"Another Poor Man's" Fairchild 660/670

[lewilson]

Yes, Also you dont want to ground the centertap to the transformer. You need the bias voltage . unless you do something different with the voltage at the cathodes. At least with the fairchilds schematic. Also if you want to do de-essing it might be better to change the response of the sidechain amp later in the circuit, after a gain stage.  you dont want to load down or change the response of the signal amp. Just some thoughts.
  From building a fairchilds I learned that the control amp is really the most critical part of the compressor, If theres noise there or bad hum or bad feedback it really messes with the signal amp.

 

[MeToo2]

Also you dont want to ground the centertap to the [input] transformer.

May I kindly enquire "why not?"

Isn't this basically what happens when the unit is first turned on, or if  there is no compression taking place (output signal below threshold)?

Won't the signal amp then simply auto bias at the point where Vg = - Ic * 8 * (620+~200/2+33)?

Yes I realise this will turn it into a near linear amplifier, but that was the idea, so that you could temporarily disable the compression and just listen to the uncompressed signal but still going through the signal amp.

I saw this applied on the UREI 176 which has a very similar topology. It is the switch marked S4 "limiting action" which bypasses the capacitor controlling the compression. http://www.waltzingbear.com/Schematics/Urei/UA_176.htm

Also if you want to do de-essing it might be better to change the response of the sidechain amp later in the circuit

Thought about that, but the problem with that is that most filters / tone controls are 3 port i.e. one common port (to ground), which messes up the nice symmetry of the side chain amp if you placed the filter there, so you'd have to build a stereo filter. That was why I suggested putting it on the output of the signal amp but before the side chain amp between the two transformers (which should not mind having one common leg between them as generally they are floating wrt dc), but I appreciate it may load the signal amp additionally, although there are 330 ohm resistors in there at that point.

So indeed you could end up putting in a buffer for the side chain filter to drive the tone stack or another inter-stage transformer which makes it more complex than it needs to be.... maybe it's simply just a feature that is not worth implementing at all on this vari mu, as you're in danger of getting too far away from the original concept.

 

[strangeandbouncy]

Hi,


  i am following this with great interest. Can someone explain why there is no DC bias on this? The attached picture shows it on the Fairchild, adjusted by R242. Have i missed something?


    Many thanks,



      ANdyP

 

[MeToo2]

Hi,


 i am following this with great interest. Can someone explain why there is no DC bias on this? The attached picture shows it on the Fairchild, adjusted by R242. Have i missed something?


   Many thanks,



     ANdyP

Good question. I think I know the answer. But maybe I'm wrong.

In the Fairchild, the resistor R242 controls the voltage at the bottom of the rectifier. Remember that it is driven by a negative voltage. More negative switches off the gain in the signal chain. The control voltage relative to ground is the sum of the voltage across multiple components i.e. capacitor C215 + (C209/R237 in parallel with whichever other mode is switched in via S202). The voltage across C215 is unaffected by the rectified signal in the side chain, and thus provides the DC bias for the signal chain. Also bear in mind that the cathodes of the signal chain in the Fairchild are actually connected to the -6.2V power supply via the 680 ohm resistor. Since the resistor R242 is adjusted for -6.2V on the centre tap, that also gives equivalent of Vg + V cathode resistor = 0V on those tubes.

In the Rotheu's filter model (see page 4), the control voltage is generated by a single capacitor relative to ground. This side chain capacitor discharges to ground via one of the selected release resistors. The quiescent state therefore is that the control voltage applied to the grids via the 1M resistor is 0V. The signal stage therefore self - biases around where the drop across the cathode resistor matches the grid voltage that would produce that amount of current. Hence my assertion that Vg = - Ic * 8 * (620+~200/2+33) [*8 because there are 8 tubes providing current to a common cathode resistor] Looking at my tube curves for a 6bc8, that should be around Vg= -4V and quiescent current of .66mA per tube

In short it's self biasing relative to ground, whereas a Fairchild apparently doesn't and needs to have that -6.2V reference adjusted manually.

