How do I sum or combine 4 filters into a coherent EQ?

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

loss1234

Well-known member
Joined
Aug 28, 2009
Messages
48
I am working on a home-made/designed EQ. I have decided on the filters I want. The filters are Not the issue.

Connecting them in some logical way is the problem. The type of EQ I am closest too in filter selection would be a parametric with a strange filter for the top end. and a low cut.

I have studied many schematics but they vary quite a bit and without owning many EQS I am not sure which style of connection is the best.

I've looked a lot at Orban, Calrec, Moog, Harrison, API, Neve, and others , focusing most on schematics which shed
light on opamp based designs without inductors

Surprisingly, with all the filter books I have (and books on pro audio too) not a single one of them covers the process of
integrating Low, Low Mid, Hi Mid and High into some kind of summing structure. They usually just show you what Low Mid and a high mid, and a low shelf look like individually. Maybe I am overthinking this and all it takes is summing the different filters outputs into one opamp like you would do designing a simple mixer. But I would be (happily) surprised if that was all it took.

So -questions

1. One setup I have looked at seems to involve a balanced in which then sends a feed to the first filters gain pot. the wiper of this pot enters the LOW sections filter input. one side of the pot then attaches to the next Gain pot (this goes to Low Mid) etc etc

the OUTS of all the EQS then connect to the Inverting input of staggered opamps (inverting in) so Low goes to one opamp, this feeds into the next opamp which then has LOW MID plug into its inverting input, etc


I assume you always  want to start with the lower bands feeding into the higher?

2. but are there other ways? I know in graphic eqs some designs use just one opamp and have all the filters connect to its summing point. But is this problematic? I am wondering if there are other ways. Self talks about a RTF setup where your curves always go back to flat outside of the range you are interested in. But he only shows how to do this for a LOW, HIGH setup (more of a tone control setup than a parametric eq----but maybe there is no difference?)

Id also love to find out more about fixed Q options or gain dependent Q. (I have read the Api Q changes as you increase amplitude)

But once again, I can't find any info on these except in schematics. I am torn between trying to design an EQ that is just
a typical parametric, or one that uses more of a API style of pre-chosen frequency points. I have never owned an API eq but I have tried them and of course I have used the plugins (not the same I know but I like having the choices narrowed)

One problem is determining the best CHOICE points. And of course this adds the cost of switches, etc. Unless of course I just use pots that are set to a much smaller range of frequency choices....
sorry thinking out loud here

I know this is asking many things.

any info appreciated

also if anyone knows of ANY books which cover this (Self comes closest in Small signal audio) in detail, I sure would be happy. Or an article online.
I have all the typical filter books. Lancaster, Self, Berlin, Valkenburg, Jungs Audio book, etc.

but maybe there are great articles online I need to read

THANKS SO MUCH FOR ANY HELP
 
OK

when you say in series, do you recommend that each filter out then goes into a separate  opamp in the series?
So wed have LOW return to opamp 1, which would connect to opamp 2 and the other side of opamp 2 would get the return from LOW mid, etc etc


I was looking a the Rane schematics (its nice that they post them all) and their PE-15 looks to be using just two opamps for the whole setup. One opamp SENDS to all of the filters, and then there is ONE summing opamp that they all return to.

this saves a few opamps but Id be interested in knowing if its a good method to pursue

THANKS!
 
loss1234 said:
OK

when you say in series, do you recommend that each filter out then goes into a separate  opamp in the series?
So wed have LOW return to opamp 1, which would connect to opamp 2 and the other side of opamp 2 would get the return from LOW mid, etc etc


I was looking a the Rane schematics (its nice that they post them all) and their PE-15 looks to be using just two opamps for the whole setup. One opamp SENDS to all of the filters, and then there is ONE summing opamp that they all return to.

this saves a few opamps but Id be interested in knowing if its a good method to pursue

THANKS!

Good design covers all bases effectively... less opamps in the path is generally a good thing unless the opamps are doing something important like buffering impedance, referencing differential signals, correcting polarity, etc.

I do not feel like designing an EQ today, been there done that, so you need to figure out how few you can get away with.

In general for a one-off DIY project it isn't worth killing a lot of brain cells to eliminate a few opamp stages,  for a 36 input console, every unneeded opamp matters when multiplied by 36x.

Good Luck.

JR
 
great

i am starting to get the impression that maybe there isn't much written about this aspect of Equalizers because it isnt really
that complicated ONCE you actually start connecting the stages.

So my plan is just to attack the breadboard and see what sounds best


thanks

 
I've looked a lot at Orban, Calrec, Moog, Harrison, API, Neve, and others

I assume you always want to start with the lower bands feeding into the higher?
If you check again some of the schematics mentioned above you'll see that it's not always the case...


