Compute filter estimates for inputs using Kalman adaptive filter algorithm
dspobslib
Note The Kalman Adaptive Filter block is still supported but is likely to be obsoleted in a future release. We strongly recommend replacing this block with the Kalman Filter block. |
The Kalman Adaptive Filter block computes the optimal linear minimum mean-square estimate (MMSE) of the FIR filter coefficients using a one-step predictor algorithm. This Kalman filter algorithm is based on the following physical realization of a dynamic system.
The Kalman filter assumes that there are no deterministic changes to the filter taps over time (that is, the transition matrix is identity), and that the only observable output from the system is the filter output with additive noise. The corresponding Kalman filter is expressed in matrix form as
$$\begin{array}{l}g(n)=\frac{K(n-1)u(n)}{{u}^{H}(n)K(n-1)u(n)+{Q}_{M}}\hfill \\ y(n)={u}^{H}(n)\widehat{w}(n)\hfill \\ e(n)=d(n)-y(n)\hfill \\ \widehat{w}(n+1)=\widehat{w}(n)+e(n)g(n)\hfill \\ K(n)=K(n-1)-g(n){u}^{H}(n)K(n-1)+{Q}_{p}\hfill \end{array}$$
The variables are as follows
Variable | Description |
---|---|
n | The current algorithm iteration |
u(n) | The buffered input samples at step n |
K(n) | The correlation matrix of the state estimation error |
g(n) | The vector of Kalman gains at step n |
$$\widehat{w}(n)$$ | The vector of filter-tap estimates at step n |
y(n) | The filtered output at step n |
e(n) | The estimation error at step n |
d(n) | The desired response at step n |
Q_{M} | The correlation matrix of the measurement noise |
Q_{P} | The correlation matrix of the process noise |
The correlation matrices, Q_{M} and Q_{P}, are specified in the parameter dialog by scalar variance terms to be placed along the matrix diagonals, thus ensuring that these matrices are symmetric. The filter algorithm based on this constraint is also known as the random-walk Kalman filter.
The implementation of the algorithm in the block is optimized by exploiting the symmetry of the input covariance matrix K(n) . This decreases the total number of computations by a factor of two.
The block icon has port labels corresponding to the inputs and outputs of the Kalman algorithm. Note that inputs to the In and Err ports must be sample-based scalars with the same complexity. The signal at the Out port is a scalar, while the signal at the Taps port is a sample-based vector.
Block Ports | Corresponding Variables |
---|---|
In | u, the scalar input, which is internally buffered into the vector u(n) |
Out | y(n), the filtered scalar output |
Err | e(n), the scalar estimation error |
Taps | $$\widehat{w}(n)$$, the vector of filter-tap estimates |
An optional Adapt input port is added when you select the Adapt
port check box in the dialog. When this port is enabled,
the block continuously adapts the filter coefficients while the Adapt
input is nonzero. A zero-valued input to the Adapt
port
causes the block to stop adapting, and to hold the filter coefficients
at their current values until the next nonzero Adapt input.
The FIR filter length parameter specifies the length of the filter that the Kalman algorithm estimates. The Measurement noise variance and the Process noise variance parameters specify the correlation matrices of the measurement and process noise, respectively. The Measurement noise variance must be a scalar, while the Process noise variance can be a vector of values to be placed along the diagonal, or a scalar to be repeated for the diagonal elements.
The Initial value of filter taps specifies the initial value $$\widehat{w}\left(0\right)$$ as a vector, or as a scalar to be repeated for all vector elements. The Initial error correlation matrix specifies the initial value K(0), and can be a diagonal matrix, a vector of values to be placed along the diagonal, or a scalar to be repeated for the diagonal elements.
The length of the FIR filter.
The value to appear along the diagonal of the measurement noise correlation matrix. Tunable (Simulink).
The value to appear along the diagonal of the process noise correlation matrix. Tunable (Simulink).
The initial FIR filter coefficients.
The initial value of the error correlation matrix.
Enables the Adapt
port.
Haykin, S. Adaptive Filter Theory. 3rd ed. Englewood Cliffs, NJ: Prentice Hall, 1996.
Double-precision floating point
Single-precision floating point
LMS Adaptive Filter (Obsolete) | DSP System Toolbox |
RLS Adaptive Filter (Obsolete) | DSP System Toolbox |
See Adaptive Filters in Simulink for related information.