CN114978200A - High-throughput large-bandwidth general channelized GPU algorithm - Google Patents
High-throughput large-bandwidth general channelized GPU algorithm Download PDFInfo
- Publication number
- CN114978200A CN114978200A CN202210894341.2A CN202210894341A CN114978200A CN 114978200 A CN114978200 A CN 114978200A CN 202210894341 A CN202210894341 A CN 202210894341A CN 114978200 A CN114978200 A CN 114978200A
- Authority
- CN
- China
- Prior art keywords
- signal
- module
- calculation
- processing
- channelized
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B1/00—Details of transmission systems, not covered by a single one of groups H04B3/00 - H04B13/00; Details of transmission systems not characterised by the medium used for transmission
- H04B1/0003—Software-defined radio [SDR] systems, i.e. systems wherein components typically implemented in hardware, e.g. filters or modulators/demodulators, are implented using software, e.g. by involving an AD or DA conversion stage such that at least part of the signal processing is performed in the digital domain
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/20—Processor architectures; Processor configuration, e.g. pipelining
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B1/00—Details of transmission systems, not covered by a single one of groups H04B3/00 - H04B13/00; Details of transmission systems not characterised by the medium used for transmission
- H04B1/06—Receivers
- H04B1/16—Circuits
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B1/00—Details of transmission systems, not covered by a single one of groups H04B3/00 - H04B13/00; Details of transmission systems not characterised by the medium used for transmission
- H04B1/06—Receivers
- H04B1/16—Circuits
- H04B1/18—Input circuits, e.g. for coupling to an antenna or a transmission line
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The invention discloses a general channelized GPU algorithm with high throughput and large bandwidth, which is based on a CUDA computing platform and comprises the following steps: s1: a calculation initialization module is used for applying for storage space of input, output and intermediate variables required in the channelization process in advance; s2: setting related parameters of the target narrowband signal by using a parameter configuration module and configuring the related parameters and the storage resources allocated in the calculation initialization module according to the related parameters; s3: the core function module is used to execute the actual channelization calculation, and according to step S2, corresponding calculation resources are allocated to each path of narrowband signal, and the execution processing of each sub-function module in the channelization flow is started. The invention can keep higher calculation performance and meet the real-time processing index of the relevant application scene. The invention can be deployed and configured on a multi-generation GPU hardware platform supporting CUDA, and the algorithm performance can be self-adaptively matched with the computing capability of GPU hardware, thereby supporting flexible configuration of channelized scenes.
Description
Technical Field
The invention relates to the technical field of communication, in particular to a general channelized GPU algorithm with high throughput and large bandwidth.
Background
The channelization technology is one of the key technologies in a digital wideband receiver, and is used for extracting a single or multiple mutually independent target narrowband signals contained in a digital wideband sampling signal, wherein the processing procedures relate to digital down-conversion (DDC), filtering (Filter), Sample Rate Conversion (SRC), and the like.
With the continuous progress of modern communication technology, the input bandwidth of digital broadband sampling signals in related channelization processing services is larger and larger, and target narrowband signals to be processed are more and more, so that resources consumed by an early channelization platform scheme realized based on a dedicated hardware module are more and more unacceptable. Moreover, the flexible and variable requirements for functional design in modern software radio promote the channelization processing to be realized by using a program algorithm, thereby not only reducing the complexity of system design, but also enabling the later-stage functional adjustment to be more flexible and convenient.
The demand of large-scale scientific research and engineering application on computing performance is continuously increased, and computer technology develops to present 'many-core' systems through 'multi-core' systems in the past decade, and particularly heterogeneous parallel acceleration platforms represented by a CPU (Central processing Unit) and a GPU (graphics processing Unit) are more and more widely applied.
The GPU has hundreds of processing cores, the memory access bandwidth is far higher than that of the CPU, the theoretical peak floating point computing performance of the GPU is improved by orders of magnitude compared with that of the CPU, with the emergence of various heterogeneous programming languages, the programming technology in the aspect of universal parallel computing on the GPU is gradually popularized to the public of developers, the GPU is gradually evolved from a special image acceleration processor into a general computing accelerator nowadays, a unique system architecture of the GPU is specially optimized for data large-scale parallel computing scenes similar to SIMD (single instruction multiple data) and SPMD (distributed computer program) styles, particularly a CUDA (compute unified device) computing platform and a programming model which are introduced by NVIDIA (network virtualization information systems) companies bring a wind and rain-fast computing revolution to the computer industry, and the GPU promotes the accelerated development and ecological prosperity of a plurality of industries such as HPC, AI (analog to digital assistant), automatic driving, finance, medical treatment and the like.
