Multi-thread-based vita packet real-time processing method
By adopting multi-threaded processing and deep neural network recognition technology in the radar signal detection system, combining power spectral density and cycle-related processing algorithms, the problems of vita packet loss and buffer area backlog in the existing system are solved, and efficient signal reception is achieved and processing meets the system's real-time needs.
Patent Information
- Application Number
- CN202411133925.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-19
- Publication Date
- 2024-12-03
AI Technical Summary
The existing radar signal detection and identification systems cannot receive and process vita packets at the same time during signal analysis and processing, resulting in vita packet loss and buffer area backlog, which cannot meet the system's real-time needs.
The real-time processing method of multi-threaded vita packets is adopted to channelize the radar signal through the signal processing board to generate continuous and burst signal vita packets, and the data receiving thread, data analysis thread and signal processing thread are used for parallel processing. Combining power spectral density, cyclic correlation processing algorithms and double sliding window burst signal detection algorithms, convolutional neural networks, cyclic neural networks and deep belief neural networks are used for intelligent target recognition.
It improves the accuracy and real-time nature of vita packet reception and processing, reduces false alarms and missed detection rates, improves the utilization rate of computing resources, and achieves good real-time and identification stability.
Smart Images

Figure CN119064866A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of information technology, and in particular to a multi-thread-based vita package real-time processing method. Background Art
[0002] In the field of radar, signal detection and recognition technology is widely used in target detection and recognition. How to accurately and quickly detect and recognize signals has become an urgent technical problem that needs to be solved.
[0003] At present, in practical applications, the signal receiving thread of the signal detection and recognition system reads the vita packet in real time into a first-in-first-out circular buffer. The reception and analysis of the vita packet mostly adopt a single-threaded mode. In this mode, the vita packet cannot be received simultaneously during the signal analysis and processing, resulting in the loss of the vita packet and an increase in the missed detection rate. Similarly, signal analysis and processing cannot be performed while the vita packet is being received, resulting in a backlog of vita packets in the buffer and delays in signal analysis and processing, which cannot meet the real-time requirements of the system.
[0004] In order to meet the system's real-time requirements for signal analysis and processing, it is necessary to find a solution with fast computing speed and good real-time performance to receive, process and analyze the vita package. Summary of the invention
[0005] In view of the above analysis, an embodiment of the present invention aims to provide a multi-threaded based vita packet real-time processing method to solve the problem that the accuracy and real-time performance of vita packet reception and processing need to be improved during the signal analysis and processing in the existing radar signal reception and processing.
[0006] The embodiment of the present invention provides a multi-threaded vita packet real-time processing method, which uses a signal processing board to perform channelization processing on a radar signal to obtain multiple DDC channel signals, performs signal detection on each DDC channel signal, generates a vita packet of a continuous signal if a continuous signal is detected, and generates a vita packet of a burst signal if a burst signal is detected; stores the generated multi-channel vita packet data in a DDR on a signal processing board; uses a development tool to create a data receiving thread, a data parsing thread and multiple signal processing threads, and creates a buffer area, each signal processing thread corresponds to a DDC channel; the data receiving thread transfers the vita packet data in the DDR to the buffer area; the data parsing thread parses the vita packet data according to information in the vita packet header, and stores the vita packet data in a storage unit corresponding to a channel number; the signal processing thread performs conventional signal recognition or target intelligent recognition on the data in the storage unit of the corresponding channel number according to the signal type in the vita packet header; the data receiving thread, the data parsing thread and the multiple signal processing threads perform parallel processing on the data stored in the buffer.
[0007] Further, the signal detection for each DDC channel signal includes: performing continuous signal detection using the power spectral density, and performing burst signal detection by combining the cyclic correlation processing algorithm and the double-sliding window burst signal detection algorithm.
[0008] Further, the DDC channel signal is a time-domain signal; performing burst signal detection by combining the cyclic correlation processing algorithm and the double-sliding window burst signal detection algorithm includes: taking the original time-domain signal of the DDC channel as the input original sequence of the cyclic correlation processing algorithm, comparing the sequence generated after circularly shifting the input original sequence with the input original sequence, calculating the correlation between the shifted sequence and the original sequence to obtain the cyclic autocorrelation result, taking the DDC channel cyclic autocorrelation result sequence as the input signal power spectral density sequence of the double-sliding window burst signal detection algorithm, setting a sliding window A and a sliding window B for the signal power spectral density sequence, and performing a double-threshold comparison on the data in the sliding window A and the sliding window B to determine the start position and the end position of the burst signal.
