Error sources and error propagation in the Levinson-Durbin algorithm

Two kinds of numerical error occur during the execution of the Levinson-Durbin algorithm for solving a linear systems of equations. The first has an erratic nature and is generated by a particular iteration's erroneous value of input autocorrelation. Its effect is constrained to a specific recursion. The second is a systematic error that is enlarged with subsequent computations. Both, however, may be traced via a methodology that yields the exact number of erroneous digits during the computation.