tape hiss / digital silence... auditory "bias"?

[Paul Frindle]

Originally Posted by Kiwiburger

'mkaaay ... here is the original quote from Paul Frindle (lifted from PSW) that I obviously misunderstood - but I think it's relevant to the discussion:



It was specifically this part that I was trying to convey: "Well the digital genny plug-in produces sinewaves correctly - but when in noise setting it is just a random number generator driving the output. So although when set to -6dB peak value no sample ever gets to be greater than 50% modulation - a reconstruction of the undecoded SAMPLE VALUES produces nearly full level SIGNAL. Reconstruction means filtering and so the filter plug-in is acting like a partial reconstruction filter (much like a DAC) - which in turn is now feeding a more legitimate SIGNAL which the sample value only meter can read more correctly.
"

Sorry I cocked this up - i'm still not 100% sure what part I misunderstood.

Yes this is true - a digital noise generator should not be just a random number generator on it's own (that is the problem with the test I described in the PSW post - the PT noise from the oscillator plug is a good example of an illegal signal when set at high levels).

What is should be is a random number generator which is reconstructed by a steep filter (like a DAC filter) in order to turn it into a signal :-)

Also there is an issue with the probability density of noise - random number gennys are square probability, more complex noise can be made by adding several uncorrelated number gennys together etc..

Noise that comes from analogue sources in the physical world is pretty much Gaussian distribution - which is theoretically speaking equivalent to an infinite number of gennys added together.

We call the number of gennys added together the 'order'. By the time you reach 3 or 4 orders you cannot hear the difference between this and true Gaussian noise - even if it is very loud and on it's own.

Noise is a fascinating subject and making digital noise of one kind or another has been a major feature of much of my work since the late 1980's...

[Kiwiburger]

Gotcha - thanks. I guess my experience with digital white noise has mainly been with dodgy VST synths and freeware app's that generate white noise.

Would I be correct in assuming that if a white noise signal is 'illegal' at high levels, it could still be a problem at low levels because it is summing with the entire mix? Sooner or later, the entire mix is going to be slammed up close to 0dBFS, and I'm thinking that a layer of illegal activity down in the noise floor is probably not a good thing ...

Maybe it doesn't matter - I was just speculating that, given the choice, any noise used on purpose is probably better off coming from analog sources. Or at least made 'legal' with a filter.

On a slightly related topic ... do you think that two channels of unrelated noise, panned left/right adds a widening effect to a mix? A lot of people talk about using outboard summing for wider mixes. A lot of these summing boxes seem to have fairly active noise floors, compared to digital infinity, and i'm wondering if it's simply the noise (hiss & hum) that is adding to the experience. Crosstalk and leakage would surely shrink the stereo image - so i'm guessing this is what is doing it. Or the fact that the panning and frequency response wouldn't be 100% dead centred, because component tolerances aren't that accurate.

I'm thinking of something like a plugin that provides a choice of "high end" analog noise floors. It would probably take the place of a dither plugin.

[Paul Frindle]

Originally Posted by Kiwiburger

Gotcha - thanks. I guess my experience with digital white noise has mainly been with dodgy VST synths and freeware app's that generate white noise.

Would I be correct in assuming that if a white noise signal is 'illegal' at high levels, it could still be a problem at low levels because it is summing with the entire mix? Sooner or later, the entire mix is going to be slammed up close to 0dBFS, and I'm thinking that a layer of illegal activity down in the noise floor is probably not a good thing ...

Maybe it doesn't matter - I was just speculating that, given the choice, any noise used on purpose is probably better off coming from analog sources. Or at least made 'legal' with a filter.

On a slightly related topic ... do you think that two channels of unrelated noise, panned left/right adds a widening effect to a mix? A lot of people talk about using outboard summing for wider mixes. A lot of these summing boxes seem to have fairly active noise floors, compared to digital infinity, and i'm wondering if it's simply the noise (hiss & hum) that is adding to the experience. Crosstalk and leakage would surely shrink the stereo image - so i'm guessing this is what is doing it. Or the fact that the panning and frequency response wouldn't be 100% dead centred, because component tolerances aren't that accurate.

I'm thinking of something like a plugin that provides a choice of "high end" analog noise floors. It would probably take the place of a dither plugin.

Some good points here - and it is a big subject, but let me try to be brief :-)

Illegal noise signals will be made legal again by the reconstruction filter in the DAC - providing that they don't get big enough to cause intersample peaks that the DAC may not decode properly. The problem is that level meters on DAW look at sample values only and do not show you what will be reconstructed and decoded. You need something like the metering I provided in the Oxford Limiter if you want to push signal to very high levels and avoid DAC overs - whether it's programme, noise or both.

As for noise on top of large modulated signal; if for instance your added noise is at say -40dB, that means it's only 1% of the total signal - so it's unlikely to be significant in respect to the rest of the programme.

On the business of correlation between L and R and stereo image; depending on how you listen to the music, you are right that complete decorrelation of noise at higher levels can seem to widen the stereo image, particularly on headphones where there is no acoustic mixing between L and R..

This is because for a 'flat and uniform' field the mind seems to want to hear some bias towards the centre - without it we hear a partial 'hole' in the centre instead. You can try this out roughly - if you have 2 channels of independent noise, by providing another channel to mix these together and injecting it into the centre at around -6dB . It then sounds more spatially neutral and balanced.

But you must be very careful about what you think is decorrelated noise! Whilst we go to loads of trouble when making our plug-ins to decorrelate dither noise between channels, things like the PT oscillator do not. So if you have two of these running on L and R there is no guarantee that they are decorrelated at all - and in fact the only difference there may be could be nothing more than a delay depending on when the instances were initiated at start up - i.e. exactly the same number sequences started at different times!
And even more confusingly, mixes saved away and restored might force the oscillators to start at exactly the same moment - so what seemed decorrelated when you were making a mix might become correlated again the next time you load the mix!!

If you are using noise as an effect in mixing, we need to make you a 'real' stereo noise generator with the right decorrelation and L, C, R weightings :-)

This is a complex subject - I hope my short description helps..