XS902 - now with added "everything" PDF!

Edit: The first released version of the "everything" PDF is now available:

Editedit: Minor corrections to pdf, now up to version 3a:

http://www.mediafire.com/?mdj2zywxndk

Layouts, instructions, suggested front panels etc.

Ta Daaaaa! :sam:


Version 2 is built, although I've already had to get stuck into it with the wire jumpers. :roll:



The biggest problem I've had is that the D4 diode in parallel with R43 and C13 upsets the final sidechain opamp stage (IC8b) and causes it to oscillate badly. See here for sidechain schematic. The series Rxx/Cxx filter on the VCA CV input makes the opamp oscillate too, despite THAT's recommendations in their datasheets, although that could be more to do with the way the CV is tapped off the inverting terminal of IC8b. I'm also finding that I'm getting a cleaner sidechain signal (according to my scope) if I take the CV signal from the output of IC8b rather than that weird inverting-terminal arrangement.

However, I've done a couple of passes with some audio to see how it sounds. (44.1KHz, 16-bit mono WAV's in both cases):

Original untreated female vocal samples

De-essed female vocal samples

The de-essed version is done with the 902 set to 2.5KHz and knocking off 4-5dB at its highest peaks, mode switch set to "full range". Apologies for the slight background "grumbling" on the effected version - The 902 is yet to be installed in a case and it's probably picking up radiated noise from the PC.

Request for Keith, Drpat and others: Can I get you guys to download the untreated version and run it through your 902's with the same settings and see how it compares? Ta.

Nice progress!

Just one thing: if you start a new thread with every new revision, you stand the risk of losing the attention of all members who were receiving e-mail notifications for the older threads, plus it gets harder to look up the progress of the development once the threads have fallen off the bottom of the first page.

JDB.

IC8b "should" be stable with or without the diode.

The series RC hanging off the - input if anything should also increase opamp stability due to attenuating the negative feedback path at HF.

I would look for possible noise on the + input and stray capacitance or noise pickup to that - input. In general it is not good design practice to hang an antenna off the - input, so a long trace there could pick up garbage.

Obvious tweak to compensate for stay capacitance on - input is to increase value of 10 pF cap across that opamp feedback net. Since the voltage at the output of that opamp stage isn't being used the C value there is arbitrary.

I still find it surprising that adding the diode and RC to gnd has any effect at all on that opamp stability. I would not rule out some other cause. The scope probe itself has some (small) capacitance and could upset a marginal circuit, perhaps another clue that the 10pf could be larger.

Something does not compute...

JR

I still find it surprising that adding the diode and RC to gnd has any effect at all on that opamp stability. I would not rule out some other cause. The scope probe itself has some (small) capacitance and could upset a marginal circuit, perhaps another clue that the 10pf could be larger.

Click to expand...

I did think of the probe as a possible cause, but scoping the output driver (IC3a/b) showed a very distorted waveform - looked like the CV oscillation was riding on top of the audio.

Removing the R/C filter and diode while retaining the inverting terminal driving arrangement of IC8b cleaned up the CV. Putting the R/C filter back in made the chip oscillate. Taking the R/C filter out and putting the diode back in made the chip oscillate. Taking the R/C filter and diode out, and moving the take-off point for the CV to the output of IC8b, the chip was stable. I've yet to try re-installing the R/C filter and diode with the CV being driven from the output of the opamp.

With no input signal, or with the unit in bypass (Q1 switched on, shunting the CV to ground) there is no oscillation. Maybe there's some funky layout problem with the board itself?

I'll try upping the feedback cap on IC8b and re-installing the diode/RC combo and see what differences I get.

Just one thing: if you start a new thread with every new revision, you stand the risk of losing the attention of all members who were receiving e-mail notifications for the older threads, plus it gets harder to look up the progress of the development once the threads have fallen off the bottom of the first page.

Click to expand...

Thanks for the tip, JDB I'll continue using this thread from here on in, and stick a link in the old one for now.

Curtis,

Do you have an e-mail address that I can send this file to???

[quote author="Curtis"]

I still find it surprising that adding the diode and RC to gnd has any effect at all on that opamp stability. I would not rule out some other cause. The scope probe itself has some (small) capacitance and could upset a marginal circuit, perhaps another clue that the 10pf could be larger.

Click to expand...

I did think of the probe as a possible cause, but scoping the output driver (IC3a/b) showed a very distorted waveform - looked like the CV oscillation was riding on top of the audio.

Removing the R/C filter and diode while retaining the inverting terminal driving arrangement of IC8b cleaned up the CV. Putting the R/C filter back in made the chip oscillate. Taking the R/C filter out and putting the diode back in made the chip oscillate. Taking the R/C filter and diode out, and moving the take-off point for the CV to the output of IC8b, the chip was stable. I've yet to try re-installing the R/C filter and diode with the CV being driven from the output of the opamp.

With no input signal, or with the unit in bypass (Q1 switched on, shunting the CV to ground) there is no oscillation. Maybe there's some funky layout problem with the board itself?

I'll try upping the feedback cap on IC8b and re-installing the diode/RC combo and see what differences I get.

Just one thing: if you start a new thread with every new revision, you stand the risk of losing the attention of all members who were receiving e-mail notifications for the older threads, plus it gets harder to look up the progress of the development once the threads have fallen off the bottom of the first page.

Click to expand...

Thanks for the tip, JDB I'll continue using this thread from here on in, and stick a link in the old one for now.[/quote]

I am not married to that drive circuit, but it should be possible to make it stable as drawn and do all the tricks promised. Perhaps increase feedback cap.

The RC to ground on CV pin.. was just something I read in the THAT data sheet. I just checked again and they predict a 5532 (output) should be low enough impedance, but nowhere do they show anything as odd as this drive approach. In another troubleshooting note they suggest 51 ohm and 2.2 nF to stiffen up the control port.

I notice that same CV line is going to the meter PCB, so there may be something going on there if not yet, later.

JR

Do you have an e-mail address that I can send this file to???

Click to expand...

curtis<at>thethirdending.com will do nicely, thanks.

I am not married to that drive circuit, but it should be possible to make it stable as drawn and do all the tricks promised. Perhaps increase feedback cap.

Click to expand...

OK, was able to do some systematic susbtituting over the last couple of hours:

1. CV drive point moved back to inverting terminal - all OK.
2. Feedback C upped to 470pF - all OK. The 10pF cap with a 1K resistor in parallel would have given a theoretical cut off point of over 15MHz (!!!). I don't have a 5532 datasheet here at the moment, but I'm guessing that's probably not practical for this chip, nor will it help. 470pF will drop it back down to a more respectable 330KHz.
3. Diode added back in parallel with feedback R and C - all OK.
4. R/C filter (100R, 1n5) on control port added back in - all OK.

Given the above I'm guessing the feedback C was the culprit, but now that it appears to be working I'm kinda loathe to pull it all apart and put the 10pF back in to prove myself right!

Also noticed that I was getting a slightly "fuzzy" looking waveform on the output of the input diff amp (IC1a). Figuring it was my crusty old scope at first I just ignored it, but on closer inspection it was actually IC1a oscillating aswell. The non-inverting terminal looked clean, but the inverting terminal was a mess. Whacking a pair of 470pF caps in parallel with the 7k5 feedback resistor and the 7k5 shunt resistor seems to have tamed the oscillation on this opamp. I could probably decrease the cap values a bit more (470pF was the only thing I could find at the time) as the frequency response at the output of IC1a starts to droop a little bit before it gets all the way to 20KHz.

Perhaps it was a combination of things upstream of the VCA that was upsetting the CV port?

I got tugged otu of work today early because my 4-year old is sick, so it'll be Friday before I get back to work... Pat did you get that file done, or should I do it on Friday?

Keith

I am still curious about what else is hanging off that CV line since it goes off that board and connects with the meter board?

JR

Keith,

Got the file processed...

Curtis,

Check your e-mail....

With a little more experimentation I'm finding that all the opamp stages are benefitting from the addition of a few hundred pF's of capacitance in parallel with the feedback resistances, even the ones in the sidechain. I'm seeing some degree of oscillation on every inverting terminal where no feedback cap is used, which seems to add some "blurring" to the waveform as viewed on the scope - a clean, crisp sinewave from the sig gen starts getting "fuzzy" (for want of a better term) as it passes through the opamp stages without feedback C's. As soon as a feedback C is added the "fuzziness" is reduced, the sinewave starts looking more crisp again, and the oscillation on the inverting terminal disappears.

I am still curious about what else is hanging off that CV line since it goes off that board and connects with the meter board?

Click to expand...

The only other thing hanging off that point is a non-inverting opamp stage (TL071) with 10x gain for the meter driver - the CV voltage is applied directly to the non-inverting terminal. The meter driver itself is an LM3914 showing linear 2dB steps of gain reduction to a maximum of 20dB. I'm going to update the diagram later on today to show all the changes so far to the main schematic, and also include the meter schematic.

I should probably also point out that the weird CV oscillation I was experiencing before was there regardless of whether the meter daughterboard was plugged in or not.

Curtis,

Check your e-mail....

Click to expand...

Thanks Patrick, I haven't had a chance to listen to it yet but I'll check it out later on today.

If inverting input terminals look funny when you probe them... don't probe them. If the + input and output is clean... you're OK. Adding more feedback cap will allow it to tolerate the C from the scope probe.

The input of a TL071 is a few pF and no problem...

Opamps at very input stage need some feedback cap to LPF signal path. Opamps at very output need some feedback cap to help keep output impedance low at HF.

How does it sound?

JR

PS: to confirm if your probe is causing the problem, probe with a 1k resistor in series.

Up-to-date schems here:

Main board
Meter daughterboard

Summary of changes:

• New voltage divider for bypass switch added to correctly turn off comparator section when bypass engaged (R57, R8).
  
  Added compensation caps to opamp stages IC1a, IC2a, IC6a, IC6b (C27-C31).
  
  Compensation cap on CV driver upped to 470pF (C13).
  
  Input resistor to VCA (R44) and feedback resistor on following buffer (R46) changed to 20K (THAT's suggested circuit).
  
  Input cap on VCA (C14) changed to a low-leakage 10u electrolytic (more likely for the average DIY-er to get their hands on this cap than the 1u MKT I had in there before, although the PCB will still support it if they choose to install one).
  
  Feedback/shunt resistors on input diff amp (R5, R6) reduced to 5K6 to make the input-to-output insertion loss closer to 0dB when running balanced (doesn't help when running unbalanced though, gives a 6dB loss in this case).

If inverting input terminals look funny when you probe them... don't probe them. If the + input and output is clean... you're OK. Adding more feedback cap will allow it to tolerate the C from the scope probe.

Click to expand...

Fair enough, but the waveform on the output of these stages looks affected without the C, and cleaner with the C, so I'm assuming there's definitely a problem there that is being tamed by the addition of the extra capacitance.

How does it sound?

Click to expand...

Not too bad actually :grin: Decide for yourself if you like - there's "before" and "after" WAV's a little bit further up the first page of this thread.

Dr Pat did a quick pass over some samples with his original DBX902 to compare with - doesn't sound too far from what I'm getting with my unit. Mine's probably a little more aggressive on the reduction of sibilants than his original, and maybe a little less clean-sounding, but it's difficult to get an accurate picture of how much I'm de-essing as the meter I've got installed only goes up in 2dB steps which makes it a little bit coarse.

The addition of the extra caps in parallel with the feedback R's has cleaned up the output a little bit more aswell, so I might post another audio sample later on to compare with.

Request for Keith (or Dr Pat, or...): If it's still possible could I get you to do a couple of measurements on your unit for me please? I'd like to get comparisons on the rise/fall times on the outputs of the RMS detectors between your DBX902 and the new THAT2252's in mine. Would you be able to burst the inputs with a short tone and measure the output rise and fall times? It might be possible to tweak the RMS timing caps in mine to closer approximate the response of the original RMS modules (if indeed there is any discrepancy).

I also wouldn't mind knowing what what quiescent output voltage the original RMS detector sits at with no signal input. Ta very muchly.

Many thanks, it's looking good. I shall be joining you in southern hemisphere sunshine in 3 weeks - we are lurching down to Argentina, so this will all make for excellent in-flight reading (16 hours ) whilst my wife sleeps!

Any updates on this project? Were you satisfied with your 2nd prototype in the end?

Yeah, sorry. A bit of time off there as I had 4 weeks annual leave from work, and too many other concurrent DIY projects at the same time! Long story short I've incorporated a few changes I did to the layout of V2 into V3, mainly just adding frequency response limiting caps on the feedback resistors in most opamp stages. This past weekend I've etched a +/-15V PSU board but have yet to populate it. Hopefully in the next week or so I'll be able to get two channels working side-by-side and post some more updates. Sounding good so far though

OK, finally received the last few components to finish off the power supply and had a couple of spare hours last night to have a fiddle with two channels. Tying the two channels together at the point labelled "Sidechain in" on the schematic does result in stereo de-essing, but also causes a lot of buzzing and noise in the audio path. The two inverting terminals definitely don't like trying to drive each other, and it sounds like (haven't had a chance to scope it yet to verify) the buffers driving the control ports of the VCA are very unstable. Removing the stereo link, everything returns to normal.

More experimentation required. However, as a dual-mono de-esser it works really well!

If it's not too much trouble I'd be curious about what precisely the opamps are doing when linked together?

As I read the schematic, which ever channel opamp has the more positive + input control voltage will pull the common - input from both upamps up to that level. The less positive opamp should sit saturated with it's output pegged to negative supply. Since the more positive opamp is driving through a diode it should easily win that tug of war.

If the diodes weren't there, or not both in the same direction, both opamps would try to control the result.

The first thing I would look at is DC voltage at output pins of each opamp. One should be sitting .5v more positive than the common - inputs, the other output should be pegged to - supply.

I have never seen anything exactly like this circuit so one thing I can't be very confident about is if the 470pf caps are so big they cause a stability problem in that topology when linked together (just an early guess if opamps are oscillating).

Not easy to debugg over such distance. But if you want to make the link work I'm game. I know it can be done, trick is with least changes.

JR

Sorry this is going a bit slow at the moment, haven't had much free time in the last couple of weeks.

Righto, spent a little bit of time on the weekend looking into this. I can get pretty good results by joining the two inverting terminals together with shielded cable. The only drawback seems to be that the shield has to be connected at both boards, which may open up a can of worms for creating ground loops between the two channels. Using shielded cable with only one end of the shield earthed is still noisy, although not as much as no shield at all.

The noise and crap I'm seeing on the outputs seems to be much worse when I connect the link points via unshielded wire. In fact, I get similar noisy results by just clipping a wire on to the link point of one board with the other end un-terminated. The inverting terminal must act as a pretty good aerial, especially with a long piece of copper hanging off it.

Noise or not, with the two boards linked at the inverting terminals, stereo de-essing occurs as you'd expect, with the "ess-ier" of the two channels driving the other channel down by the same amount.

Please 'scuse the less-than-stellar shots, only had my camera-phone with me at the time.

Both channels linked via shielded cable, left channel driven, no GR, left channel output:



As above, unshielded cable used:



As above, GR just starting to come into play:



1KHz, 2V/div in all the above.

Noise seems to get worse as the GR starts increasing.

I would be useful to probe a few other circuit nodes in addition.

Obvious nodes to look at are the output pins of two 5532s driving the common sidechain control voltage line when linked.

If one or both of those opamps have noisy output, confirm their + inputs are clean.

The intent of the design is that only the one higher opamp will dominate and other will be pegged to opposite supply. It looks like it is sort of working.. but if oscillating that may cause noise.

JR