Technical overview
by Luca Chiomenti
I presented my first design for industrial production to the public about 25 years ago. In the introduction I wrote: “We believe that, in the last 20 years, high fidelity has ended up in a blind alley. There is a pointless race to technical perfection. For too much time the final goal seemed to be to reach the limits of measuring instruments. This choice seemed to have forgotten the real function of an audio amplifier: to reproduce the music through an electroacoustic transducer.” 25 years later the situation appeared even more confused. Today, there is a general agreement that the commonly used tests for audio frequency amplifiers do not correlate with the audio quality. Audiophiles largely reject technical specifications and bench tests as an indicator of the subjective sound quality. Nevertheless, perfect bench tests seem to be essential for any product on the market. I cannot synthesize over 25 years of (my personal) study and research on the relation between subjective experience and laboratory measurements. I based my studies on direct experiments and hundreds of pages of bibliography covering over 80 years. I hope to complete a white paper on this argument soon. Now I will try to give some focal points that are the foundation for the Riviera design. We must go back to the beginning. An audio amplifier must reproduce a signal with the highest fidelity… for the human ear, not for instruments. This is the point. It is critical: 1) to understand some aspects of the functioning of the human earing system and, consequently 2) to define the characteristics of the reproduced signal for the human ear, not for an electronic measuring system. Let’s start from the ear.
- When we ear a pure tone, many studies have verified the creation of harmonics inside the ear and, specifically, in the cochlea. This is not a new discovery. First reports on this distortion are from Fletcher (yes, the famous of the Fletcher-Munson curve, in the ‘20s); more precise reports came from H.F. Olson (Acoustics, 1947) and many others later. It is interesting to observe that the ear generates really high levels of second harmonic: about 10% for pressure levels of 90dB (not 120dB, 90dB!). Higher order harmonics levels decrease with the order of harmonics. We can define a spectrum of the harmonic distortion of the ear. The shape of harmonic distribution is very important: there is high predominance of lower order harmonics, with a decreasing spectrum. This spectrum changes with the pressure level but this point is too complex to examine here. The key points are: 1) the high level of distortion the ear self generates; 2) the ear+brain system cancel those harmonics and the perception is that of an absolutely pure tone. In other words, the hearing system suppresses the harmonics generated by itself. Interesting, isn’t it? Even more interesting is this: the system suppresses the sound of “that” range of harmonics even if they are of “external “ origin, under the condition that the shape, the pattern, is maintained. Quite obvious: the system is programmed to cancel that shape of distortion and it cannot distinguish whether the origin of the distortion is internal or external (some interesting musical phenomena are related to this behavior, i.e. the missing fundamental note). If the harmonics differs from this pattern shape, the ear+brain system detects the harmonics as different tones. Based on this, we think that an amplifier generating a distortion spectrum similar to that of human ear will result extremely transparent and clean, even if its THD level is relatively high.
- The above mentioned mechanism is dependent on the pressure level. In synthesis, the higher the pressure level, the higher the harmonics distortion that our ear generates (and accepts). This brings us to prefer amplifiers with distortion monotonically increasing with the output power.
- The higher the pressure, the higher the order of harmonics that ear generates. This means that, at higher levels, an increased quantity of higher order harmonics can be accepted.
- The detectable distortion depends, among other things, on: 1) the ratio between the peak and the average value of the signal; 2) the duration of each signal peak. Studies have proved that distortion can even reach very high values and still remains inaudible if the peak length is very short.
- Masking is the well-known phenomenon where a low level tone in close proximity to a higher level one is not audible. This can limit the influence of IM distortion in an amplifier with a predominance of low order harmonics. This effect seems to be even higher if the distortion spectrum shape of the amplifier is close to the ear’s one.
- Feedback is the classical method used to reduce THD and upgrade circuit performances (THD, IMD, bandwidth, noise and more). Unfortunately feedback reduces the lower order harmonics (less harmful and more “benign” for the ear) much more than the higher order harmonics. More and better: high levels of feedback act as a generator and “multiplier” of higher order harmonics (harmonics that ear detects as “dissonant” and noisy). Even if these high order harmonics are under the audibility level, the mechanism generates a “noise floor” that seems to be absolutely undesirable for the human ear. All this suggest to avoid feedback (specially overall feedback) or to try to reduce its use to the minimum.
Starting from this theoretical base, we defined the ideal characteristics for “human ear dedicated” amplifiers. Riviera amplifiers are optimized for the bench test that demonstrated an actual relation with the listening experience only, careless of pure technical virtuosity. We focused on the following points.
- Optimization of distortion for amplitude and frequency: THD does not need to be extremely low, but it must absolutely follow the human ear distortion shape. This means predominance of lower order harmonics, with their regular distribution in frequency (the higher the order, the lower the harmonic level, with a relation between harmonics similar to the human ear one). Furthermore, the distortion level must be monotonically increasing with power level. The amp should have a soft clipping and, if possible, the distortion spectrum must be similar to the ear spectrum even in this area (or to the higher level possible before losing the right shape).
- The use of Zero Overall Feedback and minimum level of local feedback, to minimize the above mentioned negative effect that this technique has on sound result. This is the best way to reach the desired distortion behavior.
- A good open loop BW.
- A reasonable DF, like in best tube amplifiers: between 15 to 20, without looking for useless records. This point also contributes to provide articulation and harmonics richness in the low frequencies.
- Total stability on every load.
- Absence of protection, to avoid their negative effects on sound and dynamics.
- Extreme care to the power supply.
These theoretical points are then implemented in a real amplifier design. Here the way we implemented these points.
- The Zero Overall Feedback and the minimum use of local feedback, when absolutely indispensable, is the first solution to reach the desired distortion behavior.
- The use of the A class in all stages is the logical consequence: we need to have the maximum linearity without “tricks”.
- The hybrid solution is the next logical consequence: the triode is the best voltage amplifier and, mostly in a s.e. configuration, it offers a “natural” distortion shape, very close to the required one. Silicon devices (and specially Mosfets) are the best choice to handle power and low impedances; if properly used and driven, they can offer a good distortion shape too. In the topology adopted they also give the desired output impedance.
- Protection circuits damage sound. In our thinking this is a fact. So… no protection circuits (only fuses on PSU rails). Consequently a real large dimensioning, not for appearance (coke-like capacitors…) but in the real key points (PSU, transformers, power devices). A secondary consequence is the need for an adequate mechanical dimensioning too.
- Accurate PSU design. 2 transformers and 5 separate PSU. The tube PSU is PI-filtered and stabilized (with a MOSFet); in the audio power PSU stage there is a PI filter too and the distributed capacitance: instead of 2 big-and-slow capacitors, many little-and-fast capacitors, with the last ones extremely close to each power device.
- There was a long and accurate musical and bench optimization (with continuous shift from the bench to the listening room and vice versa).
- Mechanical and “physical” manufacturing with a quality level of a laboratory instrument and a made in Italy design.
October 17th, 2017