88 results about "Cyclic spectrum" patented technology

The invention discloses a radar radiation source recognition method based on a deep learning strategy and a multitask learning strategy and mainly aims to solve the problem that recognition accuracy is low in the prior art. The method comprises the implementation steps that 1, an original radar radiation source signal is subjected to data preprocessing; 2, envelope features, fuzzy function features, slice features, cyclic spectrum features and frequency spectrum features of the preprocessed radar radiation source signal are extracted, and values of the features are linearly transformed into [0,255] and saved as an image set; 3, a convolutional neural network (CNN) is designed, and the multitask learning strategy and a random inactivation strategy are introduced into the CNN; and 4, four feature training sets are used to train the CNN, then four trained CNN models are utilized to classify four feature test sets, and a radar radiation source recognition result is output. The method is high in recognition accuracy and can be applied to electronic intelligence reconnaissance, electronic support reconnaissance and radar threat warning systems.

The invention relates to an idle frequency spectrum detecting method by using cyclic spectrum statistic value in a cognitive radio. The invention relates to a method for detecting idle frequency spectrums by using cyclic spectrum statistic value. The method solves the error judgment problem caused by that in the existing cognitive radio, owing to the influence of factors such as shadow, shading depth and the like, a cognitive user detects feeble signals of an authorized master user. The method comprises the following steps of: firstly, modeling the real part of the cyclic spectrum of signals received by the cognitive user to obeyed mean value and random variable of variance in a gauss white noise channel; step two, by probability distribution under the H0 assumed condition, and obtaining signal judgment threshold under the given false-alarm probability index; step three, allowing the idle frequency spectrum to be used when C is larger than or equal to 0 and the real part Z of the cyclic spectrum of the received signals is less than the judgment threshold T, or when C is less than 0 and the real part Z of the cyclic spectrum of the received signals is larger than the judgment threshold T; otherwise, not allowing the non idle frequency spectrum to be used. The method can cause the cognitive user to detect the signals of the authorized master user under the lower signal-to-noise ratio condition.

The invention discloses a cooperative spectrum sensing method of a cognitive radio network considering malicious nodes. The cooperative spectrum sensing method is characterized in that the nodes involved in cooperative sensing are subjected to cyclic spectrum sensing to obtain characteristics of spectrum resources of a master user; normally sensed nodes and maliciously sensed nodes report sensed information to a data fusion center through an orthorhombic common control channel; the data fusion center perform data fusion on the collected sensed information and calculate the false alarm probability of the false alarm probability according to the malicious attach pattern of the malicious nodes; a secondary user and the master user share the spectrum resources, build a optimizing model, determine the constraint conditions of the transmitting power and the sensing time, solve the built optimizing problem, and repeat the steps to make the obtained average value to be the sensed parameter of the spectrum sensing model. The cooperative spectrum sensing method is fast in calculation and strong in pertinence and universality, and can correctly select the optimizing sensing period and the signal transmitting power in the sensing environment with the malicious nodes.

The invention discloses a digital modulation identifying method based on fractional lower order cyclic spectrum related coefficient under the non-gaussian noise in a cognitive radio, and the digital modulation identifying method is capable of solving the problems of bad modulation identifying performance and high calculating complexity under the background of the non-gaussian noise in the cognitive radio. The method comprises the following steps: sampling received signals; calculating relative coefficients rho 1, rho 2, rho 3, rho 4 and rho 5 of projections of fractional lower order cyclic spectrums of the fractional lower order cyclic spectrum calculating signals of the sampled signals at a section of the cycle frequency epsilon=0, a section of frequency f=0, and the cycle frequency epsilon face, and the projection at the frequency f face; and arranging a judgment threshold of a signal set, and identifying the signals in different modulating manners through a classifier based on a judging tree. Under a non-gaussian alpha stably distributed noise, the digital modulation identifying method is relatively high in identification rate, good in stability, and lower in calculation complexity, and is more suitable for a cognitive radio system.

