Hovland Radia power amplifier Measurements
As the gorgeous-looking Hovland Radia has a balanced output stage, I took care during the measurements to ensure that the speaker terminals were not connected to ground or to each other. After the usual 60-minute preconditioning period at 1/3 power into 8 ohms, with both channels driven, the Radia's chassis was warm, and the circular perforated metal grilles over the internal heatsinks were too hot to touch.
The review sample of the Radia had the optional balanced inputs, with XLR jacks. The amplifier has low gain for a modern power amplifier, at 19.4dB into 8 ohms, which is 6dB lower than specified. A preamp will need, therefore, to be able to output up to 4V to drive the Radia to its maximum output. The input impedance at 1kHz was 46.5k ohms for each phase of the balanced input, giving a usefully high 93k ohms overall. The amplifier preserved absolute polarity, with pin 2 of the XLR jacks wired as "hot."
The output impedance was a low 0.2 ohm across most of the band, this rising inconsequentially to 0.26 ohm at 20kHz. As a result, the modification of the Radia's frequency response by the Ohm's Law interaction of its source impedance with the speaker impedance was less than ±0.25dB with our simulated speaker load (fig.1, top trace). The amplifier offered a wide small-signal bandwidth into 8 ohms, with a -3dB point above 100kHz that ensured excellent squarewave performance (fig.2). However, this bandwidth was increasingly curtailed as the load impedance dropped, as can be seen from fig.1. Into 2 ohms, for example, the Radia's output was 3dB down at 55kHz, -0.5dB at 20kHz. This is probably inconsequential, but it suggests some modulation of the amplifier's operating parameters by the output-stage current.
Fig.1 Hovland Radia, frequency response at (from top to bottom at 2kHz): 2.83V into simulated loudspeaker load, 1W into 8 ohms, 2W into 4 ohms, 4W into 2 ohms (0.5dB/vertical div., right channel dashed).
Fig.2 Hovland Radia, small-signal 10kHz squarewave into 8 ohms.
Channel separation at 1kHz (not shown) was 105dB in both directions, which is excellent. This decreased at 6dB/octave above that frequency by what appeared to be the usual capacitive coupling, but was a still good 82dB at 20kHz. Below 1kHz, power-supply-related spuriae at 60Hz and its harmonics obscured the crosstalk measurement, though these lay at what might be thought an inconsequential -95dB. As a result of these spuriae (which I could not eliminate by experimenting with the grounding between the Radia and my Audio Precision System One), the Hovland's wideband signal/noise ratio (ref. 1W into 8 ohms) was a modest 66.7dB. This improved to 87.8dB when A-weighted.
Fig.3 shows how the Radia's THD+noise percentage changes with output power when both channels are driven at 1kHz into loads ranging from 16 ohms to 4 ohms, and with one channel driven into 2 ohms. The amplifier more than gives out its specified power into 8 ohms, with 195W (22.9dBW) available at the clipping point (defined as 1% THD). The maximum power increases to 290W into 4 ohms (21.6dBW), which is again higher than specified, though the fact that this is less than double the 8 ohm delivery suggests that the power supply has a higher-than-expected source impedance. The maximum power drops to 230W into 2 ohms, with higher levels of distortion apparent above 20W into this load. The internal 5A rail fuses blew when I performed this test, which, in conjunction with the 2 ohm distortion results, suggests that speakers with impedances below 4 ohms are best avoided with the Radia.
Fig.3 Hovland Radia, distortion (%) vs 1kHz continuous output power into (from bottom to top): 16 ohms, 8 ohms, 4 ohms, 2 ohms.
I assessed how the measured THD+noise percentage changed with frequency at 4V output, the level where the actual distortion harmonics begin to rise out of the noise floor. The results are shown in fig.4, with traces taken into 16, 8, 4, and 2 ohms. The distortion percentage at 1kHz is respectably low, just reaching 0.1% into 2 ohms. However, THD can be seen to rise at both high and low frequencies, with the latter curves marked by a notch at 120Hz. This is due to cancellation between the signal frequency and the residual 120Hz component of the noise, which persisted no matter how I arranged the test setup grounding, and even when I used battery-powered test gear.
Fig.4 Hovland Radia, THD+N (%) vs frequency (from bottom to top): 2.83V into 16 ohms, 8 ohms, 4 ohms, 2 ohms (right channel dashed).
Fig.5, taken at 1kHz, indicates that the Radia's residual distortion consists of both high- and low-order harmonics. Because of the human ear's masking characteristic, the former are the more audible, and can be quite disturbing even at low measured levels. Taken in conjunction, figs.4 and 5 again suggest that the Radia not be used with speakers that have low impedances, at least in the upper midrange and treble. But what about the increase in distortion at low frequencies seen in fig.4?
Fig.5 Hovland Radia, 1kHz waveform at 2W into 8 ohms (top), 0.0274% THD+N; distortion and noise waveform with fundamental notched out (bottom, not to scale).
I assess an amplifier's low-frequency linearity by driving it with a 50Hz tone at around two-thirds its rated output, then examining the spectrum of its output. Fig.6 shows what happened with the Radia when I tried this: an alarming picket fence of both harmonic and intermodulation products! Intermodulation? With a single tone? Look at the small peak lying at -54dB to the right of the 50Hz fundamental. At first I thought this was 60Hz hum, but I was wrong—the peak has a frequency of 70Hz, which is the difference between the 50Hz test signal and 120Hz, the frequency of the notch in the fig.4 traces. Similarly, all the other spikes in the fig.6 spectrum are mathematically related to the two frequencies of 50Hz and 120Hz.
Fig.6 Hovland Radia, spectrum of 50Hz sinewave, DC-1kHz, at 67W into 8 ohms (linear frequency scale).
