Fourth and last part of the interview with Mr. Giuseppe Parolo started here. Taking a cue from his research Properties of Nonlinear Distortions and Related Measures in Audio Amplifiers, published in the Journal of the Audio Engineering Society, in the previous parts we were able to find how nonlinear distortions in amplifications play a crucial role in perceived listening quality.
Thanking again Mr. Parolo for the availability shown to us and not taken for granted, given that it is a matter of making a simple and informative paper far beyond the type of work already done, we therefore conclude with a series of questions and ideas as much as possible oriented to understand how technicians and ourselves audiophiles can move in this field.
Question: At this point in your research, can it be said that the "historical" measures are perhaps limited, if not wrong, and must and can be updated or changed?
Giuseppe Parolo: From what has been reported in various research, classical measurements offer only a modest correlation with sound quality. As I have described, this situation depends mainly on the exclusion of psychoacoustic effects in the calculations, as well as their view limited to simple stationary test signals.
But there is also another, more insidious aspect to consider: these measures are not able to quantify the overall magnitude of nonlinear distortion on the test signals. In fact, the hidden distortion components in the fundamentals are excluded from their calculation. To get an idea of the effect of this lack on the measurement of the THD, we can look at Fig. 3(b): the value of the THD refers to the curve in green and not to that of the real distortion in red... a big difference!
This can also be analysed in the frequency domain by considering Fig. 9 and Fig. 11: the classical measures photograph only the distortions in the lower graph. Sometimes the missing hidden components are negligible, as in Fig. 9 where graphs (a) and (b) are almost identical; sometimes they are not, as in Fig. 11 where graphs (a) and (b) differ significantly. Thus, in some circumstances the classical measures are able to capture all nonlinearities; in others, energy components are lost, resulting in a lower value than the true value. An undesirable effect of this situation is to make comparisons between measures of distortions with different structures, and thus of different devices, inconsistent, especially when the deviations are small, about 10 dBs.
To the best of my knowledge, none of the measurement tools currently on the market involve the use of models capable of estimating missing distortion components. To remedy this situation, I proposed:
- a computational procedure to estimate the level and phase of the hidden components;
- a new set of dual measures (metrics) of the current ones, identified with the prefix “True-,” that includes these components in the calculation.
In my research, I have outlined some relatively simple methods for estimating hidden components, based on identifying models - as in Fig. 7 - that approximate device behaviour using data already collected in classical measurements. In the simplest case where harmonics of order greater than third are negligible, the correction factor to be applied to the THD depends only on the ratio between the levels of the third and second harmonics (HD3/HD2) and not on the absolute levels of each. Fig. 12 shows graphically the trend of this factor in dB, obtained by simulations.
Fig. 12 - Corrective factor to be applied to the THD as a function of the ratio between the intensity of the third and second harmonics. The curve is applicable in cases where the distortions in the harmonics of the upper orders are less than about 30 dBs.
From the figure, it can be seen that when the second harmonic predominates (HD3/HD2 < -20 dB), the THD is corrected; vice versa, when the third harmonic predominates (HD3/HD2 > 10 dB), it is necessary to add 10 dB of correction, that is, the effective distortion is 3 times higher. In intermediate situations, the correction is gradual. Thus, in the case of Fig. 3, where THD = 77.5 dB and HD3/HD2 = -84.8+78.3 = -6.5 dB, a correction of 4.2 dB will result from the graph and thus a True-THD = -77.5+4.2 = -73.3 dB. Note that for an amplification chain the HD3/HD2 ratio is not fixed, but grows as the volume increases: in fact, the level of the third harmonic depends on the cube of the signal level; the second on the square. Therefore, excluding the situation in which third harmonic distortion is absent, the correction at THD also increases by increasing the volume, with a maximum value of +10 dBs. The curve does not depend on the distortion phases. However, the hidden distortion phase can be compressive (in phase), expansive (in counterphase), or an intermediate situation. In the latter case, the models show that distortion levels will change depending on frequency. This information may be used as an additional factor to consider when classifying the device.
For the THD+N metric, the curve shifts horizontally to the right depending on the overall noise level.
The IMD requires milder corrections because it contemplates only the tones in non-harmonic ratio: the variations are due only to the reference of the fundamentals, which must be changed for the hidden component. The True-TD+N metric, on the other hand, differs from TD+N, whose value depends on the selected signal. In general, given that in this case there are many intermodulation components (see Figs. 9 and 11), the numerical contribution of hidden distortions has less influence on the final result than THD.
If the device exhibits distortions beyond the third order, the situation becomes considerably more complicated, with the phases of the distortions playing an important role. To get an idea of how they behave, we can simulate a system in which only the odd harmonics HD3 and HD5 appear (the even ones do not affect HD1). Fig. 13 shows the trend of True-THD and THD as a function of the ratio HD3/HD5, with HD5 set at -90 dB (different values shift the curves vertically) and two phases’ combinations:
- dashed curve: fase HD₃ = 180°, fase HD₅ = 0°;
- continuous curve: fase HD₃ = 180°, fase HD₅ = 180°.
Fig. 13 – True-THD, THD e HD1 in funzione del rapporto dei moduli HD3 e HD5.
In the first case, True-THD is approximately 15 dB higher than THD, and both curves increase as HD3 distortion increases. In the second case, however, True-THD initially decreases as HD3 increases (unusual behavior) and then rises again after a point of contact with THD when HD3/HD5 = 4.2 dBs. This effect is due to the phase variation of the hidden component HD1, which changes from expansive to compressive, as shown by the HD1 curve in the same graph, which has a downward cusp at the transition point. A change in the distortion phases results in intermediate trends between these two curves, with consequent dependence of distortion levels on frequency.
For the other metrics THD+N, IMD, and TD+N, considerations similar to those outlined for distortions up to the third order apply.
