helperModClassFrameGenerator.m

function y = helperModClassFrameGenerator(x, windowLength, stepSize, offset, sps)
%helperModClassFrameGenerator Generate frames for machine learning
%   Y = helperModClassFrameGenerator(X,WLEN,STEP,OFFSET) segments the
%   input, X, to generate frames to be used in machine learning algorithms.
%   X must be a complex-valued column vector. The output, Y, is a size
%   WLENxN complex-valued array, where N is the number of output frames.
%   Each individual frame has WLEN samples. The window is progressed STEP
%   samples for the new frame. STEP can be smaller or greater than WLEN.
%   The function starts segmenting the input, X, after it calculates an
%   initial offset based on the OFFSET value. OFFSET is a two-element
%   real-valued vector, where the first element is the deterministic offset
%   value and the second element is the maximum value of the random offset
%   value. Total offset is OFFSET(1)+randi([0 OFFSET(2)]) samples. The
%   deterministic part of the offset removes transients, while the random
%   part enables the network to adapt to unknown delay values. Each frame
%   is normalized to have unit energy.
%
%   See also ModulationClassificationWithDeepLearning.

%   Copyright 2018 The MathWorks, Inc.


numSamples = length(x);
numFrames = ...
  floor(((numSamples-offset)-(windowLength-stepSize))/stepSize);

y = zeros([windowLength,numFrames],class(x));

startIdx = offset + randi([0 sps]);
frameCnt = 1;

while startIdx + windowLength < numSamples
  xWindowed = x(startIdx+(0:windowLength-1),1);
  framePower = sum(abs(xWindowed).^2);
  xWindowed = xWindowed / sqrt(framePower);
  y(:,frameCnt) = xWindowed;
  frameCnt = frameCnt + 1;
  startIdx = startIdx + stepSize;
end