Build runnable differential 3-tap TX FIR reference path

[ingram-chen]

Parent

#2

What to build

Deliver a complete, runnable LTspice reference path for an ideal differential 3-tap TX FIR. The path must start with deterministic NRZ PRBS7 stimulus at 20 Gb/s, pass through the causal pre/main/post FIR, include the agreed differential source impedance and matched termination, and produce both directly inspectable waveforms and scalar measurements consumable by the repository's existing LTspice runner.

Use the agreed defaults: 50 ps UI, coefficients -0.15/0.70/-0.15, 1 UI model latency, 5 ps rise/fall time, 0.6 V common-mode, 50 Ω per-leg source impedance, 100 Ω differential load, and 1.0 Vpp differential at the loaded TX pins. Calibrate the open-circuit Thevenin swing to preserve that loaded definition.

Acceptance criteria

• The runnable artifact contains one reusable ideal differential 3-tap TX FIR behavioral model and one companion testbench.
• The model uses only causal delays of 0 UI, 1 UI, and 2 UI, with the main cursor aligned to the middle path and documented 1 UI latency.
• The default stimulus is deterministic 127-bit PRBS7 NRZ at 20 Gb/s with 5 ps rise and fall times.
• The tap coefficients default to pre=-0.15, main=0.70, and post=-0.15.
• Positive and negative outputs remain centered at 0.6 V common-mode.
• The TX presents 50 Ω series resistance per leg and drives a 100 Ω differential load.
• The loaded TX differential output reaches 1.0 Vpp worst case; the open-circuit source is calibrated for matched-load division.
• All eight adjacent pre/main/post symbol combinations are exercised and resolve to normalized levels in ±0.4, ±0.7, and ±1.0.
• The corresponding loaded differential levels are verified as ±0.20 V, ±0.35 V, and ±0.50 V.
• Cursor spacing is verified as 50 ps and overall model latency as 1 UI.
• FIR-on and main-only FIR-off modes use the same loaded peak differential swing.
• The simulation covers at least eight PRBS7 periods and limits transient time step sufficiently to resolve 5 ps edges.
• The repository's existing LTspice runner completes with zero exit status and parses the scalar measurements.
• Manual QA opens the raw result in LTspice and confirms the differential TX waveform, cursor placement, common-mode, and loaded swing.

Blocked by

None - can start immediately

[ingram-chen]

Implemented in local commit d3eb789 (feat: add LTspice 3-tap TX FIR model). The LTspice runner completed with return code 0 and no syntax errors.

Verified: loaded TX 1.0 Vpp differential, open-circuit 2.0 Vpp, VCM 0.6 V, FIR latency 50 ps, equal FIR-on/off loaded swing, and all eight three-symbol combinations at the expected ±0.20/±0.35/±0.50 V levels.

The issue remains open for the final LTspice waveform-viewer overlay checkbox.