[0009] Further, the double-threshold comparison of the data in the sliding window A and the sliding window B to determine the start position and the end position of the burst signal includes: setting sliding windows A and B with a length of L for the signal power spectral density sequence, sliding the sliding windows A and B synchronously, and having an overlap of L / 2 between the sliding windows A and B; calculating the root mean square values of the data in the sliding windows A and B respectively to obtain PA and PB each time of sliding; when PB / PA > Th1, taking the position where the window B is located as the start position of the burst signal in the input sequence, and when PB / PA < Th2, taking the position where the window B is located as the termination position of the burst signal; where Th1 and Th2 are adaptive double thresholds, Th2 = 1.5 / Th1, Th1 > Th2, and the range of L is [300, 400].
[0010] Further, the continuous signal detection using the power spectral density includes: dividing the finite-length DDC channel signal into multiple overlapping segments with a length of L1, having an overlap of L1 / 4 between adjacent two segments; estimating the power spectral density function of each segment of data using the average modified periodogram method, and if the power spectral density of each segment of data with a length of L1 is greater than the set threshold within the preset bandwidth range, considering the entire finite-length channel signal as a continuous signal, and the range of L1 is [300, 500].
[0011] Further, when the data in the vita packet is a continuous signal, the signal processing thread performs conventional signal recognition; when the data in the vita packet is a burst signal, the signal processing thread uses a deep neural network to perform target intelligent recognition of the target individual, platform type, and communication system.
[0012] Furthermore, the deep neural network includes: a convolutional neural network, a recurrent neural network, and a deep belief neural network cascaded in sequence; the data of each channel obtained by parsing the vita package data by the data parsing thread is input into the convolutional neural network; the signal features of each channel extracted by the convolutional neural network model are spliced to form a tensor as the input of the recurrent neural network model; the recurrent neural network model outputs a sequence containing time series features to the deep belief neural network, and the deep belief neural network identifies the target individual, platform type, and communication system based on the sequence of time series features.
[0013] Furthermore, the vita packet consists of a header and a valid data payload; the first 7 words of each vita data packet are the header, and the valid data payload starts from the 8th word; the header includes the packet length, data packet sequence number, timestamp, data packet type, channel number, signal ID, and reserved words. The data packet type includes two types: continuous signal and burst signal. The channel number corresponds to the DDC channel number, and the signal ID is the number of burst data packets detected by the current channel.
[0014] Furthermore, when the vita data packet type is burst data, if the data length point is greater than L2, the signal processing thread discards the burst data, where the L2 range is [1M, 1.8M].
[0015] Furthermore, the data receiving thread determines the current receiving data time t by obtaining the value of the high-precision performance counter and the frequency of the performance counter. Current The time t when the last data was received Previous Whether the comparison exceeds the specified time Δt Special , when t Current -t Previous >Δt Special When the communication interface is in operation, the internal function is called to complete the batch reading of DDR data, each time reading data of a specific size until all data in the DDR memory are read and stored in the first-in-first-out circular buffer.
[0016] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:
[0017] 1. The real-time processing method of the vita package adopted in the present invention distributes tasks to a data receiving thread, a data parsing thread and multiple signal processing threads for parallel execution on the Windows operating system, thereby improving the utilization rate of computing resources, and timely and effectively processing the vita package data collected by the FPGA and transmitted through the USB chip. A large number of experiments have verified that the false alarm and missed detection rates are extremely low.
[0018] 2. The present invention uses a combination of a cyclic correlation processing algorithm and a double sliding window burst signal detection algorithm to detect burst signals, which can accurately determine the start and end positions of burst signals. The present application uses cyclic autocorrelation results as power spectrum density sequences on the FPGA side, uses time domain signals, eliminates Fourier transforms, and has good real-time performance.