A Wireless Regional Area Network (WRAN) receiver comprises a transceiver for communicating with a wireless network over one of a number of channels, and a signal detector for use in forming a supported channel list comprising those ones of the number of channels upon which an Advanced Television Systems Committee (ATSC) DTV (digital television) broadcast signal was not detected. The signal detector performs spectrum sensing as a function of power spectral density (PSD) and cyclic spectrum.

The invention relates to a broadband signal detection and identification method based on Nyquist under-sampling. The method includes the following procedures that Nyquist under-sampling data of a front end are simulated and serve as the input of a signal reconstruction module, signal reconstruction is based on an SOMP algorithm, an energy observed value is generated each time iteration is conducted and used for spectrum detection, and meanwhile recovered frequency domain signals are used for cyclic spectrum estimation. In spectrum detection, a constant false alarm detector is adopted for making a broadband spectrum binary judgment and a multi-user identification module uses user bandwidth constraint for eliminating glitches generated by the constant false alarm detector. A cyclic spectrum estimation module uses the recovered signals and a multi-user identification result for estimating a cyclic spectrum of each user and finally, the modulation format identification, the symbol rate estimation and the carrier estimation of each user signal are achieved according to the features of cyclic spectrums of various digital communication signals. According to the method, wide spectrum detection and digital communication signal detection can be achieved at the same time.

An orthogonal frequency division multiplexing (OFDM) system sampling frequency shift blind estimation method under a multipath fading channel includes: according to features of stable circulation of OFDM signals with relevant pilot frequency, an improved spectral function overcomes fading brought by sampling frequency shift and data influenced by the multipath fading channel to estimate relative value points and estimate the sampling frequency shift according to cyclic spectrum value phase shift amount at a estimated value point. The OFDM system sampling frequency shift blind estimation method under the multipath fading channel can overcome influence of frequency shift, white gaussian noise and the multipath fading channel, makes full use of data by using jump transformation, and greatly improves estimate performance of sampling frequency shift of the OFDM system.

The invention discloses a method and system for monitoring an electromagnetic spectrum, wherein the method comprises the following steps of: obtaining a time-domain signal output by an intermediate-frequency receiver; and converting the time-domain signal into a frequency-domain signal; computing each characteristic cross section of the cyclic spectrum of the frequency-domain signal; and monitoring the electromagnetic spectrum by utilizing the conversion result of the signal and / or each characteristic cross section of the cyclic spectrum obtained through computation. The invention can be utilized for completely realizing the monitoring processing for the electromagnetic spectrum in a frequency domain and can be simply and effectively applied to sensor nodes.

The invention discloses a method for constructing a cyclic spectrum characteristic parameter extraction model and a method for identifying a signal modulation mode. The method comprises the followingsteps: preprocessing an input modulation signal to obtain a cyclic spectrum; Extracting characteristic parameters of the cyclic spectrum, and training, verifying and testing the CNN model by utilizingthe characteristic parameters to obtain a characteristic parameter extraction model; Inputting the modulation signal into a characteristic parameter extraction model to obtain a characteristic parameter sample set; Training an identification model by taking the characteristic parameter sample set as input and taking a corresponding modulation mode as output so as to obtain a signal modulation mode identification model; Inputting the to-be-detected signal into the characteristic parameter extraction model, inputting the output characteristic parameter into the signal modulation mode recognition model, and obtaining a signal modulation mode of the to-be-detected signal. According to the method, the CNN architecture is adopted to identify the signal modulation mode, and the characteristic extraction of the modulation mode is embedded into the convolutional neural network, so that the characteristics can be automatically extracted in the training process, and the identification of the signal modulation mode is realized.

