A distributed data transmission system and method based on FPGA
By using FPGA in a distributed computing system to realize full duplex communication between the host and slave, using SFP and Aurora interfaces and PCI Express protocols, the problem of inefficiency in traditional centralized computing is solved, and high-speed and secure data is achieved Transmission and parallel processing improve computing efficiency and accuracy.
Patent Information
- Application Number
- CN202110125304.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-01-29
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2041-01-29
AI Technical Summary
Traditional centralized computing has inefficient problems in data processing and high-speed transmission, especially the complexity of the Ethernet protocol and physical layer limitations, which lead to insufficient data transmission rates to meet the needs of modern applications.
Design a distributed data transmission system based on FPGA, adopts full duplex communication between the host and slave, and realizes high-speed data transmission through the SFP interface, Aurora interface and PCI Express protocol, and utilizes the low latency and high connectivity of FPGA Features: combining DDR4 memory and DMA technology to realize high-speed data transmission and parallel processing.
It realizes low-cost and efficient data transmission, with a transmission speed of 410MB/s, supports various file formats, is highly secure, shortens the calculation time of convolutional neural network and random forest algorithm, and improves the accuracy and speed of data processing. .
Smart Images

Figure CN112947156B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of distributed data transmission, and in particular relates to a distributed data transmission system and method based on FPGA. Background Art
[0002] With the development of computing technology, people have put forward higher requirements for both high-speed data processing and high-speed data transmission. Many practical applications require a huge amount of computing to support. If traditional centralized computing is used, it will take a long time to complete data calculation. Therefore, a faster data processing method is needed to replace traditional centralized computing. Distributed computing came into being.
[0003] Distributed computing relies on the high-speed interconnection of multiple computing devices, so there are certain requirements for data transmission rates. Most of the existing distributed computing platforms rely on the Internet to achieve data transmission. The biggest advantage of transmitting through the Internet is that it achieves low-cost data interaction. However, with the increasing demand for data transmission rates, Ethernet is increasingly difficult to meet the requirements. Due to the complexity of the Ethernet protocol and the physical layer, its transmission rate will not be too high, which greatly affects the real-time data transmission of data volume calculations. FPGA, or field programmable gate array, is a product of the development of relatively mature programmable devices after PAL, GAL, and CPLD. Its low latency, high connectivity, and parallel transmission characteristics make FPGA the best platform for switching platforms to achieve high-speed data communication. Summary of the invention
[0004] In view of the above problems, the present invention studies and explores the data transmission of distributed computing, and proposes a reliable technical solution for full-duplex communication, high-speed data transmission, and distributed processing between the host and the slave: a distributed data transmission system based on FPGA, including several FPGAs and several PCs;
[0005] Among them, each FPGA is connected to each PC for communication, one PC among several PCs is used as a host, and the rest of the PCs are used as slaves, and each FPGA is connected in turn;
[0006] The FPGA includes an SFP interface, a receiving control module, an Aurora receiving end, a receiving FIFO, a sending control module, a sending FIFO, an Aurora sending end, a control register, a GPIO, a DMA, an XDMA, an AXI Interconnect and a DDR4 memory; wherein, at least two SFP interfaces are provided, which are respectively connected to the Aurora receiving end and the Aurora sending end; the Aurora receiving end is controlled by the receiving control module to receive data, which is stored in the receiving FIFO after receiving; the Aurora sending end is controlled by the sending control module to send data, and the sent data is taken out from the sending FIFO; the control register receives data in the DMA via the GPIO, the control register is respectively connected to the receiving control module and the sending control module, and the DMA is also respectively connected to the receiving FIFO and the sending FIFO; the DMA and the XDMA are both connected to the AXI Interconnect, and the AXI Interconnect is also connected to the DDR4 memory.
[0007] Preferably, the FPGAs communicate with each other via an SFP connection.
[0008] Preferably, the FPGA communicates with the PC via a PCIE connection.
[0009] Preferably, the DMA is connected to the GPIO via an AXI_Lite bus.
[0010] Preferably, the DMA is connected to the AXI Interconnect via an AXI bus.
[0011] Preferably, the XDMA is connected to the AXI Interconnect via an AXI bus.
[0012] Preferably, the AXI Interconnect is connected to the DDR4 memory via an AXI bus.
