Entropy allocation optimization for PS-OFDM with constellation partitioning based modeling

We propose and experimentally demonstrate an entropy (bits/symbol) allocation scheme for orthogonal frequency division multiplexing (OFDM) signal in a millimeter wave (mmW) over optical fiber fronthaul link. Under the condition of additive white Gaussian noise (AWGN) channel, a closed-form expression between the symbol error rate (SER) of probabilistically shaped quadrature amplitude modulation (QAM) signal and signal-to-noise ratio (SNR) is derived. With a fixed averaged target entropy of the overall signal frame, we use a projected mirror descent algorithm to maximize the averaged SER of the channels based on the pre-measured SNR information. The gradient information is estimated by finite difference using the modeling result. Experimental results in a fiber-mmW converged system validate the proposed method, which allocates more entropy for higher SNR regions compared to a simple scheme that allocates the entropies proportionally based on Shannon's formula. It yields a smoother NGMI and can meet the 0.86 NGMI threshold over the operation frequencies, while uniform QAM signal and Shannon formula based algorithm fails the threshold requirement. Up to 1.8-dB and 0.6-dB received optical power (ROP) gain are achieved as compared with uniform QAM signal and signal with Shannon formula-based allocation, respectively. Compared with the uniform signal, up to 0.7-bits/symbol GMI improvement is also demonstrated.