Kronts
Sign Up Sign In

Channel Correlations: Creating Realistic Multi-Sensor Data

Channel correlations allow you to define statistical relationships between channels in multi-channel time series, essential for creating realistic data from multi-sensor systems where measurements influence each other.

Prerequisites

What Are Channel Correlations?

Correlation measures how two variables move together. In multi-sensor data, correlations arise from: - Shared environment (temperature sensors in same room) - Physical coupling (pressure and temperature in closed system) - Common causes (vibration axes on same machine) - Process dependencies (upstream affects downstream)

Correlation Coefficient Range

Correlation values range from -1.0 to +1.0:

+1.0  Perfect positive correlation
      When one increases, other increases proportionally

+0.7  Strong positive correlation
      Generally move together, some independence

+0.3  Weak positive correlation
      Slight tendency to move together

 0.0  No correlation
      Completely independent

-0.3  Weak negative correlation
      Slight tendency to move opposite

-0.7  Strong negative correlation
      Generally move in opposite directions

-1.0  Perfect negative correlation
      When one increases, other decreases proportionally

Why Correlations Matter

Without Correlations (default): - Each channel is statistically independent - Unrealistic for sensors measuring related phenomena - Example: 3 temperature sensors in same room showing completely different patterns

With Correlations: - Channels move together realistically - Mimics real multi-sensor behavior - Example: 3 temperature sensors in same room showing similar patterns with local variations

How Phoenix Applies Correlations

Phoenix uses Cholesky decomposition to apply correlations mathematically:

  1. Generate independent channels with configured properties
  2. Normalize each channel (zero mean, unit variance)
  3. Build correlation matrix from your specifications
  4. Apply Cholesky transformation to create correlations
  5. Denormalize back to original scale and variance

This ensures: - Correlations are mathematically exact - Original signal properties (mean, variance) are preserved - Only the noise components are correlated

Configuring Channel Correlations

Step 1: Create Multiple Channels

Correlations require at least 2 channels:

  1. Navigate to "Channels" section
  2. Click "Add Channel" to create Channel 1
  3. Click "Add Channel" again for Channel 2
  4. Configure each channel's properties

The "Channel Correlations" section appears automatically when you have 2+ channels.

Step 2: Add a Correlation

  1. Scroll to "Channel Correlations" section
  2. Click "Add Correlation" button
  3. A new correlation entry appears with:
  4. Channel A dropdown
  5. Channel B dropdown
  6. Correlation value input
  7. Configure the correlation (see below)

[Screenshot Required: Channel Correlations Interface] 1. With 3 channels configured 2. Add 2 correlations 3. Capture: Channel Correlations section showing correlation list 4. Purpose: Show correlation configuration UI

Step 3: Select Channel Pair

Choose which two channels to correlate:

Channel A Dropdown: Select first channel Channel B Dropdown: Select second channel

Important: - Cannot correlate a channel with itself - Order doesn't matter (A-B same as B-A) - Each pair should only be defined once

Example:

Channel 1 ↔ Channel 2: Valid
Channel 2 ↔ Channel 1: Same as above (don't add both)
Channel 1 ↔ Channel 1: Invalid (channel with itself)

Step 4: Set Correlation Value

Enter correlation coefficient between -1.0 and +1.0:

Choosing Values: - 0.9 to 1.0: Nearly identical sensors/conditions - 0.7 to 0.9: Strong relationship, common in physics - 0.4 to 0.7: Moderate relationship - 0.1 to 0.4: Weak relationship - -0.4 to 0.1: Little to no relationship - -0.7 to -0.4: Moderate inverse relationship - -0.9 to -0.7: Strong inverse relationship - -1.0 to -0.9: Nearly perfect inverse

Step 5: Add More Correlations (Optional)

For N channels, you can define up to N×(N-1)/2 unique pair correlations:

2 channels: 1 possible correlation
3 channels: 3 possible correlations
4 channels: 6 possible correlations
5 channels: 10 possible correlations
10 channels: 45 possible correlations

You don't need to define all possible pairs. Undefined pairs default to correlation = 0 (independent).

Step 6: Validation

Phoenix validates your correlation matrix:

Automatic Checks: - Values between -1.0 and +1.0 - No self-correlations - Matrix is "positive semi-definite" (mathematically valid)

Positive Semi-Definite Requirement: Some correlation combinations are mathematically impossible. Phoenix prevents these.

Example of Invalid Matrix:

A-B: +0.9
B-C: +0.9
A-C: -0.9

This is impossible: if A and B move together (+0.9),
and B and C move together (+0.9), then A and C must
also move somewhat together, not opposite (-0.9).

Phoenix will show an error if your correlations create an impossible matrix.

