Integrated Amp Reviews

To measure the Harman Kardon HK 990, I used, as always, Stereophile's loan sample of the top-of-the-line Audio Precision SYS2722 system (see the January 2008 "As We See It" and www.ap.com). Before performing any measurements, I ran the HK 990 at one-third its maximum power into 8 ohms for one hour, which maximally stresses an amplifier's ability to shed heat. At the end of the hour the front panel was cool, but the exposed heatsinks on the top panel were at 130°F (54°C). My only problem measuring the HK 990 was the tiny legends above the front-panel pushbuttons, which I found very difficult to read. Fortunately, almost every function is duplicated on the remote control.

Kal Rubinson auditioned the HK 990's phono inputs only briefly, but I performed a full set of measurements on them, assessed at the Pre Out jacks with the volume control set to 0dB and the speakers muted. Fig.1 shows the phono stage's RIAA error. The two channels match very well, with a slight shelving down of the low bass, a relative lack of mid-treble energy, and a small ultrasonic peak. In absolute terms, the error is low at ±0.15dB. While the moving-magnet input preserved absolute polarity, the moving-coil input inverted polarity. The MC input impedance measured 82 ohms across the band. I got anomalous results with the MM input: it was an appropriately high 43k ohms at 20kHz and 81k ohms at 1kHz, but at 20Hz, instead of a voltage drop when I switched the generator source impedance from 20 to 600 ohms, I got a voltage rise, which implies a negative input impedance. The gains at 1kHz were on the low side, at 29.8dB (MM) and 50.8dB (MC), though the latter figure was affected by the low input impedance.

The MM wideband signal/noise ratio was okay, at 62dB left and 64dB right, ref. 1kHz at 5mV input; the MC ratios, ref. 1kHz at 500µV, were about 10dB worse. A-weighting improved these ratios to 73.5 and 75.5dB, respectively. Overload margins were very good across the audioband, at typically 24.5dB (MM) and 26.9dB (MC), with the minimum THD+noise at around 0.043%. The primary distortion harmonic was the subjectively benign second harmonic, which lay at –66dB (0.05%) with a 1kHz signal at 5mV. Intermodulation distortion was also relatively low; a combination of 19 and 20kHz tones with a peak level of 360mV gave rise to a 1kHz difference tone at –60dB (0.1%).

Turning to the HK 990's digital input, also assessed at the Pre Out jacks with the volume control set to 0dB, a 1kHz tone at 0dBFS gave a level of 1.88V. This gave a level of 16.75V at the speaker outputs, equivalent to a power of 70Wpc into 8 ohms, which meant that the volume control couldn't be set higher than "7dB" with its digital inputs without the HK 990 clipping. The digital input preserved absolute polarity.

I could get reliable locking via TosLink only with sample rates up to 96kHz. The coaxial inputs did lock to 176.4 and 192kHz datastreams, however. Fig.2 shows the digital input's frequency response with sample rates of 44.1kHz (cyan and magenta traces) and 96kHz (blue and red). Note the level mismatch in this graph: the right channel is 0.4dB lower in level than the left. The response follows the same basic shape with both sample rates, but with a sharp cutoff at just below half of each rate. The Direct Path is locked out with digital inputs and the DSP Path appears to be limited to 96kHz. So while the Analog Devices AD1955 DAC chip used in this amplifier will decode data with sample rates of up to 192kHz, the frequency response with 192kHz data was identical to that with 96kHz. Channel separation via the digital input was good, at 100dB in both directions below 300Hz, though this decreased to 65dB at 20kHz, due to the usual capacitive coupling between channels.

Fig.3 shows my usual test for resolution, performed by sweeping a 1/3-octave bandpass filter from 20kHz to 20Hz while the DAC decodes dithered 16-bit data (top pair of traces) and dithered 24-bit data (middle traces) at –90dBFS, each data set representing a 1kHz tone. The increase in bit depth drops the noise floor by 10dB or so in the treble, but there are peaks evident in the left channel at the power-supply–related frequencies of 60 and 180Hz. These obscure the left channel's reproduction of a dithered 24-bit tone at –120dBFS, but the tone is resolved only in the right channel (bottom solid and dotted traces, respectively). FFT analysis confirms these results (fig.4), and makes visible some very low-level enharmonic products with 24-bit data. Linearity error with 16-bit data was minimal to below –105dBFS, and the HK 990's combination of low noise and high resolution allowed the three DC voltage levels that describe an undithered tone at exactly –90.31dBFS to be readily resolved (fig.5). With 24-bit data at the same level, the Harman Kardon provided a good facsimile of a sinewave (fig.6).

