random signals and the inherent noise-buried nature of direct-sequence spread spectrum (DSS) signals, traditional detection methods require long-term accumulation of the DSS signal to obtain its statistical characteristics at extremely low signal-to-noise ratios (SNRs). Simulations show that to achieve a reasonably ideal detection effect under 0 dB SNR conditions, approximately 10... 6 At a signal-to-noise ratio of -20dB, approximately 10 are required. 7 The number of sampling points required increases significantly with lower signal-to-noise ratios (SNR). Therefore, low SNR and limited sampling sequences are two major challenges in cyclic spectrum detection. It is evident that detecting finite-length direct-sequence spread spectrum (DSSS) sampling sequences at extremely low SNRs is a critical problem that urgently needs to be solved. Summary of the Invention

[0008] To address the shortcomings of existing energy detection methods, such as poor performance and long detection times for direct-sequence spread spectrum (DSSS) signals with extremely low signal-to-noise ratios (SNR), this invention aims to provide a noise cancellation-based method for detecting DSS signals with low SNR. This method utilizes an LMS adaptive filter to cancel noise, thereby improving the SNR and making the characteristics of the DSS signal clearer in the frequency domain. Due to the difference in cyclostationary characteristics between DSS signals and noise, this method can effectively distinguish between signal and noise in low SNR environments. It enables efficient detection with shorter sampling times and lower computational complexity, improving the real-time performance and noise immunity of the DSS signal detection system.

[0009] The objective of this invention is achieved through the following technical solution.

[0010] This invention discloses a low signal-to-noise ratio (SNR) direct-sequence spread spectrum (DSSS) signal detection method based on noise cancellation. Noise cancellation is achieved through an LMS adaptive filter, and iterative learning is used to automatically adjust weights to match interfering noise. This improves the SNR of the DSSS signal while ensuring a clear outline and significant spectral peaks in the frequency domain, thus achieving efficient noise suppression. Furthermore, leveraging the cyclostationary nature of the DSSS signal, the cyclic spectrum is calculated to distinguish between signal and noise, and target signal features are extracted to improve detection accuracy in low SNR environments. Based on this, the presence of the DSSS signal is determined using specific frequency characteristics of the cyclic spectrum, and parameters such as carrier frequency and pseudocode rate are estimated. Compared to traditional energy detection methods, this invention exhibits stronger anti-interference capabilities and higher detection stability in low SNR environments, improving the detection accuracy and reliability of DSSS signals.

[0011] The present invention discloses a method for detecting low signal-to-noise ratio direct-sequence spread spectrum signals based on noise cancellation, comprising the following steps:

[0012] Step 1: Construct a noise cancellation method based on minimum mean square error adaptive filtering. The method estimates noise in real time by constructing an adaptive filter and adjusts the filter parameters adaptively using the error minimization criterion. Noise cancellation is achieved through LMS-based adaptive filter. The weights are automatically adjusted through iterative learning to match the interference noise, thereby enhancing the characteristics of the direct-sequence spread spectrum (DSSS) signal, improving the signal-to-noise ratio (SNR) of the DSSS signal, clearly presenting the DSSS signal profile in the frequency domain and forming significant spectral peaks, thus achieving efficient noise suppression.

[0013] In low signal-to-noise ratio (SNR) environments, direct-sequence spread spectrum (DSSS) signals are easily masked by background noise, leading to decreased detection accuracy. This invention constructs a noise cancellation method based on minimum mean square error (MMS) adaptive filtering. By constructing an adaptive filter to estimate noise in real time and using an error minimization criterion to adaptively adjust filter parameters, the output signal is made as free of noise interference as possible, thereby enhancing the characteristics of the DSS signal and improving the accuracy of subsequent cyclic spectrum detection. This method can achieve adaptive noise suppression and improve the SNR of the sampled sequence by dynamically adjusting filter parameters through recursive calculation even when the signal characteristics are unknown.

[0014] The adaptive noise canceller has two input channels: a main input channel and a reference input channel. The main input channel contains the useful signal s(n) and uncorrelated interference noise n0(n), while the reference input channel contains only noise n1(n) that is correlated with n0(n) but uncorrelated with s(n). The direct-sequence spread spectrum signal detection system adjusts the tap weights ω using an adaptive filter. i (n), to make the filter output y(n) approximate n0(n), and obtain the optimized signal estimate through error calculation:

[0015] e(n)=d(n)-y(n)=s(n)+n0(n)-y(n) (1)

[0016] For the mean square error E[e 2 Minimize:

[0017] E[e 2 ] = E[(s+n0-y) 2 ]=E[s 2 ]+E[(n0-y) 2 (2)

[0018] When the filter output y(n) best approximates n0(n), E[e 2 The minimum output signal e(n) of the direct-sequence spread spectrum (DSS) signal detection system is mainly composed of s(n), achieving effective noise reduction. The specific implementation method of adaptive noise cancellation is as follows:

[0019] Step 101: Initialize the filter parameters, with the initial tap weights ω(0) = 0.

