Monthly 2026, 08 - Microphones
I have a project that concerns itself with cameras and microphones, and to start what I'm doing, I need to understand how exactly the sensors work on a physics and decoding level. I looked into a history of cameras before on this series, but this time I want to get into the details of what is common nowadays. I'll start with Microphones. Sound transmits as transversal "acoustic" waves through a medium. I'm primarily interested in what a person hears, vs. what a technological solution can pick up as a signal. Ignoring any higher modes, a person's audible spectrum is between 20Hz to 20kHz. Anything beyond that is considered not perceivable, but given high enough amplitudes can still have physiological effects one would generally want to avoid. Infrasound (i.e. sound below 20Hz), similarly can permeate biological tissues due to their frequency, which can directly interfere with the strength of muscle contractions, and stimulations of hormonal glands.
Microphones function initially in a very similar function to the human ear, in that there is a membrane that vibrates with the transversal waves in the medium (usually air) and that frequency and amplitude is translated into pitch and volume. Because of that, there is also a range of frequencies that a microphone can usually pick up. This band is the "frequency range". Modern microphones can pick up frequencies from 1Hz to around 140kHz, which eclipses the human hearing range, by about two orders of magnitudes. On the consumer side, most microphones have a high-frequency cut-off at around 20kHz, mostly because that's what the person interested in recording the sound hears as well.
When recording or replaying audio, the data is actually a series of samples that are chained together so fast that it feels like a continuous piece of music to a human listener. Because we don't really want huge amounts of data required for lossless encoding in most cases, so I thought it would be interesting to also take a look at how the usual audio encoding algorithms work. Audio recordings come with a bitrate, which actually describes the number of audio samples played per second. The standard bitrates are 44.1 kHz, 48 kHz, 96 kHz or 192 kHz. It is also true, that technically a human biological estimate is several orders of magnitude higher, but people can't really tell the difference between small compression differences. On protocol layer, an mp3 file consists of repeated blocks of mp3 headers and mp3 data, which together are described as an mp3 frame. The headers are mostly created to keep each of the frames in correct order. There is also a checksum, though it's optional.
The fact that the checksum is optional is really interesting for people who want to mess with audio files with relatively low overhead, which would be beneficial for real-time high-fidelity machine-editing of the file, which is primarily what I was going for.