You asked why to use modulation.
Modulation changes the frequencies of the signal so it "fits" better with the frequency response of the system.
For example, radio antennas are (more-or-less) band-pass filters, and you're only allowed to use a certain band anyway. When you transmit a signal by radio you have to make it fit within that band.
With Visible Light Communication you should have some more freedom to choose your own frequency, but your system still might not like certain frequencies.
Ambient light consists of very low frequencies <2Hz including 0Hz. So if your system relies on 0Hz for communication, it's likely to be confused by ambient light. If light off means 0 and light on means 1, then your system will get confused when I turn on the light in the room. So you design it to avoid very low frequencies. If you make it so a 100Hz flash means 0 and a 200Hz flash means 1 (that's FSK), it's not going to be confused by ambient light.
You might discover additional constraints when you construct your system and measure its frequency response. Most likely, the LED and/or the photodetector won't be able to respond to high frequencies. When you measure it, you'll find a "window" of frequencies that can get from the signal source, through the transmit amplifier (if you have one), LED, optical path, detector, and receive amplifier (if you have one), and to the signal receiver.
This "window" is your band, and the width of the band is the bandwidth (who woulda thunk?). It won't be a sharp transition, though. It won't be like 990kHz works and 991kHz doesn't. Rather, frequencies will get transmitted worse and worse as you move closer to the edge of the band. So you have to cut it off at some point and say "past this frequency the transmission is too poor". Traditionally, we find where the frequency response is 3dB weaker than the middle, and we call that the edge - though that rule of thumb does assume the edge is nicely shaped.
Then, to send data as quickly as possible, you want to work out a modulation scheme that packs your data into the band as efficiently as possible. That's a whole separate topic and out of scope here. The Shannon-Hartley formula gives a theoretical maximum amount of data that you can send in a certain bandwidth with a certain signal power and noise power.
Of course, you don't have to design an optimal modulation scheme - if you're not too concerned about data rate you can use something simple, like FSK. Many systems do.
One gotcha when thinking about modulation for beginners: you need to realize that every signal has frequency content. It's tempting to think that AM or ASK has only one frequency, since it modulates the amplitude, not the frequency. But when you take your 1MHz carrier frequency, and modulate it with a 0.2MHz signal, and then measure the frequencies in your modulated signal, you find that it does actually use the whole band 0.8MHz-1.2MHz. If you try to send that through a communications link whose frequency response only works for the band 0.95MHz-1.05MHz, the other side will get a mangled signal.
Another gotcha: if you have, say, a 200kbps digital signal, as a wave it actually contains frequencies up to 2MHz and higher, not just 200kHz. To limit the signal to 200kHz you have to "smear out" the edges, e.g. with a low-pass filter. When you measure the frequencies in a square wave you see that it contains not just the square wave frequency, but also odd-numbered multiples of it. If you send it through a low-pass filter to remove those, the wave looks a lot more sine-y.