38-EFT.WP.EDX.EMI v1.0 | 第15章 数据与复现（DataSpec / Methods.Repro）

目录 ／ 文档-技术白皮书 ／ 38-EFT.WP.EDX.EMI v1.0

第15章 数据与复现（DataSpec / Methods.Repro）

I. 章节目标与结构

1. 目标：定义 EDX.EMI 的数据契约、数据/管线卡与审计轨导出，锁定环境与等价性检验，使 Z_eft(omega)、ΔZ_rad(omega)、P_rad(omega)、E_rad/H_rad、I_CM(omega) 在统一数据口径下可复现、可审计、可发布。
2. 结构：数据契约 → 字段与模板 → 管线卡与环境锁 → 复现流程 → 审计轨与发布 → 可证伪准则 → 合规模板 → 跨章闭环。
3. 共享到达时两口径（等价，须显式路径/测度并记录 delta_form）：
   • 常量外提：T_arr = ( 1 / c_ref ) * ( ∫ n_eff d ell )
   • 一般口径：T_arr = ( ∫ ( n_eff / c_ref ) d ell )

II. 数据契约（EMI 增量）

• 工件类型：raw/cleaned/aligned S-parameters、Z_eft/argZ、Znorm(omega)、（可选）T_mm/Z0_mm、arrival、binding_ref、deemb/sync、ΔZ_rad(omega)、I_port(omega)、P_rad(omega)、E_rad/H_rad（含 AF/PF 与测距 R）、I_CM(omega)、HF_KPI{E_phase/GDR/ΔW}、qa_gates、audit_manifest。
• 最小一致性：SI 单位；check_dim=pass；Re{Z_eft}≥0、KK_consistency=pass；Re{ΔZ_rad}≥0；两口径 T_arr 一致；corr(P_rad,E/H)≥ρ_gate（建议 ρ_gate≥0.8）。
• 版本与校验：semver + 每文件 sha256；记录 created_at、toolchain 与 env_lock 指纹。

III. 数据集卡（Dataset Card v1.0，EMI 字段）

最小模板（可直接粘贴）

dataset_card:

dataset_id: "EDXEMI-001"

version: "1.0.0"

source: "field-site"

instruments:

- {type:"VNA", model:"—", cal:"2025-08-01"}

- {type:"EMI_receiver", detector:"qpk", if_bw_kHz:120, time_const_ms:1.0}

- {type:"antenna", AF_file:"/cal/AF_3m.yaml", version:"2025.1"}

- {type:"LISN", spec:"5µH CM", version:"v2.0"}

deemb: {method:"TRL", version:"1.2", artifact:"/artifacts/deemb.json", baseline_id:"BLSN-EMI-001"}

sync: {ref:"10MHz", scheme:"shared_ref+trigger", dt_sync_s: 2.0e-12}

binding_ref: "LAY2PATH-EMI-0001"

arrival:

form: "n_over_c" # or "one_over_c_times_n"

gamma: "explicit"

measure: "d_ell"

c_ref: 299792458.0

Tarr_s: 1.234e-09

u_Tarr_s: 6.0e-12

delta_form: "n_over_c"

Znorm_ohm: [50.0, 50.0]

sparams: "/artifacts/S.s2p"

mixed_mode:

enabled: true

T_mm: "/cfg/T_mm.yaml"

Z0_mm_ohm: [100.0, 25.0]

impedance:

Z_eft: {real:[...], imag:[...]}

deltaZ_rad: {Re_ohm:[...], Im_ohm:[...]} # Re ≥ 0

I_port_A: [ ... ]

P_rad_W: [ ... ]

fields:

site: {distance_m: 3.0, coords:"...", AF_file:"/cal/AF_3m.yaml", PF_file:"/cal/PF.yaml"}

E_rad_peak_dBuV_m: [ ... ]

H_rad_peak_dBA_m: [ ... ]

currents:

I_CM_A: [ ... ]

HF_KPI:

E_phase_rad: 0.043

GDR_s: 1.8e-10

ΔW: 0.17

qa_gates: {check_dim:"pass", passivity:"pass", KK:"pass"}

hashes:

files:

- {path:"/artifacts/S.s2p", sha256:"..."}

- {path:"/artifacts/deemb.json", sha256:"..."}

created_at: "2025-09-16T10:00:00Z"

toolchain: {simstack_emi:"1.0.0", kernels:"0.9.3"}

IV. 管线卡（Pipeline Card v1.0）与环境锁（Env-Lock v1.0）

• 标准阶段：ingest → align(I30-EMI) → deembed/renorm → s2z(I40-EMI) → sync_corr → path_correct → rad_equiv(P_rad) → field_convert(E/H) → kpi_eval(Ω) → export(cards)。
• 环境锁：记录 OS/解释器/依赖/容器摘要与硬件，统一单位制；提供随机种子。

pipeline_card:

pipeline_id: "EDXEMI-PIPE-001"

version: "1.0.0"

steps:

- {id:"ingest", cmd:"load_sparams --input /artifacts/S.s2p"}

- {id:"align", cmd:"map_ports --binding LAY2PATH-EMI-0001"}

- {id:"deembed", cmd:"deembed --fixture /artifacts/deemb.json --baseline BLSN-EMI-001"}

- {id:"renorm", cmd:"renorm --znorm Znorm.yaml --tmm /cfg/T_mm.yaml --z0mm Z0_mm.yaml"}

- {id:"s2z", cmd:"map_S_to_Z --mode mixed"}

- {id:"sync_corr", cmd:"phase_corr --dt_sync_s 2.0e-12"}

- {id:"path_correct", cmd:"pathfix --binding LAY2PATH-EMI-0001"}

- {id:"rad_equiv", cmd:"z2prad --zbase Z_base.npy"}

- {id:"field_conv", cmd:"prad2field --af /cal/AF_3m.yaml --pf /cal/PF.yaml --dist 3.0"}

- {id:"kpi_eval", cmd:"kpi --omega ω1,ω2 --gates gates.yaml"}

- {id:"export", cmd:"export_cards --out /release"}

env_lock:

os: {name:"Ubuntu", version:"22.04"}

interpreter: {name:"Python", version:"3.11.6"}

packages:

- {name:"numpy", version:"1.26.4", hash:"sha256:..."}

- {name:"scipy", version:"1.13.1", hash:"sha256:..."}

- {name:"pandas", version:"2.2.2", hash:"sha256:..."}

container: {image:"registry/edxemi:1.0.0", digest:"sha256:..."}

hardware: {cpu:"x86_64", gpu:"none", ram_GB:32}

seeds: {global: 20250916}

V. 复现流程（Methods.Repro，执行口径）

• 获取并校验：拉取 dataset_card/pipeline_card/env_lock，核对 sha256。
• 重建环境：按 env_lock 启动容器/虚拟环境。
• 对齐与改正：执行 align → deembed/renorm → s2z → sync_corr → path_correct，确保 passivity/KK 与 arrival 完整。
• 辐射等效：ΔZ_rad = Z_eft − Z_base，强制 Re{ΔZ_rad}≥0；计算 P_rad = ½·Re{ΔZ_rad}|I_port|^2。
• 场量换算与一致性：用 AF/PF 与 R 换算 E/H，验证 corr(P_rad,E/H) ≥ ρ_gate。
• KPI 再计算：在相干窗 Ω 上重算 E_phase/GDR/ΔW；相位先做时间基改正与解缠。
• 到达时一致性：两口径 T_arr 差值 ≤ u(T_arr)。
• 等价性判定：按阈值 ε_Z/ε_φ/ε_PRad_E/ε_PRad_H/ε_ICM/(ε_ΔW) 比较复现与发布结果，出具 qa_gates。

VI. 审计轨与发布

audit_manifest:

run_id: "RUN-2025-09-16-001"

toolchain: {simstack_emi:"1.0.0", kernels:"0.9.3"}

diffs: {params_changed:["Ks_tau"], reason:"tuning within prior"}

seeds: {global: 20250916}

qa:

check_dim: "pass"

passivity: "pass"

KK: "pass"

logs: "/logs/run_001.txt"

release_manifest:

normative_refs: ["EFT.WP.Core.DataSpec v1.0","Methods.Repro v1.0"]

artifacts:

- "/release/dataset_card.yaml"

- "/release/pipeline_card.yaml"

- "/release/env_lock.yaml"

- ["/release/audit_manifest.yaml", "/release/qa_report.json"]

hashes:

dataset_card: "sha256:..."

pipeline_card: "sha256:..."

env_lock: "sha256:..."

audit: "sha256:..."

