Low-complexity and low-latency equalization technique - probabilistic noise cancellation.

Area of Science:

• Optical communications
• Data center networking
• Digital signal processing

Background:

• Intensity-modulation and direct-detection (IMDD) systems with four-level pulse amplitude modulation (PAM4) are set to dominate data center (DC) and short-reach optical links.
• Future 100 Gbaud PAM4 transceivers necessitate advanced digital signal processing (DSP) and forward error correction (FEC) codes.
• Bandwidth limitations challenge conventional DSP, requiring upgrades from linear feed-forward equalizers (FFE) to more complex algorithms like Volterra equalizers, maximum likelihood sequence estimation (MLSE), or maximum a posteriori probability (MAP).

Purpose of the Study:

• To introduce a novel, low-complexity, low-latency equalization algorithm for data center networks (DCNs).
• To address the limitations of complex DSP algorithms like decision feedback equalizer (DFE) and MLSE/MAP due to power and latency constraints in DCNs.
• To enhance the performance of optical transceivers in bandwidth-limited environments.

Main Methods:

• Development of a novel feedforward equalization algorithm named probabilistic noise cancellation (PNC).
• PNC algorithm weights noise patterns based on their probabilities under bandwidth limitations.
• Efficiently corrects correlated errors caused by noise coloring in the feed-forward equalizer (FFE).

Main Results:

• The proposed PNC algorithm achieves performance between DFE and MLSE.
• It offers a low-complexity and low-latency solution for DSP in DCNs.
• Demonstrates effective correction of errors induced by noise coloring in bandwidth-limited channels.

Conclusions:

• The PNC algorithm presents a viable alternative for advanced DSP in next-generation data center transceivers.
• It meets stringent power consumption and latency requirements for DCNs.
• PNC enhances the reliability of optical links operating under bandwidth constraints.