How do you make a discrete convolution if you can't analyze an index less than 1

I'm trying to make a convolution filter but I can't find the sum becuase when g(n-k are less than zero matlab doesn't know what to do.

n = 0:1:100;
g1 = .25*cos((n/6)*pi).*unit(n);
g2 = .5*cos((pi/2)*n).*unit(n);
g3 = .25*cos(((6*pi)/5)*n).*unit(n);

for i = 1:21
g(i) = g1(i)+g2(i)+g3(i);
end

h = [-.027 0 .033 -.038 0 .053 -.067 0 .133 -.266 .356 -.266 .133 0 -.067 .053 0 -.038 .033 0 -.027];
yi = zeros(21,21);
y = zeros(1,21);

for i = 1:21
    for j = 1:21
yi(j) = cumsum(g(i-j)*h(j));
    end
    y(i) = yi(j);
end
subplot(2,1,1)
plot(g)
subplot(2,1,2)
plot(y)

function [s]= unit(x) 
for i = 1:length(x)
    if x(i) > 0
        s(i) = 1;
    elseif x(i) == 0
        s(i) = 1;
    else
        s(i) = 0;
    end
end
end

Answer 1

[Image Analyst]

No need to write a convolution routine yourself. Simply use the built in conv() function.

[Charles Moody]

for this assignment I have to use a routine I make