The R20 loop to the driver cathode has potentially higher gain if not loaded down by R16. So raising R16 increases NFB overall. The question should be why did RCA use the "shunt Schade" loop at all. It's counter productive.
The only reason I can imagine is that with the restriction of finite loop gains, does each N Fdbk loop type leave some curvature residual of opposite curvature, that can be adjusted so that they roughly cancel. Even so, I doubt it would work beyond 3rd harmonic. (ie. trying to cancel one error with another on each side of the P-P could fix odd harmonics) Considerable tweaking in Spice or the actual circuit would be required to optimize.
This is a similar situation to the Citation II, where two different feedbacks come to the same driver grid, both cannot be satisfied completely, but possibly the residual errors from each can be tweaked to mostly null each other out at least.
Such schemes typically would increase higher harmonics since the residual error curves are usually more curved at opposite ends and won't fully cancel.
The only reason I can imagine is that with the restriction of finite loop gains, does each N Fdbk loop type leave some curvature residual of opposite curvature, that can be adjusted so that they roughly cancel. Even so, I doubt it would work beyond 3rd harmonic. (ie. trying to cancel one error with another on each side of the P-P could fix odd harmonics) Considerable tweaking in Spice or the actual circuit would be required to optimize.
This is a similar situation to the Citation II, where two different feedbacks come to the same driver grid, both cannot be satisfied completely, but possibly the residual errors from each can be tweaked to mostly null each other out at least.
Such schemes typically would increase higher harmonics since the residual error curves are usually more curved at opposite ends and won't fully cancel.
Last edited:
Hey smoking-amp, thanks for your thoughts. Perhaps that's why RCA added that 1.3M resistor from driver tube cathode to grid. I think I don't want to go down the path of nested NFB loops, unless I build that RCA 50-Watt Amplifier exactly as shown in the tube manuals.
I wish I understood more about the workings of negative feedback as applied here. I don't have a solid grounding in mathematics, which puts me at a severe disadvantage. I have to rely heavily on computer simulations in the design phase.
In an attempt to simplify things and reduce the parts count, I'm looking at the "E-Linear" way of connecting a feedback loop from the screen taps on the output transformer to the control grids on the output pentodes. I have some 35k ohm wirewound resistors, so I used that value in the simulation, figuring they would get me to the desired DC conditions. Here's what happened:
The DC conditions came out much the same as before. The 6EJ7s have 195V on their plates, Ip = 6.6mA. There's 170V on their screen grids, Ig2 = 2.7mA.
The gain comes out to 13X, 22dB (2.83V peak out for 216mV peak in). That means the amount of NFB is about the same as for the version with the nested NFB loops (post #37).
The predicted THD comes out very low, but is all H3 (with a trace of H5). In real life I'm sure H2 and H4 would be present due to tube variation, etc.
What's attractive about this is that it has a better chance of fitting on the small PCB required to fit in a Dyna ST70 chassis.
Getting the poles and zeroes right might take some doing. There is a bump up in the frequency response below 20Hz, which I was able to smooth out by reducing the value of C5 and C6 way down to 33uF (from 220uF originally). I'll need to work on that some more...
I'm also worried about the choice of operating point for the 6EJ7 LTP. According to the Philips EF184 data sheet, I'm biasing the driver stage very cold:
Does that look wrong?
My understanding is that a good operating point for a pentode puts the load line with its left-hand end up near the 'knee' on the Vg1 = 0V plate curve. However in this case that would mean a really low value of Rload. too vertical of a load line, therefore much reduced voltage swing. Am I starving the 6EJ7 like this?
That's exactly what I do not want.
I wish I understood more about the workings of negative feedback as applied here. I don't have a solid grounding in mathematics, which puts me at a severe disadvantage. I have to rely heavily on computer simulations in the design phase.
In an attempt to simplify things and reduce the parts count, I'm looking at the "E-Linear" way of connecting a feedback loop from the screen taps on the output transformer to the control grids on the output pentodes. I have some 35k ohm wirewound resistors, so I used that value in the simulation, figuring they would get me to the desired DC conditions. Here's what happened:
The DC conditions came out much the same as before. The 6EJ7s have 195V on their plates, Ip = 6.6mA. There's 170V on their screen grids, Ig2 = 2.7mA.
The gain comes out to 13X, 22dB (2.83V peak out for 216mV peak in). That means the amount of NFB is about the same as for the version with the nested NFB loops (post #37).
The predicted THD comes out very low, but is all H3 (with a trace of H5). In real life I'm sure H2 and H4 would be present due to tube variation, etc.
What's attractive about this is that it has a better chance of fitting on the small PCB required to fit in a Dyna ST70 chassis.
Getting the poles and zeroes right might take some doing. There is a bump up in the frequency response below 20Hz, which I was able to smooth out by reducing the value of C5 and C6 way down to 33uF (from 220uF originally). I'll need to work on that some more...
I'm also worried about the choice of operating point for the 6EJ7 LTP. According to the Philips EF184 data sheet, I'm biasing the driver stage very cold:
Does that look wrong?
