Memristor-Based Large-Scale High-Radix FFT Circuit Design in NR System

Large-scale FFT operations in NR system are highly resource-intensive and computationally complicated, constituting a significant aspect of signal processing. Using high-radix to realize large-scale FFT can reduce the algorithm complexity and the number of stages, however, it introduces vector-matrix multiplication (VMM) operations. Memristor based circuits can efficiently perform VMM operations and have great development prospects. Therefore, this paper reports the design of a high-radix FFT circuit based on memristor arrays for large-scale FFT in NR system, from the perspective of analog in-memory computing. Firstly, we introduce the algorithm and implementation of high-radix FFT and its application prospect in NR communication systems. Then, we construct the memristor model based on practical devices, propose a bisection pulse strategy for resistance modulation based on the error threshold, and point out the existence of the over-precision paradox for the first time. After that, we propose the single-memristor mapping scheme aimed at accurate matrix operations. Compared with the memristor pair mapping scheme, it can reduce the number of memristor cells used by nearly 50%. Meanwhile, the computing unit circuit is jointly designed with the mapping scheme to enhance computational efficiency. This design achieves a direct one-step FFT radix operation process. Ultimately, the proposed circuit is applied to address FFT/IFFT in OFDM system modulation and demodulation. Hereon, we propose the Monte Carlo pilot to equalize the memristor array calculations. The memristor array calculation with an error threshold of 1% can also approach the ideal performance. It solves the problem of constellation point shrinkage caused by non-ideal mapping and greatly improves the BER performance of the system.