Modern radio communication and digital signal processing are moving towards the directions of high, fine and sharp, and the computational bottleneck faced by the front end of the digital broadband receiver is more and more obvious, so that the development of the channelized GPU algorithm has extremely important practical significance.
The prior art has the following defects:
although the channelization solution realized in the traditional CPU (single core, multi core) has the advantages of flexibility, easy use and low development cost, the time consumption is serious and the real-time performance is poor due to the computational power limitation of the CPU, and the channelization service under the condition of heavy computational load cannot be satisfied;
although the channelized platform solution realized based on relevant special hardware (such as ASIC and FPGA) can meet the performance requirement of real-time performance, the development cost is high, the period is long, the resource overhead is large, and the harsh condition limit in the actual application scene can not be met;
with the existing communication application processing system having more and more abundant integrated functions and flexible and variable services, the related functional modules need to be frequently adjusted, so that the scheme based on the dedicated hardware mechanism cannot meet the requirement.
Disclosure of Invention
Aiming at the problems, the invention provides a high-throughput large-bandwidth general channelized GPU algorithm which is realized based on a CUDA computing platform and a programming model, and provides auxiliary acceleration processing for load processing of related channelized tasks by carrying a GPU product of NVIDIA company in a system platform.
The invention adopts the following technical scheme:
a high-throughput large-bandwidth general channelized GPU algorithm comprises the following steps:
s1: setting the number of supported parallel maximum channels and the signal length of single processing by using a calculation initialization module, and then applying for a storage space of input, output and intermediate variables required in channelization process processing in advance according to the parameter values, wherein the storage space comprises a host memory area and an equipment memory area;
s2: setting related parameters of a target narrowband signal by using a parameter configuration module and configuring the related parameters and storage resources allocated in a calculation initialization module according to the related parameters, wherein the related parameters comprise center frequency, bandwidth, output sampling rate and target gain, and the configuration processing comprises a batch configuration mode before calculation and a dynamic configuration mode during calculation;
s3: performing actual channelized computation by using a core function module, allocating corresponding computing resources to each path of narrow-band signal according to the step S2, and starting execution processing of each sub-function module in a channelized process, where the computing resources include a thread Grid and a thread Block in a CUDA, the step S3 sub-function module includes a frequency mixing sub-module, a half-band filtering sub-module, a resampling processing sub-module, and a low-pass filtering sub-module, and the specific operation steps are as follows:
s301: transmitting broadband input signal data from a host memory buffer area to a corresponding equipment memory buffer area, wherein the host memory buffer area uses a page locking memory in order to accelerate the data transmission efficiency;
s302: performing heterodyne frequency mixing operation on the signal data of the device memory buffer area in the step 301 according to the frequency mixing factor of each path of signal, wherein in order to improve the performance, the frequency mixing factor corresponding to each path of signal is calculated in advance when the parameter configuration module sets the information of each path of signal;
s303: performing half-band filtering processing by taking the heterodyne mixing calculation intermediate result corresponding to each path of signal in the step S302 as input;
s304: taking the intermediate result of the corresponding half-band filtering processing of each path of signal in the step S303 as input to perform resampling processing;
s305: taking the intermediate result of the resampling processing corresponding to each path of signal in the step S304 as an input to carry out low-pass filtering processing;
s306: the low-pass filtering result corresponding to each path of signal in step S305 is transmitted from the corresponding device memory buffer to the host memory buffer and used as the final channelization result.
Preferably, the storage space of step S1 is organized as: all the input, output and intermediate results of the narrow-band signals are distributed with a large continuous storage space, wherein each path of signal sequentially occupies an interval section with the size of the dataStep, and the corresponding input signal length countIn and output signal length countOut cannot exceed the size of the dataStep when each functional module in the channelized service flow processes, so that data collision and pollution cannot be generated among the narrow-band signals in each path when data are processed in parallel in the algorithm.