[0019] 3. The present invention adopts a combination of convolutional neural network, recurrent neural network and deep belief neural network, and inputs the data of each channel obtained by parsing the data of the vita package by the data parsing thread into the convolutional neural network; the convolution layer in the convolutional neural network can combine the features extracted by the previous layers to extract more complex features. The signal features of each channel extracted by the convolutional neural network model are spliced to form a tensor as the input of the recurrent neural network model; the recurrent neural network model outputs a sequence containing time series features to the deep belief neural network, and the deep belief neural network recognizes the target individual, platform type and communication system in real time based on the sequence of the time series features, and the recognition result is stable and reliable.
[0020] In the present invention, the above-mentioned technical solutions can also be combined with each other to achieve more preferred combination solutions. Other features and advantages of the present invention will be described in the subsequent description, and some advantages can become obvious from the description, or can be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained through the contents particularly pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] The drawings are only for the purpose of illustrating particular embodiments and are not to be considered limiting of the present invention. Like reference symbols denote like components throughout the drawings.
[0022] Figure 1 It is a principle flow chart of multi-threaded real-time analysis and processing of a multi-threaded vita package real-time processing method according to the present invention;
[0023] Figure 2 A flow chart of burst signal detection on a signal processing board of a multi-threaded vita package real-time processing method according to the present invention;
[0024] Figure 3 A schematic diagram of a signal processing and analysis method of a multi-threaded vita package real-time processing method according to the present invention;
[0025] Figure 4 A schematic diagram of the buffer writing principle of a multi-threaded vita package real-time processing method of the present invention;
[0026] Figure 5 The present invention is a schematic diagram of a buffer reading principle of a multi-threaded vita package real-time processing method. DETAILED DESCRIPTION
[0027] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.
[0028] A specific embodiment of the present invention discloses a multi-threaded vita package real-time processing method, such as Figure 1 As shown, the specific steps include S1-S3.
[0029] S1. Use the signal processing board to channelize the radar signal to obtain multiple DDC channel signals, perform signal detection on each DDC channel signal, and generate a vita packet of the continuous signal if a continuous signal is detected; if a burst signal is detected, generate a vita packet of the burst signal; store the generated multi-channel vita packet data in the DDR on the signal processing board.
[0030] Specifically, after receiving the broadband sampling signal, the signal processing board performs channelization processing on the broadband sampling signal, divides the broadband sampling signal in the frequency domain, and converts it into multiple narrowband signals after the channelization processing. Each narrowband signal is the result of digital down-conversion of the input broadband sampling signal, which is called a DDC channel signal. The signal processing board includes FPGA and DDR memory; FPGA creates two threads for simultaneously performing continuous signal and burst signal detection on the multiple DDC channel signals, generating vita packets of continuous signals and vita packets of burst signals, and storing them in the DDR memory.
[0031] The signal detection for each DDC channel signal includes: using power spectrum density to perform continuous signal detection, and using a combination of a cyclic correlation processing algorithm and a double sliding window burst signal detection algorithm to perform burst signal detection.
[0032] The continuous signal detection using power spectrum density includes: dividing a finite-length DDC channel signal into a plurality of partially overlapping segmented data with a length of L1, and two adjacent segmented data have an overlap of L1 / 4 length; estimating the power spectrum density function of each segmented data using the average modified periodogram method, if the power spectrum density of each segmented data with a length of L1 is greater than a set threshold within a preset bandwidth range, it is considered that the entire finite-length channel signal is a continuous signal, and the range of L1 is [300,500].
[0033] Specifically, the FPGA extracts the time-domain continuous signals existing in the channel through power spectral density features. Specifically, the average modified periodogram method is used to estimate the input power spectral density function. The finite-length observation data is segmented into overlapping segmented data, the number of segments in the average periodogram method is increased, and the variance of the power spectral estimation is reduced. The power spectral distribution histogram is statistically analyzed, and then the threshold method is used to detect the presence or absence of signals. The threshold is set to μ + kδ, where μ is the mean of each segmented data segment, δ is the variance of the segmented data segment, and the range of k is [10, 15].