The invention discloses an estimation method of DS (direct sequence) / FH (frequency hopping) spread spectrum signal parameters based on a cyclic spectrum theory and a splicing technology, which is mainly used for solving the problem of poor estimation performance of DS / FH spread spectrum signal parameters in the existing method. The estimation method comprises the following steps: firstly pre-estimating carrier frequency and code rate of DS / FH spread spectrum signals according to the cyclic spectrum theory; dividing and extracting the DS / FH spread spectrum signals based on the pre-estimated carrier frequency and code rate; splicing the divided and extracted signals; and finally estimating the carrier frequency and code rate of the spliced signals according to the cyclic spectrum theory. The estimation method has the advantages of strong anti-jamming capability, high parameter estimation accuracy and capability of effectively reducing estimation errors, thus the method can be used for processing communication signals.

The invention discloses a method for identifying a common digital modulation signal based on cyclic spectrum correlation. The reliability of signal analysis is improved by utilizing the noise-proof feature of a signal cyclic spectrum; the steps of alpha section wavelet de-noising and averaging through superposition are introduced into a calculation process of a signal spectral correlation function, so that the random fluctuation caused by the limited sampling number and the external disturbance in an original spectrum correlation estimation algorithm result is effectively weakened to facilitate identification and extraction of modulation features; and meanwhile, an alpha section and an f section of an obtained spectral correlation diagram are computed by utilizing signal spectral correlation, and appropriate features and parameters (such as a ratio of maximum absolute values of spectral correlation functions, namely the alpha section and the f section, the number of intense lines of the alpha section, a coefficient of fluctuation of the alpha section, the normalized area of the f section, a predominance ratio of spectral lines of the alpha section and the like) are selected to construct a classification method to identify the modulation mode of the communication signal.

The invention provides a carrier and clock combined synchronization method for OQPSK modulation. The method comprises the following steps of: performing coarse estimation on carrier frequency by use of the cyclic spectrum characteristic of an OQPSK signal; performing series expansion on the error function of a COSTAS loop and simplifying the error function; and estimating the residual carrier frequency offset based on the maximum likelihood principle, and meanwhile, estimating the clock. According to the carrier and clock combined synchronization method for OQPSK modulation, the coarse frequency offset of the carrier is obtained by use of a square spectrum estimation method and quick locking of a large frequency offset signal is realized; the error function of a carrier synchronization loop is expanded in Taylor series and optimized so that the complexity of an algorithm is greatly reduced under the condition of guaranteeing the synchronization performance, and the accuracy of a signal synchronization result is ensured by use of the carrier synchronization loop so that the signal length needed by calculating the cyclic spectrum of the signal can be reduced; based on the maximum likelihood principle, sliding accumulation is performed on two paths of IQ signals after filtering time delay and an angle of amplitude is calculated, and meanwhile, the residual carrier frequency offset and the clock phase are estimated, so that mutual influence of clock estimation and carrier estimation is avoided.

The invention discloses a cooperative spectrum sensing method in a cognitive vehicular ad-hoc network. The main process is that each cognitive vehicle receives authorized user signals in a band of interest and the following operation is carried out: (1) a cyclic ambiguity function method is adopted to carry out Doppler frequency shift estimation on the received authorized user signals; (2) a double-threshold cyclic spectrum energy detection method is adopted for spectrum sensing, and when the cyclic spectrum energy value is larger than a large threshold value or smaller than a small threshold value, the local judgment result acquired by the cognitive vehicle and the position information are transmitted to a road side unit on a public control channel; and finally, the road side unit fuses information of cognitive vehicles participating in cooperation through a position relative decision method to judge whether the authorized user band is free. The method considers influences on detection by the Doppler frequency shift, the spectrum information at all cyclic frequencies is used, a cooperative weighting factor is changed dynamically according to the real-time change of the relative position between the cognitive vehicles, and the detection performance is improved.

