Paul White explains exactly what noise is and offers hints on keeping it to a minimum in your recordings.
One of the major problems in recording is keeping electrical noise to a minimum. Most people are aware of electrical noise as unwanted hiss, but what is it, what causes it and how can it be kept to an absolute minimum?
A Quick Science Lesson
In electrical terms, noise is simply a randomly fluctuating voltage, producing harmonics that cover the whole audio spectrum instead of one specific tone; you can think of noise as all possible frequencies being played at the same time. Noise is generated by the random motion of electrons within electrical components, and is statistical in nature. If you take noise and feed it into a loudspeaker system without filtering it in any way, the result is known as White Noise, and if you were to analyse the energy content of this noise, you'd find an equal amount of energy in every 1Hz of bandwidth across the spectrum. Because the human hearing system is logarithmic (we judge pitch increase by octaves, not by equal increments of frequency), and because each successive octave spans twice as many Hertz as the previous one, the energy content of white noise will rise by 3dB with every octave. This is the main reason why we tend to perceive white noise as mainly high‑frequency hiss, though the frequency response of the human hearing system also has some bearing on this.
There is another kind of noise used in audio testing, which is produced by filtering white noise via a 3dB per octave roll‑off filter. This counteracts the ear's logarithmic response, resulting in noise which has equal energy per octave rather than equal energy per Hz. This so‑called Pink noise is useful in various areas of electrical testing, including PA speaker calibration and equalisation.
Sadly, there is no way to eliminate electrical noise entirely, because the mechanism that produces it is bound up with the same laws of physics that hold the universe together. However, noise can be kept to a bare minimum by careful circuit design, and just as importantly, by proper use of equipment.
Gain Structure
In a typical analogue audio circuit, the level of background noise is largely independent of the level of the wanted signal being processed, so it stands to reason that the larger you can make the wanted signal, the better the ratio between the signal and the noise will be. Not surprisingly, this figure is known as the signal‑to‑noise ratio. Armed with this information, you might assume that keeping the noise low in comparison to the signal is just a matter of making the signal level really huge. Surely, that way, the noise would be swamped into insignificance? To an extent, this is true, but there's a limit to how big the signal can be made before the capacity of the circuitry is exceeded and the signal runs into clipping distortion. A nice analogy is to visualise the signal trying to pass through a doorway with a deep pile carpet at the bottom, the carpet representing the noise floor. If the signal is too small, it will be partially obscured by the noise (sorry — carpet!), but if it is higher than the top of the doorframe, it won't be able to get through unless you chop the top off. From this, you can see that the signal‑to‑noise ratio will be different depending on how big the signal is, so when you see a written spec, you'll notice that the signal‑to‑noise figure is related to a specific signal level.
The best signal‑to‑noise ratio prevails when the signal level is as high as it possibly can be without clipping, but this is not a realistic operating situation, because it leaves no safety margin for signal peaks. In practice, a nominal operating level is chosen which may be 25dB or so lower than the clipping level, in which case we have 25dB of safety margin, or 'headroom' as it is officially called. Figure 1 shows the relationship between noise, clipping and the nominal operating level.
From the above, it should now be apparent that every time a signal is fed from one piece of equipment into another, whether a mixer, effects unit or tape recorder, the input gain must be set so as to bring the signal up to its optimum working level. Fortunately, most studio devices have some form of input metering, which makes this relatively easy. Failure to set even one input gain control correctly may lead either to excessive noise (if the gain is too low) or unpleasant distortion (if the gain is too high). The ritual of optimising input gain controls is known as setting the gain structure, and it doesn't matter whether you're working with a Portastudio or a top commercial studio system, if you don't pay attention to gain structure, your recordings won't do justice to the equipment you're using.
In a typical home recording session, the first step is to set up the mixer input gain trim to optimise the mic or line input level, and most mixers enable this level to be metered via the PFL (Pre Fade Listen) system. The mixer output must then be adjusted so that the right level is going onto tape. Any effects units plugged into the mixer should be optimised independently.
From the facts I've mentioned here, it should be evident that keeping recorded noise low depends both on the quality of the equipment being used and on how well it is set up. It is also important to ensure that no unnecessary noise is present in the original signal; for example, if you're recording an acoustic instrument in the studio, try to ensure that the mics aren't also picking up noise from traffic, your hard disks, cooling fan, air conditioning, and so on. Even if you only allow a little noise onto each tape track, the cumulative effect of 16 or 24 channels of noise can be significant.
Tape And Digital Noise
Tape machines suffer from both electrical noise and tape noise, but both should be treated in the same way. Tape behaves very much like an electrical circuit; if you don't use a high enough signal level, the signal won't hide the residual tape noise, and if you use too high a level, the tape will saturate, causing distortion. Tape noise is a statistical effect related to the individual tiny particles which make up the tape's surface. The more particles that pass over the tape heads each second, the higher the resolution of the 'sound picture' being recorded, and the lower the level of noise. In practical terms, this means that machines that run at high tape speeds, with wide tape tracks, produce less noisy recordings than low‑speed, narrow‑format machines.
Digital circuitry also suffers from noise problems, though the noise generating mechanism is a little different from that of analogue circuitry. In a digital system, the signal is represented by a string of numbers, and a useful analogy is to imagine the signal as being built up out of little Lego bricks. A six foot high sine wave made up from little half‑inch bricks might look reasonably smooth and accurate, but a low level signal of only a couple of inches in height would look like pretty crude approximation, because the available brick size is almost as large as the signal being built. Technically, this difference between the input signal and its digital representation is a form of distortion (known as quantisation distortion), but for reasons too mathematical to go into, it happens to sound like noise. The rules, therefore, are the same as for analogue; too little signal and you get noise, too large a signal and you run out of bricks and end up with a flat‑topped or clipped signal. Figure 2 should make this clear.