The audio profession has a place for both science and art. An observation of other technical fields (medicine, for instance) reveals that art tends to give way to science as more is learned about the systems being observed and practiced. As the microscope provided the medical community with a view of the finer structure of life processes, a new generation of audio analyzers is allowing sound practitioners to analyze sound in greater detail. This series of Tech Topics is devoted to familiarizing those new to acoustic measurement with the fundamental principles of time domain analysis. A special thank you to Dr. Sidney Bertram for his assistance in proofing and clarification. pb

New technologies have brought advanced acoustic measurement tools within the financial reach of many audio practitioners, allowing them to move beyond the traditional sound level meter and real-time analyzer. As such tools are integrated into the day-to-day routines of audio personnel, greater levels of system performance should be realized for the end users of sound systems.

Many of those who acquire a modern analyzer are surprised at the apparent complexity of their new measurement system. Their traditional real-time analyzer or sound level meter did not require a great deal of expertise in either taking or interpreting the measurements. With advanced measurement systems, the user is faced with numerous choices and parameters that must be properly selected to get any type of meaningful display on the instrument. The purpose of this Tech Topic is to provide a primer for getting "up and running" with modern sound measuring instruments. The subjects addressed are universal in nature, and can be readily applied to most any measurement platform.

The Big Picture

Let's begin by contrasting modern analyzers with the more traditional sound level meter (SLM) and its cousin the real-time analyzer (RTA). A sound level meter simply measures the average L_p (sound pressure level) at one point in space. The measured pressure fluctuations are converted into levels and displayed on the readout of the instrument. Such a measurement is useful for determining the loudness of a sound, but is relatively blind with regard to the frequency content of the sound. The use of frequency weighting curves make the meter less sensitive to extremes in frequency, but aside from these SLMs have no frequency resolution.

A real-time analyzer can be thought of as a bank of frequency-selective sound level meters. The sound pres sure level can be displayed at fractional octave intervals (usually one-third octave), yielding considerable information about the frequency content of the sound. These individual levels are then summed to display the sound level (L_p) at the measurement microphone.

Both types of measurements are unable to differentiate between sounds arriving at different time intervals, and show an averaged level of all sounds arriving at the microphone within the integration time of the instrument. As such, the data displayed by the instrument is a composite of the signal source, signal processing chain, power amplifier, loudspeaker and room response. If an anomaly or deficiency is observed, such as missing high frequencies, it will not be immediately apparent which part of the system is at fault. Sound personnel have a need to differentiate between the responses of each part of the sound system. Of course, the analyzer input could be directly connected to each part of the system to check its response, a method that works fine until we must gather acoustical data at an audience seat. Real-Time analyzers simply cannot distinguish between the sound arriving from the loudspeaker and sound arriving from the room. A "bad seat" in an audience area could result from a flaw in the loudspeaker, or phase cancellation caused by echoes from various elements in the room. Time domain analysis can reveal which one is causing the problem, and point toward a solution.

Time-Selective Measurements

One way of "sorting out" the sound from the sound system and the reflected sound from the room is to take advantage of the fact that echoes arrive later than the direct sound from the loudspeaker. The loudspeaker's sound is gathered and the data acquisition ended before any reflected sound arrives at the microphone, a relatively easy task for most modem measurement systems. This can be achieved by "gating" the microphone at the proper time, or by gathering the entire time record and then separating the energy arrivals as a post process (a process that takes place after the measurement has been completed). All modem measurement systems include a method for taking advantage of the time difference between the sound of interest and the echoes, yielding a "quasi-anechoic" view of the loudspeaker's performance.

Acquiring the Data

Let's take a general look at what's involved in performing a basic acoustic measurement.

Time Resolution - Sampling Rate

Numerous methods have been developed to acquire the sound and sort it out. A common element among most modem methods is the use of digital sampling in determining the response of a system. We are already some what familiar with the principles because they are already in use in other gadgets. Few people, when watching a movie, stop to think that they are actually watching a fast succession of still photos. If the frame rate is made sufficient (24-30 frames per second) the motion on the screen appears to be continuous to the viewer, an illusion but a very convincing one, as long as the frame rate is fast enough to capture the fastest motion of the action on screen

Modern acoustic measurement systems use the same principle, but the “frame rate” used is much higher. In these systems, the analyzer acquires data by making discrete digital voltage measurements at the sample points. As with the movie example, the interval between samples must be short enough so that the signal change in an interval between samples is never significant. The sampling must allow at least two samples of the highest frequency present (this is the Nyquist criterion). Considering that the audible range extends to 20 kHz, the sampling rate must be at least 40 kHz. In the movie example, the sampling rate for the spokes of moving cars is generally too low, causing the spoke rate to be "aliased" into the picture range, often making the wheel appear to be rotating backwards. The most commonly used sampling rates in digital audio are 44.1 kHz and 48 kHz. Some measurement systems allow the sampling rate to be user selected, and others fix it to allow full-bandwidth data to always be acquired. For full-bandwidth audio measurements, there can be over 40,000 individual data points for each one second of measured room response that must be processed by the analyzer. This is a lot of data, but the task is well within the capabilities of today's analysis systems.

Amplitude Resolution - Quantization

Another parameter of interest when sampling audio is the numerical value assigned to each sample. As you might guess, a large number of available values would allow for better precision and better correlation with the analog signal. If too many values are made available, processing speed is slowed.

In a gray-scale photograph, one of 2⁸ (or 256) values of gray is assigned to each element of the picture. This value was chosen because humans cannot distinguish between more values than this. Once the resolution exceeds the human ability to distinguish smaller increments, it is sufficient for the task at hand. In digital sampling of sound the quantization level is determined by the required dynamic range, rather than by the required number of recognizable levels. Since there are about 100 dB between the minimum level and can be heard and where it is un comfortable, the required voltage ratio is 2OlogR=100 or R=l 00,000. A ratio of 2¹⁶ is considered the minimum required for high fidelity systems. In most measurement instruments, the number of bits of resolution is not a user-selectable parameter, and is fixed by the analyzer or computer used to gather the data.