Low-Power, Phase-Slip Tolerant, Multilevel Coding for M-QAM

For a cost efficient, high capacity, coherent dense wavelength division multiplex transmission in data center interconnect applications, the number of optical interfaces should be minimized, the throughput of each modulated channel should be maximized, and the system density should be increased. These goals are supported by using high baud rates, spectrally efficient modulation, and low-power solutions. Allowing independent design of non-binary modulation and binary forward error correction (FEC) coding, bit-interleaved coded modulation (BICM) eases the system design for software-defined reach (or capacity) by combining a programmable modulation with a single binary FEC code. Unfortunately, as we demonstrate, the suboptimality of BICM implementations tends to increase both with the higher modulation order and also the higher FEC overhead. This motivates a search for alternatives to BICM. In this paper, we examine a low-power multilevel coding architecture for higher order quadrature amplitude modulation (M- QAM). Limiting ourselves to non-iterative low-power solutions, we show that two-level coding using a low-complexity convolutional code concatenated with a powerful hard decoded FEC may be an attractive approach. We demonstrate a pilot-free, phase-slip tolerant inner coding scheme with low complexity, attractive performance, and low latency. In a real-time prototype, we study the performance and implementation aspects for 64 QAM, resulting in 12.9 dB net coding gain with less than 1 W power dissipation per coded 100 G in 14 nm ASIC technology.