Mark Levinson No.23.5 power amplifier Measurements

Sidebar 3: Measurements

One of the things that distinguishes expensive power amplifiers from the rest is that they rarely break or shut down on the test bench, even when driving 2 ohm loads at high levels at high frequencies. Such was the case with the No.23.5, which proved a powerhouse. With one channel driven, it comfortably exceeded its 23dBW power specification, the 1% distortion point being reached at 305W into 8 ohms (24.8dBW), 505W into 4 ohms (24dBW), and 750W into 2 ohms (22.7dBW), this behavior shown graphically in fig.1. With both channels driven, these figures dropped slightly, to 290W into 8 ohms and 500W into 4 ohms, but, as with the 2 ohm delivery, I feel this to be more due to the wall voltage drooping, ultimately from 116V to 110V, than to any deficiency in the No.23.5's power supply. (We do not hold the wall voltage constant during power tests, feeling that this more accurately reflects the situation in the owner's home.) The amplifier's calculated output impedance was very low, at 0.05 ohms (both channels) at 20Hz and 1kHz, rising to 0.07 ohms at 20kHz. Though this is higher than specification, experimental error could account for the discrepancy, the difference being inconsequential.

Fig.1 Mark Levinson No.23.5, distortion vs output power into (from bottom to top at 10W): 8 ohms, 4 ohms, and 2 ohms.

Fig.2 Mark Levinson No.23.5, THD+noise vs frequency at (from top to bottom): 4W into 2 ohms, 2W into 4 ohms, and 1W into 8 ohms.

Fig.1 used a signal frequency of 1kHz; it implies a measured distortion and noise much better than spec, which was confirmed by looking at how the THD and noise content of the signal varies with frequency and load. The fact that the behavior hardly changes at any frequency, other than above 5kHz as the load reduces to 2 ohms, suggests that the 23.5 will not be fazed by anything coming down the pike.

What distortion there is is almost pure third-harmonic, as can be seen by fig.3, which shows the waveform of a 1kHz tone at 135W into 4 ohms (upper trace) and the THD+Noise waveform (lower trace). The level of the distortion has been exaggerated for the purposes of plotting a readable graph; its actual level was just 0.004%! The relatively benign third-harmonic nature of the 23.5's transfer function is confirmed by fig.4, which shows the spectrum up to 1kHz of a 50Hz tone at 72W into 4 ohms. The only harmonics sticking up above the FFT noise floor are the third (–83dB (0.007%) with respect to the fundamental level) and the sixth. This characteristic behavior didn't seem to change with frequency, load, or level. Looking at the intermodulation behavior at a similar level into 4 ohms with a CD-sourced 1:1 mix of 19 and 20kHz tones revealed a slight amount of 1kHz difference product, this not present in the CD player output. The marker in fig.5 shows this to be 72dB down, however, representing just 0.025% of the signal level.

Fig.3 Mark Levinson No.23.5, 1kHz waveform at 135W into 4 ohms (top); distortion and noise waveform with fundamental notched out (bottom).

Fig.4 Mark Levinson No.23.5, spectrum of 50Hz sinewave, DC–1kHz, at 72W into 4 ohms (linear frequency scale).

Fig.5 Mark Levinson No.23.5, HF intermodulation spectrum, DC–22kHz, 19+20kHz at 48V p–p into 4 ohms (linear frequency scale).

Turning to measurements of linear changes wrought by the 23.5 on a signal, fig.6 shows the waveform of a low-level 10kHz squarewave, this excellent in shape with no ringing, and a rounded leading edge due to the amplifier's restricted ultrasonic bandwidth. This was confirmed by the small-signal frequency response (fig.7), which shows the HF output to be 3dB down at 129kHz. Rolloff at the extremes of the audio band was negligible, at – .1dB at 20Hz and 20kHz.

Fig.6 Mark Levinson No.23.5, 10kHz squarewave at 2W into 8 ohms.

Fig.7 Mark Levinson No.23.5, frequency response at 2.83V into 8 ohms (0.5dB/vertical div., right channel dashed).

Finally, the No.23.5 was non-inverting both when used single-ended via the Camac input or used balanced with pin 2 of the XLR "hot." The measured input impedance at 1kHz was 47k ohms. It had excellent unweighted audio-band S/N ratios (relative to 1W into 8 ohms) of 92/91.5dB (L/R), and, due to its dual-mono construction, had low levels of crosstalk over most of the band, though, as can be seen from fig.8, capacitive coupling from somewhere does lead to the figure rising above –98dB (L on R) and –109dB (R on L) above 1kHz. All measurements were performed after the amplifier had been driving an 8 ohm load at one-third power for 60 minutes. This thermally stresses to the limit an amplifier with a class-A/B output stage. At the end of that period, the 23.5's heatsinks were just too hot to touch, implying a temperature of 65°C or so. (The amplifier's thermal protection kicks in at 80°C.) Its voltage gain when loaded by 8 ohms was 26.8dB, implying an input of 2.25V to give full output (1% THD), 129mV to give 1W into 8 ohms. DC offsets were very low, at +2mV (L) and –2mV (R), though it should be noted that these figures were averages, there actually seeming to be a similar level of very low-frequency noise present.

Fig.8 Mark Levinson No.23.5, channel separation: L–R (top) and R–L (bottom).

With the exception of the high-frequency modulation test, which did produce some difference product (at a still low level), the ML No.23.5 is one of the best-measuring amplifiers I've had my hands on.—John Atkinson

Harman Specialty Group
3 Oak Park Drive
Bedford, MA 01730-1413
(781) 280-0300