Analytical accuracy evaluation of fixed-point systems

To satisfy cost constraints, application implementation in embedded systems requires fixed-point arithmetic. Thus, applications defined in floating-point arithmetic must be converted into a fixed-point specification. This conversion requires accuracy evaluation to ensure algorithm integrity. Indeed, fixed-point arithmetic generates quantization noises due to some bits elimination during a cast operation. These noises propagate through the system and modify computing accuracy. In this paper, an accuracy evaluation model based on an analytical approach is presented and valid for all systems including arithmetic operations. The LMS algorithm example is developed and its validity is verified through experimentations.