ruairioflaherty
Well-known member
A quick cheer from the sidelines. This thread is just as good as I had hoped and everything this forum came from.
GroupDIY Bridge Compressor.
- Thread starter DaveP
- Start date
Help Support GroupDIY Audio Forum:
Thanks ruairioflaherty,
Glad we're pleasing someone, I do wonder sometimes.
Anyway, I would really like some more contributions on attack and release times, I've only had 3 so far. I've just worked out the first 2 fairchild positions to be 0.1 and 0.3S release so they have to go in the pot. (Fairchilds published figures are for more than the usual first time constant.)
best
DaveP
Glad we're pleasing someone, I do wonder sometimes.
Anyway, I would really like some more contributions on attack and release times, I've only had 3 so far. I've just worked out the first 2 fairchild positions to be 0.1 and 0.3S release so they have to go in the pot. (Fairchilds published figures are for more than the usual first time constant.)
best
DaveP
> unless we try new approaches then we are condemned ... clone slaves
I agree about copy/stealing last year's products.
As you know, we can't "just clone" a 50 year old specialty box... so there is room for thought, modest innovation, and make-do.
> 12AU7 is fully on its output Z is 15k// with 10K (6K) so the inductance required for 20Hz would be 48H
Or if the resistor is 1K the gain and output is lower but you only need 4+H.
Lower output also means lower power-level stress in the iron, a smaller core.
Gain is cheaper than iron, we can find 6X gain in some other stage.
Maybe someone can tell me how this "Bridge" works. As I see it, it will attenuate to a null and then come back up (in reverse phase). No? How do they avoid that overdrive burst-out? Trims? Picking values so it won't ever get close to null (or deep GR)?
I agree about copy/stealing last year's products.
As you know, we can't "just clone" a 50 year old specialty box... so there is room for thought, modest innovation, and make-do.
> 12AU7 is fully on its output Z is 15k// with 10K (6K) so the inductance required for 20Hz would be 48H
Or if the resistor is 1K the gain and output is lower but you only need 4+H.
Lower output also means lower power-level stress in the iron, a smaller core.
Gain is cheaper than iron, we can find 6X gain in some other stage.
Maybe someone can tell me how this "Bridge" works. As I see it, it will attenuate to a null and then come back up (in reverse phase). No? How do they avoid that overdrive burst-out? Trims? Picking values so it won't ever get close to null (or deep GR)?
PRR,
I have had a bridge working on the bench and the tube and its balancing resistor are chosen so that they balance with 0V on the grid. The tube is normally biased at -30V and the sidechain has a +ve output instead of the usual -ve. As output increases it thus gradually turns the bridge off by turning the tube on! The closer you get to zero null point the greater the compression available.
I still have the cheap interstage I couldn't use for my BA-6A so I will check to see if I can get a decent response out of that by reducing the load resistor as you suggest.
best
DaveP
I have had a bridge working on the bench and the tube and its balancing resistor are chosen so that they balance with 0V on the grid. The tube is normally biased at -30V and the sidechain has a +ve output instead of the usual -ve. As output increases it thus gradually turns the bridge off by turning the tube on! The closer you get to zero null point the greater the compression available.
I still have the cheap interstage I couldn't use for my BA-6A so I will check to see if I can get a decent response out of that by reducing the load resistor as you suggest.
best
DaveP
> contributions on attack and release times
You don't need specific numbers. It's DIY, you give a table of R C values builders can select to taste.
You do need the extremes. Fairchild's goal was ANY audio bandwidth event, 0.05mS (50uS) so that a cutting-lathe would NOT over-cut. On one hand we now have faster channels, 48KHz and 96KHz. OTOH an "over" in good digital gear is a non-event. Good arguments have been made that <5mS clipping goes un-noticed. Some engineers disagree so 1mS could be a goal.
At the other extreme, >100mS release ratchets-down the average audio level, opposite of what we think a limiter should do. If every P-pop drops level and it takes a long time to recover, a P-popper results in a weak sea-sick level (much like my over-processed cable-TV operator). OTOH sometimes the goal is clean and natural not pumped-up.
Short-release piano is awful (even in pop music; why the Rhodes was much-loved).
Piano initial transient is sudden, so attack must be short.
We can change the magnitude of _both_ attack and release by changing the capacitor.
But the _ratio_ of attack to release determines the ratio of cap-charge current to cap-discharge current. This sets requirements for cap driver and cap load.
With simple tube limiters, attack current may be <10mA. Discharge is limited by ~~0.1uA grid-current, but this is very variable so the user-control discharge must be much higher. Taking 1mA and 1uA we can have 1,000:1 ratio of attack to release. 1mS/1Sec, 5mS/5Sec, 0.05mS/50mS, etc.... but such diverse choices require cap-change. To handle 0.05mS:10Sec plus variable leakage we need 2,000,000:1 ratio of charge to discharge current.
With the six grids in a Fairchild we need a really hefty 2x6V6 sidechain driver.
For the much narrower time choices in a Federal we can get away with a single triode.
(Yes, arguably such wide ratios could be done with dual-constant networks, or even Audimax/Volumax dual-boxes, but too late tonight to get into that.)
BJTs do big current, JFETs do small current, better/cheaper than any affordable tube sidechain. This does multiply the problem of fancy limiters needing multiple supply nodes at various polarities at wide range of voltage.
So what is the sweet-spot of practical charge/discharge currents? Obviously each designer has to work this out so the users will be pleased.
You don't need specific numbers. It's DIY, you give a table of R C values builders can select to taste.
You do need the extremes. Fairchild's goal was ANY audio bandwidth event, 0.05mS (50uS) so that a cutting-lathe would NOT over-cut. On one hand we now have faster channels, 48KHz and 96KHz. OTOH an "over" in good digital gear is a non-event. Good arguments have been made that <5mS clipping goes un-noticed. Some engineers disagree so 1mS could be a goal.
At the other extreme, >100mS release ratchets-down the average audio level, opposite of what we think a limiter should do. If every P-pop drops level and it takes a long time to recover, a P-popper results in a weak sea-sick level (much like my over-processed cable-TV operator). OTOH sometimes the goal is clean and natural not pumped-up.
Short-release piano is awful (even in pop music; why the Rhodes was much-loved).
Piano initial transient is sudden, so attack must be short.
We can change the magnitude of _both_ attack and release by changing the capacitor.
But the _ratio_ of attack to release determines the ratio of cap-charge current to cap-discharge current. This sets requirements for cap driver and cap load.
With simple tube limiters, attack current may be <10mA. Discharge is limited by ~~0.1uA grid-current, but this is very variable so the user-control discharge must be much higher. Taking 1mA and 1uA we can have 1,000:1 ratio of attack to release. 1mS/1Sec, 5mS/5Sec, 0.05mS/50mS, etc.... but such diverse choices require cap-change. To handle 0.05mS:10Sec plus variable leakage we need 2,000,000:1 ratio of charge to discharge current.
With the six grids in a Fairchild we need a really hefty 2x6V6 sidechain driver.
For the much narrower time choices in a Federal we can get away with a single triode.
(Yes, arguably such wide ratios could be done with dual-constant networks, or even Audimax/Volumax dual-boxes, but too late tonight to get into that.)
BJTs do big current, JFETs do small current, better/cheaper than any affordable tube sidechain. This does multiply the problem of fancy limiters needing multiple supply nodes at various polarities at wide range of voltage.
So what is the sweet-spot of practical charge/discharge currents? Obviously each designer has to work this out so the users will be pleased.
ruairioflaherty
Well-known member
DaveP said:
Well, my goal may be different to others. I'm not here for a clone or new wonderbox, I'm here to watch and learn as you guys build this project from the ground up and as such I'm having lots of fun.
I'll comment more on my modified 26C times as I can, might be a week. I think my fastest release was obtained with the stock 0.25 cap and a 150K R. Sounds pretty amazing on something like James Brown or The Meters as program material. Fastest attack will pretty much erase a snare drum, far more than any vari-mu I've used, even though 2mS is quoted. I built out a rotary switch of even release values, picked by ear, and I've never done the calculations to see where I landed. I think I maxed the release around 2M instead of the stock 10M.
Remember the 26C meter at least only tells you about the first 5 dB of reduction, and is pretty easy to bury. The manual calls 13 dB the max, but that's related to overload of the input stage preceding. Overload in 26C may well be part of the limiting effect on purpose.
With an SA-39 I seldom use a release more than it's fastest, which I think is 0.2S. The fastest attack gets you that Jimmy Page acoustic attack sound, but a bit slower gets you good pop piano attack.
The dual timing of the NBC vari-mu AGC I have is 20 seconds for 90% recovery after prolonged 35 dB reduction, or 1 second for 90% recovery after a 0.1 second 20 dB reduction. This was the model that guided later RCA limiter dual timing methods.
Remember the 26C meter at least only tells you about the first 5 dB of reduction, and is pretty easy to bury. The manual calls 13 dB the max, but that's related to overload of the input stage preceding. Overload in 26C may well be part of the limiting effect on purpose.
With an SA-39 I seldom use a release more than it's fastest, which I think is 0.2S. The fastest attack gets you that Jimmy Page acoustic attack sound, but a bit slower gets you good pop piano attack.
The dual timing of the NBC vari-mu AGC I have is 20 seconds for 90% recovery after prolonged 35 dB reduction, or 1 second for 90% recovery after a 0.1 second 20 dB reduction. This was the model that guided later RCA limiter dual timing methods.
MagnetoSound
Well-known member
DaveP said:
Thumbs up here, Dave. I will try to remember to pipe up now and again.
Thanks for the info so far, I think a table of values is a good idea.
In the meantime, I think I have worked out why the 26W has a tapped OPT for the sidechain.
The stock tube has an rp of 2.6K and in parallel with 5K/4 is 2.6K//1.25 which is 844 ohms.
This governs the minimum attack time, so with a 0.25uF cap it would be 0.2mS but the published minimum is 0.1mS. They have used a tapped OPT at about 2K to give 2.6K//500 which is ~400 ohms, this gives the stated 0.1mS attack.
best
DaveP
In the meantime, I think I have worked out why the 26W has a tapped OPT for the sidechain.
The stock tube has an rp of 2.6K and in parallel with 5K/4 is 2.6K//1.25 which is 844 ohms.
This governs the minimum attack time, so with a 0.25uF cap it would be 0.2mS but the published minimum is 0.1mS. They have used a tapped OPT at about 2K to give 2.6K//500 which is ~400 ohms, this gives the stated 0.1mS attack.
best
DaveP
plumsolly
Well-known member
DaveP said:
Don't wonder too hard. This thread is fantastic and I really appreciate your time and efforts!
On my 26C, using a 0.22uf cap, I ended with a 5M rheostat with 1M in series for release. I think the I chose the 1M because anything smaller (faster) than that gave me objectionable audible distortion at faster attack settings. Looking at EMRR's makes wonder though - I like a really fast release for making toms huge or whatever. I can't say I end up in the really slow release settings ever.
For attack I used a 2M rheostat with nothing in series. I hang out in the middle a lot here, but I am not a very fast attack guy usually.
Best,
Ben
Thanks for your input Plumsolly,
What is emerging is a very wide taste in timings and as PRR suggested, its probably better to make a table so that builders can choose the spread they want.
Thanks again for your encouragement, people give of their best with a positive vibe.
best
Dave
What is emerging is a very wide taste in timings and as PRR suggested, its probably better to make a table so that builders can choose the spread they want.
Thanks again for your encouragement, people give of their best with a positive vibe.
best
Dave
I have been working on this tonight.
Its a much simplified version of the 26W; a single bridge instead of a double.
It has just caps to feed the following push-pull stage. I would appreciate some advice on their size/ the value of the resistors, snags etc.
I have paralleled the 6N7 to improve the drive and low frequency response, but I'm only using half a 12AU7. Although it is not DC balanced it is AC balanced with those values.
Its getting late here and I'm going to turn in soon, so I'd like some fresh brains to look at this for me to carry on with it tomorrow.
Regarding the compression, 0V gives an output, in the make-up amp, (not shown) of 0.2V but this rises to 20V when the grid is on -26V. This means it can vary 20/0.2=100:1 or 40dB.
I'm knackered!
best
DaveP
Its a much simplified version of the 26W; a single bridge instead of a double.
It has just caps to feed the following push-pull stage. I would appreciate some advice on their size/ the value of the resistors, snags etc.
I have paralleled the 6N7 to improve the drive and low frequency response, but I'm only using half a 12AU7. Although it is not DC balanced it is AC balanced with those values.
Its getting late here and I'm going to turn in soon, so I'd like some fresh brains to look at this for me to carry on with it tomorrow.
Regarding the compression, 0V gives an output, in the make-up amp, (not shown) of 0.2V but this rises to 20V when the grid is on -26V. This means it can vary 20/0.2=100:1 or 40dB.
I'm knackered!
best
DaveP
Is anyone out there able to put the bridge circuit into simulation for me, Spice etc?
The 6N7 data for those parameters is rp 21k, mu 33, gm 1.57.
The 12AU7 at 0V CV is rp 9050, mu 20.5 gm 2.265.
Just need to check what happens when CV is varied from 0 to -30V
Choose Cap values for flat response.
Thanks
DaveP
The 6N7 data for those parameters is rp 21k, mu 33, gm 1.57.
The 12AU7 at 0V CV is rp 9050, mu 20.5 gm 2.265.
Just need to check what happens when CV is varied from 0 to -30V
Choose Cap values for flat response.
Thanks
DaveP
Had a good evening on the bench.
Tried using an interstage to couple the bridge to the next stage but low end losses were too great.
Finally twigged how to do it, thanks to good old Otto Schmitt. He pubished his papers in 37 & 41, probably too late for the early designs made before his dif amp caught on.
I used the other half of the 12AU7, works just as well as the 6N7 for V1.
I have not optimised all the values or the tube for the dif amp but it does work with a 6BK7-B.
The advantage with this set-up is that it does not need an interstage and that the caps can be quite small due to the bootstrapping raising the R values, frequency response is looking good.
No-one seems to have done a simulation yet but it looks like it won't be necessary now.
best
DaveP
Tried using an interstage to couple the bridge to the next stage but low end losses were too great.
Finally twigged how to do it, thanks to good old Otto Schmitt. He pubished his papers in 37 & 41, probably too late for the early designs made before his dif amp caught on.
I used the other half of the 12AU7, works just as well as the 6N7 for V1.
I have not optimised all the values or the tube for the dif amp but it does work with a 6BK7-B.
The advantage with this set-up is that it does not need an interstage and that the caps can be quite small due to the bootstrapping raising the R values, frequency response is looking good.
No-one seems to have done a simulation yet but it looks like it won't be necessary now.
best
DaveP
Things are starting to take shape now.
Regarding the control Amp: I have listed a table of tubes suitable for Octal, B9A & B7G formats. These are all cheaper types less popular with the HIFI guys, but fine for our use. Octals are metal.
I will be breadboarding the various types. The input is basically single ended but you can put any transformer of your choice in front to make it balanced.
The output Amp will be the standard configuration and will follow as soon as the tubes arrive.
I am also waiting for an Edcor 8K:600 transformer to arrive, (shipping costs more than the iron!)
The timing circuit still needs some work but hopefully should be finalised next week.
As soon as all this has come together I will transfer to the lab for the construction of a prototype.
I shall probably make it in octals as a control, if the tubes are vintage then any sound changes from tube changes will be more easily detectable.
I will have to rely those of you with 26C's and 26W's to make comparisons and assesments.
best
DaveP
Regarding the control Amp: I have listed a table of tubes suitable for Octal, B9A & B7G formats. These are all cheaper types less popular with the HIFI guys, but fine for our use. Octals are metal.
I will be breadboarding the various types. The input is basically single ended but you can put any transformer of your choice in front to make it balanced.
The output Amp will be the standard configuration and will follow as soon as the tubes arrive.
I am also waiting for an Edcor 8K:600 transformer to arrive, (shipping costs more than the iron!)
The timing circuit still needs some work but hopefully should be finalised next week.
As soon as all this has come together I will transfer to the lab for the construction of a prototype.
I shall probably make it in octals as a control, if the tubes are vintage then any sound changes from tube changes will be more easily detectable.
I will have to rely those of you with 26C's and 26W's to make comparisons and assesments.
best
DaveP
Jonte Knif
Well-known member
I am confused, are you really going to skip the original PP?
This SE version will not work properly. You will get maximum amount of side chain feed through, and you are going to totally block that differential pair during attack.
In case you find this difficult to believe, I checked in simulation that this is the case, and it is also true what I and PRR said about the original, namely good tube balance matters if side chain action is not to be heard at all.
This SE version will not work properly. You will get maximum amount of side chain feed through, and you are going to totally block that differential pair during attack.
In case you find this difficult to believe, I checked in simulation that this is the case, and it is also true what I and PRR said about the original, namely good tube balance matters if side chain action is not to be heard at all.
Jonte,
Are you sure you understand how this works?
What do you mean by sidechain feed through? Thump?
In a vari-mu, the tubes are difficult to match throughout their range, perfectly matched tubes would not generate any thump. As I understand it, thump is a low frequency intermodulation distortion and you need two mismatched tubes for that. So you need a push pull stage to cancel in a vari-mu compressor, but this is not a vari-mu.
In the bridge, the second tube is just a variable resistance, it is not amplifying signal. It is the equivalent to the second T4 opto unit in the bridge around the VU meter in an LA-2A.
The dif tubes will only block if I make the coupling caps too large, but they can be quite small because of the bootstrapping in the cathode circuit.
best
DaveP
Are you sure you understand how this works?
What do you mean by sidechain feed through? Thump?
In a vari-mu, the tubes are difficult to match throughout their range, perfectly matched tubes would not generate any thump. As I understand it, thump is a low frequency intermodulation distortion and you need two mismatched tubes for that. So you need a push pull stage to cancel in a vari-mu compressor, but this is not a vari-mu.
In the bridge, the second tube is just a variable resistance, it is not amplifying signal. It is the equivalent to the second T4 opto unit in the bridge around the VU meter in an LA-2A.
The dif tubes will only block if I make the coupling caps too large, but they can be quite small because of the bootstrapping in the cathode circuit.
best
DaveP
Jonte Knif
Well-known member
Dave,
I do understand how this works.
Yes, side chain feed through = thumb in my book. This is what you get. It is not about distortion or modulation, it is about different "gains" in any kind of gain reducing elements (or, as in this case, the lack of one) in a topology that has "signal riding on top of DC"
Forget for a moment what this topology is or isn't. You seem to be confused because of this.
Just meditate on what happens when you apply fast compression in a SE topology to a signal *that has DC added to it*. This is fundamental. Where do you get rid of the modulated DC, i.e. side chain signal? It has to happen somewhere if you have gain control signal frequency content overlapping with audio band i.e. pretty much always. Placing some caps somewhere to filter it is not cutting the mustard with fast attack times.
It does not matter whether gain control element is amplifying or not. It does not matter if it is optical, tube or fast midgets with sub miniature potentiometers.
You got to have *two signal paths*, which differ in the phase relation between audio and side chain, what I mean is that in the summing operation following a gain reduction signals add up and side chain is nullified. Just like in the original. It is symmetrical, PP, for a reason, and that reason is not just about removing 2nd harmonic content (which BTW is also very important in Vari Mu circuits) but getting rid of the SC signal.
I can not explain this any better. Of course we could also look at your schema and follow the signal from junction to junction, but that is something you could do and I already did it in the simulation with unforgiving fast attack time.
Best,
Jonte
I do understand how this works.
Yes, side chain feed through = thumb in my book. This is what you get. It is not about distortion or modulation, it is about different "gains" in any kind of gain reducing elements (or, as in this case, the lack of one) in a topology that has "signal riding on top of DC"
Forget for a moment what this topology is or isn't. You seem to be confused because of this.
Just meditate on what happens when you apply fast compression in a SE topology to a signal *that has DC added to it*. This is fundamental. Where do you get rid of the modulated DC, i.e. side chain signal? It has to happen somewhere if you have gain control signal frequency content overlapping with audio band i.e. pretty much always. Placing some caps somewhere to filter it is not cutting the mustard with fast attack times.
It does not matter whether gain control element is amplifying or not. It does not matter if it is optical, tube or fast midgets with sub miniature potentiometers.
You got to have *two signal paths*, which differ in the phase relation between audio and side chain, what I mean is that in the summing operation following a gain reduction signals add up and side chain is nullified. Just like in the original. It is symmetrical, PP, for a reason, and that reason is not just about removing 2nd harmonic content (which BTW is also very important in Vari Mu circuits) but getting rid of the SC signal.
I can not explain this any better. Of course we could also look at your schema and follow the signal from junction to junction, but that is something you could do and I already did it in the simulation with unforgiving fast attack time.
Best,
Jonte
Jonte,
Well you may be right or wrong in your predictions, I'm not convinced either way.
I'm not sure if you think the modulated DC is coming via an unsmoothed sidechain signal or from the DC in the bridge. Because the sidechain DC current flow is so small, the CV cap has a much greater smoothing effect than you might imagine.
There are 1uF capacitors that have to be charged in 1 or 2 mS in the original 26W, so capacitors are not inherently bad, its the size and context which decide whether they work satisfactorily. I shall do my best to see that they are used wisely.
I will build a breadboard version first when some more components arrive and test it thoroughly before committing to metal. The design is certainly not set in stone at this stage.
best
DaveP
Well you may be right or wrong in your predictions, I'm not convinced either way.
I'm not sure if you think the modulated DC is coming via an unsmoothed sidechain signal or from the DC in the bridge. Because the sidechain DC current flow is so small, the CV cap has a much greater smoothing effect than you might imagine.
There are 1uF capacitors that have to be charged in 1 or 2 mS in the original 26W, so capacitors are not inherently bad, its the size and context which decide whether they work satisfactorily. I shall do my best to see that they are used wisely.
I will build a breadboard version first when some more components arrive and test it thoroughly before committing to metal. The design is certainly not set in stone at this stage.
best
DaveP