[0034] The DDC channel signal is a time-domain signal; a combined method of a cyclic correlation processing algorithm and a double-sliding window burst signal detection algorithm is used for burst signal detection, including: using the original time-domain signal of the DDC channel as the input original sequence of the cyclic correlation processing algorithm, comparing the sequence generated after circularly shifting the input original sequence with the input original sequence, calculating the correlation between the shifted sequence and the original sequence to obtain the cyclic autocorrelation result, using the DDC channel cyclic autocorrelation result sequence as the input signal power spectral density sequence of the double-sliding window burst signal detection algorithm, setting a sliding window A and a sliding window B for the signal power spectral density sequence, and performing a double-threshold comparison on the data in the sliding window A and the sliding window B to determine the start position and the end position of the burst signal.
[0035] The double-threshold comparison of the data in the sliding window A and the sliding window B to determine the start position and the end position of the burst signal includes: setting the sliding window A and the sliding window B with a length of L for the signal power spectral density sequence, the sliding window A and the sliding window B slide synchronously, and the sliding window A and the sliding window B have an overlap of L / 2; each time they slide, the root mean square values are calculated for the data in the sliding window A and the sliding window B respectively to obtain PA and PB; when PB / PA > Th1, the position where the window B is located is used as the start position of the burst signal in the input sequence, and when PB / PA < Th2, the position where the window B is located is used as the termination position of the burst signal; where Th1 and Th2 are adaptive double thresholds, Th2 = 1.5 / Th1, Th1 > Th2, and the range of L is [300, 400].
[0036] The burst signal detection is performed using the double-sliding window burst signal detection method based on cyclic autocorrelation processing as Figure 2 shown.
[0037] Specifically, the cyclic correlation processing algorithm is adopted to effectively suppress the in-band noise of the channelization and reduce the detection discrimination threshold of the burst signal. For the double-sliding-window burst signal detection algorithm, by establishing a statistical distribution model of clutter and noise, an adaptive double threshold is set to maintain a constant false alarm rate. The detection threshold changes from the absolute value of the envelope of the original cyclic autocorrelation to the relative jump value of the energy between the burst signal and the noise floor. Assume that the two windows are window A and window B, and the sliding step of the sliding window A and the sliding window B is synchronized at L / 2. PA and PB represent the energy values of the two windows respectively. The double threshold values are Th1 and Th2. When PB / PA > Th1, there may be a burst signal in window B. According to PB / PA < Th2, the termination position of the burst signal is found. The setting of the Th1 threshold is determined by an empirical value coefficient and an optimization factor that changes with the window size. The optimization factor reaches the maximum when the window size reaches the minimum. The value of the Th2 threshold is related to the Th1 threshold, and Th2 = 1.5 / Th1. Combining the cyclic autocorrelation and the double-sliding-window technology can achieve high-sensitivity and high-probability detection of burst signals in a complex electromagnetic environment. The detected continuous signals and burst signals are respectively packed into corresponding vita packets and stored in the DDR address through the DMA technology, and the data is stored at the specified address in the DDR memory space.
[0038] The vita packet consists of a packet header and a valid data payload; the first 7 words of each vita data packet are the packet header, and the valid data payload starts from the 8th word; the packet header includes the packet length, data packet sequence number, timestamp, data packet type, channel number, signal ID, and reserved word. The data packet type includes two types: continuous signal and burst signal. The channel number corresponds to the DDC channel number, and the signal ID is the number of the burst data packet detected on the current channel.
[0039] Specifically, each word of vita is 32bit, and the reserved word facilitates the addition of new variables in the future.
[0040] S2. Use the development tool to create a data receiving thread, a data parsing thread, and multiple signal processing threads, and create a buffer. Each signal processing thread corresponds to a DDC channel.
[0041] A specific embodiment of the present invention uses VS2013 development tools under Windows operating system to develop a user operation interface and create multiple threads in the application program, including a data receiving thread, a data parsing thread, and 32 signal processing threads corresponding to 32 DDC channels. A DDC channel parameter class object is created for each DDC channel, and DDC channel parameter information is recorded and passed to the thread as a function input parameter when the signal processing thread is created. At the same time, a first-in-first-out loop buffer and an FPGA parameter update synchronization class object are created. After the application is initialized, the working frequency band is selected in the user operation interface, the DDC data update time, the DDC channel center frequency, the DDC channel bandwidth, the DDC analysis bandwidth are set, the normalized window coefficient, etc., and the parameter information is transmitted to the FPGA (the signal processing board includes a USB interface circuit for communicating the signal processing board data with the computer) by calling the USB communication interface function, and the FPGA performs data acquisition and processing according to the parameter information.
