Abstract
Representative of subspace decomposition DOA estimation algorithm, the MUSIC
algorithm has ultra-high resolution capability than traditional DOA estimation algorithm,
because MUSIC algorithm uses the orthogonal characteristics of signal subspace and
noise subspace. However, in practical applications due to the subspace decomposition
algorithm requires eigenvalue decomposition operator, It’s difficult for DSP processing
chip to meet the requirements of real-time, one way to meet this requirements is
FPGA-based parallel eigenvalue decomposition. This paper studied the eigenvalue
decomposition of parallel computing in depth. Proposed an improve scheme Based on
BLV parallel array structure. did the main work as follows:
In-depth studied the Jacobi algorithm for eigenvalue decomposition, Eigenvalue
decomposition need arctangent function and rotation operations,Therefore, based on the
CORDIC algorithm designed processing unit studied convergence and calculation
precision of CORDIC unit, and has carried on the simulation and error analysis.
Conducted a detailed analysis of the computing time and resource use about BLV
parallel array structure. BLV structure has a low utilization efficiency of resources , and
resource requirements increase square-fold as the matrix dimension increases. To solve
these problems, The paper presents an improved pipeline structure,Completed two
calculation module design, Compared the computation time, resource consumption and
efficiency of per unit area of BLV structure and pipeline structure. The two structures
are implemented on FPGA, comparison illustrates the effect of improved
structure.According to the synthesis reporting and timing simulation,The improved
structure in computing time increased by 6%, but 51.8% reduction in the use of
resources.The computational efficiency of the unit area of 1.957 times to parallel BLV
structure.
Keyword:DOA Jacobi EVD BLV array structure Pipeline BLV CORDIC