The invention discloses a spectrum sensing method based on a quantum particle swarm optimization extreme learning machine. The method relates to the field of cognitive radio, solves the problems thatthe main user signal detection rate is low under the condition of low signal-to-noise ratio in the existing wireless channel environment, a traditional extreme learning machine algorithm is only basedon empirical risk minimization and is easy to overfit, the network structure is poor and the like, and comprises the following steps of extracting the signal cyclic spectrum characteristics and the energy characteristics; constructing a training data set; training a QPSO-ELM spectrum sensing model according to the obtained training data set; inputting the extracted energy characteristics and cyclic spectrum characteristics of the received signals into the spectrum sensing model trained in the step 3 as detection data to realize the spectrum sensing of the main user signals, and determining that a main user exists when the output of the spectrum sensing model is 1; and when the output is 0, determining that the main user does not exist. According to the method, through the optimization ofthe quantum particle swarm and the introduction of structural risks, the algorithm can extract the input features more effectively, and the false alarm probability is relatively lower.

The invention provides an unmanned aerial vehicle signal recognition method and device, electronic equipment, a storage medium and an unmanned aerial vehicle signal recognition method. The unmanned aerial vehicle signal recognition method comprises the steps that discrete target image transmission signals of a target unmanned aerial vehicle are acquired; determining a target cyclic spectrum corresponding to the target image transmission signal; and identifying the target cyclic spectrum by using a trained recognition model, and determining model information of the target unmanned aerial vehicle. According to the method, the type information of the unmanned aerial vehicle is identified on the basis of the cyclic spectrum, and independent extraction of signal features can be avoided, so thatthe influence of noise and interference can be weakened, the accuracy of signal recognition is improved, and relatively high recognition accuracy can also be guaranteed in a low signal-to-noise ratioenvironment.

The invention provides a deep learning intelligent modulation identification method based on cyclic spectrum estimation. The method comprises the following steps of generating a modulation signal according to a carrier frequency and a code element rate; performing cyclic spectrum estimation on the modulation signal, and extracting a sectional drawing of a cyclic spectrum function; training a deepneural network by taking the sectional view as a feature; and identifying the modulation mode of the unknown signal by using the deep neural network. Cyclic spectrum estimation and the deep neural network are combined, and the performance of the whole signal modulation identification system is improved by utilizing the intelligent processing capability of the neural network and the relatively goodclassification identification capability of the cyclic spectrum. Only the sectional view of the cyclic spectrum function is utilized, the step of extracting cyclic spectrum features is omitted, and the time complexity of the algorithm is reduced.

The invention relates to a cognitive radio frequency spectrum sensing method based on circulation symmetry, belonging to the field of communication. The invention aims at solving the problems that the calculated quantity is high, the operation is complex and the accuracy of frequency spectrum sensing is low under the condition of low signal to noise ratio in the traditional method for realizing radio frequency spectrum sensing by judging whether cyclic spectrum of a received signal has symmetry. The method provided by the invention comprises the following steps of: 1, sampling a radio signal,and acquiring the cyclic spectrum of the radio signal by adopting an SSCA (stochastic sparse-grid collocation algorithm) algorithm; 2, selecting 15 pairs of symmetric points on the cyclic spectrum acquired in the step 1; 3, calculating the sum of the amplitude differences of the 15 pairs of symmetrical points selected in the step 2; and 4, judging whether the sum of the amplitudes differences of the 15 pairs of symmetrical points is less than a symmetry decision threshold, if the sum is less than the symmetry decision threshold, judging that a master user signal is existed in a channel; and if the sum is not less than the symmetry threshold, judging that no master user signal is existed in the channel. The accuracy of spectrum sensing under the condition of low signal-to-noise ratio is obviously improved.

