one-time-slot-not-two-our-patented-iq-code-for-pdl-and-crosstalk-mitigation

The Impairment Coherent DSP Can't Touch

Modern coherent receivers are very good at undoing chromatic dispersion and polarization-mode dispersion — both are unitary effects, and unitary effects can be inverted analytically at the receiver. Polarization-dependent loss (PDL) is different: it attenuates one polarization more than the other, which is a non-unitary operation. No amount of conventional equalization, MIMO processing, or carrier-phase estimation recovers information that PDL has already discarded.

Nyquist wavelength-division multiplexing built on digital sub-carriers (DSC) — the architecture behind today's 800G-and-beyond superchannels — makes this worse, not better. Packing sub-carriers near Nyquist spacing to maximize spectral efficiency introduces inter-carrier interference (ICI) between adjacent sub-bands whenever laser drift narrows the guard band. PDL and crosstalk end up compounding each other in exactly the systems most motivated to push capacity.

Why the Existing Fixes Don't Fit

Space-time (ST) codes — the Golden code and Silver code among them — mitigate PDL by linearly combining signals across multiple time slots. They work, but that time diversity is the problem: it's built around a coding structure that doesn't map cleanly onto single-carrier or OFDM-style DSP chains in use today, and it roughly doubles decoding complexity by design. One-time-slot alternatives have been proposed — pairwise coding, four-dimensional signaling — and they do fit inside existing DSP. But every one of them was designed for exactly two channel streams. Scale a DSC system to eight sub-carriers and sixteen streams, and none of them generalize.

IQ-Code: Separate, Rotate, Recombine

IQ-code closes both gaps at once. It splits every transmitted QAM stream into its in-phase (I) and quadrature (Q) components, applies a tailored linear combination — built from real rotation matrices derived using algebraic number theory — to the real and imaginary parts independently across all 2N streams (N sub-carriers × 2 polarizations), then recombines them into an expanded constellation before transmission. The receiver reverses the same rotation with a simple zero-forcing decoder.

Two design choices make this practical rather than merely elegant. First, it runs in a single time slot — there's no second symbol interval to wait for, no incompatibility with single-carrier DSP, and none of the latency Golden- or Silver-code schemes carry. Second, because the rotation matrices are defined for arbitrary dimension, IQ-code scales to any number of sub-carriers N, not just two streams. And it costs nothing in rate or power: the same number of QAM symbols is transmitted — only the shape of the constellation changes, and mean transmitted power is conserved.

What the Numbers Show

• Robust to PDL orientation.
  In back-to-back single-carrier tests at PDL = 6 dB, IQ-code delivers a 0.4 dB OSNR gain at the FEC limit, and — unlike the uncoded case — its performance barely moves with the PDL rotation angle θ (0.1–0.4 dB spread vs. a much wider swing uncoded). The gain also grows with PDL severity: more loss imbalance, more benefit.
• Holds up at 800G.
  On a 96 GBaud DP-16QAM, 8 sub-carrier system over 1000 km (ten 100 km spans) with a distributed, randomly-oriented PDL channel, IQ-code adds 0.5 dB OSNR gain at the FEC limit with no ICI, and up to 0.5 dB with ICI present.
• Doesn't disturb the nonlinear regime.
  IQ-code reshapes the constellation, which can raise its kurtosis and in principle change nonlinear sensitivity. In practice, the shift is small (~0.1) and the system still gains 0.5 dB in Q-factor at 4 dBm launch power.
• Stacks with FEC and PCS.
  Paired with an LDPC(1944,1458) code, IQ-code adds up to 1.7 dB OSNR gain at BER = 10^-3, and — notably — keeps performance stable across PDL orientation where the uncoded worst case degrades sharply. Combined with probabilistic constellation shaping and FEC together in a full 800G system, it adds 1.0 dB without ICI and 1.5 dB with ICI.
• Beats the Golden code where it matters for DSP.
  Head-to-head, IQ-code outperforms the Golden code by 0.5 dB when there's no ICI; the Golden code edges ahead at high OSNR once ICI is present. The real differentiator is architectural, not just OSNR: IQ-code needs one time slot where the Golden code needs two.

That last trade-off is worth stating precisely rather than rounding up. Zero-forcing decoding complexity scales with system size for both codes: IQ-code is cheaper at a single carrier (9 vs. 27 operations) and at four sub-carriers (87 vs. 108), but crosses over at eight sub-carriers, where IQ-code's complexity (303) exceeds the Golden code's (216). What IQ-code preserves at every scale is the one-time-slot structure — meaning it drops into existing single-carrier DSP chains without the redesign a two-time-slot code demands.

Why This Matters