Filter data with filter object


y = filter(hd,x)
y = filter(hd,x,dim)


Fixed-Point Filter Syntaxes

y = filter(hd,x) filters a vector of real or complex input data x through a fixed-point filter hd, producing filtered output data y. The vectors x and y have the same length. filter stores the final conditions for the filter in the States property of hdhd.states.

When you set the property PersistentMemory to false (the default setting), the initial conditions for the filter are set to zero before filtering starts. To use nonzero initial conditions for hd, set PersistentMemory to true. Then set hd.states to a vector of nstates(hd) elements, one element for each state to set. If you specify a scalar for hd.states, filter expands the scalar to a vector of the proper length for the states. All elements of the expanded vector have the value of the scalar.

If x is a matrix, y = filter(hd,x) filters along each column of x to produce a matrix y of independent channels. If x is a multidimensional array, y = filter(hd,x) filters x along the first nonsingleton dimension of x.

To use nonzero initial conditions when you are filtering a matrix x, set the filter states to a matrix of initial condition values. Set the initial conditions by setting the States property for the filter (hd.states) to a matrix of nstates(hd) rows and size(x,2) columns.

y = filter(hd,x,dim) applies the filter hd to the input data located along the specific dimension of x specified by dim.

When you are filtering multichannel data, dim lets you specify which dimension of the input matrix to filter along — whether a row represents a channel or a column represents a channel. When you provide the dim input argument, the filter operates along the dimension specified by dim. When your input data x is a vector or matrix and dim is 1, each column of x is treated as a one input channel. When dim is 2, the filter treats each row of the input x as a channel.

To filter multichannel data in a loop environment, you must use the dim input argument to set the proper processing dimension.

You specify the initial conditions for each channel individually, when needed, by setting hm.states to a matrix of nstates(hd) rows (one row containing the states for one channel of input data) and size(x,2) columns (one column containing the filter states for each channel).


collapse all

Filter a signal using a filter with various initial conditions (IC) or no initial conditions.

x = randn(100,1);    % Original signal.
b = fir1(50,.4);     % 50th-order linear-phase FIR filter.
hd = dfilt.dffir(b);    % Direct-form FIR implementation.

Do not set specific initial conditions.

y1 = filter(hd,x);   % 'PersistentMemory'='false'(default).
zf = hd.states;      % Final conditions.

Now use nonzero initial conditions by setting ICs before you filter.

hd.persistentmemory = true;
hd.states = 1;      % Uses scalar expansion.
y2 = filter(hd,x);
stem([y1 y2])       % Different sequences at beginning.

Looking at the stem plot shows that the sequences are different at the beginning of the filter process.

Here is one way to use filter with streaming data.

reset(hd);           % Clear filter history.
y3 = filter(hd,x);   % Filter entire signal in one block.

As an experiment, repeat the process, filtering the data as sections, rather than in streaming form.

reset(hd);              % Clear filter history.
yloop = zeros(20,5);    % Preallocate output array.
xblock = reshape(x,[20 5]);
for i=1:5
   yloop(:,i) = filter(hd,xblock(:,i));

Use a stem plot to see the comparison between streaming and block-by-block filtering.

stem([y3 yloop(:)]);

Filtering the signal section-by-section is equivalent to filtering the entire signal at once.


collapse all

Quantized Filters

The filter command implements fixed- or floating-point arithmetic on the quantized filter structure you specify.

The algorithm applied by filter when you use a discrete-time filter object on an input signal depends on the response you chose for the filter, such as lowpass or Nyquist or bandstop. To learn more about each filter algorithm, refer to the literature reference provided on the appropriate discrete-time filter reference page.


dfilt/filter does not normalize the filter coefficients automatically. Function filter supplied by MATLAB does normalize the coefficients.

Adaptive Filters

The algorithm used by filter when you apply an adaptive filter object to a signal depends on the algorithm you chose for your adaptive filter. To learn more about each adaptive filter algorithm, refer to the literature reference provided on the appropriate adaptive filter object reference page.


[1] Oppenheim, A.V., and R.W. Schafer, Discrete-Time Signal Processing, Prentice-Hall, 1989.

See Also


Introduced in R2011a