Forssell's EQ paper - some questions

[JoshuaUnitt]

first off,  I'm a technician- not an engineer (and I don't endeavour to be one! ). So I'm learning and open-minded, and feel free to talk to me like I'm 5.

Basically I've been commissioned to come up with a preamp/EQ. I've got a pretty good grasp on the preamp, but I'm not so confident in my EQ design abilities so I consulted Fred Forssell's White Paper. A very challenging, very fun read, if I may say! I would like to use his one-amp, gyrator-based approach in a 4-band fully parametric EQ, but I do have a (probably dumb) question about his circuit that I'm having trouble resolving in my head.

First off, here's the basic circuit-

That's well and good, but I'm having trouble understanding the actual actuation of boost and cut. What I gleaned from his paper was that if your filter network exists as part of the shunt from the negative feedback loop to ground, then you get boost functionality; and that if your filter network exists in the shunt from the non-inverting input of your opamp to ground, you'll get cut functionality. This circuit seems to be connected in both fashions, yet the paper seems to speak of the circuit having discrete "cut" and "boost" modes, implying a switch? Which I don't see. Thoughts? Help?

here's a link to the paper in case anyone needs a quick reference- http://www.forsselltech.com/media/attachments/Evolution_of_an_EQ_Design2.pdf

 

[Piedwagtail]

I built it with switches for different frequency cap/inductor settings.
Makes a great effect eq.
Main drawback I found was having the cut/boost pot in this position is very prone to noise.
Fine for set and forget, but don't try changing it during a mix.

I get the impression that he didn't take it further because of that.
Two vactrols could work to alleviate it, but it's research and development and time.

 

[abbey road d enfer]

Basically I've been commissioned to come up with a preamp/EQ. I've got a pretty good grasp on the preamp, but I'm not so confident in my EQ design abilities so I consulted Fred Forssell's White Paper. A very challenging, very fun read, if I may say! I would like to use his one-amp, gyrator-based approach in a 4-band fully parametric EQ,

This structure does not lend itself to fully parametric. First, the gyrator needs varying two resistors that are very different in value, so you would need pots like 100k revlog + 1k revlog that is definitely not standard and very difficult to source even as a custom part.  Then making variable Q would involve an additional stage with a dual pot. Ther are good reasons why almost every parametric EQ uses a variation or another of the state-variable filter. Very few use a variation of the Wien bridge, implying a more complex implementation.
Then the one-stage approach suffers from interaction between bands, which is not a problem when these bands are fixed AND properly spaced. But with fully parametric, there are very often cases where the bands are close, producing "interactive" results.

That's well and good, but I'm having trouble understanding the actual actuation of boost and cut. What I gleaned from his paper was that if your filter network exists as part of the shunt from the negative feedback loop to ground, then you get boost functionality; and that if your filter network exists in the shunt from the non-inverting input of your opamp to ground, you'll get cut functionality. This circuit seems to be connected in both fashions, yet the paper seems to speak of the circuit having discrete "cut" and "boost" modes, implying a switch? Which I don't see. Thoughts? Help?

I think this is for the sake of explaining the operation. At each end of the travel, the pot acts almost like a switch, which helps understanding what happens. In practice, any intermediate position is the result of both modes interacting; in particular, the "neutral" position is the result of an identical amount of boost and cut, resulting in flat response. 

here's a link to the paper in case anyone needs a quick reference- http://www.forsselltech.com/media/attachments/Evolution_of_an_EQ_Design2.pdf

  I'll read the paper.

 

[JoshuaUnitt]

Appreciate the help so far, thanks!!

This structure does not lend itself to fully parametric. First, the gyrator needs varying two resistors that are very different in value, so you would need pots like 100k revlog + 1k revlog that is definitely not standard and very difficult to source even as a custom part.  Then making variable Q would involve an additional stage with a dual pot. Ther are good reasons why almost every parametric EQ uses a variation or another of the state-variable filter. Very few use a variation of the Wien bridge, implying a more complex implementation.
Then the one-stage approach suffers from interaction between bands, which is not a problem when these bands are fixed AND properly spaced. But with fully parametric, there are very often cases where the bands are close, producing "interactive" results.

This definitely makes sense, and perhaps would explain why stuff like the API 553 had fixed bands.
As I understand it, state-variable filter implementation relies on a bit more differential equation knowledge than I've got in this little head. I'll have to do some more research of course.