VII. 可证伪准则（Repro 对应）

1. R-EMI-1（两口径一致）：|T_arr(n_over_c) − T_arr(one_over_c_times_n)| > u(T_arr)。
2. R-EMI-2（正实/因果失败）：复现得到 Re{Z_eft}<0 或 KK_consistency 失败。
3. R-EMI-3（功率—场量一致性失败）：corr(P_rad,E/H)<ρ_gate。
4. R-EMI-4（等价性失败）：
   • ε_Z = max_ω |Z_eft^{rep} − Z_eft^{pub}| / |Z_eft^{pub}| > ε_Z_gate
   • ε_φ = max_ω |arg Z^{rep} − arg Z^{pub}| > ε_φ_gate
   • ε_PRad_E = 1 − corr(P_rad^{rep}, E^{pub}) > ε_PRad_E_gate
   • ε_PRad_H = 1 − corr(P_rad^{rep}, H^{pub}) > ε_PRad_H_gate
   • ε_ICM = 1 − corr(I_CM^{rep}, Re{ΔZ_rad}^{pub}) > ε_ICM_gate
   • ε_ΔW = max_ω |ΔW^{rep} − ΔW^{pub}| > ε_ΔW_gate
5. R-EMI-5（环境不可再现/审计缺失）：env_lock 无法重建或缺关键 hashes/logs。

VIII. 合规模板（可直接粘贴）

# 1) 两口径 T_arr

T1 = (1/c_ref) * sum(n_eff[i]*Δell[i]) # n_over_c

T2 = sum((n_eff[i]/c_ref) * Δell[i]) # one_over_c_times_n

assert abs(T1 - T2) <= u_Tarr

# 2) S→Z / ΔZ_rad / P_rad / E/H

Z = map_S_to_Z(S_ren, Znorm(ω))

ΔZ_rad = Z - Z_base

assert min(Re(ΔZ_rad)) >= 0.0 and KK_consistency(Z)

P_rad = 0.5 * Re(ΔZ_rad) * abs(I_port)**2

ρE, ρH = corr(P_rad, E_env), corr(P_rad, H_env)

assert ρE >= ρ_gate and ρH >= ρ_gate

# 3) KPI

phi = unwrap(arg(Z) - ω*Δt_sync)

E_phase = max_abs(phi - (ω*Tarr + φ0_opt))

# ΔW 对比

eps_dW = max_over_ω(abs(ΔW_rep - ΔW_pub))

# 4) 等价性集

eps_Z = max_abs(Z_rep - Z_pub) / max_abs(Z_pub)

eps_phi = max_abs(phi_rep - phi_pub)

eps_PrE = 1 - corr(P_rad_rep, E_env_pub)

eps_PrH = 1 - corr(P_rad_rep, H_env_pub)

eps_Icm = 1 - corr(I_CM_rep, Re(ΔZ_rad)_pub)

assert eps_Z<=ε_Z_gate and eps_phi<=ε_φ_gate and eps_PrE<=ε_PRad_E_gate and \

eps_PrH<=ε_PRad_H_gate and eps_Icm<=ε_ICM_gate and eps_dW<=ε_ΔW_gate

阈值建议（可在数据卡细化）

equivalence_gates:

eps_Z_gate: 0.05 # 5% 相对误差

eps_phi_gate: 0.05 # rad

eps_PRad_E_gate: 0.20 # 1−corr 上限（越小越严）

eps_PRad_H_gate: 0.20

eps_ICM_gate: 0.20

eps_ΔW_gate: 0.10

rho_gate: 0.80

IX. 跨章引用与闭环

• 依赖：第8–9章（映射/计量链）、第10章（证伪）、第13章（多域协同）、EDX.HighSpeed 第15章（数据与复现）。
• 对接：第16章（设计规程与工程清单，将本章的等价性/审计门纳入签核）与 EFT.WP.Core.DataSpec / Methods.Repro 统一规范。