[Screenshot Required: Correlation Validation Error] 1. Configure impossible correlations (e.g., above example) 2. Attempt to preview 3. Capture: Error message showing invalid correlation matrix 4. Purpose: Show validation feedback

Step 7: Preview with Correlations

Click "Preview" to generate correlated multi-channel data:

  1. Phoenix generates independent channels
  2. Applies correlation transformation
  3. Renders multi-channel chart
  4. Displays statistics per channel

Verifying Correlations in Chart: - Positively correlated channels move together (similar shapes) - Negatively correlated channels move opposite - Uncorrelated channels move independently - Zoom in to see correlation patterns more clearly

[Screenshot Required: Correlated Channels] 1. Configure 3 channels with high positive correlations (0.8-0.9) 2. Preview 3. Capture: Chart showing 3 traces moving together 4. Purpose: Demonstrate positive correlation visually

Common Correlation Patterns

High Positive Correlation (0.85+)

Use Cases: - Temperature sensors in same room - Pressure sensors on same vessel - Collocated environmental sensors

Example: 4 Temperature Sensors in Server Room

All sensors measure same ambient temperature with local variations

Correlations:
  Sensor 1 ↔ Sensor 2: 0.90
  Sensor 1 ↔ Sensor 3: 0.88
  Sensor 1 ↔ Sensor 4: 0.85
  Sensor 2 ↔ Sensor 3: 0.92
  Sensor 2 ↔ Sensor 4: 0.87
  Sensor 3 ↔ Sensor 4: 0.89

Interpretation: All sensors track room temperature closely
with small local differences

[Screenshot Instructions: Temperature Array] 1. Create 4 channels: - Names: "Sensor 1" through "Sensor 4" - Means: 22, 22.5, 21.8, 22.2 (slight differences) - Noise: All 0.4 - Trend: All 0.001 (slow warming) 2. Add correlations as above 3. Duration: 2 hours, Sampling: 0.01 Hz 4. Preview and capture 5. Purpose: Realistic correlated temperature array

Moderate Positive Correlation (0.4-0.7)

Use Cases: - Related but not identical processes - Partial physical coupling - Shared but not dominant influences

Example: Multi-Zone HVAC System

Different zones partially influence each other via air circulation

Correlations:
  Zone 1 ↔ Zone 2: 0.55 (adjacent zones)
  Zone 1 ↔ Zone 3: 0.32 (distant zones)
  Zone 2 ↔ Zone 3: 0.48 (adjacent zones)

Low/No Correlation (0-0.3)

Use Cases: - Independent measurements - Unrelated processes - Isolated sensors

Example: Multi-Parameter Process Monitor

Temperature, Flow, and Pressure measuring independent aspects

Correlations:
  Temperature ↔ Flow: 0.1 (weak relationship)
  Temperature ↔ Pressure: 0.15 (weak relationship)
  Flow ↔ Pressure: 0.05 (nearly independent)

Negative Correlation (-0.4 to -0.9)

Use Cases: - Inverse relationships - Compensating systems - Thermodynamic laws (pressure vs. volume)

Example: Thermal System with Cooling

As temperature increases, cooling valve opens more (inverse)

Correlations:
  Temperature ↔ Cooling: -0.75

Example: Seesaw/Balance

Two ends of a balance move opposite

Correlations:
  End A ↔ End B: -0.95

[Screenshot Instructions: Inverse Correlation] 1. Create 2 channels: - Channel 1: "Temperature", Mean=80, Noise=2, Trend=0.01 - Channel 2: "Cooling", Mean=50, Noise=1, Trend=-0.005 2. Correlation: -0.75 3. Duration: 1 hour, Sampling: 0.1 Hz 4. Preview and capture 5. Purpose: Show negative correlation pattern

Mixed Correlation Patterns

Example: 3-Axis Vibration with Partial Coupling

X and Y partially coupled (mechanical), Z independent (gravity)

Correlations:
  X ↔ Y: 0.35 (mechanical coupling)
  X ↔ Z: 0.05 (minimal)
  Y ↔ Z: 0.08 (minimal)

Advanced Correlation Techniques

Correlation vs. Channel Configuration

Important: Correlations affect only the noise components, not oscillations or trends.

Example:

Channel 1: Mean=100, Noise=5, Oscillation=10 Hz
Channel 2: Mean=200, Noise=3, Oscillation=20 Hz
Correlation: 0.8

Result:
- Noise components are 80% correlated
- Oscillations remain independent (10 Hz vs. 20 Hz)
- Means remain different (100 vs. 200)

To create fully correlated oscillations: 1. Use same frequency on both channels 2. Use same or similar amplitude 3. Use same phase 4. Add correlation for noise

Partial Correlation Networks

You don't need to define all pairwise correlations. Undefined pairs default to 0.

Example: Chain Pattern

Channels: A, B, C, D

Define only:
  A ↔ B: 0.7
  B ↔ C: 0.7
  C ↔ D: 0.7

Result:
  A ↔ D: ~0 (no direct correlation)
  A influenced by B, B by C, C by D (indirect chain)

Symmetric Correlation Matrices

Phoenix automatically ensures symmetry: - If you define A ↔ B: 0.5 - Then B ↔ A is automatically 0.5 - Don't define both directions

Troubleshooting

"Correlation matrix is not valid" error

Cause: Your correlation combination is mathematically impossible (not positive semi-definite)

Common Scenarios:

Scenario 1: Triangle Inequality Violation

Problem:
  A ↔ B: +0.9
  B ↔ C: +0.9
  A ↔ C: -0.8  ← Impossible

Fix: If A and B are similar (+0.9), and B and C are similar (+0.9),
then A and C must also be somewhat similar.
Change A ↔ C to +0.5 or higher.

