lag threshold influence values

[nervusvagus]

I have a sinewave from mpu9250 sensor which I need to calculate the max of during live sensor readings. I'm trying to understand how lag threshold influence values work in your library and what i should set them to. is there any other sources? In example they are peakDetection.begin(48, 3, 0.6); and have no idea why.

[frankcohen]

Just another user here... The library source is 142 lines long. I'd look there for how it works. And surprisingly ChatGPT answered my question "Thanks. A sample of the calculated magnitude values is: "99
305
-487
-125
112
-2457
-1204
638
861
-1032
-61
162
9
443
150
-361
-305
-216
50
-675
568
35
-348
6". What values should I use for lag, threshold, influence?" with this answer:

To determine the optimal values for lag, threshold, and influence, we need to analyze the provided data. Here's a breakdown of how to choose these parameters:

1. Parameter Descriptions
   lag: Number of previous samples used for smoothing and calculating averages and standard deviations. It determines the window size for trend detection.
   threshold: The sensitivity multiplier for detecting peaks. Higher values make peak detection stricter, reducing false positives.
   influence: The weight given to new data when a peak is detected. Lower values make the algorithm less influenced by peaks and stabilize future trend calculations.
2. **Analyzing the Data
   Given your sample values, let’s use the following guidelines:

Lag:
Based on your data, where peaks occur relatively quickly but some noise exists, start with a small window (e.g., 10 samples) to balance responsiveness and smoothing.
T
To determine the optimal values for lag, threshold, and influence based on the sample values you provided, let's analyze them step by step.

1. Understanding the Parameters
   lag: The size of the sliding window used for smoothing. It defines how many previous samples are considered to compute the moving average and standard deviation.
   threshold: Controls how significant a deviation must be to qualify as a peak. A higher value reduces false positives.
   influence: Determines how much detected peaks influence the trend calculation. A lower value reduces the impact of peaks on subsequent calculations.
2. Analyzing Your Data
   Your data contains large variations, with positive and negative values ranging from -2457 to 861. This suggests:

Lag: Peaks occur relatively quickly, so a lag of around 10 samples should work well.
Threshold: Given the wide range, a threshold of 2 to 3 (2-3 times the standard deviation) can help detect significant deviations.
Influence: To stabilize the trend while still adapting to peaks, set the influence to around 0.5.
3. Recommended Starting Values
Based on your data:

lag = 10: A small window for responsive peak detection.
threshold = 2.5: Moderate sensitivity to reduce noise-induced peaks.
influence = 0.5: Balanced impact of peaks on future calculations.
4. Code Implementation
Here’s how to initialize and test the algorithm with your data:

cpp
Copy code
#include
#include
#include
#include "PeakDetection.h"

std::vector magnitudes = {
99, 305, -487, -125, 112, -2457, -1204, 638, 861,
-1032, -61, 162, 9, 443, 150, -361, -305, -216,
50, -675, 568, 35, -348, 6
};

void setup() {
Serial.begin(115200);

// Initialize peak detection with chosen parameters
PeakDetection peakDetector;
peakDetector.begin(10, 2.5, 0.5); // lag=10, threshold=2.5, influence=0.5

for (double magnitude : magnitudes) {
    double filteredValue = peakDetector.add(magnitude);
    int peak = peakDetector.getPeak();

    Serial.print("Magnitude: ");
    Serial.print(magnitude);
    Serial.print(" | Filtered: ");
    Serial.print(filteredValue, 2);
    Serial.print(" | Peak: ");
    if (peak == 1) {
        Serial.println("Positive Peak Detected!");
    } else if (peak == -1) {
        Serial.println("Negative Peak Detected!");
    } else {
        Serial.println("No Peak");
    }
}

}

void loop() {
// Do nothing
}
5. Fine-Tuning
If the algorithm detects too many false positives or misses peaks:

Increase threshold to reduce false positives.
Decrease lag for quicker responsiveness.
Adjust influence to balance the impact of peaks.
These values should provide a good starting point for detecting peaks in your magnitude data. Let me know if you'd like help tuning further or analyzing the results!

[frankcohen]

By the way, ChatGPT's answer says to use "peakDetector.begin(10, 2.5, 0.5);" which is wrong, since the threshold is an int. ChatGPT is helpful and not perfect.