The first really decent stereo audio system I owned had a pushbutton on the front panel called "loudness." I pushed it. I decided the audio sounded better so I left it pushed in; never changed it. OK, occasionally I would push it to the off position; nope, sounded better on. Thinking back, I really didn't care what it was for. I left it pushed ON because it made the system sound better, so I continued to concentrate on my video career. All the while, my stereo system "sounded good."
Fast-forward 30 years. Here I am writing about that pushbutton. That pushbutton recently created considerable flurry here at Extron. Since the inclusion of much more audio support in our products, that loudness pushbutton finally crept into a prototype product. Someone in Product Management pushed it, but it didn't do what it was supposed to do: make the system "sound good." That launched an investigation into the real intent of the loudness control function and a bit of re-evaluation by those individuals designing audio products.
We learned something. Different people and different companies have different design philosophies toward functionality of the loudness control. So, when you push it, what should the loudness control do? Let's begin answering that question with a look at the historical basis for the loudness control and why it should make an audio system "sound good."
Equal Loudness for All
Research characterizing the human hearing range generated an areal map of auditory response as shown in Figure 1. The shape of the minimum audibility curve tells us something about the frequency response of the ear at low sound pressure levels. At the boundary of minimum audibility, perception of low frequencies and high frequencies requires a significant level boost as compared to the midrange frequencies. Conversely, at the upper boundary the threshold of feeling represents a much flatter response. The incomplete outline of the graph indicates regions of extreme variability where performance data collected on human subjects is not altogether consistent.
In the 1930s, two Bell Labs researchers, Harvey Fletcher and Wilden Munson, organized a research project in which they asked that each participant match the perceived level of two pure tones by adjusting the level of one tone source against a 1000 Hz tone at a pre-determined reference level, until the two tones were perceived as equal in loudness. Their test results embodied data at many frequencies at various sound pressure levels across the human hearing range. Since this is a highly subjective way to conduct research, many subjects performed the experiment. Results were averaged to obtain loudness contours intended to represent a "normal" response. Their results, called the Fletcher-Munson equal loudness contours, provide significant insight toward our understanding of how humans perceive relative sound levels. In 1956, D.W. Robinson and R.S. Dadson refined this work in a similar study described as having more reliable measurement results. The Robinson and Dadson equal loudness contours shown in Figure 2 are most widely used today, but are often referred to as the Fletcher-Munson curves. Eventually, the International Standards Organization adopted the Robinson-Dadson curves as a basis for "Normal Equal-Loudness Level Contours," ISO 226:1987, which is now the current standard. Since these curves describe the perception of only pure tones in a free field, they do not necessarily apply to noise band analysis or diffused random noise.1 Additional research continues in an attempt to further characterize our hearing perceptions in real audio applications.
If we invert any of these contour curves at a particular intensity level, we now have the relative frequency response plot of the human ear for all tones across the frequency range on that particular contour. Inversion of the lower curves illustrates human ear frequency response deficiencies at low sound intensities. Conversely, invert any of the upper, higher intensity curves to realize a more flattened frequency response. Fletcher and Munson found human hearing response consistently deficient at low sound intensities, for both low frequencies and higher frequencies compared to the 1000 Hz reference point used to establish these curves. But, the ear is particularly sensitive in the range of about 300 to 6000 Hz. This happens to be the frequency range that includes the sound of most human speech patterns and, curiously, the pitch of a crying baby. For average background sound pressure levels, the rolled-off response at lower sound pressures allows us to more easily listen and understand human speech in the presence of low or high frequency noise.
Meet the Phons
Each contour is identified as a level in "phons." A phon is a subjective unit for loudness equal to the sound pressure level in decibels when compared to an equally loud standard note. The standard note is a 1000 Hz pure tone or narrowband noise centered at 1000 Hz.2 Note that the level in phons matches the sound pressure level in decibels only at the 1000 Hz standard reference point on the graph. Therefore, the 40 phon contour represents a 40dB SPL at 1000 Hz, but a different SPL at most other frequencies. Essentially, each phon contour represents a 10dB step that we perceive as about twice as loud as the previous level. This may be a bit confusing, since we know that a measured 3dB increase represents a doubling of sound power, but only a perceptible increase in volume to the ear. The red dashed line at the bottom of Figure 2 describes the minimum audible level for hearing sensitivity in a free field.
The effect of applying these curves suggests that some form of filtering is required within measurement equipment if we wish to synthesize the normal hearing performance of the ear when calibrating systems or making value judgments as to sound quality. Most often, the sound pressure level (SPL) meter is used to setup listening levels for an audio system. The SPL meter includes selectable filters that modify its calibration so it approximates the ear's response at a given range of sound pressure levels. The most often-used filter settings are the A-weighted and C-weighted. What are these and how do they relate to our hearing response?
