XS902 De-esser - First Prototype!

[JohnRoberts]

Sorry I didn't mean to hijack this thread or turn it into a tutorial on side chain design but let me back up a little and fill in some gaps.

First, yes that looks like an huge improvement in driving the VCA with a cleaner and lower impedance drive. To maintain stereo linking you need to connect your sidechain/link node to the + input of the opamp so they will sum resistively. It's worth note that doing so will result in a 50:50 contribution from each channel. With some additional opamp tricks you could make both track the most de-essed of both channels but not tonight.

OK back to sidechains and opamps 101. The main issue is dealing with control voltages on the order of a 6 mV/dB with opamps so that the opamp's noise and dc offsets don't cause errors. The 5532 and TL074 are both old soldiers and old favorites of mine from a back in the day.

Door #1 TL074
++ high input impedance
++ low/no bias current due to FET input
++ fast (13 V/usec)
+ moderately low noise (approx 3 uV)
- wimpy drive capability (approx 2k)
-- poor DC offsets (as bad as 15mV for some grades)

Door #2 5532
++ low noise (approx .5 uV)
++ fast (10 V/usec)
++ good drive capability (approx 500 ohms)
++ good DC offset (perhaps < 1 mV)
- bipolar input impedance (lower than FET opamp)
- bipoloar input bias and offset current

Finally an observation about the THAT (dBx) RMS chip. I haven't looked inside to see if there's ways to game it, but it looks like it's designed to work in minimum part configuration comp/NR or whatever. Full scale voltage is on the order of 300 mV with limited drive capability (looks like 2K but data sheet describes out specific limits as some multiple of bias current).

With out mapping out a full proof, I see two alternate approaches using those part choices. To use TL074 you want to scale up the CV at least 10x to make bifet opamp noise and offset voltages insignificant. Alternately using the 5532 the noise and DC offset voltages are not problematic, but DC errors due to bipolar input bias and offset currents can introduce errors depending on resistor values and matching so play is just keep impedances low with stepping up.

Another consideration mitigating against scaling up the control voltage for bifet is the one time scaling of the DC also multiplies the offset voltage of that ifrst opamp stage, With only one offset trim for two RMS detectors trimming both is not judicious (in fact I don't like trims at all from years of design for production).

So that said and guessing you are perhaps looking for specific advice more than a professorial lecture, I would be inclined to stay with the 5532s and not scale up the 300mV.

Note: The roughly 3x in the initial design at that intermediate stage was not for DC accuracy but to provide the following stage with a useful compliance for the shunt FET to work into, and for stereo linking(?). Pushing it up 10x, and then down again in the very next stage buys little to nothing.

Back to specifics. Since the 2252 will drive 2k (I think I saw that in app notes) , using that lower impedance will reduce noise and bias current errors in 5532 unity gain inverter. Note: the resistor from + pin to ground is needed with 5532 and should be 1k (2k feedback in parallel with 2k input).

For the 5532 at the final sum/calculation amp the input bias correction is not clear cut due to diode steering and following configuration. Since the sources coming in are already corrected to zero, I would connect the + pin directly to 0V with no resistor and use the offset trimpot to correct for that, and all other errors.

For the feedback resistor at that stage something like 3x is OK but since that is not a round value with 5% resistors I'll leave that to you to figure out something specific in that range and following divider. With the bipolar opamps you might want to keep impedances of following divider a little lower, but this has to be traded off against shunt capability of JFET.

Correcting the bias current related DC errors at buffer 5532 is rather conflicted also since the DC impedance at the - input is now well defined (1k), but the DC impedance at the + input varies depending on the state of the FET shunt and whether diodes are forward or reverse biased in operation.

Again I'm inclined to just let the offset trim this out. but it can only be trimmed completely for one mode.

I hope this is making some sense and I'm not making this sound too mysterious. The old opamps required a little more math to get good results. The modern hot stuff opamps don't have bias current errors, and do have low offset and low noise... but they cost more $$ than the old soldiers.

Good luck
JR

 

[JohnRoberts]

After thinking about that CV buffer it just looked just too odd to be a chance accident so I'm forming an alternate theory... It's either poorly copied from something that worked or perhaps depopulated from a stereo unit for mono use only and just left hanging.

Going back to that original buffer topology, if we add a diode from output to - input in parallel with the 1k and 10 pf, it would act quite a bit differently in stereo link mode when two similar circuits are tied together. Now while the resistors might still fight each other, the diodes can only pull in one (the same direction) so the one commanding the most gain reduction would win and now establish the control voltage for both. Whichever channel was hottest would now seamlessly control both.

