Experimental Assessment of Digital Predistortion Using Reinforcement Learning for 5G Analog Radio over Fiber Links

Radio over Fiber (RoF) is pivotal for extending reliable 5G connectivity to enhanced remote area communications (ERAC) use cases, that can be used for transporting analog signals from the central office to a simplified remote base station, composed only of an optical detector and radio-frequency front-end. However, the RoF link introduces undesired nonlinear effects that can severely degrade overall system performance and prohibitively increase the out-of-band emissions. We investigate and propose the use of reinforcement learning (RL) algorithms based digital predistortion (DPD) called as RLDPD method for linearizing next-generation Analog Radio over Fiber (A-RoF) links within the 5G landscape. We experimentally compare the proposed RLDPD with the conventional methods including generalized memory polynomial (GMP), canonical piecewise linearization (CPWL) and deep learning based convolutional neural networks (CNN). The experiment evaluation involves multiband 5G new radio (NR) flexible-waveform signals at 3 GHz and 10 GHz carrier signal transmitted over a 10 km single mode fiber length. The performance is compared in terms of error vector magnitude (EVM), adjacent channel leakage ratio (ACLR) and computation complexities. The RLDPD achieves a EVM of 2.85% for 5G NR waveform, surpassing GMP's 4.8%, CPWL's 3.5%, CNN's 3.08% and 11.25% without linearization, while also reducing ACPR by 19 dBc when compared to absence of linearization.