Evaluation of aerosol deconvolution algorithms for determining submicron particle size distributions with diffusion battery and condensation nucleus counter

Determining the particle size distribution of airborne particles is important in many contexts, including understanding and thus reducing the deposition of particles on micro-electronic components during their manufacture. Sizing of particles smaller than 0.1 μm is usually done with a diffusion battery, which requires use of a deconvolution algorithm to obtain particle size distributions. The following algorithms were evaluated: CINVERSE (Crump and Seinfeld, Aerosol Sci. Technol. 1, 363, 1982). Twomey's iterative procedure (Twomey, J. Comput. Phys. 18, 188, 1975), expectation maximization (Maher and Laird, J. Aerosol Sci. 16, 557, 1985), constrained least-squares fit (Nelder and Mead, Comput. J. 7, 308, 1965; Cooper and Spielman, Atoms. Envir. 10, 1976). Expectation maximization and constrained least-squares fit are more suited to this use than are the other two. The non-monotonic response of the diffusion battery with respect to particle size cannot be corrected for by any such algorithm. One could modify the diffusion battery to prevent entrance of the larger particles or one could use an independent measurement of the larger particles to correct the diffusion battery data. Using the latter approach provided an improved estimate of the particle size distribution in a clean room.