[0042] S3, the data receiving thread transfers the vita package data in the DDR to the cache area; the data parsing thread parses the vita package data according to the information in the vita package header, and stores the vita package data in the storage unit of the corresponding channel number; the signal processing thread performs conventional signal recognition or target intelligent recognition on the data in the storage unit of the corresponding channel number according to the signal type in the vita package header; the data receiving thread, the data parsing thread and multiple signal processing threads process the data stored in the cache in parallel.
[0043] In a specific embodiment of the present invention, in the VS2013 development tool, different functions of implementing the real-time analysis and processing tasks of the vita package are respectively implemented by multiple functions. For the multiple threads created in step S2, each thread executes a function, and multiple threads are executed concurrently. Data sharing and communication are realized through mutual exclusion fusion. The real-time analysis and processing tasks of the vita package are thus executed by different threads at the same time. The data analysis thread stores the vita package data in the member variables of the DDC channel parameter class corresponding to the channel number according to the channel number in the vita package header, and passes it to the corresponding signal processing thread as an input parameter. In the callback function, the data in the vita package passed by the data analysis thread is stored in another loop buffer area. The signal processing thread, as a thread function of the upper application class, reads data from the loop buffer area in real time, and uses the Windows API function to save the data for subsequent signal analysis and processing. The signal processing thread is responsible for signal processing of the data in the vita package, including conventional signal recognition and target intelligent recognition. Multi-threaded data sharing adopts a single-producer-multiple-consumer design pattern, and the circular buffer area for vita package transmission adopts a "first-in-first-out" data management method to improve the stability of vita package real-time analysis and processing.
[0044] The data receiving thread obtains the value of the high-precision performance counter and the frequency of the performance counter to determine the current receiving data time t Current The last time data was received is t Previous Whether the comparison exceeds the specified time Δt Special , when t Current -t Previous >Δt Special When the communication interface reads the data in batches from the DDR of the signal processing board, a function is called internally, reading data of a specific size each time until all the data in the DDR memory is read and stored in the first-in, first-out circular buffer.
[0045] In a specific embodiment of the present invention, first, the data receiving thread reads the FPGA register flag bit through the USB3.0 interface function to obtain the value stored at the specified address in the DDR storage space of the signal processing board. When the specified address has been written with data, the current receiving data time t is determined by obtaining the value of the high-precision performance counter and the frequency of the performance counter. Current The last time data was received is t Previous Whether the comparison exceeds the specified time Δt Special , condition t Current -t Previous >Δt Special When the conditions are met, the vita packets in the DDR storage queue are read and stored in the first-in-first-out circular buffer.
[0046] like Figure 4 As shown, when the relationship between the current write pointer position index writeindex (starting from 0) in the circular buffer, the length of the vita packet read from the DDR storage queue this time, and the first-in-first-out circular buffer size RING_BUFFER_SIZE satisfies the following two situations: Case I: writeindex+readlength≤RING_BUFFER_SIZE, the vita packets are written sequentially into the circular buffer, and the write pointer position index is updated to writeindex+readlength; Case II: writeindex+readlength>RING_BUFFER_SIZE, after the vita packet is written from the write pointer position index writeindex position to the end of the queue, the excess data is written from the head of the queue into the circular buffer, and the write pointer position index is updated to writeindex+readlength-RING_BUFFER_SIZE.
[0047] Determine the current receiving data time t Current The last time data was received is t Previous In comparison, when tCurrent -t Previous <Δt Special When the CPU is in use, the sleep function is used to put the current thread computing program to sleep, making it inactive for a period of time and giving the CPU usage rights to other threads.
[0048] The data parsing thread reads the vita packet from the circular buffer, the packet header length headsize, the read pointer subscript writeindex, the write pointer subscript readindex, and the read-write circle difference cntdiff (the read-write circle difference cntdiff is the difference in circles between the read and write pointers, and the value is 0 or 1. 0 means that the difference between the read and write pointers is within the buffer length, and 1 means that the write pointer exceeds the read pointer value and has exceeded the buffer length.) When the following conditions are met:
[0049] When the readindex+headsize+packetlength≤writeindex+cntdiff×RING_BUFFER_SIZE condition is met, the signal processing thread reads the vita packet from the circular buffer, where packetlength is the length of the valid data payload in the packet.