Thus, the adoption of this new set of metrics, combined with information about the phase of the hidden distortion, can highlight phenomena that the classical ones do not detect. I am currently working on further refinements of the measurement method to make the estimation of hidden components more accurate, using ad hoc test signals and more complex models.
Question: Can correctness of measures versus listening pleasure coexist then? How can they be reconciled?
Parolo: It is common when it comes to sound quality to confuse pleasantness with correctness. To sum up, you can't automatically assume that the audio chain with the lowest distortion is the most appreciated: listening habits for certain types of distortion in domestic environments make them appear normal and pleasant. However, my research does not aim to define global indicators for the pleasantness or naturalness of the sound, but only to qualify more precisely some alterations on the signal to be related to particular effects on the perceived sound. In fact, many other variables intervene in the feeling of pleasantness and naturalness.
To get an idea, a study on the subject – Ref. [4] – tries to define indicators with sound quality gives the effects due to distortions a weight of 30%.
Ref. [4] F. Toole, Sound Reproduction - Loudspeaker and Rooms, Focal Press, 2008, available for full download here
More than half the weight lies in the spatial feeling of sound created by the interaction between the loudspeaker and the listening environment. This includes effects such as the size of the soundstage and the immersion effect. The physical elements that influence this can be found in the relationship between the quantity and quality of sound produced by the loudspeakers and the sound that is reverberated from the environment. Without going into too much detail about room acoustics, it is important to note that the room in which an audio system is installed can significantly alter the frequency response through different mechanisms for low and medium-high frequencies. In the former, the resonance modes of the room, identified by its dimensions, dominate; in the latter, the effect of sound reflections on the different surfaces of the room prevails. Alterations in the frequency response measured with a microphone can be as much as ±10 dB, but here again psychoacoustics comes into play, which determines different perceptions depending on how the energy of the reverberated sound is distributed over time, see Haas Effect here.
Therefore, physical measurements, in all their forms, are always the starting point for understanding the reality that surrounds us, providing us with objective indications. Learning to interpret measures, understanding what they tell us and what they don't tell us, can only be useful for anyone, reducing the risk of making major errors of evaluation.
Question: Yours is scientific research. It was conducted with complex mathematical tools not available to most people. It has been checked, verified, and endorsed by your peers. It is therefore objectively difficult to enter into the merits. But, jumping to his conclusions, could his scientific approach possibly suggest to us simple audiophiles some "tricks" or best practices of approach to listening?
Parolo: It is not easy to offer advice for audiophiles derived directly from my study because it requires detailed measurements of the devices in the listening chain, which are not normally available.
More generally, an initial recommendation concerns the approach to comparing devices. Since acoustic memory is not very reliable, it is advisable to always perform direct, blind comparisons, in a calm environment, on audio chains where only one component is changed, using the same music tracks and controlled playback levels. This is the only way to evaluate any differences, sometimes subtle, between the behaviour of a device inserted into a specific chain, in which particular synergies can be created: the shortcomings of one component can be mitigated by those of another. Therefore, there is nothing particularly new compared to common practices in this field.
There is also nothing new in recommending, as far as the amplification part is concerned, the use of a solid-state solution for power amplifiers, which can provide more neutrality and drive capability for speakers than a tube solution. It's the preamp that you can play the most: solid state or tube based on your preference. This combination gives you more control over the overall distortions in general.
The part that may probably seem new to the less experienced is that in the audio chain there is also the listening environment to consider: it must be taken care of like any other component (indeed, even more, possibly relying on professionals) since the reverberated sound heavily affects the perceived sound quality.
Finally, I would like to suggest a couple of experiments with your system to verify some effects of nonlinear distortions:
- Phase inversion of even-order distortions - The effect can be experienced if the DAC or preamplifier allows phase inversion of the output signal. This operation, combined with the inversion of the connection of the positive with the negative in the cables to the loudspeakers, has no effect on the fundamentals and distortions of odd order, while it reverses all those of even order.
- Changing the mix of distortions - Considering that nonlinear distortions multiply each other when the related systems are cascaded, a good potentiometer can be inserted between the output of the preamplifier and the input of the power amplifier. By compensating for an increase in volume on the preamplifier with an attenuation on the potentiometer in order to maintain the same listening volume, you can experience what happens by increasing the distortion effects of the preamplifier.
Of course, to understand exactly the characteristics of what you are listening to in any given situation, it is always necessary to make measurements.
Question: Finally, what practical, constructive advice would you also feel able to give to audio amplification manufacturers in the light of your research, in the current state of your knowledge of the problem?
Parolo: The goal of trying to reduce nonlinear distortion to inaudible levels, at least in amplifiers, could be a waste of time and money, since many people like the sound of certain types of distortion. Therefore, it is necessary to research the right amount and structure of distortion that meets the needs of audiophiles, considering the variability of elements in the audio chain. The result of my research provides the designer with more refined tools for the physical analysis of nonlinear distortion that can help him in this task. However, it does not delve into the circuit types, components, layouts, working points, etc. that lead to a certain type of behavior: it is up to the skill of the designer to figure out how to act on the design to achieve a certain result.
As a personal experience, I can add some aspects that help to reduce the effects of memory on distortions, surely already known:
- frequency response - flat and extended well beyond the audible bandwidth
- feedback - in case this type of circuit is used, it is good to minimize the length of the signal path
- electromagnetic interference - to be avoided by acting on the layout and with shielding
- vibration control – to be reduced as much as possible
Finally, order and simplicity - they are always rewarding.
Ref. [4] F. Toole, Sound Reproduction - Loudspeaker and Rooms, Focal Press, 2008, available for full download here
End part 4 of 4 - Back to the first part
For further info:
write to Eng. Parolo
to JAES website
to AES website