[0013] Preferably, the PC includes a file selection module, a distribution mode selection module, a file list display module, a site calculation selection module and a file transfer speed measurement module.
[0014] Preferably, the file selection module selects the file type and sends it to the FPGA through PCIE, or receives data from the FPGA through PCIE; the distribution mode selection module is used to select the distribution method, and a back-end node can be fixed or sent evenly; the file list display module is used to display the name, type and size of the sent file; the site calculation selection function is used to select the type of distributed computing, including CNN convolutional neural network training and random forest data analysis; the file transfer speed measurement module is used to display the transmission speed during the transmission process.
[0015] Based on the above purpose, the present invention also provides a distributed data transmission method based on FPGA, using the above distributed data transmission system based on FPGA, comprising the following steps:
[0016] S10, first select the file to be distributedly processed on the PC serving as the host, and then upload the file;
[0017] S20, the file is transmitted to the FPGA connected to the PC as the host through PCIE, and then transmitted to the DDR4 memory of the FPGA connected to the slave PC through the SFP interface of the FPGA;
[0018] S30, the slave PC selects a receiving file location and starts receiving;
[0019] S40, the FPGA connected to the slave PC receives the signal and transmits the data to the slave PC through PCIE. When all the files are saved on the slave PC, it means that one transmission is completed.
[0020] The beneficial effects of the present invention include at least:
[0021] 1. The present invention has low cost and relatively complete functions. It has 4 SFPs and one PCIE. The actual transmission speed can reach 410MB / s. It can recognize various file formats, so it can realize high-speed transmission of various files. It is equipped with upper computer software and is relatively easy to operate.
[0022] 2. The data exchange between FPGAs of the present invention is carried out using the Aurora 64b / 66b high-speed serial protocol, and the data exchange between FPGA and PC uses the PCI Express 3.0X8 standard protocol. The transmission speed is faster than that of USB3.0, and data is not easily lost, with higher security.
[0023] 3. The platform is used to build convolutional neural networks on the host and slave machines and perform parallel distributed computing, and the time is greatly shortened to half of the original time.
[0024] 4. Through the high-speed interconnection of the FPGA's distributed data transmission platform and the parallel calculation of the random forest algorithm, large data are processed and analyzed in blocks, and finally processed by two hosts through voting, which greatly improves the speed and accuracy of the random forest algorithm. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Figure 1 It is a structural block diagram of the distributed data transmission system based on FPGA of the present invention;
[0026] Figure 2 It is a schematic diagram of the FPGA structure of the distributed data transmission system based on FPGA of the present invention;
[0027] Figure 3 This is a distributed file transmission system interface diagram of the FPGA-based distributed data transmission system of the present invention;
[0028] Figure 4 This is a PC interface diagram of a sending end node of a distributed data transmission system based on FPGA of the present invention;
[0029] Figure 5 This is a PC interface diagram of a receiving end node of the FPGA-based distributed data transmission system of the present invention;
[0030] Figure 6 This is an interface diagram of a data acquisition unit of a distributed data transmission system based on FPGA of the present invention;
[0031] Figure 7 It is a DMA pin diagram of the distributed data transmission system based on FPGA of the present invention;
[0032] Figure 8 A format diagram of a file transmission data packet of the FPGA-based distributed data transmission system of the present invention;
[0033] Fig. 9 It is a Verilog HDL hierarchical schematic diagram of the control logic of the distributed data transmission system based on FPGA of the present invention;
[0034] Fig.10 This is a data waveform diagram of the Aurora sending end of the FPGA-based distributed data transmission system of the present invention. DETAILED DESCRIPTION
[0035] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
[0036] See also Figure 1 , Figure 2 , a distributed data transmission system based on FPGA, comprising a plurality of FPGAs 10 and a plurality of PCs 20;
[0037] Among them, each FPGA 10 is respectively connected to each PC 20 for communication, one PC 20 among the several PCs 20 acts as a host, and the other PCs 20 act as slaves, and each FPGA 10 is connected in turn;
[0038] FPGA 10 includes an SFP interface 11, a receiving control module 111, an Aurora receiving end 112, a receiving FIFO 113, a sending control module 114, a sending FIFO 115, an Aurora sending end 116, a control register 110, a GPIO 19, a DMA 13, an XDMA 15, an AXIInterconnect 16, and a DDR4 memory 17; wherein at least two SFP interfaces 11 are provided, which are respectively connected to the Aurora receiving end 112 and the Aurora sending end 116, and the Aurora receiving end 116 is connected to the Aurora receiving end 112. The receiving end 112 is controlled by the receiving control module 111 to receive data, and the receiving data is stored in the receiving FIFO 113. The Aurora transmitting end 116 is controlled by the transmitting control module 114 to send data, and the transmitted data is taken out from the transmitting FIFO 115. The control register 110 receives the data in the DMA 13 via the GPIO 19. The control register 110 is connected to the receiving control module 111 and the transmitting control module 114 respectively. The DMA 13 is also connected to the receiving FIFO 113 and the transmitting FIFO 115 respectively. Both the DMA 13 and the XDMA 15 are connected to the AXI Interconnect 16, and the AXI Interconnect 16 is also connected to the DDR4 memory 17.