What's interesting is that it seems like from Forssell's gyrator setup, he doesn't even make accommodations for a variable-Q- only stepped Q (unless you use variable caps? ew.). In all honesty, I'm not opposed to using, say, 3 switchable Q modes even though at that point we aren't technically looking at full-blown parametric.

If I may- how does the one-amp topology relate to a design like the 550A which, if I'm not mistaken, uses buffers between bands to stop interaction? Could any of that methodology be ported over to this?

I think this is for the sake of explaining the operation. At each end of the travel, the pot acts almost like a switch, which helps understanding what happens. In practice, any intermediate position is the result of both modes interacting; in particular, the "neutral" position is the result of an identical amount of boost and cut, resulting in flat response.

That answers my question perfectly, thank you!!

 

[AusTex64]

Is there a reason why the EQ has to be fully parametric? If we look around at some of the most successful combos (1073, API 312 + 550/560) the EQ was limited to what the experienced designer decided would "sound good". As an example, I bought a pair of the Rupert Neve 5015 mic pre/dynamics modules and a pair of the 5033 parametric EQ's, thinking I required the maximum level of control and flexibility with my EQ. Interestingly, I find myself only using the high and low shelving most of the time, rarely messing with the midbands. Now, it would be fair to argue that I choose mics and placements pretty carefully to avoid those kinds of problems later. But I think there are good reasons that arguably the most popular mic pre/EQ combo, Neve 1073, has high and low shelving and a mid band control. I am really starting to appreciate the power in that simple approach. Another benefit is the Neve EQ is passive with a makeup amp. Since using Neve and Pultec EQ's on a regular basis, it's hard to go back to active designs. The hype with passive is real - it's not a subtle difference, they sound WAY smoother and musical to my ears.

But if you insist on fully parametric, nothing I've ever used sonically surpassed the Sontec/Massenburg parametrics. There's a very good reason Sontecs are the industry standard parametric EQ in mastering houses.

 

[abbey road d enfer]

Is there a reason why the EQ has to be fully parametric? If we look around at some of the most successful combos (1073, API 312 + 550/560) the EQ was limited to what the experienced designer decided would "sound good". As an example, I bought a pair of the Rupert Neve 5015 mic pre/dynamics modules and a pair of the 5033 parametric EQ's, thinking I required the maximum level of control and flexibility with my EQ. Interestingly, I find myself only using the high and low shelving most of the time, rarely messing with the midbands. Now, it would be fair to argue that I choose mics and placements pretty carefully to avoid those kinds of problems later. But I think there are good reasons that arguably the most popular mic pre/EQ combo, Neve 1073, has high and low shelving and a mid band control. I am really starting to appreciate the power in that simple approach.

I respect your opinion, but I can't help noting that the inventor of the parametric EQ, George Massenburg, is known for being fastidious with mic choice and placement. So there must be a good reason for using parametric EQ's. I use parametrics on snare and toms because I hate the resonances drummers love. Since I don't want to interfere with drummer's playing, I let them tune their drums to their liking, but I need to doctor their tracks. In a perfect world, where drummers have sound engineer ears (and brains), I would use a preamp without EQ.  8)

Another benefit is the Neve EQ is passive with a makeup amp. Since using Neve and Pultec EQ's on a regular basis, it's hard to go back to active designs. The hype with passive is real - it's not a subtle difference, they sound WAY smoother and musical to my ears.

Again I would not fight your opinion, but I have found that in most cases, passive EQ's have broader curves. In fact, for a given amount of boost/cut and a given bandwidth, there should be no difference between passive or active designs, unless one is not minimum-phase (which is pretty difficult - or senseless - to achieve).

 

[JohnRoberts]

Fixed EQ are good if you guess right but if you aren't sure about how you want to voice your eq to satisfy all users, full parametric means never having to say you're sorry.  8)

Massenberg wrote an AES article about his classic EQ in '72 but some credit Flikinger with inventing it (He has a patent #3752928). here is a schematic for a Massenberg Parametric around the WWW that looked pretty respectable IMO especially considering it's age.

