by Luca Chiomenti

In 1994, on the occasion of the public presentation of my first important industrial project in the audio world, I wrote these lines that still remain relevant today: ‘It is our conviction that high fidelity, in the last 20 years, has ended up in a blind alley. We are witnessing a sterile search for technical perfection as an end in itself. For too long, an analysis aimed only at reaching the limits of measuring instruments has been proposed. This has made us lose sight of the very function of an audio amplifier: that of reproducing musical signals through an electroacoustic transducer.’
Many years later, the situation seems even more confusing. Let me explain. After a period of blind faith in measurements (the 1970s and 1980s), the opinion had spread that the measurements normally used to test an audio amplifier had no real correlation with the perceived musical quality (late 1990s to early 2010s): for years, audiophiles refused to consider technical specifications and laboratory tests as indicators of actual musical quality.
Today, this is still the case for some. For others there seems to have been a resurgence in measurement, with a widespread perception that ‘flawless’ instrumental results are a prerequisite for any device being put on the market.

I cannot adequately summarise over 35 years of my studies and research into the correlations between listening sensations and laboratory measurements. But at the basis of my research are both experiments carried out directly and hundreds (thousands) of pages of bibliography relating to studies carried out over the last 80 to 90 years.
I hope sooner or later to be able to complete a more in-depth treatment that I have been preparing on the subject for some time. For now, I would like to limit myself to giving a few remarks on the salient points underlying the Riviera project. It is necessary to go back to the beginning, to the origins.
An audio amplifier must reproduce an audio signal as faithfully as possible… for a listener, for a human being. Not for measuring instruments. This is the point. It is fundamental: 1) to understand certain aspects of the functioning of the human auditory system and, consequently, 2) to identify which characteristics the reproduced signal must have for the ear and not for an electronic measuring system. Let us begin with the functioning of the ear.

• Numerous studies have verified that when listening to a pure tone, harmonics not present in the original signal (i.e. harmonic distortion) are generated inside the ear and specifically in the cochlea. This is certainly not a new discovery. The first investigations into this distortion date back to Fletcher (yes, the famous Fletcher-Munson curve, in the 1920s); more accurate analyses were made by H.F. Olson (Acoustics, 1947) and then by many others to follow. It is interesting to note how high the level of distortion (2nd harmonic) generated by the ear is: about 10% for pressures of 90dB (not 120dB, 90dB!). For higher order harmonics, the level decreases in proportion to the order of the harmonic and one can arrive at a spectrum of ear distortion. The shape of this spectrum, i.e. the distribution of these harmonics, is extremely important: there is a strong predominance of low-order harmonics and the spectrum has a decreasing trend with frequency. The spectrum then changes as a function of the pressure at which the measurements are taken, but this point becomes a little too complex to be discussed in detail here. The key points, however, are these: 1) the ear generates a high level of harmonics within it and 2) the auditory system (ear+brain) cancels these harmonics and we have the perception of a pure sound (like the original). In other words, the auditory system is able to suppress the harmonics it generates within itself. Interesting, no? But that’s not enough, there’s something even more interesting: the system cancels out the contribution of ‘that’ distribution of harmonics even if these harmonics have an ‘external’ origin, as long as the shape of the distribution is maintained. It’s logical, if you think about it: the system is designed to cancel a distortion with that shape and is unable to distinguish whether the origin of the distortion is internal or external to the ear (there are well-known phenomena in the field of music relating to this behaviour, one of which, for example, is that of the missing fundamental). If, on the other hand, the distribution of harmonics has a different shape from the ear, the auditory system perceives the harmonics as different sounds and recognises them. Based on this behaviour, it is our opinion that an amplifier that generates a distortion spectrum similar to that of the human ear is extremely transparent and clean to the ear, even if its measurable THD value is relatively high.
• The mechanism now mentioned depends on the sound pressure level. In short, the higher the acoustic pressure, the higher the distortion generated by the ear (and therefore acceptable, if it has the right shape). This leads us to the conclusion that amplifiers with a monotonically increasing distortion rate with output power are preferable.
• As the sound pressure increases, higher order harmonics also increase in the distortion generated by the ear. This means that at higher power levels a somewhat higher amount of higher order harmonics can be accepted.
• Perceivable’ distortion depends, among other things, on: 1) the ratio of the peaks to the average signal level; 2) the duration of each signal peak. There are studies showing that distortion can reach even very high values without being audible, as long as this occurs in peaks of sufficiently short duration.
• Masking is that well-known phenomenon whereby a lower level signal that is very close in frequency to a higher level signal is inaudible. This may lead one to consider the influence of intermodulation distortion less important in an amplifier with a distortion characteristic where low-order harmonics prevail. This effect would seem to be even more pronounced if the amplifier’s distortion spectrum is very similar to that of the human ear.
• Counter-reaction (negative feedback) is the classic technique used to reduce THD and improve the overall performance of a circuit (THD, IMD, bandwidth, noise and more). Unfortunately, feedback reduces low-order harmonics (less harmful and more acceptable to the auditory system) much more than it does high-order harmonics. In other words, it distorts the distortion spectrum making it very different from that of the human ear. It gets worse: high rates of feedback create a mechanism of ‘generation and multiplication’ of high-order harmonics (harmonics that the auditory system perceives as dissonant and annoying). Even if the level of these high-order harmonics is below the threshold of audibility, the mechanism ends up generating a ‘carpet of noise’ that seems to be very unpleasant to the ear. All this leads us to eliminate the use of feedback as much as possible (especially of total or overall feedback) and to try to reduce its use to a minimum.