[0039] The FPGAs 10 communicate with each other via SFP, and the FPGAs 10 and the PC 20 communicate with each other via PCIE.
[0040] DMA13 is connected to GPIO19 via AXI_Lite bus 18; DMA13 is connected to AXI Interconnect16 via AXI bus 14; XDMA15 is connected to AXI Interconnect16 via AXI bus 14; AXI Interconnect16 is connected to DDR4 memory 17 via AXI bus 14.
[0041] Among them, the single-channel small fiber optic pluggable transceiver SFP (Small Form-factor Pluggable), the high-speed serial computer expansion bus PCIE (Peripheral Component Interconnect Express), the fourth-generation double-rate synchronous dynamic random access memory DDR4 (Dual Data Rate SDRAM 4), DMA13 (Direct Memory Access), XDMA15 (DMA13 / Bridge Subsystem for PC20I Express, DMA13 / Bridge Subsystem for PC20I Express), AXI Interconnect16 (AXI bus type 14 interface connector.
[0042] PC20 includes a file selection module, a distribution mode selection module, a file list display module, a site calculation selection module, and a file transfer speed measurement module. The file selection module selects the file type and sends it to FPGA10 through PCIE, or receives data from FPGA10 through PCIE; the distribution mode selection module is used to select the distribution method, and a backend node can be fixed or sent evenly; the file list display module is used to display the name, type and size of the sent file; the site calculation selection function is used to select the type of distributed computing, including CNN convolutional neural network training and random forest data analysis; the file transfer speed measurement module is used to display the transmission speed during the transmission process.
[0043] FPGA10 is used as a conversion medium for interaction between PC20 hosts. The data exchange between FPGA10 and PC20 adopts the PC20I Express 3.0X8 standard protocol, with a theoretical speed of more than 50Gbps. The Aurora 64b / 66b protocol is used for data transmission between FPGA10 platforms, and the single-channel transmission speed of this protocol can reach more than 10Gbps. Each PC20 is connected to an FPGA10, and the FPGA10s are also connected through the SFP interface 11. For the convenience of description, the PC20 that sends data is called Master PC20, the FPGA10 connected to the Master PC20 is called Master FPGA10, the PC20 that receives data is called Slave PC20, and the FPGA10 connected to the Slave PC20 is called Slave FPGA10; it is obvious that the Master FPGA10 and the Slave FPGA10 are connected through the SFP interface 11 to transmit data.
[0044] The entire data flow is as follows: Master PC20 writes the data and instructions to be sent to Slave PC20 into the DDR4 memory 17 space of Master FPGA10 through the PCIE interface, and then the data is routed through the fiber optic data of the Aurora64b / 66b protocol; Slave FPGA10 receives the corresponding fiber optic data for analysis, caches it into the DDR4 memory 17 space of the Slave FPGA10 and notifies Slave PC20 to read the data. Slave PC20 obtains the data through the PCIE interface and then analyzes the data to form the corresponding file information, indicating that the transmission is completed. As a transmission intermediary, FPGA10 not only needs to implement the read and write control of PCIE, Aurora64b / 66b protocol, and DDR4, but also needs to cache data, parse instructions, route ports, and ensure seamless coordination between interfaces.
