Experimental Analysis and FPGA Implementation of the Real Valued Time Delay Neural Network Based Digital Predistortion

Yesil S., Sen C., YILMAZ A. Ö.

26th IEEE International Conference on Electronics, Circuits and Systems (ICECS), Genoa, Italy, 27 - 29 November 2019, pp.614-617 identifier identifier

  • Publication Type: Conference Paper / Full Text
  • Volume:
  • Doi Number: 10.1109/icecs46596.2019.8964743
  • City: Genoa
  • Country: Italy
  • Page Numbers: pp.614-617
  • Keywords: RVTDNN, DPD, activation function
  • Middle East Technical University Affiliated: Yes


This paper presents an FPGA implementation of the Real Valued Time Delay Neural Network (RVTDNN) based digital predistortion with a very low resource utilization and high throughput. The implementation exploits efficient utilization of FPGA primitives and approximation of activation functions that can be realized with simple logic operations. The proposed modifications and constraints on the algorithms have been decided and verified based on a closed-loop adaptive hardware setup including RFHIC RWP03040-1H PA as the unit under test together with injected RF impairments. Over 20dB of in-band performance gain has been reported with a (4,8,2) RVTDNN configuration utilizing only 52 DSP48 blocks, 224 LUTs and 120 FFs.