Microphone Clipping, The Three Types Explained

[@goaway2174]

Thank you for the video. I have read your content at the link as well and have been working with digital electronics since 1984, specifically industrial & vehicular electronic control systems. My degrees are in physics. While you make many valid points, I think there is a misunderstanding about what 32 bit floating point can and cannot solve. It can’t fix a microphone diaphragm excursion limitation or a preamplifier that is gained so high in level that it is distorting the original performance. The problem it solves is exceeding 0 dBfs during recording 24 bit in the digital domain. You may have set levels properly and used appropriate gain but an unexpected exceptionally loud event occurs and causes digital clipping. In a 24 bit recording this cannot be corrected as the data captured exceeded the maximum value that can be represented. The data is incoherent and the sound cannot be reconstructed. The way 32 bit solves it is it has 0 dBfs set equivalent to the loudest level digitally represented by 24 bit, so the clipped event is captured unclipped at a level exceeding 24 bit but easily represented by 32 bit. That is it in a nutshell and I own the sound devices mixpre 6 II but don’t really ever use the 32 bit float because I don’t record performances in such uncontrolled circumstances. If you want more information, this is a good explanation: www.sounddevices.com/32-bit-float-files-explained/ Best of luck to you!

[@MaxoticsTV]

As I mention in my essay, you cannot exceed whatever voltage is expected as a maximum, 0dbfs. So you'd need to explain to me in greater detail how that is possible. Further, 32-bit-float is a different animal than 24-bit fixed. It uses 8-bits to widen the scale which has no effect on the precision. From a fidelity point of view, 24bit and 32-bit float resolve to the same data. And most analog sources don't resolve past 16 bits--and even that a stretch. The sounddevices link you mentioned is flawed IMO. Why I took such great pains to go through each step. To me it is fraudulent. Sorry!

[@goaway2174]

I have 2 A/D converters in parallel fed by my analog audio signal. One converter has a reference voltage of 1V representing O dBfs the other has a reference voltage of 5V representing 0 dBfs. Something that primitive isn’t how sound devices are doing it but it is along those lines, perhaps the reference is variable or perhaps they have broken the digital data across multiple converters. I agree that 16 bits is adequate for reproduction.

[@MaxoticsTV]

@@goaway2174 There can only be one analog stream of electrons from a microphone. Any split of that stream would add noise and/or reduce resolution. I've seen diagrams from Rode where they show a split analog stream...which angers me ;) Of course, you can amplify a source into as many streams as you want, and set them to any voltage, but again, without the inherent resolution between each high and low (more than 16 bits) what can you accomplish? I've asked Sound Devices to clarify that question. No answer. They sent me some files. They must not know I can tell when data has been doctored. All that said, I understand how one can use multiple streams for production reasons. I understand why one might use 32-bit float just to keep the whole workflow standardized. I always want to repeat. Do what works! Just don't make up your own physics ;)

[@goaway2174]

Who says there can only be one stream? If you wanted you could have several outputs using a custom transformer. But let’s say it is a normal microphone and you have one balanced output. What’s the big deal? We measure voltage into high impedances not current (the current actually drawn in this case is seen as insignificant). So sorry but I am not understanding your position of 1 current stream from a microphone, it would get split in any number of different ways in the analog domain anyway. You have attenuators, voltage dividers, feedback networks on op amps, etc. all in the preamp. The signals from microphones are viewed as low voltage (at least in my experience) with voltage gain needed to get them to a useable level. E.g. a Shure SM7B has 1.12mV output with 1 pascal pressure (94 dB) at the microphone diaphragm. I guess I don’t comprehend why you have the belief that splitting this current stream has such significant consequences. There are so many areas that have multiple orders of magnitude greater impact (impedance mismatch, poorly terminated cable, emi fields, wrong microphone choice (pattern, type, sensitivity)).

[@maxcanthelpit]

@@goaway2174 When a magnet moves across the wires it vibrates (theoretically) single electrons. Transformers, impedances, attenuators, voltage dividers, feedback networks, op amps--all word salad. They all have impacts, yes, and all negative to the original signal. Although I have studied each one in turn, none add any information, or improve the accuracy of measuring those electrons, which must be amplified to write memory (in our real world). Some use such terms as these to create an argument that supports what they want to believe--that 32-bit float improves recorded microphone fidelity (reduced clipping). I have not seen any experimental proof that any proposed technique works; that is, anything recorded on two of the same mics going into two pieces of recording equipment, one 24-bit and the other 32-bit float, where, once the gains were set properly for each, resulted in reduced clipping. I have done experiments myself and have found no difference whatsoever. So I have my theoretical explanation of why it's not possible and my tests. Of course, people love to create words and theory about why something works. You can especially see this in skin care products. And btw, I couldn't care less about this subject from an audio point of view. What fascinates me is how so many people can't let go of their preconceptions no matter how much evidence to the contrary they are exposed to.