[0020] Step 102, the filter tap weights at time n are ω(n) = [ω0(n), ω1(n), ..., ω M-1 (n)] T The input vector is represented as u = [u(n), u(n-1), ..., u(n-M+1)]. T Calculate the output of the LMS adaptive filter.

[0021] y(n)=ω H (n)u(n) (3)

[0022] Step 103, calculate the error between the main channel signal d(n) and the filter output:

[0023] e(n)=d(n)-y(n)=d(n)-ω H (n)u(n) (4)

[0024] The error signal e(n) reflects the noise components that the filter has not yet eliminated.

[0025] Step 104, Tap Weight Update. Given the prior information of the unknown signal, the steepest descent algorithm is used to adjust the filter weight vector along the steepest descent direction of the performance surface, searching for the minimum point of the performance surface. The iterative formula for calculating the weight vector is:

[0026]

[0027] Where μ is the step size factor. The gradient descent is calculated using the following formula:

[0028]

[0029] Step 105, Error Correction and Convergence Judgment. If the mean square error E[e 2 ] = E[(s+n0-y) 2 If the value is less than the set threshold, stop the iteration; otherwise, continue adjusting the filter weights.

[0030] Step 106: Repeat steps 102-105 until the system output signal e(n) is mainly composed of s(n), that is, the noise suppression reaches the optimal state.

[0031] This step dynamically adjusts the weights through an iterative learning mechanism to make the filter output optimally approximate the interference noise, thereby achieving real-time adaptive noise suppression. This effectively improves the signal-to-noise ratio of the direct-sequence spread spectrum (DSSS) signal, allowing the DSSS signal to clearly present its outline in the frequency domain and form a significant spectral peak at the carrier frequency. Ultimately, this achieves efficient and accurate noise suppression, enhancing the signal detection capability in low signal-to-noise ratio environments.

[0032] Step two utilizes the cyclostationary characteristics of the direct-sequence spread spectrum (DSSS) signal to distinguish between signal and noise by calculating the cyclic spectrum and extracting target signal features. Having already improved the signal-to-noise ratio (SNR) through adaptive noise cancellation in step one, this step further extracts the cyclic characteristics of the DSSS signal from the frequency domain, enhancing detection accuracy in low SNR environments.

[0033] Because direct-sequence spread spectrum (DSSS) signals are modulated using pseudo-random noise (PN) sequences, they exhibit a cyclic frequency corresponding to the carrier frequency. α Cyclic spectrum analysis exhibits a stable energy distribution at a given frequency, while traditional Gaussian white noise has no significant component at a non-zero cyclic frequency f. Therefore, cyclic spectrum analysis can effectively highlight the direct-sequence spread spectrum (DSB) signal while suppressing noise, making signal detection more stable and reliable.

[0034] This invention converts the signal with improved signal-to-noise ratio in step one to the cyclic spectral domain, and then uses the cyclic frequency dimension to further suppress noise, thereby significantly improving the detection accuracy of direct-sequence spread spectrum signals in low signal-to-noise ratio environments.

[0035] Due to the instability of random signals, the variance is relatively large. This invention utilizes the Welch algorithm to divide a long sequence into multiple fixed-length random sequences, calculate the cyclic spectrum of each segment, and finally average them. By using the segmented averaging method, the final cyclic spectrum acquires certain statistical properties, reduces variance, and achieves a smoother time domain.

[0036] By replacing the corresponding quantities in the continuous formula of the cyclic spectral correlation function with discrete values, we obtain the discrete implementation of the spectral correlation function:

[0037]

[0038] The specific steps for step two are as follows:

[0039] Step 201: Sample sequence preprocessing.

[0040] To improve the stability of frequency domain estimation, the input sampling sequence is segmented. When the total number of sampling points is N, it is divided into N1 groups, with N2 data points in each group:

[0041]

[0042] This segmentation strategy uses the Welch smoothing method for preprocessing, which effectively reduces the variance of spectral estimation and improves signal stability.

[0043] Step 202: Short-time Fourier Transform (STFT).

[0044] Perform a Fast Fourier Transform (FFT) on each set of data to transform the time-domain signal into the frequency domain:

[0045]

[0046] The transformation is used to obtain the energy distribution of the signal at different frequencies, providing a basis for subsequent cyclic spectrum calculations.

[0047] Step 203: Calculate the spectral correlation function.

[0048] In cyclic spectrum calculation, the periodicity of a signal can be extracted through the correlation between different frequency components. Calculate the cross-correlation between different frequency components:

[0049]

[0050] At this point, if α ≠ 0, the spectral correlation of white noise tends to 0, while the direct-sequence spread spectrum signal, due to its periodicity, will exhibit a correlation at α = 2f. c A significant spectral peak is observed at the carrier frequency.

