Orchard Article Series - #1 - Signal-to-Noise Ratio (SNR) Explained

[orchardaudio]

I decided to write a series of articles that would help people out in understanding audio. I have found through my experiences that good information is hard to come by. Anyway, the articles will be ranging in topics and technical levels.

Questions, comments, recommendations welcome.

Link to first article:
https://www.facebook.com/notes/orcha...3667253242373/

[Patrick Dixon]

For illustration, let's say that you're having a conversation with someone in a kitchen that also happens to have a particularly loud refrigerator. Let's also say the refrigerator generates 50 dB of hum (consider this as the noise) — a loud fridge. If the person you are speaking with chooses to converse in whispers (consider this as the signal) at 30 dB, you won't be able to hear a single word because it is overpowered by the refrigerator hum! So, you ask the person to speak louder, but even at 60 dB, you may still be asking them to repeat things. Speaking at 90 dB may seem more like a shouting match, but at least words will be clearly heard and understood.

This is misleading because it doesn't take into account that the 'noise' is not related to the signal and you are often able to filter it out even if it's a relatively high level.

If what you've written was true, you would only be able to hear the loudest instrument in a piece of music at any time, because it would overpower all the quieter ones.

[orchardaudio]

This is misleading because it doesn't take into account that the 'noise' is not related to the signal and you are often able to filter it out even if it's a relatively high level.

If what you've written was true, you would only be able to hear the loudest instrument in a piece of music at any time, because it would overpower all the quieter ones.

For the particual example let's assume the refrigirator noise is broadband and cannot be filtered. That is the way the article is intended to be read.

[Patrick Dixon]

It's not whether it's broadband or not, it's whether it's correlated with the signal.

Refrigerator 'noise' is never going to be correlated with the conversation, so it's misleading.

Will removing the refrigerator noise make the conversation easier to follow? Yes it will, but that's not to say it stopped you from hearing the conversation at all, even if it's at a higher level than the speech. The problem is that on reading your explanation, people might think that because musical information is below the SNR of a system it doesn't matter as it can't be heard, but that's not the case.

[orchardaudio]

I don't disagree with you, this article was written with a very different target audience in mind.

[Patrick Dixon]

I appreciate that, and good on you for trying, but I still think it's important to write what's correct even if you are trying to simplify a complex subject.

If you say that it will be harder to make out the conversation with the refrigerator noise, rather than you won't hear it at all, then you won't be being incorrect.

As an example of unintended consequences I have seen people justifying that the benefits of 24 bit digital audio cannot possibly be heard because the extra resolution is below the SNR ratio of the reproducing equipment :-(

[Pharos]

Well, my fridge is always trying to interrupt me.

[Barry]

During the War, signal interceptors could discern Morse code and transcribe it without error, even when it was 'buried' in noise that was 20dB louder than the signal. This was because the noise was 'white', broadband and uncorrelated, and being 'constant' can be ignored by the brain.

A similar phenomenon is the 'cocktail party effect', whereby one can easily follow a conversation being held by a couple across a crowded and noisy room, despite the fact the conversation going on near you is at a greater level.

[Macca]

As an example of unintended consequences I have seen people justifying that the benefits of 24 bit digital audio cannot possibly be heard because the extra resolution is below the SNR ratio of the reproducing equipment :-(

Not the best counter example, seeing as no-one can hear the difference when put to the test, and bit rate relates to dynamic range, it has nothing to do with frequency response in any case.

[Patrick Dixon]

Not the best counter example, seeing as no-one can hear the difference when put to the test, and bit rate relates to dynamic range, it has nothing to do with frequency response in any case.

Thanks for that. I didn't mention FR in any case, but it seems my point was just too technical not to pass you completely by.