 
> Blonder-Tongue Audio/Baton

If your filter is mostly-flat, flat-passes everything it isn't supposed to bend, then you just run output to input.

A special-case: many popular EQs have an explicit "everything" path, then use band-passes to add/subtract from there.

The BT AB filters are band-pass, they do NOT pass anything except the band they are tuned for. The idea is that with so many band-passes, summed-together equally, you end up with "everything" (and not-equal, a bent-up/down "everything").

Filter-order: first knock-down anything which gives trouble. If you have subsonic rumble, low-cut as soon as you amplify out of hiss, but before you get a nominal level which may distort subsonics into the audio band. Same for pure 60Hz hum. Last, knock-down any dirt the filters add, typically you high-cut hiss and distortion overtones.

You can do a LOT with a $69 BOSS/ROSS graphic EQ pedal. If only to verify the general shape you want and eye-ball the inflection frequencies (Boss's 1KHz too low, 2KHz too high, so your design bends at 1.4KHz). A $69 graf-EQ won't do steep slopes or infinite dips; but severe EQ-bendings are usually for technical problems (woofer slap, ADC alias) not musical flavoring.
 
Maybe you should check the schemes of the sontec, it uses an summingamp and active feed forward/feed back where the filters are working... Maybe it helps.

Could you say something more about your filters or some diagram to follow you better.

JS
 
I looked around to find a 4 band EQ that was clearly drawn so it is easy to understand, and found this:

http://groupdiy.com/index.php?action=dlattach;topic=45134.0;attach=8521

This is a "classic" approach, using a unity gain inverter for each band with the boost/cut pot "wrapped around" each inverter.  The wiper of each pot feeds the individual filter circuits.  In this case, the low and high EQ filters are Sallen-Key two pole low and high pass filters and the two mid bands are state variable filters (using the bandpass outputs of each).

The filtered outputs are then fed back into the opamp for that particular band.  I have seen a jillion variations of this basic concept, including some designs where the low and high EQ boost/cut functions were handled by a single opamp.

Bri
 
loss1234 said:
I am working on a home-made/designed EQ. I have decided on the filters I want. The filters are Not the issue.

Connecting them in some logical way is the problem. The type of EQ I am closest too in filter selection would be a parametric with a strange filter for the top end. and a low cut.

I have studied many schematics but they vary quite a bit and without owning many EQS I am not sure which style of connection is the best.

I've looked a lot at Orban, Calrec, Moog, Harrison, API, Neve, and others , focusing most on schematics which shed
light on opamp based designs without inductors

For what it's worth, the Rane parametrics (PE-15 and PE-17) have all of the filter sections in parallel. I used to own a pair of the PE-15s and I measured their filter responses and interactions and the results were what one would expect.

The schematics are available on the Rane web site.

-a
 
Andy Peters said:
loss1234 said:
I am working on a home-made/designed EQ. I have decided on the filters I want. The filters are Not the issue.

Connecting them in some logical way is the problem. The type of EQ I am closest too in filter selection would be a parametric with a strange filter for the top end. and a low cut.

I have studied many schematics but they vary quite a bit and without owning many EQS I am not sure which style of connection is the best.

I've looked a lot at Orban, Calrec, Moog, Harrison, API, Neve, and others , focusing most on schematics which shed
light on opamp based designs without inductors

For what it's worth, the Rane parametrics (PE-15 and PE-17) have all of the filter sections in parallel. I used to own a pair of the PE-15s and I measured their filter responses and interactions and the results were what one would expect.

The schematics are available on the Rane web site.

-a
That's kind of interesting... The topology Rane used where a bandpass filter feeds the wiper of a pot across two inverting inputs is more commonly used with fixed GEQ bandpass filters not parametric that can sweep and overlap. I notice they split the EQ sections up into two smaller groups in parallel. One with two and one with 3 bands. probably with adjacent bands alternating into the two different groups to reduce likely overlap.

FWIW this approach only uses one less opamp than putting them all in series, while 5 inversions would require one more to get back in polarity. I suspect Rane engineers had their own different reasons for using that topology.

JR
 
loss1234 said:
Surprisingly, with all the filter books I have (and books on pro audio too) not a single one of them covers the process of
integrating Low, Low Mid, Hi Mid and High into some kind of summing structure.
Most "serious" books are geared towards telecoms, broadcast, radar... Audio is often considered a minor genre.
Maybe I am overthinking this and all it takes is summing the different filters outputs into one opamp like you would do designing a simple mixer. But I would be (happily) surprised if that was all it took.
That's one possibility, but it has so many drawbacks it is seldom used. Most of the implementations of parallel EQ have severe flaws.
1. One setup I have looked at seems to involve a balanced in which then sends a feed to the first filters gain pot. the wiper of this pot enters the LOW sections filter input. one side of the pot then attaches to the next Gain pot (this goes to Low Mid) etc etc