I would still add the 100 ohm and 1.5nF (especially now) to deliver low source impedance at HF but now the stereo link would work in a more reasonable fashion giving you all the de-essing you need for the worst case channel, and maintain a solid stereo image.

I much prefer this explanation of a copy or translation error from a decent design than a flawed original design, but I'm still pretty much guessing here.

JR

 

[Curtis]

OK, so if I'm understanding you correctly we're now looking at something like this

Summary: resistors on unity inverting buffer (HPF RMS'ed section) changed to R27=2K, R28=2K, R29=1K to lower input bias errors+noise in IC6a; non-inverting input of the rectifier section tied directly to ground; feedback R of rectifier back to 30K (3x amplification); voltage divider on output of FW rectifier back to 0.3x (R41, R42), CV signal taken from inverting input of last opamp stage (as was originally), diode D4 added in parallel with R43 and C13.

Side note, thinking out loud: that bypassing arrangement I've got isn't going to work (SW2) as it doesn't pull the inverting input of IC6b negative enough to turn on Q1.

 

[SSLtech]

I just realised something...

Take a look at the picture:

There's no provision for stereo coupling or linking 902s whatsoever.

Therefore, I'm of a mind to incline myself towards John's suspicion that this may have been a vestigial remnant from a stereo 'exploration'.

I'm therefore developing the notion that it could almost certainly be 'improved' as a result.

Keith

 

[JohnRoberts]

That looks better (in my mind). Of course it needs to be bread boarded to see if it actually works.

It seems a simpler bypass, would be to just put the mechanical switch in place of the FET shunt. That would prevent stereo link from working so perhaps switching a resistor from +15V into the 10k/30k opamp to swamp it off... In that case link mode would overpower the bypass so there is a judgement call regarding how you want it to work in link mode.

To bypass one and defeat link so other channel just works as if it wasn't linked might be accomplished with a switch pole in series with CV pin on VCA that toggles between ground and the opamp drive. That way bypass and link don't interact.

JR

 

[jdbakker]

[quote author="JohnRoberts"]After thinking about that CV buffer it just looked just too odd to be a chance accident so I'm forming an alternate theory... It's either poorly copied from something that worked or perhaps depopulated from a stereo unit for mono use only and just left hanging.[/quote]
I just had a look at the schematics for the 903 and 904, and they have a CV buffer/stereo link hookup that is very similar if not exactly the same.

FWIW, the 902 circuit was designed to drive a 202-class canned VCA; these have an internal CV buffer which relaxes CV drive specs considerably compared to the THAT218x chips.

JDB.

 

[JohnRoberts]

[quote author="jdbakker"][quote author="JohnRoberts"]After thinking about that CV buffer it just looked just too odd to be a chance accident so I'm forming an alternate theory... It's either poorly copied from something that worked or perhaps depopulated from a stereo unit for mono use only and just left hanging.[/quote]
I just had a look at the schematics for the 903 and 904, and they have a CV buffer/stereo link hookup that is very similar if not exactly the same.

FWIW, the 902 circuit was designed to drive a 202-class canned VCA; these have an internal CV buffer which relaxes CV drive specs considerably compared to the THAT218x chips.

JDB.[/quote]

By "same" I assume you mean VCA connected to the - input but without diode in feedback path.

In that case, when linked, the opamp trying to drive closest to 0V will have the slight advantage. In the resistor tug of war, with both opamps in complete opposition the result is 0V. This would favor the channel doing the "least" gain reduction and trying to command closer to 0V. I don't see how this locking to the least is preferable to the simple arithmetic average of the two, which at least gets you part way there to correcting the worst channel.

FWIW, if one were to appropriately scale the values of two opamp feedback resistors, you could make it so one channel always won the tug of war, which is another way to go, and not discernible from looking at only one channel schematic.

I personally preferred the simple arithmetic mean, but appreciate that users may find it confusing in use. Letting one channel arbitrarily always win, while only responding to the input from one channel also seems less than optimal.

So there are a number of options. I vote against the original appraoch with equal value resistors as there could be some squirrely states around zero if both opamps clip, without the additional shunt impedance at control port, and it seems the "least wins" is the worst tradeoff of the several options.

JR

PS: This all hinges on my visualization of how these opamps work in opposition. For example one opamp trying to command +2V would peg its output to +15v, while the other opamp only trying to command say +1V could overpower the common node and establish +1V while only swinging it's output down to -13V. These are not typical voltages but demonstrate the relationship. Closer to midway between the rails wins.