[0051] Step 204: Calculate and output the cyclic spectrum.

[0052] The mean value of the spectral correlation functions obtained from all segmented calculations is calculated to obtain the final cyclic spectrum estimate:

[0053]

[0054] The statistical stability of the cyclic spectrum is ensured by obtaining the final cyclic spectrum estimate, making signal detection more robust.

[0055] This step uses cyclic spectrum analysis to clearly reveal the characteristics of the direct-sequence spread spectrum (DSS) signal at non-zero cyclic frequencies, while noise, lacking periodicity, exhibits energy distributions approaching zero at these frequency points. This characteristic enables cyclic spectrum analysis to accurately extract key parameters of the DSS signal, including carrier frequency and spreading code rate, even in low signal-to-noise ratio environments, thereby effectively improving detection performance.

[0056] Step 3: Based on the specific frequency characteristics of the cyclic spectrum, determine the presence or absence of a direct-sequence spread spectrum (DSSS) signal and perform parameter estimation. The parameter estimation result is the DSS signal detection result under low signal-to-noise ratio (SNR), i.e., DSS signal detection under low SNR is achieved based on noise cancellation. The parameters include carrier frequency and pseudocode rate.

[0057] In the preceding steps, adaptive noise cancellation improves the signal-to-noise ratio (SNR), and cyclic spectrum is used to enhance the characteristics of the direct-sequence spread spectrum (DSSS) signal. Next, based on the characteristics of the cyclic spectrum, it is necessary to determine whether the DSSS signal exists and further extract parameters such as carrier frequency and pseudocode rate. Traditional energy detection methods are easily affected by noise at low SNR, while the cyclic spectrum-based method can effectively distinguish between signal and noise, making the detection more stable and reliable.

[0058] In the cyclic spectrum, due to its periodicity, the direct-sequence spread spectrum (DSSS) signal forms a distinct spectral peak at twice the carrier frequency (α = 2f0), while Gaussian white noise is mainly concentrated at the zero cyclic frequency. Therefore, the presence of the signal can be determined by analyzing the distribution of the cyclic spectrum across different cyclic frequency dimensions.

[0059] Step 301: Obtain two-dimensional slices of the cyclic spectrum and extract features.

[0060] Since the energy of the direct-sequence spread spectrum (DSSS) signal is concentrated at the cyclic frequency α = 2f0, while Gaussian white noise only exists at the zero cyclic frequency, the distribution characteristics of the cyclic spectrum in different frequency dimensions can be used to detect the DSS signal.

[0061] From the calculated three-dimensional cyclic spectrum, a two-dimensional slice with frequency f = 0 is extracted, i.e.:

[0062] |S x (α,0)| (12)

[0063] In this slice, the direct-sequence spread spectrum (DSS) signal exhibits significant spectral peaks at non-zero cyclic frequencies, while the noise component is concentrated at zero cyclic frequencies.

[0064] Step 302: Signal detection and decision.

[0065] Set a detection threshold and search for the maximum value S on the cycle frequency axis. max Simultaneously calculate the average energy S of the cyclic spectrum. avg .

[0066] To improve the accuracy of the judgment, the peak-to-average power ratio (PAPR) is calculated:

[0067]

[0068] and the set decision threshold P th The comparison is performed. If the threshold is exceeded, it is determined that a direct-sequence spread spectrum signal exists in the sampling sequence; if the threshold is not reached, the detection is determined to have failed, and the current signal detection process is terminated.

[0069] Compared to directly comparing the maximum spectral peak value, using the peak-to-average power ratio (PAPR) for decision-making can more effectively avoid the influence of background noise and reduce the possibility of false detections, making it particularly suitable for signal detection in low signal-to-noise ratio (SNR) environments. This method utilizes the periodicity of the signal to avoid the problem of traditional energy detection methods being susceptible to noise uncertainty in low SNR environments.

[0070] Step 303: Estimate the carrier frequency of the direct-sequence spread spectrum signal.

[0071] The maximum value S obtained from step 302 max Its corresponding cycle frequency is 2f0:

[0072]

[0073] Where α max The maximum value of the cyclic spectrum S max The corresponding cycle frequency.

[0074] Based on the peak characteristics of the cyclic spectrum, a priori-free blind estimation of carrier frequency is achieved.

[0075] Step 304: Estimate the pseudocode rate.