The invention discloses an interference signal modulation recognition method for a communication carrier monitoring system, and the method comprises the following steps: S1, carrying out the offline learning, constructing a cascaded ResNet neural network classifier, and carrying out the modulation type classification recognition of a generated two-dimensional signal analysis graph; building a full-connection BP network classifier to capture characteristic parameter information of the residual signal r; and S2, performing online learning, calculating and forming a two-dimensional signal analysis graph and cyclic spectrum statistical characteristic parameters of r, then inputting the two-dimensional signal analysis graph and the cyclic spectrum statistical characteristic parameters into thecascaded ResNet neural network and the full-connection BP network for modulation type classification and identification, and providing a final prediction result. The invention provides a neural network modulation recognition classifier based on parallel connection of the cascaded ResNet neural network and the full-connection BP neural network. Learning and mining structural features of a two-dimensional signal analysis graph of the interference signal are performed by the cascaded ResNet neural network; the full-connection BP neural network learns to mine cyclic spectrum parameter characteristics of the interference signals; and by combining the two neural networks for processing and judgment, the modulation recognition rate of interference signals is effectively improved.

The invention discloses a motor rolling bearing partial pitting fault diagnosis method and a diagnosis system thereof. The motor rolling bearing partial pitting fault diagnosis method comprises the following steps: arranging a measuring point at an end cover of a bearing seat of the motor rolling bearing, wherein a vibration acceleration sensor collects a vibration signal of the measuring point; generating a kurtosis distribution diaphragm based on the vibration signal, determining filter center frequency and frequency band width corresponding to the maximum kurtosis value; performing filtering processing on the vibration signal based on the filter center frequency and frequency band width so as to obtain the filtered characteristic enhancement signal; performing cyclostationary analysis based on the characteristic enhancement signal so as to generate a cyclic spectrum density map; detecting whether the bearing fault characteristic frequency is existent in a cyclic frequency domain ofthe cyclic spectrum density map, and determining the vibration frequency of the fault impact waveform based on the cyclic frequency domain, thereby diagnosing the motor rolling bearing pitting fault.

The invention discloses a method for acquiring a cyclic spectrum alpha section based on frequency domain smoothing. The method comprises the following steps of: setting a certain time length for the received signal to perform discrete Fourier transform (DFT); and fixing a Fourier frequency value k, utilizing the calculating result of the discrete Fourier transform, setting a value of a cyclic frequency variable alpha as m1+m2 and selecting two smoothing windows to perform cross transposition and performing frequency domain smoothing calculation to acquire the cyclic spectrum alpha section corresponding to the cyclic frequency variable alpha, wherein the centers of the two smoothing windows are k+m1 and k-m2 respectively and the lengths of the two smoothing windows are 2M+1. In the method,the cyclic spectrum alpha section is acquired through the cross transposition of the two smoothing windows in the frequency domain, the spectrum information at the odd number position of the section can be effectively recovered, so that the spectrum information of the odd position in the cross section can be effectively restored, alpha section spectrum information loss caused by a DFT barrier effect can be reduced, the spectrum information of the acquired cyclic spectrum alpha section is more complete, and the requirements of subsequent signal processing processes, such as parameter extraction, modulation mode identification and the like, can be met.

The invention relates to a signal estimation method in a non-reconstruction framework, belonging to the field of cognitive radio parameter identification and estimation. In order to solve the problems of slow reconstruction speed and poor accuracy in using an existing reconstruction algorithm to restore a signal, the method comprises a step of establishing an association between a sample signal cyclic spectrum vector Sx(c) and a sample signal cyclic autocorrelation vector rx, a step of establishing an association between a sampling signal compression measurement value autocorrelation vector rz and the sample signal cyclic autocorrelation vector rx, a step of establishing the relation between the sampling signal compression measurement value autocorrelation vector rz and the sample signal cyclic spectrum vector Sx(c), a step of deleting the redundant elements in the sample signal cyclic spectrum vector Sx(c), and obtaining a simplified sample signal cyclic spectrum vector Sxs(c), a step of reconstructing the simplified sample signal cyclic spectrum vector Sxs(c) by using the sampling signal compression measurement value autocorrelation vector rz and an orthogonal matching tracking algorithm based on block sparse, and obtaining an original signal cyclic spectrum, and a step of extracting the parameter information of the original signal according to the original signal cyclic spectrum, and a step of extracting the parameter information of the original signal according to the original signal cyclic spectrum. The method is mainly used for extracting the signal parameter information.

