Roundoff error effects on spatial lattice algorithm

The floating-point roundoff error effect under finite word length limitations is analyzed for the time updates of reflection coefficients in the spatial lattice algorithm. It is shown that recursive computation is superior to direct computation under finite word length limitations. Moreover, the forgetting factor, which is conventionally used to smooth the time variations of the inputs, is also a crucial parameter in the consideration of the system stability and adaptability under finite word length constraints.