[0076] Based on the signal carrier frequency f0 estimated in step 303, another two-dimensional slice of the cyclic spectrum is selected, namely the cyclic spectrum slice at frequency f = f0:

[0077] |S x (α,f0)| (15)

[0078] In this slice, the pseudocode rate R of the direct-sequence spread spectrum signal is... c This is reflected in the peak value of the cycle frequency of the maximum value of this slice:

[0079] R c =α max (16)

[0080] This step overcomes the limitations of traditional energy detection methods under low signal-to-noise ratio conditions by employing a cyclic spectrum-based signal detection method, making direct-sequence spread spectrum (DSSS) signal detection more accurate. The extraction of carrier frequency and pseudocode rate requires no prior information, making it suitable for signal analysis in complex environments. Combined with the preceding steps, the complete signal detection method can accurately separate DSS signals in strong noise backgrounds and estimate key parameters, providing reliable data for subsequent communication or signal processing.

[0081] Beneficial effects:

[0082] 1. This invention discloses a low signal-to-noise ratio (SNR) direct-sequence spread spectrum (DSSS) signal detection method based on noise cancellation. It employs an adaptive noise cancellation method based on minimum mean square error, which can track and eliminate non-stationary noise interference in real time. It automatically optimizes filtering parameters in unknown channel environments, enabling the signal detection system to adapt to different background noise conditions. Compared to traditional fixed filtering methods, the adaptive filtering strategy of this invention can minimize the impact of noise and improve detection sensitivity without losing the characteristics of the target signal, allowing the system to detect even weaker DSSS signals.

[0083] 2. This invention discloses a low signal-to-noise ratio (SNR) direct-sequence spread spectrum (DSSS) signal detection method based on noise cancellation. Utilizing cyclic spectrum analysis, it extracts the cyclic stationary characteristics of the DSSS signal to achieve robust signal detection in low SNR environments. Unlike traditional energy detection methods that rely on signal strength, this invention determines the presence of a signal by identifying characteristic spectral peaks at non-zero cyclic frequencies and further estimates the carrier frequency and pseudocode rate, achieving more accurate signal identification and parameter extraction. This avoids the impact of noise uncertainty on detection performance and improves the robustness and reliability of the detection.

[0084] 3. This invention discloses a low signal-to-noise ratio (SNR) direct-sequence spread spectrum (DSSS) signal detection method based on noise cancellation. By combining Welch smoothing and short-time Fourier transform, the variance of spectral estimation is reduced during cyclic spectrum calculation, improving detection robustness. Even in extremely low SNR environments, the system maintains good detection performance, avoiding signal submersion due to noise dominance in traditional methods and improving signal detection success rate. The peak-to-average power ratio (PAPR) decision method optimizes the signal detection strategy, which, compared to the traditional maximum peak value threshold method, more effectively distinguishes DSS signals from random noise interference, reducing false detection rate. Attached Figure Description

[0085] Figure 1 This is the overall flowchart of the direct-sequence spread spectrum signal detection method with low signal-to-noise ratio based on noise cancellation disclosed in this invention;

[0086] Figure 2 This is a system structure diagram for noise cancellation in this invention;

[0087] Figure 3 This is a block diagram of the LMS adaptive filter principle in this invention;

[0088] Figure 4 This is a diagram illustrating the effect of the noise canceller in this invention on a direct-sequence spread spectrum signal with a signal-to-noise ratio of -15dB. Figure 4 (a) is the original signal. Figure 4 (b) is the power spectral density of the direct-sequence spread signal after noise cancellation;

[0089] Figure 5 This invention is for Figure 5 A three-dimensional image of cyclic spectral density obtained by processing the noise-reduced signal;

[0090] Figure 6 This is a two-dimensional image obtained by slicing a three-dimensional image of cyclic spectral density at specific frequencies, as described in this invention. Figure 6 (a) The figure shows a two-dimensional slice when f = 0. Figure 6 (b) The figure shows f = f c Two-dimensional slice. Detailed Implementation

[0091] To better illustrate the purpose and advantages of the present invention, the invention will be further described below in conjunction with the accompanying drawings and examples.

[0092] Example 1:

[0093] The technical problem to be solved by the noise cancellation-based method for detecting the cyclic spectrum of a direct-sequence spread signal under low signal-to-noise ratio conditions in this embodiment is: to detect a finite-length direct-sequence spread signal sampling sequence under extremely low signal-to-noise ratio conditions.

[0094] The implementation parameters are as follows: 200 information code points, pseudo-random sequence length of 31, pseudo-code rate of 46.5 kHz, signal code rate of 1 kHz, carrier frequency of 93 kHz, sampling rate of 744 kHz, signal-to-noise ratio of -15 dB, and the power spectrum of the original sampled sequence is as follows. Figure 4 As shown in (a). At this time, the direct-sequence spread spectrum (DSSS) signal will be submerged in noise at a signal-to-noise ratio of -15dB, and the spectral peak of the DSS signal at the carrier frequency cannot be observed.