[0045] The FPGA10 is divided into two parts. One is the DMA13 part: this part has two data paths, one is the memory mapping of the AXI bus 14, and the other is the register mapping interface of AXI_Lite. Through the former, the PCIE data transmitted by PC20 can be directly mapped to the DDR4 memory 17 space of FPGA10, and through the latter, PC20 can control the Slave end of the AXI_Lite bus 18 interface as GPIO19, that is, large data transmission can be achieved through the AXI bus 14, and the transmission of control instructions can be achieved through the AXI_Lite bus 18. In addition, 5 GPIO19 are added, which are used to receive PC20 control information, FPGA10 control information, transmission file size, DDR4 memory 17 space and other control information. PC20 can monitor these five control registers 110 in real time and exchange lightweight data with them, which will not affect the normal PCIE large data stream transmission. The second is the data transmission part: in order to realize the full-duplex working mode, and because the sending and receiving ends of the optical fiber not only need to send data information, but also to send instruction information (such as the size of a transmission, the source ID of the transmission, the destination ID of the transmission, etc.), it is necessary to write the corresponding control logic for read and write arbitration and parse the corresponding information. In addition, the transmitted data needs to be exchanged with the memory through the DMA13 controller, so the DMA13 controller needs to design the corresponding control logic; the data from the memory to the Aurora module belongs to the cross-clock domain transmission, so the input and output ends need to add FIFO to realize the cross-clock domain. Therefore, the SFP interface 11, the receiving control module 111, the Aurora receiving end 112, the receiving FIFO 113, the sending control module 114, the sending FIFO 115, the Aurora sending end 116 and the control register 110 are set.
[0046] On the PC20 side, select the local file to be transferred in the file selection module; then select the distribution method in the distribution mode selection module, you can choose to fix a backend node, or send it evenly; select the calculation you want to perform in the site calculation selection module: CNN (Convolutional Neural Networks) convolutional neural network training or random forest data analysis; right-click the file in the file list display module and select Start Sending, and the data will start to be transmitted; during the transmission process, the file transmission speed measurement module will automatically pop up to display the size of the transmitted file and the transmission speed. In this process, the file and memory mapping mmap (memory map) technology is used to greatly improve the reading efficiency of large files. In order to avoid a large amount of data passing through the PCIE interface at once and causing data blocking, the size of the data block distributed each time is controlled. Use the file memory pointer index to read in blocks to reduce unnecessary CPU burden during the transmission process. After the receiving node completes the reception, it can use the advantages of distributed storage to perform block calculations. The CNN convolutional neural network uses the advantages of parallel convolution of pixels to distribute a large number of image data sources to different hosts to make up for the limits of the computing power of a single computer memory and CPU. The Python language rPC20 remote procedure call method is used to perform calculation adjustments on the slave nodes and the return value results of the distributed calculations. The host node then summarizes the return values to achieve a total-to-sum structure. Users can select the required transfer files and corresponding functions in the host computer without having any knowledge of FPGA principles and VIVADO software operations.
[0047] Method Embodiment
[0048] A distributed data transmission method based on FPGA10, using the above-mentioned distributed data transmission system based on FPGA10, see Figure 3 The method is a PC20 interface diagram, and includes the following steps:
[0049] S10, first select the file to be processed in a distributed manner on PC 20 as the host, and then upload the file, see Figure 4 ;
[0050] S20, the file is transmitted to the FPGA 10 connected to the PC 20 as the host through the PCIE, and then transmitted to the DDR4 memory 17 of the FPGA 10 connected to the slave PC 20 through the SFP interface 11 of the FPGA 10;
[0051] S30, the slave PC20 selects a receiving file location and starts receiving. Figure 5 ;
[0052] S40, the FPGA 10 connected to the slave PC 20 receives the signal and transmits the data to the slave PC 20 through the PCIE. When all the files are saved on the slave PC 20, it indicates that one transmission is completed.
[0053] See also Figure 6 The figure shows the speed measurement interface of the file transfer of the present invention. After the file transfer is completed, the interface shows that the file saving speed reaches about 410Mb / s. To achieve high-speed data transmission, the present invention selects the PC20I Express 3.0 protocol standard, whose communication rate is theoretically 8GT / s and the measured speed is 3.5GB / s. FPGA10 sets 4 SFP fiber interfaces, uses G bit high-speed transceiver GTH (one of Gigabyte Transceiver) as the physical layer, uses Aurora64b / 66b as the transport layer protocol, and the theoretical line speed of a single channel is about 2GB / s. The maximum line speed measured by the board is 1.3GB / s. Through theoretical calculation, the actual speed of a single channel is up to 3.5G / 4≈800MB / s. After removing the fiber duplex read-write arbitration and memory read-write arbitration time, the bandwidth of a PCIE X8 interface can carry a maximum speed of about 700MB / s for 4-channel SFP fiber.
