Add lossy channel and TX FIR eye-diagram acceptance

[ingram-chen]

Parent

#2

What to build

Extend the runnable TX FIR reference path with a replaceable, bypassable lossy channel and complete the FIR-on versus FIR-off eye-diagram acceptance flow. The default channel must preserve DC gain and produce approximately -12 dB insertion loss at the 10 GHz Nyquist frequency. Run equalized and unequalized cases with identical PRBS7 stimulus, source impedance, channel, and receiver termination so the received-eye comparison is meaningful.

The finished slice must remain directly runnable through the existing LTspice flow and manually reviewable in the LTspice waveform viewer.

Acceptance criteria

• The testbench provides a channel bypass mode that leaves the TX FIR behavior unchanged.
• The default behavioral channel has approximately 0 dB low-frequency gain.
• The default channel measures approximately -12 dB insertion loss at 10 GHz between the loaded TX and loaded RX measurement planes.
• The initial channel implementation is a minimal first-order low-pass response with a pole near 2.59 GHz.
• The RX remains terminated by 100 Ω differential impedance.
• FIR-on and main-only FIR-off runs use the same PRBS7 sequence, channel, source impedance, termination, simulation duration, and loaded peak swing.
• LTspice eye folding is configured for 20 Gbaud and its alignment compensates for the 1 UI FIR latency plus channel group delay.
• Initial settling data is excluded from eye evaluation.
• Scalar measurements report channel loss and the received-eye acceptance quantities through the existing LTspice result parser.
• FIR-on normalized RX eye height (eye height divided by received Vpp) is greater than FIR-off normalized RX eye height for the default channel.
• FIR-on RX eye width is greater than FIR-off RX eye width for the default channel.
• A failure to improve normalized eye height or eye width fails the requested coefficient/channel pairing; the implementation does not silently retune coefficients or weaken the criteria.
• Manual QA opens the raw results in LTspice, overlays FIR-on and FIR-off received eyes, and confirms the normalized eye-height and eye-width improvement.
• The channel boundary remains replaceable by future Touchstone data without changing the TX FIR behavioral contract.

Blocked by

• Build runnable differential 3-tap TX FIR reference path #3

Decision note

Absolute eye height is not the acceptance metric for this de-emphasis case because the fixed coefficients intentionally reduce steady-state amplitude. The user approved normalized eye height plus eye width as the meaningful comparison.

[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 default channel: DC gain ≈0 dB, insertion loss −12.0006 dB @10 GHz, bypass error 20 nV. Settled PRBS7 acceptance: normalized eye-height gain 2.404× and eye-width gain 2.062×. The channel boundary owns its 100 Ω differential input termination and is isolated from the direct TX characterization path.

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