Preferably, the calculation initialization module is executed only once according to the set parameters, and the applied related storage resources are multiplexed in the subsequent calling of the core function module.
Preferably, in a scenario where a narrow-band signal parameter is dynamically added in the calculation, the parameter configuration module adopts a lock synchronization mechanism, and only after the parameter configuration module is set and unlocked, the core function module can safely access the dynamically updated signal parameter information.
Preferably, the data of the memory buffer area called by the core function module each time is overwritten, so that the output result needs to be subjected to other subsequent calculation processing or copy storage before the core function module is called next time for processing.
The invention has the beneficial effects that: by utilizing the powerful parallel data processing capacity of the GPU, the high calculation performance can be still kept in the channelized processing situation with high throughput, large bandwidth and more total number of target 'narrow-band' signals, and the real-time processing index of a related application scene is met. In addition, due to the compatibility and the expandability of the CUDA programming model, the method can be deployed and configured on a multi-generation GPU hardware platform supporting CUDA, and the algorithm performance can be adaptively matched with the computing capacity of GPU hardware, so that the flexible configuration of channelized scenes can be supported, namely the GPU hardware with low cost is configured in a simple scene, and the GPU hardware with higher configuration level in a complex scene is configured.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings of the embodiments will be briefly described below, and it is apparent that the drawings in the following description only relate to some embodiments of the present invention and are not limiting on the present invention.
FIG. 1 is a block diagram of the system architecture of the present invention;
FIG. 2 is an organization of the device memory space allocated inside the computing initialization module according to the present invention;
FIG. 3 shows the configuration of the internal signal and corresponding storage resource of the parameter configuration module according to the present invention;
FIG. 4 is a diagram illustrating an organization of computing resources by threads within a core function module according to the present invention;
fig. 5 is a block diagram of the signal processing flow of each path inside the core function module according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the drawings of the embodiments of the present invention. It is to be understood that the embodiments described are only a few embodiments of the present invention, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the described embodiments of the invention without any inventive step, are within the scope of protection of the invention.
Unless otherwise defined, technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this disclosure belongs. The use of the word "comprising" or "comprises", and the like, in this disclosure is intended to mean that the elements or items listed before that word, include the elements or items listed after that word, and their equivalents, without excluding other elements or items. "upper", "lower", "left", "right", and the like are used merely to indicate relative positional relationships, and when the absolute position of the object being described is changed, the relative positional relationships may also be changed accordingly.
The invention is further illustrated with reference to the following figures and examples.
As shown in fig. 1, a high-throughput and large-bandwidth general channelized GPU algorithm is implemented based on a CUDA computing platform and a programming model, and provides auxiliary acceleration processing for load processing of related channelized tasks by mounting a GPU product of NVIDIA corporation in a system platform. The embodiment performs the following steps through two execution platforms of GTX1660 and RTX3060 Ti:
s1: using a calculation initialization module to set the supported parameter values such as the number of parallel maximum channels, the signal length of single processing, etc., and then pre-applying for the storage space of input, output and intermediate variables required in the channelization process according to the parameter values, where the storage space includes a host memory area and a device memory area, and the organization form of the storage space is as shown in fig. 2: all the input, output and intermediate results of the narrow-band signals are distributed with a large continuous storage space, wherein each path of signal sequentially occupies an interval section with the size of the dataStep, and the corresponding input signal length countIn and output signal length countOut cannot exceed the size of the dataStep when each functional module in the channelized service flow processes, so that data collision and pollution cannot be generated among the narrow-band signals in each path when data are processed in parallel in the algorithm.
