The highest representable frequency really is the Nyquist rate: you only need 1 bit of sample depth to generate a digital signal whose corresponding band-limite

The highest representable frequency really is the Nyquist rate: you only need 1 bit of sample depth to generate a digital signal whose corresponding band-limited continuous signal is a sine wave that oscillates at that rate. Sample depth roughly tells you how far away a sample can be before the magnitude of the ideal sinc function used to reconstruct the continuous signal falls below your quantization threshold. That distance in turn gives you an idea of the frequency resolution you have... You could in theory run an FFT over that many samples and detect that the corresponding change was not just due to quantization noise.