CN111683111B

CN111683111B - Interferometry multi-phase channelization baseband conversion system based on GPU

Info

Publication number: CN111683111B
Application number: CN202010306476.3A
Authority: CN
Inventors: 马宏; 焦义文; 陈永强; 吴涛; 刘燕都; 李贵新; 史学书; 张威; 蔡洋; 曹玉凡
Original assignee: Peoples Liberation Army Strategic Support Force Aerospace Engineering University
Current assignee: Peoples Liberation Army Strategic Support Force Aerospace Engineering University
Priority date: 2020-04-17
Filing date: 2020-04-17
Publication date: 2021-04-06
Anticipated expiration: 2040-04-17
Also published as: CN111683111A

Abstract

The invention provides a multi-phase channelized baseband conversion system for interferometry based on a GPU (graphics processing unit). firstly, based on a GPU platform, data and a process are decoupled to the maximum extent by utilizing multi-thread block scheduling and multi-phase shunt, so that the parallelism of the system is improved; then, by utilizing a multi-stage streaming processing mode, the space complexity is exchanged for the data processing time complexity, the parallelization of the processing flow is realized, the reliability is higher, the expansibility is better, the precision is higher, the development difficulty of a baseband conversion system is effectively reduced, and the requirement of deep space measurement and control interferometry on the baseband conversion system can be met.

Description

Interferometry multi-phase channelization baseband conversion system based on GPU

Technical Field

The invention belongs to the technical field of communication, and particularly relates to an interferometric multi-phase channelized baseband conversion system based on a GPU.

Background

The interferometric technique has extremely high angular resolution, and is widely applied to deep space measurement and control systems. The baseband converter is an important subsystem of the interference measurement system and is responsible for completing the functions of AD conversion, amplitude adjustment, channelized reception, sub-band extraction, formatted transmission, storage and the like of analog measurement and control signals. The traditional baseband converter is mostly realized by adopting a hardware board card based on FPGA, and a core baseband conversion algorithm is completed by the FPGA.

The traditional interference measurement baseband converter generally meets the current observation requirements of deep space exploration and geodetic survey, plays an important role in lunar exploration engineering and geodetic survey in China, but along with multidirectional expansion of deep space measurement and control network frequency bands, bandwidth, functions and the like, a large gap exists between the state of the traditional baseband converter and task requirements, and the traditional baseband converter is mainly embodied in the following aspects:

(1) the system development period is long, the reconstruction is difficult, the upgrading difficulty is high, and the price is high;

(2) the digital signal processing system adopts fixed-point operation, and the finite word length effect cannot be ignored;

(3) FPGA resources are limited, and the performance and the efficiency are difficult to balance.

Disclosure of Invention

In order to solve the problems, the invention provides the interference measurement multi-phase channelized baseband conversion system based on the GPU, which has higher reliability, better expansibility and higher precision and can meet the requirements of deep space measurement and control interference measurement on the baseband conversion system.

An interferometric multi-phase channelized baseband conversion system based on a GPU (graphics processing unit), comprises an acquisition module, an interface module, a control module and an operation module, wherein the control module comprises a CPU (central processing unit) control unit and a data cache unit, and the operation module comprises more than two GPUs;

the acquisition module is used for converting an interference measurement analog signal acquired from the outside into an interference measurement digital signal and then sending the interference measurement digital signal to the data cache unit through the interface module;

the CPU control unit is used for partitioning the interferometry digital signals stored in the data cache unit, and then distributing the obtained data blocks to different GPUs in a multi-thread parallel mode, wherein each GPU can be partitioned into at least one data block;

each GPU is used for carrying out multi-phase shunting on the received data block, then filtering each path of obtained data in a multi-stream parallel mode, and then carrying out thread synchronization on each path of filtered data, wherein the quantity of each path of data is expressed by M; forming a two-dimensional matrix by the M paths of synchronized data, and then performing M-point FFT (fast Fourier transform) according to columns to obtain M subchannels with equal bandwidths, wherein the bandwidths between adjacent subchannels are overlapped;

each GPU is also used for selecting a required sub-channel from the M sub-channels according to the center frequency of a signal required by a user, wherein when the center frequency of the signal required by the user is positioned at a non-overlapping part of a certain sub-channel, the sub-channel is selected as the required sub-channel; when the center frequency of the signal required by the user is positioned in the first half section of the overlapping part of some two sub-channels, selecting the sub-channel arranged in the front as the required sub-channel; when the center frequency of the signal required by the user is positioned in the second half of the overlapping part of the two sub-channels, selecting the sub-channel arranged at the back as the required sub-channel;

each GPU is further configured to sequentially perform down-conversion, filtering, and subband extraction on each required subchannel in a multi-stream parallel manner, so as to obtain a baseband signal of a center frequency and a bandwidth required by a user, and implement baseband conversion of an interferometric analog signal.

