AUTOMATIC C CODE GENERATION FOR IMPLEMENTATION OF IIR EMBEDDED SYSTEMS IN FIXED-POINT ARITHMETIC

Due to constraints in speed, size, and power consumption, IIR digital systems are often implemented in fixed-point arithmetic, where the programmer must take into consideration quantization of variables and finite precision issues. Manual writing or conversion of the code is a very error-prone task, besides requiring a great deal of time and effort and automatic conversion from floating point is a robust alternative but the comprehensiveness required for such a generic enterprise frequently renders the final product too long and sophisticated. Then, the objective of this work is to develop an automatic C code generator for the implementation of IIR systems in fixed-point hardware, where the code must be efficient with respect to precision and memory requirements without harming execution time. Generated directly in fixed point and targeting a specific class of systems, it should also be lean, simple, and easy to read and customize. To accomplish this, a broad theory review was conducted to create a scheme that governs the process of code generation in MATLAB environment, including the decisions and trade-offs intrinsic to the system. Lastly, the final code was evaluated analytically and via simulation with respect to precision, by means of statistical metrics.