Deep learning-based defense against phase-injection attacks in pilot-multiplexed continuous-variable quantum key distribution

System
  • T. Shen, J. Ma, X. Wang
  • 2026
Note

Read in full for the A-0103 review. Two roles. It adds a fifth lever on the same phase-reference-alignment surface, injecting structured non-Gaussian perturbations into a published pilot tone, with the same signature of biasing the estimate without noticeable changes in the estimated channel parameters. And it settles A-0103's maturity question. Sec. 2.3 adds the perturbation numerically to recorded phase values, but Sec. 3.1 performs it optically, at a threat-actor terminal of a polarisation controller, a polarising beam splitter with a power meter and an FPGA-driven phase modulator, on one pilot pulse in ten over 20 km of fibre, and Sec. 3.2 measures the resulting biased estimates. A-0103 stays theoretical on scope, not on evidence: this route biases the phase estimate, not the trusted phase-noise term A-0103 exploits, and the four trusted-term routes are simulation only. Its own maturity belongs to a pilot-tone row SQOUT does not yet hold.

Identifier: Shen2026
Type: Journal Article
Year: 2026