Equalization for Rookies (special guest Christian Siedschlag, DDMF)
Christian Siedschlag, the man behind the DDMF, knows a thing or two about equalization since he created a respected EQ plug-in. He shares some of his expertise with us here.
DDMF – it is a name that suddenly popped in a field crowded with a big names, representing an independent “one man” company that become a subject of a many rumors on various music related forums, such as “Did you hear about that cheap equalizer that sounds better than most of the expensive ones?” At the time, I was involved in an extensive search for my main equalizer.I tried most of the expensive ones and they sounded good. But some of them didn’t had enough bands, other offered editing only through the knobs, some of them even don’t sound as they should. Some sounded good but were CPU intensive. I didn’t expect too much from this newcomer. What a mistake.
To make long story short, Christian Siedschlag, the man behind the DDMF, created an equalizer that became my main secret weapon, one that I’ve used for many years on every song on literally every track. His IIEQ Pro can sound analog or digital, it offers a wide range of filters, and what’s more important, it sounds right. It is light on CPU, easy to operate and no matter what you use as a source, it always sounds natural, no matter if you cut or boost the material.
We all use equalizers, but what do we know about them? Let’s draw back the curtain to discover some basic facts about this tool, and even more, to unveil the secret – how those tools are made in the first place. So, let’s hear the truth from the big master – Christian from DDMF.
More about DDMF on http://www.ddmf.eu/
Equalization for Rookies…
Remember that old saying “Talking about music is like dancing about architecture”? Well, to some extent that statement is probably also true when talking about the art of mixing music: it remains an abstract matter when you don’t have actual sound examples at hand, which is necessarily the case in an article like this. Plus, many of the world’s greatest mixing engineers “feel” more than they “know” what to do in which situation. Fortunately (for our purpose), I’m definitely not one of the world’s greatest mixing engineers but just a developer of audio software. By nature, I’m forced to translate my intuitive understanding of things into working software code, which is why there’s good reason to hope that I might be able to explain a thing or two about digital signal processing, and about equalization specifically, which is the subject of the present article. The target audience, as stated by the title of the article, is “rookies” so for the more experienced readers a lot of things will probably sound familiar.
What is Equalization?
Let’s start with the definition: what is equalization? These days we fortunately have Wikipedia, so we can readily answer “Equalization (British: equalisation) is the process of adjusting the balance between frequency components within an electronic signal“. This definition implies that any electronic signal has frequency components, and indeed it is this mathematical fact that lays the foundation for all techniques of equalization. Sound, if viewed from an engineering perspective, is nothing but a sum of individual frequencies (sine waves) played together. The relative strength and timing of these frequencies with respect to each other uniquely describe a certain sound (or a whole piece of music) and is called the “spectrum”. Equalizers are devices (real or virtual) that allow the mixing engineer to lower or raise parts of the spectrum without affecting other parts of the spectrum.
In Fig. 1 we see an example of a spectrum, frozen in time. The horizontal axis is the frequency axis, showing at what frequency a contribution to the overall sound is oscillating, while the vertical axis measures the strength (loudness) of each contribution. The frequencies are measured in Hertz (which are just oscillations per second) and the loudness in decibels (dB). The huge peak at about 80 Hz is the bass drum. We see a lot going on in the range between 200 and 5000 Hz, which is where vocals and all the other instruments are fighting to be heard, and then a slow drop off towards higher frequencies. The sharp drop off at about 16 kHz is a result of mp3 compression, by the way. The flat white line is the equalization curve, as the screenshot is taken from one of DDMF’s EQs, without any equalization applied.
The Task of EQing
You’ve probably already come across the following situation a few times: you have a track with a number of instruments and maybe some vocals which all sound great when played individually, but after mixing them together, it just sounds muddy and crowded. That great guitar line suddenly is barely noticeable, and when you raise the volume on the guitar track, the vocals start to disappear. This is where you need to reach for your EQ. EQing is all about carving space in your mix so that the individual tracks don’t get in each other’s way.
