I try to use Ozone-5 Multiband Compressor, but it looks very confused to me. I really want to know what is the purpose of doing Mutiband compression and How do you usually use it.
The concept of multiband compression can be easier understood when you think of them as multiple compressors packaged into one device. They are just like single compressors in the sense that they are controlling the dynamic range of the signal they are applied to. The same settings apply - threshold, ratio, attack/release.
The biggest difference is that each compressor in a multiband is only being applied to a slice of the frequency spectrum. So you'll likely have a compressor for the low freqs, low mids, high mids, and highs. Some multibands will also have floating comps that can be applied to specific freq areas regardless of where the others are set. This is where your freq. settings come into play as to where one compressor ends and the next one begins.
After that understanding, the concept is then to "tighten" the frequency spectrum of the signals together - where a single comp would apply compression to the whole signal across the board regardless of frequency spectrum, with a multiband you can get for example the lows (bass+kick) comp'd/pumping together without the same compression being applied to the rest of the frequency spectrum.
Hope this helps!
Well, the purpose of a compressor in general is mostly to bring up the RMS level while avoiding excessive peaks.
The purpose of the multiband splitting is mostly to avoid the ducking effect. When a compressor reduces one peak, it affects the whole signal. So if there is one very loud sound and some quieter ones at the same time, the loud one will seemingly “push away” everything else. Sometimes this is exactly what you want – for instance to make voices better understandable it helps to make them duck away some of the other sound. However, too obvious ducking sounds pretty weird. And it's particularly obvious when the different signals are far apart in the frequency spectrum, because then they don't mask each other very much. In particular, a fat bass drum needs a lot of peak level to sound loud, and when it pumps a compressor you'll often hear very clearly how the rest of the signal goes down.
The classical way to solve this particular problem is to use a sidechain EQ in the compressor: cut away some of the low-end before it can trigger the compressor. That works pretty well, however the output signal will then still have large peaks at those bass drum hits. If your medium is peak-limited (like CDs, and indeed almost all other digital media) then this will be the limiting factor to how loud you can make the master. Here's where the multiband compressor comes in: it compresses the bass drum a lot, but without reducing the gain in the higher frequency ranges as strongly.
Of course, it works the other way around, to – a long bass drone won't get chopped by high-peak treble transients if they only trigger the high bands of a multiband compressor.
The flip side is that you loose “signal integrity”: a normal compressor is guaranteed to keep the ratios of all frequency components in a signal consistent. A multiband compressor keeps changing the frequency response; in extreme settings this can give results resembling actual dynamic-morphed filters (auto-wah etc.). Normally it's not that bad – sometimes it can even subtly improve1 signals (in particular for bass drum and slap-bass, it brings out dynamics in a particular way that's often quite useful).
However, in mastering, even slightly uneven response can make the overall mix sound off. That's why I'd would recommend that, like brickwall limiting, multiband compression should be used as gently as you can afford. It's a useful tool, but unless you notice in a particular mix that there is one “frequency blob” that you want to reduce, better apply multiband compression only to shave up a bit of the loudest peaks in each frequency band, perhaps a bit more in the bass range, and otherwise rely more on well set up singleband compression stages (with suitable attack/decay and sidechain EQ settings), and perhaps some tape-style soft clipping, to bring up a master.
1Also, the band splitting can substantially change the waveform – quite relevant for loud transients when you're feeding a soft-clipping afterwards. In my experience, a multiband often makes such nonlinearities sound better, but it depends a lot on the precise choice of band edge frequencies and the relative attack setting.