 

[JohnRoberts]

[quote author="jdbakker"][quote author="JohnRoberts"]After thinking about that CV buffer it just looked just too odd to be a chance accident so I'm forming an alternate theory... It's either poorly copied from something that worked or perhaps depopulated from a stereo unit for mono use only and just left hanging.[/quote]
I just had a look at the schematics for the 903 and 904, and they have a CV buffer/stereo link hookup that is very similar if not exactly the same.

FWIW, the 902 circuit was designed to drive a 202-class canned VCA; these have an internal CV buffer which relaxes CV drive specs considerably compared to the THAT218x chips.

JDB.[/quote]

The 202 AFAIK was several selected discrete resistors on a common heatsink. The control port was not buffered per se but resistively divided to scale for 20 dB/V with an input impedance of roughly 800 ohms. This would also make side chain processing less sensitive to small DC errors. The impedance seen by the transistors in the 202 is the short leg of this divider so probably <100 ohms and resistive, so no stability issue from that. Probably another small resistor internally to match impedance at other control port not being used (instead of grounding like in new VCAs).

The 1k resistor in the opamp feedback path would drive more than +/- 100dB control range, into a circa 202 VCA.

Still dual channels wouldn't play together nicely, in my estimation, based on what I have learned here about the schematics so far.

To literally interface newer VCA with older circa 202 designs, main tweak is to scale 50mV/dB down to 6 mV/dB, while I suspect old design didn't have 300 mV full scale RMS logger. either.

JR

PS: the actual behavior of this stereo linking should be easy enough to listen for on an actual (original design) unit. How the behavior changes between linked and not linked when one channel is commanding significantly more gain reduction than other would be instructive. The hot channel could be alternated for another data point.

I'd love to be wrong. I have lots of respect for Blackmer and crew, and I never looked very closely into their actual schematics. My speculation is based on second hand (to me) information.

 

[SSLtech]

AS I mentioned earlier, I have a pair of 902's here, a 900 rack, and plenty of test gear. -If you'd like any measurements taken to confirm or correct any assumptions, I'd be only too happy to do what I can.

-But there's no obvious way TO link them... I would have to strap them, but certianly there's no link switch to enable/disable a link... and the 900 rack doesn't link them de facto.

Keith

 

[JohnRoberts]

Thanx, but If there is no link capability, there is nothing to determine.

I'm back to a stereo formerly link capable design that was perhaps depopulated (?) for mono use.

Further if original design was for 202 VCAs the port drive impedance stability issue would be moot due to internal resistive divider in old VCA. Only drive consideration then was ability to drive 800 ohms to a few volts.

If this is not going to ever be linked for stereo it doesn't much matter how they play together in pairs.

Adding RC at control port for stability should adequately mitigate against any problems from unconventional drive topology.

Simply scaling down side chain to run at 6 mV/dB instead of 50mV/dB to mate with new generation VCA and RMS/rectifier/log chip suggests to me the need for some care with DC precision.

JR

 

[jdbakker]

[quote author="SSLtech"]-But there's no obvious way TO link them... I would have to strap them, but certianly there's no link switch to enable/disable a link... and the 900 rack doesn't link them de facto.[/quote]
Yup. I re-read the 902 manual, and nowhere do they mention the words 'stereo' or 'link'. What they do say is:

[quote author="The dbx902 manual"]The CV input and output connections come from the factory jumpered together for normal de-esser operation. To use the CV input, this jumper (R74) must be removed.
[...]
The CV input provides direct access to the control port of the dbx VCA of the 902 for gain control. This provides the user with the ability to use the 902 as an automatable level control. The dbx VCA provides linear-gain tracking in decibels over a range of -100 to +40 dB with 1% accuracy. An output of 0 V produces 0 dB of gain through the VCA. The control port should be driven by an op-amp or other low-impedance source.
[...]
The control voltage output is a DC voltage which changes in proportion to the amount of gain change being produced by the 902. The relationship between this voltage and the amount of gain change is expressed as CV=50mV/dB gain change. The CV output is low impedance and capable of driving bridged loads of more than 10k and less than .01µf without affecting the operation of the 902.[/quote]
So maybe the odd op-amp circuit was put in to (a) avoid damage to equipment if a user were to drive CVin without first removing R74 and (b) have the circuit still working as advertised with a moderate load on CVout?

JD 'just speculating here' B.

