Multiband Compressor Idea

[alexc]

Just taking a look now  ...

The 'Compressor' section was as I described.

The 'Limiter' section also 3 band, has individual threshold, attack and release but the 'lim' setting  (ratio) is only switchable 'on-off' for each band.  There is an overall 'Digital Ceiling' setting.

Is this what you are meaning ?

 

[Gold]

Yes that’s how I remember MD2. I never had a finalizer.

The limiter sections are only practical to use as peak limiters. Not as a peak detecting compressor. I believe buried in the menu you can change the compressor sections from rms detection to peak detection.

 

[alexc]

And more generally, regarding multiband processing ..  on 'plug-ins'.

There's a lot of them out there, some better than others but certainly some very capable ones.

Are there some that do what are looking for ?

I've used a few and in terms of control, they are very comprehensive.

In terms of sound, I guess it's subjective, but I think a few work less obtrusively than the Finalizer  .... 

 

[Gold]

Are there some that do what are looking for ?

Not that I’ve seen.  I’ve never seen anything like this. The GML can do it but in a roundabout way.  It’s not multiband though.

 

[Gold]

The unique thing about this is more about ergonomics than anything else. It should have a very wide range of control with fairly simple controls.

Since it is five bands it should be possible to design in useful time constants per band for a wide range of program material. This also helps by making it harder to sound bad.

If all controls in each section are matched it should act mostly like a wide band compressor with a dual time constant. Bypass the rms section and it’s like an 1176. Bypass the peak section and it’s a classic VCA comp. that’s just with a bypass button and a threshold knob.

Then you can dig in with very fine control of dynamic behavior within each band. Even this has a fairly simple set of controls for the amount of possible control. Ehh?

 

[alexc]

Hope you do get to further develop your ideas in this area.  mb gain control is very interesting indeed.

......

As I understand it, at certain point, ones 'paradigm' shifts from 'several band-limited dynamics controllers'  to 'several bands of dynamic eqs' 

I think some of the more modern 'software realizations'  go from three to five or seven bands of 'dynamic equalisation'.

In a sense, I can 'comprehend' a little more clearly in this 'view'  ...  more like an specialised para eq  5+ filter bands,  and fully variable dynamic response.

I think some also do the whole 'transient shaping' thing too .. in a mb context. So there's that as well!

...

Certainly there are software makers doing 'aggregations of mastering blocks'  in conjunction with successful music producers, in particular, 'mastering guys'. They can be very useful.

In hardware ..  all the better!  If you can combine some analog qualities and maybe even digital qualities,  with new ways of doing things, then  fabulous.

Certainly the tech out there, both analog and digital, can be superb.

 

[gyraf]

So, to condense the idea: It's about having a multiband compressor with dual time constants and ratios for each band?

I like the idea of dual time constants (use it in e.g. my G24) but extending this to multiband would certainly be quite a forest of knobs to navigate.

I'm not sure I agree that e.g. timings can be reduced to single controls, as much of the musicality of compression is about accurately timing both grab and let-go, both when and how-much. attack/release/threshold/ratio.

Don't let the easy user interface of MD2/Finalizer trick you into believing that it is straightforward and simple under the hood: Kim Rishøj (the man) once described for me how they reduced some 95 original dynamics parameters into the five control knob groups available - they did this by recording large amounts of (qualified) user interaction on an all-parameter version, and then factor-analyzed (multivariate correlation) the groupings of interdependent setting until they were reduced to the simple parameters that could be brought out to front panel.

..this also means that whenever you think you change one parameter in the unit, in reality you shift the whole unit into a different "pattern of behavior" on a palette that was systematically derived from the original user's way of using it and put into 5-dimensionally space  - but still it feels very simple and intuitive in use (to the degree that people who really shouldn't, do anyway, but that's a different story).

Jakob E.

 

[Gold]

I have experience with the THAT VCA. I know how it behaves in simple RMS mode and in a peak detection mode. The Maselec uses 2181’s. I’m pretty sure what I’ve proposed will work and produce the results I expect with the controls I’ve suggested.

 

[JohnRoberts]

So, to condense the idea: It's about having a multiband compressor with dual time constants and ratios for each band?

I like the idea of dual time constants (use it in e.g. my G24) but extending this to multiband would certainly be quite a forest of knobs to navigate.

I'm not sure I agree that e.g. timings can be reduced to single controls, as much of the musicality of compression is about accurately timing both grab and let-go, both when and how-much. attack/release/threshold/ratio.

Easier to say than do...  (dbx had success with their "one-knob squeezer", mainly because it worked good enough.)

Years ago I speculated about how many parameters "could" be useful to control to customize a dynamic processor's behavior. At the time my goal was to mimmic other successful processors, provided as presets that could be recalled . Besides the dialed in factory presents, users would be allowed to use them as starting points to make their own custom processors.

For a hint about how long ago this was, I planned to save and recall the custom settings using FSK encoding and saving the data to the tape recorder printed to the leader between songs with the slate tones (for easy recall months later). This was wildly impractical and obviously never happened.

