I think sampling rate derives this sub carrier spacing.
In LTE, sampling rate 30.72 = 2048 * 15 KHz.
I would like to say since symbol time is reciprocal of sub carrier spacing. 2048 IDFT is fit of LTE since 1024 is insufficient due to 1200 sub carrier (100 RB * 12 sub carrier per RB).
So, Actually 1200 sub carriers used to capture 1200 sine waves and rest padded with 0.
Not have more knowledge about 2048 IDFT. It seems somehow 2048 derives 15KHz spacing.
But I am not sure, looking for others too to respond .