This study introduces a practical approach to implement real-time signal processing algorithms for general surveillance radar based on NVIDIA graphical processing units (GPUs). The pulse compression algorithms are implemented using compute unified device architecture (CUDA) libraries such as CUDA basic linear algebra subroutines and CUDA fast Fourier transform library, which are adopted from open source libraries and optimized for the NVIDIA GPUs. For more advanced, adaptive processing algorithms such as adaptive pulse compression, customized kernel optimization is needed and investigated. A statistical optimization approach is developed for this purpose without needing much knowledge of the physical configurations of the kernels. It was found that the kernel optimization approach can significantly improve the performance. Benchmark performance is compared with the CPU performance in terms of processing accelerations. The proposed implementation framework can be used in various radar systems including ground-based phased array radar, airborne sense and avoid radar, and aerospace surveillance radar.
This abstract is for the academic institution profiles session
This presentation will focus on radar research programs at the University of Oklahoma, the radar research in OU has more than 50 years history of collaboration with NOAA, and has been through tremendous growth since early 2000. Before 2010, the focus was weather radar and weather surveillance, and since the Defense, Security and Intelligence (DSI) initiative in 2011, there have many new efforts on the defense and military radar applications. This presentation will focus on the following information: (1) The history, facilities and instrumentations of Advanced Radar Research Center, (2) Focus area of polarimetric phased array systems, (3) Focus area of airborne and spaceborne radars, (4) Intelligent radar information processing, (5) Innovative antenna and components.
KEYWORDS: Radar, Signal to noise ratio, Transceivers, Digital signal processing, Clocks, Calibration, Matrices, Field programmable gate arrays, Signal processing, Radar signal processing, Advanced radar, Expectation maximization algorithms
The new challenges originated from Digital Array Radar (DAR) demands a new generation of reconfigurable backend processor in the system. The new FPGA devices can support much higher speed, more bandwidth and processing capabilities for the need of digital Line Replaceable Unit (LRU). This study focuses on using the latest Altera and Xilinx devices in an adaptive beamforming processor. The field reprogrammable RF devices from Analog Devices are used as analog front end transceivers. Different from other existing Software-Defined Radio transceivers on the market, this processor is designed for distributed adaptive beamforming in a networked environment. The following aspects of the novel radar processor will be presented: (1) A new system-on-chip architecture based on Altera’s devices and adaptive processing module, especially for the adaptive beamforming and pulse compression, will be introduced, (2) Successful implementation of generation 2 serial RapidIO data links on FPGA, which supports VITA-49 radio packet format for large distributed DAR processing. (3) Demonstration of the feasibility and capabilities of the processor in a Micro-TCA based, SRIO switching backplane to support multichannel beamforming in real-time. (4) Application of this processor in ongoing radar system development projects, including OU’s dual-polarized digital array radar, the planned new cylindrical array radars, and future airborne radars.
With the rapid proliferation of small unmanned aerial systems (UAS) in the national airspace, small operational drones are being sometimes considered as a security threat for critical infrastructures, such as sports stadiums, military facilities, and airports. There have been many civilian counter-drone solutions and products reported, including radar and electromagnetic counter measures. For the current electromagnetic solutions, they are usually limited to particular type of detection and counter-measure scheme, which is usually effective for the specific type of drones. Also, control and communication link technologies used in even RC drones nowadays are more sophisticated, making them more difficult to detect, decode and counter. Facing these challenges, our team proposes a “software-defined” solution based on noise and LPI radar. For the detection, wideband-noise radar has the resolution performance to discriminate possible micro-Doppler features of the drone versus biological scatterers. It also has the benefit of more adaptive to different types of drones, and covertly detecting for security application. For counter-measures, random noise can be combined with “random sweeping” jamming scheme, to achieve the optimal balance between peak power allowed and the effective jamming probabilities. Some theoretical analysis of the proposed solution is provided in this study, a design case study is developed, and initial laboratory experiments, as well as outdoor tests are conducted to validate the basic concepts and theories. The study demonstrates the basic feasibilities of the Drone Detection and Mitigation Radar (DDMR) concept, while there are still much work needs to be done for a complete and field-worthy technology development.
