On Multiple Bit-Width IRC Receiver Optimization by Genetic Approach

• Published in: 2024 26th International Conference on Digital Signal Processing and its Applications (DSPA) pp. 1 - 5
• Main Authors: Lyashev, Vladimir, Potekhin, Robert
• Format: Conference Proceeding
• Language: English
• Published: IEEE 27.03.2024
• Subjects: arithmetic operator design; Degradation; discrete optimization; genetic algorithms; graph theory; IRC receiver; LDL decomposition; low power architectures; matrix inversion; Modulation; multiple input multiple output; Power demand; Receivers; Search problems; Sensitivity; Signal processing algorithms; signal processing design
• DOI: 10.1109/DSPA60853.2024.10510061

In this work, we are going to discuss customized bit-width optimization on the most complicated part of Interference Rejection Combining (IRC) receiver - matrix inversion operation. We are going to reduce the total number of bits required for the operation while keeping the sensitivity of the IRC receiver (matrix inversion precision). It is a developed bit-width optimization tool that can selectively change the bit-width for each individual operation instead of block-wise or system-wise. This is carried out by the example of the LDL decomposition 1 1 Special type of classical Cholesky decomposition [1] and triangular matrix inversion procedure in the IRC receiver. Lower bit-width calculations mean less chip area usage and lower power consumption, although it increases the error level, hence an appropriate level of calculation precision should be established. Therefore, selecting the appropriate algorithm and defining a suitable fit-function for such a contentious task is challenging. The gradient descent algorithm is not applicable for that kind of task because it is a multi-convex optimization problem. Moreover, the brute force algorithm is not applicable too due to the large number of hypotheses. One of the possible solutions to IRC receiver design problem is to use directed search algorithms.