Broad-band background noise is a source of corruption in the recording and transfer of a wide range of signals.  For instance, an audio signal may contain unwanted environmental noise as well as noise caused by random sensitivities within the recording equipment.  Such noise typically has adverse effects which limit the usability of the information contained within a signal.

​

Broad-band noise, which covers a wide range of the frequency spectrum, presents a unique challenge compared to narrow-band noise.  When the original signal covers a wide, unpredictable frequency range, it is much less feasible to suppress broad-band noise which overlaps the spectrum of the original signal.  The following example provides a simple illustration of a situation where windowed filtering is sufficient, and one where it does not adequately reconstruct a noisy signal.

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

Clearly, this band-pass filter reconstructs the original signal very accurately.  However, in order to size and place the pass-band, it is necessary to know the frequency spectrum of the signal in advance.

​

​

For our project, we chose to investigate and implement the Wiener filter, which suppresses broad-band noise using statistical estimates of the background noise and original signal, both extracted from the frequency spectrum of the observed, "corrupted" signal.

​

In doing this, we hoped to design a digital filter that is very robust in terms of the range of signals that it can be used to process.  No prior knowledge of the signal is required, and the only information that one needs to know in order to suppress the majority of the noise in the signal is that it has been corrupted by broadband additive noise.

​

​