Extracting Time-Constant Spectra by the Subspace Barzilai and Borwei Non-Negative Least Square Algorithm
Joosun Yun, Byongjin Ma, Guesuk Lee, Tae-Hee Jung, Dong‐Soo Shin, Youngbeom Kim, Hyundon Jung · 2024
The subspace Barzilai and Borwei non-negative least square (SBB-NNLS) algorithm is adopted to extract time-constant spectra from thermal impedance curves. The standard method to convert from Foster to Cauer network based on JEDEC 51-14 is used to convert from time-constant spectra to thermal structure functions. Extracted results are compared with those from the T3Ster Master and the Richardson-Lucy algorithm. Comparison results show that the SBB-NNLS algorithm may offer guidelines to decide the effective iteration number of Bayesian-based deconvolution methods to extract time-constant spectra and structure functions when the noise in curves of thermal impedance Zthbecomes severe. Such capability comes from the projection property of the SBB-NNLS algorithm, allowing to reduce the total data points describing time-constant spectra with acceptable accuracy.