Ultra-wideband radio astronomical pulse radio achromatic terminal

Abstract

The utility model discloses an ultra-wideband radio astronomical pulse radio achromatic terminal, which is characterized by comprising an ADC (Analog to Digital Converter) sampling module used for carrying out multi-channel ultra-wideband concurrent sampling on baseband signals after radio frequency mixing; sampled data passes through the PFB multiphase filtering module, the FFT operation module and the frequency spectrum accumulation control module to obtain superposed spectral line data, and the superposed spectral line data is packaged and then sent to the GPU video memory module. The GPU video memory module comprises a shared memory and a global memory; the GPU high-spectral-resolution operation module processes the cached data to obtain ultrahigh-spectral-resolution frequency spectrum data and caches the ultrahigh-spectral-resolution frequency spectrum data into a shared memory; the GPU coherent achromatic operation module carries out alignment multiplication on the spectrum data with the ultrahigh spectral resolution and the CHIRP dispersion function parameter matrix to obtain pulse radio signal spectrum data after correlation achromatic; and the GPU inverse Fourier transform operation module carries out inverse Fourier transform on the pulse radio signal frequency spectrum data after correlation achromatic dispersion to complete coherent achromatic dispersion.

Description

Technical Field

[0001] This utility model belongs to the fields of astronomical observation and ultra-wideband communication technology, specifically relating to an ultra-wideband radio astronomy pulse radio dispersion apochromatic terminal. Background Technology

[0002] In the field of radio astronomy, pulsar observations have contributed to two of the seven astrophysical achievements awarded the Nobel Prize in Physics to date. Pulsars' highly stable pulsed radio emissions not only provide a natural timekeeping tool but also serve as effective probes for cutting-edge astronomical explorations such as interstellar electron density, interstellar magnetic fields, and gravitational waves. In the development of pulsar research, the core issue is improving observational capabilities and methods; finding more and fainter pulsar samples is the most urgent task. When pulsed radio signals reach the observation equipment through the interstellar medium, they undergo dispersion. To obtain the original pulsar signal, the observed pulsed radio signal must be de-dispersioned, which also improves the accuracy of pulsar timing.

[0003] Traditional dedispersion methods primarily employ incoherent dedispersion, which involves dividing the pulsed radio signal into narrowband channels using traditional analog filtering devices based on time series. Then, the time-domain sequence corresponding to each channel is shifted for alignment and dedispersion. While incoherent dedispersion involves less data and computation time, it cannot completely eliminate dispersion, and the resulting signal has lower time resolution. In recent years, with the development and advancement of high-speed data signal processing devices and new data processing technologies, the international mainstream has increasingly adopted coherent dedispersion. This involves performing coherent dedispersion on the pulsed radio signal across the entire passband. In the frequency domain, the inverse propagation function of the interstellar medium is used to achieve dedispersion across the entire observation passband. Coherent dedispersion results in higher time resolution and superior performance. However, due to the large data volume required for coherent dedispersion, especially for ultra-wideband (UWB) applications, it relies more heavily on system-level designs based on high-speed processing chips.

[0004] The principle of pulsar coherent achromatic design addresses the issue that when periodic pulse signals generated by pulsed radio radiation propagate through space, they are affected by the interstellar medium (ISM). This interaction with the ISM causes the received pulse profile to broaden and become blurred, a phenomenon known as pulsar dispersion. The ISM mainly refers to the matter widely distributed in interstellar space, primarily composed of neutral hydrogen, ionized hydrogen, and helium. Because free electrons exist in the ISM, the propagation time of pulsar signals is delayed due to their influence, with the degree of delay varying with frequency. Lower frequencies result in longer delays. Therefore, the dispersion effect of the ISM causes the energy of the same pulse at different frequencies to arrive at the radio telescope at different times, such as... Figure 1As shown, this manifests as pulse width broadening. In the design of a coherent achromatic terminal, the energy transfer rate is known to be the group velocity, and its relationship with frequency is as follows:

[0005]

[0006] in:

[0007] This refers to the plasma frequency in the ISM. e,n e ,m e These represent the electron charge, interstellar electron density, and electron mass, respectively. e Generally between 0.01 and 1 cm -3 The corresponding f p The frequency ranges from 5 to 50 kHz. Pulsar observation frequencies, however, are typically between 100 MHz and 10 GHz. Therefore, the observation frequency f is much greater than f0. p .

[0008] Perform a Taylor expansion on equation 1-1, and ignore the quadratic term (f >> f). p )get:

[0009]

[0010] Therefore, the time required for a pulse signal to travel a distance d is:

[0011]

[0012] in This is called dispersion measure.

[0013] Therefore, the time delay caused by dispersion is:

[0014]

[0015] In the formula t DM Let t be the delay time and f be the signal frequency. Dispersion will broaden the pulse profile of the pulsar, making it blurry. If the delay time t... DM When the pulse period is similar, it is almost impossible to observe the outline of the pulse, that is, it is impossible to accurately reconstruct the pulse radio signal X.

