If the signal is strongly band-limited around 40kHz you can always mix it down to audio frequency
with for instance a SA612 or Tayloe mixer and a suitable local oscillator frequency (from an Arduino
PWM pin perhaps?)
At the risk of further confusing the original poster, if the signal is inherently band-limited (or a band pass filter is employed prior to the digitizer to make it so), bandpass sampling can be employed to effectively down shift the signal without an analog mixer. That is, for a bandpass signal, one can intentionally alias it in useful ways.
The bigger point is that one is unlikely to find an Arduino-ready plug and play microphone that has flat response to/beyond 40 kHz*. What I found in a quick look at datasheets was microphone elements that had useful response up to perhaps 100 kHz, but which had higher response at normal audio frequencies. This implies that executing the project probably involves picking such a microphone element and design/build of analog bandpass filtering and amplification to condition the signal for an ADC. Specifying the bandpass filter gets back to the question of what exactly are the expected features of the source signal to be captured.