[0050] like Figure 5 As shown, there are three situations: Ⅰ. The read pointer subscript plus the length of the vita header and the effective data load length in the packet does not reach or just reaches the end of the buffer area. The vita header is read first, and the data in the packet is read in sequence according to the length of the data in the packet and the channel number recorded in the vita header, and stored in the member variable of the DDC channel parameter class of the corresponding channel number. Ⅱ. The read pointer subscript plus the length of the vita header does not reach the end of the buffer area, and the length of the data in the packet exceeds the length of the buffer area. When the data in the packet reaches the end of the buffer area, the read pointer returns to the beginning of the buffer area to continue reading. According to the length of the data in the packet and the channel number recorded in the vita header, the data in the packet is stored in the member variable of the DDC channel parameter class of the corresponding channel number. Ⅲ. The read pointer subscript plus the length of the vita header exceeds the length of the buffer area. When the data in the packet reaches the end of the buffer area, the read pointer returns to the beginning of the buffer area to continue reading until the vita header is read. Then, according to the data length and channel number in the vita packet header, the data in the packet is read sequentially and stored in the member variable of the DDC channel parameter class corresponding to the channel number.
[0051] When the data in the vita package is a continuous signal, the signal processing thread performs conventional signal recognition; when the data in the vita package is a burst signal, the signal processing thread uses a deep neural network to perform intelligent target recognition of the target individual, platform type, and communication system.
[0052] Specifically, conventional signal recognition can be based on observing harmonics, analyzing instantaneous amplitude, frequency, phase, frequency-time domain waveform, signal spectrum, etc. using constellation diagrams, or it can be identified using neural networks.
[0053] When the vita data packet type is burst data, if the data length point is greater than L2, the signal processing thread discards the burst data, where the L2 range is [1M, 1.8M].
[0054] Schematic diagram of signal processing and analysis method Figure 3 shown.
[0055] The present invention is a specific embodiment, simulates and generates continuous numerical value or single tone signal in FPGA, VS2013 develops VC++ multithreaded program, receives vita bag, and parses and processes vita bag data, returns the data in the package of vita bag to upper application by callback function, and stores corresponding file according to channel number, uses matlab tool to read data in storage file, then judges whether there is loss in continuous numerical value or single tone signal by matlab tool programming, and can also use its own drawing function to visualize data. From the simulation result, it can be seen that whether continuous numerical value or single tone signal is continuously without loss, thus reflecting that the present invention has very high real-time performance, which is greatly improved compared with previous single-thread system.
[0056] The deep neural network includes: a convolutional neural network, a recurrent neural network, and a deep belief neural network cascaded in sequence; the data of each channel obtained by parsing the vita package data by the data parsing thread is input into the convolutional neural network; the signal features of each channel extracted by the convolutional neural network model are spliced to form a tensor as the input of the recurrent neural network model; the recurrent neural network model outputs a sequence containing time series features to the deep belief neural network, and the deep belief neural network identifies the target individual, platform type, and communication system based on the sequence of the time series features.