S2: setting relevant parameters of a target narrowband signal by using a parameter configuration module and configuring the relevant parameters and storage resources allocated in a calculation initialization module according to the relevant parameters, wherein the relevant parameters comprise center frequency, bandwidth, output sampling rate and target gain, and the configuration processing comprises a batch configuration mode before calculation and a dynamic configuration mode during calculation, which are shown in fig. 3:
the batch configuration before calculation refers to that parameter information of all target narrow-band signals is known and set in advance before broadband input signal data is subjected to channelized flow processing, parameters of all target narrow-band signals are configured in advance before a core function module is called to perform DDC flow service processing, and according to the sequence added during batch configuration, all paths of signals correspond to corresponding storage spaces one by one;
the dynamic configuration in the calculation refers to a parameter configuration mode that a broadband input signal data stream is executing channelization process processing, but a narrowband signal needs to be added or deleted and the mode takes effect instantly, and includes dynamic deletion in the calculation and dynamic addition in the calculation, it can be seen from the figure that in the mode, the added signal has a globally unique number, a dynamic mapping relation is established between the added signal and a storage space area identifier, when the added signal is dynamically deleted, the storage area number mapped with the added signal is searched according to the number of the deleted signal and is set to be in an idle state, and when a core function module is called later, the path signal will not be allocated with corresponding thread computing resources. When the deleted signals are more, the module can carry out ascending arrangement management on the idle storage areas according to the identifiers, when the target narrow-band signal is dynamically added at the later stage, the idle storage area identifier at the head of the queue is preferentially selected to establish a mapping relation with the signal, and the idle state mark of the idle storage area identifier is simultaneously cancelled, so that when the core function module is called later, the signal is allocated with corresponding thread computing resources, and the corresponding DDC service flow processing is started.
S3: the core function module is used to execute the actual channelized computation, and according to step S2, corresponding computing resources are allocated to each path of narrowband signal and the execution processing of each sub-function module in the channelized flow is started, where the computing resources include a thread Grid and a thread Block in the CUDA, and as shown in fig. 4, the channelized parallel processing of multiple paths of narrowband signals is organized into a two-dimensional thread Grid form, and in addition, considering processing is a one-dimensional signal, so the thread Block is also selected as a one-dimensional form. All the thread blocks with the same y-direction index (blockidx.y) in the two-dimensional thread grid are responsible for channelizing processing tasks of signals mapped by corresponding storage space region identifiers (non-idle states), and a plurality of thread blocks with different x-direction indexes (blockidx.x) are responsible for processing data points in different signal periods.
As shown in fig. 5, the step S3 sub-function module includes a frequency mixing sub-module, a half-band filtering sub-module, a resampling processing sub-module, and a low-pass filtering sub-module, and includes the following specific operation steps:
s301: the broadband input signal data is transmitted from the host memory buffer to the corresponding device memory buffer, and in order to accelerate the data transmission efficiency, the host memory buffer uses a page lock memory. The page locking memory is distributed on the memory of the host by a CUDA function cudaHostAlloc, and the important attribute of the page locking memory is that the operating system of the host can not perform paging and exchange operation on the memory of the host, so that the memory is ensured to be always resident in the physical memory;
s302: performing heterodyne frequency mixing operation on the signal data in the device memory buffer area in step 301 according to the frequency mixing factor of each path of signal, in order to improve performance, calculating the frequency mixing factor corresponding to each path of signal in advance when the parameter configuration module sets information of each path of signal, and moving the frequency spectrum structure corresponding to each path of target narrow-band signal to a baseband after frequency mixing processing, namely, the center frequency of the signal becomes "0";
s303: performing half-band filtering processing by taking the intermediate result of heterodyne mixing calculation corresponding to each path of signals in the step S302 as input, configuring a plurality of half-band filtering cycles for each path of signals according to the relationship between the output sampling rate of each path of signals and the original input sampling rate, adopting a constant memory in a CUDA programming model for storage because the half-band filtering coefficients are relatively fixed, reducing excessive global memory access operations (long-delay and high-overhead operations in the CUDA model) during kernel function calculation, and in addition, in order to improve the performance, performing segmentation processing by using a linear convolution operation corresponding to the filtering processing according to an x-direction index organization form of a thread block, storing data corresponding to each segment by using a shared memory in the CUDA programming model, similarly avoiding excessive global memory access operations, and filtering out most of interference signal components through half-band filtering processing, the actual sampling rate of the resulting signal is also very close to the target output sampling rate parameter;
s304: performing resampling processing by taking the intermediate result of the corresponding half-band filtering processing of each path of signal in the step S303 as input, performing interpolation/sampling processing on the intermediate result according to a pre-calculated factor in a resampling processing sub-module, and after resampling processing, enabling each path of corresponding output result signal to meet the output sampling rate;
s305: performing low-pass filtering processing by taking the resampling processing intermediate result corresponding to each path of signal in the step S304 as an input, performing convolution operation on the low-pass filtering intermediate result by a low-pass filtering submodule according to an FIR low-pass filter coefficient designed by preset parameters, performing performance optimization processing by adopting a shared memory mechanism, and after the low-pass filtering processing, filtering all other signal frequency spectrum components except for the corresponding target narrow-band signal in each path of output signal result to only contain the complete frequency spectrum structure of the target narrow-band signal;
s306: the low-pass filtering result corresponding to each path of signal in step S305 is transmitted from the corresponding device memory buffer to the host memory buffer and used as the final channelization result, because the target narrowband output sampling rate parameters of each path are different, the length of the final output result of each path is different, if the whole memory interval is directly transmitted from the device to the host as shown in fig. 2, a large amount of invalid data is inevitably transmitted, therefore, in order to improve the data processing result transmission efficiency, additional processing is performed inside the core module, the channelization results of each path are reorganized into a head-to-tail continuous storage form, and then are transmitted back to the host memory buffer, and the master regulator returns the parameters according to the result length, the data head address offset and the like to the corresponding host memory buffer index and accesses the final channelization result corresponding to each path of signal.