In the beginning (and not only in the beginning) it is very helpful to have a chart that approximately shows where in “frequency space” instruments typically have the strongest components. There are a lot of these charts available on the internet, and one of them is shown in Fig. 2. These types of charts can help you when deciding where you need to apply equalization. Also shown are typical attributes that are often associated with certain frequency ranges, e.g. “warmth” at about 150-220 Hz. This means that when a track is lacking “warmth” that frequency area might be a good starting point for a little peak filtering. Which brings us to the next subtopic, namely …
Types of EQ Curves
There are equalizers out there which, using a technique called FFT, allow you to change the spectrum of a track in a free-form way. While this approach clearly offers the greatest amount of freedom and flexibility, you can also quickly experience a phenomenon called “choice paralysis”, where you have so many options and variables that in the end you can’t really decide in which direction you want to move. Especially in the beginning, it is much better to stick to tried-and-true “EQ curves” which over the years have been implemented time and again, partially because they were relatively easy to achieve with hardware, but also because they produce predictable and pleasant sonic results. These types of curves can be controlled with a very limited number of parameters, depending on the type of EQ you have. An equalizer typically offers a number of “bands” which are individual filters through which the signal passes in series. The simplest form of an equalizer, a “graphical” equalizer, only allows the change of one parameter per band, namely, the gain of the band (in dB). In the age of software EQs, however, this is an unnecessary restriction: here we’ll be dealing with the more general “parametric” EQs in which each band can be controlled by setting gain, frequency and something called “Q” or “width”, which determines the width of the range in the spectrum the band is operating on.
1.) High-pass/Low-pass Filters
These filters gradually block all frequencies below (high-pass) or above (low-pass) the band frequency. There’s no gain control, and the width can be used to change the filter response around the band frequency from very smooth to (in the extreme case) resonance-like. These filters are very useful to clean up the upper and lower end of the frequency spectrum. It is often a good idea, for example, to high-pass all but the kick and the bass at around 80-120 Hz. The lowest string of a guitar, for instance, has a frequency of 82 Hz, so everything below that can be safely discarded. Also, many engineers apply a low-cut filter to the whole bus signal at around 5 Hz, as this area can’t be heard by humans anyway and is only eating up unnecessary energy.
Shelving filters boost or cut the frequencies above/below the threshold frequency by a fixed amount of dB. When cutting, they are a little less drastic than low- or high-cuts in the sense that they do not progressively lower all frequencies above or below the threshold frequency. Again, Q is used to shape the response around the threshold region.
3.) Peaking Filters
Peaking filters can be used to treat an isolated range of the frequency spectrum. They have a center frequency around which the response is symmetrical. There’s a boost or cut by the specified amount of dB at the center frequency, with a smooth fall-off around the center frequency. The width of the “active” frequency window is set by the Q value. Peaking filters can either be used to enhance specific areas of a track’s spectrum in order to make it heard more clearly (something which is easily overdone, though, so be careful!) or to reduce “annoying” areas. For instance, in order to decrease the “muddiness” of a mix, it is often a good advice to apply a broad cut by 2 or 3 dB at around 300-500 Hz to the master (sum) signal.
4.) Bandpass Filters
Bandpass filters leave the center frequency untouched but cut out an increasing amount of dB with increasing distance from the center frequency. This type of filter can also be used to effectively decrease the necessary bandwidth to transmit a signal. A famous effect is the “telephone voice” which can be generated by a single band pass filter set to about 1000 Hz, with a bandwidth of about one octave. While often sounding thin when applied to a single instrument, it can help to make the instrument sit better in a mix and create space for the other competing tracks.
5.) Notch Filters
A notch filter is the exact opposite of a band-pass filter: it completely cuts the spectrum at the center frequency, and gradually less so around the center frequency, until it reaches 0 dB gain outside a window determined by Q. A notch filter is very useful for removing annoying or problem frequencies (for instance a 50 Hz humming from the power supply). A nice technique when you have the feeling that a track has some annoying component but you can’t exactly figure out where it sits is to apply a large-gain, narrow peaking filter and slowly sweep it across the frequency spectrum. When you have localized the problematic area, apply a notch filter there, with a width that’s large enough to be effective but small enough to avoid any unwanted side effects.
These are the basic filter types that are available in almost all of today’s (software) EQs. The implementation details may vary, especially when you start comparing Q values … almost every developer uses his own definition. Another point that’s influenced by the implementation is the CPU consumption. When the number of tracks to be equalized in your projects is huge, this is something that will definitely become important at some point.
Design of (Software) EQs
Although you can happily use your EQs without knowing too much what’s going on under the hood (pretty much like you can drive a car without having to be a mechanic), it can be useful or at least interesting to know at least a bit about how these filters are actually made. A thorough exploration requires quite a lot of background knowledge in math and would be beyond the scope of this article, but I’ll try to explain the process briefly, and in laymen terms as much as possible.
