Anal. Chem. 2004, 76, 2859-2868
Two-Dimensional Nonlinear Wavelet Compression of Ion Mobility Spectra of Chemical Warfare Agent Simulants Libo Cao, Peter de B. Harrington,*,‡ and Chang Liu†
Center for Intelligent Chemical Instrumentation, Department of Chemistry and Biochemistry, Ohio University, Athens, Ohio 45701-2979, and School of Electrical Engineering and Computer Science, Russ College of Engineering and Technology, Ohio University, Athens, Ohio 45701-2979
Ion mobility spectrometry (IMS) affords miniaturized hand-held devices that can be used for monitoring and remote measurement. Because such instruments have limits on storage capacity or bandwidth for wireless transmission, data compression is important. Furthermore, all instruments should be operated with the fastest possible sampling rates because a signal-to-noise gain can be achieved by wavelet compression. Linear wavelet compression (LWC) applied to IMS data may cause peak distortion when the spectra are reconstructed. Nonlinear wavelet compression (NLWC) precisely preserves the peak location (i.e., drift time), height, and shape. IMS data of three chemical warfare simulants, dimethyl methylphosphonate, triethyl phosphate, and dipropyleneglycol monomethyl ether, were collected from an Ion Track ITEMISER and a Graseby Ionics detector CAM. Twodimensional NLWC was used to compress the IMS data in the drift time and data acquisition dimensions on IMS data of chemical warfare simulants. NLWC was applied to achieve a compression factor of 1/128 with relative error of root-mean-square of F 0 0.545 3
Randomly Chosen Wavelet Filter Combinations Used for Linear Wavelet Compression, Nonlinear Wavelet Compression no.
row filter
column filter
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
daublet 18 coiflet 2 coiflet 5 daublet 16 coiflet 1 daublet 20 daublet 8 coiflet 5 coiflet 2 coiflet 5 symmlet 7 coiflet 3 symmlet 7 coiflet 4 daublet 14
daublet 20 symmlet 6 daublet 10 coiflet 5 symmlet 6 coiflet 5 daublet 14 daublet 10 daublet 14 daublet 8 daublet 16 symmlet 6 coiflet 1 symmlet 6 daublet 16
high fidelity. Baseline noise and small peaks below the threshold are neglected during compression, which could pose problems for miniscule analytical responses. When the analyte exposures are infrequent, the same benefit of intensity threshold compression to the sample acquisition dimension will be obtained. Because the threshold is applied to the two-dimensionally compressed matrix, a synergistic benefit Analytical Chemistry, Vol. 76, No. 10, May 15, 2004
2867
is obtained that arises from compression characterizing regions (channels and spectra) with large features. CONCLUSIONS A two-dimensional nonlinear wavelet compression method was developed and applied to IMS data. A compression factor of 1/128 with relative error of root-mean-square (RERMS) of 0.25% was achieved for ITEMISER data. With same compression factor, a RERMS of 0.09% was achieved for CAM data, which had larger noise levels. The ALS models from reconstructed and unprocessed data sets gave indistinguishable spectra and concentration profiles. The wavelet filters for both drift time and acquisition time dimensions were optimized, although optimization may not be necessary to obtain good compression. NLWC avoids peak distortion and eliminates artifacts caused by linear wavelet compression. Peak shifting was reduced to
