The above system is sufficient for the implementation of a single metafeature. However, in practice, there are two observations that necessitate important additions to TClass.
Firstly, the set of domains where a single metafeature is sufficient for learning purposes is limited. Typically, there will be many metafeatures for a given problem domain, and as we will see later, the same metafeature may be applied to a number of channels within the same problem domain. Hence we have to expand our architecture to support multiple metafeatures and multiple applications of metafeatures.
Secondly, while in general it is the temporal characteristics of a signal that are important, there are many cases when other non-temporal attributes are useful. These will be termed global attributes.