My understanding is that a good operating point for a pentode puts the load line with its left-hand end up near the 'knee' on the Vg1 = 0V plate curve. However in this case that would mean a really low value of Rload. too vertical of a load line, therefore much reduced voltage swing. Am I starving the 6EJ7 like this?
Last edited:
In simulation, with 100uF cathode resistor bypass caps C5 and C6, frequency response at [OUT] looks like this (with 2.83V peak across 4 ohms):
As you can see, there's a +0.8dB boost at around 6Hz (way down there). It's not a big boost, but it's definitely not desirable.
Reducing the value of C5 and C6 to 33uF, the frequency response now looks like this (note the scale has changed a little):
That looks better, right? Now there's practically no LF bump at all (just very slight now).
However, the NFB loop is boosting the bass frequencies in a big way:
That's a predicted +9.5dB boost below 10Hz. That can't be good.
By changing the values of the interstage coupling caps C3 and C4, and changing the values of the output stage cathode bypass caps C5 and C6, I can change the shape of that bump, and move it up in frequency a bit. But I can't reduce the amplitude of the bump very much. Could this bump be caused by the input transformer?
Trying to place a high-pass filter before the input transformer, I found the capacitance interacts with the transformer's input inductance in unwanted ways (increases the Q of the LF bump and moves it up or down in frequency, but doesn't get rid of it). So that's not an easy fix.
I'm using a model of an Edcor 10k:15k input transformer because that's the only model I have available right now. I'll need to measure the Jensen...
As you can see, there's a +0.8dB boost at around 6Hz (way down there). It's not a big boost, but it's definitely not desirable.
Reducing the value of C5 and C6 to 33uF, the frequency response now looks like this (note the scale has changed a little):
That looks better, right? Now there's practically no LF bump at all (just very slight now).
However, the NFB loop is boosting the bass frequencies in a big way:
That's a predicted +9.5dB boost below 10Hz. That can't be good.
By changing the values of the interstage coupling caps C3 and C4, and changing the values of the output stage cathode bypass caps C5 and C6, I can change the shape of that bump, and move it up in frequency a bit. But I can't reduce the amplitude of the bump very much. Could this bump be caused by the input transformer?
Trying to place a high-pass filter before the input transformer, I found the capacitance interacts with the transformer's input inductance in unwanted ways (increases the Q of the LF bump and moves it up or down in frequency, but doesn't get rid of it). So that's not an easy fix.
I'm using a model of an Edcor 10k:15k input transformer because that's the only model I have available right now. I'll need to measure the Jensen...
If you don't need all the gain, you can reverse the Edcor so it's step-down, 15K:10K, and see what happens.
How did you connect the suppressor and screen grids? To cathode to plate? With a resistor?
Steve
Hmmm... Good idea. So I tried it in simulation, and...
Slightly less gain.
THD predictions remain about the same.
With 100uF cathode bypass caps the infrasonic boost is lessened, but it's still there.
I spent a few hours on a PCB layout to fit the Dyna ST70 chassis. It's a pain. The PCB is small, and I need to leave room for ventilating the tubes and the plate load and screen grid load resistors. The 4.7uF 400V screen grid bypass caps (one per channel) also need to fit on there somehow. Perhaps I could mount the 4.7uF caps on the underside of the PCB, inside the chassis. They'd get hot, though. That's probably not a good idea.
Unfortunately, now there's no room for the transformers on the PCB. I'm worried that if I mount those inside the chassis, they'll pick up hum from the power transformer, choke, and AC heater windings.
Layout is going to be difficult.
Loading to the knee is for maximum power output, not applicable for small signal voltage amplification. The Philips curve is misleading, add in the 2.5W dissipation limit and you will see that there is a large zone that exceeds this. Operating point looks fine.
Thanks!
It makes sense that the Philips curve shows a lot of unusable territory, because the operating points I chose are at about 3/4 of the max plate and screen dissipation ratings of 6EJ7.
This amp designing thing can be difficult.
I'm finding that I don't have enough room in the space allotted for the PCB to fit the big plate and screen resistors, the big screen bypass caps, and the input transformers, all at the same time. Something's gotta give...
Now I have a version that uses 2.2uF 400V caps for the screen grid bypass (I have some metalized polypropylene ERO box caps that should do).
I'm also back to the Long-Tailed Pair (LTP) with a negative voltage supply and a CCS for the tail. Should I do that with a plain old LM317? Maybe an LM1085? I have a 12V-0-12V 450mA transformers I can throw at this. I could use the whole secondary with a diode bridge for 24VDC at 225mA, which should be plenty. Or do I really have to use a DN2535?
Everything is a compromise.
Everything is a compromise.
Everything is a compromise.
I'm finding that I don't have enough room in the space allotted for the PCB to fit the big plate and screen resistors, the big screen bypass caps, and the input transformers, all at the same time. Something's gotta give...
Now I have a version that uses 2.2uF 400V caps for the screen grid bypass (I have some metalized polypropylene ERO box caps that should do).