 

[JohnRoberts]

[quote author="jdbakker"][quote author="SSLtech"]-But there's no obvious way TO link them... I would have to strap them, but certianly there's no link switch to enable/disable a link... and the 900 rack doesn't link them de facto.[/quote]
Yup. I re-read the 902 manual, and nowhere do they mention the words 'stereo' or 'link'. What they do say is:

[quote author="The dbx902 manual"]The CV input and output connections come from the factory jumpered together for normal de-esser operation. To use the CV input, this jumper (R74) must be removed.
[...]
The CV input provides direct access to the control port of the dbx VCA of the 902 for gain control. This provides the user with the ability to use the 902 as an automatable level control. The dbx VCA provides linear-gain tracking in decibels over a range of -100 to +40 dB with 1% accuracy. An output of 0 V produces 0 dB of gain through the VCA. The control port should be driven by an op-amp or other low-impedance source.
[...]
The control voltage output is a DC voltage which changes in proportion to the amount of gain change being produced by the 902. The relationship between this voltage and the amount of gain change is expressed as CV=50mV/dB gain change. The CV output is low impedance and capable of driving bridged loads of more than 10k and less than .01µf without affecting the operation of the 902.[/quote]
So maybe the odd op-amp circuit was put in to (a) avoid damage to equipment if a user were to drive CVin without first removing R74 and (b) have the circuit still working as advertised with a moderate load on CVout?

JD 'just speculating here' B.[/quote]

Yes, that makes sense... The opamp topology uses the feedback resistor to current limit the opamp's ability to regulate that node. Driving it externally with a low impedance would just sink the few mA and ignore the internal opamp.

IMO not much utility for the modern design, with already explored downsides.

Thank You

JR

edit: my apologies for my increasingly apparent wrong guess that the unusual topology was for stereo linking, "and" flawed. External VCA control was even further down my short list of guesses.

At least the design now makes sense as drawn and is not some ham fisted mistake. For reasons already beat to death it is perhaps inappropriate but won't be a problem with additional RC at VCA control port to insure stability.

I'm sorry to whipsaw the original poster around. The last version is IMO most likely to give a good result, and facilitate stereo linking if desired (perhaps not common for de-essing).

Since i doubt external VCA control is desirable, and if it was, I would scale the VCA law to much higher than 6 mV/dB for obvious reasons. This is probably a non-issue.

For simple mono use control voltage driver version #2 is slightly better but in practice I don't expect any audible difference between #2 and #3, so choose based on your intended use.

This has been an interesting journey and finally makes some sense.../edit

 

[Curtis]

Thanks John, JDB and Keith.

edit: my apologies for my increasingly apparent wrong guess that the unusual topology was for stereo linking, "and" flawed. External VCA control was even further down my short list of guesses.

Hehe, I believe it was my guess that the inverting terminal point was for stereo linking, so the apology is mine for starting such a wild goose chase in the first place :guinness:

I'm sorry to whipsaw the original poster around. The last version is IMO most likely to give a good result, and facilitate stereo linking if desired (perhaps not common for de-essing).

Not a problem John. I am quite grateful to have some more experienced eyes check all this out at this early stage. Input offsets and rising output impedances of opamp circuits are things that were somewhere in my dim, dark memory from my electronic schooling, but long (and foolishly) forgotten.

AS I mentioned earlier, I have a pair of 902's here, a 900 rack, and plenty of test gear. -If you'd like any measurements taken to confirm or correct any assumptions, I'd be only too happy to do what I can.

Thanks Keith. At this stage I'll re-work the current PCB to accommodate the changes discussed, but I may yet take you up on that offer.

 

[SSLtech]

You're most welcome... after all, I -ADORE the 902, and I'm curiouis as to why the 263X has never risen to the same level of appreciation.

People who use channel-strip dynamics with the EQ inserted into the sidechain (a la SSL E, G and J-series, Neve V series, Calrec UA8000 and other -mainly digital- consoles) seem to be labouring under the misapprehension that they can thus create a de-esser. -Well, I suppose they can... but if you've ever USED a 902 and then tried to match it with a channel setup, it takes VERY little time to arrive at an admiration of just exactly how 'self-adjusting' the 902 is... specially for vocals.

-A slight fluctuation in vocal 'effort' changes the spectral balance, but heavy compression/limiting -as seems to be so much in favour these days- removes the dynamic envelope 'clues'... the 902 sees past the 'trick' and just does the job 'right'... in other words, once you've told it what region is bothering you, and how much you want it pulled down by, it makes STUNNINGLY agreeable 'decisions' about how and when to do it.

If the 263X (which is available for pennies-on-the-dollar on eBay) is not the equal of the 902 (which still commands VERY respectable eBay money) -and I suspect that it may not be- then I'm curious as to why... They used a comparable topology, and yet I suspect that something was 'lost' in the "translation".

It has been a subject of some pondering on my part over a few years now, how much home recordists here could more closely approximate 'professional' results, if they had a 902...