[0016] Coherent dedispersion treats the influence of the interstellar medium on pulsed radio signals as an equivalent filter. This is achieved by performing a Fourier transform on the signal received by the radio antenna, multiplying it by the inverse function of the equivalent filter's transfer function (the chirp function), and then performing an inverse Fourier transform on the result to obtain the original pulsar signal in the time domain. The interstellar medium transfer function can be decomposed into amplitude and frequency responses. The equivalent filter needs to correct both the amplitude and phase of the signal, thus yielding the discrete chirp function:

[0017]

[0018] Where A k and H k These are the amplitude correction for the pulse radio signal at frequency k and the equivalent filter transfer function of the ISM, respectively, where f0 is the observed center frequency (MHz) and D is the dispersion constant. d is the distance of the pulsar from the observer, n e X is the electron density in space.

[0019] With the increasing demand for ultra-wideband (UWB) and ultra-high resolution pulsar observations, the parallel computation of massive FFTs has become a new bottleneck between the requirements for coherent dedispersion of pulsed radio signals and practical applications. Compared to the analog filtering devices and CPU coherent dedispersion processing systems used in traditional international radio astronomy, the high cost and complex MPI collaborative processing have limited the further promotion of the algorithm. With the development of parallel algorithms and processing architectures in recent years, emerging platforms such as programmable gate arrays (FPGAs) and graphics processing units (GPUs) have begun to be researched and adopted, especially in the field of parallel processing of large amounts of data for ultra-wideband applications.

[0020] Meanwhile, the mainstream digital terminal equipment for current radio astronomy research is the ROACH2 spectrum analysis terminal. It uses a Xilinx Virtex-6 series FPGA as the main processing unit and a PowerPC 440EPx as the SOC for user interaction and control. Through the combination of multiple FPGA functional modules, it completes radio astronomy spectrum analysis and observation functions.

[0021] The ROACH2 spectrum analysis terminal for radio astronomy mainly consists of six modules, each with the following specific functions:

[0022] 1. ADC sampling module, which uses a 1Gsps@8bit ADC chip to sample RF signals with a bandwidth of 0-1GHz;

[0023] 2. The FFT operation module performs spectral channelization processing on the sampled data of the ADC through FFT operation. By controlling the channelization spectral operation processing of the 2-8K frequency points, it meets the frequency resolution control requirements of radio astronomy observations.

[0024] 3. The spectrum accumulation control module achieves the time resolution control requirements of radio astronomy observations by performing spectrum accumulation (integration) operations;

[0025] 4. Bit truncation module: performs 8-bit truncation operation on the 32-bit single-precision floating-point number after FFT operation to meet the 8-bit input and 8-bit output requirements of the spectrum analysis terminal, and can also adjust the dynamic range value of the output spectrum.

[0026] 5. A 10 Gigabit transmission module, which encapsulates and transmits UDP or TCP / IP data after processing the spectrum, and can control and set the transmission packet header file, destination IP address and MAC value, etc.

[0027] 6. The receiving and storage module unpacks and transfers the transmitted data, directly storing it to the target computer's hard drive via the RDMA mode of the 10 Gigabit network card, thus completing the data storage function for spectrum analysis.

[0028] In response to the development of next-generation ultra-wideband radio astronomy receiver technology, the existing ROACH2 radio astronomy digital terminal system, due to the capacity limitations of its hardware ADC and FPGA chips, can no longer meet the key technical requirements of observation and application, such as 1GHz, 2K frequency, and 10GbE transmission. It cannot perform coherent dedispersion processing on ultra-wideband pulsed radio signals emitted by pulsars, cannot meet the high spectral resolution observation requirements of 16K-point FFT operations, and cannot meet the high-speed data transmission requirements of 100Gbps. This invention comprehensively solves the requirement of coherent dedispersion processing of ultra-wideband pulsed radio signals through a complete programmable gate array (FPGA) and graphics processing unit (GPU) design scheme incorporating two key innovations. Utility Model Content

[0029] To address the problems existing in the prior art, this utility model provides a coherent achromatic terminal for ultra-wideband radio astronomy pulsars, realizing a coherent achromatic terminal for ultra-wideband radio astronomy pulsars and pulse radio applications with dual polarization 5Gsps@14bit, 16K~1M frequency point resolution, and 90Gbit / s high-speed transmission rate.

[0030] This application addresses the development needs of radio astronomy pulsar observation by developing an ultra-wideband radio astronomy pulsar coherent achromatic terminal system based on FPGA and GPU platforms. It completes a set of ultra-wideband radio astronomy pulsar coherent achromatic terminal systems based on PCIe high-speed acquisition cards and GPU platforms that meets design parameters. Its main innovations are:

[0031] 1) For the first time, the design specifications of 5GHz ultra-wideband fully cover the needs of radio astronomy pulsar and pulse radio observation. It realizes the design of multi-ADC channel ultra-wideband concurrent module based on time-staggered multi-ADC channel. Through the timing optimization and concurrent design of four high-speed sampling ADCs of 2.5Gsps / 14bit, it fully covers the ultra-wideband specifications of 5GHz (10Gsps).

[0032] 2) Ultra-wideband coherent achromatic processing of radio astronomy pulsars is achieved through GPU-CUDA architecture. In the core design of the achromatic chirp function matrix multiplication module, the sub-module design of memory merging and block multiplication is optimized, which improves efficiency by about 10 times compared with the traditional CPU processing technology.

[0033] 2) Using a PCIe high-speed acquisition card based on the PCIe 3.0x16 architecture, high-speed data transmission from FPGA to GPU is achieved, realizing a high-speed bus transmission rate of up to 90Gb / s;

[0034] 3) The FPGA core implements the primary channelization design of PFB and 16K-point FFT. By introducing a multiphase filter bank design, high-performance spectrum processing and flexible parameter settings are achieved.

[0035] 4) The GPU-CUDA architecture enables ultra-high spectral resolution computation of secondary ~1M point channels of data, which greatly improves the spectral resolution of coherent achromatic data.

[0036] The technical solution of this application is as follows:

[0037] An ultra-wideband radio astronomy pulse radio dispersion apochromatic terminal, characterized in that it includes:

[0038] The ADC sampling module 1 is connected to the PFB multiphase filter module 2 and is used to perform multi-channel ultra-wideband concurrent sampling on the baseband signal after RF mixing and send the sampled data to the PFB multiphase filter module 2.

[0039] The PFB multiphase filtering module 2 is connected to the FFT operation module 3 and is used to perform multiphase filtering on the input sampled data to obtain channelized time-domain data and transmit it to the FFT operation module 3.

[0040] The FFT operation module 3 is connected to the spectrum accumulation control module 4 and is used to perform fast Fourier transform spectrum processing on the input channelized time domain data to obtain primary spectrum data.

[0041] The spectrum accumulation control module 4 is connected to the PCIe data encapsulation module 6 and is used to perform accumulation operations on the primary spectrum data to obtain superimposed spectral line data and transmit it to the PCIe data encapsulation module 6.

[0042] The clock signal control module 5 is connected to the ADC sampling module 1, the PFB polyphase filter module 2, the FFT operation module 3, and the spectrum accumulation control module 4 respectively. It is used to provide clock control signals to the ADC sampling module 1, the PFB polyphase filter module 2, the FFT operation module 3, and the spectrum accumulation control module 4, and to coordinate the transmission rate of the ADC sampling module 1 with the processing timing clock of the PFB polyphase filter module 2, the FFT operation module 3, and the spectrum accumulation control module 4.

[0043] The PCIe data encapsulation module 6 is connected to the GPU memory module 7 and is used to encapsulate the input superimposed spectral data and send it to the GPU memory module 7.

[0044] GPU memory module 7 is used to cache input packaged data; GPU memory module 7 includes shared memory and global memory;

[0045] The GPU high-spectral-resolution computing module 8 is connected to the GPU memory module 7 and is used to perform secondary spectral channelization subdivision processing on the cached data stream in the GPU memory module 7. Then, the result of the secondary spectral channelization subdivision processing is subjected to FFT operation to obtain the ultra-high spectral-resolution spectrum data matrix of radio astronomy observation and cache it in the shared memory.

[0046] The GPU coherent dedispersion operation module 9 is connected to the GPU memory module 7 and is used to perform aligned multiplication of the ultra-high spectral resolution spectrum data matrix of radio astronomy observations and the CHIRP dispersion function parameter matrix to obtain the spectrum data of pulsed radio signals after relevant dedispersion. The CHIRP dispersion function parameter matrix is ​​calculated based on the known path loss of strong sources in radio astronomy observations and the pulsar dispersion calculation formula and stored in the global memory.

[0047] The GPU inverse Fourier transform operation module 10 is connected to the GPU memory module 7 and is used to perform inverse Fourier transform on the spectrum data of the pulse radio signal after correlation dedispersion to complete coherent dedispersion processing.

[0048] The GPU data folding module 11 is connected to the GPU memory module 7 and the GPU inverse Fourier transform operation module 10, respectively. It is used to perform radio astronomy standard periodization folding operation on the pulsar profile of the relevant dedispersion pulse radio signal spectrum data and send it back to the GPU memory module 7 for caching.

[0049] The radio astronomy pulsar data standardization format generation and storage module 12 is connected to the GPU memory module 7 and is used to convert and store the coherent dedispersion data cached in the GPU memory module 7 in the PSRFITS standard format.

[0050] Furthermore, the GPU coherent dedispersion operation module 9 includes a multi-threaded continuous memory access submodule and a block matrix multiplication submodule. The multi-threaded continuous memory access submodule utilizes multiple threads to read a column of data from the CHIRP dispersion function parameter matrix from the global memory within the same clock cycle, with different threads reading different columns of data from the CHIRP dispersion function parameter matrix. Then, each column of data is stored in a memory block of shared memory, with different columns of data stored in different memory blocks of shared memory. The block matrix multiplication submodule divides the ultra-high spectral resolution spectral data matrix of radio astronomy observations into multiple spectral data blocks according to the number of columns in the CHIRP dispersion function parameter matrix. Each thread multiplies each spectral data block with the column data stored in the corresponding memory block in the shared memory.

[0051] Furthermore, the block matrix multiplication submodule divides and completes the obtained spectrum data blocks and CHIRP dispersion function parameter matrix blocks with a power of 2 value of 16 bytes.

