10 Gbps PAM4 transmission for fiber wireless access in Ka-band based on long short-term memory neural network equalizer

Abstract

In this paper, 28GHz millimeter wave (MMW) in n257 band which is one of the recommended frequency bands for beyond 5G, has been demonstrated by experiment for optical wireless access. The MMW antenna with the bandwidth of 26.5 ~ 29.5GHz and an envelope detector with a 3dB bandwidth of about 500MHz are used to enable 4-ary pulse amplitude modulation (PAM4) signal transmission over 5km fiber and 1.6m wireless distances in our experiment. In order to compensate the linear and nonlinear impairments of the optical wireless links, the long short-term memory (LSTM) neural network nonlinear equalizer is adopted in the receiver DSP. Additionally, the traditional linear equalizer (LE) and Volterra equalizer (VE) are also conducted for comparison. The results show that neither the performances of LE nor VE can reach the 7% hard-decision forward error correction (HD-FEC) threshold (3.8×10^-3) in the case of 5 Gbaud PAM4 transmission over 5km fiber and 1.6m wireless distances. Instead, after adopting the LSTM equalizer, the bit error rate can be reduced to approximate 1×10^-3, which reveals a noticeable performance improvement. Moreover, the performances of the three kinds of equalizers at different transmission rates are further studied. We find that the LSTM can help improve the system capacity from below 9Gbps to above 10Gbps at 7% HD-FEC threshold, which means more than 10 percent improvement has been achieved. According to the above results, we believe that the LSTM equalizer will facilitate the large-capacity communication for the upcoming 5G MMW in fiber wireless access networks.