XS902 - now with added "everything" PDF!

[Curtis]

Thanks John, I'll have a closer look over the next day or two.

The intent of the design is that only the one higher opamp will dominate and other will be pegged to opposite supply.

Yes, at the time I did check the 5532 output pins to see if it was indeed doing this, but only with a voltmeter. I have yet to to scope it to see if there are any other gremlins lurking on the O/P of those chips. From memory I think I scoped the preceeding sidechain opamp output pins, but they all appeared clean, so it must be that last stage just before the CV input to the VCA that's causing the trouble.

I do remember scoping the outputs of the audio stages, and there was definitely nothing untoward until I got to IC2b (the buffer hanging off the VCA output), where I started seeing stuff like in the above photos.

 

[Curtis]

OK, had another chance to scope things a bit. With no stereo link, all sidechain pins are clean. With the stereo link inserted I'm seeing approx 12mV pk-pk noise on the CV input to the VCA (my crusty old scope won't go any better than 20mV/div, so I can't get any more detail than that). Slightly more weird is that I'm also seeing a slight increase in noise on ALL the sidechain opamps with the stereo link in, maybe 3-5mV pk-pk. With the stereo link in, the two channels' CV buffers operate as you describe - the more positive opamp driving both channels down, and the less positive opamp sitting with it's output pegged hard-negative. They're just...noisy.

Using shielded cable for the stereo link with the shield connected at both ends to earth improves things, but it's still not as clean as having the two channels operating independantly.

I'm starting to think that simply joining the two channels in this way is not going to work without some kind of intermediate circuitry, some kind of buffer to provide isolation between the two channels sidechains.

 

[SSLtech]

I'm with you. While it's not something that I've ever tried myself, I've always assumed that if you attempt to combine two inverting input nodes, then the amplifiers will play tug-of-war with the (now common) node...

I think that since the 902 is only ever used in mono mode in the 900 series racks, this may be a somewhat academic exercise, although I am enjoying the exploration of the idea...

The only other thing which I think looks suspect on the schematic is the hard-switching of the inverting input through SW2 (bypass). -In the 'halfway-there' position, the op-amp output will likely slam to the rail... there may perhaps be a more elegant way to 'shunt' it to ground, for more gently behaviour during switching?

Keith

 

[jdbakker]

I would think that for proper stereo de-essing you'd want something like True RMS Summing for both the full-band and HPF chains, as we're looking for relative energy levels. This would mean that you can stereo-link, only not through the point you're using now. As far as I can tell, the original 902s were never meant to be linked at this point, right ?

JDB.
[not to say that it wouldn't be a Good Thing to get those noise issues resolved]

 

[JohnRoberts]

That greater of two circuit combined at - inputs is surely unconventional, but that doesn't mean it can't be made to work. I would be tempted to experiment with value of 470 pf up or down. I can also imagine a way with addition of a resistor in series with minus input and moving 470 pf back to the actual - input pin, would isolate the opamp from capacitance on that line. That said, there may be an alternate solution, not that I'm giving up on this approach.

Starting with a clean sheet of paper instead of trying to make that work as close as possible to originally drawn, I would observe that when linking CV between two PCBs, ground differences will look like control voltage modulations, and show up in outputs.

:idea: From a quick glance, it looks like instead linking at the buffers, linking the two channels at the junction of D3 cathode, R41, and R37, should also deliver the "greater of either" channel function. This way the .33x pad is after the side chain link, so pad reduces impact of noise and ground potential differences between channels.

If one channel is bypassed it can still influence the other channel when linked but that may be useful in a side chain ducking like application.

Moving the stereo link to here, you can lose the diodes from buffer amps, ASSuming this works quietly.

JR

 

[Curtis]

As far as I can tell, the original 902s were never meant to be linked at this point, right ?

That's right, but it ain't gonna stop me from tryin'! :razz: I can fit two of these things in a 1u case, thus it seemed logical to me that I should be able to offer this as a dual mono/linked stereo option for such occasions where DIY'ers wanted to do some stereo de-essing aswell.

The only other thing which I think looks suspect on the schematic is the hard-switching of the inverting input through SW2 (bypass). -In the 'halfway-there' position, the op-amp output will likely slam to the rail... there may perhaps be a more elegant way to 'shunt' it to ground, for more gently behaviour during switching?

You mean IC8a? It's meant to switch hard-positive to hard-negative, so an extra slam during the switch transition probably isn't going to upset it too much. I certainly can't hear significant pops and splutters when engaging bypass.

I would be tempted to experiment with value of 470 pf up or down.

...

linking the two channels at the junction of D3 cathode, R41, and R37, should also deliver the "greater of either" channel function

These are all easy tweaks to do, so I should be able to try them out this evening.

Thanks all!

 

[Curtis]

No dice. :? Tried increasing the feedback cap as far as 4.7nF on both channels - no real improvement. Moved the stereo link point to the junction of D3/R41/R37 - a little bit better, but still very prone to noise pickup unless shielded cable is used, and even then the shield still has to be connected at both PCB's grounds for best results. Removed diode D4, still no real improvement.

The two channels are definitely happiest when they don't know about each other! :razz:

My poor brain can't easily picture how two dogs can share a bone in this case... I suppose it's going to be current-sharing rather than a voltage 'fight'. but if I were drawing something which I wanted to work simply, I'd never try and join those two nodes...

Keith

 

[JohnRoberts]

It is always possible to teach an old dog a new trick, it just takes a little more effort.

#1, confirm that connecting ground between the two circuits is not causing a problem. If they use a common PS the ground should already be connected, so connecting again could cause a loop.

#2, the main problem in sharing a control voltage that is tens of mV is that small ground potential differences get added to CV. When I was designing dynamics processors I always scaled up log of signal level voltage so FS analog, would be represented by a DC voltage of 10v or so. In this design since you're dealing with two log convertors that would require a second gain stage (or tweaking values inside 2252). It's a lot easier to scale up 30k at R37 for more voltage swing there. Of course you then also need to increase R41 so output scales back down.

#3 Next problem is impedance of the control line that is being shared. In this case it's roughly 18k with current values used and even higher if we scale up R41. These relatively high values are used so shunt FET to command bypass will be effective. FWIW there is no reason why we couldn't just add a resistor to ground at cathode of D3 to drop that node impedance. Say a 3k to ground on both channels drops that impedance a factor of ten.

===========
So simplest thing to try is adding 3k to ground from cathode(s) D3 and use shielded wire to link. Shield should be grounded on one end. If two channels don't already have grounds connected that should probably be a different ground wire.

Additional noise immunity can be gained by scaling up log domain voltage (increase R37), and scaling back down again after link node (increase R41). This will be limited by max attenuation accommodated.

The meter(s) should be driven from the high voltage, (relatively) low impedance node (cathode D3) with their own voltage divider(s) at meter input on meter board(s) as needed.

JR

 

[Curtis]

OK, that's weird - somehow Keith's reply ended up inside mine!

#1, confirm that connecting ground between the two circuits is not causing a problem. If they use a common PS the ground should already be connected, so connecting again could cause a loop.

No, definitely not the problem. The noise problem exists when using unshielded wire, and is in fact worse. Using shielded cable with the earth at one end only improves things a little but not by much. Using two pieces of shielded cable with one end earthed at each PCB only (ie, not connected all the way from one PCB to the other), same result. Shield at both ends seems best.

It's also worth pointing out that with a single unshielded, unterminated wire hanging off the stereo link point I still get garbage on the output, which immediately disqualifies it from being a solution if I have to run the wire to a switch on the front panel.

So simplest thing to try is adding 3k to ground from cathode(s) D3 and use shielded wire to link. Shield should be grounded on one end. If two channels don't already have grounds connected that should probably be a different ground wire.

Isn't that going to screw up my CV scaling going into the VCA? Or do you mean the 3K is only added when the stereo link is engaged?

 

[JohnRoberts]

[quote author="Curtis"]

So simplest thing to try is adding 3k to ground from cathode(s) D3 and use shielded wire to link. Shield should be grounded on one end. If two channels don't already have grounds connected that should probably be a different ground wire.

Isn't that going to screw up my CV scaling going into the VCA? Or do you mean the 3K is only added when the stereo link is engaged?[/quote]

No.

A 3k to ground from cathode of D3 does not change feedback path back to opamp - input, or divider forward... It only makes that point lower impedance. Actually it should already be low impedance but doesn't have much sink current capability.

try adding 3k to ground on both channels (all the time) and link those points with shielded wire.

Later if you are adventuresome you can try scaling up voltages.

JR

 

[jdbakker]

bumpbumpbumpBUMPbumpBUMPbumpbump

JDB.

 

[Curtis]

Heh, it has been a while hasn't it?

Basically, the current status is I can't get it to play nicely as a stereo unit with that wacky CV summing circuit - it just refuses to do it cleanly. As a mono de-esser it works great, it just sucks in stereo :razz: The original unit obviously wasn't meant to be used this way either, so it's not surprising I've had all this trouble with two units strapped together.

So the upshot is I'm thinking of simply finishing it off as a single channel unit and letting it be done. I should be able to post the current layout fairly soon if anyone is interested.

 

[SSLtech]

Yes yes yessss!

-Count me IN!

I loves me some 902!

Keith

 

[jdbakker]

[quote author="Curtis"]Basically, the current status is I can't get it to play nicely as a stereo unit with that wacky CV summing circuit - it just refuses to do it cleanly.[/quote]
If you have the time (and if you can be bothered) you may want to try connecting pin 6 of IC4 of one channel to pin 6 of IC4 on the other channel, and likewise with pin 6 of IC5. This will yield True-RMS summing of both channels' sidechain inputs, which should give properly working stereo operation. Pin 6 of the THAT2252 is a current output, so it should be somewhat less sensitive to capacitive loading and ground noise.

(In a production system you'd want to have a multipole switch disconnect C6/C11/R18/R26 on one of the channels and possibly bypass one of the sidechains, but for a 5-minute test the abovementioned recipe should be fine.)

JDB.

 

[JohnRoberts]

I am a little frustrated because I know control voltage manipulations are not heavy lifting. Without specific feedback from my last suggestions (in May) I have nothing to add.

Trying to make it work with minimal changes from the original drawings can be a little harder, than just doing it right.


JR

 

[JohnRoberts]

[quote author="jdbakker"][quote author="Curtis"]Basically, the current status is I can't get it to play nicely as a stereo unit with that wacky CV summing circuit - it just refuses to do it cleanly.[/quote]
If you have the time (and if you can be bothered) you may want to try connecting pin 6 of IC4 of one channel to pin 6 of IC4 on the other channel, and likewise with pin 6 of IC5. This will yield True-RMS summing of both channels' sidechain inputs, which should give properly working stereo operation. Pin 6 of the THAT2252 is a current output, so it should be somewhat less sensitive to capacitive loading and ground noise.

(In a production system you'd want to have a multipole switch disconnect C6/C11/R18/R26 on one of the channels and possibly bypass one of the sidechains, but for a 5-minute test the abovementioned recipe should be fine.)

JDB.[/quote]

That would average the result from both channels. In an ideal linkage it should respond to the worse of both channels. But simple ave would be easy enough to do and maintain solid stereo image center.

JR

 

[king-rb]

i never tested the 902 ... is it also good for mastering use=? or is it posible to do some mods to work with smaler bandwidth=?

 

[SSLtech]

Not great for mastering in my experience.

If you're talking analog De-Esser for mastering, you'll NEVER treat anything else as being useful once you've tried the Maselec.

But for vocal De-Essing, the 902 is simply without equal.

Keith

 

[mikep]

[quote author="JohnRoberts"]In an ideal linkage it should respond to the worse of both channels. But simple ave would be easy enough to do and maintain solid stereo image center.
[/quote]

youve made some good points, but what IS the ideal linking behavior of a stereo de-esser? who knows. I like the 902 for vocals but on a mix? why even bother with the linking?

from a mostly-mastering perspective, Id have to say M/S de-essing is the most useful. de-ess only the mid or the sides. usually the mid. I dont have a masselec, any one got a schematic of that one? Ive always wondered what is going on in that box.

mike p

 

[JohnRoberts]

IMO the proper operation for a stereo linked de-esser is to de-ess the offending channel the same as if it was working in mono, but apply the same correction to both channels so stereo image doesn't get pulled one way or the other.

Averaging the result protects the stereo image, but it now has a different de-essing effect in stereo mode as mono..

YMMV

JR

EDIT : for vocals which are usually panned roughly center the ave approach is probably fine... perhaps if trying to manage cymbal crashes or some source less centered it might be noticeable,,,

For DIY projects you can always say I meant to do that... :wink: