[Interview] An intersting interview with Dr. Earl Geddes

Among the most influential figures in the field of psychoacoustics, Earl Geddes undoubtedly stands out. A physicist by training, with a Ph.D. in Acoustics from Pennsylvania State University, Geddes has built a career that bridges research, industrial innovation and a deep passion for high-fidelity audio. After a long tenure at Ford Motor Company, he continued his professional journey at companies such as Knowles Electronics and Visteon, holding leadership roles in research and development.

In 1996, together with Dr. Lee, he founded GedLee Associates, a consulting firm specializing in acoustics, noise control, audio design and loudspeaker development. Since then, he has collaborated with major companies including Motorola, Resound, Tellabs and Ford. He is co-author, along with Dr. Lee, of: Audio Transducers and Premium Home Theater.

Member of the Audio Engineering Society since 1978, Geddes holds numerous patents in the fields of transduction and active noise control. His scientific publications and technical presentations are numerous and internationally recognized. Over the course of his career, he has received several prestigious honors, including the Henry Ford Technological Achievement Award.MB Many audiophiles and reviewers are used to evaluating a loudspeaker on traditional measurements, but you proposed a radically different approach… The starting point of your research was not the measurements themselves, but rather how humans recognize and perceive music through audio systems. Dr. Geddes' website, offers a wealth of highly interesting material free of charge.

MB Many audiophiles and reviewers are used to evaluating a loudspeaker through traditional measurements, but you proposed a radically different approach. The starting point of your research was not the measurements themselves, but rather how humans recognize and perceive music through audio systems. Dr. Geddes, I would like to ask how you came to develop the GedLee Metric. In what ways does this metric differ from more widespread evaluation methods and why do you think we still insist on relying on these parameters?

EG I began looking at nonlinear systems very early in my career, perhaps back in the mid 70's. I suspected that nonlinearity played a crucial role in the sound quality that we perceive from an audio playback system. I did extensive reading on how nonlinear systems modify the signal through them, but when I combined this knowledge to what I was also learning about how the ear works (with much help from Lidia, my wife,) I realized that how we would perceive a systems nonlinearity would not be well represented by the “standard” measures, THD and IMD. Lidia and I went about setting up a test to determine how humans perceived various nonlinearities. I learned some years previous that very small nonlinearities, such as crossover distortion [intended as distortions introduced in the crossover region between the two half-waves in class B and AB amplifiers], which can be obvious, while they would have a very small THD level can have very pronounced effects on sound quality. Clearly, perception and THD had some issues with being a valid relationship.

Following this idea, we set up a test to evaluate how the various orders of nonlinearity (the terms in a Taylor series description of a nonlinear Transfer Function - the standard description in the literature) were related to perception. The results were not surprising to me, but most certainly were to many others.

Basically, we found that there was not a statistically valid correlation between either THD or IMD and perception as evaluated in our study. What little correlation there was, was actually negative! - meaning that people tended to prefer some nonlinearity. We now know this to be true in many cases - many people have shown a preference for some 2nd order nonlinearity - they like distortion. I'm a purest kind of guy, looking for truly “clean” reproduction regardless of preference.

What was to be done? We had the standard measures, which we showed to be meaningless at describing sound quality. How can we do better? By applying concepts in psychoacoustics to the perception of nonlinearity I determined several things:

1) Masking in our hearing would mask nonlinear generated products that were close to the source signal

The first point told me that our hearing will be more sensitive to higher orders of nonlinearity and that this effect would be most pronounced for lower-level signals. Without going too deep into the mathematical logic, I proposed the GedLee metric. This metric treated the resulting nonlinear signal quite differently than standard THD or IMD measure do.

It first heavily weights low level portions of the waveform while downweighing the louder portion. In many ways (although this is not entirely correct) it appears to weight the harmonics of a sinusoid at about 6 dB per order. When tested this metric gave a significant correlation between the measure and the perception, which is not the case with THD or IMD.

Why the industry has ignored this and continues to use THD and IMD as quality measures has always baffled me. It's like the drunk looking for his keys under a streetlight (which are, of course, elsewhere) because “the light is better.” THD and IMD are so easy to do - let's just keep doing it that way. Never mind that it doesn't mean anything. People think that they understand it.

MB Do you think the principles behind the GedLee Metric could also be extended upstream, to electronics such as amplifiers, DACs, or preamps? Do you believe a psychoacoustic approach might offer a more meaningful perspective than traditional methods in these areas as well?

