Hello!

Just like we can use a SRRC filter for pulse shaping BPSK symbols, why is it that we cannot use a SRRC filter for pulse shaping a BPSK modulated OFDM symbol? What do we generally do for pulse shaping an OFDM symbol, considering the fact that OFDM symbols have cyclic prefixes in them to get rid of ISI. However, I guess it doesn't help fully in removing the ISI and hence we do need some sort of filter to shape the symbols. I heard generally a matched filter operation can help in that case. But, can anyone please tell me in brief if matched filtering is done both at the transmitter and the receiver, and what are the factors that we have to keep in mind while designing a one. 

Thanking you in advance!

The short answer is no, matched filtering is not done in OFDM like it is in single carrier communications.  Consider that the single-carrier matched filters only match at the center of the symbols, to allow sampling there with zero-ISI, but the remaining parts of the symbol may have large amounts of ISI.

This isn't possible in OFDM since a large fraction of the OFDM symbol has to be sampled many times in order to recover the subcarriers.

Instead, the transmit and receive filters are typically a flat-top filter with as little rolloff as practical in order to reject adjacent energy.  The spectrum occupied by the subcarriers generally requires as flat a frequency-domain response as practical since the multipath response will be (generally) unknown at the transmitter, so you can't assume which subcarriers can be attenuated and which can't.

The purpose of the base-band filter in OFDM TX is to smooth the transition from one symbol to another in the time domain.  The main reason is to help band-limit the edges of signal to meet spectral masks.

In Docsis 3 a raised cosine shape is used.

Adding my voice to all other good answers:

ofdm has long symbols and the ISI is inherently removed by choice of subcarrier spacing to be inverse of symbol time. cyclic prefix is for multpath which is different issue...

ofdm signal is generated by ifft and so it is brick wall but symbols are not continuous in phase in the stream and needs filter to remove unwanted side lobes.

non ofdm QAM (FDM) has very short symbols and a filter is needed for two purposes; one is to reduce user band and improve SNR as any filter but a second purpose is to reduce ISI i.e. shape symbols in time domain such that their side lobe power cancels out (after DAC). 

If you are just experimenting a model without actual DAC/ADC you can ignore ISI and just use any filter.

Great question and great answers!  Yes, I'm about a month and a half behind in reading up on these DSP Related questions.