Higher precision integer operations instead of floating-point operations in computers or microprocessors
Jintao Zhang, Gang Li, Yongshun Luo, Ling Lin · Review of Scientific Instruments · 2021
The rounding errors of floating-point operations are inevitable in computers or microprocessors, and this issue will make the calculation results inaccurate, unreliable, or even completely incorrect. For this purpose, this paper proposes to replace floating-point operations with integer operations to improve the operation precision. The key lies in not only controlling the variable type as the integer to avoid the automatic conversion of intermediate operation results into floating-point numbers but also converting floating-point operations in the operation process into integer operations using some numerical calculation methods. Lock-in amplifier is one of the most widely used instruments in the field of weak signal detection. This paper only takes the digital lock-in amplifier (DLIA) as an example for detailed analysis and proposes a DLIA based on integer calculation. The experimental results show that replacing floating-point operations with integer operations can obtain higher operation precision without "wasting" memory, and the improvement will be more significant as the calculation amount increases. The research will help to further improve the calculation accuracy of digital signal processing and other scientific computations in computers or microprocessors.