| 5. A Question of Fidelity |
(Page 4 of 7) |
When sound is converted into digital form, the sound is sampled at periodic instants of time. Sample rate conversion ideally produces a new sequence of samples mathematically equivalent to the original. Any error is the result of the finite precision of the arithmetic operations performed. The important thing is to preserve the fidelity of the continuous signal that the set of samples represents, by making the arithmetic errors negligible compared to the signal's inherent analog noise.
There are two types of signal degradation that can occur in the sample rate conversion process. Linear errors result in a frequency response that is not completely flat, but instead having "ripple". This error is due to the fact that a sample rate converter contains an implicit filter. When the magnitude of the ripple approaches about 1dB, it becomes clearly audible. Since multiple sample rate converters may be used in the signal processing chain in a manner that causes their ripple to accumulate, an extremely small ripple is required for each conversion stage.
Non-linear errors are caused both by the specifications of the implicit filter and the errors in the computations in implementing those filters. The most common measure of the magnitude of non-linear processing errors is "THD + Noise." This measurement sums all of the distortion components and all the random noise in a signal, and compares it to the level of the intended signal, typically a pure sine waveform.
