Averaging an analog value.

paulw2 · February 6, 2020, 4:32pm

Hi There,

I'm measuring a analog input value which in turn converts and display a voltage qualtity to my display. I found that the measurement is mostly correct in a range but fluctate a lot.

I've implemented signal smoothing as per example below.

My question is , how do I truncate the sample value? I want to discard the highest and lowest value in the sample and use the rest for an average calc. Any advise would be appreciated.

Thanks

//
// Helper function to smooth out the light sensor data.
// Takes any number of readings and smoothes out the data to an average value.
//
// Returns 8-bit value (0-255).
//
int smooth(){
  int i;
  int value = 0;
  int numReadings = 10;

  for (i = 0; i < numReadings; i++){
    // Read light sensor data.
    value = value + analogRead(sensor);

    // 1ms pause adds more stability between reads.
    delay(1);
  }

  // Take an average of all the readings.
  value = value / numReadings;

}

dustin02rsx · February 6, 2020, 4:42pm

First I'll say that I wouldn't throw out data. I'd take more samples and average more samples.

With that said, to do what you're asking, I would store all the analogRead() values in an array, then for loop through the array to find the the highest and lowest value.

From there it would just be a matter of calculating the average of all the indices in the array (except the Max/min).

I'd do something like this (completely untested, just for illustration)

int sampleSize = 10;
int sampleRate = 10;


int myArray[sampleSize];
float average = 0;
int maxVal = 0; //set low so any value read will be higher
int minVal = 1024; //set arbitrarily high so any value read will be lower




for(int i=0; i < sampleSize; i++)
{
myArray[i] = analogRead(0);

if(myArray[i] >= maxVal)
maxVal = myArray[i];

if(myArray[i] <= minVal)
minVal = myArray[i];

delay(sampleRate);
}


for(int i; i < sampleSize; i++)
{
sum = sum + myArray[i];
}

average = (sum - (maxVal+minVal))/(sampleSize-2);

paulw2 · February 6, 2020, 5:32pm

I will play around with multiple average of samples and give your idea a try. ...did not think about that.

I did notice that there pops in a couple of low and high values over a decent size sample. So I dont think throwing out one high and one low will make thay a big differnce in say a sample of 50 readings?

jremington · February 6, 2020, 7:38pm

Look up "median filter".

This can be very effective in cases where occasional large outliers cause conventional smoothing to fail, but requires constant sorting of the input data.

Idahowalker · February 6, 2020, 8:13pm

I use the Simple Kalman Filter library GitHub - denyssene/SimpleKalmanFilter: A basic implementation of Kalman Filter for single variable models., which may be of interest to you.

An example of use:

void TrackSun( void * pvParameters )
{
  int Altitude = 1500;
  int Azimuth = 1500;
  int maxAltitudeRange = 2144;
  int minAltitudeRange = 900;
  int maxAzimuthRange = 2144;
  int minAzimuthRange = 900;
  SimpleKalmanFilter kfAltitude0( 10.0, 10.0, .01 ); // kalman filter Altitude 0
  SimpleKalmanFilter kfAltitude1( 10.0, 10.0, .01 ); // kalman filter Altitude 1
  SimpleKalmanFilter kfAzimuth0( 5.0, 10.0, .01 ); // kalman filter Azimuth 0
  SimpleKalmanFilter kfAzimuth1( 5.0, 10.0, .01 ); // kalman filter Azimuth 1
  float filteredAltitude_0 = 0.0f;
  float filteredAltitude_1 = 0.0f;
  float filteredAzimuth_0 = 0.0f;
  float filteredAzimuth_1 = 0.0f;
  int64_t AzimuthEndTime = esp_timer_get_time();
  int64_t AzimuthStartTime = esp_timer_get_time(); //gets time in uSeconds like Arduino Micros,
  int64_t AltitudeEndTime = esp_timer_get_time();
  int64_t AltitudeStartTime = esp_timer_get_time();