Scenario 2: Extreme Values

Problem:
  A ↔ B: +0.99
  B ↔ C: +0.99
  C ↔ D: +0.99
  A ↔ D: 0.0  ← Unlikely to be valid

Fix: With such high pairwise correlations, A and D must also correlate.
Change A ↔ D to +0.9 or remove some correlations.

General Fix Strategy: 1. Remove all correlations 2. Add them back one at a time 3. Preview after each addition 4. Find which correlation causes the error 5. Adjust that correlation value

Can't see correlation effect in chart

Possible Causes:

  1. Correlation too weak (< 0.3)
  2. Weak correlations are subtle
  3. Zoom in to see patterns

  4. Or increase correlation value

  5. Noise too small compared to oscillations

  6. Correlation affects noise, not oscillations
  7. If noise is tiny, correlation effect is tiny

  8. Increase noise amplitude to see correlation

  9. Different channel scales obscure correlation

  10. Channel 1: range 0-10
  11. Channel 2: range 1000-1010
  12. Hard to see correlation visually

  13. Use similar scales or normalize mentally

  14. Duration too short

  15. Need enough data to see correlation patterns
  16. Try longer duration or more samples

Correlation in wrong direction

Check: 1. Verify sign: positive (+) vs. negative (-) 2. Confirm correct channels selected (A vs. B) 3. Preview shows expected relationship: - Positive: move together - Negative: move opposite

Cannot add correlation

Possible Causes:

  1. Only 1 channel exists
  2. Need at least 2 channels

  3. Add another channel first

  4. Maximum correlations reached

  5. For N channels: max N×(N-1)/2 correlations

  6. Delete unused channels to free space

  7. Duplicate pair

  8. Already defined A ↔ B
  9. Don't also define B ↔ A
  10. Delete one and keep the other

Correlation value won't accept input

Check: 1. Value must be between -1.0 and +1.0 2. Use decimal point (.) not comma (,) 3. Check for typos (1.5 invalid, 0.15 valid)

Best Practices

Start with Physical Understanding

Base correlations on real-world relationships: - Measure actual correlations if possible - Use domain knowledge (thermodynamics, mechanics) - Consult SMEs familiar with the system

Use Realistic Ranges

Too High (> 0.95): - Rarely occurs in real systems - Suggests sensors might be redundant - Use only for nearly identical conditions

Too Perfect (±1.0): - Never occurs in real data (always some noise) - Use 0.95-0.99 for very strong correlations

Too Precise (0.873): - False precision - Round to 1 decimal: 0.9 - Or 2 decimals max: 0.87

Don't Over-Correlate

You don't need to define every possible pair: - Define primary relationships - Let others default to 0 - Simpler is better for troubleshooting

Test Incrementally

  1. Start with no correlations (all independent)
  2. Add one correlation
  3. Preview and verify
  4. Add next correlation
  5. Repeat until done

Document Your Choices

When saving time series, include correlation info in description:

Good description:
"3-channel temp array, all correlated 0.85-0.9, simulates server room"

Poor description:
"Test data"

Correlation Examples by Industry

Manufacturing

Assembly Line Temperatures

Stations in sequence, each slightly warmer than previous

Station 1 ↔ Station 2: 0.75
Station 2 ↔ Station 3: 0.75
Station 3 ↔ Station 4: 0.75
Station 1 ↔ Station 3: 0.45 (indirect)

HVAC

Multi-Zone Building

Adjacent zones correlate more than distant zones

Zone 1 ↔ Zone 2: 0.68 (adjacent)
Zone 2 ↔ Zone 3: 0.65 (adjacent)
Zone 1 ↔ Zone 3: 0.42 (one zone apart)
Zone 1 ↔ Zone 4: 0.25 (far apart)

Chemical Process

Reactor with Controls

Temperature and pressure inversely related (thermodynamics)
Flow rate independently controlled

Temp ↔ Pressure: -0.65
Temp ↔ Flow: 0.15
Pressure ↔ Flow: -0.10

Vibration Monitoring

3-Axis on Rotating Equipment

X and Y partially coupled (rotation plane)
Z more independent (axial direction)

X ↔ Y: 0.40
X ↔ Z: 0.10
Y ↔ Z: 0.12

Summary

Channel correlations transform independent multi-channel data into realistic multi-sensor datasets:

  • Correlation range: -1.0 (perfect inverse) to +1.0 (perfect direct)
  • Affects: Noise components (not oscillations/trends)
  • Validates: Must be mathematically possible (positive semi-definite)
  • Common values: 0.7-0.9 (strong), 0.4-0.7 (moderate), 0-0.3 (weak)
  • Best practice: Base on physical understanding, test incrementally

Next Steps

Mastering correlations enables creation of highly realistic multi-sensor test data that accurately represents complex industrial systems.