In many real-time applications, we don't have time to wait for the whole signal to be recorded before we start analyzing it, we need answers now. But the moment we start cutting up a signal to get those fast results, we compromise the very data the FFT needs to be accurate.

How can we mitigate this issue? What is the secret to balancing speed with accuracy? And what is the hidden "price" we pay the moment we try to increase our resolution?

Why do your frequency peaks look smeared or "wrong"? In the previous video, we saw how spectral leakage creates those messy tails that hide your signal. In this video, we're going to fix it.

The secret lies in a technique called Windowing.

But what is windowing? How does it help? What problems does it cause?

So now we understand all the problems with the FFT and we've looked at how these problems might be mitigated.

How do we choose the best window for the job? How do we ensure that our FFT size is large enough to give us the frequency resolution we need? How do we put all this knowledge together for the most accurate FFT possible?