  while (1)
  {
    //Altitude
    AltitudeEndTime = esp_timer_get_time() - AltitudeStartTime; // produce elasped time for the simpleKalmanFilter
    kfAltitude0.setProcessNoise( (float)AltitudeEndTime / 1000000.0f ); //convert time of process to uS, update SimpleKalmanFilter q
    kfAltitude1.setProcessNoise( (float)AltitudeEndTime / 1000000.0f ); //convert time of process to uS, update SimpleKalmanFilter q
    filteredAltitude_0 = kfAltitude0.updateEstimate( (float)adc1_get_raw(ADC1_CHANNEL_3) );
    filteredAltitude_1 = kfAltitude1.updateEstimate( (float)adc1_get_raw(ADC1_CHANNEL_0) );
    if ( filteredAltitude_0 > filteredAltitude_1 )
    {
      Altitude -= 1;
      if ( Altitude < minAltitudeRange )
      {
        Altitude = 1500;
      }
      fMoveAltitude( Altitude );
      vTaskDelay( 12 );
      AltitudeStartTime = esp_timer_get_time();
    }
    if ( filteredAltitude_0 < filteredAltitude_1 )
    {
      Altitude += 1;
      if ( Altitude >= maxAltitudeRange )
      {
        Altitude = 1500;
      }
      fMoveAltitude( Altitude );
      vTaskDelay( 12 );
      AltitudeStartTime = esp_timer_get_time();
    }
    //// AZIMUTH
    AzimuthEndTime = esp_timer_get_time() - AzimuthStartTime; // produce elasped time for the simpleKalmanFilter
    kfAzimuth0.setProcessNoise( (float)AzimuthEndTime / 1000000.0f ); //convert time of process to uS, update SimpleKalmanFilter q
    kfAzimuth1.setProcessNoise( (float)AzimuthEndTime / 1000000.0f ); //convert time of process to uS, update SimpleKalmanFilter q
    filteredAzimuth_0 = kfAzimuth0.updateEstimate( (float)adc1_get_raw(ADC1_CHANNEL_5) );
    filteredAzimuth_1 = kfAzimuth1.updateEstimate( (float)adc1_get_raw(ADC1_CHANNEL_6) );
    if ( filteredAzimuth_0 > filteredAzimuth_1 )
    {
      Azimuth += 1;
      if ( Azimuth >= maxAzimuthRange )
      {
        Azimuth = 900;
      }
      fMoveAzimuth( Azimuth );
      vTaskDelay( 12 );
      AzimuthStartTime = esp_timer_get_time();
    }
    if ( filteredAzimuth_0 < filteredAzimuth_1 )
    {
      Azimuth -= 1;
      if ( (Azimuth >= maxAzimuthRange) || (Azimuth <= minAzimuthRange) )
      {
        Azimuth = 900;
      }
      fMoveAzimuth( Azimuth );
      vTaskDelay( 12 );
      AzimuthStartTime = esp_timer_get_time();
    }
    // Serial.println();
    vTaskDelay( 30 );
    //
  } // while(1)
} //void TrackSun()

Instead of using a static time, I use a dynamic time, which works much better.

PieterP · February 6, 2020, 10:16pm

In my experience, using a cheap exponential moving average filter followed by a hysteresis stage works very well to eliminate the fluctuations in an analog reading.

You can find an implementation and an example as part of the Arduino Filters library:
https://tttapa.github.io/Arduino-Filters/Doxygen/d3/dbe/1_8FilteredAnalog_8ino-example.html

[color=#434f54]// Include the library[/color]
[color=#5e6d03]#include[/color] [color=#434f54]<[/color][b][color=#d35400]Filters[/color][/b][color=#434f54].[/color][color=#000000]h[/color][color=#434f54]>[/color]
 
[color=#5e6d03]#include[/color] [color=#434f54]<[/color][b][color=#d35400]AH[/color][/b][color=#434f54]/[/color][color=#000000]Hardware[/color][color=#434f54]/[/color][b][color=#d35400]FilteredAnalog[/color][/b][color=#434f54].[/color][color=#000000]hpp[/color][color=#434f54]>[/color]
[color=#5e6d03]#include[/color] [color=#434f54]<[/color][b][color=#d35400]AH[/color][/b][color=#434f54]/[/color][color=#000000]Timing[/color][color=#434f54]/[/color][color=#000000]MillisMicrosTimer[/color][color=#434f54].[/color][color=#000000]hpp[/color][color=#434f54]>[/color]
 
[color=#434f54]// Create a filtered analog object on pin A0, with the default settings:[/color]
[b][color=#d35400]FilteredAnalog[/color][/b][color=#434f54]<[/color][color=#434f54]>[/color] [color=#000000]analog[/color] [color=#434f54]=[/color] [color=#000000]A0[/color][color=#000000];[/color]
 
