benchmark suite radar signal & data processing ceres epc workshop 2008-10-01
TRANSCRIPT
BENCHMARK SUITERADAR SIGNAL & DATA PROCESSING
CERES EPC WORKSHOP 2008-10-01
THE BENCHMARK SUITE
The purpose is to evaluate processing architectures with regard to radar signal & data processing requirements
The suite comprises
Signal processing kernels• “front-end” processing• data-independent, stream-oriented
Information and knowledge processing kernels• “back-end” processing• data-dependent, thread oriented
Application examples• some of the kernels are used• illustrates complications in data access/movement
It is to a large extent based on the HPEC Challenge benchmark suite
THE HPEC CHALLENGE BENCHMARK SUITECreated under the DARPA PCA program, introduced 2005
Nine kernel benchmarks:
Signal processing• Time-domain and frequency-domain FIR filters
• QR factorization
• Singular value decomposition
• Constant false-alarm rate detection
Information and knowledge processing• Pattern matching
• Graph optimization via genetic algorithm
• Real-time database operation
Communication kernel• Corner turn (memory rearrangement) of a data matrix
Metrics
Latency, throughput, efficiency
MORE KERNELS
Complement to the HPEC Challenge suite
Fast Fourier Transform• The free FFTW package from MIT
• C subroutine library for computing the DFT in one or more dimensions
• Benchmark source code and methodology are available
Interpolation kernels• Cubic interpolation
• Bi-cubic interpolation
• Source code is available
APPLICATIONS
signal processing kernels
processing chain 1
processing chain 2
different processing directions in chain
channel
range
pulse
data access/movement complications when combining kernels
APPLICATIONS
A simplified Doppler signal processing chain• problem: processing along different directions in data set
• benchmark: Doppler filtering, pulse compression, CFAR detection
Space-Time Adaptive Processing (STAP)• problem: weight calculations based on a sliding volume in a 3D data
set
• benchmark: QR decompositions of matrices formed from data in a sliding volume
Synthetic Aperture Radar (SAR) processing• problem: 2D interpolations along tilted paths in memory
• benchmark: elementwise addition of data from two matrices accessed along tilted lines
THE PROVIDED SOURCE CODE
Single processor code
For comparisons/reference
Excecutable ”spec”
Basis for parallel code, if applicable
Click to edit Master title style
Click to edit Master text styles• Second level
• Third level
• Fourth level
• Fifth level