• The N-Ifft points and number of subcarriers are not same all the time. Why we introduce unmodulated or zero padded subcarrier between the modulated or useful subcarrier.
For different reasons. Mainly:
- Most OFDM systems are based on direct-conversion receivers. In those receivers, you mix a signal down to complex baseband by multiplying it with a complex tone of the signal's center frequency – but that leads to the signal at exactly 0Hz in the baseband to be unrecoverable, technologically. For that reasons, the DC carrier (or carriers around DC) (DC=direct current, a very Electrical Engineering way of saying "0Hz") is not used
- OFDM systems often have to deal with problems of strong Out-Of-Band radiation. By leaving the "edge" carriers unused, you can mitigate that.
• Should the training sequence be added before ifft operation or after ifft+cp inclusion operation? If both can be possible then what is the best approach between these two.
I assume you want to train your equalizer? There's a lot of approaches to channel equalization with OFDM, and every one has its advantages and disadvantages.
The same applies to adding a timing and frequency synchronization sequence: there's a lot of approaches to doing synchronization for OFDM, and hence, as many trade-offs for advantages and disadvantages to having something in time- and frequency domain. Note that fine frequency estimation is often done with known time-domain signal techniques (because we can't do an FFT overly well if we don't even know where), and coarse frequency correction (ie. correcting an offset of integer multiples of the subcarrier spacing) can then be done afterwards (in time- or frequency domain, whichever is more appropriate for the system.
I personally would recommend Schmidl&Cox' classic paper Robust frequency and timing synchronization for OFDM – it's a great read, and I personally think it's a good paper, with a short, concise intro, a well-explained method, well-condensed mathematical proofs, and useful explanation and derivation of performance parameters. Also, if you're doing OFDM in your thesis, and want to actually have read at least one of the papers you cite, this should probably be the one.
• For the DAC conversion, raised cosine pulse shaping filters are considered to be the best among other shaping filters. What should be its roll off factors, No. of symbols and No of samples in each symbol (for OFDM). How much should be an oversampling factor for upsampling process.
Generally, as you've learned, Roll-off etc are parameters of your design that you choose to make your design work according to your requirements. So, no matter for which method you'd be asking, "what are the values of the Roll off factor for ?", the answer will always be, "what you design them to be!".
However: OFDM was partially "invented" to eliminate the need for a set of pulse shaping filters. So your question regarding a raised cosine doesn't make much sense. If you pulse shape the time domain OFDM signal, you destroy the "O" in OFDM! You mustn't do that.
• Is Low Pass Filter should be used at the transmitter side too what is the purpose of using it at the transmitter side ?
I think you're sitting in front of some specific example, and think it is "all possible OFDM transceivers". I can't tell you what your example is – I don't know it. Maybe that low-pass filter is just part of the direct conversion mixer that converts the OFDM baseband signal to a passband signal?!