[0095] like Figure 1 As shown in the figure, this embodiment discloses a method for detecting low signal-to-noise ratio direct-sequence spread spectrum signals based on noise cancellation. The specific implementation steps are as follows:

[0096] Step 1: Construct a noise cancellation method based on minimum mean square error adaptive filtering. The method estimates noise in real time by constructing an adaptive filter and adjusts the filter parameters adaptively using the error minimization criterion. Noise cancellation is achieved through LMS-based adaptive filter. The weights are automatically adjusted through iterative learning to match the interference noise, thereby enhancing the characteristics of the direct-sequence spread spectrum (DSSS) signal, improving the signal-to-noise ratio (SNR) of the DSSS signal, clearly presenting the DSSS signal profile in the frequency domain and forming significant spectral peaks, thus achieving efficient noise suppression.

[0097] The adaptive noise canceller has two input channels: a main input channel and a reference input channel, such as... Figure 2 As shown. The main input channel contains the useful signal s(n) and uncorrelated interference noise n0(n), while the reference input channel contains only noise n1(n) that is correlated with n0(n) but uncorrelated with s(n). The direct-sequence spread spectrum (DSS) signal detection system adjusts the tap weights ω using an adaptive filter. i (n), to make the filter output y(n) approximate n0(n), and obtain the optimized signal estimate through error calculation:

[0098] e(n)=d(n)-y(n)=s(n)+n0(n)-y(n) (17)

[0099] Since s(n) is independent of n0(n) and y(n), we can obtain the mean square error of the above equation.

[0100] E[e 2 ] = E[(s+n0-y)2 ]=E[s 2 ]+E[(n0-y) 2 (18)

[0101] To minimize the mean square error, the tap weights ω of the LMS adaptive filter need to be adjusted. i (n), to minimize the mean square error, is

[0102] minE[e 2 ]=E[s 2 ]+minE[(n0-y) 2 (19)

[0103] The above equation shows that y(n) achieves the best estimate of the noise n0(n) in the main channel d, with its mean square error approaching zero. The error e(n) = s(n) + n0(n) - y(n) is statistically close to s(n). Therefore, the system's output signal e(n) is close to the useful signal s(n), and the noise in the system's output signal will be greatly reduced.

[0104] Adaptive filter structure as follows Figure 3 As shown, the adaptive filter order is set to 12. The filter tap weights at time n are... The input vector is represented as d(n) is the main channel input signal, y(n) is the desired output response of the filter, e(n) is the error output signal, and M is the length of the filter. Achieving optimal filter weighting is impossible without prior knowledge of the unknown signal. The instantaneous gradient value will be used instead of the expectation factor in the steepest descent method.

[0105] Step one, the specific implementation method of adaptive noise cancellation is as follows:

[0106] Step 101: Initialize tap weight ω(0) = 0.

[0107] Step 102: At time n, calculate the output of the LMS adaptive filter.

[0108] y(n)=ω H (n)u(n) (20)

[0109] Step 103, estimate the error at time n.

[0110] e(n)=d(n)-y(n)=d(n)-ω H (n)u(n) (21)

[0111] The error signal e(n) reflects the noise components that the filter has not yet eliminated.

[0112] Step 104, Tap weight update

[0113]

[0114] Where μ is the step size factor, used to control the convergence rate and stability. During the implementation phase, the step size factor μ is set to 0.0008. The gradient descent is calculated using the following formula:

[0115]

[0116] Step 105, Error Correction and Convergence Judgment. If the mean square error E[e 2 ] = E[(s+n0-y) 2 If the value is less than the set threshold, stop the iteration; otherwise, continue adjusting the filter weights.

[0117] Step 106: Repeat steps 102-105 until the system output signal e(n) is mainly composed of s(n), that is, the noise suppression reaches the optimal state.

[0118] This step dynamically adjusts the weights through an iterative learning mechanism to optimally approximate the interference noise in the filter output, achieving real-time adaptive noise suppression. This effectively improves the signal-to-noise ratio (SNR) of the direct-sequence spread spectrum (DSS) signal, allowing the DSS signal to clearly present its outline in the frequency domain and forming a significant spectral peak at the carrier frequency. Ultimately, this achieves efficient and accurate noise suppression, enhancing signal detection capabilities in low SNR environments. The results are as follows... Figure 4 As shown, Figure 4 (a) is the original spectrum. Figure 4 (b) is the spectrum after processing in step one. The signal-to-noise ratio has been significantly improved from the initial setting of -15dB to approximately 0dB.

[0119] Step two utilizes the cyclostationary characteristics of the direct-sequence spread spectrum (DSSS) signal to distinguish between signal and noise by calculating the cyclic spectrum and extracting target signal features. Having already improved the signal-to-noise ratio (SNR) through adaptive noise cancellation in step one, this step further extracts the cyclic characteristics of the DSSS signal from the frequency domain, enhancing detection accuracy in low SNR environments.

