Thanks for the link Steve. That nerd-sniped me, and I spent a few hours trying to grok the Quisk S-meter, the AD9866 LNA signal path, and thermal noise limits, and the limitations of a 12-bit A/D for an SDR receiver. All fascinating! I also checked my own antenna vs 50 Ohm dummy load.
The TL;DR: The HL2 seems to have enough resolution given the QRM on my 40m long wire antenna. Also, IIUC, an "S9" signal should be reported for any signal rising about 27 dB over the FFT noise floor.
My antenna noise appears about 4 dB over noise from my 50 Ohm dummy load, as seen in the Quisk graph. More resolution would simply increase the difference in the noise I see from my dummy load vs my antenna. So, until I move out to the country and buy a tower and Yaggi, a more expensive SDR with better resolution won't do me any good.
The S-meter is Quisk took me a while to understand. Folks should know that Quisk displays voltage, not power, so 3 dB is a factor of 2 in voltage, which is 6 dB in power, which is roughly 1 S-unit. If you see a signal rising 27 dB on the graph over the FFT noise floor, IIUC, that should be reported as S9. Does that sound right?
I found with the dummy load that on all bands, the FFT noise floor was about -136 on the main graph, with the LNA set to -12. With the dummy load, there was no increase in the noise floor as I increased the LNA gain until around -8 dB. After that, the FFT noise floor increased roughly the same as the LNA increase. However, with my antenna plugged in, HL2 set to 10M, ATU in the HR50 active, AGC off, and LNA = -12, the noise floor increased to about -132 (higher for higher bands, especially 80m). It increased linearly with LNA as I increased LNA until Quisk reported clipping at about LNA = 29 dB. So, IIUC, QRM from my antenna dominates even with LNA = -12, while internally generated noise in the AD9866 dominates from about -12 to -8 dB if the antenna input is grounded through 50 Ohms. The thermal noise generated by my 50 Ohm load should be about 5uV up to 30MHz. An A/D with a 2-volt range would need 19 bits to consistently see the LSB toggling. With the HL2's 8-to-1 transformer in the RX path, a 16-bit A/D could reliably detect the thermal noise.
BTW, my second and third Quisk receivers that I typically display using the "Add Rx" button, had a higher FFT noise floor, by about 3 dB. I suspect Quisk uses an FFT 2X smaller for all additional receivers displayed.
As for applications of libsonic, I am considering writing a hack to let me listen to two conversations at the same time. I normally listen to my screen reader at 4X speedup (about 600 wpm). I prefer listening to humans at around 3X speedup. Listening at 2X speedup, I should be able to listen to two different conversations at the same time. I'll buffer both receivers, increase the speech rate by 2X, and merge both streams heuristically, at low-signal points corresponding hopefully to the end of sentences. I'll lag behind by several seconds, but if I am not expected to respond, that should be no problem. It could, for example, let listen to a conversation on HF while taking part in a net on our local VHF repeater. I'm not sure how that will feel in my head, multi-tasking across conversations, but I'm looking forward to finding out.
73, Bill ak3q