[0052] Furthermore, the ADC sampling module 1 also includes four sub-modules to improve sampling performance and stability. Among them, the ADC clock allocation sub-module is used to maintain the sampling time interval between each channel; the gain matching sub-module is used to monitor and adjust the gain difference of each channel to achieve stable output signal amplitude and improve signal-to-noise ratio; the offset mismatch adjustment sub-module is used to adjust and align the zeros of each ADC unit; and the timing deviation alignment sub-module is used to monitor the sampling time of each ADC unit and adjust and align the sampling time of the ADC unit when the sampling time of an ADC unit deviates from the sampling clock deviation amount set for that ADC unit.

[0053] Furthermore, the ADC sampling module 1 includes multiple channels, each channel is equipped with an ADC unit, and the sampling times of each ADC unit are arranged in a time-staggered manner.

[0054] Furthermore, the ADC sampling module 1 includes four channels, and the ADC sampling module 1 performs four-channel 2.5GSPS@14bit ultra-wideband concurrent sampling on the baseband signal after RF mixing.

[0055] Furthermore, the radio astronomy pulsar data standardization format generation and storage module 12 performs PSRFITS standard format conversion on the coherent achromatic data cached in the GPU memory module 7 through a RAID disk array and then stores it in parallel.

[0056] Furthermore, the PCIe data encapsulation module 6 performs high-speed PCIe bus memory read and write operations on the encapsulated data in DMA mode, and directly stores it into the memory of the GPU memory module 7.

[0057] Furthermore, by using a memory-shared parallel computing framework to call the cufftExecC2C API algorithm library, a 1M-point FFT operation is performed on the secondary spectral channelization subdivision results to obtain an ultra-high spectral resolution spectral data matrix for radio astronomy observations.

[0058] The advantages of this patent are as follows:

[0059] In response to the needs of radio astronomy pulsar observation and the development of next-generation radio astronomy ultra-wideband receiver technology, the key technical indicators of the existing radio astronomy digital terminal ROACH2 system, such as 1GHz, 2K frequency point and 10GbE transmission, can no longer meet the observation and application requirements. It cannot realize the application of ultra-wideband pulsar coherent dedispersion, cannot meet the high spectral resolution observation of 16K-point FFT operation, and cannot meet the high-speed data transmission requirements of 100Gbps. Therefore, it is urgent to carry out corresponding research, design and development work.

[0060] Based on this requirement and the inherent contradictions, we have developed a complete ultra-wideband radio astronomy pulsar coherent achromatic terminal, achieving a bandwidth of 5–8 GHz, 8–14 bit accuracy, a primary frequency of 32 kHz, high spectral resolution up to 1 MHz at the GPU end, and a high-speed transmission rate of 90 Gbit / s. Through the design of the latest generation FPGA and GPU cores, we have achieved the development of a new generation of high-performance digital terminals that meet the requirements of wide bandwidth, high spectral resolution, large dynamic range, and long timescale stability. Attached Figure Description

[0061] Figure 1 This is a pulse width expansion plot caused by pulsar dispersion.

[0062] Figure 2 Design diagram of a four-channel time-interleaved ADC for ultra-wideband applications.

[0063] Figure 3 This is a diagram of the coherent achromatic terminal structure for an ultrawideband radio astronomy pulsar.

[0064] Figure 4 This is a comparison chart of GPU data transfer rates in the CUFFT library.

[0065] Figure 5 The time test graph is for the relevant dedispersion algorithm submodule.

[0066] Figure labels: 1-ADC sampling module, 2-PFB multiphase filtering module, 3-FFT operation module, 4-spectrum accumulation control module, 5-clock signal control module, 6-PCIe data encapsulation module, 7-GPU memory module, 8-GPU high spectral resolution operation module, 9-GPU coherent dedispersion operation module, 10-GPU inverse Fourier transform operation module, 11-GPU data overlap folding module, 12-Radio astronomy pulsar data normalization format (PSRFITS) generation and storage module. Detailed Implementation

[0067] The present invention will now be described in further detail with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0068] The main functions of the ultra-wideband hyperspectral resolution (UHVHD) terminal for radio astronomy are as follows: After receiving the input intermediate frequency (IF) signal, performing heterodyne sampling, quantization, and encoding for digital measurement, the signal is transmitted via a data bus to the FPGA chip core. Digital signal processing, including polyphase filtering and fast Fourier transform operations, is then performed to calculate the signal's power spectrum data, obtaining the measurement results. These results are then stored in the terminal system's built-in memory chip, arranged in a data matrix with frequency on the X-axis and power spectrum on the Y-axis. Data is output to the terminal receiving computer via the FPGA's PCIe interface, and then transmitted via the PCIe bus to one or more GPUs for further measurement processing, graphics rendering, and data storage. The terminal system rack integrates and packages the FPGA and GPU systems, and includes a unified power supply and clock module.