Certainly, for EVERY important mix that I've ever done -and I have a goodly number of silver, gold and platinum albums to my name over the years- there's been AT LEAST one 902 in the rack... and doing a brief 'bypass-check' (to see if I'm just inculding it in the chain out of habit or if it's actually 'serving the greater good' by double-checking the signal path WITHOUT it) has inevitably confirmed that it's just a GREAT piece, with no peer that I've EVER met.

A couple of years ago, I was up in Nashville at Harrison's factory for a few days, and I got to chat with a couple of the DSP guys there. They were offering the 'tools' section of their MPC-4D with a De-Esser as one 'tool'. -It was a fixed-threshold device, I discoverd when I asked them how it worked. -I suggested to them that they might want to investigate how the 902 did it, and when I explained how it worked, they were most intrigued. -Eventually they developed a SECOND version of their digital de-esser tool, with the 'adaptive' threshold mimicking the 902... -It's ALWAYS been that important to me, and once you 'dial your ears in' to how much a heavy limiter starts to call for de-essing, you never really turn back.

In a way, I can't help but wonder if the quest for 'shaped' sidechains on the cloned GSSL compressors is out of an attempt to reduce that specific side-effect... and while it's no less valid than the old 'compress and De-Ess" approach, it's another brush to use on the canvas, and I'd LOVE to see this come to fruition as a DIY project...

So I suppose that this is a long-winded way of explaining why I'm so keen to help if necessary... The offer stands. I can graph sweeps, run static measurements etc... whatever you need... -I'd just like to be CERTAIN that this doesn't end up being perceived as a 263X-type 'second-rate' implementation.

Keith

 

[JohnRoberts]

While a minor semantic distinction, I wouldn't call the threshold "adaptive" but more something like "(fixed) relative".

The specific sound character of this approach is due to responding to the difference between the two bandpasses mostly independent of their level rather than comparing the highpass to a fixed threshold. Making the crossover point between the bandpasses adjustable adds even more flexibility to shape spectral balance.

This strikes me as something more than a de-esser, but more is good, and it certainly can de-ess too.

JR

 

[jdbakker]

[quote author="SSLtech"]If the 263X (which is available for pennies-on-the-dollar on eBay) is not the equal of the 902 (which still commands VERY respectable eBay money) -and I suspect that it may not be- then I'm curious as to why...[/quote]
As am I. Looking over the 263X schematic, it looks like a budget version of the 902. Topology is essentially the same, but the 263X uses cheaper and lower specced VCA/RMS blocks, a 5534 output stage instead of the hybrid line driver the 902 has, no FET or other switch to kill the de-essing of tape noise and other very low level signals, and lower-spec decoupling and RMS integrator caps. The CV is 14mV/dB, and there's an extra inverter in the filter chain (U1) which isn't immediately obvious to me. Other than that they look the same.

Maybe the lower-spec parts (and the associated drift of a lower CV range) really make the unit sound worse, or maybe the 263X is yet another piece of gear that's underappreciated for no good reason. Keith, have you actually heard these side-by-side? Has anyone else?

JDB.
[I share Keith's concern -- the 902 and the 263X are close enough to make me want to ID any difference in sound (and its cause) lest we unintendedly end up with the ugly sister]

 

[SSLtech]

Can't say that I have ever side-by-sided the two units...

...but it makes me want to try.

Keith

 

[JohnRoberts]

While they look similar, there can be differences in the implementation details. Some might not embrace a 4558 in the audio path, while admittedly it is only in the LPF so shouldn't be too stressed. :wink:

JR

 

[drpat]

Keith, You're "spot on" about the 902, and I've been following this thread closely, as I've always wondered about how the 902 seems to "know" the difference between loud sibilants and soft ones.

I'm not sure if too many people would build it, as it's so cheap to buy anyway, but a great thread none the less.

 

[Curtis]

So I suppose that this is a long-winded way of explaining why I'm so keen to help if necessary... The offer stands. I can graph sweeps, run static measurements etc... whatever you need... -I'd just like to be CERTAIN that this doesn't end up being perceived as a 263X-type 'second-rate' implementation.

Thanks Keith, I'll certainly keep that in mind. I've etched the new revision of the PCB's this morning, but I've yet to drill them (have to borrow the drill press at my day-job's workshop), so we'll see how this one turns out.

Personally I've always found it extremely useful to have a de-esser on-hand for that extra "polishing", not just for it's obvious uses on overly "spitty" vocals, but also for things like de-emphasising a reverb send on a vocal, or cleaning up the bleed from hihats in the snare mike without resorting to heavy gating. The fact that you can also perform fancier moves on the 902 by just having it work on the HF signal (the mode switch), or extreme HF gain reductions makes it an interesting shaping tool to be playing with too.