To check this pathological behavior, I drove the Radia with other frequencies, and in each case got a slew of intermodulation products related to the signal frequency and 120Hz. Fig.7, for example, shows what happens with a 100Hz input: sum and difference components appear at 20Hz and 220Hz, with many other intermodulation products apparent. But with this input frequency, true harmonic-distortion products can also now be seen. The third harmonic (300Hz) and the fifth (500Hz) both lie at around -70dB (0.03%), with the seventh at -80dB (0.01%) and the ninth at -86dB (0.005%).
Fig.7 Hovland Radia, spectrum of 100Hz sinewave, DC-1kHz, at 67W into 8 ohms (linear frequency scale).
A similar picture emerged when I tried a 1kHz test signal (fig.8). The 1kHz peak can be seen to be surrounded by 120Hz-related intermodulation spikes, with a regular series of harmonic spikes extending out to the ninth. And with my usual intermodulation test—in which I drive the amplifier under test close to its clipping point with an equal mix of 19kHz and 20kHz tones (fig.9)—while the actual signal intermodulation products are respectably low, again, the plethora of 120Hz-related sidebands appear.
Fig.8 Hovland Radia, spectrum of 1kHz sinewave, DC-10kHz, at 11W into 8 ohms (linear frequency scale).
Fig.9 Hovland Radia, HF intermodulation spectrum, DC-24kHz, 19+20kHz at 210W into 4 ohms (linear frequency scale).
Usually, when I encounter this kind of behavior—though it has never been nearly as extreme as with the Radia—I suspect a grounding incompatibility between the amplifier under test and the Audio Precision test setup. But, as I said, I experimented with every possible combination of grounding—from no ground connection at all to a direct grounding strap between the Audio Precision's reference ground and the Radia's chassis—with no change in the Radia's behavior. I also tried battery-powered measurement gear, so that the only connection between amplifier and ground was via the Radia's AC cord. Again, no change in behavior.
The 6dB difference between the specified and actual voltage gains was suspicious, in that it's what you might expect if one side of the balanced input was not connected. But this didn't appear to be correct because: 1) both channels behaved identically; 2) driving one side of the XLR input and grounding the other didn't produce any different result; 3) driving the amplifier with a balanced signal produced identical but opposite-polarity voltages from each speaker terminal when referenced to ground, and 4) the amplifier could deliver its full output voltage.
Assuming the Radia was behaving correctly—I wait to hear from Hovland on that matter—what could be predicted about its sound quality from these measurements? Interestingly, at the start of his auditioning comments, Paul Bolin described pretty much what I would have expected: "a definite lack of authority...haziness and insubstantiality in the upper treble—a somewhat gauzy, misty character....The lowest bass was soft, and the transition from the upper bass to the lower midrange was even more so." But this was before the amplifier had broken in, begging the question of whether the "break-in period" was really Paul getting used to what the Radia does wrong and learning to appreciate what it does right.—John Atkinson
As Paul Bolin describes, after we had finalized the Radia review and sent the manufacturer the preprint for them to prepare a "Manufacturer's Comment" for publication, Alex Crespi of Hovland contacted me. He informed me that the 120Hz hum problem had been resolved and that our review sample of the amplifier was therefore not representative of current production.
My policy in such cases is to postpone publication of the review to allow us to audition and measure a second sample, which is what happened here. PB's auditioning comments apply in the main to the second sample. I am appending my measurements of that sample to those of the first.
The second sample had virtually the same low gain as the first: 19.6dB vs 19.4dB, both measured at 1kHz. Its input and output impedances were the same as the first, as was its frequency response. This sample's maximum output power was slightly but insignificantly higher, at 300W into 4 ohms vs 290W, for example, but it showed a similar drop of power into 2 ohms.
The fundamental difference between the two samples was that the second one (serial no. 06105803) did not suffer from the first's (serial no. 03105103) 120Hz hum. This can be seen from the second sample's plot of THD+noise against frequency (fig.10). Comparing this graph with fig.4, the amplifier now offers much better linearity in the bass and midrange. However, the increase in THD in the high treble and above the audioband is still evident.
Fig.10 Hovland Radia, sample 2, THD+N (%) vs frequency (from bottom to top): 4V into 16 ohms, 8 ohms, 4 ohms, 2 ohms.
Spectral analysis of 1kHz and 50Hz tones at reasonably high powers (figs.11 and 12, respectively) are virtually free from the 120Hz power-supply component and the resulting intermodulation products that could be seen in figs.6 and 8. However, the 1kHz spectrum (fig.11) still reveals the presence of high-order harmonic components.
Fig.11 Hovland Radia, sample 2, spectrum of 1kHz sinewave, DC-10kHz, at 67W into 8 ohms (linear frequency scale).
Fig.12 Hovland Radia, sample 2, spectrum of 50Hz sinewave, DC-1kHz, at 67W into 8 ohms (linear frequency scale).
The amplifier's spectrum just below clipping with a mix of 19kHz and 20kHz tones is shown in fig.13. That the amplifier is working very hard with this signal and load is revealed by the existence of 120Hz-spaced sidebands around the primary tones, but these are very much lower in level than in fig.9; overall, the spectrum is much cleaner.
Fig.13 Hovland Radia, sample 2, HF intermodulation spectrum, DC-24kHz, 19+20kHz at 180W into 8 ohms (linear frequency scale).
Judging by the second sample of the Radia, Hovland's first solid-state amplifier gets a clean bill of health (as long as it's not used to drive speakers with impedances dropping much below 4 ohms). Yes, those higher-order harmonics in its output still bother me a little, but they shouldn't keep anyone from giving the gorgeous-looking Radia a listen.—John Atkinson