[0069] Ultra-wideband radio astronomy pulsar coherent achromatic terminal such as Figure 3 As shown, it mainly includes 12 modules, and the system consists of 13 modules. Modules 1-6 are FPGA operation modules, and modules 7-12 are GPU operation modules. The FPGA and GPU are connected via PCIe protocol. The specific functions of each module are as follows:

[0070] ADC sampling module 1 is used to perform four-channel 2.5GSPS@14bit ultra-wideband concurrent sampling of the baseband signal after RF mixing and send it to PFB polyphase filter module 2; the basic architecture of the time-interleaved ADC is based on the concurrent operation of multiple ADC channels, that is, each different channel is equipped with a separate ADC unit. These ADC units arrange their sampling times in a time-interleaved manner. Figure 2The design incorporates a four-channel time-interleaved ADC that meets 10GHz ultra-wideband requirements: the first channel samples at t=0, t=4T, t=8T, etc.; the second channel samples at t=T, t=5T, t=9T; the third channel samples at t=2T, t=6T, t=10T; and the fourth channel samples at t=3T, t=7T, t=11T, with this sequence continuing. Here, T represents the sampling period of each ADC. This parallel configuration effectively expands the total bandwidth by multiplying the sampling rate by the total number of channels. If the sampling rate of each channel is Fs, then the effective sampling rate of a time-interleaved ADC with N channels is ultimately N*Fs. Four 5Gsps sampling channels achieve 20Gsps of complete 10GHz coverage.

[0071] The ADC sampling module 1 also includes four sub-modules to improve performance and stability: a. ADC clock allocation sub-module: Each channel has an independent ADC clock allocation sub-module to ensure that the sampling interval of the current ADC channel differs from that of the adjacent ADC channel by a quarter cycle; b. Gain matching sub-module: Monitors and fine-tunes the gain difference of each ADC channel. By referring to the signal gain of the first channel, it periodically monitors and compensates for the digital gain values ​​of all channels (with a 1pps reference clock at 1s intervals) to achieve stable output signal amplitude and improve signal-to-noise ratio (SNR); c. Offset error adjustment sub-module: Used to adjust and align when the starting zeros of each ADC channel cannot be precisely aligned, i.e., to ensure that when the starting zeros of the first channel ADC are not precisely aligned, the offset error adjustment sub-module is achieved. At reference time 0, the DC monitoring and fine-tuning of the subsequent three channels strictly follows the 1 / 4, 2 / 4, and 3 / 4 cycle intervals after reference time 0. Alignment is achieved by adding fractional delays. This module function can avoid introducing DC offset into the output signal, which would cause baseline offset in the reconstructed signal. d. Timing deviation alignment submodule is used to monitor and fine-tune the alignment reference clock after reference time 0, when the sampling time of each channel ADC deviates from the reference clock by a relative amount of 1PPS (i.e., the 1 / 4, 2 / 4, 3 / 4, and 4 / 4 cycles) during continuous sampling. Alignment is achieved by adding fractional delays. This module function avoids waveform distortion in the reconstructed signal caused by single-channel differences (Reference: S. Thongmee et., “5-bit 5-GS / s Noninterleaved Time-Based ADC in 65-nm CMOS for Radio-Astronomy Applications”, IEEE Transactions on Very Large Scale Integration (VLSI) Systems, 2018). The above four sub-modules will enhance timing and phase control in concurrent ADC design, ensuring timing consistency between the output waveform and the ADC channel.

[0072] The PFB multiphase filtering module 2 is used to perform multiphase filtering on the input sampled data to obtain channelized time-domain data and transmit it to the FFT operation module 3.

[0073] FFT operation module 3 is used to perform fast Fourier transform spectralization processing on the input channelized time domain data to obtain primary spectralized data at 16K frequency points (secondary spectralization will be implemented in GPU high spectral resolution operation module 8 to achieve high-resolution spectralization of 1M points and millions of channels); the spectral data is a 32-bit single-precision floating-point number.

[0074] The spectrum accumulation control module 4 is used to accumulate the primary spectrum data obtained by the FFT operation module 3 to obtain superimposed spectral data and transmit it to the PCIe data encapsulation module 6.

[0075] The clock signal control module 5 is used to provide clock control signals to the ADC sampling module 1, the PFB polyphase filter module 2, the FFT operation module 3, and the spectrum accumulation control module 4, and to coordinate the transmission rate of the ADC sampling module 1 with the processing timing clock of the PFB polyphase filter module 2, the FFT operation module 3, and the spectrum accumulation control module 4.

[0076] PCIe data encapsulation module 6 is used to encapsulate the input superimposed spectral data in real time using PCIe (High-Speed ​​Serial Computer Expansion Bus) data stream.

[0077] GPU memory module 7 is used to cache the data stream transmitted from PCIe data encapsulation module 6; GPU memory module 7 includes global memory and shared memory.

[0078] The GPU high spectral resolution computing module 8 is used to perform secondary spectral channelization subdivision processing on the cached data stream in the GPU memory module 7. Through the parallel computing framework of shared memory, the cufftExecC2C API algorithm library is called to perform 1M-point FFT operation on the secondary spectral channelization subdivision processing result, so as to obtain the ultra-high spectral resolution spectrum data matrix of radio astronomy observation and cache it in shared memory for the GPU related de-dispersion module 9 to call.