I saw the design equations for SVF sections in an ap note from some IC company selling their op amps in the mid '70s. My parametric EQ kit in '79 would have surely infringed on Flickinger if his patent was issued only 8 years earlier. Something useful to know.  ;D

JR

 

[abbey road d enfer]

Indeed, when I credited George M. as the inventor of the parametric EQ, he certainly didn't invent the bridged-T filter that is at the heart of it; the bridged-T has probably predated GM's birth by a few decades; he admittedly took it from a 1940's Bell Labs filter handbook.
Actually, GM is credited for the lay-out and functionalities AND creating the name "parametric equalization". When UREI had come with the Little Dipper, which was a tentative approach for a one band variable EQ, GM had been working for over two years on what became the 4 band unit that has become a de-facto standard for equalization for many users.
It is interesting to note that he didn't use the SVF structure, which was around at the time. I believe he was aware of the existence of the SVF, but probably the large number of opamps needed for its implementation may have been the determining factor. In comparison, the SVF requires 3 opamps when only one is needed for bridged-T. The motivation there is not performance, it's purely economical. GM had contempt for the monolithic IC's of the period so had to make his own DOA's.

This allows me veering into the Forsell paper, where he presents his choice of single opamp as an indisputably better solution than a string of series opamps.
The answer he gives is definitely tainted by his credo: "Fortunately for us minimalist designers, the answer is a definite yes! "
This bias leads him to deliberately ignore the practical consequences of his remark: "the only reason that I could see to use a topology using series connected EQ sections is to make their transfer functions additive.", followed by "It must be said however, that the EQ design presented here will not boost the signal by 24 dB if all 4 of the filter sections with set as described above. It will boost it by 6 dB". 
This is precisely where it differs from the essential approach of parametric equalization, where all parameters are utterly independant and the final result is the strict sum of all the dialed process.
The final nail in the coffin is "Many graphic equalizers with octave frequency centers use a single amplifier topology."
This last comment is valid only because the filters are spaced by one octave instead of being allowed to overlap.

The use of a parametric EQ is not only making broad adjustments, in fact it's very often making surgical treatments. I mentioned earlier my daily use of parametric filters (now most often digital) for killing nasty resonances in drums. That involves putting narrow dips (about 1/10 oct wide according to digital designers definition, about 1/4th octave according to mine - and to my ears) spaced by about one half octave. I just couldn't do that with a "single opamp" EQ, where the interaction between bands would fight against me.

Forsell presents his choices as based on undisputable truth, which in fact is the result of his basic choices as to what an EQ should do and how it should be used. I don't disagree with his options when it comes to producing broad and smooth alterations to the frequency response, as in mastering or audio sweetening, but there is room and use for other types of EQ's that allow more dramatic operations.

The minimalist approach is right only if the number of components is enough for the job. If proper achievement of the goals requires a string of 5 opamps, I will use them without blinking.
The defects of opamps do not combine linearly: noise combine quadratically, distortion may in fact cancel partially - depending on structure, slew-rate-related artefacts happen just once.
The performance of most audio opamps (including the venerable TL0's and 5532/4) is perfectly compatible with stringing a dozen of them without any seriously objectionable effects - indeed, provided their implementation is correct.
Just think about the signal path for a single recorded signal in analog: mic pre, HPF, fader amp, output stage, recorder's input stage, 2 or 3 more before hitting tape, one or two in the repro amp, the output stage, then goes to line input, fader amp again, summing amp, another fader amp and output stage; that's about 15 stages before hitting your ears - I haven't counted the power amp, which is probably responsible for as much as the rest in terms of distortion, except tape and loudspeakers. How noticeable will be another 4 or 5 opamps in the EQ, compared to one?

 

[JohnRoberts]

Oops I was thinking of the Barry Porter Parametric that used SVF.

FWIW I was aware of any of these until much later... My first exposure to parametric was in commercial recording products (Ashley, et al).

JR

 

[AusTex64]

This pretty much tells the story of who invented parametric EQ:

http://www.massenburg.com/wp-content/uploads/2011/12/GM_ParaEQ.pdf

Certainly Burgess Macneal was heavily involved:

http://euphonicmasters.com/misc/BURGESS.pdf

I've had the opportunity to discuss this topic with both Burgess and Massenburg, and both were pretty quick to credit the other.