[0120] In the cyclic spectral density function, the frequency domain is extended from the traditional coordinate domain of power spectrum analysis to a dual-frequency coordinate system of spectral frequency and periodic frequency. In non-stationary signals, some spectral lines overlap in the power spectrum, and the envelope of the spectral information covers these characteristic spectral lines. When detecting signals using the power spectral density function, the peak values ​​of the line spectrum cannot be observed, thus preventing the acquisition of their characteristic values. However, when using the cyclic spectral density function, at α≠0, some signal characteristics are re-represented as discretely distributed periodic frequency spectral lines. Therefore, periodic information such as the symbol rate, which is unavailable in power spectrum analysis, can be obtained.

[0121] When data is limited, noise instability often significantly impacts the results of cyclic spectrum estimation, and the results can only be obtained through a single estimation, leading to large variance. This paper utilizes the Welch algorithm to divide the long sequence into multiple fixed-length random sequences, calculate the cyclic spectrum for each segment, and finally average them. By using a segmented averaging method, the final cyclic spectrum can acquire certain statistical properties, reduce variance, and achieve a smoother time domain.

[0122] By replacing the corresponding quantities in the continuous formula of the cyclic spectral correlation function with discrete values, we can obtain the discrete implementation of the spectral correlation function:

[0123]

[0124] According to the formula, the general calculation steps for calculating the cyclic spectrum of the sampled sequence in step two are as follows:

[0125] Step 201: Split the sequence after noise removal in Step 1. The total number of sampling points, 99200, is divided into 96 groups, with 1024 sequence points in each group. The extra 896 points are discarded.

[0126] x 1024 (1,f+α / 2),x 1024 (2,f+α / 2),...,x 1024 (96,f+α / 2) (25)

[0127] Step 202: Perform a fixed-point FFT of length 1024 for each set of sampling points.

[0128]

[0129] Step 203: Multiply the frequency components together to calculate the spectral correlation function value.

[0130]

[0131] Step 204: To reduce the influence of randomness, the mean of the spectral correlation function results obtained from all groups is calculated to obtain a more accurate estimate of the spectral correlation function, which is then used as the final output.

[0132]

[0133] The cyclic spectral density function obtained through the above steps is as follows: Figure 5Peaks appear on both the cyclic frequency axis and the spectral frequency axis. Superimposed Gaussian stationary white noise in the signal has very little interference with the signal characteristic peaks at non-zero cyclic frequencies because the cyclic spectral density of the stationary noise is mainly concentrated at the zero cyclic frequency. Especially against a strong background noise background, the signal spectrum at the zero cyclic frequency is often submerged in the noise. Therefore, using the cyclic spectral density function to detect direct-sequence spread spectrum signals in noise is effective, thus overcoming the drawbacks of conventional signal energy detection methods.

[0134] Step 3: Based on the specific frequency characteristics of the cyclic spectrum, determine the presence or absence of a direct-sequence spread spectrum (DSSS) signal and perform parameter estimation. The parameter estimation result is the DSS signal detection result under low signal-to-noise ratio (SNR), i.e., DSS signal detection under low SNR is achieved based on noise cancellation. The parameters include carrier frequency and pseudocode rate.

[0135] To make the observation more intuitive and reduce the amount of computational simulation, it is necessary to slice the three-dimensional plot of the cyclic spectrum and perform a one-dimensional search on the non-zero cyclic frequency axis, which can effectively estimate the carrier frequency and pseudocode rate.

[0136] By performing a two-dimensional slice when f=0, we obtain

[0137]

[0138] Among them, T c Let f(f) be the spreading symbol period of the direct-sequence spread spectrum signal, f0 be the carrier frequency of the signal, and Q(f) be the spectrum of the signal.

[0139] In the above formula, the zero-frequency slice is symmetrically distributed, so we only need to consider the case where α > 0. The maximum value occurs at α = 2f0. At f=0, the second largest value appears. Therefore, by searching for the maximum and second largest values ​​on the non-zero cyclic frequency axis at f=0, the carrier frequency and pseudocode rate of the direct-sequence spread spectrum signal can be estimated.

[0140] By performing a two-dimensional slice when f = f0, we obtain

[0141]

[0142] In the above formula, the carrier frequency slices are also symmetrically distributed, and we only need to consider the case where α > 0. When in The maximum value appears at f = f0. Therefore, the pseudocode rate of the direct-sequence spread spectrum signal can be estimated by searching for the maximum value on the non-zero cyclic frequency axis at f = f0.

[0143] According to the formula, the general calculation steps for step three—determining the presence or absence of a direct-sequence spread spectrum signal and estimating parameters—are as follows:

[0144] Step 301: Extract the center plane of frequency f in the three-dimensional cyclic spectrum; this is the two-dimensional slice where f = 0. For example... Figure 6 As shown in (a).

