Adaptive filter

An adaptive filter whose output characteristics correspond to the sum of the characteristics of a plurality of elementary filters comprising:

• a first memory for storing sets of m input signals and m wanted output signals, with m an integer .gtoreq.1, said m wanted output signals defining together an Fm vector, and representing a function (F) of said input signals to be approximated, with said function defining a function space;
• a second memory containing modifiable parameters;
• filter calculation means cooperating with said second memory, for defining therewith n elementary filters, where n is an integer .gtoreq.1 and i has all of the n possible values, the elementary filters spanning a sub-space in the function space;
• intermediate calculating means comprising:
• i) means for applying said m input signals to said filter calculation means thereby obtaining nm values, with each one g.sub.i of the n elementary filters delivering m outputs arranged as an m-component filter vector g.sub.i ;
• ii) n scalar multiplication means for multiplying each of the m-component filter vectors g.sub.i with the Fm vector to obtain n resulting quantities representing the normal projections of the vector Fm onto the respective vectors g.sub.i, said resulting scalar multiplication quantities together forming an n-component vector A.sub.i of normal projection;
• iii) linear algebraic calculation means receiving said filter vectors g.sub.i and said vector A.sub.i of normal projection for calculating therefrom intermediate results representing the projection of the function (F) to be approximated on said spanned sub-space in the function space;
• a third memory for storing said intermediate results of said intermediate calculating means;
• calculating means cooperating with the first and third memories for calculating the difference D between the function (F) to be approximated and a corresponding recombination of said intermediate results of said intermediate calculating means;
• and
• minimization means, responsive to the output of said calculating means for changing said modifiable parameters in accordance with one of a deterministic and a stockastic calculation method;
• whereby the desired approximated function (F) is obtained at said output of the filter, from the superimposition of the elementary filters adjusted by the minimization in accordance with the intermediate results of said intermediate calculating means.