FWIW, I'm not anti parametric EQ. Hell, I sold most of the Sontec EQ's in Hollywood in the late '80's. They are great tools, in the right hands. But they also provide ample opportunity to badly screw things up in the wrong hands, and for most tracking applications are over complicated.

Regarding passive vs. active EQ - while in theory they should sound the same, in my experience they don't. The most widely used room tuning EQ before  the Meyer was the White 4400, which was cut only. My Neve 5033's are pretty darned nice active EQ's, but my Pultec clones sound better boosting high frequencies. Much smoother. This is my experience though, YMMV.

And this quote is pretty much the reason people started using  short signal paths directly to the recorder:

<Just think about the signal path for a single recorded signal in analog: mic pre, HPF, fader amp, output stage, recorder's input stage, 2 or 3 more before hitting tape, one or two in the repro amp, the output stage, then goes to line input, fader amp again, summing amp, another fader amp and output stage; that's about 15 stages before hitting your ears - I haven't counted the power amp, which is probably responsible for as much as the rest in terms of distortion, except tape and loudspeakers. How noticeable will be another 4 or 5 opamps in the EQ, compared to one?>

John Hardy sums it up nicely with the diagram on page 8 of the M1 manual:

http://www.johnhardyco.com/pdf/M1M2M1p.pdf

 

[sahib]

And this quote is pretty much the reason people started using  short signal paths directly to the recorder:

Just think about the signal path for a single recorded signal in analog: mic pre, HPF, fader amp, output stage, recorder's input stage, 2 or 3 more before hitting tape, one or two in the repro amp, the output stage, then goes to line input, fader amp again, summing amp, another fader amp and output stage; that's about 15 stages before hitting your ears - I haven't counted the power amp, which is probably responsible for as much as the rest in terms of distortion, except tape and loudspeakers. How noticeable will be another 4 or 5 opamps in the EQ, compared to one?

John Hardy sums it up nicely with the diagram on page 8 of the M1 manual:

http://www.johnhardyco.com/pdf/M1M2M1p.pdf

All those stages Abbey mentions serve specific operational requirements.

Your argument will hold when it will be possible to plug the guitar jack straight into the tape itself, by-passing everything in the way.

 

[AusTex64]

And this quote is pretty much the reason people started using  short signal paths directly to the recorder:

Just think about the signal path for a single recorded signal in analog: mic pre, HPF, fader amp, output stage, recorder's input stage, 2 or 3 more before hitting tape, one or two in the repro amp, the output stage, then goes to line input, fader amp again, summing amp, another fader amp and output stage; that's about 15 stages before hitting your ears - I haven't counted the power amp, which is probably responsible for as much as the rest in terms of distortion, except tape and loudspeakers. H ow noticeable will be another 4 or 5 opamps in the EQ, compared to one?

John Hardy sums it up nicely with the diagram on page 8 of the M1 manual:

http://www.johnhardyco.com/pdf/M1M2M1p.pdf

All those stages Abbey mentions serve specific operational requirements.

Your argument will hold when it will be possible to plug the guitar jack straight into the tape itself, by-passing everything in the way.

Huh? I've been using outboard mic pres patched directly to the recorder, sometimes with compression, EQ, de-essing, etc inline since 1985, monitoring back through the console and avoiding fader buffers, summing amps, etc whenever possible. Pretty sure I'm not the only one nor the first either. My argument holds up just fine, thank you.

 

[Gold]

I've been looking at that Forssell paper for years. I am in need of a shelving EQ. I thought I might give this a try.

I want a four band with two high bands and two low bands. It would be used like you would use a tilt EQ. Both the low bands would be spaced far apart and both the high bands would be spaced far apart.

One of the two low bands would be say 40Hz-100Hz,  the other would be around a tilt frequency like 500Hz-900Hz.

The high bands would be something like 1k-3k and 10k-16k.

Would there be interaction between the low shelf and high shelf bands that are spaced fairly close together (500Hz-900Hz and 1kHz -3kHz)?

 

[ruairioflaherty]

Paul,

A few questions..

- Is your new console balanced or unbalanced?
- Do you need variable frequencies on each shelf or is it a fixed frequency for each of the four shelves?
- What kind of gain range do you need?

Cheers,
Ruairi

 

[abbey road d enfer]

I've been looking at that Forssell paper for years. I am in need of a shelving EQ. I thought I might give this a try.