[0057] In a specific embodiment of the present invention, 32 signal processing threads make real-time judgments. When the serial number corresponding to the current working parameter is the same as the serial number corresponding to the data stored in this thread, the corresponding callback function is called according to the data packet type to return the relevant information in the header of the vita packet and the data in the packet to the upper-layer application. When the data packet type of the vita packet is a continuous signal, the callback function corresponding to the continuous signal processing is called and sent to the upper-layer application for conventional signal recognition; when the data packet type is burst data, if the data is too long, it is discarded, otherwise the callback function corresponding to the burst signal processing is called and sent to the upper-layer application for target intelligent recognition, including recognition of target individuals, platform types, communication systems, etc. This target intelligent recognition method breaks through tradition and does not use the demodulation and translation method, so it can achieve a high recognition accuracy rate. The identification of target individuals and platform types adopts the target intelligent identification technology based on raw data, relying on deep neural networks to automatically learn and extract the signal depth features of target signals. In order to enable the network to learn the internal features of the signal, multiple paths are adopted, and multiple neural network models such as deep convolutional neural networks, recurrent neural networks and deep belief neural networks are cascaded at the same time. The convolutional neural network model processes the input raw collected data. As the network level deepens, the convolution layer can combine the features extracted by the previous level to extract more complex features. The signal features of each channel extracted by the convolutional neural network model are spliced to form a tensor as the input of the recurrent neural network model. The recurrent neural network model uses the cyclic kernel to realize time series feature extraction, and the extracted information is sent to the fully connected layer of the recurrent neural network model, and is converted into an abstract feature representation with the help of the fully connected layer to support the recognition task. The deep belief neural network is combined with other neural network structures for model training and classification, which can not only extract the local and global features of the signal at the same time, but also extract the time-invariant features and time-series related features at the same time, so as to represent the target from multiple angles and in depth. The recognition of the communication system adopts intelligent recognition technology based on multi-task learning. For the identification of communication systems, it is necessary to identify signal specifications such as modulation mode, channel bandwidth, coding method, network protocol, etc., and adopt multi-task learning based on deep neural network. Deep neural network regards multiple tasks as a whole, and realizes joint learning and optimization of multiple tasks by sharing the bottom network layer and training the top network layer separately.
[0058] Compared with the prior art, the real-time processing method of the vita package provided in this embodiment distributes tasks to the data receiving thread, the data parsing thread and multiple signal processing threads for parallel execution on the Windows operating system, thereby improving the utilization rate of computing resources, timely and effectively processing the vita package data collected by the FPGA and transmitted through the USB chip, and verifying through a large number of experiments that the false alarm and missed detection rates are extremely low. This embodiment uses a combination of a cyclic correlation processing algorithm and a double sliding window burst signal detection algorithm to detect burst signals, and can accurately determine the starting and ending positions of the burst signal. This application uses the cyclic autocorrelation result as the power spectral density sequence on the FPGA side, uses the time domain signal, omits the Fourier transform, and has good real-time performance. The convolutional neural network, recurrent neural network, and deep belief neural network combination used in this embodiment input the data of each channel obtained by parsing the vita package data through the data parsing thread into the convolutional neural network; the convolutional layer in the convolutional neural network can combine the features extracted by the previous layers to extract more complex features. The signal features of each channel extracted by the convolutional neural network model are spliced to form a tensor as the input of the recurrent neural network model; the recurrent neural network model outputs a sequence containing time series features to the deep belief neural network, and the deep belief neural network identifies the target individual, platform type, and communication system in real time based on the sequence of time series features, and the recognition results are stable and reliable.
[0059] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.
[0060] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.
Claims
1. A multi-threaded vita package real-time processing method, characterized in that: The signal processing board is used to perform channelization processing on radar signals to obtain multiple DDC channel signals. Signal detection is performed on each DDC channel signal. If a continuous signal is detected, a vita packet of the continuous signal is generated. If a burst signal is detected, a vita packet of the burst signal is generated; The generated multiple vita packet data is stored in the DDR on the signal processing board; A data receiving thread, a data parsing thread, and multiple signal processing threads are created using a development tool, and a buffer is created. Each signal processing thread corresponds to a DDC channel; The data receiving thread transfers the vita packet data in the DDR to the buffer; The data parsing thread parses the vita packet data according to the information in the vita packet header and stores the vita packet data in the storage unit corresponding to the channel number; The signal processing thread performs conventional signal recognition or target intelligent recognition on the data in the storage unit corresponding to the corresponding channel number according to the signal type in the vita packet header; The data receiving thread, the data parsing thread, and multiple signal processing threads perform parallel processing on the data stored in the buffer.
2. The real-time processing method of vita package according to claim 1, characterized in that: The signal detection for each DDC channel signal includes: using the power spectral density for continuous signal detection, and using a combination of the cyclic correlation processing algorithm and the double sliding window burst signal detection algorithm for burst signal detection.
3. The real-time processing method of vita package according to claim 2, characterized in that: The DDC channel signal is a time-domain signal; using a combination of the cyclic correlation processing algorithm and the double sliding window burst signal detection algorithm for burst signal detection includes: taking the original time-domain signal of the DDC channel as the input original sequence of the cyclic correlation processing algorithm, comparing the sequence generated after circularly shifting the input original sequence with the input original sequence, calculating the correlation between the shifted sequence and the original sequence to obtain the cyclic autocorrelation result, taking the DDC channel cyclic autocorrelation result sequence as the input signal power spectral density sequence of the double sliding window burst signal detection algorithm, setting a sliding window A and a sliding window B for the signal power spectral density sequence, and performing a double-threshold comparison on the data in the sliding window A and the sliding window B to determine the start position and end position of the burst signal.
