I was thinking the other day about how to increase the usage of an ADC and I came up with this idea. It is mathematically solid but I have no idea if it is practical.
The idea is to use the fact that a shift in the frequency domain is equivilent to multiplying by exp(+-I*a*t) in the time domain.
So by using that property we can shift the higher frequencies of the function into a lower region that can be handled by the ADC. Sampling rate can be increased by running ADC's in parallel.
To do this requres one to first compute the signal,
s_k(t) = [exp(I*k*O*t)*s(t)] ** [O/sqrt(2*Pi)*exp(I*(a + O/2)*t)*sinc(O/2*t)]
where ** means convolution and [a,O] represents the interval of the bandwidth of the ADC. (a represents the lowest frequency and O the highest)
So one would take k = 0 and pass it to one ADC, k = 1 to another, etc and then to get the original signal back one just as to compute in the digital domain,
s(t) = sum(exp(-I*k*O*t)*s_k(t),k=-oo..oo).
Thats basically it. The theory is very simple.
The hard part come with implementing it. First off, the circuitry that actually computes the convolution and does the multiplication would have to have enough bandwidth not to destory the signal and the time delay would have to be taken into account. Theres also the issue with generating the sin and cos waves to use in the exp() factors along with the sinc function.
the formula for s_k(t) does not involve information about s(t) except at the point t so it is causal.
Certain approximations will, ofcourse, come into play like having to restrict the convolution to a finite range(which will in effect make the filtering non-ideal...which might be able to be taken care of after conversion).
All this rests on how hard it is to implement the operational circuitry to do the "shifting". Mathematically its easily provable that it works but implementing the necessarily mathematical operations might be impossible with circuits or might not be practical enough. (although I'm still varifying the mathematics and making sure some things are correct)
If one can create a operational circuit that can do the above equation for s_k(t) that can "out perform" the ADC(i.e., that it can compute s_k(t) that will be good enough for the ADC). Then maybe we can try to impelement it and see how well it works.
Overall it might be easier just to make a better ADC than the method I'm proposing but thats why I'm asking since I don't have much experience with implementing analog mathematics.
Jon