This study introduces a practical approach to develop real-time signal processing chain for general phased array radar on NVIDIA GPUs(Graphical Processing Units) using CUDA (Compute Unified Device Architecture) libraries such as cuBlas and cuFFT, which are adopted from open source libraries and optimized for the NVIDIA GPUs. The processed results are rigorously verified against those from the CPUs. Performance benchmarked in computation time with various input data cube sizes are compared across GPUs and CPUs. Through the analysis, it will be demonstrated that GPGPUs (General Purpose GPU) real-time processing of the array radar data is possible with relatively low-cost commercial GPUs.
KEYWORDS: Radar, Signal to noise ratio, Super resolution, Detection and tracking algorithms, Doppler effect, Digital filtering, Image filtering, Optical resolution, Antennas, Resolution enhancement technologies
Matched Filter sidelobes from diversified LPI waveform design and sensor resolution are two important considerations in radars and active sensors in general. Matched Filter sidelobes can potentially mask weaker targets, and low sensor resolution not only causes a high margin of error but also limits sensing in target-rich environment/ sector. The improvement in those factors, in part, concern with the transmitted waveform and consequently pulse compression techniques. An adaptive pulse compression algorithm is hence desired that can mitigate the aforementioned limitations. A new Matched Filter based Iterative Adaptive Approach, MF-IAA, as an extension to traditional Iterative Adaptive Approach, IAA, has been developed. MF-IAA takes its input as the Matched Filter output. The motivation here is to facilitate implementation of Iterative Adaptive Approach without disrupting the processing chain of traditional Matched Filter. Similar to IAA, MF-IAA is a user parameter free, iterative, weighted least square based spectral identification algorithm. This work focuses on the implementation of MF-IAA. The feasibility of MF-IAA is studied using a realistic airborne radar simulator as well as actual measured airborne radar data. The performance of MF-IAA is measured with different test waveforms, and different Signal-to-Noise (SNR) levels. In addition, Range-Doppler super-resolution using MF-IAA is investigated. Sidelobe reduction as well as super-resolution enhancement is validated. The robustness of MF-IAA with respect to different LPI waveforms and SNR levels is also demonstrated.
This paper investigates the feasibility of real-time, multiple channel processing of a digital phased array system backend design, with focus on high-performance embedded computing (HPEC) platforms constructed based on general purpose digital signal processor (DSP). Serial RapidIO (SRIO) is used as inter-chip connection backend protocol to support the inter-core communications and parallelisms. Performance benchmark was obtained based on a SRIO system chassis and emulated configuration similar to a field scale demonstrator of Multi-functional Phased Array Radar (MPAR). An interesting aspect of this work is comparison between “raw and low-level” DSP processing and emerging tools that systematically take advantages of the parallelism and multi-core capability, such as OpenCL and OpenMP. Comparisons with other backend HPEC solutions, such as FPGA and GPU, are also provided through analysis and experiments.
New radar applications need to perform complex algorithms and process large quantity of data to generate useful information for the users. This situation has motivated the search for better processing solutions that include low power high-performance processors, efficient algorithms, and high-speed interfaces. In this work, hardware implementation of adaptive pulse compression for real-time transceiver optimization are presented, they are based on a System-on-Chip architecture for Xilinx devices. This study also evaluates the performance of dedicated coprocessor as hardware accelerator units to speed up and improve the computation of computing-intensive tasks such matrix multiplication and matrix inversion which are essential units to solve the covariance matrix. The tradeoffs between latency and hardware utilization are also presented. Moreover, the system architecture takes advantage of the embedded processor, which is interconnected with the logic resources through the high performance AXI buses, to perform floating-point operations, control the processing blocks, and communicate with external PC through a customized software interface. The overall system functionality is demonstrated and tested for real-time operations using a Ku-band tested together with a low-cost channel emulator for different types of waveforms.
In order to achieve low probability-of-intercept (LPI), radar waveforms are usually long and randomly generated. Due to the randomized nature, Matched filter responses (autocorrelation) of those waveforms can have high sidelobes which would mask weaker targets near a strong target, limiting radar’s ability to distinguish close-by targets. To improve resolution and reduced sidelobe contaminations, a waveform independent pulse compression filter is desired. Furthermore, the pulse compression filter needs to be able to adapt to received signal to achieve optimized performance. As many existing pulse techniques require intensive computation, real-time implementation is infeasible. This paper introduces a new adaptive pulse compression technique for LPI waveforms that is based on a nonparametric iterative adaptive approach (IAA). Due to the nonparametric nature, no parameter tuning is required for different waveforms. IAA can achieve super-resolution and sidelobe suppression in both range and Doppler domains. Also it can be extended to directly handle the matched filter (MF) output (called MF-IAA), which further reduces the computational load. The practical impact of LPI waveform operations on IAA and MF-IAA has not been carefully studied in previous work. Herein the typical LPI waveforms such as random phase coding and other non- PI waveforms are tested with both single-pulse and multi-pulse IAA processing. A realistic airborne radar simulator as well as actual measured radar data are used for the validations. It is validated that in spite of noticeable difference with different test waveforms, the IAA algorithms and its improvement can effectively achieve range-Doppler super-resolution in realistic data.
