Live Sound University Article Mon, December 01, 2008

How do you correct phase shift? You can only do so much to an EQ before you lose all your power so how do you correct it correctly(?)?

Mike G.

Reply posted by Tom Young on December 10, 2001
Any filter (equalizer, crossover, high/low pass) exhibits phase shift that is increased as the gain of the filter is boosted or cut. Zero boost or cut equals zero phase shift. Despite some aggressive marketing hype, there are no filters or equalizers that do not cause phase shift. Digital filters emulate analog filters and they therefore also have phase shift.

So, unfortunately the answer to your question is: “reduce the degree that you are eq-ing and the numbers of filters you have implemented”.

But you may be describing a phenomena that is also or mostly caused by the attenuation that has resulted from over-EQ’ing. The more you EQ out (cut, specifically), the more sucked-out the overall sound becomes. Phase shift is also pretty bad in this situation, but it is primarily a case of your having removed too much of the signal.

Yes, this may be deemed ‘necessary’ due to poor equipment mixed with horrible acoustics and/or hopeless mic techniques/requirements… but there is no magical way to reduce the suck-out other than minimizing the EQ.

Tom Young
Electroacoustic Design Services

Reply posted by Paul Henderson on December 10, 2001
Something worth noting here...as long as the system is behaving in a minimum-phase manner (at some physical point of measurement), the phase characteristic of the system is directly related to the magnitude response.

By applying minimum phase corrective equalization filters to the system (analog or digital), you are then effectively correcting both the magnitude and phase response of the system simultaneously (at that specific acoustic position). Without getting into the details of what makes up a minimum phase system and how this may vary over space in electroacoustic systems, suffice it to say that the nonlinear phase response in corrective equalization is a necessary and useful product of the application of filters.

Without it, you would be dealing with a highly nonminimum phase electrical response, which brings with it drastic time-domain and acoustical difficulties.

As Mr. Young pointed out, over-EQ’ing will be the cause of problems, specifically when the filter is not matched to the minimum-phase anomaly it is attempting to correct. This brings with it both problems in magnitude response, but also increased group delay and hence time smear near the filter resonance. This all goes back to the ability to accurately measure a system and intelligently interpret results using a transfer function measurement platform like SmaartLive or SIM.

Paul Henderson

Reply posted by Mike G. on December 11, 2001
EQ does come into the picture though. CD horn EQ is radical. This causes lots of phase shift. IF you look at Servodrive subs, they are not acoustically flat from 30-160hz therefore they need EQ. Now we’ve introduced phase shift. How do you correct for this problem outside of your EQ?

Mike G.

Reply posted by Andy Peters on December 11, 2001
-"CD horn EQ is radical. This causes lots of phase shift.”

CD Horn EQ is a 6 dB/8ve rise above its cutoff. Not very radical. And, I believe the CD horn’s roll-off is minimum phase, which means it CAN be corrected by electronic EQ.

Reply posted by Thomas Danley on December 11, 2001 at 02:44:04:
-"EQ does come into the picture though. CDhorn EQ is radical. This causes lots of phase shift. IF you look at Servodrive subs, they are not acoustically flat from 30-160hz therefore they need EQ. Now we’ve introduced phase shift. How do you correct for this problem outside of your EQ?”

A side point, once you have 2 or more BT-7’s (the horns) together, the response is reasonably flat across the band, a group of 6 is + - 1.5 dB 28 -110 Hz with an acoustic phase near zero degrees over much of that band. (see TEF measurements at web site if interested) Even 2 have less frequency response deviation than one would see on most conventional VC sub boxes going from a cold to hot voice coil.

Your real issue, however, is with phase shift, it seems to me.

An important concept, a “minimum phase” thing like a speaker or most electrical networks (not including all pass filters) has a fixed relationship between amplitude and phase. One cannot change the amplitude without also changing the phase a corresponding amount.

If one had a minimum phase speaker which had a non flat response, one could use a normal filter or EQ to make both the amplitude AND phase flat at the same time as the filter correction has the right amplitude and phase to counter act the speakers response .

In the case of your CD horn, many drivers do have some non-minimum phase response (due to the mechanical bits) but for the most part, the horn and driver response can be EQ’d reasonably flat leaving only the driver’s non minimum phase part.

Where it becomes much more complicated is when you include more than one source (or a reflection), now the problem is in 3D. Now, any separation between the two sources is the same as a phase shift between them, if the phase shift is significant (more than 1/4 wavelength) then the output of the array is directed according to the pattern of cancellation and addition. As music is a time variant signal, one can also look at the difference as one in time.

An EQ cannot fix a problem in Time, which was causing a response problem (a reflection for example usually seen as a dip at some frequency). An attempt to correct a time problem WILL put a large amount of phase shift where there wasn’t any needed.

While one can improve the look of a response curve, adding a lot of phase shift may also have a negative effect on the sound quality.

Bottom line, so far as I see it, under the right circumstances a filter correction or EQ can be totally seamless and transparent when used to correct an inverse problem at the source.

On the other hand, when used to compensate a time related interference, then EQ can make a flatter response but is also more likely to screw up the sound by adding phase shifts where they don’t belong.

Cheers,
TD

Reply posted by Mikael Holm on December 11, 2001
- “How do you correct for this problem outside of your EQ?”

By stacking more boxes.

Miffe

Reply posted by Mike G on December 11, 2001
This is more a system question then a filter question. If you have a 180-degrees phase shift at 45hz, what do you do about it?