The calculation initialization module is executed only once according to the set parameters, and the applied related storage resources are multiplexed in the subsequent calling of the core function module.
In a scene that a certain narrow-band signal parameter is dynamically added in calculation, the parameter configuration module adopts a lock synchronization mechanism, and only after the parameter configuration module is set and unlocked, the core function module can safely access the dynamically updated signal parameter information.
The data of the memory buffer area called by the core function module each time can be overwritten, so that the output result needs to be subjected to other subsequent calculation processing or copy storage before the core function module is called next time for processing.
The execution results of the two execution platforms GTX1660 and RTX3060Ti are shown in the following table:
as can be seen from the above table, under different processing path number scenarios, the performance indexes of the method for executing channelized task load are all ideal, and the performance of the algorithm is enhanced along with the improvement of the computing capability of the GPU hardware platform.
Although the present invention has been described with reference to a preferred embodiment, it should be understood that various changes, substitutions and alterations can be made herein without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (6)
1. A high-throughput large-bandwidth general channelized GPU algorithm is characterized by being based on a CUDA computing platform and comprising the following steps:
s1: the method comprises the steps that a calculation initialization module is used for applying storage spaces of input, output and intermediate variables required in channelization process processing in advance, wherein the storage spaces comprise a host memory area and an equipment memory area;
s2: setting related parameters of a target narrowband signal by using a parameter configuration module and configuring the related parameters and storage resources allocated in a calculation initialization module according to the related parameters, wherein the related parameters comprise center frequency, bandwidth, output sampling rate and target gain, and the configuration processing comprises a batch configuration mode before calculation and a dynamic configuration mode during calculation;
s3: and executing actual channelized calculation by using the core function module, allocating corresponding calculation resources to each path of narrow-band signal according to the step S2, and starting execution processing of each sub-function module in the channelized flow, wherein the calculation resources include a thread Grid and a thread Block in the CUDA.
2. A high throughput large bandwidth general purpose channelized GPU algorithm as claimed in claim 1 wherein said step S1 storage space is organized in the form of: and the input and output of all narrow-band signals and the intermediate result allocate continuous storage space, wherein each path of signal sequentially occupies an interval paragraph with the size of the dataStep, and ensures that the corresponding input signal length countIn and output signal length countOut do not exceed the size of the dataStep when each functional module in the channelized service flow processes.
3. The high-throughput large-bandwidth general-purpose channelized GPU algorithm according to claim 1, characterized in that said step S3 sub-functional modules include a frequency mixing sub-module, a half band filtering sub-module, a resampling processing sub-module and a low pass filtering sub-module, and the specific operation steps are as follows:
s301: transmitting broadband input signal data from a host memory buffer area to a corresponding device memory buffer area, wherein the host memory buffer area uses a page lock memory;
s302: performing heterodyne frequency mixing operation on the signal data of the device memory buffer area in the step 301 according to the frequency mixing factor of each path of signal, wherein the frequency mixing factor corresponding to each path of signal is calculated in advance when the parameter configuration module sets the information of each path of signal;
s303: performing half-band filtering processing by taking the heterodyne mixing calculation intermediate result corresponding to each path of signal in the step S302 as input;
s304: taking the intermediate result of the corresponding half-band filtering processing of each path of signal in the step S303 as input to perform resampling processing;
s305: taking the intermediate result of the resampling processing corresponding to each path of signal in the step S304 as an input to carry out low-pass filtering processing;
s306: the low-pass filtering result corresponding to each path of signal in step S305 is transmitted from the corresponding device memory buffer to the host memory buffer and used as the final channelization result.