Super-computing based on Graphic Processing Unit (GPU) has become a booming field both in research and industry. In this paper, GPU is applied as the main computing device on traditional RADAR super resolution algorithms. Comparison is provided between GPU and CPU as computing architecture and MATLAB, as a widely used scientific implementation, is also included as well as C++ implementation in demonstrations of CPU part in the comparison. Fundamental RADAR algorithms as matched filter and least square estimation (LSE) are used as standard procedure to measure the efficiency of each implementation. Based on the result in this paper, GPU shows an enormous potential to expedite the traditional process of RADAR super-resolution applications.
The traditional radar RF transceivers, similar to communication transceivers, have the basic elements such as baseband waveform processing, IF/RF up-down conversion, transmitter power circuits, receiver front-ends, and antennas, which are shown in the upper half of Figure 1. For modern radars with diversified and sophisticated waveforms, we can frequently observe that the transceiver behaviors, especially nonlinear behaviors, are depending on the waveform amplitudes, frequency contents and instantaneous phases. Usually, it is a troublesome process to tune an RF transceiver to optimum when different waveforms are used. Another issue arises from the interference caused by the waveforms - for example, the range side-lobe (RSL) caused by the waveforms, once the signals pass through the entire transceiver chain, may be further increased due to distortions. This study is inspired by the two existing solutions from commercial communication industry, digital pre-distortion (DPD) and adaptive channel estimation and Interference Mitigation (AIM), while combining these technologies into a single chip or board that can be inserted into the existing transceiver system. This device is then named RF Transceiver Optimizer (RTO). The lower half of Figure 1 shows the basic element of RTO. With RTO, the digital baseband processing does not need to take into account the transceiver performance with diversified waveforms, such as the transmitter efficiency and chain distortion (and the intermodulation products caused by distortions). Neither does it need to concern the pulse compression (or correlation receiver) process and the related mitigation. The focus is simply the information about the ground truth carried by the main peak of correlation receiver outputs. RTO can be considered as an extension of the existing calibration process, while it has the benefits of automatic, adaptive and universal. Currently, the main techniques to implement the RTO are the digital pre- or –post distortions (DPD), and the main technique to implement the AIM is the Adaptive Pulse Compression (APC). The basic algorithms and experiments with DPD will be introduced which is also the focus of this paper. The discussion of AIM algorithms will be presented in other papers, while the initial implementation of AIM and correlation receiver in FPGA devices will also be introduced in this paper.
The large utility-scale wind turbines are reported to have negative impact on nearby radars due to complex scattering mechanisms, which is usually referred to as the radar Wind Turbine Clutter (WTC). Extremely complicated time-varying Doppler spectrum have been observed. Conventional ground clutter filter techniques thus have failed in mitigating the non-stationary components in the frequency domain. Rotation of the blades is a micro-motion as the wind turbine always stays at the same location. The time-evolving spectrum associated with the blade rotation is therefore a Micro-Doppler signature, which is important in characterizing radar WTC. This paper will disclose some latest findings from our recent studies in characterizing the Micro-Doppler radar signatures of wind turbine through electromagnetic modeling.
Super-resolution (SR) is a radar processing technique closely related to the pulse compression (or correlation receiver). There are many super-resolution algorithms developed for the improved range resolution and reduced sidelobe contaminations. Traditionally, the waveforms used for the SR have been either phase-coding (such as LKP3 code, Barker code) or the frequency modulation (chirp, or nonlinear frequency modulation). There are, however, an important class of waveforms which are either random in nature (such as random noise waveform), or randomly modulated for multiple function operations (such as the ADS-B radar signals in ). These waveforms have the advantages of low-probability-of-intercept (LPI). If the existing SR techniques can be applied to these waveforms, there will be much more flexibility for using these waveforms in actual sensing missions. Also, SR usually has great advantage that the final output (as estimation of ground truth) is largely independent of the waveform. Such benefits are attractive to many important primary radar applications. In this paper the general introduction of the SR algorithms are provided first, and some implementation considerations are discussed. The selected algorithms are applied to the typical LPI waveforms, and the results are discussed. It is observed that SR algorithms can be reliably used for LPI waveforms, on the other hand, practical considerations should be kept in mind in order to obtain the optimal estimation results.
