Correlation and cross-correlation in Signal Processing

Introduction

Correlation helps us find how two signals or data sets move together. Cross-correlation shows how one signal matches or shifts compared to another.

To check if two sensors are measuring similar patterns over time.

To find the delay between two audio signals recorded at different places.

To detect repeating patterns in a noisy signal by comparing with a known pattern.

To measure similarity between two time series in finance or weather data.

To align images or signals by finding the best matching position.

Syntax

Signal Processing

cross_correlation = np.correlate(signal1, signal2, mode='full')

Use mode='full' to get correlation at all possible shifts.

The result shows similarity values; the highest value indicates best alignment.

Examples

Calculate cross-correlation of two short signals and print the similarity values at different shifts.

Signal Processing

import numpy as np
signal1 = [1, 2, 3]
signal2 = [0, 1, 0.5]
corr = np.correlate(signal1, signal2, mode='full')
print(corr)

Calculate auto-correlation of a signal to see how it matches with itself at different shifts.

Signal Processing

import numpy as np
signal = [1, 2, 3, 2, 1]
auto_corr = np.correlate(signal, signal, mode='full')
print(auto_corr)

Sample Program

This program calculates the cross-correlation between two signals. It prints the correlation values and the shift where the signals match best. It also shows plots of the signals and their cross-correlation for easy understanding.

Signal Processing

import numpy as np
import matplotlib.pyplot as plt

# Two simple signals
signal1 = np.array([1, 2, 3, 4, 5])
signal2 = np.array([0, 1, 0.5, 0, 0])

# Compute cross-correlation
cross_corr = np.correlate(signal1, signal2, mode='full')

# Find the shift with maximum correlation
max_index = np.argmax(cross_corr)
shift = max_index - (len(signal2) - 1)

print(f"Cross-correlation values: {cross_corr}")
print(f"Maximum correlation at shift: {shift}")

# Plot signals and cross-correlation
plt.figure(figsize=(10, 6))
plt.subplot(3, 1, 1)
plt.title('Signal 1')
plt.stem(signal1, use_line_collection=True)
plt.subplot(3, 1, 2)
plt.title('Signal 2')
plt.stem(signal2, use_line_collection=True)
plt.subplot(3, 1, 3)
plt.title('Cross-correlation')
lags = np.arange(-len(signal2) + 1, len(signal1))
plt.stem(lags, cross_corr, use_line_collection=True)
plt.xlabel('Lag')
plt.tight_layout()
plt.show()

Important Notes

Correlation values can be positive or negative depending on signal similarity.

Cross-correlation helps find time delay between signals.

Auto-correlation is a special case where a signal is correlated with itself.

Summary

Correlation measures how two signals relate or match.

Cross-correlation finds similarity at different shifts or delays.

Use cross-correlation to detect delays, patterns, or align signals.

Practice

1. What does the cross-correlation between two signals primarily measure?

easy

A. The similarity between the signals at different time shifts

B. The sum of the signals' amplitudes

C. The difference in frequency between the signals

D. The product of the signals' maximum values

Solution

1. Step 1: Understand cross-correlation concept

   Cross-correlation compares two signals by sliding one over the other to find similarity at different shifts.

2. Step 2: Identify what cross-correlation measures

   It measures how well the signals match at each time delay, not just their sums or frequency differences.

3. Final Answer:

   The similarity between the signals at different time shifts -> Option A

4. Quick Check:

   Cross-correlation = similarity at shifts [OK]

Hint: Cross-correlation slides signals to find best match [OK]

Common Mistakes:

• Confusing correlation with frequency difference
• Thinking it sums amplitudes only
• Assuming it multiplies max values

2. Which Python function from NumPy computes the cross-correlation of two 1D signals correctly?

easy

A. numpy.convolve(signal1, signal2)

B. numpy.correlate(signal1, signal2, mode='full')

C. numpy.cross(signal1, signal2)

D. numpy.corrcoef(signal1, signal2)

Solution

1. Step 1: Identify NumPy functions for correlation

   numpy.correlate computes cross-correlation; numpy.cross is for vector cross product; numpy.convolve does convolution; numpy.corrcoef computes correlation coefficient matrix.

2. Step 2: Confirm correct function and parameters

   numpy.correlate with mode='full' returns cross-correlation at all shifts.

3. Final Answer:

   numpy.correlate(signal1, signal2, mode='full') -> Option B

4. Quick Check:

   Use numpy.correlate for cross-correlation [OK]

Hint: Use numpy.correlate with mode='full' for cross-correlation [OK]

Common Mistakes:

• Using numpy.cross which is vector product
• Confusing convolution with correlation
• Using numpy.corrcoef which gives correlation matrix

3. Given two signals x = [1, 2, 3] and y = [0, 1, 0.5], what is the cross-correlation output using numpy.correlate(x, y, mode='full')?

medium

A. [0.5, 2.5, 4, 3, 0]

B. [0, 1, 2, 3, 4]

C. [0.5, 2, 3.5, 3, 0]

D. [1, 2, 3, 0, 0]

Solution

1. Step 1: Calculate cross-correlation values manually

   Cross-correlation at lag -2: x[0]*y[2]=1*0.5=0.5; lag -1: x[0]*y[1]+x[1]*y[2]=1*1+2*0.5=1+1=2; lag 0: x[0]*y[0]+x[1]*y[1]+x[2]*y[2]=1*0+2*1+3*0.5=0+2+1.5=3.5; lag 1: x[1]*y[0]+x[2]*y[1]=2*0+3*1=3; lag 2: x[2]*y[0]=3*0=0

2. Step 2: Arrange results in full mode order

   Full mode output is [lag -2, -1, 0, 1, 2] = [0.5, 2, 3.5, 3, 0]

3. Final Answer:

   [0.5, 2, 3.5, 3, 0] -> Option C

4. Quick Check:

   Cross-correlation values match calculated array [OK]

Hint: Calculate lag by lag multiplying overlapping elements [OK]

Common Mistakes:

• Mixing convolution with correlation
• Incorrect lag order in output
• Forgetting zero padding in full mode

4. You run numpy.correlate(signal1, signal2) but get a shorter output than expected. What is the likely cause?

medium

A. The function requires 2D arrays, not 1D

B. signal1 and signal2 must be the same length

C. You need to normalize signals before correlation

D. The default mode is 'valid', which returns only overlapping parts

Solution

1. Step 1: Check default mode of numpy.correlate

   By default, numpy.correlate uses mode='valid', which returns only positions where signals fully overlap, resulting in shorter output.

2. Step 2: Understand effect of mode on output length

   To get full cross-correlation including partial overlaps, mode='full' must be specified.

3. Final Answer:

   The default mode is 'valid', which returns only overlapping parts -> Option D

4. Quick Check:

   Default mode='valid' shortens output [OK]

Hint: Specify mode='full' for complete cross-correlation output [OK]

Common Mistakes:

• Assuming signals must be same length
• Thinking normalization affects output length
• Believing function needs 2D arrays

5. You have two noisy signals recorded with a delay. How can cross-correlation help you find the delay between them?

hard

A. By finding the lag where cross-correlation is maximum, indicating best alignment

B. By subtracting the signals and measuring the difference

C. By computing the Fourier transform and comparing magnitudes

D. By averaging the signals and checking the mean value

Solution

1. Step 1: Understand cross-correlation use for delay detection

   Cross-correlation slides one signal over the other and measures similarity at each lag.

2. Step 2: Identify lag with maximum cross-correlation

   The lag where cross-correlation peaks shows the delay that best aligns the two signals despite noise.

3. Final Answer:

   By finding the lag where cross-correlation is maximum, indicating best alignment -> Option A

4. Quick Check:

   Max cross-correlation lag = signal delay [OK]

Hint: Max cross-correlation lag reveals signal delay [OK]

Common Mistakes:

• Using subtraction which ignores alignment
• Confusing frequency domain methods here
• Averaging signals loses timing info