"""M0 基准实验:单指数信号 + 10ns 直接采样 + 噪声 + NLS 寿命拟合。 场景 A:理想无噪 → 拟合应恢复设定寿命(M0 验收①,对应原项目"标准拟合"基准); 场景 B:SNR 50 dB + 12bit 量化,重复 N 次 → τ̂ 分布(M0 验收③)。 运行:python -m fiber_delay_sim.experiments.exp01_m0_baseline [--repeats 20] [--seed 42] """ import argparse import numpy as np from .. import pipeline from ..export import record from ._plot import setup_style def run(tau=52.8, window_ns=400.0, snr_db=50.0, bits=12, repeats=20, seed=42, runs_dir="runs"): plt = setup_style() # ---------- 场景 A:理想无噪 ---------- ideal_params = {"preset": "single", "tau": tau, "snr_db": None, "bits": None, "window_ns": window_ns} rec_ideal = pipeline.run_m0(ideal_params, seed=seed) # ---------- 场景 B:加噪重复 ---------- noisy_params = {"preset": "single", "tau": tau, "snr_db": snr_db, "bits": bits, "window_ns": window_ns} rec_noise = pipeline.run_m0(noisy_params, seed=seed) reps = pipeline.run_m0_repeats(noisy_params, repeats=repeats, seed=seed) # ---------- 图 1:理想采样与拟合(第一张拟合图) ---------- fig1, ax = plt.subplots(figsize=(7, 4.2)) ax.plot(rec_ideal["t_cont"], rec_ideal["y_cont"], "-", color="#9aa5b1", lw=1.2, label="连续信号(真值)") ax.plot(rec_ideal["t_samples"], rec_ideal["y_samples"], "o", color="#2D5F8A", ms=6, label=f"10 ns 直接采样({rec_ideal['n_samples']} 点)") t_fit = rec_ideal["t_samples"] f = rec_ideal["fit"] ax.plot(t_fit, f["A"] * np.exp(-t_fit / f["tau"]) + f["B"], "r--", lw=1.5, label=f"NLS 拟合:τ̂ = {f['tau']:.4f} ns") ax.set_xlabel("时间 (ns)") ax.set_ylabel("归一化强度") ax.set_title(f"M0 理想无噪基准:设定 τ = {tau} ns,恢复偏差 " f"{rec_ideal['rel_error']:.2e}(相对)") ax.legend() ax.grid(alpha=0.3) # ---------- 图 2:加噪单次 ---------- fig2, ax = plt.subplots(figsize=(7, 4.2)) ax.plot(rec_noise["t_cont"], rec_noise["y_cont"], "-", color="#9aa5b1", lw=1.2, label="连续信号(真值)") ax.plot(rec_noise["t_samples"], rec_noise["y_samples"], "o", color="#C25B2A", ms=5, label=f"采样点(SNR {snr_db} dB,{rec_noise['bits']} bit)") fn = rec_noise["fit"] ax.plot(t_fit, fn["A"] * np.exp(-t_fit / fn["tau"]) + fn["B"], "r--", lw=1.5, label=f"NLS 拟合:τ̂ = {fn['tau']:.3f} ns") ax.set_xlabel("时间 (ns)") ax.set_ylabel("归一化强度") ax.set_title(f"M0 加噪单次:SNR = {snr_db} dB,12 bit 量化") ax.legend() ax.grid(alpha=0.3) # ---------- 图 3:重复实验 τ̂ 分布 ---------- fig3, ax = plt.subplots(figsize=(7, 4.2)) ax.hist(reps["tau_hats"], bins=max(10, repeats // 2), color="#2D5F8A", alpha=0.75, edgecolor="white") ax.axvline(reps["tau_true"], color="#C25B2A", ls="--", lw=2, label=f"真值 τ = {reps['tau_true']:.1f} ns") ax.axvline(reps["mean"], color="#1A5276", ls=":", lw=2, label=f"均值 {reps['mean']:.3f} ns") ax.set_xlabel("拟合寿命 τ̂ (ns)") ax.set_ylabel("次数") ax.set_title(f"M0 重复 {repeats} 次:相对标准差 {reps['rel_std'] * 100:.2f}%," f"偏差 {reps['bias'] * 100:+.2f}%") ax.legend() ax.grid(alpha=0.3) # ---------- 落盘记录包 ---------- run_dir = record.save_run( runs_dir, "exp01_m0_baseline", records=[rec_ideal, rec_noise, reps], figures=[("fig1_ideal_fit", fig1), ("fig2_noisy_fit", fig2), ("fig3_repeat_hist", fig3)], notes=(f"M0 基准:①理想无噪恢复 τ={tau} ns(偏差 {rec_ideal['rel_error']:.2e});" f"②SNR {snr_db} dB 重复 {repeats} 次分布:均值 {reps['mean']:.3f} ns、" f"相对标准差 {reps['rel_std'] * 100:.2f}%。"), seed=seed, ) print(f"[M0] 理想无噪:τ̂ = {rec_ideal['fit']['tau']:.6f} ns " f"(设定 {tau},相对偏差 {rec_ideal['rel_error']:.2e})") print(f"[M0] 加噪单次:τ̂ = {rec_noise['fit']['tau']:.3f} ns") print(f"[M0] 重复 {repeats} 次:均值 {reps['mean']:.3f} ns," f"σ(τ̂)/τ = {reps['rel_std'] * 100:.2f}%,bias = {reps['bias'] * 100:+.2f}%") print(f"[M0] 记录包:{run_dir}") return run_dir if __name__ == "__main__": ap = argparse.ArgumentParser(description="M0 基准实验") ap.add_argument("--tau", type=float, default=52.8) ap.add_argument("--window-ns", type=float, default=400.0) ap.add_argument("--snr-db", type=float, default=50.0) ap.add_argument("--bits", type=int, default=12) ap.add_argument("--repeats", type=int, default=20) ap.add_argument("--seed", type=int, default=42) ap.add_argument("--runs-dir", default="runs") a = ap.parse_args() run(tau=a.tau, window_ns=a.window_ns, snr_db=a.snr_db, bits=a.bits, repeats=a.repeats, seed=a.seed, runs_dir=a.runs_dir)