Master of Industrial Sciences 2013- 2014
Faculty of Engineering Technology, Campus Group T
EXTENDED EIGEN VALUE DECOMPOSITION FOR
MUSIC IMPLEMENTATION OF TRANSIENT SIGNAL
Xiaoyi Lu
1
, Bingtao Chen
1
1
Master student of Industrial Engineering: Intelligent Electronics,
Faculty of Engineering Technology, Campus Group T, Leuven
Vesaliusstraat 13, 3000 Leuven, Belgium
Supervisor: Prof.Ir.Luc Bienstman
Faculty of Engineering Technology, Campus Group T, Leuven
Vesaliusstraat 13, 3000 Leuven, Belgium
<Luc.Bienstman@kuleuven.be>
Co-supervisors: Ir. Lieven Philips, Dr.Ir.Adnan Al-adnani
Electronic Measurements R&D Department , Agilent Technologies, Wingepark 51, 3010 Rotselaar, Belgium
ABSTRACT
This paper considers spectrum analysis of transient signal of short data sets. The classical Fast Fourier Transform
(FFT) technique requires large time domain data to achieve high frequency resolution, however time resolution is
reduced. Therefore, subspace method Multiple Signal Classification (MUSIC) is chosen as it models signals with
short data sets and achieves high frequency resolution. The Jacobi algorithm is applied in FPGA-based Eigen
Value Decomposition based on CORDIC Rotation algorithm for hardware-efficient solutions. Existed 4×4
processing core has already designed by Shi Huimin. To achieve higher frequency resolution, 8×8 processing
blocks is developed in this paper. A pipelined systolic array to decompose a 8×8 symmetric matrix is implemented
with Ping-Pong method. Parallel computation is adopted with multiprocessor structure. The design is verified by
hardware co-simulation. Higher sinusoids dimension result in final diagonal matrix, due better spectrum resolution
achieved and more signals can be handle at the same time. Furthermore, the design saves FPGA area and reduces
hardware cost.
Keywords
Ping-Pong algorithm, MUSIC algorithm, spectrum estimation, eigenvalue decomposition, systolic array,
CORDIC algorithm, FPGA, real-time computation, multiprocessor