 

[strangeandbouncy]

..... and I thought that that bias set the threshold(?), or it would compress at even very low level, right down to nothing . . .

 

[MeToo2]

..... and I thought that that bias set the threshold(?), or it would compress at even very low level, right down to nothing . . .

Not according to the diagram I have. Again I'm prepared to be corrected, but I don't see how a floating DC voltage on the bottom of a circuit connected to a transformer could affect threshold.

I thought the DC threshold was set by R117 / R217 in the side chain amp of the Fairchild. This effectively clips the bottom off the side chain signal if it is too low, resulting in no feed to the rectifier, no capacitor charging, no control voltage increase (and thus no compression).

I haven't built this but I suspect the signal at that point would look something like this during correct operation (class AB under biased)
http://www.ionaphysics.org/lobby/robotics/Textbooks/liechtml/Exper/05324.png

In Rotheu's circuit the equivalent is B100K at the entrance to his side chain amp.

 

[strangeandbouncy]

I am really not qualified to answer, and i would hate to be making a chump of myself, but how else is the threshold of compression set? as I understand, no compression will occur 'til the SC voltage exceeds -6.2V.


      No?. Otherwise it would compress at ANY level wouldn't  it?



    So soory if I am wrong!


  ANdyP

 

[lewilson]

Check the schematic. The voltages from the cathode to grids are not 0vdc. Maybe some very small coupling caps would work to make a 660 a de-esser. You could have them on a switch

 

[MeToo2]

Check the schematic. The voltages from the cathode to grids are not 0vdc. Maybe some very small coupling caps would work to make a 660 a de-esser. You could have them on a switch

Just to be clear we are talking about proposed modifications to the schematic for this Poor Man's Fairchild 660/670
http://i572.photobucket.com/albums/ss163/rotheu/Fairchild660Mod2.jpg and this timing chain
http://i572.photobucket.com/albums/ss163/rotheu/Fairchild660ModTiming.jpg: not whether you could do this mod with the original Fairchild.

When talking about a DC voltage you have to also have a reference. You can't just say 0V DC.

Let's see if I can convince myself and you in the process.

The only DC reference for the grids of the signal chain in Rotheu's circuit is the common voltage of the top of the attack capacitor (to signal ground) and the top of the release resistor, again to signal ground. Otherwise we're talking 10M+ parasitic leakage resistors. When the attack capacitor discharges, the grids should sit around 0V compared to signal ground.

When the unit is initially turned on the valves will see 0V on their grid via the 1M resistor and the release resistor, 0V on their cathode (no plate current), and 100V on their plate, so they'll begin to conduct and try to force around 16mA of plate current to flow (from the 6bc8 curves). If all 8 tubes fed 16mA that would raise the cathode voltage by 8 * 16mA * 100+620+33 ~ 100V meaning the grid to cathode voltage would be -100V, turning off the valve completely, so that isn't going to happen.

The effect of the plate current is to cause cathode current, which in turn develops negative feedback across the tail resistor: effectively a load line of approx 6K (because of 8 tubes forcing plate current through the common tail resistor of 753 ohms).

The effective plate voltage is also 100V (power supply) - the voltage across the cathode resistor = 100 - Ip * 6K but this drop is going to be pretty negligible as a first approx. so the quiescent state should be somewhere near where 6K*Ib crosses the grid voltage versus plate current curve of the 6bc8 with Vplate = 100V

So rather than reaching the full 16mA of plate current, the equilibrium will be reached at somewhere around 0.66mA of plate current = 0.66mA cathode current per tube or a total of around 5mA generating ~4V across the common cathode resistor.

Relative to signal ground the cathode will sit between  0.66mA * 8 * 753 ohms = 4V
Relative to signal ground, the grid will sit at about 0V.

This will maintain a grid to cathode voltage of around -4V (which of course magically corresponds to around .66mA of plate current per tube at Vplate around 100V according hte the 6bc8 curves) thus maintaining the equilibrium.

