does oversampling physcially increases the number of bits (or quantization level) in oversampling ADC's or it is a virtual increase ie. am i getting only the accuracy of a 16 bit with my 12 bit or does my bits can be really increased to 16 bits???

there's another way of looking at it.

Oversampling will increase the resolution within the signal band, providing a finer representation of the signal there. You will see this enhanced resolution if you observe only this highly oversampled signal band. You could this by:

1. Decimating the ADC output with a filter that discards most of the spectrum, leaving behind the signal band. This process will automatically result in a higher resolution as the filter you are using will need to have a wider bitwidth than that of the input signal, e.g. a crude averaging filter which takes a 10..10 bit stream and averages the last 4 bits will need to represent values (0, 0.25, 0.5, 0.75, 1). This corresponds to 2 bit additional resolution.

2. You can read out the signal band information from an FFT by looking only at the signal band bins. This is an implicit brick wall filtering that you are doing in your head or on paper.

Note that oversampling and decimation/filtering to increase resolution in a band which is narrower than the full bandwidth of your ADC (fs/2) is something you can use with any ADC output, not necessarily one which is built as an oversampling converter. The oversampling factor then gives you a processing gain (PG). However, a oversampled ADC intrisically offers you a much better sampling rate vs. resolution tradeoff than such a rudimentary method.

Vivek