Decades later with digital medium capability this makes a little more sense but the interface is still awkward. It would need to be menu based to drill down several layers to reach every tweak, and would surely be beyond the capability of most end users to easily understand and fully exploit. This would also break my rule of never sounding bad, because that much control could easily sound horrible. Of course providing presets that don't suck would keep many users happy and occupied.

For todays technology the platform to do this is a digital software plug-in but why bother with a one stop solution if they can already mix and match from other existing classic processing mimics. 

Sorry about the veer...

JR 

Don't let the easy user interface of MD2/Finalizer trick you into believing that it is straightforward and simple under the hood: Kim Rishøj (the man) once described for me how they reduced some 95 original dynamics parameters into the five control knob groups available - they did this by recording large amounts of (qualified) user interaction on an all-parameter version, and then factor-analyzed (multivariate correlation) the groupings of interdependent setting until they were reduced to the simple parameters that could be brought out to front panel.

..this also means that whenever you think you change one parameter in the unit, in reality you shift the whole unit into a different "pattern of behavior" on a palette that was systematically derived from the original user's way of using it and put into 5-dimensionally space  - but still it feels very simple and intuitive in use (to the degree that people who really shouldn't, do anyway, but that's a different story).

Jakob E.

 

[Gold]

One feature that may not be obvious is that when using the rms and peak sections together it can act like dual detector but depending on the program it can act almost like a second compressor the further the threshold levels are from each other.

 

[JohnRoberts]

One feature that may not be obvious is that when using the rms and peak sections together it can act like dual detector but depending on the program it can act almost like a second compressor the further the threshold levels are from each other.

I have been chewing on this for a while trying to make sense of it.

By "peak" and "rms"  I suspect you are implying different nominal time constants for each. Peak presumably is fast attack, while RMS infers integration or smoothing for a slower attack. To normalize the amplitude between both suggests a 3 dB scale offset for same input signal.

FWIW when I was developing a microprocessor based console meter, I invested some time and effort into coding up a "true RMS" algorithm for the VU portion of my Peak/VU meter.  In production this would cost me zero, just the sunk cost of software NRE.  To prove to myself that this was worthwhile I coded up a comparison between a meter with straight average VU and my true RMS VU. To keep it fair I used the exact same attack and release time constants for both.

I am not sure what I expected but using a number of different program sources I saw zero difference between the simple average and more complex true RMS. I ended up removing it from my code as extra complexity with no benefit.

At the risk of editorializing I suspect some of the love heaped on rms for dynamic processing is related to the pervasive use of theTHAT corp detector (after dbx).  There are probably some unique characteristics related to integration time constants applied inside the log domain (but it is probably more complicated than that). IIRC Paul Buff (RIP) did some interesting tricks back in the 80s with side chain dynamics operating on the log domain control voltage. 

You can pretty much do anything you can imagine.. I would budget some bench time to explore these "differences" and anticipated benefits more rigorously.  I have had many promising concepts fail on the bench over the decades (perhaps a personal problem of mine ).

JR

PS: Back in the day DBX made some hyperbolic claims about the benefit of RMS detection in their tape noise reduction compandors. In my experience and judgement those claims were undeserved. 

 

[Gold]

By "peak" and "rms"  I suspect you are implying different nominal time constants for each. Peak presumably is fast attack, while RMS infers integration or smoothing for a slower attack. To normalize the amplitude between both suggests a 3 dB scale offset for same input signal.

Yes, I would expect the values to be different for each band. Determined by ear.

I am not sure what I expected but using a number of different program sources I saw zero difference between the simple average and more complex true RMS. I ended up removing it from my code as extra complexity with no benefit.

The off the shelf detector from THAT says RMS on it so might as well use it. I know how that detector behaves too. If this was a side chain design exercise it might not matter so much.

At the risk of editorializing I suspect some of the love heaped on rms for dynamic processing is related to the pervasive use of theTHAT corp detector (after dbx).  There are probably some unique characteristics related to integration time constants applied inside the log domain (but it is probably more complicated than that). IIRC Paul Buff (RIP) did some interesting tricks back in the 80s with side chain dynamics operating on the log domain control voltage.

Sure but I don't see what the point is. That's a big part of why what  I imagine will happen with this design  should be correct.  I have experience with the part over about a ten year period. Both using it in a commercial piece of gear like the Maselec and playing around with it when building the Pico compressors.

 

[JohnRoberts]

Yes, I would expect the values to be different for each band. Determined by ear.

You may be missing my point, or I may be reading more into your comments than intended (entirely possible).

The off the shelf detector from THAT says RMS on it so might as well use it. I know how that detector behaves too. If this was a side chain design exercise it might not matter so much.

It sounds like a side chain design exercise to me...

Sure but I don't see what the point is. That's a big part of why what  I imagine will happen with this design  should be correct.  I have experience with the part over about a ten year period. Both using it in a commercial piece of gear like the Maselec and playing around with it when building the Pico compressors.