There are two basic approaches: the time-based approach and the frequency-based approach. What does this mean? Well, digital audio is, as you probably know, represented by samples that are being delivered at a certain sample rate, typically 44.1 KhZ. The time-based approach calculates the output of a filter at any time by “simply” calculating a weighted sum of the current sample, a certain number of sample before the current sample and (in general) a certain number of previous outputs of the filter. This is called “time-based” since it only looks at how the samples are coming in one after each other, there’s no direct attempt to measure the frequencies that are present in the signal. Nevertheless, with the correct weighting of samples, it’s possible to enhance or decrease the contribution of only a range of frequencies, just like in the filter examples shown in the previous section.
A very simple, intuitive example is a filter that calculates its output by simply summing the current and the previous sample. You wouldn’t expect that there’s anything useful coming out of this operation, but actually this is the simplest form of a low-pass filter! You can convince yourself of this fact when you consider a signal that only consists of the highest possible frequency that is available for the sample rate at hand (the Nyquist frequency). This signal consists of the sequence +1, -1, +1, -1, +1, -1 … wildly oscillating, as you can see. Now what’s the output of our simple filter? It’s always either (+1-1) or (-1+1) depending on the sample position; in any case, it’s identical to zero. This means that the Nyquist frequency is completely blocked. On the other hand, if we only have a DC component in our signal, the sequence of samples would look like this: 1 1 1 1 1 1 1 1… (or any other number different from 0, depending on the strength of the signal). Clearly, the output of the filter would be 2 2 2 2 2 2 … so the DC component is enhanced. All frequencies between are interpolated, which gives the frequency response of a low pass filter.
The time-based approach is working well in most situations, and there’s a whole theory behind it that is also concerned with how to find the weighted coefficients for the summing of the samples to match a given electric circuit as closely as possible (which opens the door for simulating “classic” or not-so-classic pieces of hardware in software). One issue, however, is that close to the Nyquist frequency, the filter responses often become less ideal since you’re always translating a system with continuous time (your analog filter) to a system with discrete time (your simulated filter). There are remedies and tricks to avoid this to some extent, but ultimately the “cleanest” option is to use frequency-based filters (or FFT filters, named after the technique of Fast Fourier Transformation which is usually used for this). Briefly, what is done is instead of summing and subtracting the sample values “live” as they are floating in, one waits a little while until one has enough samples at hand to perform an analysis of the frequencies that are contained in the sample set. The nice thing is that there is a mathematical operation (the Fourier Transform) which, for a number of samples, calculates the contained frequencies (the spectrum of the signal) but also, for a given spectrum, calculates the samples that produced the spectrum. There’s a one-to-one correspondence. This means that one can shape and bend the spectrum in any way one wants, and then calculate the samples that result. So if one wants more gain at around 100 Hz, no problem, just add a peak there, easy enough. Within certain limits, any desirable spectrum can be generated, which is why FFT based filters are usually what’s being used in free-draw EQs. But also the area around the Nyquist frequency, which is critical with the time based approach, poses no problem for FFT filters.
There are two draw-backs, however: the FFT operation is taking more CPU power than the time-based approach, and, since you always need a certain number of samples to get some precision in your frequency analysis, a delay (latency) is inevitable. This is why FFT filters are usually not considered “tracking EQs” (of which you place at least one instance on any track in your project), but they rather belong on the master bus.
While there’s obviously a lot more to say about equalizers and equalization, the material presented here should give you a good starting point to begin with your own journey into this field. I recommend that you not go into gear-hunting mode during the first few months, but rather find a cheap or even freeware parametric EQ and try to learn the basic principles first. You’d be surprised what a good engineer is able to produce already with low budget plugins. One thing to look for is the option for mid/side EQing, which is a simple yet effective method to treat your center and side signal separately (especially useful for creating a solid, “centered” bass and a more “airy” stereo field).
In the beginning though, the most important part is to learn how your specific mixing setup sounds (that includes the speakers and the room). It is definitely advisable to always compare your own mixes to one or more reference tracks which sound more or less the way you would like to sound, and to switch back and forth between the reference tracks and your own material frequently. The ear quickly adapts to changes in the frequency spectrum, and after a few minutes a 10 dB boost at 1 kHz will sound almost natural if you don’t have any standard reference to compare it to. Also, when it comes to frequency analyzers (like the one presented in Fig. 1), you shouldn’t use it too much initially, but rather train your ears first.
That’s about it! Hope you enjoyed the material, and happy mixing!
by Christian Siedschlag, DDMF