Dear Abdul.

From the point of view of electronics, you have to have a good differential amplify isolated.

With a filter that rejects mains frequency 50 or 60 Hz, and can have filters specific frequencies to study each of the waves of interest, whether alpha, beta, theta, and so on. A good chipset is texas ADS1299.

The tensions on the scalp is in the range of micro volts. and fequencia between 1 and 300Hz.

Consider this book, which explains the concepts, math, and implementations of many methods to isolate frequency components of EEG data: http://mitpress.mit.edu/9780262019873

EEG rhythms are generated by neural loops in the brain that work like bandpass filters. These loops are driven by random non-rhythmic activity from elsewhere. Simulations by mathematical models of these loops, show that these loops generate spindling (waxing and waning) rhythms although all parameters of the model are kept constant. The rhythm waxes when the input activity accidentally co-operates with the present rhythm in the loop. The rhythm wanes when the input activity accidentally counteracts with the present rhythm. So my conclusion is: do not detect separate rhythmic spindles. Instead, detect whether the neural network is in a bandpass filtering mode or not. How to do that? Read my 2000 article "Analysis of a sleep-dependent neuronal feedback loop: the slow-wave microcontinuity of the EEG". This describes the detection of the slow-wave mode but also works perfectly for other rhythms like sleep spindles and alpha rhythm. Software for this "Neuro-loop analysis" can be downloaded for free: see http://www.edfplus.info/downloads/index.html which includes an open-source version.