Commendably, the HK 990's digital decoder was free from noise modulation. However, fig.7 shows that there was a little more harmonic distortion than I was expecting, a 50Hz tone at 0dBFS being accompanied by second-harmonic distortion at –80dBFS (0.01%) and third-harmonic at –87dBFS (0.04%). These figures were for the right channel (red trace), again assessed at the Pre Out jacks. The left channel (blue) had a little less second harmonic, but overall these levels of distortion are still low. Intermodulation distortion was also low, and the HK 990 offered good rejection of jitter, any sidebands accompanying the Fs/4 tone with the J-Test signal lying at the residual level of the odd-order harmonics of the Fs/192 squarewave (fig.8).

Moving on to the HK 990's measured performance as a conventional integrated amplifier, the volume control operated in accurate 1dB steps, with a maximum gain of 9.8dB measured at the Pre Out jacks, and 40.65dB into 8 ohms measured at the speaker terminals. These were for unbalanced signals; the maximum gain for a balanced input was 34.3dB. Both preamp and speaker outputs preserved absolute polarity (ie, were non-inverting). The unbalanced input impedance was 42k ohms at low and mid frequencies, dropping slightly to 30k ohms at 20kHz. The balanced impedance was twice this value, as expected.

The output impedance from the Pre Out jacks was a low 169 ohms at all frequencies. The output impedance from the speaker terminals, including 6' of cable, was a very low 0.085 ohm, rising slightly at 20kHz to 0.13 ohm. The modification of the amplifier's frequency response due to the Ohm's Law interaction between this impedance and that of our standard simulated loudspeaker (see www.stereophile.com/reference/60/index.html) was very small, at ±0.1dB (fig.9, gray trace). The HK 990's response in its Direct Path mode rolls off gently above the audioband, but its reproduction of a 10kHz squarewave featured short risetimes and an absence of overshoot or ringing (fig.10). The 1kHz squarewave was essentially perfect (fig.11).

The DSP Path offers Bass and Treble controls; the effects of these, set to their maximum and minimum ±10dB positions, are shown in fig.10. These are shelving controls rather than usual Baxandall type, and the maximum boost and cut are each exactly 10dB (blue and red traces). The sharp cutoff above 40kHz in the graph indicates that when the DSP Path is invoked, the HK 990 digitizes analog input signals at 96kHz or so. The green and gray traces in fig.11 show the response with the subwoofer crossover switched into circuit; the output is now –3dB at 40Hz, and the ultimate slope of the high-pass filter is 12dB/octave.

Channel separation for line-level inputs was good at 90dB or greater below 4kHz, and still 80dB at 20kHz. The wideband, unweighted S/N ratio, measured in Direct Path mode with the input shorted but the volume control at its maximum "10dB" setting and ref. 1W into 8 ohms, was good, at 67dB. A-weighted, this improved to 77.6dB in the right channel and 75.9dB in the left. The main noise components were supply-related spuriae at 60Hz and its odd harmonics (fig.12), which suggests that these arise from magnetic interference from the twin AC transformers.

Specified as delivering 150Wpc into 8 ohms (21.75dBW), the H/K delivered 190W at clipping into 8 ohms with both channels driven (22.8dBW), with clipping defined as 1% THD+N (fig.13). Again with both channels driven, the HK 990 met its 300Wpc specification into 4 ohms (21.75dBW, fig.14). With one channel driven, the amplifier clipped at 450W into 2 ohms (20.5dBW, fig.15). Fig.16 shows that the HK 990's THD+N figure is dominated by noise below 40W or so. Fig.17 plots the THD+N percentage against frequency at 10V; the distortion starts to rise out of the noise only in the top octave.

With an analog input signal, even at high powers into 4 ohms, the distortion spectrum is predominantly the subjectively innocuous second and third harmonics (figs.18 and 19). The third is the highest in level, at –90dB (0.003%), which is still very low in absolute terms. Intermodulation distortion is also extremely low; fig.20 shows the spectrum of the amplifier's output when it drove an equal mix of 19 and 20kHz tones at 120Wpc into 4 ohms. The 1kHz difference product lies below –100dB, and the higher-order products at 18 and 21kHz are almost 90dB down from peak level.

Harman Kardon's HK 990 is a powerhouse of an amplifier with, overall, superb measured performance. While high-quality outboard phono stages and D/A processors will give better measured performance than the HK 990's equivalents, there is no other sign of any aspect of its performance having been compromised to include so many versatile features. I am impressed.—John Atkinson