.... classic self biasing

So the basic point of all of that is to show that the grid voltage in the quiescent state is at 0V relative to signal ground, and so the grid could be safely connected to signal ground in THIS CIRCUIT (not in the original Fairchild) and thus to turn off compression.

am really not qualified to answer, and i would hate to be making a chump of myself, but how else is the threshold of compression set? as I understand, no compression will occur 'til the SC voltage exceeds -6.2V.

    No?. Otherwise it would compress at ANY level wouldn't  it?

   So soory if I am wrong!

  ANdyP

Again, you have to talk about a voltage and a reference point. The absolute level of the 6.2V and the adjustment of R142/242 has nothing to do with determining the threshold of the Fairchild IMHO. IMHO The side chain rectifier is floating compared to its input signal (due to the inter stage transformer T104/T204) and will switch at the normal ~0.6V of silicon diodes. That voltage is then added to the control voltage by charging C109/C209. The fact that the whole circuit is raised 6.2V relative to signal ground via C215/C215 R142/R242 in the quiescent state is neither here nor there when it comes to determining the threshold of compression (but it does determine DC biasing of the signal amp, so yes you're right in a sense that if the control voltage remains at -6.2V = no charge on C109/C209, then nothing gets compressed, but that is a symptom, not the cause.)

The threshold for DC compression is determined at the input to the side chain amp, not at the output. Have a good look at the biasing arrangement of the 12ax7 of the side chain amp. Draw out just half of one stage, i.e. the upper half of transformer T103, the ac pot R115a, R116 270K, R117 100K pot, 1M R118, V105a (1/2 12AX7) R119 18K & R120 100K. Set the AC pot R115 half way. Turn the DC threshold pot R117 fully left, fully right, and then half way and check/predict what happens for no signal input, and then add on small AC signal.

Is the 12AX7 running in class A, AB, B or is it actually completely under biased in Class C, and only switches on once a significant ac signal arrives?

[
Hint 1. The AC pot has no DC across it. Both sides are at 0V relative to signal earth as the transformer secondary has low impedance at DC. On the other hand, the DC pot has negligible AC across it, as it is connected to the centre tap,
Hint 2. After much crumpling of paper, I came up with grid - cathode bias voltages of -16V, -3.6V & -13V respectively assuming a linear pot and applying Kirchoff's laws.
Hint 3. Check out the 12AX7 data sheet here http://lib.store.yahoo.net/lib/thetubestore/JJ-ECC83-S.pdf and look carefully at the graph of plate current against Ug....
]

 

[lewilson]

I was in reference to a real 660.  Im loosing track of this conversation.
 Its  a diy so build one and try your ideas.
After seeing this schematic it seems the grids are just floating there on top of the bridge rec. when there is no signal. Maybe in missing something.

 

[MeToo2]

I was in reference to a real 660.  Im loosing track of this conversation.
 Its  a diy so build one and try your ideas.
After seeing this schematic it seems the grids are just floating there on top of the bridge rec. when there is no signal. Maybe in missing something.

The release resistor connected to signal ground in the modified timing chain.....

I am thinking on building a stereo pair, but I prefer to think first, then do. Just my way. Different people learn in different ways.

 

[lewilson]

It looks like the release resistor changes as the different release times are selected.
 As long as the transition from self bias to sidechain amp bias is smooth and the diodes switching on doesn't create an audiable bump, I guess it could work. Do the signal amp tubes operate in class "A" for low signals and then switch to  class a/b as the signal gets stronger? It looks like the creator has made one.

 

[rotheu]

Hi everybody,

very interesting discussion.

Is it worth putting in a 2 way switch to disconnect the side chain output from the centre tap of the input transformer and instead ground it (to temporarily turn off compression, so you can easily hear / monitor the amount of mojo being applied)?

That will work, but I personally like to have a hard bypass to see what it's really doing and compare it with original signal, unless you want to use it as a linear amp. But then you can just set the AC threshold not to compress.

By changing the gain of the side chain amp, would it be possible to change the compression ratio in a meaningful way?

[edit]
oops the side chain input pad pot is already there. silly me

The reason I exactly duplicated the front end (12AX7) of the original Fairchild is to have the DC threshold control (100K pot), which in conjunction with AC threshold will allow to change the ratio from 2:1 to somewhere between 20:1 and 30:1. It also changes the sound of compression. IMHO that part of the circuit and timing network have by far the most influence on the sound. Much more than all "bottles and iron"

Hi,


  i am following this with great interest. Can someone explain why there is no DC bias on this? The attached picture shows it on the Fairchild, adjusted by R242. Have i missed something?


    Many thanks,



      ANdyP

Good question. I think I know the answer. But maybe I'm wrong.

In the Fairchild, the resistor R242 controls the voltage at the bottom of the rectifier. Remember that it is driven by a negative voltage. More negative switches off the gain in the signal chain. The control voltage relative to ground is the sum of the voltage across multiple components i.e. capacitor C215 + (C209/R237 in parallel with whichever other mode is switched in via S202). The voltage across C215 is unaffected by the rectified signal in the side chain, and thus provides the DC bias for the signal chain. Also bear in mind that the cathodes of the signal chain in the Fairchild are actually connected to the -6.2V power supply via the 680 ohm resistor. Since the resistor R242 is adjusted for -6.2V on the centre tap, that also gives equivalent of Vg + V cathode resistor = 0V on those tubes.

In the Rotheu's filter model (see page 4), the control voltage is generated by a single capacitor relative to ground. This side chain capacitor discharges to ground via one of the selected release resistors. The quiescent state therefore is that the control voltage applied to the grids via the 1M resistor is 0V. The signal stage therefore self - biases around where the drop across the cathode resistor matches the grid voltage that would produce that amount of current. Hence my assertion that Vg = - Ic * 8 * (620+~200/2+33) [*8 because there are 8 tubes providing current to a common cathode resistor] Looking at my tube curves for a 6bc8, that should be around Vg= -4V and quiescent current of .66mA per tube

In short it's self biasing relative to ground, whereas a Fairchild apparently doesn't and needs to have that -6.2V reference adjusted manually.

You are exactly right, the cathodes are at around +4V relative to the ground.

Original 670 has -6.2V bias on the grids, but 660's grids are 0V relative to the ground. That cold be the reason for signal amp gain with no compression 670=7db, 660=16db? Correct me if I'm wrong, but I think the reason for negative bias in 670 is to be able to zero the gain reduction meter individually on each channel using that very 2.5K biasing pot on the front panel(zero adjustment). Just my guess. ??? Please, correct me if I'm wrong.

It looks like the release resistor changes as the different release times are selected.
  As long as the transition from self bias to sidechain amp bias is smooth and the diodes switching on doesn't create an audiable bump, I guess it could work. Do the signal amp tubes operate in class "A" for low signals and then switch to  class a/b as the signal gets stronger? It looks like the creator has made one.

Yes, release resistor changes as the different release times are selected, as it does in the original 660. Otherwise, how can you change release time independent from the attack time?
I still have the prototype on my desk (don't have time to build it into chassis and add second channel ). There are no audible bumps, it runs very smooth and have very transparent compression, unless you squash things hard with very short attack/release settings. Then it gets aggressive. Very cool on the drum room mics.

 

[MeToo2]

Hi everybody,

very interesting discussion.

....

You are exactly right, the cathodes are at around +4V relative to the ground.

Nice to know after some hard slog that despite what some people write that this circuit actually is predictable and it isn't all magic

Really appreciate being able to talk to you on this! Thanks once again.

