DIAGNOSING THE PATIENT
Jul 1, 2000 12:00 PM, Sam Berkow
Using acoustical measurement to optimize system performance.
There are many types of acoustical measurements. There are also an unlimited number of applications for the use of acoustic measurements. Within the professional audio community, acoustical measurement systems are most commonly used as tools in the sound system optimization and equalization process. In this article, I will cover the evolution of these tools and the role of acoustic measurement systems in sound system optimization.
Sound system optimization, also known as system tuning, is generally the last step taken when installing the sound system. For permanently installed sound systems and systems set up for temporary use, system tuning can play a critical role in both the perceived quality and the stability and responsiveness of the sound system. Given that this process so greatly affects the perceived quality of the sound system, it is both curious and unfortunate that there are few if any standardized techniques or even a widespread understanding of the acoustic principles involved with system optimization.
The system optimization process usually consists of setting such system controls as EQs, crossovers, delays and amp levels. Given the current state of the art in DSP-based devices, this process can involve hundreds of controls, each with its own wide range of settings. A fundamental question that one might ask when faced with all of the choices on the control surface of a modern sound system is how these parameters relate to the perceived subjective quality of the sound system.
In the good old days, sound system optimization involved listening to the playback of music and turning knobs until the system sounded right. Although this satisfied some people, systems set up using this technique often did not provide maximum gain before feedback or optimum tonal balance. Enter the widespread use of real-time analyzers (RTAs). Real-time analyzers are usually single-channel devices with an input mic and a display that provides a graphic representation of the frequency content of a signal arriving at the input mic.
Real-time analyzers became popular for several reasons. They make wonderful ear training devices. They are easy to understand, and usually easy to operate. Real-time analyzers are, however, severely limited as tools for sound system optimization. Although an RTA will reveal the frequency content of the signal arriving at the measurement mic, there is little information about the quality of that data available. Further, when used with music as a source, there is little or no information about the correct spectrum of the music. To address this limitation, an interesting approach was adopted; a test signal that looked flat on an RTA could be sent to the loudspeaker system. Because the input signal was known to be flat, users merely looked at the spectrum of the signal arriving at the mic and adjusted EQs and other system controls until the measured signal looked flat. At first glance, this procedure appears to make sense, but it has almost universally been found to yield unsatisfactory results in the form of poor sound quality.
As a result of the failure of this pink noise/RTA single-channel measurement method, it is extremely common to hear someone say that equalizing to flat sounds bad. Operators who have this opinion have most likely played pink noise through a sound system, adjusted EQs so that the resulting spectrum appears to be flat, stopped the pink noise, played some music and found that the music sounded tonally unbalanced. In many cases, a system operator will abandon pink noise at this point and adjust EQs while listening to some familiar music. The standard configuration for the RTA measurement is shown in Figure 1. The failure of the pink noise/RTA method is caused by the lack of attention this technique pays to the time relationships of signals sent to and received from a sound system. This technique also offers no information as to how sound is perceived in various acoustical environments.
To address these shortcomings, two acoustical measurement systems became popular in the 1980s within the acoustical and sound contracting communities. These systems, along with several others, provided the ability to compare a signal being sent to a sound system within the received signal in a time-coherent and time-windowed manner. Although these two systems used different mathematical techniques to reach these goals, both the TEF Machine and SYSiD provided a stimulus and response measurement tool that allowed the user to compare the input and output of a sound system in such a way that the user could, to a certain extent and within several limitations, define a ratio of how much direct sound from the sound system and the sound system-to-room interaction were included in each measurement.
When compared with the RTA method, stimulus and response measurement systems offered several key advantages. The ability to compare the input and output of a system in either the time domain or the frequency domain allowed users to measure, evaluate and understand more clearly how the sound system was interacting within itself or with its acoustical environment. Measurements that compare the input and output of the system may provide a number of advantages when compared to single-channel analysis. Such measurements provide the ability to calculate both measured results and quality of data indicators, such as phase relationships and the coherence function. Measurements showing this comparison in the time domain provide a graphical representation of the signal that represents the performance characteristics of the sound system and the acoustical environment. This measurement, called impulse response, clearly displays such important information as direct-to-reverberant levels, decay rates at various frequencies and the presence of potentially disruptive reflections. These time domain results represent either the actual impulse response of a system or the energy envelope of a systems impulse response, known as the energy time curve (ETC) The standard configuration for a stimulus and response measurement with a dedicated test signal is shown in Figure 2.
For many consultants, contractors and live sound engineers, stimulus and response measurements provided a new way of understanding the interaction of the sound system and the acoustical environment in which the sound system functioned. These stimulus and response measurement systems were not without their limitations, however. In this case, a dedicated test signal is required. While in some cases there is flexibility in the nature of this test signal, the need to use a dedicated test signal severely limits the ability to make measurements during a performance. Further, test signals are often difficult for users to listen to in a critical way, requiring users to rely on graphical results rather than interactive listening for subjective judgments. Lastly, the ability of the systems to provide time windowing, where the user is able to mathematically (to some degree) isolate the sound system from the acoustical environment, often results in measurements that provide sufficient frequency resolution over only a limited portion of the audio bandwidth.