The invention discloses a method for improving the accuracy of signal recognition of an unmanned aerial vehicle. The method comprises the following steps of: initializing a signal processing environment and framing a time domain signal; and passing the framed signal unit through a band pass filter; calculating variable delay autocorrelation function of a signal unit to combine the peak search to obtain the useful symbol duration of the signal; calculating the fixed delay cyclic spectrum function of the signal unit to combine the peak search to obtain the symbol duration of the signal; calculating the cyclic prefix length; calculating the subcarrier spacing of the signal; and calculating the number of subcarriers of the signal. The characteristic parameters calculated according to the abovesteps are compared with the signals in the spectrum feature library to achieve the classification and recognition of the signals transmitted by the unmanned aerial vehicle; the accuracy of the calculation of the characteristic parameters is effectively improved; the useful symbol duration is calculated at a faster speed; the amount of calculation is greatly reduced, and the system response sensitivity is improved; and a powerful support is provided for the subsequent demodulation and decoding of signals.

The invention discloses a communication fingerprint identification method integrating multi-layer sparse learning and multi-view angle learning, which comprises the following steps: 1) adopting a sparse automatic encoder to suppress noise for an original steady-state signal and an original transient-state signal; Carrying out bispectrum analysis and cyclic spectrum analysis on the de-noised signal, and obtaining characteristics on a transform domain by using a sparse coding method based on an over-complete signal dictionary; 2) for the second-order matrix form features on the transform domain,adopting a sparse coding method to obtain low-dimensional features which describe the fine features of the signal more simply and accurately; 3) for the radio station with multiple frequency points and multiple modulation modes, in order to comprehensively extract the common characteristics of the radio station under different working carrier frequencies and modes, adopting tree structure sparsecoding, and 4) from the characteristics of different visual angles, adopting multi-visual-angle canonical correlation analysis to carry out fusion of multiple sparse coding characteristics, and adopting a full connection neural network to carry out classification.

The invention provides a spectrum sensing method, a spectrum sensing system, a client and a server. The spectrum sensing method includes the following steps that: an energy sensing mode is adopted to judge whether interference signals exist; a cyclic spectrum characteristic sensing mode is adopted to judge whether interference signals exist; and whether interference users exist is judged according to the judgment results of the energy sensing mode and the cyclic spectrum characteristic sensing mode. According to the spectrum sensing method, the energy sensing mode and the cyclic spectrum characteristic sensing mode are adopted to perform mixed sensing on spectra, so that coarse sensing and fine sensing combined spectrum hybrid sensing can be realized, and therefore, the precision and speed of detection for the interference users can be improved.

The invention relates to a method for identifying a digital modulation signal containing frequency deviation and phase deviation under a multipath fading channel, which comprises the steps of (1) sampling a received MPSK (Multi-Phase Shift Keying) signal y(t) to obtain y(n); (2) calculating a cyclic spectrum shown in the abstract; (3) calculating an amplitude (shown in the abstract) of a spectrum coherent equation of a digital modulating signal; (4) searching the maximum value of the amplitude on a frequency shaft f, and taking the maximum value as an identified characteristic value; (5) setting a judgment threshold of a signal set to be as the formula shown in the abstract, wherein delta i is a threshold values of adjacent signals or signals Y1 and Y2, max (rY1) is the maximum value of the mean value of characteristic values of the signal Y1, and max (rY2) is the minimum value of the mean value of the characteristic values of the signal Y2; and (6) identifying BPSK (Binary Phase Shift Keying), QPSK (Quadrature Phase Shift Keying ) and 8PSK (8-Phase Shift Keying) signals containing frequency deviation and phase deviation according to the set judgment threshold.

