Detailed Description

Simple one dimensional linear Kalman filter operator

This CDP operator can be used to filter measurement noise or inaccuracies using Kalman filter mathematics.

Arguments

When operator is used inside a signal its default input is automatically tied to signal's InternalValue or previous operator's output. Its default output is automatically tied to next operator's input or to signal's Value. See also CDP Operator Usage In CDP Signals.

Example

Below is an example plot of noisy signal (cyan) and the same signal with KalmanFilter operator applied (orange):

Actual Processing Code of the KalmanFilter

unsigned int KalmanFilter<T>::Process()
{
  // Set process noise covariance
  double procNoiseCov = 0.0;
  if (d->procNoiseCov)
  {
    CDPVariantValue procNoiseCov_v = d->procNoiseCov->GetVariantValue();
    procNoiseCov = procNoiseCov_v.GetConvertedValue<T>();
  }
  if (procNoiseCov != 0.0)
    d->SetIdentity2x2(d->KF.processNoiseCov, procNoiseCov);
  else
    d->SetIdentity2x2(d->KF.processNoiseCov, d->procNoiseCovDefault);

  // Set measurement noise covariance
  double measNoiseCov = 0.0;
  if (d->measNoiseCov)
  {
    CDPVariantValue measNoiseCov_v = d->measNoiseCov->GetVariantValue();
    measNoiseCov = measNoiseCov_v.GetConvertedValue<T>();
  }
  if (measNoiseCov != 0.0)
    d->KF.measurementNoiseCov(0) = measNoiseCov;
  else
    d->KF.measurementNoiseCov(0) = d->measNoiseCovDefault;

  // Set estimation error covariance (on first run only, because it starts to correct at every cycle)
  if (!d->errorCovInitialized)
  {
    double estErrorCov = 0.0;
    if (d->estErrorCov)
    {
      CDPVariantValue estErrorCov_v = d->estErrorCov->GetVariantValue();
      estErrorCov = estErrorCov_v.GetConvertedValue<T>();
    }
    if (estErrorCov != 0.0)
      d->SetIdentity2x2(d->KF.errorCovPost, estErrorCov);
    else
      d->SetIdentity2x2(d->KF.errorCovPost, d->estErrorCovDefault);
    d->errorCovInitialized = true;
  }

  // Do Kalman filtering cycle
  Matrix<double> prediction = d->KF.predict();
  d->measurement(0) = (double)m_input;
  d->KF.correct(d->measurement);
  m_output = static_cast<T>(prediction(0));

  return STATUS_OK;
}