Fitting one line to another

I am attempting to do some basic noise cancelation by using a reference signal. I know there are more elegant solutions but this should work:

    t = linspace(-2*pi,2*pi,200);
    signal = sin(t)'; %original signal
    sigma = 0.07; %noise standard deviation
    noise = sigma*randn(size(signal)); %noise
    rover = signal + noise;
    base = noise; 
    hold off;
    figure(1)

    plot(rover,'color','red'); 
    hold on;
    plot(base,'color','blue');

    E = [ones(size(base)) base ];
    ls_solution = lsqr(E,rover ,1e-6,20)
    fit = E*ls_solution;
    solution = rover -fit; 
    plot(fit,'color','green');
    plot(solution,'color','cyan');

Here is the output: lsqr converged at iteration 2 to a solution with relative residual 0.99.

 ls_solution =

   -0.0030
    1.5903

While it seems to work conceptually the convergence is so bad that it doens't remove much of the noise. Why does this only go to iteration 2? The tolerance is set to 1-e6 and the max iterations are 20.