Accurate Power Estimation of Modulation Systems for FPGAs using Learning-Augmented IP Modeling

The market for Field Programmable Gate Arrays (FPGAs) has grown as an outcome of its adoption in wearable technology and cell phones. FPGAs have become increasingly complex, allowing for the realization of multiple communication systems. Because of their intricate architecture, FPGAs use a lot more power than the Application Specific Integrated Circuits (ASICs) counterparts, despite their superior performance. Consequently, power estimation is becoming a crucial design metric. Although power estimation is possible with the current commercial technologies, it takes longer because the designs must go through the entire design cycle. Furthermore, power assessment of the whole system is still in its infancy, with the majority of the literature concentrating on power estimation of individual blocks. Binary Phase Shift Keying (BPSK) and Quadrature Phase Shift Keying (QPSK) modulation methods, which are employed in numerous satellite and wireless communication applications, have therefore been designed in this study. A new identity that may assess the power of entire systems is also put forth. User-defined intellectual property (IP) cores and embedded IP have been used in the design of the QPSK and BPSK systems. With the use of MATLAB R2016b and a supervised machine learning technique, a power estimation model of various blocks of QPSK and BPSK modulation systems has been created. The models are then verified against a commercial product. Finally, the proposed identity is used to estimate the power of complete QPSK and BPSK modulation systems. Its accuracy is confirmed using the Vivado 2014.2 tool, targeting the Zynq family device, and the most recent research. The results show that, when compared to other identities that are currently available in the literature, the proposed identity performs better power estimation of entire systems and produces results that are aligned with the commercial tool.