[0054] PCIE is directly connected to DDR4 memory 17 through AXI Interconnect 16, so PC20 can easily exchange data with DDR4 memory 17: large data transmission is realized through AXI bus 14, and control instruction transmission is realized through AXI_Lite bus 18, and the IP core of XDMA15 needs to make corresponding conversions to the base address and offset address of AXI_Lite bus 18. However, if the user module wants to obtain DDR4 memory 17 data, it must also encapsulate the module as AXI Master (AXI bus type master interface) to read and write DDR4 memory 17 through AXI Interconnect 16 (AXI bus type interface connector). This is because the data of DDR4 memory 17 is transmitted according to the AXI bus type interface, and data needs to be read and written with different modules, so AXI Interconnect 16 is needed to route the data. The function realized by DMA13 is direct memory acquisition, which is to complete the conversion function from AXI bus 14 to user interface in terms of the essential implementation method. The present invention adopts the open source DMA control module provided by MSXBO, and reads and writes data to the DDR4 memory 17 by controlling the corresponding request signal, and can also design information such as burst length, bus width, and burst quantity. In addition, the data interface of the module can be directly used as the read and write enable signal of the input and output buffer FIFO. The port of the module is shown in FIG. Figure 7 .
[0055] The FPGA distributed transmission designed by the present invention is a full-duplex working mode. Because the sending and receiving ends of the optical fiber not only send data information, but also send instruction information, such as the size of a transmission, the source end ID of the transmission, and the destination end ID of the transmission, it is necessary to write corresponding control logic for read-write arbitration and parse the corresponding information. Figure 8 The data packet format shown. In addition, the transmitted data needs to be exchanged with the memory through the DMA controller, so the DMA controller needs to design the corresponding control logic; the data from the memory to the Aurora module belongs to the cross-clock domain transmission, so the input and output ends need to add FIFO to realize the cross-clock domain, and the control lines of the two FIFOs are also classified as control logic design. The control logic of the present invention is designed in Verilog HDL language, and the design level is shown in Fig. 9 The data waveform of the Aurora64b / 66b sending part inside the FPGA during the transmission process can be seen in Fig.10 .
[0056] In practice, a transmitted file may fail to verify the entire file due to a one-byte data transmission error. In addition, the file formats are diverse, so a consistent communication protocol must be defined for reception and transmission. When the system transmits data, there is only raw memory data, and the receiving end cannot know the type of file, so the files must be classified in advance, such as pdf, mp4, jpg, png, etc. Simple encoding and marking operations are performed according to different file types, so that the receiving end can save the corresponding file type through marking. Two bytes are opened in front of the data to be transmitted to store the encoding corresponding to the file type. Different file types correspond to different data headers, and the data header is judged in advance to select the file type to be saved. In order to prevent data sticking, the transmitted data header not only has the file type but also records the file size to avoid receiving too much or too little data that is inconsistent with the capacity of the original data, resulting in recovery failure. Therefore, eight bytes of space are opened after the file type, and the remainder calculation is performed according to the file capacity size, and a set operation is performed at the corresponding position of the eight bytes. Therefore, the present invention is designed Figure 8 communication protocol.
[0057] During transmission, the receiving end first reads the memory header data to determine the file type, and decodes the corresponding different file types through the pre-specified protocol standard. Then read the file size of the following few bytes, and calculate the accurate file size through the conversion between hexadecimal and decimal. After obtaining the file size, change the corresponding type of file saved in the segment through the index of the memory pointer to ensure the integrity of the transmitted data. Because the working efficiency of the PC's CPU is limited, if the capacity of a local file transmitted by the sender is too large, it will still read the hard disk data multiple times according to the traditional data stream reading method of the transmission file, which will waste a lot of time and even cause the reading process to be blocked. At this time, the PC is always in the state of reading the file, and the reading process will have the consequence of not responding for a long time. If the blocking time is too long, the program will crash and the file reading will fail. For this reason, it is necessary to make a judgment first. If the file is too large, such as the file size is greater than 500MB, it is necessary to use mmap file memory mapping, and directly map the file data to the memory segment, so that the CPU does not need to switch to read the memory data directly, and the reading speed is greatly improved. This operation requires the use of the PC's virtual memory and ensures that the mapped file size is smaller than the virtual memory reserved by the PC. If it is larger than the virtual memory, it needs to be segmented and read sequentially through the head pointer to improve the system's transmission efficiency and avoid the system's "pseudo-death" state.
