Sounding Reference Signal i.e. SRS is a reference signal transmitted by the UE in the uplink direction which is used by the eNodeB to estimate the uplink channel quality over a wider bandwidth. The eNodeB may use this information for uplink frequency selective scheduling.
The eNodeB can also use SRS for uplink timing estimation as part of timing alignment procedure, particularly in situations like there are no PUSCH/PUCCH transmissions occurring in the uplink for a long time in which case, the eNodeB relies on SRS for uplink timing estimation.
There are largely two kinds of SRS, Common SRS and Dedicated SRS. Common SRS is also called Cell Specific SRS and Dedicated SRS is also called UE Specific SRS.
Now coming to your second part why we need this kind of signal (I believe this is what you want to ask)?
In LTE eNodeB often allocates only a partial section of full system bandwidth for a specific UE and at a specific time. So it would be good to know which section across the overall bandwidth has better channel quality comparing to the other region. In this case, Network can allocate the specific frequency region which is the best for each of the UEs. (If we always have to use full bandwidth, we may not need this kind of reference signal since there is no choice even when there is a better or worse spots within the bandwidth).