[0079] The GPU coherent dedispersion operation module 9 is used to perform coherent dedispersion operation on the ultra-wideband high-spectral-resolution radio astronomy spectral data transmitted by the GPU high-spectral-resolution operation module 8. Specifically, it performs a multiplication process by aligning and multiplying the pulse radio signal spectral data within the entire ultra-wideband bandpass with the CHIRP dispersion function parameter matrix (see TH Hankins, “Coherent Dedispersion: History and Results,” Proceedings of the International Astronomical Union, 2017) to obtain the dedispersion-corrected pulse radio signal spectral data. The CHIRP dispersion function parameter matrix is ​​calculated in advance using the pulsar dispersion calculation formula from the known path loss of strong radio astronomy sources (Formulas 1-1 to 1-5) and stored in the global memory of the GPU memory module 7. In the design of the dedispersion CHIRP function parameter matrix multiplication module, an innovative multi-threaded continuous memory call submodule and a block matrix multiplication submodule are implemented, such as... Figure 4 and Figure 5 Actual measurements show that it improves efficiency by about 10 times compared to traditional, well-known CPU processing technologies.

[0080] a. The multi-threaded sequential memory access submodule is designed for the uncached global memory. It utilizes multiple threads to sequentially access memory data to improve data transfer efficiency. GPU memory is Dynamic Random Access DRAM, which is most efficient for continuous reading and writing. However, traditional CUDA execution follows a thread-by-thread sequence: thread0 -> thread1 -> thread2… This means that when the first thread loads a block of global memory data, the GPU switches to the next thread, preventing continuous memory access within the same thread. This innovative design uses multiple threads to sequentially access memory data. Thread 0 reads the first column of the matrix, thread 1 reads the second column, and so on, with different threads reading different columns of the matrix data synchronously. Within the same clock cycle, each thread stores its read columns into a shared memory block, thus achieving multi-threaded sequential memory access and improving data transfer efficiency.

[0081] b. Block Matrix Multiplication Submodule: This module optimizes global memory read rates and reduces overall computational complexity by implementing block matrix multiplication. The ultra-high spectral resolution spectral data matrix from radio astronomy observations is divided into multiple spectral data blocks. These blocks, along with the CHIRP dispersion function parameter matrix blocks, are configured for optimal transmission based on the memory bandwidth, using power-of-two values ​​of 16 bytes to divide and complete the sub-matrices. For example, a 512x512 array can be divided into 16 32x32 arrays. The value of array C [0,0]-[15,15] after matrix multiplication in the first array is A[0,15]*B[0,15]+A[0,31]*B[16,15]+... and so on. After this division, each thread can complete the matrix multiplication calculation by accessing only the same shared memory. That is, under multi-threaded parallelism, each block of spectral data is multiplied by the corresponding block of CHIRP dispersion function parameter matrix, and the resulting achromatic block matrix data is sorted according to the thread order and stored back into global memory, completing the achromatic spectral data matrix calculation process. This module innovatively realizes the synchronous calculation of multiple submatrices, while also reducing global memory access time and increasing the throughput of matrix operations.

[0082] The GPU inverse Fourier transform operation module 10 is used to perform inverse Fourier transform on the data after dedispersion in the GPU coherent dedispersion operation module 9. It completes the time-domain transformation of the dedispersion spectrum data by calling the dedispersion spectrum data in global memory to perform the CUFFT_INVERSE transform operation.

[0083] The GPU data overlap folding module 11 is used to perform periodic overlap folding operations on the data in the GPU coherent dedispersion operation module 9 according to the radio astronomy standard time domain data based on the pulsar outline to further improve the signal-to-noise ratio, and then transmit it back to the global memory of the GPU memory module 7 for data caching.

[0084] The PSRFITS (Pulsar Data Standardization Format) generation and storage module 12 is used to convert and store the coherent dedispersion data obtained by the folding module 11 in the global memory of the GPU memory module 7 into the PSRFITS standard format, and finally complete the entire ultra-wideband pulsar coherent dedispersion processing.

[0085] Furthermore, the PCIe data encapsulation module 6 performs high-speed memory read and write operations on the PCIe bus in DMA (Direct Memory Access) mode, directly storing the encapsulated data into the target GPU memory module 7.

[0086] Furthermore, the radio astronomy pulsar data standardization format (PSRFITS) generation and storage module 12 uses a RAID disk array to store the data cached by the GPU memory module 7 in parallel.

[0087] Table 1. Summary of Technical Parameters for Ultra-Wideband Radio Astronomy Pulsar Coherent Achromatic Terminal

[0088]

[0089] Table 2. PFB filter coefficients for coherent dedispersion termination in ultra-wideband radio astronomy pulsars.