The past two decades has witnessed a renaissance in passive radar research. One of the areas of research in passive radar that has received recent attention is the use of reflected GNSS signals as the signal-of-opportunity for bistatic synthetic aperture radar (BSAR), known as space-surface BSAR (SS-BSAR) [1-7]. SS-BSAR is unique because it uses GNSS signals, which, in the case of the US owned and operated Global Positioning System (GPS), provide almost constant coverage to almost the entire earth [8-10]. Furthermore, the GPS satellites transmit left and right-hand circularly polarized signals combined during transmission to form a right-hand circularly polarized (RHCP) signal; the benefits being, when compared to horizontal or vertical polarized waveforms, is the signal reflection re-radiates in the opposite, or left-hand circularly polarized (LHCP), polarization with signal loss ranging from 15 to 20 dB [5, 10]. One major drawback to using GPS as the signal-of-opportunity is that the received signal level is extremely low, and lower when reflected (see Table 1).
A unified digital pulse compression processor is introduced as a radar-application-specific-processor (RASP) architecture for the next generation of adaptive radar. Based on traditional pulse compression matched filter and correlation receiver, the processor integrates specific designs to handle waveform diversities, which includes random noise waveforms, as well as digital transceiver self-reconfiguration for adaptive radars. Initial prototype of this processor is implemented with the latest Xilinx FPGA device and tested with an RF spaceborne radar transceiver testbed. Initial validation results show the effectiveness of real-time processing and engineering concepts.
A hybrid-aperture radar system is being developed for passive, GNSS-based sensing and imaging missions. Different from previous work, the real aperture (RA) array has excellent cross-range resolution and electronic scanning capability, and synthetic aperture processing is applied for the dimension along the UAV/aircraft flight path. The hybrid aperture thus provides real-time, combined sensing capability and multiple functions. Multi-level signal synchronization and tracking is used to ensure the signal phase coherency and integrity. The advantages of covert radar sensing and reduced onboard computing complexity of this sensor are being demonstrated through experiments.
KEYWORDS: Radar, Signal to noise ratio, Unmanned aerial vehicles, Super resolution, Detection and tracking algorithms, Interference (communication), Signal processing, Smoothing, Antennas, Spatial resolution
The sense and avoidance (SAA) and due-regard radar systems have strict requirements on size, weight and power (SWaP)
and target localization accuracies. Also, the multi-mission capabilities with both weather and hard targets are critical to
the survivability of unmanned aerial vehicles (UAV) in the next generation national airspace. The aperture limitations of
the aircraft sensor installation, however, have prevented large antennas/arrays to be used. The tradeoffs among
frequencies, resolutions and detection range/accuracies have not been fully addressed. Innovative concepts of
overcoming the aperture limitation by using a special type of super-resolution technology are introduced. The first
technique is based on a combination of thinned antenna array, an extension to the traditional Multiple Signal
Classification (MUSIC) technique, and applying a two-dimensional sidelobe mitigation technique. To overcome the
degradation of MUSIC-type of approach due to coherent radar signals, a special waveform optimization procedure is
used. The techniques for mitigating artifacts due to "thinned" array are also introduced. Simulated results of super-resolution
techniques are discussed and evaluated, and the capability of separating multiple targets within aperture-constrained
beamwidth is demonstrated. Moreover, the potential capabilities of autonomous weather hazard avoidance
are also analyzed.
An airborne radar sensing technology for detecting and monitoring of multiple types of external hazards is investigated.
Antennas with spatial and polarimetry diversity are adopted in the radar sensor to support the comprehensive hazard
monitoring requirements. A knowledge-aided joint space-time processing approach is developed for monitoring wind
hazard as well as estimating target direction and Doppler spectrum simultaneously. The hazard microphysics information
can be retrieved through polarimetric data processing. In addition to the intelligent processing algorithms, the system
design and the tradeoffs are considered.