The Fourier transform is a powerful tool for understanding the frequency content of a signal. It takes a signal from the time domain and shows us which frequencies are present. However, it has a significant limitation: it tells you what frequencies are in the signal, but not when they occur. This is fine for stationary signals, where the frequency content doesn't change over time. But what about signals where the frequencies change, like a piece of music or a seismic reading? For those, we need a different approach.

Imagine trying to read a sheet of music. The Fourier transform would be like getting a list of all the notes played in the entire song, but with no information about their order or timing. You'd know an A flat was played, but not whether it was in the first measure or the last. Wavelet transforms solve this problem by providing a time-frequency analysis. They show you which frequencies are present and precisely when they appear.

At the heart of the wavelet transform is the wavelet itself. Unlike the sine and cosine waves used in Fourier analysis, which oscillate forever, a wavelet is a short, wave-like oscillation with a beginning and an end. It's localized in time.

Think of a single clap in a quiet room. It's a brief event. A wavelet is like a mathematical representation of that clap. It has energy concentrated in a short time interval. To analyze a signal, we use a single prototype wavelet, often called the "mother wavelet." We then create a whole family of "daughter wavelets" by scaling (stretching or compressing) and shifting (moving) the mother wavelet along the signal.

There are two main types of wavelet transforms: the Continuous Wavelet Transform (CWT) and the Discrete Wavelet Transform (DWT).

The Continuous Wavelet Transform (CWT) compares the signal to shifted and scaled versions of a mother wavelet continuously over all possible scales and positions. This process generates a lot of data, making it highly redundant but also very detailed. It's excellent for in-depth signal analysis where you need to see every nuance.

The Discrete Wavelet Transform (DWT) is a more practical and efficient alternative. Instead of using all possible scales and positions, the DWT uses a specific subset of them, chosen to be just enough to reconstruct the original signal perfectly. This eliminates the redundancy of the CWT and makes it much faster, which is why it's widely used in applications like data compression (like the JPEG 2000 image format), denoising, and communications.

The magic behind the DWT is a concept called Multiresolution Analysis (MRA). MRA provides a framework for analyzing a signal at different levels of resolution, or detail. Think of it like a set of sieves. The signal is passed through a high-pass filter and a low-pass filter.

The high-pass filter, using a wavelet function, captures the fine details or high-frequency components of the signal. The low-pass filter, using a scaling function, captures the coarse approximations or low-frequency components. The output from the low-pass filter can then be passed through another set of filters to get an even coarser approximation and the next level of detail. This process can be repeated, breaking the signal down into many different resolution levels.

The scaling function is key here. It's related to the wavelet function and is responsible for capturing the general shape or trend of the signal at a certain scale. The wavelet function then captures the 'difference' or the detail that the scaling function missed. Together, they allow us to represent a signal as a combination of a low-resolution approximation plus details at various finer resolutions.

This ability to zoom in on high-frequency, short-lived events while also seeing the low-frequency, long-term behavior is what makes the wavelet transform so versatile for analyzing real-world, non-stationary signals.

Ready to check your understanding?

Understanding these fundamental concepts is the first step toward using wavelets to analyze signals and build powerful filters.