EG Of course, the ideas are independent of what stage in the audio system that we are talking about. They apply just as well to microphones and amplifiers as they do to loudspeakers. It's just that my interest was mostly loudspeakers. In fact, the nature of nonlinearities in mechanical systems like loudspeakers suggests that the nonlinearity will be of the lower orders. Higher orders don't tend to exist and shouldn't exist because they are so audible. I have seen loudspeakers with a snap through problem [mechanical instability of the cone or suspension] where the cone snaped through its rest position and this sounds simply awful. Loudspeakers tend to be dominated by the 2nd and 3rd orders which are almost inaudible and some even find pleasant. This is true up until the loudspeaker is working at its design limits. Then the orders can rise quickly, and poor quality is perceived. How this growing set of nonlinear orders grows is significant.

Compare a hard clipped signal with a soft clipped one - we did this study. People thought the soft clipped amp sound (louder) fine well past the point where the clipped amp was deemed poor. Both amps had the same rails.

At the other end is crossover distortion. It hardly shows up in a THD test unless one tracks the THD level down into the noise (a problem since what is usually measured is THD plus noise). I found that significant issues can be seen when one does a synchronous average of the test signal at the output, which suppresses the noise. The nonlinear harmonics can thus be seen below the noise level. I think that this is a significant issue. What I am saying is this: the GedLee metric is probably more relevant to electronics than to loudspeakers. Although I do have to agree with Floyd Toole that nonlinearity these days is simply not a significant issue in well designed components, not that it can't be an issue.

MB The importance of the listening room is often underestimated. What role does room acoustics play in determining playback quality, and what is your view on room correction systems? Additionally, what would you personally recommend to improve listening environments: acoustic treatments, DSP solutions, subwoofers, or other approaches?

EG The listening room plays a critical role in in sound quality of a playback system, but in different ways across the frequency spectrum. At low frequencies (LF) the room dominates the situation and at high frequencies (HF) the loudspeaker is the major contributor, although in both cases it's how the room and the loudspeaker interface with each other. At LFs we have strong modal effects and in particular small rooms [with this adjective, Geddes refers to the rooms of any domestic environment]. Basically, the rooms volume dictates the LF modal situation and while each room is different equal volume rooms will behave very similarly. What is most important in this regime is the damping of the room modes (critical), the placement of the sources and how many. Here more is better, up until three, four or more LF sources at which time we see diminishing returns. The subject of multiple subs is vast, and I can't get into it all in this short of space, but it is widely becoming a critical design aspect of high-quality playback systems. At LFs the source types and specifications get dominated by the rooms characteristics and only DSP can properly correct issues here.

At HFs we have a different situation. Here damping is not so desirable as it reduces spaciousness, but what is critical are the reflection levels and timing, especially the Very Early Reflections (VER) (because these meld with the direct sound degrading imaging.) The loudspeakers' directivity and frequency response then become very important criteria. It seems obvious to me that one wants the spectral image of the reflections to be the same as that of the direct sound although there are some that may object to this position. The measure of directivity is best stated as the Directivity Index (DI.) DI varies with frequency and can be as low as 0 dB or in some cases as high as 12 dB (although this would be rare.) The higher the DI the more directional the sound. What is critical is that the DI vary smoothly and I believe should be as flat, i.e. the concept of “Constat Directivity”. In all systems the DI goes to 0 dB at LFs and will rise to some higher DI at HFs. How flat and how high it should be gets very complicated and there is likely no one correct answer as it depends on many personal factors - mostly musical genre type. A low DI will tend to enhance spaciousness as it increases the reflective energy levels at very early delay times. This will also tend to degrade imaging as the VER blend with the direct sound to “blur” the image. As the DI goes higher the higher directivity tends to diminish the VER and thus degrades spaciousness but enhances imaging. If it seems like this is a tradeoff that should be resolved, that's because it is.

Large venue recordings in-situ, like orchestral works will benefit by lower DI values to enhance spaciousness and the degradation of imaging is not so critical. On the other hand, for studio recording work where a great deal of care is taken in placing the sources in the sound stage, a high DI will benefit imaging. And if care is taken to have a less damped listening room at HFs the spaciousness can be enhanced by a great level of later reflections. This is exactly what I do in my rooms. But then I hardly ever listen to large venue works. Imaging is also very important in film playback as one would want in a Home Theater.

One needs to know what they are looking for in a playback system before they can answer the DI level issue. There is also the issue of how the major directivity axis is pointed. I always cross the major axis of the stereo pair in front of me, which means that I am well off the direct axis of the sources. This is another reason why I prefer a flat DI (Constant Directivity) as this won't work for a rising DI. The “toed-in” situation also enhances the size of the “sweat spot” since moving laterally moves one into the higher signal levels of the farther speaker. This again can only be accomplished with a flat DI.