The concept of weighting refers to the relative shaping of the filter's response so as to mimic the ear at a given loudness level. Four weighted filter functions, A, B, C, and D, are used to simplify and apply regions of the loudness contours that are most meaningful for describing the frequency response of the human ear toward real world applications. Refer to Figure 3 for the following discussion. A-weighting defines the shape of the filter (and the human ear response) at low sound pressure levels, namely the 40 phon loudness contour curve. Sound level measurements in decibels relating to A-weighting are denoted with the units – dB(A). Shaping for this curve means that low frequencies are attenuated and the speech frequencies are amplified within the measuring equipment. B-weighting describes an intermediate level approximating the 70 phon curve. Notice how the ear's response begins to flatten. C-weighting utilizes the 100 phon curve, which describes the nearly flat response of the ear for high levels. The C-weighting response is most useful for typical home theater listening levels and for evaluating system performance for flat response characteristics. The D-weighting Curve is a special case developed for aircraft fly-over noise testing, which penalizes high frequencies.3 Likewise, sound level measurements in decibels relative to these weighting curves are recorded as dB(B), dB(C), and dB(D), respectively. The A and C weightings are most often used since the former relates to normal everyday sound pressure levels and the latter relates to higher listening levels where the ear's response is nearly flat.
We've covered some significant background, but how does all of that relate to the loudness control feature on an audio system? Understanding how the ear perceives sound intensity versus frequency leads us directly to that loudness feature. The loudness control is simply intended to significantly boost low and high frequencies when listening at low levels so that the ear perceives an overall flatter sound pressure level. In other words, if the loudness contouring control is not enabled at low volume levels, bass and treble appear to be lacking. This effect corresponds to the recently described A-weighted condition where low and high frequencies require additional amplification so the audio "sounds good."
Since the ear's frequency response is relatively flat at high sound levels, the compensating effect of the loudness contouring control is not required. The loudness feature is a kind of equalizing function that, ideally, should adjust itself to have greater compensation effect at low sound pressure levels and less effect as sound pressure increases.
From Figure 4, you can see that the amount of power needed (green shaded area bounded by LA curve) to compensate for low frequencies is significant. For this reason, in home theater audio system design, it is not uncommon to use fairly large, separate amplification just for the low frequency channel. The shaded area within the high frequency range indicates relative compensation required for this portion of the spectrum when at a lower volume level. At high loudness levels, where the ear's response is nearly flat, compensation requirements decrease to nearly zero as shown by the LC curve.
Loudness Made Too Simple
The issue is whether the implementation of the loudness control feature merely boosts lows and highs using one fixed setting as some simplistic designs might do; or is it dynamic and capable of modifying the amount of equalization depending on the setting of the volume control?
Historically, most loudness controls were analog implementations using discrete resistors, capacitors, and even inductors intended to approximate the compensation curve (curve LA in Figure 4) for the A-weighting function. Most were designed around the volume control. Figure 5 illustrates one simple approach using a volume control incorporating a fourth tap located about halfway through rotation. Resistor-capacitor networks, when switched into the volume control circuit, provided amplitude compensation. For really low cost circuits, only the low end frequencies may have been boosted. Or, perhaps the midrange was "cut" to make it sound more like the level of the low end. Certainly, analog implementations of the loudness feature vary widely. Full compensation for the A-weighted response requires a relatively complex compensation network.
The basic approach with the circuit in Figure 5 is: 1) use C1 to boost high frequencies where it is connected across the top half of the volume control when the loudness switch is ON; 2) select the value of C2 so its reactance is lower at high and mid frequencies, and; 3) select R so that high and mid frequencies are attenuated; but, as frequency decreases, the reactance of C2 will rise and reduce attenuation of low frequencies. This is a simple, low-cost design built totally around performance trade offs.
DSP: Just Made for Loudness
Modern implementations of loudness equalization circuits fall comfortably into the realm of digital signal processing, or DSP. Within the vast possibilities afforded by digital processing, the creation of filters capable of replicating a near-exact compensating response is not only possible, but generally straightforward. DSP-based algorithms allow for continuously adaptable functions that will compensate in real time as sound pressure level is varied over its normal excursion.
High speed digital signal processing, in all its forms, provides the means for the best implementation of loudness compensation contouring in today's sophisticated audio systems. With tools like this, engineers must return and study the fundamental knowledge base developed by researchers like Fletcher and Munson; et al. Taking a fresh look at "what once was" will ensure us the best shot at developing digital-based products that perform to the closest approximation of the original concept. But, no matter what, all that you and I really care about is that when we push that loudness button, the system "sounds good."
- Fletcher-Munson definition. Rane Corporation at en.wikipedia.org/FletcherMunson_curves
- Weik, M. H., Communications Standard Dictionary, 1997, Chapman & Hall
- Lord, H., Gatley, W., Evensen, H., Noise Control for Engineers, Robert Krieger Publishing, 1987
- Ballou, Glen M., Handbook for Sound Engineers, Third edition, 2002, Butterworth-Heinemann, Chapter 2, Psychoacoustics, F. Alton Everest