The invention relates to an MIMO-SCFDE self-adaptive transmission scheme based on model-driven deep learning. According to the method, a self-adaptive transmission model is established based on an MIMO SCFDE system. AMNet and ADNet are adopted to replace a signal modulation part and a modulation identification part in a traditional system respectively. The AMNet adopts a combined network taking a2D CNN, an LSTM and an FCDNN as sub-networks to form an integrated neural network model, a modulation mode of a sending end is adjusted according to a channel condition of a receiving end, feature information extracted from a received signal is input into the plurality of sub-networks, and conversion between features and an optimal modulation scheme are achieved according to network parameters obtained by training. Meanwhile, the receiving power under different path delays is selected as an adaptive factor to achieve adaptive integration of each sub-network result. The ADNet completes adaptiveselection of a modulation identification scheme based on the complexity of a cyclic spectrum according to the advantage that the cyclic spectrum has accurate detection on the signal type under a lowsignal-to-noise ratio. The system is more suitable for performance requirements of a 5G communication system.

The invention provides a rolling bearing fault diagnosis method and system based on discrete cosine cyclic spectrum coherence, and belongs to the technical field of mechanical equipment fault diagnosis. The method comprises the steps: obtaining a time domain vibration signal of a rolling bearing; extracting discrete cosine cyclic spectrum coherent features from the time domain vibration signals to obtain a two-dimensional discrete cosine cyclic spectrum coherent feature map; and obtaining a final diagnosis result according to the two-dimensional discrete cosine cyclic spectrum coherent feature map and a preset convolutional neural network model. Based on the discrete cosine cyclic spectrum coherent features and the improved convolutional neural network model, accurate and rapid diagnosis of the rolling bearing fault can be realized under data distribution change conditions such as data imbalance and working condition change.

The invention relates to a frequency spectrum detection method based on characteristic circulation frequency in wireless medical monitoring. In the invention, when signals are processed with limited-length cyclic spectrum treatment, only cyclic spectrum numerical value at the signal characteristic circulation frequency part is processed based on characteristic of fixed frequency points of the wireless medical system, thereby reducing processing complexity and solving low power consumption of wireless medical devices; in addition, the signal detection probability and false-alarm probability are controlled by regulating decision threshold. Based on the basic theory of cyclic spectrum detection, a limited signal length is adopted and frequency spectrum detection is conducted at the signal characteristic circulation frequency part; when the cyclic spectrum numerical value at the characteristic circulation frequency part is greater than a preset threshold value, the frequency spectrum is deemed to be occupied; and when the cyclic spectrum numerical value at the characteristic circulation frequency part is less than a preset threshold value, the frequency spectrum is deemed to be not occupied. The method improves frequency spectrum detection performance greatly, and realizes frequency spectrum detection under extra-low signal to noise ratio.

The invention provides a modulation signal identification method based on deep learning. The method comprises the following steps: generating different types of modulation signals containing noise; carrying out wiener filtering noise reduction on the noise-containing modulation signal; performing cyclic spectrum estimation on the modulated signal after noise reduction, and extracting a cyclic spectrum two-dimensional sectional view; constructing a deep neural network, inputting the cyclic spectrum two-dimensional sectional view as an input feature into the deep neural network, and training the deep neural network; and identifying a modulation mode of an unknown signal by using the trained deep neural network. The invention further provides a modulation signal recognition system based on deep learning, noise reduction processing is performed on the modulation signal through Wiener filtering, and the influence of noise on recognition precision can be effectively reduced; meanwhile, the cyclic spectrum two-dimensional sectional view is used as an input feature, on one hand, the cyclic spectrum two-dimensional sectional view is not sensitive to noise, so that the influence of the noise on an identification result can be effectively reduced, and on the other hand, the complexity of an algorithm can be greatly reduced, and the identification efficiency can be improved.