[0145] Step 302: Based on the current experimental environment and the established signal-to-noise ratio (SNR) parameters, set a threshold as the decision threshold for the presence or absence of a signal. Search for the maximum value on the cyclic frequency axis and calculate the peak-to-average power ratio (PAPR).

[0146] By observing the two-dimensional slice where f=0, if the peak-to-average power ratio (PAPR) is greater than the set threshold, it is determined that a direct-spread signal exists in the sampling sequence. A distinct symmetrical peak is observed on the cyclic frequency axis. Therefore, it is determined that a direct-spread signal definitely exists in the sampling sequence, consistent with the experimental settings.

[0147] The cycle frequency corresponding to the maximum value obtained in steps 303 and 302 is twice the carrier frequency, i.e., 2f0. In this experiment, 2f0 = 186kHz. Dividing the peak frequency by 2, we obtain the carrier frequency f0 = 93kHz.

[0148] Step 304: Extract a two-dimensional slice from the three-dimensional cyclic spectrum at frequency f = 93 kHz, as shown below. Figure 6 As shown in (b), the maximum value is searched on the cycle frequency axis. The cycle frequency value corresponding to the maximum value is the pseudocode rate R. c Therefore, the pseudocode rate R is obtained. c =46.5kHz, consistent with the experimental design.

[0149] Thus, the above example uses a low signal-to-noise ratio (SNR) direct-sequence spread spectrum (DSSS) signal detection method based on noise cancellation to complete the detection and parameter estimation of a finite short sequence under extremely low SNR conditions. The flowchart is as follows. Figure 1 As shown, accurate detection was achieved with a signal-to-noise ratio of -15dB, 200 short sequence information code points, and a pseudo-random sequence length of 31. The example demonstrates that this invention has excellent detection performance for short sequence signals at extremely low signal-to-noise ratios, and the capability shown in this example far exceeds the limits of detection capability.

[0150] The above detailed description further illustrates the purpose and technical solution of the invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting low signal-to-noise ratio direct-sequence spread spectrum signals based on noise cancellation, characterized in that: Includes the following steps, Step 1: Construct a noise cancellation method based on minimum mean square error adaptive filtering. The method estimates noise in real time by constructing an adaptive filter and adjusts the filter parameters adaptively using the error minimization criterion. Noise cancellation is performed based on the adaptive filter LMS. The weights are automatically adjusted through iterative learning to match the interference noise, thereby enhancing the characteristics of the direct-sequence spread spectrum signal, improving the signal-to-noise ratio of the direct-sequence spread spectrum signal, clearly presenting the outline of the direct-sequence spread spectrum signal in the frequency domain and forming significant spectral peaks, thus achieving efficient noise suppression. Step 2: Utilize the cyclostationary characteristics of the direct-sequence spread spectrum to distinguish between signal and noise, and extract the features of the target signal; The specific steps for step two are as follows: Step 201: Sample sequence preprocessing; To improve the stability of frequency domain estimation, the input sampling sequence is segmented; when the total number of sampling points is N, it is divided into segments. Groups, each group Data points: The segmentation strategy employs the Welch smoothing method for preprocessing to reduce the variance of spectral estimation and improve signal stability. Step 202: Short-time Fourier Transform; Perform a Fast Fourier Transform on each set of data to convert the time-domain signal to the frequency domain: The transformation is used to obtain the energy distribution of the signal at different frequencies; Step 203: Calculate the spectral correlation function; In cyclic spectrum calculation, the periodicity of a signal is extracted through the correlation between different frequency components; Calculate the cross-correlation of different frequency components: At this time, if Then the spectral correlation of white noise tends to 0, while the direct-sequence spread spectrum signal, due to its periodicity, will... A significant spectral peak is observed at the carrier frequency; Step 204: Calculate and output the cyclic spectrum; The mean value of the spectral correlation functions obtained from all segmented calculations is calculated to obtain the final cyclic spectrum estimate: The statistical stability of the cyclic spectrum is ensured by obtaining the final cyclic spectrum estimate, making signal detection more robust; Step 3: Based on the specific frequency characteristics of the cyclic spectrum, determine whether a direct-sequence spread spectrum (DSSS) signal is present and perform parameter estimation. The parameter estimation result is the DSS signal detection result under low signal-to-noise ratio (SNR), i.e., DSS signal detection under low SNR is achieved based on noise cancellation. The parameters include carrier frequency and pseudocode rate. Step 301: Obtain two-dimensional slices of the cyclic spectrum And perform feature extraction; Step 302: Use peak-to-average power ratio (PAPR) for signal detection and decision-making; Step 303: Estimate the carrier frequency of the direct-sequence spread spectrum signal; Step 304: Estimate the pseudocode rate; The adaptive noise canceller has two input channels: a main input channel and a reference input channel; The main input channel contains the useful signal. and unrelated interference noise The reference input channel only contains... Related but with irrelevant noise The direct-sequence spread spectrum (DSS) signal detection system adjusts the tap weights using an adaptive filter. This makes the filter output Approaching And obtain the optimized signal estimate through error calculation: For mean square error Minimize: When the filter output Optimal approximation hour, Minimum output signal of direct-sequence spread spectrum signal detection system Depend on Composition to achieve effective noise reduction.

