Python中Cross Correlation信号同步异常问题求助

我尝试对两个形状相同但长度不同的信号做同步，采用互相关（Cross Correlation）方法后得到的最佳滞后约为269000，但实际能让信号完美对齐的最佳滞后应为129000，无法确定互相关失效的原因。

实现代码 #

def sync_signals(self):
        accel_path = self.accel_path_text.text()

        samrate, data = wavfile.read(str(accel_path))
        accelData = pd.DataFrame(data)

        rafale_path = self.rafale_path_text.text()
        rafaleData = pd.read_csv(rafale_path, delimiter=';', header="infer")

        stat_path = self.stat_path_text.text()
        df = pd.read_csv(stat_path)

        start_time = df.iloc[1][2]
        end_time = df.iloc[1][3]
        measurement_time = int(end_time)-int(start_time)

        temps_entre_points_accel = measurement_time/(len(accelData))
        accelData.insert(0, 'Heure', accelData.index * temps_entre_points_accel)
        
        def str_to_milliseconds(time_str):
            # 拆分时间字符串组件
            hours, minutes, seconds, milliseconds = map(int, time_str.split(':'))

            # 计算总毫秒数
            total_milliseconds = (hours * 3600 + minutes * 60 + seconds) * 1000 + milliseconds

            return total_milliseconds
        
        rafaleData.insert(0, 'Temps', [str_to_milliseconds(time_str) for time_str in rafaleData.iloc[:, 0]])

        # 将信号转为numpy数组
        accel_signal = accelData.values[:, 1]
        accel_time = accelData.values[:, 0]
        rafale_signal = rafaleData['GAMMAZ (g)'].values
        rafale_time = rafaleData['Temps'].values

        # 判断哪个信号更短
        if len(accel_signal) < len(rafale_signal):
            short_signal = accel_signal
            short_time = accel_time
            long_signal = rafale_signal
            long_time = rafale_time
        else:
            short_signal = rafale_signal
            short_time = rafale_time
            long_signal = accel_signal
            long_time = accel_time

        long_time_simulated = np.arange(short_time[0],short_time[len(short_time)-1],20)
        
        interpolation_function = interp1d(short_time, short_signal, kind='linear')
        resampled_signal = interpolation_function(long_time_simulated)

        print(pd.DataFrame(long_time))
        print(pd.DataFrame(long_time_simulated))

        shifted_versions = [
            (pd.DataFrame(long_signal*0.00390625), pd.DataFrame(resampled_signal))
        ]

        print(shifted_versions)
        def shift_for_maximum_correlation(x, y):
            correlation = correlate(x, y, mode="full")
            lags = correlation_lags(x.size, y.size, mode="full")
            lag = lags[np.argmax(correlation)]
            print(f"Best lag: {lag}")
            if lag < 0:
                y = y.iloc[abs(lag):].reset_index(drop=True)
            else:
                x = x.iloc[lag:].reset_index(drop=True)
            return x, y
        
        def plot_correlation(x, y, text):
            # 绘制对比图
            plt.subplots(figsize=(10, 6))
            plt.plot(x)
            plt.plot(y)
            plt.legend(loc="best")
            plt.show()

        for x, y in shifted_versions:
            shifted_x, shifted_y = shift_for_maximum_correlation(x, y)
            plot_correlation(shifted_x, shifted_y, text="after shifting")

相关数据 #

long_time（长信号时间轴） #

0             0.0
1            20.0
2            40.0
3            60.0
4            80.0
...           ...
475245  9504900.0
475246  9504920.0
475247  9504940.0
475248  9504960.0
475249  9504980.0

long_time_simulated（重采样后的短信号时间轴） #

0      1742887
1      1742907
2      1742927
3      1742947
4      1742967
...        ...
86236  3467607
86237  3467627
86238  3467647
86239  3467667
86240  3467687

long_signal（长信号数据） #

0       0.382812
1       0.371094
2       0.425781
3       0.410156
4       0.503906
...          ...
475245 -0.074219
475246 -0.078125
475247 -0.062500
475248 -0.046875
475249 -0.046875

resampled_signal（重采样后的短信号数据） #

0      0.935748
1      0.935736
2      0.935724
3      0.935712
4      0.935700
...         ...
86236  0.164273
86237  0.143966
86238  0.123658
86239  0.103350
86240  0.083042

同步结果 #

• 当前互相关计算得到的同步信号未对齐
• 无法达到预期的完美同步效果
• 原始未同步信号可作为参考对比

已尝试的排查措施 #

• 对含噪声的长信号进行滤波处理，无改善
• 更换互相关的不同计算模式，结果依旧不符合预期

内容的提问来源于stack exchange，提问作者gabriel bastier