Further, when each GPU selects a required sub-channel from M sub-channels according to the center frequency of a signal required by a user, if the required sub-channel is a sub-channel arranged at the rear M/2, directly and sequentially performing down-conversion, filtering and sub-band extraction on data in the required sub-channel in a multi-stream parallel mode; and if the required subchannel is the subchannel arranged in the front M/2, turning the data in the required subchannel and then sequentially performing down-conversion, filtering and subband extraction in a multi-stream parallel mode.

Furthermore, each GPU is further configured to perform format conversion on the baseband signal according to a requirement of an external interferometric data interface, and then output the baseband signal to the outside through the interface module.

Further, the interface module includes an ethernet card and a PCIE bus;

the PCIE bus is configured to receive the baseband signal after format conversion, and then output the baseband signal to the outside via the ethernet card.

Further, the interface module includes an ethernet card and a PCIE bus;

the Ethernet card is used for receiving the interference measurement digital signals sent by the acquisition module through the Ethernet and then forwarding the interference measurement digital signals to the data cache unit through the PCIE bus.

Further, the acquisition module converts the interferometric analog signal to an interferometric digital signal at a sampling rate of 1024 Msps.

Has the advantages that:

Drawings

FIG. 1 is a schematic block diagram of a GPU-based interferometric multi-phase channelized baseband conversion system according to the present invention;

FIG. 2 is a block diagram of a parallelization implementation process of the multi-phase channelized baseband conversion algorithm based on the GPU according to the present invention;

FIG. 3 is a schematic diagram of a classical architecture for polyphase shunt channelization provided by the present invention;

FIG. 4 is a schematic diagram of selecting any channel of a local spectrum in an overlap mode according to the present invention;

FIG. 5 is a schematic diagram of the selection of a full-spectrum arbitrary channel in the overlay mode according to the present invention;

FIG. 6 is a schematic diagram of a channel division structure according to the present invention;

FIG. 7 is a schematic diagram of channel division in a channel overlap mode according to the present invention;

fig. 8 is a schematic diagram of an apparatus composition and a connection relationship of an antenna array test system provided in the present invention;

FIG. 9 is a schematic diagram of the PCAL signal spectrum provided by the present invention;

FIG. 10 is a schematic illustration of a PCAL signal local spectrum provided by the present invention;

FIG. 11 is a schematic diagram illustrating the amplitude-frequency characteristics of the CuDBE output signal provided by the present invention;

FIG. 12 is a schematic diagram illustrating the amplitude-frequency characteristic of the CuDBE output signal according to the present invention;

fig. 13 is a schematic diagram of the PCAL phase frequency curve provided by the present invention.

Detailed Description

In order to make the technical solutions better understood by those skilled in the art, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application.

As shown in fig. 1, an interferometric multi-phase channelized baseband conversion system based on a GPU includes an acquisition module, an interface module, a control module and an operation module, wherein the control module includes a CPU control unit and a data cache unit, and the operation module includes more than two GPUs.

The acquisition module is used for converting an interference measurement analog signal acquired from the outside into an interference measurement digital signal and then sending the interference measurement digital signal to the data cache unit through the interface module.

In the acquisition module, the interferometric analog signal is subjected to amplitude adjustment and then sent to an analog-to-digital conversion (ADC) module for sampling, and the sampling rate is single-channel 1024 Msps; the acquisition module is externally connected with a 10MHz reference clock and a second pulse signal; the sampled interference measurement digital signals are sent to a gigabit network card of a high-speed interface module through a gigabit network, after the high-speed interface module receives data, the acquired high-speed interference measurement digital signals are input to a temporary cyclic data cache unit of a control module through a 16x PCIE bus, and the cyclic cache of the data is started under the scheduling of a CPU control unit.

The CPU control unit is used for partitioning the interferometry digital signals stored in the data cache unit, and then distributing the obtained data blocks to different GPUs in a multi-thread parallel mode, wherein each GPU can be partitioned into at least one data block.

Based on a GPU platform, data and a flow are decoupled to the maximum extent by data multithreading block scheduling and digital local oscillator multiphase shunting, and the parallelism of a system is improved; then, a multi-level stream processing (stream) mode is utilized, space complexity is exchanged for data processing time complexity, and parallelization of processing flows is achieved, wherein signal processing flows in each GPU are shown in fig. 2, specifically, each GPU is used for performing multiphase shunting on each received data block, then filtering each obtained data in a multi-stream parallel mode, and performing thread synchronization on each filtered data, wherein the number of each data is represented by M; and forming a two-dimensional matrix by the M paths of synchronized data, and then performing M-point FFT (fast Fourier transform) according to columns to obtain M subchannels with equal bandwidths, wherein the bandwidths between adjacent subchannels are overlapped, and the overlapping range is not less than the maximum subband bandwidth of a GPU (graphics processing Unit) capable of outputting undistorted signals.

That is, after the polyphase split channelization process, the channel is divided into bandwidths of₂pi/M, 2 pi/M uniform sub-channels with frequency interval of pi/M, and the passbands of the sub-channels are overlapped by pi/2M.

It should be noted that, the polyphase splitting means that the data sequence is split into parallel multipaths according to the original filter polyphase decomposition method in the polynomial filter bank. The specific implementation mode is that if the input signal is x (n) and the decimation multiple is D, the output signal of the i (i ═ 0,1, 2.., D-1) channel of the multi-phase shunt is x (nD + i); as shown in fig. 3, which is a classical structure of the multi-phase shunt; second, multi-stream parallelism, where stream (stream) is a parallel concept in the CUDA environment, can be used in the GPU to handle concurrent operations. A stream is essentially a queue of operations that can be issued by different threads on the host and executed in sequence on the GPU in an asynchronous manner. Different streams can be executed concurrently, and a plurality of streams can execute operations such as frequency mixing, filtering and the like of different paths of data concurrently, so that the parallelism of the operations can be greatly improved.

Each GPU is also used for selecting a required subchannel from the M subchannels according to the center frequency of a signal required by a user, wherein when the center frequency of the signal required by the user is positioned in a non-overlapping part of a certain subchannel, the subchannel is selected as the required subchannel, when the center frequency of the signal required by the user is positioned in the first half of the overlapping part of the certain two subchannels, the subchannel arranged in the front is selected as the required subchannel, when the center frequency of the signal required by the user is positioned in the second half of the overlapping part of the certain two subchannels, the subchannel arranged in the back is selected as the required subchannel, and meanwhile, the bandwidth of the signal required by the user does not exceed the maximum subband bandwidth.

That is, after obtaining the channelized output of M sub-channels, it is necessary to select a corresponding sub-channel for further processing according to the position of the center frequency point of the signal required by the user and the required output bandwidth, and due to the existence of channel overlap, it is necessary to select a proper channel according to the position of the frequency point of the signal required by the user and the bandwidth; taking a 1024MHz sampling rate, where M is 16 as an example, channel selection of a local spectrum in an overlapping mode is performed within a range of 0-144 MHz; referring to fig. 4, the dashed lines represent the 0 th, 1 st, and 2 nd channel filter responses, the solid lines represent the 15 th and 14 th sub-channel filter amplitude-frequency responses, and the shaded portions represent the overlapping portions of the passbands of the two adjacent sub-channels. Taking the sub-channel 15 as an example, when the center frequency of the signal is within the 16MHz bandwidth range of 40-56 MHz at the center of the passband, selecting the sub-channel 15 can ensure that all the broadband signals with the center frequency of 40-56 MHz and the bandwidth of not more than 16MHz can be received without distortion. When the signal is in the transition band, without loss of generality, 24-40 MHz is selected as an example, and since the bandwidth of the undistorted signal that can be output by the baseband conversion system is defined as the maximum sub-band bandwidth, in the present invention, the maximum sub-band bandwidth is designed to be 16MHz, when the center frequency of the signal required by the user is within the range of 32-40 MHz, the requirement can be met by selecting the sub-channel 15, and when the signal is within the range of 24-32 MHz, the sub-channel 0 should be selected for subsequent processing. In summary, when the signal is at 32-64 MHz, the subchannel 15 is selected for subsequent processing, and when the signal is at 0-32 MHz, the subchannel 0 is selected, so that the entire bandwidth of 0-512 MHz can be divided into channels as shown in FIG. 5.

