The fuzzy c-means clustering method provides an accurate method for quantitative estimation of odorant concentration also as seen in Section 5.3. This is important for determination of the state of spoilage. Particularly, TMA concentration is widely recognized to be a reliable indicator of stage of fish spoilage. There exists a low concentration of TMA in fresh fish also due to bacterial activities in the fish body. After death, however, TMA concentration increases with time as the bacterial degradation advances. The TMA concentration is indicative of time elapsed after death. The TMA discrimination ability of SAW sensor array combined with its sensitivity and fast response may facilitate development
of SAW TMA meter for monitoring the freshness/spoilage of fish products. The list of polymers listed in Table 3 was also prepared by consulting wide literature on characterization and application of polymers for various organic vapors sensing under different situations, see the references mentioned there. We tried to include polymers having wide range of chemical selectivities toward target headspace volatiles for both fish and milk however there may be some significant omissions. The quality of selection will obviously improve with more potential polymers in the list. Further, in simulations we considered only the additive frequency noise. This was done to present only a representative practical
level of noise challenge for analyte detection. In reality, SAW feedback oscillators used for chemical sensing are limited by the phase fluctuations in the SAW frequency control device. The feedback configuration converts these phase fluctuations into frequency fluctuations. The spectral frequency noise power density SF (F) is related to the spectral phase noise power density S ̊(F) according to SF (F) = F2S ̊(F) where F denotes Fourier (or offset) frequency [126]. The other factor of concern to real SAW sensor operation is the frequency drift due to components aging, sorbents
retention and relaxation in the sensing polymer film. The development of a real SAW sensor system, of course, has to address these issues. In this work our objective was to chalk out a simple procedure for polymer selection by data mining before taking up actual fabrication of sensors. The analyses presented meet this objective very well. By utilizing the available literature in the context of milk and fish associated headspace volatiles we have seen that the fuzzy c-means clustering yields an efficient method for optimal polymer search from a large pool of potential polymers; the efficacy of selection will, of course, depend on the richness of the polymers list and the accuracy of the partition coefficient data. The impact of this approach on SAW electronic nose development can however be felt only after real fabrication and evaluation of application specific systems. The conclusions based on the theoretical SAW sensor array responses are not likely to go far off as the SAW sensor response model is fairly established and widely used [35,42]; worst, if not very precise, the present simulation based analysis and selection approach will certainly reduce the amount of experimentation needed for sensor selection.