Output gradually scales up over time

[samcrow]

We've seen that the time-domain reconstructed samples gradually increase in magnitude over several seconds.

This line in the frequency correct step might have a small problem: https://github.com/ucsdsysnet/sparsdr/blob/master/reconstruct/src/steps/frequency_correct.rs#L67

The magnitude of self.correction is supposed to always be 1, but because of math errors its magnitude might gradually increase.

I saw somewhere that someone was doing self.correction = self.correction / self.correction.norm() periodically (not for every sample) to keep the magnitude close to 1.