After perusing the vast store of information at minidisc.org, I got an Aiwa AM-C80 MD recorder from minidisco.com on Aug 7, 2000. This is one of the few pocket minidisc recorders which offer the option of using either AGC or manual level control while recording. Here are some measurements of its audio response.
Equipment: I used a HP 89410A (DC-10 MHz vector signal analyser) and Agilent 34401A true RMS meter for the analog recording and playback measurements. I also used the Sony MDS-JB930 to play the recorded tracks into my computer via Fiji soundcard S/PDIF digital in, and analysed the digital signal with Cool Edit 2000.
Recording: I set the 89410A signal source to 1 kHz and connected that to the input of the Aiwa AM-C80, set to "line input" level. The Aiwa's manual record level runs from 0 to 20. With a manual recording level set at 5, a 1 kHz sine at +8 dBV (+10.2 dBu) just reached an indicated 0 dB on the Aiwa's level bargraph (without an "over" indication of clipping). I recorded a minute of this tone, then substituted a 50 ohm dummy load for the signal source and recorded that (nominally silence) on another track with the same record level.
Analog Playback: I played back the two tracks into the analyser with the headphone output set at 14 (out of 20). I used 14 because any higher level caused increased distortion. You can see the signal source direct, the recorded signal, and the recorded silence in these three graphs from the analyser. I also measured total signal power from the two tracks with the Agilent multimeter in AC-RMS mode. You will note that the dominant distortion is the third harmonic at 65 dB down from the 1 kHz tone, and the total signal power from the RMS meter was -10.6 dBV (tone) and -65.2 dBV (silence), for a signal-to-noise ratio (SNR) of 54.6 dB. The main limitation is the headphone amplifier, which had about -65 dBV output even with the disc stopped. With the unit turned off, the meter read -110 dBV, so this wasn't external measurement noise.
Digital Playback: I suspected the signal actually recorded on the MiniDisc had higher dynamic range than the Aiwa's headphone amplifier could reproduce, so I put the disk into my Sony deck and played the digital signal into my computer. Cool Edit 2000 (from Syntrillium.com) showed that the 1 kHz tone spectrum (-1 dBfs) is cleaner, with the 2nd harmonic dominant at -76 dBfs (...which is exactly what the analyser reported as the distortion of its own source, connected in direct loopback.) The noise spectrum from the silent track has a total power of -93 dBfs, so the actual recorded signal on the disc has a 92 dB dynamic range. This is 37 dB better than the headphone output! It is interesting to note the shape of the noise floor in the tone spectrum, where some of the frequency bins of the ATRAC algorithm show up as a "plateau" in the noise. Notching out 1, 2, and 3 kHz from the "tone" spectrum yields a residual noise power of -82 dBfs, or a 81 dB SNR.