Nonlinearity and Distortion

Linear operations have inherent advantages: a sinusoid remains a sinusoid, a waveform can be analyzed into Fourier components and synthesized from Fourier components, operations can be cascaded in any order, and the result of such cascaded operations is derived easily and uniquely by convolution or correlation. However, no system is completely linear. Consequently we need a measure of the departure of a system from linearity both as a check on the validity of processes we apply, and as a figure-of-merit for the instruments we use. When a sinusoid is injected into a nonlinear process, the output is no longer sinusoidal. For most types of nonlinearity, however, the output still repeats at the same frequency. Therefore the output can be regarded as the fundamental of frequency f plus the harmonics of frequency 2f, 3f, 4f, ... This is a matter of everyday observation; if we turn up the volume of an inferior radio we can actually hear the high frequencies buzzing on the louder passages. We say that the sound is "distorted"; the effect is called harmonic distortion. A nonlinear operation changes its nature according to either the magnitude or the polarity (or both) of the input. Thus, Figure 1 ( such a mechanism generates only odd-order harmonics )(b)

Figure 1

shows an input-output relationship which turns the sinusoid of Figure 1 (a) into a "clipped" sinusoid; the nonlinearity depends on the magnitude, and the output waveform remains symmetrical about the axis. Figure 1 ( such a mechanism generates even-order harmonics )(c) shows an input-output relationship which

turns each negative half-cycle into a positive half-cycle ("rectification"); the nonlinearity is one of polarity, and the output waveform is asymmetrical about the axis. These two examples illustrate an important distinction: Symmetrical nonlinearity generates only odd-order harmonics (3f, 5f, 7f, ...); asymmetrical nonlinearity generates only even-order harmonics (2f, 4f, 6f, ...). Since no energy is added by the nonlinearity, the appearance of the harmonics is accompanied by a loss of the fundamental. Figure 1 (c) is an extreme case of this; the output is all harmonics, of even order, and the fundamental is destroyed. In general, any recording instrument generates both odd-order and even-order harmonics. Usually, the contribution of the harmonics decreases with increasing order, so that the most serious harmonics are the second and the third. The usual measure of nonlinearity is called total harmonic distortion. In effect, it is the ratio of the rms amplitude of all the harmonics to the rms amplitude of the fundamental; the details are given in the section titled "Measurement of Distortion," which appears under the heading "References and Additional Information." The total harmonic distortion of a recording instrument ordinarily increases with amplitude in the manner of Figure 2 ; the maximum signal the equipment can handle is therefore defined at an acceptable level of total harmonic distortion.

Figure 2

The total harmonic distortion is often a function of frequency also; it follows that an instrument specification of so-many percent of total harmonic distortion is meaningless unless the amplitude and frequency of the input are given also. There are two cautions in deciding what level of harmonic distortion is acceptable. The first is concerned with filtering. If the nonlinear process is followed by filtering that cuts the low frequencies (or by inverse filtering that boosts the high frequencies), the harmonics are aggravated relative to the fundamental. Instrument designers therefore try to place any low-cut filtering before the signal is exposed to harmonic distortion. The second caution is concerned with intermodulation distortion. If we consider an input signal containing two frequencies, instead of one, we remember that the output of a linear system is a beat, formed of sinusoids whose frequencies are the sum and the difference of the input frequencies f1 and f2. The output of a nonlinear system, however, also contains the sums and differences of all their harmonics: 2f1 + f2, 2f1 - f2, f1 + 2f2, f1 - 2f2, and so on. For transient signals containing a band of frequencies, this obviously becomes too complicated to think about; our normal practice is to insist on a low level of total harmonic distortion as measured on a sinusoid, and to trust that this ensures a low level of intermodulation distortion.

Applications
Before leaving the subject of nonlinearity, we should note that nonlinear processes can be very useful in the right place. One illustration occurs when we are stacking many versions of a signal immersed in noise, and when one of those versions contains a large burst of impulsive noise ( Figure 1 (a)).

Figure 1

The noise is so large that it cannot possibly be signal, but short enough that it must be shorter than the signal. Then a good plan is to clip the noise waveform, as in Figure 1 (b), or even to zero it, as in Figure 1 (c), before using that version in the stack. A highly nonlinear operation, but generally a beneficial one. More frequently, the right place for a nonlinear operation is at the end of a sequence of operations. Thus, if our sequence of operations has been successful in raising the peak signal amplitude above the peak noise amplitude (but not otherwise), we may choose to exaggerate the large amplitudes relative to the small amplitudes, and so to increase the obviousness of the signal; we could do this, for example, by raising all amplitude values to the third power. And it is standard practice, in seismic exploration, to give more visual emphasis to the positive half-cycles of a waveform than to the negative half-cycles. Again a highly nonlinear operation, but a beneficial one.