[0090] realw1 0.1460971 0.146097 0.146097 0.146097 0.146097 0.146097 0.146097 0.146097 0.146097 0.146097 0.146097 0.146097 0.146097 0.146097 realw2 -0.417904 -0.4179 -0.4179 -0.4179 -0.4179 -0.4179 -0.4179 -0.4179 -0.41789 -0.41788 -0.41787 -0.41786 -0.41784 -0.41782 realw3 0.7594459 0.759446 0.759446 0.759444 0.75944 0.759432 0.759418 0.759394 0.759357 0.759303 0.759229 0.759128 0.758996 0.758827 realw4 -1 -1 -1 -0.99999 -0.99998 -0.99996 -0.99992 -0.99985 -0.99974 -0.99958 -0.99936 -0.99906 -0.99867 -0.99817 realw5 1 1 0.999998 0.999991 0.999971 0.999928 0.999852 0.999725 0.999532 0.99925 0.998857 0.998327 0.997631 0.996739 realw6 -0.759446 -0.75945 -0.75944 -0.75943 -0.75941 -0.75936 -0.75927 -0.75912 -0.75889 -0.75856 -0.75809 -0.75746 -0.75664 -0.75558 realw7 0.4179042 0.417904 0.417902 0.417896 0.417877 0.417837 0.417765 0.417646 0.417464 0.417199 0.41683 0.416332 0.415678 0.41484 realw8 -0.146097 -0.1461 -0.1461 -0.14609 -0.14608 -0.14607 -0.14603 -0.14597 -0.14589 -0.14576 -0.14559 -0.14535 -0.14504 -0.14464 imagw1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 imagw2 5.12E-17 5.00E-05 0.0002 0.00045 0.0008 0.00125 0.001799 0.002449 0.003198 0.004047 0.004996 0.006044 0.007192 0.008439 imagw3 -1.86E-16 -0.00018 -0.00073 -0.00164 -0.00291 -0.00454 -0.00654 -0.0089 -0.01162 -0.01471 -0.01816 -0.02197 -0.02613 -0.03067 imagw4 3.67E-16 0.000359 0.001435 0.00323 0.005741 0.00897 0.012916 0.017578 0.022956 0.029049 0.035856 0.043376 0.051607 0.060547 imagw5 -4.90E-16 -0.00048 -0.00191 -0.00431 -0.00766 -0.01196 -0.01722 -0.02344 -0.03061 -0.03873 -0.0478 -0.05782 -0.06879 -0.08069 imagw6 4.65E-16 0.000454 0.001817 0.004088 0.007267 0.011354 0.016347 0.022247 0.029052 0.03676 0.045368 0.054873 0.06527 0.076553 imagw7 -3.07E-16 -0.0003 -0.0012 -0.0027 -0.0048 -0.0075 -0.01079 -0.01469 -0.01918 -0.02427 -0.02995 -0.03622 -0.04308 -0.05051 imagw8 1.25E-16 0.000122 0.000489 0.001101 0.001957 0.003058 0.004402 0.005991 0.007823 0.009897 0.012212 0.014766 0.017558 0.020584

[0091] Performance testing:

[0092] We used the ASP (Arecibo Signal Processor) system to generate 2.5Gb dual-polarized 8-bit complex random numbers to perform performance testing on the ultra-wideband radio astronomy pulsar coherent dedispersion terminal system, such as... Figure 4 , Figure 5 As shown.

[0093] The data is transferred from main memory to the GPU and divided into Nfft sample lengths corresponding to the length of a CHIRP, simulating the original data superimposed in the time domain. The length of the called CHIRP ranges from 16 to 32K samples. For each set of CHIRP data, the FFT call length is 2*(chirp length) to 4M samples. In the operation, the 8-bit data is first converted to single-precision floating-point numbers, and each set of data is subjected to FFT operation, multiplied by the CHIRP function, and then subjected to inverse FFT operation. Finally, the square root of the output data is calculated to obtain the energy spectrum, which is then transferred from the GPU to main memory to complete the entire coherent dedispersion process. The system has been tested and verified to fully realize the coherent dedispersion operation of ultra-wideband radio astronomy pulsars, achieving the designed performance.

[0094] Although specific embodiments of the present invention have been disclosed for illustrative purposes to aid in understanding and implementing the invention, those skilled in the art will understand that various substitutions, variations, and modifications are possible without departing from the spirit and scope of the invention and the appended claims. Therefore, the invention should not be limited to the content disclosed in the preferred embodiments, and the scope of protection claimed by the invention is defined by the claims.