the OUTS of all the EQS then connect to the Inverting input of staggered opamps (inverting in) so Low goes to one opamp, this feeds into the next opamp which then has LOW MID plug into its inverting input, etc
That's probably the most common implementation. You have to understand what each element does. The wiper of each pot feeds a filter, which output goes back to the inverting opamp, and that's the amount and polarity of the signal that governs the amount of boost or cut.
I assume you always  want to start with the lower bands feeding into the higher?
No. It doesn't really matter. Organisation is based on best lay-out of the PCB and position of the pots.
2. but are there other ways? I know in graphic eqs some designs use just one opamp and have all the filters connect to its summing point. But is this problematic? I am wondering if there are other ways.
There are many ways, but those with only one opamp suffer from interaction between bands. This is not an issue with graphic EQ's because the bands are unmovable and the interaction is well predicted and taken into account, but for parametrics, where there is a risk of having several bands on teh same frequency, this can bea problem. that's one of the issues with parallel EQ's.
Self talks about a RTF setup where your curves always go back to flat outside of the range you are interested in. But he only shows how to do this for a LOW, HIGH setup
This type of behaviour is necessary for bell-shape EQ, that's why he doesn't mention it. Using unity-asymptote filters for LF & HF tone controls is relatively unusual, particularly in the audiophile domain where Self exercises his preaching. In pro-audio, many channel EQ's offer the choice between bell and shelving curves.
Id also love to find out more about fixed Q options or gain dependent Q. (I have read the Api Q changes as you increase amplitude)
The very notion of Q for an equalizer is seriously debatable. Q is a notion that applies only to zero-asymptote filtersn not to equalizers that are unity-asymptotic. I condemn the use of Q and favour the use of bandwidth, which is more instinctive. Anyway, when you change the boos or cut, the BW changes, whatever the type of filter. The term "constant Q" has been coined by marketeers but has no operational benefit. It is true that these so-called "constant Q" EQ's react differently than the others ("variable Q"?) but no one could ever prove one or the other to be indiscutably "better".
But once again, I can't find any info on these except in schematics. I am torn between trying to design an EQ that is just a typical parametric, or one that uses more of a API style of pre-chosen frequency points.
Both type exist, fixed frequency and variable ones. They represent different choices and compromises between reproductibility, accuracy, versatility, cost, noise performance... you have to assess your own needs and make your choice.
also if anyone knows of ANY books which cover this (Self comes closest in Small signal audio) in detail, I sure would be happy. Or an article online.
You may want to have a look at Steve Dove's articles "designing a professinal mixing console", but really the best source is analysing all the schematics you can find. For example check the differences between a KT DN27 or Rane GE60 and a dbx series 20 graphic, and after that check the similitudes between the dbx graphic EQ and a Rane PE17 parametric.
 
abbey road d enfer said:
The very notion of Q for an equalizer is seriously debatable. Q is a notion that applies only to zero-asymptote filtersn not to equalizers that are unity-asymptotic. I condemn the use of Q and favour the use of bandwidth, which is more instinctive. Anyway, when you change the boos or cut, the BW changes, whatever the type of filter. The term "constant Q" has been coined by marketeers but has no operational benefit. It is true that these so-called "constant Q" EQ's react differently than the others ("variable Q"?) but no one could ever prove one or the other to be indiscutably "better".

It has (had) been a personal campaign of mine to get AES to establish one or more precise definitions for Q or bandwidth in the context of boost/cut EQs. This has become a serious issue with generic outboard DSP being applied to live sound reproduction, to execute published presets for multi-way speaker systems. While center frequency, corner frequency, boost/cut,  and Q/BW of most crossover dividing filters are well understood. The lack of industry standards for boost/cut EQ Q/BW makes generic published presets unreliable between sundry DSP platforms.

My personal preference is to define Q/BW for spot corrective EQ using the half power bandwidth of just the EQ's boost/cut contribution (i.e. subtract the EQ'd signal from flat, and measure half power of just that difference signal). As you note a physical EQ circuit can deliver a Q/BW that changes with amount of Boost/Cut. Right now we have a tower of babel, where end users need to try to figure out what platform the manufacturer's corrective EQ Q/BW preset advice was developed from. 

JR

PS: If we ever want to realize virtual consoles that run in generic (dsp) environments and model classic console EQ, we need an industry definition for Q/BW for boost/cut EQ sections.
 