[0058] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made in form and details without departing from the scope defined by the claims of the present invention.
Claims
1. A distributed data transmission system based on FPGA, characterized in that: Including several FPGAs and several PCs; Among them, each FPGA is connected to each PC for communication, one PC among several PCs is used as a host, and the rest of the PCs are used as slaves, and each FPGA is connected in turn; The FPGA includes an SFP interface, a receiving control module, an Aurora receiving end, a receiving FIFO, a sending control module, a sending FIFO, an Aurora sending end, a control register, a GPIO, a DMA, an XDMA, an AXI Interconnect and a DDR4 memory; wherein at least two SFP interfaces are provided, which are respectively connected to the Aurora receiving end and the Aurora sending end, the Aurora receiving end is controlled by the receiving control module to receive data, and the data is stored in the receiving FIFO after receiving, and the Aurora sending end is controlled by the sending control module to send data, and the sent data is taken out from the sending FIFO; the control register receives data in the DMA via the GPIO, the control register is respectively connected to the receiving control module and the sending control module, and the DMA is also respectively connected to the receiving FIFO and the sending FIFO; the DMA and the XDMA are both connected to the AXI Interconnect, and the AXI Interconnect is also connected to the DDR4 memory; The PC includes a file selection module, a distribution mode selection module, a file list display module, a site calculation selection module and a file transfer speed measurement module; The file selection module selects the file type and sends it to the FPGA through PCIE, or receives data from the FPGA through PCIE; the distribution mode selection module is used to select the distribution method, and can optionally fix a back-end node or send it evenly; the file list display module is used to display the name, type and size of the files that have been sent; the site calculation selection function is used to select the type of distributed computing, including CNN convolutional neural network training and random forest data analysis; the file transfer speed measurement module is used to display the transmission speed during the transmission process.
2. The distributed data transmission system based on FPGA according to claim 1, characterized in that: The FPGAs communicate with each other via SFP connection.
3. The distributed data transmission system based on FPGA according to claim 1, characterized in that: The FPGA communicates with the PC via a PCIE connection.
4. The distributed data transmission system based on FPGA according to claim 1, characterized in that: The DMA is connected to the GPIO via an AXI_Lite bus.
5. The distributed data transmission system based on FPGA according to claim 1, characterized in that: The DMA is connected to the AXIInterconnect via an AXI bus.
6. The distributed data transmission system based on FPGA according to claim 1, characterized in that: The XDMA is connected to the AXI Interconnect via an AXI bus.
7. The distributed data transmission system based on FPGA according to claim 1, characterized in that: The AXIInterconnect is connected to the DDR4 memory via an AXI bus.
8. A distributed data transmission method based on FPGA, using the distributed data transmission system based on FPGA according to any one of claims 1 to 7, characterized in that: The following steps are involved: S10, first select the file to be distributedly processed on the PC serving as the host, and then upload the file; S20, the file is transmitted to the FPGA connected to the PC as the host through PCIE, and then transmitted to the DDR4 memory of the FPGA connected to the slave PC through the SFP interface of the FPGA; S30, the slave PC selects a receiving file location and starts receiving; S40, the FPGA connected to the slave PC receives the signal and transmits the data to the slave PC through PCIE. When all the files are saved on the slave PC, it means that one transmission is completed.
Citation Information
Patent Citations
Aurora protocol based method for conducting high-speed interconnection between field programmable gate arrays (FPGAs)
CN102681971A
Magnetic resonance echo signal simulator and simulation system and signal processing method of magnetic resonance echo signal simulator
CN103592611A
Data packet switching method and device
CN104219591A
Multi-protocol high-speed pseudo-random signal loopback test system based on FPGA
CN111010241A
Distributed data transmission system based on FPGA
CN215006296U