MB You have proposed loudspeakers that use a technology - compression drivers - often dismissed by many audiophiles as outdated or inferior to more commonly used solutions today, such as dome tweeters. Starting from these drivers, you developed waveguides specifically designed to overcome some of the typical shortcomings of traditional horns. Could you tell us what led you to choose professional-grade drivers (Dr. Geddes chose drivers produced by the Tuscan company B&C, specifically, the 15NBX100 for low frequencies and the DE500 for mid-high frequencies of the New Summa 15) instead of more “audiophile-approved” ones? And what, in your view, are the real advantages of this approach when properly implemented?

EG Now that I have described those attributes of a loudspeaker that I seek to create I will describe why only a compression driver on a proper waveguide can achieve these objectives. First, we need to consider directivity, i.e. the DI of the system. It is important to understand that direct radiating drivers; cones, domes always have a rising DI - the physics dictates this result. That means that using direct radiators as the sources means that the DI can never be flat of smooth unless it is very low - does not meet my requirement for a high flat DI. Thus, one can see that direct radiators cannot be used for the entire bandwidth of the loudspeaker. Physics simply won't allow it. However, a waveguide can produce a flat DI, i.e. constant directivity. In years past this was achieved by diffraction in the horn creating a curved wavefront which is then controlled by the side walls. As far as directivity is concerned this works well. But there is a serious downside. Because of the diffraction, a portion of the wavefront at the diffraction slit will be sent backwards towards the diaphragm creating an internal standing wave in the device. This standing wave causes frequent peaks and dips in the frequency response as well as a serious delay in the signals decay and arrival at the listener. These resonances make horns of this type sound resonant and of poor sound quality. This is the issue that has plagued horns for Hi-Fi since the beginning of their use.

Many (many!) years ago my friend John Eargle challenged me to build a model of a compression driver/horn system as I had done some substantial work in loudspeaker modeling creating the computer simulation called Speak. In creating this model (which was published in JAES, Journal of the Audio Engineering Society) I came to realize that in order to calculate the directivity of the device I needed to know the wavefronts shape at the mouth. The problem was that the existing theory of horns, i.e. Websters Horn Equation (WHE) could not provide this description since it was a one-dimensional solution (I needed two.) I also realized that WHE had some serious built-in assumptions that made it inapplicable to the vast majority of devices then in use.

I began to research other approaches to horns that were fully 2D and discovered that certain coordinate systems would have this property and in specific the Oblate Spheriodal (OS) coordinates fit the bill. There are many other coordinates that work such as spherical, etc. but they either required specific wavefronts at the throat (like spherical which requires a spherical wavefront) or they didn't really produce very interesting devices.

The OS ones required a flat circular disk at the throat which was ideal. Many types of sources meet this requirement, in particular compression drivers. Domes don't actually meet this requirement but are none the less often used with some benefits. Why use a compression driver as opposed to a flat disk? Actually, my first test of the theory did use a flat disk which proved the concept. (Coincidentally I found that there was still resonance in the system because of a reflection from the mouth. But otherwise, the idea worked well.) One feature of compression drivers that is not always appreciated is that they have a wider usable bandwidth than a direct radiator and they can handle much much great power output.

For home theater, where high SPLs are required to achieve theater sound levels is a must, the higher output from these devices was again ideal. By rounding the edge of the mouth and using an OS contour otherwise I was able to achieve a very high DI that was flat and smooth without any of the coloration found in traditional horns. Over the years this has been found to be almost the ideal as witnessed by many more advanced simulations seeking to find the optimum.

Mating the high DI of the waveguide with the high DI of a large LF radiator (at the right crossover frequency) yielded a system with virtually ideal characteristics, very high SPL capability, smooth frequency response, high, flat and smooth DI. From this point on I never looked back and the speakers that I designed and built have been held in very high regard.

DISCLAIMER. TNT-Audio is neither a shop, nor a HiFi company or a repair laboratory for HiFi components. We don't sell anything. It is a 100% independent magazine that neither accepts advertising from companies nor requires readers to register or pay for subscriptions. If you wish, you can support our independent reviews via a PayPal donation. After publication of reviews, the authors do not retain samples other than on long-term loan for further evaluation or comparison with later-received gear. Hence, all contents are written free of any “editorial” or “advertising” influence, and all reviews in this publication, positive or negative, reflect the independent opinions of their respective authors. TNT-Audio will publish all manufacturer responses, subject to the reviewer's right to reply in turn.

An Interview with Dr. Earl Geddes

A Conversation on Psychoacoustics and High Fidelity. Part I.