Do you have any data to back up or to deny my suspect that it has a fixed
cut off frequency in its anti alias input filter?
The short story: All 192kHz capable audio ADCs that I have seen have anti-aliasing filters which are less than ideal from a Linrad point-of-view.
Background: I'm looking into adding WBFM support. To do this comfortably, you'd need more bandwidth than the 80..90kHz you effectively get with a 96ksps soundcard[1]. I'm using ksps (kilo-samples per second) here, to reduce confusion between bandwidth and sampling rate[2].
The logical step would be to use a 192ksps soundcard. However, while good audio ADCs have a digital anti-aliasing filter with a transition band less than 5kHz wide in their 48ksps and 96ksps sampling modes, the transition band (from -3dB to -100dB) can easily be as wide as 48kHz ! This means that, for sampling complex signals, a 192ksps soundcard will only offer 2*((192ksps/2[Nyquist]) - 48kHz) = 96kHz of of usable bandwidth (where usable is defined as having an adjacent channel rejection of >100dB).
Why is this the case ? As far as I can tell, it's because they are *audio* converters, and no-one seems to care if 'audio' between 96kHz and 144kHz aliases into the band between 48kHz and 96kHz. For our purposes it's more harmful.
Why don't the soundcard tests pick up on this ? Because (a) as noted by someone else, many if not most soundcard tests only consider the 0-20kHz audible range, and (b) those nice programs you can use to test your own soundcard by looping it back to itself are by definition unable to measure aliasing. For linrad usage, you should really sweep a sine generator from zero to beyond the Nyquist frequency of a given converter to determine its aliasing behaviour.
Of course, none of this matters if your receiver's (analog) RF/IF/AF filters are steeper and/or narrower than your ADC's anti-aliasing filter...
By the way, I am thinking about doing a backend for a linrad/SDR-1000 like receiver with A/D converters integrated in the final stage (behind the I/Q demodulator). Not only does this allow to put the ADC in the mast (running S/PDIF-like signals into the shack over Cat5 or fiber), but it also allows hard synchronisation between the sample clock and other receiver clocks, reducing spurs and drift.
Regards, JDB.[1] You *can* get 96kHz (-48...+48kHz complex bandwidth) from a stereo-sampled I/Q signal on a 96kHz soundcard, but only if you're either willing to accept aliasing noise at the band edges, or if you can build a dual tracking anti-aliasing filter sharper than the sound chip's digital filter (think > 250dB/octave for a good converter).
[2] And even 'sampling rate' is a misnomer with modern audio ADCs, seeing that they all use some form of sigma-delta conversion, which (very simply said) is an oversampling lower-order ADC (sometimes 1-bit, sometimes multibit) inside a feedback loop, combined with a digital filter/decimator to produce the output samples.
-- LART. 250 MIPS under one Watt. Free hardware design files. http://www.lart.tudelft.nl/LINRADDARNIL