Stochastic Approximation Aided Adaptive Thresholding for Optical Detection in PAM4 Based FSO Transmission

Abstract

This study proposes an adaptive thresholding optical detection technique using a stochastic approximation (SA) to enhance receiver performance in multi-level pulse amplitude modulation (PAM) based free-space optical (FSO) transmission. When the receiver detects the symbol by fixed threshold decision (FTD) scheme, the detection performance of high order modulation, including PAM4, becomes degraded due to a random fluctuation of the received signal. Owing to the slow varying characteristic of atmospheric turbulence, we used an adaptive linear prediction (ALP) filter to predict the decision threshold of the upcoming frame. The experiment was conducted on various degrees of turbulence channels. A Mach-Zehnder modulator (MZM) based turbulence channel emulator was used to experimentally investigate the turbulence-induced scintillation effect. The experiment results confirmed that the proposed technique yielded improved receiver performance compared to FTD. In the case of considering the amplified spontaneous emission (ASE) noise of erbium-doped fiber amplifier (EDFA), the adaptive threshold decision (ATD) scheme obtained about 3dB gain compared with an even-valued threshold (EVT) decision scheme when the average EDFA input optical power was below -16dBm.

RZ‐PPM impulse tone‐based CSI estimation for adaptive threshold decision in FSO communications

“…Adaptive linear prediction (ALP) filters are used to predict the next frame of CSI. 12 However, the convergence speed of the stochastic approximation (SA) algorithm is comparatively slow, requiring more time to adapt to rapidly changing channels. Pulse position modulation (PPM) is a widely used binary modulation scheme in FSO communication.…”

“…However, the black‐box nature of DL results in poor interpretability of estimation results. Adaptive linear prediction (ALP) filters are used to predict the next frame of CSI 12 . However, the convergence speed of the stochastic approximation (SA) algorithm is comparatively slow, requiring more time to adapt to rapidly changing channels.…”