I expect the dbx/THAT rms detector is ubiquitous in many (most) modern analog dynamics designs. These days I would just use the THAT chip working in that milieu too (we did at Peavey in the 90s), but you will still need to roll your own peak detector. IMO the benefit of using the THAT detector chip is #1 a decent FW rectifier (not trivial for low level signals), #2 a VCA friendly log conversion, and #3 if at all the RMS function which is a little sleight of hand relying upon the VCA log gain control input as the square root function.

To make a proper Peak detector you would need to roll your own but you could cheat another THAT detector chip to be peak-like by using a very tiny integration time capacitor. Thinking out loud you could probably extract peak information for a peak side chain threshold using an undersized cap, before following with smoothing circuitry (even with different att/rel times) for average/RMS after that to execute the second side chain threshold... one chip-two functions (being cheap I like that).  ;D
======
I published a proforma circuit on Wayne's forum for how to present a variable capacitance at that integration node, for adjustable time constants since attack/release time constants are controlled by the same cap in that chip, unless you hang extra circuitry off it. Sorry I never melted solder over my variable cap circuit, because I haven't messed with actual dynamics processor design this century.

Sorry if I am getting down into the weeds while you're taking a fly over perspective, but at some point the details will have to be worked out.

JR

 

[Gold]

To make a proper Peak detector you would need to roll your own but you could cheat another THAT detector chip to be peak-like by using a very tiny integration time capacitor.

That’s what I thought would be the solution. Maybe on the high bands it wouldn’t be fast enough but on the lower bands that will work well. I played around with that when building the Pico compressor.

I published a proforma circuit on Wayne's forum for how to present a variable capacitance at that integration node, for adjustable time constants since attack/release time constants are controlled by the same cap in that chip, unless you hang extra circuitry off it.

Continuously variable attack and release is not in this design. No knobs for it. That’s on purpose.

 

[abbey road d enfer]

To me, the big issue is not in the dynamics, it's in the band-splitting.
I designed a 4-band unit in the 80's (when private FM became legal) and stumbled on the cross-over design. 1st-order (6dB/octave) is not acceptable for most users, which also rules out conventional substractive filters with their asymmetrical slopes.
The accepted "standard" at the time was the BSS combo of their standard modular Linkwitz-Riley x-over MCS200, 4xDPR402 comp-lim and a summer (I guess it sold only in France), so I took on the fact that a 4-way LR x-over would be accepted by the market.
Measurements (and calculations - at the time I didn't have the benefits of a Spice simulator) quickly showed that interaction between bands result in two major issues;
First, obtaining flat response is impossible. I could achieve about +/- 0.7dB, which is acceptable for FM transmission - I doubt it would be for mastering 8).
Second, and more significantly, the narrower a band is the higher the interaction from the adjacent bands.  A gain reduction of 3dB in one band can actually result in a 6dB cut at the recombined output.
That puts a serious limit to the choice of cross-over frequencies.
A more promising techniques is the use of phase-compensated substractive cross-overs, where the output of a filter is substracted to an all-passed version of the input signal. Unfortunately this works only for a restricted number of filter alignments, which result in less sharp crossover points. In consideration of the intended application and the significant cost increase it was decided not to pursue.
Today I'm a daily user of the 4-way multi-band in Samplitude; it took me some time to set it up but now it takes me only a couple of minutes to adjust it to a particullar program.
I also use, but less often the Reaper multi-band, that I have configured as a 7-band, that correspond, to my ear, to a sensible separation between sub-low (<63), low (63-150), high low (150-300), low-mid (300-800), hi-mid (800-2k), low treble (2k-5k) and high (>5k). The use of such narrow bands is made possible by digital processing. Indeed, it is possible to make it sound absolutely rubbish. ;D
Now if 1st-order filters are acceptable and useful in mastering, forget all I've said...

 

[Gold]

Now if 1st-order filters are acceptable and useful in mastering, forget all I've said...

First order is acceptable to me. The Maselec uses first order slopes and It's a very useful box. The only setting on the MBC in Samplitude that I've ever been happy with is bypass. I've spent hours trying to get useful results out of that thing. I never could. Not to say that you don't.

 

[JohnRoberts]

First order is acceptable to me. The Maselec uses first order slopes and It's a very useful box. The only setting on the MBC in Samplitude that I've ever been happy with is bypass. I've spent hours trying to get useful results out of that thing. I never could. Not to say that you don't.

with the derived crossover the slope is only -6dB/oct for one direction, other cut off can be steeper and still sum mathematically (ignoring gain/level changes between passes)

JR

 

[abbey road d enfer]

with the derived crossover the slope is only -6dB/oct for one direction, other cut off can be steeper and still sum mathematically (ignoring gain/level changes between passes)

JR

Samplitude uses a choice of almost Linear-Phase filters with steep slopes on both sides. In the analog domain, substractive filters with phase-compensated substractors are not too bad.

 

[abbey road d enfer]

Gold, I have tried to answer your PM but your inbox is full.
Basically I said I'm interested in starting this project.

 

[Gold]

I was out of town and off grid for a couple of weeks. I missed this. Wow, super fantastic. I sent you an email.