Modulo-N circular convolution
c = cconv(a,b,n)
c = cconv(gpuArrayA,gpuArrayB,n)
Circular convolution is used to convolve two discrete Fourier transform (DFT) sequences. For long sequences, circular convolution can be faster than linear convolution.
c = cconv(a,b,n) circularly
the length of the resulting vector. If you omit
it defaults to
= length(a)+length(b)-1, the circular convolution is equivalent
to the linear convolution computed with
You can also use
cconv to compute the circular
cross-correlation of two sequences.
c = cconv(gpuArrayA,gpuArrayB,n) returns
the circular convolution of the input vectors of class
See Establish Arrays on a GPU for
details on gpuArray objects. Using
requires Parallel Computing Toolbox™ software and a CUDA-enabled
NVIDIA GPU with compute capability 1.3 or above. See http://www.mathworks.com/products/parallel-computing/requirements.html for
details. The output vector,
c, is a gpuArray object.
See Circular Convolution Using the GPU for
an example of using the GPU to compute the circular convolution.
Generate two vectors and compute their modulo-4 circular convolution.
a = [2 1 2 1]; b = [1 2 3 4]; c = cconv(a,b,4)
c = 14 16 14 16
Generate two signals of different lengths. Compare their circular convolution and their linear convolution. Use the default value for
a = [1 2 -1 1]; b = [1 1 2 1 2 2 1 1]; c = cconv(a,b); % Circular convolution cref = conv(a,b); % Linear convolution dif = norm(c-cref)
dif = 9.7422e-16
The resulting norm is virtually zero, which shows that the two convolutions produce the same result to machine precision.
Generate two complex sequences. Use
cconv to compute their circular cross-correlation. Flip and conjugate the second operand to comply with the definition of cross-correlation. Specify an output vector length of 7.
a = [1 2 2 1]+1i; b = [1 3 4 1]-2*1i; c = cconv(a,conj(fliplr(b)),7);
Compare the result to the cross-correlation computed using
cref = xcorr(a,b); dif = norm(c-cref)
dif = 3.3565e-15
Generate two signals: a five-sample triangular waveform and a first-order FIR filter with response .
x1 = conv([1 1 1],[1 1 1]) x2 = [-1 1]
x1 = 1 2 3 2 1 x2 = -1 1
Compute their circular convolution with the default output length. The result is equivalent to the linear convolution of the two signals.
ccnv = cconv(x1,x2) lcnv = conv(x1,x2)
ccnv = -1.0000 -1.0000 -1.0000 1.0000 1.0000 1.0000 lcnv = -1 -1 -1 1 1 1
The modulo-2 circular convolution is equivalent to splitting the linear convolution into two-element arrays and summing the arrays.
ccn2 = cconv(x1,x2,2) nl = numel(lcnv); mod2 = sum(reshape(lcnv,2,nl/2)')
ccn2 = -1 1 mod2 = -1 1
Compute the modulo-3 circular convolution and compare it to the aliased linear convolution.
ccn3 = cconv(x1,x2,3) mod3 = sum(reshape(lcnv,3,nl/3)')
ccn3 = 0 0 0 mod3 = 0 0 0
If the output length is smaller than the convolution length and does not divide it exactly, pad the convolution with zeros before adding.
c = 5; z = zeros(c*ceil(nl/c),1); z(1:nl) = lcnv; ccnc = cconv(x1,x2,c) modc = sum(reshape(z,c,numel(z)/c)')
ccnc = 0.0000 -1.0000 -1.0000 1.0000 1.0000 modc = 0 -1 -1 1 1
If the output length is equal to or larger than the convolution length, pad the convolution and do not add.
d = 13; z = zeros(d*ceil(nl/d),1); z(1:nl) = lcnv; ccnd = cconv(x1,x2,d) modd = z'
ccnd = Columns 1 through 7 -1.0000 -1.0000 -1.0000 1.0000 1.0000 1.0000 0.0000 Columns 8 through 13 -0.0000 0.0000 0.0000 0.0000 -0.0000 -0.0000 modd = -1 -1 -1 1 1 1 0 0 0 0 0 0 0
The following example requires Parallel Computing Toolbox software and a CUDA-enabled NVIDIA GPU with compute capability 1.3 or above. See http://www.mathworks.com/products/parallel-computing/requirements.html for details.
Create two signals consisting of a 1 kHz sine wave in additive white Gaussian noise. The sample rate is 10 kHz
Fs = 1e4; t = 0:1/Fs:10-(1/Fs); x = cos(2*pi*1e3*t)+randn(size(t)); y = sin(2*pi*1e3*t)+randn(size(t));
y on the
gpuArray. Obtain the circular convolution
using the GPU.
x = gpuArray(x); y = gpuArray(y); cirC = cconv(x,y,length(x)+length(y)-1);
Compare the result to the linear convolution of
linC = conv(x,y); norm(linC-cirC,2)
Return the circular convolution,
to the MATLAB® workspace using
cirC = gather(cirC);
 Orfanidis, Sophocles J. Introduction to Signal Processing. Englewood Cliffs, NJ: Prentice-Hall, 1996, pp. 524–529.