Reference

Power amplifiers fascinate me. In the past 15 years I've helped design and build over a dozen advanced models, with output powers ranging from 50 to 2500W, for a number of the UK's professional equipment manufacturers. To learn from others' ideas and mistakes, I've also repaired, measured, used, and reviewed hundreds of makes of amplifier. My experiences have led me to regard the power amplifier as one of the messiest, most imperfect pieces of electronic equipment in the record/replay path.

The Setting Apart
Why should this be so? Creating good-sounding mixing consoles, preamplifiers, volume controls, D/A converters, and crossovers is difficult enough (!). Still, there are few really ugly engineering problems; the attainment of good objective performance (a good baseline) should be relatively straightforward, leaving the designer to concentrate 70% of his mental budget on sonics, system overview, producibility, reliability, cost, and the like. Compared to a properly engineered power stage, a line-level audio stage is relatively simple, with even the worst-case load conditions defined.

Power amplifiers, on the other hand, are hooked up to complex, ill-defined loads while being asked to deliver a thousand times as much current as any other audio circuit. This is a key factor, as both electromagnetic interference and impedance-related interactions between supposedly separate conductors and circuits are proportional to the currents involved, while heating and consequent thermal interactions are proportional to the square of that current.

Moreover, the physical layout of any finely tuned amplifier design is also fundamental to its performance. (A pure circuit diagram without mechanical drawings, or even an amplifier schematic bereft of its power supply, is nothing more than audio pornography.) There are few nonessentials in power amplifiers; everything influences everything else. This "all-come-at-once-ness" not only adds considerably to the development difficulties, hence cost, but also slows the overall evolution of up-market power-amplifier design.

Unraveling the Residue
Over the past decade, it has been established in DIY audiophile circles that upgrading component parts in any reasonable amplifier design can change and generally improve sonics, often greatly—even though the circuitry itself remains unchanged. As nothing has been forthcoming from the scientific and academic community as to why this should be, I undertook an experiment of my own. I would investigate the influence of passive components on the measured performance of a typical amplifier.

My experimental field was Audio Precision's System One Dual Domain DSP test set, the "Senior Model." Since this device's introduction in the fall of 1988, a growing body of users has been able to routinely measure the levels of individual distortion harmonics, even with modern semiconductor electronics where the harmonics are generally below 0.1% and substantially buried beneath the noise floor.

Traditionally, THD (Total Harmonic Distortion, or everything that remains in a device's output after the fundamental signal has been removed) has been measured as a percentage of that fundamental signal and typically plotted vs frequency. Such tests may have value in design and review, but are basically meaningless when it comes to predicting a component's sound quality: the audibility of THD does not correlate with our perception when the levels are below 0.1% (footnote 1). But individual harmonics are different, and their order—the harmonic structure—can offer a fair correlation with what we hear (footnote 2). Correlation can be improved by referring to typical acoustic levels, then overlaying "hearing sensitivity curves" derived (for example) from Robinson and Dadson (footnote 3). Another UK designer has worked on producing still more refined psychoacoustic overlays (footnote 4) in order to model masking and other perceptual attributes. But this is a subject to itself; for the purposes of showing a meaningful difference, I felt it was sufficient to look directly at the harmonic structures of the devices under test.

The subject for my experiment was the Rauch DVT 50s 200Wpc professional amplifier. I had been involved with the design of this class-AB MOSFET unit in 1986, and it has a fairly simple signal path. For a number of years, I had supplied audiophiles with a stock upgrade which produced an improvement in sound quality that had been well received (footnote 5). It was obvious that this Rauch was ripe for use as a subject in my quest after differences—before and after modification.

The two channels of a production sample of the amplifier were initially measured with the Audio Precision Dual Domain to confirm that they were substantially matched. One channel was then upgraded as follows:

• All the 1% metal-film resistors were replaced with a highly specified type made by Holsworthy Electronics in the UK.

• The indifferent electrolytic caps were replaced with established audiophile-grade models.

• The IC op-amp (a reputable low-noise type, itself much better than the standard NE5534) was replaced by a damn fast (1000V/µs), newly released current-mode op-amp made by Linear Technology.

• Finally, the DC supply leads were unwound from the wiring harness, separated from ground, and mutually twisted; ditto the output pairs. Both steps were taken to reduce interaction, although the original loom did not exceed 7" in length. No other changes were made.

