SDR minimum ADC level

What is the minimul level of a I/Q signal at the input of tha ADC of a SDR that is required to demodulate the signal?

In the document "Software defined radio for the masses IV" is is stated:

"For a weak signal to be recovered,
the minimum analog gain must be great
enough so that the weakest signal to
be received, plus thermal and atmospheric
noise, is greater than at least
one A/D converter quantizing level (the
least-significant usable bit). "

How can be enough to recover the signal if it has the amplitude of > one bit? So 2 bits would be enough?

Yep, and many GPS receivers use a single bit quantiser to pick out a signal that can be 20dB BELOW the noise, it does NOT have to be 2 bits, greater then one is sufficient...

Consider that the input to the ADC is broadband, you are going to end up with a narrowband output from the decimator chain and every time you reduce the bandwidth by a factor of 4 you gain an extra bit of effective word length, by the time you have decimated down from a screaming fast ADC to a few tens of KHz, your word length has grown in quite a useful way.

Regards, Dan.

I know that oversampling increase the resolution. But will this work also in this case? I want to buil an QSD SDR with a 24bit ADC (PCM4202). There will be a 100kHz LPF before ADC and a sampling frequency of 216kHz.

So I will have a chunk of 200kHz spectrum sampled with 24bit (or less in reality). But then the decrease of bandwidht from these 200kHz to 3-4kHz is not so great. Or should I consider the 36MHz of the sigma delta modulator when comparing to 4kHz output signal?

In the end, I only want to have an idea of how much gain should I have before the ADC in order to demodulate a signal of let's say 100uV (at the ADC input).

Dithering and oversampling help to "see" signals whose level is below one quantization step.

Dithering, by giving to the quantization noise (that is nonrandom for very low level signals) statistical properties like random noise.

Oversampling, by distributing the noise power over a wide bandwidth (wider than the band of interest). Then, noise outside the band of interest can be removed by filtering. The oversampled signal can be further decimated in order to reduce sampling rate.

Z

I believe the question is useless without referring to signal characteristics and modulation methods. GPS mentioned by is in fact an extreme example, still working with very low SN ratio. Most digital modulation methods require at least 10 or 20 dB SNR for acceptable transmission quality, analog radio even more.

Designing a SDR with 1 or 2 bit SNR is a bad idea.

The modulation is SSB

IF these conditions are satisfied:

- signal level is "high enough" in such a way that in a sample time the signal crosses more than a quantization level at "almost all the samples", and
- the signal spectrum is "not too concentrated", so the signal is not "almost periodic"

THEN the following assumptions can be made:

- quantization noise is a random process independent of the signal
- samples of quantization noise are independent random variables uniformly distributed in [-q/2, q/2), q being the quantization step
- the power spectral density of quantization noise is uniform over the Nyquist band

With these assumptions, quantization noise can be treated as additive noise using standard methods.
Note that some characteristics noted with quotation marks are not accurately defined.

Z