The widespread use of measurement systems based upon stimulus and response within the contracting community resulted in a discussion of what should and should not be considered when making measurements in order to optimize a sound system. For many people, the ability to isolate the sound system from the acoustical environment suggested a method for making measurements in which equalization was performed using data that contained only the direct sound from the loudspeaker system. In many cases, people use strict time windowing to isolate the measurement of a loudspeaker from the influence of the room interactions as completely as possible. This attempt to use short time windows to achieve what can be called semi-anechoic measurements resulted in measurements that had little or no low-frequency information.
For some people performing system equalization, such measurements were sufficient. For others, however, the lack of low-frequency information within measurements using short time windows meant that several measurements of various time windows had to be made to equalize a sound system across the audible bandwidth. Discussions concerning which time windows to use and how much, if any, of the influence of the room should be included when making measurements intended to help system optimization were common for many years. One phrase that was often heard was: "Because you can only equalize the direct sound, you should isolate the direct sound as much as possible in your measurements." Although this phrase certainly appears reasonable, its implications are far-reaching, and its application often mistargeted.
At this point, it might be a good idea to ask why system optimization is required at all. If a manufacturer provides a loudspeaker with a given on-axis frequency response and documented polar response, why is any additional equalization required? The primary goal of system optimization is to make a system as linear as possible. Linear, in this case, refers to a system that has limited or damped resonances. The idea of making a system linear suggests that the system will have maximum gain before feedback and will provide an articulate and tonally balanced response at some point in space.
The term "optimization" implies that the system can be as linear as possible. This does not mean that the system has to be flat, but it does mean that response of the system should be without any strong localized resonances, such as frequencies ringing or feeding back or having a perceived increase in level compared to adjacent frequency ranges. In this case, resonances refer to frequencies where the system provides more energy output than the spatial balance of the input signal would suggest. Resonances in sound systems come primarily from two sources; interaction of one portion of the sound system with another portion of the sound system (often coupling of loudspeakers or loudspeaker-mic feedback loops), and interaction of the sound system with the acoustical environment (room modes, reflections or reverberation).
Although it is certainly true that using an EQ affects only the direct signal that is being sent to the loudspeaker, by examining the interaction of the sound system and the room at mid and low frequencies, where this interaction is often modal or reverberant, equalization can be used to optimize this loudspeaker-to-room interaction. At high frequencies, however, the interaction of the loudspeaker system and the room is primarily due to discrete or specular reflections, where equalization is of only limited effectiveness.
With the recent increase in available computer power at reasonable prices, the advances made in DSP technology and the ever-increasing quality of A/D converters, several practical new measurement systems have been created that provide measurements specifically designed to allow multiple time-window measurements to be made simultaneously. Such systems as the SIM system from Meyer Sound Lab and the Smaart System from SIA Software Company provide the user the ability to make measurements that incorporate several time windows in a single display. Both of these systems are based on the real-time transfer function. This measurement is similar to stimulus and response techniques in that it provides a comparison between the input and output of a system; however, in the dual-channel FFT-based transfer function, no dedicated stimulus is required. Further, the real-time transfer function can use one or more time windows simultaneously, allowing users to select how much of the sound system-to-room interaction is included in their measurement.
The standard configuration for a real-time transfer function measurement is shown in Figure 3. Notice that in this configuration, the computer does not need to generate any stimulus signal. Program material is used as a reference signal against which the output of the system is measured.
The advantages of a measurement using multiple time windows is shown in Figure 4. At high frequencies, the direct sound is as isolated as possible. To make meaningful measurements at low frequencies, longer time windows are used, allowing more of the sound system-to-room interaction to be included in the measurement. The use of longer time windows at lower frequencies not only provides increased frequency resolution, but it also correlates extremely well with the nature of human hearing where low-frequency energy requires longer periods of time for our ears to process.
A further advantage of the real-time transfer function measurement is the ability of this measurement to use music as the reference signal. The use of music has two distinct advantages. It allows users to listen to a familiar program while changes to the system controls are made, and it allows meaningful measurements to be made during actual performances.
At this point, it is critical to note that all of the techniques discussed so far are just tools. These tools can be used effectively to achieve wonderful results or misapplied to create poorly balanced, unstable systems. When using these tools it is critical that mic position, loudspeaker positions and system design be carefully considered and understood before changes to system controls are undertaken. When used in conjunction with critical listening and careful thought, all of the tools discussed here can be helpful. It is my experience that the use of the multi-windowed transfer function can provide valuable information about system performance, providing semi-anechoic information at high frequencies and detailed information about sound system-to-room interaction at low frequencies. When combined with critical listening, this data is the most useful available when optimizing sound systems.