If you are 40 degrees out between 130-160 how do you correct for this. It is correctable, at least to a certain extent. I don’t want to do this with my EQ. How else is this accomplished?

Mike G. 

Reply posted by Mikael Holm on December 11, 2001
I’d design FIR-filter for that particular box.

Miffe

Reply posted by Phill Graham on December 11, 2001
Hey Miffe,

While this is a fine suggestion, it’s not very practical for live work. The time involved for the FIR filter is approximately the same as the length of sample to resolve that frequency with a fourier transform. This is a substantial period!

Posted by Mikael Holm on December 11, 2001
I’m dealing with a subprocessor here. 35 seconds propagation delay

During my first year in university, I was told I should use recursive algorithms whenever it’s possible because it’s more powerful way to do it. Well, this recursive filter just kills me. Longest sample I run through taps is ~48ms.

Miffe

Reply posted by Mikael Holm on December 10, 2001
With FIR filters, you can make EQ filters with close to zero phase shift. Ideal filters are also possible but then you would have generated an oscillator

Miffe

Reply posted by Mike G. on December 11, 2001
Where can I find out more about FIR filters?

Mike G.

Posted by Andy Peters on December 11, 2001
I can’t think of anything “simple.” All of the books I have assume some math (calculus) background. There’s no book, a-la “Digital Filters for Dummies.”

However, Digital Filters
http://www1.fatbrain.com/asp/bookinfo/bookinfo.asp?theisbn=048665088X&vm=
by R. W. Hamming, is excellent. This is a small book, and it’s only $11. However, it’s eminently readable. Hamming eschews mathematic rigor (and he’s got a funny comment about that: “mathematical rigor...all too often leads to rigor mortis.") for explanations that make sense.

[As you read about digital signal processing, you’ll come across the important concept of windowing. There’s a handful of “standard” windows; one is the Hamming window.]

In addition to Proakis, the other canonical DSP book is Digital Signal Processing
http://www1.fatbrain.com/asp/bookinfo/bookinfo.asp?theisbn=0132146355&vm=
by Oppenheim and Schafer. It’s expensive ($100) but covers the whole subject in clear detail.

I also have an excellent book at work that’s got exercises in DSP with Matlab, but I can’t seem to find it on Fatbrain.

This may be interesting.
http://www1.fatbrain.com/asp/bookinfo/bookinfo.asp?theisbn=0130909998&vm=

andy

Reply posted by Phill Graham on December 11, 2001 at 19:29:48:
Andy,

While I was interning at AMD, I started on a book called “DSP using Matlab” in the AMD reference library. I didn’t come close to finishing it, but it looked like it was going a promising direction.

Phill Graham

Reply posted by Mikael Holm on December 11, 2001
Chapters 14 through 21 (Introduction to Digital Filters...Filter Comparison)
http://www.analog.com/technology/dsp/training/materials/dsp_book_index.html
touch the art of making different digital filters.

Also “Digital Signal Processing” by Proakis and Manolakis gets a recommendation from me. Andy Peters has some other books in his mind also.

Miffe

Reply posted by Tom Young on December 10, 2001
But why is it we see virtually no FIR processors ? (This is not bait)

Tom Young
Electroacoustic Design Services

Reply posted by Andy Peters on December 11, 2001
Good question.

Too much delay at low frequencies to be useful in real-time audio systems?

Reply posted by Mikael Holm on December 10, 2001
Probably because they are more or less purpose designed for special systems. HK Audio and German Audio Engineering first come to my mind. If you have read the DFC paper by HK Audio you’d know that after each, even slight, adjustment you have to compile the filter again causing few seconds delay before new filter can be loaded.

It’s not impossible but to correct phase you have to know extremely well the signal chain after controller. For example only few manufacturers give the phase response specs for their amplifiers. Usually those are within’ +/-25% but in some cases, like LAB4000, it’s +/-5% if i remember it right.

Miffe

Reply posted by Ron Riedel on December 11, 2001
FIR filters have phase shift. However, if they are properly designed, the phase shift is linear with frequency, and thus translates into a pure time delay. This has several ramifications:

1) The phase shift doesn’t “color” the sound. A time delayed signal sounds just like the original, just later.

2) An FIR filter is not a minimum phase system. Thus, it will not properly correct for minimum phase frequency response variations in a system component (a speaker or microphone, for example). You can correct the amplitude response of a speaker with a linear phase FIR filter, but the phase response will still be messed up. Conversely, a good old analog EQ can and will fix both the phase and amplitude response at the same time, as has been pointed out below.

3) FIR filters generate serious amounts of delay. A 1000 tap filter sampling at 44Khz will insert about 23 mSec of delay, which is equivalent to moving your speakers almost 20 feet! In many live sound situations, this kind of delay is totally unacceptable. You can reduce the delay by reducing the number of taps, but then the performance of the filter suffers, especially at low frequencies.

FIR filters find their greatest use in the digital-analog conversion process in digital audio playback. There the delay is irrelevant, and the linear phase response of the low-pass FIR filter is a decided advantage in sonic quality. For live sound use, though, you have to think of them first as primarily delay lines, that happen to have non-flat or selectable frequency response. Sort of a special purpose tool, requiring sophisticated and intelligent application.

Regards,
Ron Riedel

Reply posted by Mikael Holm on December 11, 2001
- “A 1000 tap filter sampling at 44Khz will insert about 23 mSec of delay, which is equivalent to moving your speakers almost 20 feet!”

Many vented boxes exhibit longer group delays at the tuning frequency. Actually that’s what i’m correcting with own lil’ box. It just needs so much time to work it out

Miffe