Digitizing IF for SDR/Sprectrum analysis

[prutser]

Suppose I have a downconverter  and want to digitize a spectrum with a bandwidth of 30MHz using and ADC  with 100MHz fs.

Option1: Down convert down to 0 Hz, use an LPF and digitize, so the spectrum appears between 0 and 30 MHz in the 1st Niquist zone.
Option2: Down convert to a slightly higher frequency and use a BPF, so the converted spectrum appears between 10MHz - 40 MHz in the 1st Niquist zone.

This first option seems the best with respect to the anti Alias filtering. Are there any advantages for using option 2 ?

Any insights here ?

[paul@yahrprobert.com]

People tend to like option 2 because of 1/f noise, especially if you will need some IF gain. If you look up the various RF transceiver chips that direct convert down to baseband they often do some dc offset correction to mitigate the noise near DC.

[KE5FX]

Right, a lot depends on how important the spectrum near DC is.  If you expect to preserve very low frequencies without a lot of added flicker noise, it may be better to downconvert to a nonzero IF, perhaps using an LO that's synchronous with your sampling clock if phase recovery is important. 

Where is the spectrum of interest, exactly?  Can you adjust your sampling clock frequency to place it in the middle of a higher Nyquist zone?  That can be the best strategy of all.  ADCs designed for RF digitization tend to be usable at frequencies well beyond their spec-sheet ratings, which can be a Good Thing or a Bad Thing.

[radiolistener]

This first option seems the best with respect to the anti Alias filtering. Are there any advantages for using option 2 ?

The first option is more easy, because don't needs to take into account that the frequencies are shifted.
The second option has advantage because allows to keep your signal within more safe bandwidth, because at low frequency there are a lot of noise which is hard to filter so it may reduce dynamic range at the beginning of your signal spectrum. But at the same time it is more complicated because needs additional frequency shift in digital domain.

[Kleinstein]

There is also the point that with the downconversion to zero ZF there are "posisitve" and "negative" frequencies. One can tell them appart when looking at 2 signals als I and Q part. With the non zero ZF one gets the 2 separated also with 1 ADC.   

[mag_therm]

First scrot is HackRF1 with Gqrx set LO= 14.000 MHz 8 MHz sample with Decimation of 16, zoomed out to show the 500 kHz passband centred on 14.000 HMz. (+) and (-)  250 kHz about the LO. The "DC remove" is selected to put a sharp null at 0 IF.
https://app.box.com/s/cbiz4gcwitnuby2qa7sdodpz8h1rvqbv

Second scrot is zoomed into (+) side at 14.074 MHz showing the user filter set to USB  3.1 kHz to cover the ft8 sub band.
https://app.box.com/s/k0kvfde66uqku9snk5xkyzfuazvm5jul

( Just to illustrate the discussion)

[radiolistener]

There is also the point that with the downconversion to zero ZF there are "posisitve" and "negative" frequencies. One can tell them appart when looking at 2 signals als I and Q part. With the non zero ZF one gets the 2 separated also with 1 ADC.

yes, if you use simple mixer, it will leads to images mirrored from zero frequency, but if you use quadrature mixer, it can be fixed, but quadrature mixer needs good phase balance otherwise it will reduce dynamic range.

So in average it all depends on compromises between simplicity, cost, dynamic range and images performance.

[mawyatt]

The PolyPhase Mixer might be of interest, it naturally produces I and Q Baseband Channels centered around zero.

https://www.eevblog.com/forum/rf-microwave/polyphase-or-n-path-mixer/msg3381802/#msg3381802

Best,

[glenenglish]

It'll be quite difficult to get a flat response down to zero 0Hz
also boards tend to have SMPS noise and other rubblish there

1 MHz is reasonable start, but since you have Fs= 100 MHz, perhaps 10-40.
Remember to prevent alias responses, with a 40 MHz highest freq, your aliasing filter will need to be 100dB down or whatever at 60 MHz. and stay down  ... since it will alias  to hundreds of MHz.
So for that reason, to make your aliasing filter easier, choose the lowest downconversion freq you can get a nice response.
perhaps 5 to 35. So now the alias filter only needs to be 100dB down at 65 MHz. that's easier.

Also, why not just undersample ? Have you considered that? Many good converters are still pretty  good in the 2nd nyq zone, through dynamic performance does degrade.
Sample clock phase noise will be proportionally more critical....
I would consider a direct undersample. Once you introduce mixers, there are more spurious responses.....
Also, watch out for the 2f converter response. IE harmonics produced in the converter..... IE 2 x 32 MHz = 64 MHz == (100-64)= 36 MHz