The log-add kernel operation is represented as a summation of an average and a correction factor composed of a constant and a term based on a difference between the input arguments. In a described embodiment, the correction factor is approximated using the reduction of the correction factor into a Taylor series expansion, which may be defined around the difference between the input arguments as approximately zero. The approach may be further optimized to provide the Taylor series expansion as being modified to compute the correction factor with simple additions, multiplications, and shift operations. If the input arguments are close to each other, the new computed representation may be used, and if the arguments are further apart, the max operation is used. The log-add kernel operation also may be extended to more than two arguments, for application, for example, in the kernel operation of the generalized Viterbi decoder with a branch fan-in greater than 2. Additionally, a method of computing bit soft values for symbols of a higher-order modulation using log-add operations is disclosed, where each log-add is provided as an approximation including a primary term and a parameterized correction factor. Performance close to the optimal soft value generation is achieved at roughly the same complexity as the dual-min algorithm.