www.radartutorial.eu www.radartutorial.eu Radar Basics

What Is an Alpha-Beta Filter?

Alpha-Beta Filter

Alpha-Beta Filter

An alpha-beta filter (α-β filter) is a simple mathematical method used to estimate the position and velocity of a moving target from successive position measurements. The filter is used in radar target tracking, among other applications. Based on previously estimated motion data, it predicts the target's next position and corrects this prediction using a new radar measurement. Two fixed filter parameters are used for this purpose: α (alpha) and β (beta).

In its simplest form, the method assumes that the target moves at an approximately constant velocity between two measurements. It can be applied to a single coordinate or independently to several coordinates. However, when processing multiple coordinates, differences in measurement accuracy and relationships between the coordinates may need to be considered.

Predicting the Next Target Position

The calculation starts with the previously estimated position and velocity. These values can be used to predict the position the target is expected to reach at the time of the next measurement. Assuming constant velocity, the distance traveled during the time interval is added to the previous position.

xp = xalt + valt · Δt (1)

In this equation, xp represents the predicted position, xalt is the previously estimated position, valt is the previously estimated velocity, and Δt is the time interval between two measurements. In the assumed constant-velocity motion model, the predicted velocity initially equals the previously estimated velocity.

Correction Using a New Radar Measurement

When a new radar measurement becomes available, its result is compared with the predicted position. The difference between these two values is called the residual or innovation. It indicates how far the measured position deviates from the prediction.

r = xmess − xp (2)

The filter uses this residual to make two different corrections. The factor α is used to correct the estimated position. The factor β is used to correct the estimated velocity.

xneu = xp + α · r
vneu = valt + (β / Δt) · r
(3)

The parameter α determines what proportion of the position deviation is used to correct the estimated position. The larger α is, the closer the corrected position moves toward the new measurement. A smaller value, on the other hand, gives greater weight to the previous motion prediction.

The parameter β influences the velocity correction. For this purpose, the position deviation is divided by the time interval between the measurements and then weighted by β. In this way, the filter can derive a change in the estimated velocity from successive position measurements.

A radar tracks a target whose previously estimated range is 10.0 km. The estimated radial velocity is 200 m/s. One second elapses before the next radar measurement. Assuming constant velocity, the filter therefore expects the target at a range of 10.2 km.

However, the new radar measurement indicates a range of 10.3 km. The residual is therefore +100 m. For this example, the filter parameters are set to α = 0.6 and β = 0.2.

r = 10 300 m − 10 200 m = 100 m
xneu = 10 200 m + 0,6 · 100 m = 10 260 m
vneu = 200 m/s + (0,2 / 1 s) · 100 m = 220 m/s
 

After the correction, the estimated range is therefore 10.26 km, and the estimated velocity is 220 m/s. The filter uses these corrected values for the next prediction. The original radar measurement of 10.3 km remains unchanged.

How Do α and β Affect Target Tracking?

The choice of the two filter parameters largely determines the behavior of the alpha-beta filter. Larger values generally produce a stronger response to new measurements. As a result, the filter can adapt more quickly to changes in target motion. At the same time, however, measurement errors and noise have a greater effect on the estimated target motion.

Smaller values, on the other hand, produce smoother estimates because new measurements have less influence on the previous prediction. However, the filter may then respond more slowly to actual changes in target motion. Particularly during a target maneuver, the estimated position may temporarily deviate significantly from the true position.

Therefore, the parameters must not be selected without considering the characteristics of the radar and target motion. Measurement accuracy, measurement interval, and expected target maneuverability all influence the appropriate settings. The values must also be chosen to ensure that the filter remains stable.

What Happens During Target Maneuvers?

The basic alpha-beta filter uses a constant-velocity motion model. If a target accelerates, decelerates, or changes its direction of motion, this model no longer fully represents the actual movement. This can result in larger deviations between the predicted and measured positions.

The filter responds to these deviations by correcting the estimated position and velocity. However, it cannot predict an unexpected maneuver before corresponding measurements become available. Depending on the selected filter parameters, several measurements may be required before the estimates have sufficiently adapted to the target's actual motion.

Extended methods can be used for target motion involving significant acceleration. One example is the alpha-beta-gamma filter (α-β-γ filter), which accounts for acceleration in addition to position and velocity. Other motion models and filtering methods may also offer advantages when tracking maneuvering targets.

Difference from the Kalman Filter

The alpha-beta filter and the Kalman filter use the same basic principle: They first calculate a prediction and then correct it using a new measurement. The main difference lies in how the correction gain is determined.

In the conventional alpha-beta filter, the two gain factors α and β are fixed. A Kalman filter, on the other hand, calculates its gain from the estimated uncertainties of the motion model and the measurements. These uncertainties are represented by covariance matrices and are updated at each calculation step.

Under certain conditions, an alpha-beta filter can be regarded as a simplified form of a Kalman filter with constant gain. In particular, if the model and measurement conditions remain unchanged, the gain of a Kalman filter may converge to constant values after an initial transient period. However, the alpha-beta filter does not require continuous calculations of covariance matrices and therefore has lower computational requirements.

Importance for Radar Target Tracking

The alpha-beta filter is particularly suitable for applications that require a simple and fast estimate of target position and velocity. Its calculations consist primarily of additions and multiplications. Consequently, the method requires relatively little computational effort and can also be used when tracking many targets simultaneously.

The predicted target position can also be used to define the region in which the next radar measurement is expected. However, determining which radar plots belong to which track is the task of data association, not of the alpha-beta filter itself. An incorrect assignment of measurements cannot be reliably corrected by the filter alone.

Like the Kalman filter, the alpha-beta filter also operates recursively. Each new calculation step requires only the previously estimated position and velocity, the time interval, and the newly associated measurement. Therefore, the method is suitable for automatic real-time target tracking.