4. The high-throughput large-bandwidth general-purpose channelized GPU algorithm according to claim 1, characterized in that said computation initialization module only needs to be executed once according to the set parameters, and the applied related memory resources are multiplexed in the subsequent kernel function module calls.
5. The GPU algorithm of claim 1, wherein the parameter configuration module employs a lock synchronization mechanism in a scenario where a narrow-band signal parameter is dynamically added during computation, and only after the parameter configuration module is configured and unlocked, the core function module can safely access the dynamically updated signal parameter information.
6. The GPU algorithm for generic channelization with high throughput and large bandwidth of claim 1, wherein the core function module performs calculation processing or copy storage on the output result before next call processing.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210894341.2A CN114978200B (en) | 2022-07-28 | 2022-07-28 | High-throughput large-bandwidth general channelized GPU algorithm |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210894341.2A CN114978200B (en) | 2022-07-28 | 2022-07-28 | High-throughput large-bandwidth general channelized GPU algorithm |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114978200A true CN114978200A (en) | 2022-08-30 |
CN114978200B CN114978200B (en) | 2022-10-21 |
Family
ID=82969602
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210894341.2A Active CN114978200B (en) | 2022-07-28 | 2022-07-28 | High-throughput large-bandwidth general channelized GPU algorithm |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114978200B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117648192A (en) * | 2023-12-14 | 2024-03-05 | 南宁湖畔科技有限公司 | Resource allocation method and GPU resource pool scheduling system |
CN118215223A (en) * | 2024-05-22 | 2024-06-18 | 成都派奥科技有限公司 | Solder paste screen printing clamp for complex pattern soft substrate sintering process |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100239052A1 (en) * | 2009-03-23 | 2010-09-23 | Lockheed Martin Corporation | Wideband digital receiver with integrated dynamic narrowband channelization and analysis |
CN109617631A (en) * | 2018-12-28 | 2019-04-12 | 华航高科(北京)技术有限公司 | Reconnaissance system adaptive reception method based on the measurement of digital channelizing instantaneous parameters |
US20190171965A1 (en) * | 2017-12-01 | 2019-06-06 | Deepwave Digital, Inc. | Artificial intelligence radio transceiver |
CN111552559A (en) * | 2020-04-07 | 2020-08-18 | 上海大学 | Broadband signal DDC system design method based on GPU |
CN111683111A (en) * | 2020-04-17 | 2020-09-18 | 中国人民解放军战略支援部队航天工程大学 | Interferometry multi-phase channelization baseband conversion system based on GPU |
CN111682880A (en) * | 2020-04-17 | 2020-09-18 | 中国人民解放军战略支援部队航天工程大学 | GPU-based streaming architecture broadband signal digital down-conversion system |
CN111786688A (en) * | 2020-06-16 | 2020-10-16 | 重庆邮电大学 | Broadband parallel channelization receiving method based on embedded GPU |
CN113660046A (en) * | 2021-08-17 | 2021-11-16 | 东南大学 | Method for accelerating generation of large-scale wireless channel coefficients |
CN114172527A (en) * | 2021-12-13 | 2022-03-11 | 武汉中元通信股份有限公司 | Digital channelized parallel receiving method and device, electronic equipment and storage medium |
-
2022
- 2022-07-28 CN CN202210894341.2A patent/CN114978200B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100239052A1 (en) * | 2009-03-23 | 2010-09-23 | Lockheed Martin Corporation | Wideband digital receiver with integrated dynamic narrowband channelization and analysis |
US20190171965A1 (en) * | 2017-12-01 | 2019-06-06 | Deepwave Digital, Inc. | Artificial intelligence radio transceiver |
CN109617631A (en) * | 2018-12-28 | 2019-04-12 | 华航高科(北京)技术有限公司 | Reconnaissance system adaptive reception method based on the measurement of digital channelizing instantaneous parameters |
CN111552559A (en) * | 2020-04-07 | 2020-08-18 | 上海大学 | Broadband signal DDC system design method based on GPU |