FIG. 5 is a schematic diagram of the selection of the full-spectrum arbitrary channel in the overlap mode; as can be seen from fig. 5, the whole channel selection is divided into three levels, the first layer is a mirror spectrum channel layer below the coordinate axis, the layer contains 8 channels from 0 to 7 channels, each channel of the layer is a mirror image of each channel within the range of 0 to 512MHz, and a real non-inversion signal can be obtained only by conjugating a signal in the channel during processing. And the second layer is a positive spectrum channel layer which is positioned below the mirror spectrum, all channels are positive spectrums of all channels within the bandwidth range of 0-512 MHz, and when signals are positioned in the channels of the layer, the signals of the channels can be directly taken for further processing. The third layer is an overlapped channel dividing layer, the layer considers the overlapping of all channels, and finally, the whole sampling bandwidth is evenly divided into 16 central frequency point distribution bandwidths with the bandwidth of 32MHz, when the central frequency point of a signal required by a user is given, the channel on which the frequency point is positioned can be directly judged, and then the channel is selected for processing.

Further, the positive and negative spectra formed by the channel dividing structure of the present invention are shown in fig. 6; under the channel structure, the central frequency expression of each channel is as follows:

on the basis of a classical channel division mode, the frequency spectrum is shifted to the left by pi/2M, then the

sub-channels

0,1, … and 7 take mirror image frequency spectrums, the sub-channels 15, 14, … and 8 take positive frequency spectrums, the size of channel overlapping bandwidth is guaranteed to be not less than the maximum sub-band bandwidth, and therefore all received signal bandwidths can be guaranteed to be within the range of the pass band of the filter and cannot fall into a transition band, and the maximum sub-band bandwidth of the GPU is determined by the hardware property of the GPU.

The amplitude-frequency correspondence of the channel division filter bank is shown in FIG. 7, and a solid line and a dashed line frame above a coordinate axis represent the passband range of the filter bank, wherein the solid line represents a positive spectrum signal within the range of 0-512 MHz, and the dashed line represents a mirror image projection signal of a negative half-axis spectrum within a positive half-axis range. The total bandwidth of all the filters is 64MHz, the passband is 48MHz, the two sides of the transition band are respectively 8MHz, and the passbands of every two filters are overlapped by 16MHz, so that all signals to be received can be ensured to be in the same filter range within any receiving bandwidth range which is not more than 16MHz in the whole analog receiving band, thereby eliminating the influence of the transition band of the filters on broadband signals and ensuring the quality of the received signals.

That is, when the required subchannel is a subchannel arranged at the rear M/2, the data in the required subchannel is directly subjected to down-conversion, filtering and subband extraction in sequence in a multi-stream parallel manner, so that a baseband signal with the central frequency and the bandwidth required by a user is obtained, and the baseband conversion of the interferometric analog signal is realized; when the required subchannel is a subchannel arranged in the front M/2, the data in the required subchannel is overturned, and then down-conversion, filtering and subband extraction are sequentially carried out in a multi-stream parallel mode, so that the data rate of each channel is greatly reduced, a baseband signal of the center frequency and the bandwidth required by a user is obtained, and the baseband conversion of the interferometric analog signal is realized.

Therefore, the interference measurement multi-phase channelized baseband conversion system based on the GPU is composed of four parts, namely an acquisition module, a control module, an operation module and an interface module. The acquisition module is composed of parts such as a high-speed ADC, a 10MHz frequency scale and 1PPS, and is mainly used for completing high-speed digital sampling of analog intermediate-frequency signals and assisting the amplitude regulation function of the signals. The interface module mainly comprises a high-speed Ethernet card and a data transmission network, and mainly completes the functions of data transmission and storage. The control module mainly comprises a CPU control unit and a data cache unit, and mainly completes the management, scheduling, data distribution and simple data processing functions of each module. The operation module is an operation core of the system and mainly completes channelization and sub-band segmentation operations of a large amount of data, the operation module is composed of a plurality of GPU (graphics processing unit) computing cards, the GPUs are connected with the control module and the interface module through a high-speed PCIE (peripheral component interface express) bus to complete data interaction and issue system scheduling instructions, the GPUs are connected with one another through a high-speed Nvlink bus to complete interaction of operation intermediate data, and a main parallel algorithm of the system runs on the operation module.