I'm also back to the Long-Tailed Pair (LTP) with a negative voltage supply and a CCS for the tail. Should I do that with a plain old LM317? Maybe an LM1085? I have a 12V-0-12V 450mA transformers I can throw at this. I could use the whole secondary with a diode bridge for 24VDC at 225mA, which should be plenty. Or do I really have to use a DN2535?
Everything is a compromise.
Everything is a compromise.
Everything is a compromise.
Indeed this is what I felt (I say felt because I have no way to measure that) when I tried to apply feedback loops from output tubes' plates to driver cathodes in the baby huey circuit.
Not sure what kind of transformer is the Edcor but most input transformers are shielded with a mu-metal round can so hum should not be a problem. If you have transformers with an octal plugs, like the Altec 15335A or Ampex bridging transformers, you can plug them into Dynaco's front octal sockets. I know, it might look weird....
The amp's octal sockets were removed long ago. Also, I don't have the transformers in the octal plug package.
The Edcor trannies are open bobbin, so they're not going to get used in this project. I have them in the simulation because I made a model for them a while ago, so I can use that for the sims.
I do have a pair of Jensen JT-11P4-1, which are in little mu-metal cans with L-brackets for mounting to a chassis. I really don't want to drill that chassis (it's steel, and pretty heavy), but since the transformers are in a physically small package, I might be able to rig something up.
In the meantime, I modeled a version using an LM1085 as a CCS, with -24VDC for the CCS IC.
Gain is about the same.
I re-jiggered the screen grid supply for the 6EJ7s so I could use a smaller value bypass cap (2.2uF instead of 4.7uF). I have some ERO MKP box capacitors that will fit well.
This version's THD is actually lower than the one with the input transformers.
The 32VDC supply is my estimate of the raw DC I'll get from a 25.2VAC 225mA winding drawing 35mA. That DC supply is going to have to be mounted somewhere inside the chassis. I'm not looking forward to struggling with that, but I think I can make it work.
I know some people are going to object to the voltage regulator IC being there. Are they really that bad used this way? I like that I don't need a trimpot to set the current. Just 1.25V/Ip = Rset in ohms.
1.25V / 0.0166A = 75.3R
Looks good with Rset = 75R.
The simulation didn't take much time, but the PCB layout is taking a long time. I have one channel laid out.
The Edcor trannies are open bobbin, so they're not going to get used in this project. I have them in the simulation because I made a model for them a while ago, so I can use that for the sims.
I do have a pair of Jensen JT-11P4-1, which are in little mu-metal cans with L-brackets for mounting to a chassis. I really don't want to drill that chassis (it's steel, and pretty heavy), but since the transformers are in a physically small package, I might be able to rig something up.
In the meantime, I modeled a version using an LM1085 as a CCS, with -24VDC for the CCS IC.
Gain is about the same.
I re-jiggered the screen grid supply for the 6EJ7s so I could use a smaller value bypass cap (2.2uF instead of 4.7uF). I have some ERO MKP box capacitors that will fit well.
This version's THD is actually lower than the one with the input transformers.
The 32VDC supply is my estimate of the raw DC I'll get from a 25.2VAC 225mA winding drawing 35mA. That DC supply is going to have to be mounted somewhere inside the chassis. I'm not looking forward to struggling with that, but I think I can make it work.
I know some people are going to object to the voltage regulator IC being there. Are they really that bad used this way? I like that I don't need a trimpot to set the current. Just 1.25V/Ip = Rset in ohms.
1.25V / 0.0166A = 75.3R
Looks good with Rset = 75R.
The simulation didn't take much time, but the PCB layout is taking a long time. I have one channel laid out.
On the curve tracer for triode mode, I just connect the screen lead banana plug into the plate lead banana plug. Suppressor is grounded to cathode.
From previous tests, a beam forming plate hardly makes any difference. Some true pentodes will show a tiny shift of triode curves to lower voltages if the suppressor is connected to plate too.
My curve traces don't guarantee long term stability for low voltage screen rated tubes, but I've never had any problem tracing the curves short term. Resistor or Zener to screen may be needed long term. I'm mainly interested in the tube linearity, and triode mode shows that quickly.
Hot glue, mounting tape, or cable ties under the chassis is suffice and no need to drill anything.
My thing with long-tailed pair is that the inverter phase has to go through an extra tube than the input phase. Probably makes no difference sonically but I'm OCD so it irks me.
I'd like to use these transformers. I've built a couple of other designs into this old ST70 chassis, and I'm sick of looking at it languishing on the shelf. Right now there's nothing in it. I'm hoping this goes together pretty easily, so I'm a bit obsessed with reducing the parts count. That ST70 gets really crowded inside.
I whipped up a PCB layout with a DN2535 for the CCS, with a trimpot to adjust the current. That means this design will have four trimpots: one each side for the CCS current, one each side for the LTP balance. The CCS adjustment is pretty much set-and-forget. I'm curious to see if the LTP balance adjustment makes an easily audible difference.
I whipped up a PCB layout with a DN2535 for the CCS, with a trimpot to adjust the current. That means this design will have four trimpots: one each side for the CCS current, one each side for the LTP balance. The CCS adjustment is pretty much set-and-forget. I'm curious to see if the LTP balance adjustment makes an easily audible difference.