A simple Q: is digital equalization always a lossy process? (let's not take rounding and truncating errors into account) Is analog eq always superior to digital? Also, what's the best equalizer VST (freely available) or plugin (for Winamp)? I'm not very happy with Shibatch's super EQ because when I use the parametric equalizer, it makes the treble sound very muffled. The best one I've found is http://www.geocities.jp/webmaster_of_sss/vst/#lpgeq2 but is there anything better?

Define "lossy" in this context. Basically, if you don't take rounding, truncating and clipping into account, it's lossless because it can be undone by applying an inverse EQ curve.

It won't lose any frequencies, only amplify or dampen them.

I think you mean by "lossy" the tradeoff between phase smearing and ringing artifacts. And yes, you always get one of both, for both digital and analog equalization.

Its not considered a lossy process. I wouldn't worry about it... If it needs EQ to "sound right", the alternative is to leave it as-is... sounding "worse". And, just about everything that passes through a pro recording console is EQ'd.

But, you're correct about rounding and truncation. Mathematically, it's lossy because it's not bit-for-bit reversable. The same can be said for a simple volume adjustment.


Typically, I'd expect better results in with digital. You've generally got more precise control, and more "flexibility". You are less-likely to introduce noise, and if your digital EQ doesn't offer the settings you need (i.e. center frequencies), it's a lot easier to get a new plug-in than it is to get another analog EQ.


I have no idea why that happens... Any EQ should have no effect when set to "flat". A digital EQ can be 100% lossless when set to flat.

High-tap FIR eqs like shibatch and foobar2000 tend to have very flat passbands and very sharp, high-Q transition bands. The sharper a transition band is, the more ringing will occur.

The one time I investigated this to any detail with the fb2k equalizer, I pretty conclusively showed that its eq effects were not lossless. Ringing tended to persist after an EQ cancellation as verified in an ABX. This can occur due to slight asymmetries between boost and cut eqs, slight bugs in the eq code, etc. That said the eq may have changed substantially since then.

IIR filters like biquads, or FIR filters with gentle transition bands, do not have ringing issues.

http://www.winamp.com/plugins/details/147879 - this is a great equalizer for winamp, highly recommended. And digital eq should be more or less lossless, probably better than an analog version, and much cheaper.

There are also some other EQs at this page: http://www.savioursofsoul.de/Christian/?page_id=64

I'm a big fan of Nyquist.

I'm a big fan of Naqoyqatsi.

Digital equalization is the only kind of equalization that could possibly be lossless. A digital equalizer can be set to process the signal with flat response, or the action of a non-flat equlizer can be equalized out. The only possible losses would then be things like rounding and truncation which can be practically reduced to be as small as is desired, by simply upping the precision.



You seem to be poorly informed. As a practical matter, digital eq can be far superior to analog eq. If a diqital eq does not equal or outperform a comparable analog eq, it is a design problem.

For example, digital eq with zero phase shift, or a wide range of desired phase shift characteristics can be readily implemented.

In contrast, analog eqs are generally locked into minimum phase response, which gives you any kind of phase response for a given frequency response that you want, as long as it is just one thing! ;-)

The problem with digital eq has I think been a combination of prejudice and poor design.

Thanks to all NyquistEQ was most to my liking, I love its simplicity. Has anyone ever made any digital EQ comparision regarding how "natural" they sound?

I'm not sure about that. You can losslessly convert a digital signal from time to frequency domain (and back) using (I)DTFT - but as soon as you apply a filter transform on the frequency domain data you will get rounding errors when converting back to the time domain.

Edit: Made new post out of lengthy edit.

I don't know how to interpret that question... An EQ set to flat, should have zero effect on the sound. Small/gentle settings should sound more "natural" than drastic adjustments... An EQ set for a 6dB cut at 100Hz, should "sound like" a 6dB cut at 100Hz.

A "perfect" sound reproduction system, playing "perfect" material doesn't need any EQ. EQ is used to correct frequency response deficiencies.... If there's nothing wrong, you don't need any EQ.

In sound production, it's a different story. EQ is use both for correction/compensation and to enhance particular tracks. A recording engineer might use a particular EQ setting for female singers, and different settings with different microphones or with male singers. Each instrument will have its EQ tweaked differently. Each drum & cymbal will have it's own EQ setting.

My point is that you can reduce digital rounding error to be as small as you would like it to be by just using longer data words for your calculations, but there are very definate practical limts to how accurate analog processing can be.

I should have read the rest of the paragraph .

"upping the precision" must refer not only to the filter precision (window size), but also to the bit depth of the output bitstream to reduce rounding errors to near lossless territory. Inverse DTFT maps from a possibly infinite set of values to a limited set (65536 for 16 bit). To reduce rounding errors you have to increase the precision of the target range.

So lossless in the sense of 100% reversible by an inverse filter is not possible. But I think one could eliminate phase smearing or ringing completely by increasing the window size to the complete length of a track. Gapless playback would be close to impossible to realize then, though. For (almost perfect) gapless playback your player would have to prefetch the next track in parallel to playback and preconfigure its filter banks accordingly. But that's pure speculation. I don't think that such a filter would be ABXable against a standard biquad implementation with, let's say, a window of 4k. Well, maybe for very extreme Q values.

I'm not a DSP expert, but I think most digital equalizers are "built" with filters that don't use Fourier transforms. I believe the standard graphic equalizer or parametric equalizer uses mathematically simpler high-pass and low-pass filters (combined to create bandpass & band-reject filters).

You can do amazing things with Fourier transforms, like "draw your own curve" equalizers, and very narrow notch/bandpass filters. But, I believe these are a special class of EQ.

As far as I read it you always need a z-transform, with the same implications. DTFT is only a special kind of z-transform.