4. The real-time processing method of vita package according to claim 3 is characterized in that: The double-threshold comparison of the data in the sliding window A and the sliding window B to determine the start position and end position of the burst signal includes: setting sliding windows A and B with a length of L for the signal power spectral density sequence. The sliding windows A and B slide synchronously, and there is an overlap of L / 2 between the sliding windows A and B; each time they slide, the root mean square values are calculated for the data in the sliding windows A and B to obtain PA and PB; when PB / PA > Th1, the position of window B is taken as the start position of the burst signal in the input sequence. When PB / PA < Th2, the position of window B is taken as the end position of the burst signal; where Th1 and Th2 are adaptive double thresholds, Th2 = 1.5 / Th1, Th1 > Th2, and the range of L is [300, 400].
5. The real-time processing method of vita package according to claim 1, characterized in that: The continuous signal detection using power spectrum density includes: dividing a finite-length DDC channel signal into a plurality of partially overlapping segmented data with a length of L1, and two adjacent segmented data have an overlap of L1 / 4 length; estimating the power spectrum density function of each segmented data using the average modified periodogram method, if the power spectrum density of each segmented data with a length of L1 is greater than a set threshold within a preset bandwidth range, it is considered that the entire finite-length channel signal is a continuous signal, and the range of L1 is [300,500].
6. The real-time processing method of vita package according to claim 1, characterized in that: When the data in the vita package is a continuous signal, the signal processing thread performs conventional signal recognition; when the data in the vita package is a burst signal, the signal processing thread uses a deep neural network to perform intelligent target recognition of the target individual, platform type, and communication system.
7. The real-time processing method of vita package according to claim 6, characterized in that: The deep neural network includes: a convolutional neural network, a recurrent neural network, and a deep belief neural network cascaded in sequence; the data of each channel obtained by parsing the vita package data by the data parsing thread is input into the convolutional neural network; the signal features of each channel extracted by the convolutional neural network model are spliced to form a tensor as the input of the recurrent neural network model; the recurrent neural network model outputs a sequence containing time series features to the deep belief neural network, and the deep belief neural network identifies the target individual, platform type, and communication system based on the sequence of the time series features.
8. The real-time processing method of vita package according to claim 1, characterized in that: The vita packet consists of a header and a valid data payload; the first 7 words of each vita data packet are the header, and the valid data payload starts from the 8th word; the header includes the packet length, data packet sequence number, timestamp, data packet type, channel number, signal ID, and reserved words. The data packet type includes two types: continuous signal and burst signal. The channel number corresponds to the DDC channel number, and the signal ID is the number of burst data packets detected by the current channel.
9. The real-time processing method of vita package according to claim 8, characterized in that: When the vita data packet type is burst data, if the data length point is greater than L2, the signal processing thread discards the burst data, where the L2 range is [1M, 1.8M].
10. The real-time processing method of vita package according to claim 1, characterized in that: The data receiving thread obtains the value of the high-precision performance counter and the frequency of the performance counter to determine the current receiving data time t Current The time t when the last data was received Previous Whether the comparison exceeds the specified time Δt Special , when t Current -t Previous >Δt Special When the communication interface is in operation, the internal function is called to complete the batch reading of DDR data, each time reading data of a specific size until all data in the DDR memory are read and stored in the first-in-first-out circular buffer.
Citation Information
Patent Citations
VPX-platform-based radar signal processing system and application software design method
CN107167773A
FPGA (field programmable gate array) Circuit board and telescope digital operational board
CN108196358A
Storage parsing device and storage parsing method based on 10-gigabit Ethernet interface receiver
CN109547355A
X-band multi-pulse radar echo high-speed parallel processing system
CN114280567A
Techniques for implementing a portable spectrum analyzer
US20180091336A1
Cited By
Time domain burst signal extraction method and device, electronic equipment and storage medium
CN120145028A
