SOLUTIONS:
Combatting Noise
 

If there is one subject that can pervade a data centre conference or related meeting, is the subject of noise. Or, more specifically, how to get rid of it! There are many theories that present themselves the moment noise becomes a subject (and it is rather humouring to stand back and watch who really knows their subject and who are nothing more than sheep!).

Another sphere where the most wonderful theories of noise, this time coupled with issues such as 'mains impedance', is the Hi-Fi world. Someone could make themselves millions if they could successfully start a comic strip based on the very believable yet scientifically false rubbish being presented in various forms in Hi-Fi and related publications.

Here is the no-nonsense low-down on noise and how to combat it (properly!). Data centres, please take note!

Before proceeding, it is recommended you are familiar with the way noise presents, as seen here.

In all of the following explanations, it is always assumed the noise spoken of is relative to Earth. Also, there are basically two methods of providing power being unbalanced (normal Live-Neutral) and balanced (Live-Live i.e. split-phase). For more on types of supply, see here.

The emphasis is on 120V systems as used in the Americas (including the fact 120V is often supplied as split-phase 240V), and 230V unbalanced as used in Europe.



Common-Mode Noise

The basic method of ridding any noise from a system is with capacitors strapped across the current carrying conductors and Earth. When it comes to HF, the capacitors can be rather small. But, with lower frequencies around the upper harmonics (up to 10kHz) then pretty large capacitors are required to shunt the noise sufficiently.

As the noise on all conductors is equal in common-mode, it makes no difference whether you use unbalanced (L-N) or balanced (L-L) for power - you have as much noise to contend with. - If you use L-N-L (complete split-phase) then you have an extra 50% to reduce/eliminate.

Yet common-mode noise is very easy to eliminate using an isolation transformer. Even standard transformers can offer 40 to 60dB of Common-Mode Rejection (CMR) - that's a whopping 1000 times reduction in noise level. Special isolation transformers - these have a shield between the primary and secondary - can push this to well over 100 dB CMR (that's 10 000 times reduction, minimum!).



Differential-Mode Noise

Contrary to popular belief, you cannot get rid of the differential mode noise that easily. Some say by taking 120V and making it 240V, or using the two phases of a split-phase system, the noise is reduced through the two phases being naturally cancelling. Not so!

On 240V balanced systems, the differential-mode noise automatically doubles in amplitude as opposed to 120V unbalanced. However, the system impedance quadruples. This means to reduce the noise by the same ratio, one needs only quarter the noise shunting capacitance as when running on 120V unbalanced.

The problem is, the same level (i.e. absolute volts) of noise is desired. To achieve this means raising the capacitor to half that needed on 120V. Unfortunately, the down side is the VAR of the capacitor is related to Vē. As the capacitor is only half the original, the VARs imposed by the capacitor are now twice as what was found when using the 120V unbalanced. As hi-tech load is quite often leading, this is a very undesirable trait and big inductors have to be put in place to correct the power factor to near unity.



Common-Mode Component

By using a 240V balanced system, the two common-mode components of the two opposing phases cancel each other out and the only noise one is left with is the genuine common-mode and differential-mode noises. Because it is only the common-mode component being reduced, figures of only 12 to 15dB reduction are realized. The true differential and common-mode noise still being present. Only missing is the common-mode component.



Filter and System Design

Although proven techniques are available to designers of data centres and other installations requiring low noise environments, it appears there is reluctance to implement them (probably the old IT vs. Electrical conflict - c'mon, what do IT guys know about electricity!?).

If one deals with HF type noise, then only small capacitors are needed. But when dealing with harmonic-level noise (up to about 3kHz), then things become hairy with regards the size of cap required to calm this noise. Two problems are present being the power factor, and the capacitors having to shunt the harmonic noise - the latter resulting in a very noisy Earth and often in failed capacitors.

There is one way out of the hole - using LC filters. By using a large (ish) inductor (usually a few mH) and a not-so-large capacitor, one can land up with the inductor making things look a little inductive (lagging power factor), but the capacitor (being leading) brings it back to unity. Having both the L and C doing noise reduction, the Earth has very little current presented to it (because the capacitors are a lot smaller).

Because the inductors effectively 'isolate' the capacitors from any supply noise (including harmonics), the capacitors too are not being asked to shunt harmonic content from the supply and are left to only contend with that of the load attached to the filter.

  >>


| | Ask a Question |

© 31.10.05