I want a four band with two high bands and two low bands. It would be used like you would use a tilt EQ.

I'm not sure we have the same definition for a tilt EQ; for me it boosts the highs and cuts the lows (or the contrary) in a progressive way.  I have doubts  how two of them can really coexist. You probably have another definition, that I would like to know before making any suggestion.

Both the low bands would be spaced far apart and both the high bands would be spaced far apart.

One of the two low bands would be say 40Hz-100Hz,  the other would be around a tilt frequency like 500Hz-900Hz.

The high bands would be something like 1k-3k and 10k-16k.

Would there be interaction between the low shelf and high shelf bands that are spaced fairly close together (500Hz-900Hz and 1kHz -3kHz)?
[/quote] Not much; but it is really structure dependant. Although it doesn't adhere to the minimalist's principles - which are IMO an anthropomorphic way of undertaking science and techniques - , I prefer having fully independant sections, so the results are predictable.

 

[Gold]

- Is your new console balanced or unbalanced?

Balanced. Everything that goes to the monitor drives both legs with 47R buildout resistors. Outboard gear could be any arrangement.

- Do you need variable frequencies on each shelf or is it a fixed frequency for each of the four shelves?

Variable.  Just a few choices would be fine though.

- What kind of gain range do you need?

The other EQ's are +/- 3dB. That would be fine for this too.

I have a bunch of inductors so I think I could put one together without too much trouble. I could use input and output transformers.
I just got some Tamura line input transformers from a forum member to match the output transformers I have. The output transformers are very nice. I'm assuming the input transformers are as good.

 

[Gold]

I'm not sure we have the same definition for a tilt EQ; for me it boosts the highs and cuts the lows (or the contrary) in a progressive way. 

I think we do. I just was not very clear.  I meant I would use it like one would use a tilt EQ. This would allow separate access to boosting or cutting highs or lows. When I do this I often want to cut extreme lows or highs. The lowest band and the highest band would be used almost exclusively for cut. The way I work the bands around the tilt frequencies would be used mostly for boost.

The other two EQ's are a T filter EQ and an SVF. This is a stunt EQ.

\

 

[sahib]

All those stages Abbey mentions serve specific operational requirements.

Your argument will hold when it will be possible to plug the guitar jack straight into the tape itself, by-passing everything in the way.

Huh? I've been using outboard mic pres patched directly to the recorder, sometimes with compression, EQ, de-essing, etc inline since 1985, monitoring back through the console and avoiding fader buffers, summing amps, etc whenever possible. Pretty sure I'm not the only one nor the first either. My argument holds up just fine, thank you.

You are most welcome.

And that is my point of the weakness of your argument.

Why "sometimes" and not "all the time"?

Because it is required to serve the specific operational  need of that moment.

So, if your EQ design requires two extra op-amps to serve it then you'll have them. But disregarding those requirements in the name of "minimalism" and somehow presenting it as "superior" is not really a credible argument.

 

[abbey road d enfer]

I'm not sure we have the same definition for a tilt EQ; for me it boosts the highs and cuts the lows (or the contrary) in a progressive way. 

I think we do. I just was not very clear.  I meant I would use it like one would use a tilt EQ. This would allow separate access to boosting or cutting highs or lows. When I do this I often want to cut extreme lows or highs. The lowest band and the highest band would be used almost exclusively for cut. The way I work the bands around the tilt frequencies would be used mostly for boost.

Ok, understood. If I'm not mistaken, you want a basic variable (not sweepable because you want to use you available inductors). Then the "Forsell" structure is adequate because you don't want the high and low bands to overlap. In view of the application (mastering), you don't want extreme Q's. That means that there may be some overlap but very minor.

The other two EQ's are a T filter EQ and an SVF. This is a stunt EQ.

You lost me again.  What other two? Do you mean you already have decided using a bridged-T and an SVF? If that's the case, I wouldn't want to use a combo of three different types of filters on a single opamp string. It may work, but would probably meet conflicting requirements. What is a stunt EQ?

 

[Gold]

You lost me again.  What other two?

What is a stunt EQ?

This would be part of a mastering console. There are two other EQ units  in the console. One is a four band T filter EQ and the other is a four band SVF. Both are built with parametric bands only. I need a broad shelving EQ.

By stunt EQ I meant it only needs to perform a specific narrow function.