Sampled data processing is now a fundamental part of all aspects of electronics. As well as the obvious fields of audio, video, and communications, it is also used in robotics,
AI/ML, control, and energy systems. All these systems rely on the processing of digital samples in some way.
Shannon-Nyquist sampling is so pervasive that we can forget that, like any theory, there are terms and conditions attached. In particular, it assumes that a signal is processed in a linear fashion and that there are no time-varying operations. One example of where this can break down is in the simple act of measuring a waveform's level.
Measuring level is an important part of many audio devices. Meters, limiters, compressors, etc. all require accurate, and often instantaneous, measures of a waveform's value. However, because the audio waveform is sampled, the actual level may not be a sample value, but instead may be an intermediate value between samples which may be higher, or lower, than the adjacent samples. Various methods have been proposed to ameliorate this.
This talk, which will be accessible to a wide variety of backgrounds, will discuss the issue of deriving the true signal value from the sampled audio. Sometimes it is necessary to reconstruct the analogue waveform to have accurate results. The pros and cons of various methods of measuring the peak value will be discussed, including a simple approach that provides accuracy with low computation and delay. Finally, we will conclude with a discussion of why knowledge of the intermediate sample values is required for other non-linear operations, and some methods of achieving this.
By the end of the talk, we will realise that, "The Samples Truly Are Not What They Seem!"
