Kalman Filter Tracking

The filter operates through a continuous predict and update loop. The prediction step projects the current state estimate and error covariance forward in time using a motion model. The update step then fuses a new noisy measurement with the prediction, weighting the innovation by the Kalman Gain to produce a minimum mean-square error estimate.

The standard Kalman filter is the optimal estimator when the system dynamics and measurement models are linear and both process and observation noise are additive, white, and Gaussian. Under these conditions, it minimizes the trace of the error covariance matrix, providing a mathematically proven best estimate. Extensions like the Extended Kalman Filter (EKF) and Unscented Kalman Filter (UKF) relax the linearity constraint.

The filter maintains a state vector typically containing position, velocity, and sometimes acceleration. The state transition matrix (F) defines the physical motion model, such as a constant velocity or constant acceleration model. This matrix propagates the state from time k-1 to k, enabling the prediction of a mobile user's future location for proactive handoff triggering.

The Kalman Gain (K) is the critical weighting factor computed at each step. It determines the relative trust between the prediction and the new measurement. A high gain trusts the measurement more (high measurement certainty), while a low gain trusts the prediction more (high process certainty). This adaptive weighting allows the filter to optimally fuse Received Signal Strength (RSS) measurements with a kinematic model.

A unique strength of the Kalman filter is that it provides not just a point estimate but a full error covariance matrix (P). This matrix quantifies the uncertainty in the state estimate. The diagonal elements represent the variance of position and velocity errors, giving the spectrum handoff controller a confidence region for the user's predicted location, enabling risk-aware decision-making.

In cognitive radio, the Kalman filter tracks a secondary user's trajectory from noisy RSS measurements. By predicting the user's future position and velocity, the system can determine if the user is moving toward a primary user exclusion zone or the edge of a coverage area. This spatial awareness triggers a proactive spectrum handoff before a link failure or interference event occurs.

The mathematical foundation of the Kalman filter. It operates in a two-step cycle: predict the next state using a motion model, then update the belief using a noisy measurement. This recursive nature makes it computationally efficient for real-time tracking.

• Prior Belief: The predicted position before incorporating the new RSSI measurement.
• Posterior Belief: The refined position estimate after fusing the prediction with the observation.
• Markov Assumption: The current state depends only on the immediate previous state, simplifying the probabilistic model.

The Kalman filter tracks a state vector, typically containing the mobile user's 2D position and velocity: [x, y, vx, vy]. The motion model (state transition matrix) predicts how this state evolves over time.

• Constant Velocity Model: Assumes the user moves at a steady speed between measurements, a common baseline.
• Constant Acceleration Model: Adds acceleration terms to the state vector for maneuvering targets.
• Process Noise Covariance (Q): A matrix that quantifies the uncertainty in the motion model itself, accounting for unpredictable user movements.

The measurement model maps the hidden state (user position) to the observed Received Signal Strength Indicator (RSSI). This relationship is non-linear, often requiring an Extended Kalman Filter (EKF).

• Path Loss Model: A logarithmic formula (e.g., log-distance path loss) that predicts RSSI based on distance from a base station.
• Measurement Noise Covariance (R): A matrix representing the variance of RSSI noise caused by multipath fading and shadowing.
• Innovation: The difference between the actual RSSI measurement and the predicted RSSI, used to correct the state estimate.

The standard Kalman filter assumes linear models. Since RSSI is a non-linear function of position, the EKF linearizes the measurement model around the current estimate using a first-order Taylor expansion (Jacobian matrix).

• Jacobian Matrix: The matrix of partial derivatives of the measurement function with respect to the state variables, evaluated at the predicted state.
• Linearization Error: The approximation error introduced by the EKF, which can cause divergence in highly non-linear scenarios.
• Alternative: The Unscented Kalman Filter (UKF) avoids linearization by sampling sigma points, often yielding higher accuracy for RSSI-based tracking.

The Kalman Gain is the optimal weighting factor that balances trust between the prediction and the measurement. It is computed dynamically at each time step.

• High Gain: The filter trusts the new RSSI measurement more, rapidly correcting the estimate but potentially amplifying noise.
• Low Gain: The filter trusts the motion model prediction more, smoothing the trajectory but responding slowly to maneuvers.
• Covariance Update: The gain also updates the state covariance matrix P, shrinking the uncertainty ellipsoid when a good measurement is received.

The direct application of Kalman filter tracking. By predicting a mobile secondary user's trajectory, the system can proactively reserve a target channel in the user's predicted path before a handoff is triggered.

• Geolocation Database: The predicted coordinates are cross-referenced with a radio environment map to identify idle channels at the future location.
• Handoff Trigger: Initiated when the predicted trajectory intersects a primary user's protection contour or a known interference zone.
• Link Maintenance: Reduces forced termination probability by ensuring a new channel is available before the current link degrades.