Original 670 has -6.2V bias on the grids, but 660's grids are 0V relative to the ground. That cold be the reason for signal amp gain with no compression 670=7db, 660=16db? Correct me if I'm wrong, but I think the reason for negative bias in 670 is to be able to zero the gain reduction meter individually on each channel using that very 2.5K biasing pot on the front panel(zero adjustment). Just my guess. ??? Please, correct me if I'm wrong.

To be clear: I'm referencing the original Fairchild 670 schematic in this post, not your redesign.

I have traced out the circuit quite carefully and redrawn it retracing everything back to a common point in the power supply and signal ground. Could have made a mistake so would appreciate if you'd check.

The necessary elements are
V201-204a V201b-204b
Balance pots R205a & b "BAL ADJ"
Secondary of T201.
R237 (release)
The now famous R242 & C215.
Power supply resistors R301 R302 R303.
Meter switch S201a/b

Next tracing everything back to the -6.2V and ground at the power supply.

According to my schematic, R242 is adjusted as a potential divider to split -19V and 0V to give approx -6.2V at the bottom of R237. Since R237 + T201 is negligible resistance compared to the load, the grids of V201-204 will sit at -6.2V in the quiescent state relative to signal ground (yes, via a high impedance feed, but that is irrelevant as there is only tiny leakage to signal ground)

If you set the balance pot R205a/b centre i.e. both side to wiper to 50 ohms, then for DC they do exactly what they say on the tin i.e. they do not contribute anything to the _relative_ biasing of the two halves, as any plate current through both legs V201-V204 (IBias_total) will split equally flowing to the 0V and -6.2V power supply respectively via the balance pot. But don't forget there'll also be a current across R205a/b between the 0V and -6.2V supplies. NB this -6.2V supply is hard set in the power supply chain and is low impedance via L302 to signal ground for DC.

That means you effectively have (50/2+680) * Ibias_total - (50/100* 6.2) at the top of R203 and R204 (relative to signal earth & balance pot set in the middle)

The relative grid to cathode voltage is Vgrid - Vcathode =  (V@R242 = -6.2) - ( (50/2 +680) * IBias_total - 50/100*6.2)

or simplifying -Vgrid effective = -3.1 + 705 IBias_total [note this is why I make it 3.1 instead of 6.2]

At the end of all that, the grid to cathode voltage will self bias to produce Ibias that balances the potential difference at the top of R203.

Assuming equal sharing across all 4 of the tubes that gives IBias_per_tube = IBias_total / 4

or Vgrid effective =  705*4 * IBias_per_tube - 3.1 [assuming R242 set to -6.2V, otherwise V@R242 - 3.1]

So you're equivalent circuit is self biasing a single tube relative to a variable supply with a default of -3.1V on a 2820 cathode resistor. That won't be quite equivalent to self-biasing to 0V.

Looking at the curves for a 6386 with a plate around 230-240V (using 250V curve) that'd suggest to me around 9mA Ibias and Vg_bias of -18.8 for the suggested position of R242 as 6.2V, instead of 8mA and Vg_bias of -19.5  for if the bias and zero pots didn't exist.

In summary (and IMVHO): this is not a great deal of difference considering the potential inaccuracies of the curves and valves, but may be enough to change the gain of the signal amp slightly for no compression.

I certainly agree that resistor r242 directly helps control how much total bias is being applied to both halves of the signal amp. That will of course also alter the gain with no compression. But I'm not sure how significant that'll be at this point of the curve if you keep it set around -6.2V. Obviously if you turn it all the way towards -13.1V it'd have a bigger impact (approx 10.5mA quiescent) or all the way to 0 (approx 7mA quiescent) which in turn should have an effect on mu going from around 8-10. My guess is that this biasing is primarily so you can force the signal amp very slightly into class AB operation and thus prevent any cross over distortion. Setting this pot R242 may also have a use controlling the overall gain of the Fairchild, but nowhere near as much as the 50K input pad in your modified circuit.

The "balance" setting on the meter M201 (position 1 & 3) appears to be measuring the current draw of each of the two halves of the channel for +ve & -ve signals respectively. So that is how to adjust the balance pot to ensure both sides of the signal are being amplified the same amount and there is no distortion. Whilst the "zero" on the meter (position 2) appears to be measuring the average of the two signals across the 30 ohm resistors in parallel. Since the meter M201 is a VU meter rather than a pure DC voltmeter, you could be correct that r242 is also affecting the gain enough to ensure 0VU out, but I'll freely admit I'm at the limit of my knowledge.

[edit] Just checked and the output and input transformers are apparently wound 1+1:9+9, so with the switch in postion 2 the voltage across the meter Vm is Iboth_channels *15. This current splits over 2 windings in class A operation, and reduces by the square of the winding ratio. Equivalent current on each output leg is I/2/(9/2)^2. Since each leg sees half of the 600 ohm output load Vo = 300/(9/2)^2 * I/2 * 2 windings. Vm/Vo = 300/(4.5^2*15) =0.99. In other words, the voltage measured on the VU meter with switch position 2 is equal to the voltage of the 2 output windings connected in series assuming a perfect transformer and a 600 ohm load.

BTW S301 seems to be a routing circuit to allow either L+R and L-R or L & R channels separately to be compressed. Presumably because that's how stereo record grooves used to be cut (also FM radio) and this was a mastering compressor: http://www.youtube.com/watch?v=mu0zP5s_PSo

So yes 0VU makes complete sense.
[end edit]

 

[lewilson]

Yes, in positon "a" the meter shows a level reprisenting the current fram one half the signal output transformer position "b" shows the other side. You can use the balance control to correct to a certain extent the  the differences in the tubes and output transformer windings. the center position shows gr.

  I was wondering why the 660 had more gain than the 670, Still not real sure.
Mine has about 12 db of gain . 

 

[MeToo2]

Yes, in positon "a" the meter shows a level reprisenting the current fram one half the signal output transformer position "b" shows the other side. You can use the balance control to correct to a certain extent the  the differences in the tubes and output transformer windings. the center position shows gr.

  I was wondering why the 660 had more gain than the 670, Still not real sure.
Mine has about 12 db of gain .  

I haven't been able to find a 660 schematic. Maybe someone can point one out to me. Rotheu already stated that the 660 self-biased relative to 0V, whereas the 670 clearly doesn't: it's adjustable around -3.1V (with a range of around +3 to -11 on the extremes of R242). That would largely explain the mechanics of any gain difference, together with different wiring of the trannies.

But why would they do this? This kit was obviously made by some very smart people, so they must have had a reason.

One possibility is that the 670 was specifically intended for use in mastering stereo vinyl, and the 660 was mono.

It is simple to match the appropriate total gain on a mono recording after processing with a 660. You just put another gain pot or preamplifier in series with the compressor to correct any shortfall or cut any over amplification to make sure the cutter is optimized. Since you've got tubes there, then why not give it a little bit extra gain than needed so you just need to attenuate a bit later via a passive 600 ohm matched network?

But since the original idea of the 670 was almost certainly to be able to optimize the groove depth for the stereo vinyl cutter (and not for use as an effects box) it'll normally be compressing L+R and L-R in the two channels.

Obviously that would explain why you need a feature for changing the individual channel gain of the two channels (even slightly) via manually adjusting the bias via R142/242. If you mismatch the gain balance between two (L+R) and (L-R) stages, when combining back to separate L & R channels you'd mess up the stereo separation and imaging, as you'd not get proper cancellation e.g. a 5% gain mismatch of the two compressor channels without compressing would give you ((1.05L/2+1.05R/2) + (1L/2-1R/2)) = 1.025L +0.025R on the left channel and ((1.05L/2+1.05R/2) - (1L/2-1R/2)) = 0.025L + 1.025R on the right channel, plus any compression.

And of course the L+R channel would be running at around approx 2 * L for centre-panned drums and bass, so you need to back off the total gain a bit (6db) to prevent going into clipping / distorting the output for the same input. Plus two centre-tapped halves of an input or output tranny connected in series in the L+R L-R matrix gives you 6dB gain reduction (1/2 windings gives 1/4 voltage * 2 trannies = 1/2 gain). So all other things being equal, you have to lose at least 6dB somewhere when implementing the L+R matrix.

I'm guessing (and it is clearly a guess) that this is what they'd be afraid of, and so they reduced the overall gain and added a fine gain adjustment on the 670. Comparison with a 660 schematic could help prove or disprove that.

 

[strangeandbouncy]

Hi Guys,

  That is all really interesting indeed. Thanks very much.



  ANdyP

 

[rotheu]

Hi,

this is 660 schematic. When I talked about zero adjustment I didn't mean 0VU or actual level, which will certainly slightly affected by it, because it's changing the current through the tube. It's the relative "ZERO" for gain reduction meter. When the meter switch is in position 2 called "ZERO", the 2.5K biasing pot is used to adjust that relative zero on the meter on 670. On 660 that pot (R41) is in the power supply circuitry and it doesn't use negative bias voltage.

 

[MeToo2]

Hi,

this is 660 schematic. When I talked about zero adjustment I didn't mean 0VU or actual level, which will certainly slightly affected by it, because it's changing the current through the tube. It's the relative "ZERO" for gain reduction meter. When the meter switch is in position 2 called "ZERO", the 2.5K biasing pot is used to adjust that relative zero on the meter on 670. On 660 that pot (R41) is in the power supply circuitry and it doesn't use negative bias voltage.

Not sure I understand your comment about R41. I'll redraw and try to post a cleaned up copy of the biasing scheme as it is difficult to read the jpeg.

But my guess is that R41 in the 660 is the equivalent of R313 in the 670 and adjusts the reference voltage fed back into to the HT regulator V10 / V9. i.e. it adjusts the regulated +240V supply to the amps. I don't think it affects any negative voltages on the signal amp.

According to my first look, the negative voltage in the 660 looks to be fixed, and is the the voltage developed across CR5 followed by the RC filter formed by R50(?) C16 R48 C(?)

So summarising the differences between the 660 & 670, 660 first then 670

660 transformers fixed in L & R mode only with both windings in series. 670 switchable via a matrix controlled by S301 for either L & R mode or L+R L-R mode. That explains right there the need for the reduced gain on the 670 compared to the 660 = to avoid overloading one signal amp when processing L+R.

On the 660 there is only balance to control biasing. Biasing on the 670 has both variable amount of biasing (= additional gain control) + balance.

(no equivalent in the 660 bias circuit) R142/R242 on 670 = overall bias of signal amp (also sets gain)
R3a R3b on 660 equivalent to R105a R105b on 670 = balance
R4 & R5 on 660 equivalent to R103 R104 on 670 = cathode resistors of the signal amp - set basic gain and biasing of the signal amp
R32 on 660 equivalent of R137b on 670 = release resistor
C7 on 660 equivalent of C109 on 670 = compression portion of control voltage
(no equivalent in 660) C115 on 670 = stabilising offset of overall bias of signal amp
R41 on 660 equivalent of R313 on 670 = adjust 240V for side chain amp

So it looks to me like the original 660 signal amp may be slightly differently biased compared to your mod although you may have already adjusted for that as it is difficult to read resistor values on the 660 schematic. [flame suit on] In the big scheme of things we're not even going to use the same valves either, so as long as the overall gain and output impedance comes out about right it shouldn't matter too much [flame suit off].

The more I look at this, the keener I am to build and see how it really runs

 

[lewilson]

interesting, I build my 660 from the 670 schematic. I could never find a 660 schematic.
So the grids of the signal amp are not fed a negitive voltage. They are referenced to ground through r32 in parallel with whatever happens to be selected for a time constant.
Cathodes are balanced with a negitive voltage tapped from the fixed bias supply for the output tubes.
Also bypass caps on the signal amp are very small compared to th 4uf that the 670 has.