Claims

1. An ultra-wideband radio astronomy pulse radio dispersion-averse terminal, characterized in that, include The ADC sampling module (1) is connected to the PFB multiphase filter module (2) and is used to perform multi-channel ultra-wideband concurrent sampling on the baseband signal after RF mixing and send the sampling data to the PFB multiphase filter module (2). The PFB multiphase filtering module (2) is connected to the FFT operation module (3) and is used to perform multiphase filtering on the input sampled data to obtain channelized time domain data and transmit it to the FFT operation module (3). The FFT operation module (3) is connected to the spectrum accumulation control module (4) and is used to perform fast Fourier transform spectrum processing on the input channelized time domain data to obtain primary spectrum data. The spectrum accumulation control module (4) is connected to the PCIe data encapsulation module (6) and is used to perform accumulation operations on the primary spectrum data to obtain superimposed spectral data and transmit it to the PCIe data encapsulation module (6). The clock signal control module (5) is connected to the ADC sampling module (1), the PFB multiphase filter module (2), the FFT operation module (3), and the spectrum accumulation control module (4) respectively. It is used to provide clock control signals to the ADC sampling module (1), the PFB multiphase filter module (2), the FFT operation module (3), and the spectrum accumulation control module (4), and to control the transmission rate of the ADC sampling module (1) to coordinate with the processing timing clock of the PFB multiphase filter module (2), the FFT operation module (3), and the spectrum accumulation control module (4). The PCIe data encapsulation module (6) is connected to the GPU memory module (7) and is used to encapsulate the input superimposed spectral data and send it to the GPU memory module (7). The GPU memory module (7) is used to cache the input encapsulated data; the GPU memory module (7) includes shared memory and global memory; The GPU high spectral resolution computing module (8) is connected to the GPU memory module (7) and is used to perform secondary spectral channelization subdivision processing on the cached data stream in the GPU memory module (7), and then perform FFT operation on the secondary spectral channelization subdivision processing result to obtain the radio astronomy observation ultra-high spectral resolution spectrum data matrix and cache it in the shared memory. The GPU coherent dedispersion operation module (9) is connected to the GPU memory module (7) and is used to perform aligned multiplication of the ultra-high spectral resolution spectrum data matrix of radio astronomy observations and the CHIRP dispersion function parameter matrix to obtain the spectrum data of pulse radio signals after relevant dedispersion; wherein, the CHIRP dispersion function parameter matrix is ​​calculated according to the known path loss of strong sources in radio astronomy observations and the pulsar dispersion calculation formula and stored in the global memory; The GPU inverse Fourier transform operation module (10) is connected to the GPU memory module (7) and is used to perform inverse Fourier transform on the spectrum data of the pulse radio signal after correlation dedispersion to complete the coherent dedispersion processing. The GPU data folding module (11) is connected to the GPU memory module (7) and the GPU inverse Fourier transform operation module (10) respectively. It is used to perform radio astronomy standard periodic folding operation on the pulsar profile of the relevant dedispersion pulse radio signal spectrum data and send it back to the GPU memory module (7) for caching. The radio astronomy pulsar data standardization format generation and storage module (12) is connected to the GPU memory module (7) and is used to convert and store the coherent dedispersion data cached in the GPU memory module (7) in the PSRFITS standard format.

2. The ultra-wideband radio astronomy pulse radio dispersion-averse terminal according to claim 1, characterized in that, The GPU coherent de-dispersion operation module (9) includes a multi-threaded continuous video memory call submodule and a block matrix multiplication submodule; the multi-threaded continuous video memory call submodule uses multiple threads to read a column of data of the CHIRP dispersion function parameter matrix from the global memory in the same clock cycle, and different threads read different columns of data of the CHIRP dispersion function parameter matrix; Then, each column of data is stored in a memory block of shared memory, and different columns of data are stored in different memory blocks of shared memory. The block matrix multiplication submodule divides the ultra-high spectral resolution spectrum data matrix of radio astronomy observations into multiple spectrum data blocks according to the number of columns in the CHIRP dispersion function parameter matrix. Each thread multiplies each spectrum data block with the column data stored in the corresponding memory block in the shared memory.

3. The ultra-wideband radio astronomy pulse radio dispersion-averse terminal according to claim 2, characterized in that, The block matrix multiplication submodule divides and completes the obtained spectrum data blocks and CHIRP dispersion function parameter matrix blocks using a power of 2 value of 16 bytes.

4. The ultra-wideband radio astronomy pulse radio dispersion-averse terminal according to claim 1, characterized in that, The ADC sampling module (1) also includes four sub-modules to improve sampling performance and stability. Among them, the ADC clock allocation sub-module is used to maintain the sampling time interval between each channel; the gain matching sub-module is used to monitor and adjust the gain difference of each channel to achieve stable output signal amplitude and improve signal-to-noise ratio; the offset mismatch adjustment sub-module is used to adjust and align the zero point of each ADC unit; and the timing deviation alignment sub-module is used to monitor the sampling time of each ADC unit and adjust and align the sampling time of the ADC unit when the sampling time of an ADC unit deviates from the sampling clock deviation set for that ADC unit.

5. The ultra-wideband radio astronomy pulse radio dispersion apochromatic terminal according to claim 1, 2, 3, or 4, characterized in that, The ADC sampling module (1) includes multiple channels, each channel is equipped with an ADC unit, and each ADC unit arranges its sampling time in a time-staggered manner.

6. The ultra-wideband radio astronomy pulse radio dispersion-averse terminal according to claim 5, characterized in that, The ADC sampling module (1) includes four channels, and the ADC sampling module (1) performs 4-channel 2.5GSPS@14bit ultra-wideband concurrent sampling on the baseband signal after RF mixing.

7. The ultra-wideband radio astronomy pulse radio dispersive terminal according to claim 1, 2, 3, or 4, characterized in that, The radio astronomy pulsar data standardization format generation and storage module 12 converts the coherent dedispersion data cached in the GPU memory module (7) into the PSRFITS standard format through a RAID disk array and then stores it in parallel.

8. The ultra-wideband radio astronomy pulse radio dispersion-abolition terminal according to claim 1, 2, 3, or 4, characterized in that, The PCIe data encapsulation module (6) performs high-speed memory read and write operations on the PCIe bus in DMA mode and directly stores the encapsulated data into the memory of the GPU memory module (7).

9. The ultra-wideband radio astronomy pulse radio dispersion-averse terminal according to claim 1, 2, 3, or 4, characterized in that, By using a shared-memory parallel computing framework to call the cufftExecC2C API algorithm library, a 1M-point FFT operation is performed on the secondary spectral channelization subdivision results to obtain an ultra-high spectral resolution spectral data matrix for radio astronomy observations.

Images