I wrote a Gnu Radio program to help measure the dongle. It tunes the dongle, sets its front-end gain, shows a spectrum display, and also shows the power in a 500 Hz bandwidth around the tuned frequency. That power reading is what I used for these measurements. The flowgraph is available at my Github repository.
With the dongle set to receive a 192 kHz bandwidth, and using an HP 8642A signal generator and 10 dB attenuator, I obtained these results at 144.2 MHz:
|0.0||-77 dBFS||-95 dBm||-15 dBm||80 dB|
|12.5||-77 dBFS||-112 dBm||-33 dBm||79 dB|
|22.9||-77 dBFS||-122 dBm||-44 dBm||78 dB|
|32.8||-77 dBFS||-131 dBm||-52 dBm||79 dB|
|42.1||-74 dBFS||-135 dBm||-60 dBm||75 dB|
|49.6||-71 dBFS||-138 dBm||-67 dBm||71 dB|
This shows more dynamic range than theory suggests. I’m not sure why. It does show clearly the impact of increasing the front-end gain: as the gain increases, the minimum discernible signal (“MDS”) gets lower, but so does the clipping level. All the gain does is move the operating range of the radio; it does not change it (much). But at gains >40, we do see the noise floor start to go up, and the dynamic range come down. That indicates that internal noise starts to become a factor, and suggests keeping the gain below that point.
By the way, I also did some spot tests at other frequencies. While there is a dB or two of difference, the results generally hold, at least through 1 GHz.