Daubechies-4 wavelets system-on-chip for classifications of mixed-chemical
H.S. Abdel-Aty-Zohdy, Stephanie Roth, Ephram Mebrahtu · 2013
Wavelet transforms are favorably used for compressing and filtering complex signals from chemical sensors for classifications. The Daubechies 4 (D4) transform is used and found to be more accurate especially with chemical mixtures, fairly simple to implement, and effective at pre-processing data from 32 sensors. The resulting output of this system is sent to a Spiking Neural Network processing chip for chemical classification. The system features a SPI bus for data input and output. It accepts a 32 bit fixed point value inputs for faster processing using bit shifts and adds to implement multiplication. The system is designed to be modular and capable of being used on other potential applications. Experimental measurements of mixed gases, including IEDs, are presented. This paper presents the design evaluated on a VHDL platform, to be implemented on a Si chip for fabrication through MOSIS 0.5 um CMOS technology. D4 has more capability than Haar transform, especially with gas mixtures such as 0.1% of TATP/HMTD in Acetonitrile. This high precision system design occupies a silicon chip area of less than 2 mm2. Experimental results are presented with good sensitivity, stability, and tolerance.