The performance of the baseband conversion system provided by the invention is verified by using the measured data. As shown in fig. 8, which is a schematic diagram of the equipment composition and connection relationship of the test system, the system uses an analog signal source to generate a PCAL (Phase Calibration) signal, the equipment of the present invention performs baseband conversion on the signal and records data to a magnetic disk, and finally, Matlab analysis equipment is used to output the data; the PCAL signal interval is 10KHz, the frequency range is 247-251MHz, and the input test signal is shown in FIG. 9; the sampling rate of the baseband converter is 1024MHz, the center frequency of the output sub-band is 249MHz, and the bandwidth is 2HMz, as shown in fig. 10. The-3 dB bandwidth efficiency requirement of the deep space measurement and control network interference measurement baseband converter in China is better than 90%, the attenuation of the stop band is more than or equal to 40dB, and the in-band fluctuation is less than or equal to +/-0.5 dB. The amplitude-frequency characteristic of the CuDBE output signal, the local amplification of the amplitude-frequency characteristic of the CuDBE output signal and a PCAL phase-frequency curve are respectively shown in figures 11-13, so that the effective utilization rate of a sub-band is greater than 0.93 × 2/2 × 100%, the fluctuation of the in-band amplitude is superior to 0.2dB, and the system group delay is 77.597us through phase-frequency curve analysis.

That is to say, the invention provides an interferometric multi-phase channelized baseband conversion system based on a GPU, which takes the GPU as a data processing device of a core and uses a CPU to complete the distribution and the scheduling of tasks; the system can effectively reduce the development difficulty of the baseband converter; the floating point operation can effectively reduce the influence of the finite word length effect and improve the frequency mixing and filtering operation precision; commercial devices and software operation can effectively improve the adaptability of the system to various different interface interference measurement systems, the number of GPUs can be flexibly configured according to performance requirements, the defects of the prior art are overcome, and the method has the advantages of flexibility in reconstruction, high operation precision, good expandability and the like.

The present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof, and it will be understood by those skilled in the art that various changes and modifications may be made herein without departing from the spirit and scope of the invention as defined in the appended claims.

Claims

1. An interferometric multi-phase channelized baseband conversion system based on a GPU is characterized by comprising an acquisition module, an interface module, a control module and an operation module, wherein the control module comprises a CPU control unit and a data cache unit, and the operation module comprises more than two GPUs;

the CPU control unit is used for carrying out multi-phase blocking on the interference measurement digital signals stored in the data cache unit and then distributing the obtained data blocks to different GPUs in a multi-stream parallel mode, wherein each GPU can be divided into at least one data block;

each GPU is further configured to sequentially perform down-conversion, filtering, and sub-band extraction on each required sub-channel in a multi-stream parallel manner, so as to obtain a baseband signal of a center frequency and a required bandwidth required by a user, and implement baseband conversion of an interferometric analog signal.

2. The system according to claim 1, wherein when each GPU selects a desired subchannel among M subchannels according to a center frequency of a signal desired by a user, if the desired subchannel is a subsequent M/2 subchannel, data in the desired subchannel is directly subjected to down-conversion, filtering, and subband decimation in sequence in a multi-stream parallel manner; and if the required subchannel is the subchannel arranged in the front M/2, turning the data in the required subchannel and then sequentially performing down-conversion, filtering and subband extraction in a multi-stream parallel mode.

3. The system of claim 1, wherein each GPU is further configured to perform format conversion on the baseband signal according to a requirement of an external interferometry data interface, and then output the baseband signal to the outside through the interface module.

4. The system of claim 3, wherein the interface module comprises an Ethernet card and a PCIE bus;

5. The system of claim 1, wherein the interface module comprises an ethernet card and a PCIE bus;

6. The GPU-based interferometric polyphase channelized baseband conversion system of claim 1 in which said acquisition module converts interferometric analog signals to interferometric digital signals at a sample rate of 1024 Msps.