CN111683111A (en) * | 2020-04-17 | 2020-09-18 | 中国人民解放军战略支援部队航天工程大学 | Interferometry multi-phase channelization baseband conversion system based on GPU |
CN111682880A (en) * | 2020-04-17 | 2020-09-18 | 中国人民解放军战略支援部队航天工程大学 | GPU-based streaming architecture broadband signal digital down-conversion system |
CN111786688A (en) * | 2020-06-16 | 2020-10-16 | 重庆邮电大学 | Broadband parallel channelization receiving method based on embedded GPU |
CN113660046A (en) * | 2021-08-17 | 2021-11-16 | 东南大学 | Method for accelerating generation of large-scale wireless channel coefficients |
CN114172527A (en) * | 2021-12-13 | 2022-03-11 | 武汉中元通信股份有限公司 | Digital channelized parallel receiving method and device, electronic equipment and storage medium |
Non-Patent Citations (2)
Title |
---|
ALFONSO BREGLIA等: ""Electromagnetic optimization of MIMO channels using GPUs"", 《 2017 INTERNATIONAL APPLIED COMPUTATIONAL ELECTROMAGNETICS SOCIETY SYMPOSIUM - ITALY (ACES)》 * |
赵欢欢,张润生: ""基于GPU的数字信道化设计"", 《数字技术与应用》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117648192A (en) * | 2023-12-14 | 2024-03-05 | 南宁湖畔科技有限公司 | Resource allocation method and GPU resource pool scheduling system |
CN118215223A (en) * | 2024-05-22 | 2024-06-18 | 成都派奥科技有限公司 | Solder paste screen printing clamp for complex pattern soft substrate sintering process |
Also Published As
Publication number | Publication date |
---|---|
CN114978200B (en) | 2022-10-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN114978200B (en) | High-throughput large-bandwidth general channelized GPU algorithm | |
CN107679621B (en) | Artificial neural network processing device | |
CN107679620B (en) | Artificial neural network processing device | |
US6199093B1 (en) | Processor allocating method/apparatus in multiprocessor system, and medium for storing processor allocating program | |
CN101950282B (en) | Multiprocessor system and synchronous engine thereof | |
US8392920B2 (en) | Parallel query engine with dynamic number of workers | |
CN105183698B (en) | A kind of control processing system and method based on multi-core DSP | |
US20230244537A1 (en) | Efficient gpu resource allocation optimization method and system | |
CN103019838B (en) | Multi-DSP (Digital Signal Processor) platform based distributed type real-time multiple task operating system | |
CN117058288A (en) | Graphics processor, multi-core graphics processing system, electronic device, and apparatus | |
CN107943592B (en) | GPU cluster environment-oriented method for avoiding GPU resource contention | |
CN109947416A (en) | A kind of cross-platform universal parallel of software implementation radar calculates the construction method in library | |
CN115576675A (en) | FPGA (field programmable Gate array) acceleration method for monitoring and controlling network edge cloud computing | |
JPH10111858A (en) | Method and device for making graphics pipeline parallel | |
CN113406572B (en) | Radar parallel processing system and method, storage medium and terminal | |
CN112395056B (en) | Embedded asymmetric real-time system and electric power secondary equipment | |
CN115904510B (en) | Processing method of multi-operand instruction, graphic processor and storage medium | |
CN109840306A (en) | One kind being based on recursive parallel FFT communication optimization method and system | |
Takada et al. | " Linux on ITRON": a hybrid operating system architecture for embedded systems | |
CN110955526A (en) | Method and system for realizing multi-GPU scheduling in distributed heterogeneous environment | |
CN112732634B (en) | ARM-FPGA (advanced RISC machine-field programmable gate array) cooperative local dynamic reconstruction processing method for edge calculation | |
CN113255538B (en) | FPGA-based infrared weak and small target detection tracking device and method | |
CN113923212A (en) | Network data packet processing method and device | |
JPH06236272A (en) | Method and system for enhancement of efficiency of synchronization of superscalar processor system | |
CN108304343A (en) | A kind of chip-on communication method of complexity SOC |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |