Signal integrity (SI) analysis heavily relies on simulations. To support various interfaces and an increasing number of topologies, extensive simulations have been run over years. Such SI analysis is time and resource intensive, generating a huge amount of data. Can the existing big data set be exploited to ease our simulation and analysis efforts? In this work, we utilize wavelet techniques to compress one channel response (tens of thousands of data points) into only a small set of coefficients. There are many potential applications, if the reconstruction accuracy is maintained. One such application is to construct a Neural Net model of those coefficients over physical design parameters, such that the channel impulse response for any physical design can be obtained without any circuit simulation. This application, along with several others will be discussed in this paper. Moreover, wavelet-based techniques will be compared to more traditional compression techniques.
[1]
W. Kinsner,et al.
Aquantitative comparison of different mother wavelets for characterizing transients in power systems
,
2005,
Canadian Conference on Electrical and Computer Engineering, 2005..
[2]
O. Farle,et al.
An Adaptive Multi-Point Fast Frequency Sweep for Large-Scale Finite Element Models
,
2009,
IEEE Transactions on Magnetics.
[3]
R. Haddad,et al.
Multiresolution Signal Decomposition: Transforms, Subbands, and Wavelets
,
1992
.
[4]
Martin Vetterli,et al.
Wavelets and filter banks: theory and design
,
1992,
IEEE Trans. Signal Process..
[5]
B. Muralikrishnan,et al.
Engineering Surface Analysis With Different Wavelet Bases
,
2003
.
[6]
Rolf Schuhmann,et al.
Model Order Reduction for Large Systems in Computational Electromagnetics
,
2006
.
[7]
Truong Q. Nguyen,et al.
Wavelets and filter banks
,
1996
.