JohnRoberts said:
The lack of industry standards for boost/cut EQ Q/BW makes generic published presets unreliable between sundry DSP platforms.
Absolutely. In my current line of work, I'm facing this daily. My customers wish/want to use their favorite "digital loudspeaker management system" and ask us for loudspeaker preset parameters. Having tested a number of available products, I can prove them that identical settings  produce different results. Indeed the differences are not outrageous but still unacceptable. 
My personal preference is to define Q/BW for spot corrective EQ using the half power bandwidth of just the EQ's boost/cut contribution (i.e. subtract the EQ'd signal from flat, and measure half power of just that difference signal).
  Altough certainly a method that provides undisputable results, I think it is a somewhat tedious operation that actually requests to have the unit ready for measurement. I favor a method that assesses the BW at 1/4 (in dB) the boost or cut. For example, for the usual 12dB boost I would take the +3dB points. I've found that with most designs, this assessed BW is constant enough over a range of 3 to 15dB boost or cut.
That way, it is not necessary to actually have the equipment, since the published graphs can be used.
What's more, it makes sense in terms of correlation between measurements and auditive perception.
 
Andy Peters said:
For what it's worth, the Rane parametrics (PE-15 and PE-17) have all of the filter sections in parallel.
Actually the PE17 uses a combination of series and parallels. One block with 3 filters and another with 2 filters. Rane, for some reason, seem to be the advocates of interactive filters. They actually made it a feature instead of a flaw by pontificating on "interpolating EQ", by showing that under appropriate circumstances, pushing two adjacent faders would create an intermediate curve, but failed to caution that pushing several adjacent faders (which no one should do, but almost everybody does, cf. the ubiquitous "smile" EQ) had adverse effects.
 
Brian Roth said:
I looked around to find a 4 band EQ that was clearly drawn so it is easy to understand, and found this:

http://groupdiy.com/index.php?action=dlattach;topic=45134.0;attach=8521

I've run into a few of these now where they don't specify pot taper for the different sections. Is it safe to assume it's usually linear for gain and reverse log on the frequency pot, or are there rules for different circuit arrangements? Also is rev.log on the frequency pot easily distinguishable from one of the others? I've got a circuit I'm doing now which takes 500k(C) and they only have regular log in the style I need.

Also, if anybody has any good examples of switched shelf/bell circuits, I didn't find much on that.

Hope loss1234 doesn't mind another EQ noob tacking some additional questions on his thread.  ;)

Thanks!
 
Jidis said:
Is it safe to assume it's usually linear for gain
You may find that commercial products use "S" taper pots, which provide slower variation at both ends of travel. These are generally not available to Joe Public, so you have to use linears.
and reverse log on the frequency pot, or are there rules for different circuit arrangements?
Some products use linear pots. They are not connected as variable resistors, but rather as real pots. One resistor goes from the wiper to the integrator, another from the top to the wiper. The respectuive values of the pot and the wiper resistor are so chosen that it alters the taper in a way that provides a remarkably smoothe control law.
Also is rev.log on the frequency pot easily distinguishable from one of the others?
Do you mean "does it make a difference if the pot is not RevLog? Yes, using a linear instead of RevLog cramps the upper frequency range in a very narrow angle.
I've got a circuit I'm doing now which takes 500k(C) and they only have regular log in the style I need.
I think you can get them from Audio Maintenance. If not you may just use different value and change the caps accordingly.
Also, if anybody has any good examples of switched shelf/bell circuits, I didn't find much on that.
These are a tad tricky. Check the dbx 905, Valley People Maxi-Q or SSL 82E02.
 
abbey road d enfer said:
JohnRoberts said:
The lack of industry standards for boost/cut EQ Q/BW makes generic published presets unreliable between sundry DSP platforms.
Absolutely. In my current line of work, I'm facing this daily. My customers wish/want to use their favorite "digital loudspeaker management system" and ask us for loudspeaker preset parameters. Having tested a number of available products, I can prove them that identical settings  produce different results. Indeed the differences are not outrageous but still unacceptable.
Yes there are more differences than just Q/BW definition, but I'm trying to keep on topic.
My personal preference is to define Q/BW for spot corrective EQ using the half power bandwidth of just the EQ's boost/cut contribution (i.e. subtract the EQ'd signal from flat, and measure half power of just that difference signal).
 
Altough certainly a method that provides undisputable results, I think it is a somewhat tedious operation that actually requests to have the unit ready for measurement. I favor a method that assesses the BW at 1/4 (in dB) the boost or cut. For example, for the usual 12dB boost I would take the +3dB points. I've found that with most designs, this assessed BW is constant enough over a range of 3 to 15dB boost or cut.
That way, it is not necessary to actually have the equipment, since the published graphs can be used.
What's more, it makes sense in terms of correlation between measurements and auditive perception.
It does not matter how we define it. One or more industry definitions that digital platforms can relatively easily translate between would work. Back in the bad old days we were locked into a handful of analog hardware driven control relationships, now we can make  whatever arbitrary control law we need, but someone must drive a stake into the ground. 

In the meanwhile a rosetta stone between different platforms could help. but not my job mon...

JR

PS see post #12 for more discussion of Rane's series parallel PEQ.
 
Back
Top