Feature/vz error amp

[paulQM]

PR Summary: CZ phase compensation with error amplification

Branch: feature/vz-error-amp
Base: main (ab66cf09)
Commits: 4

Commit	Message
c3e17440	feat: add files
12330449	rename files and add parameter
8053e8f4	feat: improve analysis with sinc fit
f4ff5b53	chore: remove local VS Code settings from repo

---

Summary

This PR adds a new CZ calibration node (33b) that calibrates residual single-qubit (local Z) phase shifts from the CZ macro using error amplification: the CZ gate is repeated 1…N times before a final virtual frame rotation, which makes small phase errors easier to resolve than the standard single-CZ Ramsey sequence in node 33a.

It also renames the existing phase-compensation node from 33_cz_phase_compensation.py → 33a_cz_phase_compensation.py to make room for the new variant in the CZ calibration numbering scheme.

---

What’s new

Node 33b — 33b_cz_phase_compensation_error_amp.py

• Runs a Ramsey-like sequence (x90 → N×CZ → x90) separately on control and target qubits.
• Sweeps an additional virtual frame rotation on top of the stored phase_shift_control / phase_shift_target.
• Sweeps number of CZ operations (default 1–8) to amplify phase error before readout.
• Supports state discrimination or raw IQ readout.
• Updates QUAM state by adding the fitted compensation frame to the existing phase shifts (modulo 1).

Calibration utilities — calibration_utils/cz_phase_compensation_error_amp/

New package mirroring the structure of the standard cz_phase_compensation utilities:

Module	Role
parameters.py	Node parameters: shots, frame sweep (frame_range, num_frames), number_of_operations, CZ macro type, state discrimination
analysis.py	Averages signal over repetitions, fits a sinc model to locate the optimal compensation frame; falls back to parabolic peak refinement if the sinc fit fails
plotting.py	2×N panel figure (control / target per qubit pair) with mean curve, sinc overlay, and fitted optimum
__init__.py	Public exports

Analysis model:

f(frame) = B + A · sinc(w · (frame − frame₀) / π)

For each qubit (control and target), the signal is averaged over number_of_operations, then the peak position gives the residual phase to compensate. Fit success requires both control and target fits to succeed.

Rename — node 33a

• 33_cz_phase_compensation.py → 33a_cz_phase_compensation.py
• Node name updated to 33a_cz_phase_compensation (99% content preserved; standard single-CZ phase compensation unchanged in behavior).

---

File changes

Change	Path
Added	calibration_utils/cz_phase_compensation_error_amp/ (4 files)
Added	calibrations/CZ_calibrations/33b_cz_phase_compensation_error_amp.py
Renamed	33_cz_phase_compensation.py → 33a_cz_phase_compensation.py
Modified	calibrations/1Q_calibrations/07_iq_blobs.py

Stats: 7 files changed, +705 / −3 lines (vs main).

---

Comparison: 33a vs 33b

	33a (standard)	33b (error amp)
CZ repetitions	1	1…N (default 8)
Frame sweep	Full 0→1 cycle	Narrow window around 0 (default ±0.05 in 2π units)
Analysis	Sinusoidal Ramsey fit	Sinc fit on mean-over-N curve
State update	Subtract fitted phase	Add fitted phase

Use 33a for coarse or routine compensation; use 33b when residual phases are small and need amplification to measure reliably.

---

Notes / review items

1. 07_iq_blobs.py — Includes a local debug override (node.parameters.qubits = ["qD3", "qA6"]) and a trailing # %% cell marker. Consider reverting before merge if these are not intentional for the repo.
2. Sign convention — 33b adds fitted phase to stored compensation; 33a subtracts. This matches the different experiment geometry (amplified error vs direct Ramsey phase) but is worth confirming against hardware validation.

---

Test plan

• Run 33a on a qubit pair and confirm behavior is unchanged after the rename (node name / GUI registration).
• Run 33b with use_state_discrimination=True on a calibrated pair; verify sinc fits succeed and plots show control/target peaks.
• Confirm phase_shift_control / phase_shift_target update correctly in QUAM state after a successful 33b run.
• Re-run a simple 2Q check (e.g. conditional phase or RB) to verify improved compensation vs pre-calibration.
• Test fallback path: intentionally use a narrow frame_range or noisy data and confirm parabolic fallback still returns a finite phase.
• Optional: run with use_state_discrimination=False and confirm IQ → voltage conversion and analysis still work.