[color=#00979c]void[/color] [color=#5e6d03]setup[/color][color=#000000]([/color][color=#000000])[/color] [color=#000000]{[/color]
  [b][color=#d35400]Serial[/color][/b][color=#434f54].[/color][color=#d35400]begin[/color][color=#000000]([/color][color=#000000]115200[/color][color=#000000])[/color][color=#000000];[/color]
  [color=#5e6d03]while[/color] [color=#000000]([/color][color=#434f54]![/color][b][color=#d35400]Serial[/color][/b][color=#000000])[/color]
    [color=#000000];[/color]
  [color=#434f54]// Select the correct ADC resolution[/color]
  [color=#000000]analog[/color][color=#434f54].[/color][color=#d35400]setupADC[/color][color=#000000]([/color][color=#000000])[/color][color=#000000];[/color]
[color=#000000]}[/color]
 
[color=#00979c]void[/color] [color=#5e6d03]loop[/color][color=#000000]([/color][color=#000000])[/color] [color=#000000]{[/color]
  [color=#434f54]// Read the analog input every millisecond, and print if the value has changed[/color]
  [color=#00979c]static[/color] [b][color=#d35400]Timer[/color][/b][color=#434f54]<[/color][color=#d35400]millis[/color][color=#434f54]>[/color] [color=#000000]timer[/color] [color=#434f54]=[/color] [color=#000000]1[/color][color=#000000];[/color] [color=#434f54]// ms[/color]
  [color=#5e6d03]if[/color] [color=#000000]([/color][color=#000000]timer[/color] [color=#434f54]&&[/color] [color=#000000]analog[/color][color=#434f54].[/color][color=#d35400]update[/color][color=#000000]([/color][color=#000000])[/color][color=#000000])[/color]
    [b][color=#d35400]Serial[/color][/b][color=#434f54].[/color][color=#d35400]println[/color][color=#000000]([/color][color=#000000]analog[/color][color=#434f54].[/color][color=#d35400]getValue[/color][color=#000000]([/color][color=#000000])[/color][color=#000000])[/color][color=#000000];[/color]
[color=#000000]}[/color]

Pieter

dave-in-nj · February 6, 2020, 11:00pm

Take a look at the inputs.
If they are all over the place then question the value of the signal.
You cannot fix a bad sensor in software.
Also consider signal filters before the ADC.

Wawa · February 6, 2020, 11:09pm

paulw2:
I'm measuring a analog input value which in turn converts and display a voltage qualtity to my display.

What sensor you're using.

Measuring a 'voltage' with a ratiometric A/D (what you're code is doing) is fundamentally wrong.
Measuring a ratiometric sensor with a ratiometric A/D, and convert it into a voltage (which you can't do with simple code like that) is also wrong.

Give us all the facts, including schematic/pictures/powering/etc. of the setup.
All the advice given so far is impressive, but could be wrong for your sensor.
Leo..

paulw2 · February 7, 2020, 5:37am

Hi Im using a simple voltage divider with low %tolerances.

Wawa · February 7, 2020, 5:53am

paulw2:
Hi Im using a simple voltage divider with low %tolerances.

A voltage divider to bring a higher voltage down to what the Arduino can measure?

If you measure a voltage, then you must use (one of) the internal reference voltage(s) for a stable result.
Default Aref measures a ratio (of the 5volt supply).
If the 5volt supply 'wobbles', so will the end result.

Low tolerance resistors means nothing.
Aref could be 5% or more off, and that usually sets the final tolerance.

Now tell us what you're really trying to do.
Leo..

dave-in-nj · February 7, 2020, 10:48am

What generates the signal?
What is the full scale of the signal?
Do you have a data sheet for the device that generates the signal?
This is where you should be giving excessive details. Data sheets , links to the device etc.

Also schematic of your circuit. Measuring an AC waveform with a simple voltage divider?

Topic		Replies	Views
Calculating average of analog read values Programming	3	45951	May 5, 2021
Smoothing example questions Sensors	3	883	May 6, 2021
ADC Average example query Programming	6	2489	March 23, 2022
Averaging AnalogIn values / Processing + Arduino Interfacing	4	1714	May 6, 2021
Smoothing analog sensor inputs - double arrays? Syntax & Programs	11	7237	May 6, 2021

Averaging an analog value.

Related topics