Comparing Data
Fig.1 shows the manner in which the percentage of Total Harmonic Distortion plus Noise (THD+N, footnote 6) in the amplifier's output varies with frequency, before and after modification (dashed and solid traces, respectively). The amplifier was driving an 8 ohm resistive load, about 0.6dB below its clipping point. Before modification, the curves for the two channels were nearly identical, close enough to be interlaced. It can be seen from fig.1, however, that the modified channel's THD+N curve is higher overall than the unmodified unit's. This need not mean that increased distortion per se is the sole reason for the difference. In fact, the fact that the replacement op-amp is 12–15dB noisier is bound to have an effect on this measurement. The only hint of the sonic improvement offered by the upgrade is a slight change in curvature at the edges of the modified unit's plot (top, dashed curve). Here, the rate of increase of THD (and/or noise) with frequency looks like it will level off and perhaps fall at ultrasonic frequencies, whereas the unmodified unit's THD+N (bottom, solid curve) looks like it will skyrocket above 30kHz.

Fig.2 uses the Audio Precision's DSP facility to plot the second to seventh harmonics of the unmodified channel, again driven into 8 ohms, at a fractionally lower level of 2dB below the amplifier's clipping point. In general, the spectra could be termed chaotic, noise residue making the plots below the 0.001% point erratic. But the general picture is quite clear: the second harmonic dominates, followed by the third, below 3kHz—excepting a sporadic dip and change in order between 600 and 700Hz. Above 3kHz, the fourth and fifth harmonics dominate, giving way above 18kHz to the third (footnote 7). Above 7kHz, analysis is complex, as the seventh harmonic is then the next most dominant, but the seventh harmonics of frequencies above 7kHz lie above 49kHz. The effect of these products will not be directly audible, but they can create unpleasant intermodulation products in the audible band.

By comparison with fig.2, fig.3 shows that the modified amplifier's harmonic structure has become much cleaner, even taking into account the fact that harmonic levels are higher and thus lifted away from the noise-floor. Now the subjectively innocuous second harmonic can be seen to dominate clearly at all frequencies. The even fourth and sixth harmonics are clearly the next most dominant, respectively. Overall, the even-order (second/fourth/sixth) harmonics dominate the structure; moreover, they are in their natural sequence, the lowest orders being the stronger. The odd harmonics aren't so naturally ordered; eg, the third is weaker than the fifth and seventh below 8kHz.

The upshot is a fairly nice-sounding amplifier: the benign, dominant second harmonic is a clear leader, up to an order of magnitude stronger than the others. The order of the harmonics in the unmodified channel's output (fig.2) is similar below 3kHz, though the third and fifth are now stronger than the fourth. The unmodified amplifier's sound changes at high frequencies to a dominant fourth/fifth combination, which is less pleasant while not being unlistenable.

The complexities of describing and analyzing these data are less important than the hard copy itself. An audible, "subjective" (footnote 8) change has now a clear difference mirrored in the "objective" test data. The difference is complex but glaring.

Residue Studied...
Recall the late '70s, when Japanese manufacturers' glossy advertisements promised amplifiers with THD+N levels of 0.0002% or lower? In practical production, and at power levels ranging into hundreds of watts into 4 ohms and below, you can usually attain figures of this magnitude only with carefully trimmed high or multiple feedback circuitry, and then usually only at 1kHz and below.

---

Footnote 1: This was pointed out by Stereophile's founder, J. Gordon Holt, some 25 years ago. In addition, Stereophile's Test CD 2 has a number of tracks comparing the audibility of large and small amounts of different harmonics: the second, third, and seventh. I prepared the signals for these tracks entirely in the digital domain using the Audio Precision System One Dual Domain.—John Atkinson

Footnote 2: Russell O. Hamm, "Tubes vs Transistors—Is There an Audible Difference?," JAES, May 1973.

Footnote 3: Mark Burgin & Ben Duncan, "Dynamic Loudness Compensation," Proc. IOA, Vol.8, 1987. See also Louis Fielder, "Human Auditory Capabilities and Their Consequences in Digital-Audio Converter Design," AES 7th International Conference, Toronto, May 1989, Paper 4A, in which Mr. Fielder shows that the order of a distortion harmonic is as important if not more so than its absolute level when it comes to audibility.

Footnote 4: J.R. Stuart, "Psychoacoustic Models for Evaluating Errors in Audio Systems," Proc. IOA, Vol.13 Pt.7 (1991). See also "Estimating the Significance of Errors in Audio Systems" and "Predicting the Audibility, Detectability and Loudness of Errors in Audio Systems," two papers presented by Bob Stuart at the 91st Convention of the Audio Engineering Society, held in New York in October 1991.

Footnote 5: Chris Wood, "Modifications to the Rauch DVT-50s," Audio Conversions (UK), No.14, Spring 1992.

Footnote 6: The two cannot be readily separated with standard test instruments.

Footnote 7: The harmonic levels in figs.2 and 3 are plotted up to the point where they reach the AP's 80kHz filter limit. For example, the measured amount of fourth harmonic is not plotted above 20kHz because the fourth harmonic at 20kHz is 80kHz.

Footnote 8: Quotation marks because the word has been debased to suggest rather flaky or romantic perception, rather than "human perception unfettered by instruments."