Starting from these basic principles, we came to define the technical characteristics that an amplifier designed ‘for’ and ‘dedicated to’ the human ear should have. It should be noted that many of the classic measurements (harmonic distortion, intermodulation, etc.) were proposed in the 1920s and 1930s (almost 100 years ago), only a few such as TIM and DIM in the 1970s (50 years ago, in any case) but there was never a critical review process to verify the real correlations between those measurements and listening. Riviera amplifiers were only optimised for those measurements that showed a real correlation with listening feedback, without seeking unnecessary technical virtuosity. We therefore concentrated on these main points.

• Optimisation of amplitude and frequency distortion: THD does not necessarily have to have extremely low values, but it absolutely must have an optimised trend with respect to that of the ear. This means the predominance of low-order harmonics, with a regular frequency distribution: the level of the harmonics must decrease as the order of the harmonic itself increases, with a ratio between the harmonics similar to that of the ear’s intrinsic distortion (frequency optimisation). Furthermore, the level of harmonic distortion must be monotonically increasing as a function of output power (amplitude optimisation). In the vicinity of saturation, the amplifier should exhibit soft clipping and, if possible, even in this area the distortion spectrum should maintain a shape as close as possible to that of the ear’s distortion (or as high as possible before losing its optimal shape).
  Speaking of distortion, an important point must be made: some manufacturers try to artificially ‘add’ distortion that is considered ‘euphonic’. Riviera absolutely DOES NOT follow this approach. On the contrary, our goal is an absolutely natural result. When designing and fine-tuning, we try to reduce all forms of distortion as much as possible in the way that is most appropriate for the ear, i.e. by focusing on the intrinsic linearity of circuits and devices, not on a posteriori corrections (feedback). Our aim is for residual distortions to be as low as possible and as close in shape as possible to those produced by the human ear.
• Zero Overall Feedback and as little local feedback as possible, to minimise the negative effects this technique brings to the sound result. In our opinion, this is the best way to achieve the desired behaviour in terms of distortion.
• A good bandwidth even in open loop.
• A reasonable damping factor (Damping Factor): between 15 and 30, as in the best valve amplifiers, without chasing unnecessary levels. In our opinion, this characteristic contributes to restoring articulation and harmonic richness in the low frequency region.
• Total stability on any load.
• Absence of active protection, to eliminate the negative effects that these devices often have on the sound result and dynamics in particular.
• Extreme attention to the power supply section as it is considered an integral part of the musical signal path.

These theoretical points then took shape in the design of various real devices. So here is how we have actually implemented them, bearing in mind that depending on the type of equipment (preamplifiers, power amps, integrated amplifiers) there may be differences in detail.

• Zero Overall Feedback and the use of the minimum possible local feedback where strictly necessary. This is the basic point from which we started to obtain the desired behaviour in terms of distortion as well as many other parameters. This basic choice implies a series of further decisions that become almost consequential in view of a reference result.
• The use of Class A in all stages is the first consequence: it is necessary to have the maximum intrinsic linearity in the amplification stages, since we decided not to resort to the ‘trick’ of feedback to linearise non-linear stages… Class A is the guarantee of the maximum possible intrinsic linearity.
• The hybrid solution is the next logical consequence. The triode is still the best voltage amplifier in existence and, especially in the single ended configuration, offers a form of ‘natural’ distortion that is extremely close to the desired one. This is why it was chosen as the voltage amplifier of choice. Solid-state devices (especially MOSFets) are the best solution when dealing with high power and low impedances and, if used correctly and properly driven, can provide excellent behaviour in terms of distortion. In the circuit configuration adopted, they also provide the desired output impedance.
• In our opinion, active protections are often detrimental to the sonic result. Therefore, there are no active protections: only safety fuses on the power supplies. The consequence of this choice is the need for really generous dimensioning, and not only in appearance (some components simply ‘big’ like certain capacitors the size of beer cans) but in the really significant points: power supply as a whole, transformers, power devices, heatsinks. A further consequence, secondary but not negligible, is the need for adequate mechanical dimensioning.
• Great care in power supply design. In all Riviera devices, there are always at least 2 transformers and 5 separate power supplies. The power supply of the valve sections always have pigreco filters and are stabilised. In power amps and integrated amps, the power stage power supply section also employs a pigreco filter and the distributed capacitance technique, a solution I have been adopting with satisfaction since the early 1990s: lots of small-and-fast capacitors placed close to the power devices, instead of two large-and-slow capacitors mounted far away.
• An extended and careful focus on both measurements and listening, with an optimisation that means a continuous transition from the design phase and measurement bench to the listening room and back again.
• Great care has been taken in the mechanical construction, both because it must be adequate for the generous overall sizing and because we believe that the mechanical part has an influence on the final sound. The result is a mechanism with the quality level of a laboratory instrument and a design that immediately makes it clear that it is entirely Made in Italy.

October 17th, 2017

update: October 18th, 2024