2. The method for detecting low signal-to-noise ratio direct-sequence spread spectrum signals based on noise cancellation as described in claim 1, characterized in that: The specific implementation method of the noise cancellation method of minimum mean square error adaptive filtering in step one is as follows: Step 101: Initialize filter parameters and initial tap weights. ; Step 102, filter tap weights at time n The input vector is represented as Calculate the output of the LMS adaptive filter. Step 103, Calculate the main channel signal Error with filter output: Error signal Reflects noise components that the filter has not yet eliminated; Step 104, Tap Weight Update; When the prior information of the unknown signal is unknown, the steepest descent algorithm is used to adjust the filter weight vector along the steepest descent direction of the performance surface, search for the minimum point of the performance surface, and the iterative formula for calculating the weight vector is: in Step size factor The gradient descent is calculated using the following formula: Step 105, Error Correction and Convergence Judgment; if the mean square error If the value is less than the set threshold, stop the iteration; otherwise, continue adjusting the filter weights. Step 106: Repeat steps 102-105 until the system outputs a signal. Depend on The composition is that noise suppression reaches its optimal state.

3. The method for detecting low signal-to-noise ratio direct-sequence spread spectrum signals based on noise cancellation as described in claim 2, characterized in that: In step two, Welch's algorithm is used to divide a long sequence into multiple fixed-length random sequences, calculate the cyclic spectrum of each sequence, and finally calculate the average. By using the segmented averaging method, the final cyclic spectrum has statistical properties, the variance is reduced, and the time domain is smoothed. By replacing the corresponding quantities in the continuous formula of the cyclic spectral correlation function with discrete values, we obtain the discrete implementation of the spectral correlation function: 。 4. The method for detecting low signal-to-noise ratio direct-sequence spread spectrum signals based on noise cancellation as described in claim 1, characterized in that: Step 301 describes obtaining a two-dimensional slice of the cyclic spectrum. The method for feature extraction is as follows: Because the energy of the direct-sequence signal is at the cycle frequency The cyclic spectrum is concentrated at a certain frequency, while Gaussian white noise only exists at the zero cyclic frequency. Therefore, the distribution characteristics of the cyclic spectrum in different frequency dimensions are used to detect the direct-sequence spread spectrum signal. Frequency extraction from the calculated three-dimensional cyclic spectrum Two-dimensional slices, namely: In this slice, the direct-sequence spread spectrum (DSS) signal exhibits significant spectral peaks at non-zero cyclic frequencies, while the noise component is concentrated at zero cyclic frequencies.

5. The method for detecting low signal-to-noise ratio direct-sequence spread spectrum signals based on noise cancellation as described in claim 1, characterized in that: The method for signal detection and decision using peak-to-average power ratio (PAPR) described in step 302 is as follows: Set a detection threshold and search for the maximum value on the cycle frequency axis. Simultaneously calculate the average energy of the cyclic spectrum. ; To improve the accuracy of the judgment, the peak-to-average power ratio (PAPR) is calculated: and the set judgment threshold Compare; If the preset threshold is exceeded, it is determined that a direct-sequence spread spectrum signal exists in the sampling sequence; if the threshold is not reached, the detection is determined to have failed and the current signal detection process is terminated.

6. The method for detecting low signal-to-noise ratio direct-sequence spread spectrum signals based on noise cancellation as described in claim 1, characterized in that: The method for estimating the carrier frequency of the direct-sequence spread spectrum signal in step 303 is as follows: The maximum value obtained from step 302 Its corresponding cycle frequency is : in The maximum value of the cyclic spectrum The corresponding cycle frequency; Based on the peak characteristics of the cyclic spectrum, a priori-free blind estimation of carrier frequency is achieved.

7. The method for detecting low signal-to-noise ratio direct-sequence spread spectrum signals based on noise cancellation as described in claim 1, characterized in that: The method for estimating the pseudocode rate in step 304 is as follows: Based on the signal carrier frequency estimated in step 303 Select another two-dimensional slice from the cyclic spectrum, namely the frequency. Cyclic spectrum slice at: In this slice, the pseudocode rate of the direct-sequence spread spectrum signal This is reflected in the peak value of the cycle frequency of the maximum value of this slice: 。

Images