| Home |
| Search |
| Today's Posts |
|
#1
Posted to rec.audio.tubes
|
|||
|
|||
Best tap for NFB?
Hi, Vacuumlanders,
I'd appreciate the opinions of the tube amplifier cogniscenti... I have built a semi-clone of a Williamson power amplifier, viz: -- 6SN7 1st audio and concertina phase splitter. -- 6SN7 driver stage, 1/2 a triode per phase - lots of drive available. -- 2 x 6L6 o/p tubes (can't afford KT66's at my pay grade!) -- 5U4 rectifier, of course :-) -- Recycled Fisher OPT with 0, 4, 8, 16 ohm taps (can't afford a Partridge or similar original - same reason!) -- Only 380 VDC B+, less about 24 VDC cathode bias, due to available PT (I don't want to go fixed bias unless I have to.) The original Williamson ran 450 VDC B+ on the KT66's but I'm not sure of the bias used, perhaps around 30 - 35 VDC as a guess. Nevertheless, I get about 18 watts at clipping - not so bad! Now the question: amplifier has global NFB from OPT secondary to cathode of 1st audio (stable and with small cap to knock out the ringing on square waves.) Should I take this NFB from the 16 ohm tap because it embraces all of the wire in the secondary, hence most of the magnetic flux in the core, or should I take it from the 8 ohm tap as that is what feeds every loudspeaker I have (and ever will have), so it's closest to the actual amplifier output. Needless to say, the Fisher OPT does not use efficient speaker secondaries in series and parallel to get the desired o/p impedance, rather it has just one tapped winding - the CT for 4 ohms, full secondary for 16 ohms with an intermediate tap for 8 ohms. NFB resistor would be adjusted for 8 or 16 ohm tap, of course. Presently NFB is from the 16 ohm tap. Is 8 ohm tap better, worse, or are we indifferent as to which? Thanks and cheers, Roger |
|
#2
Posted to rec.audio.tubes
|
|||
|
|||
|
Best tap for NFB?
|
|
#3
Posted to rec.audio.tubes
|
|||
|
|||
|
Best tap for NFB?
On 06/01/11 15:56, Engineer so wittily quipped:
Hi, Vacuumlanders, I'd appreciate the opinions of the tube amplifier cogniscenti... I have built a semi-clone of a Williamson power amplifier, viz: -- 6SN7 1st audio and concertina phase splitter. -- 6SN7 driver stage, 1/2 a triode per phase - lots of drive available. -- 2 x 6L6 o/p tubes (can't afford KT66's at my pay grade!) -- 5U4 rectifier, of course :-) -- Recycled Fisher OPT with 0, 4, 8, 16 ohm taps (can't afford a Partridge or similar original - same reason!) -- Only 380 VDC B+, less about 24 VDC cathode bias, due to available PT (I don't want to go fixed bias unless I have to.) The original Williamson ran 450 VDC B+ on the KT66's but I'm not sure of the bias used, perhaps around 30 - 35 VDC as a guess. Nevertheless, I get about 18 watts at clipping - not so bad! sounds good to me. You're limited by output Z matching probably. Maybe putting speakers BETWEEN taps will get you a few more watts? Depends on how they wound the secondaries, etc.. Now the question: amplifier has global NFB from OPT secondary to cathode of 1st audio (stable and with small cap to knock out the ringing on square waves.) good. Should I take this NFB from the 16 ohm tap because it embraces all of the wire in the secondary, hence most of the magnetic flux in the core, or should I take it from the 8 ohm tap as that is what feeds every loudspeaker I have (and ever will have), so it's closest to the actual amplifier output. I'd pick the output you're likely to use, and design around that. Needless to say, the Fisher OPT does not use efficient speaker secondaries in series and parallel to get the desired o/p impedance, rather it has just one tapped winding - the CT for 4 ohms, full secondary for 16 ohms with an intermediate tap for 8 ohms. good enough for 1955 probably. At least you have multiple taps. NFB resistor would be adjusted for 8 or 16 ohm tap, of course. Presently NFB is from the 16 ohm tap. Is 8 ohm tap better, worse, or are we indifferent as to which? hard to say. Again, I'd pick the output you're likely to use for your NFB. That way output is least likely to have any kinds of transformer-related artifacts (phase, etc.). HF oscillation could be worse, though, due to different (worse?) capacitive coupling and 'more NFB'. |
|
#4
Posted to rec.audio.tubes
|
|||
|
|||
|
Best tap for NFB?
On Jun 2, 8:56*am, Engineer wrote:
Hi, Vacuumlanders, I'd appreciate the opinions of the tube amplifier cogniscenti... I have built a semi-clone of a Williamson power amplifier, viz: -- 6SN7 1st audio and concertina phase splitter. -- 6SN7 driver stage, 1/2 a triode per phase - lots of drive available. -- 2 x 6L6 o/p tubes (can't afford KT66's at my pay grade!) -- 5U4 rectifier, of course *:-) -- Recycled Fisher OPT with 0, 4, 8, 16 ohm taps (can't afford a Partridge or similar original - same reason!) -- Only 380 VDC B+, less about 24 VDC cathode bias, due to available PT (I don't want to go fixed bias unless I have to.) *The original Williamson ran 450 VDC B+ on the KT66's but I'm not sure of the bias used, perhaps around 30 - 35 VDC as a guess. Nevertheless, I get about 18 watts at clipping - not so bad! Now the question: amplifier has global NFB from OPT secondary to cathode of 1st audio (stable and with small cap to knock out the ringing on square waves.) Should I take this NFB from the 16 ohm tap because it embraces all of the wire in the secondary, hence most of the magnetic flux in the core, or should I take it from the 8 ohm tap as that is what feeds every loudspeaker I have (and ever will have), so it's closest to the actual amplifier output. *Needless to say, the Fisher OPT does not use efficient speaker secondaries in series and parallel to get the desired o/p impedance, rather it has just one tapped winding - the CT for 4 ohms, full secondary for 16 ohms with an intermediate tap for 8 ohms. NFB resistor would be adjusted for 8 or 16 ohm tap, of course. Presently NFB is from the 16 ohm tap. *Is 8 ohm tap better, worse, or are we indifferent as to which? Thanks and cheers, Roger Where there is a 16,8,4 ohm sec arrangement I like to use the 16 ohm tap for NFB because its the one most tightly coupled magnetically to the primary. For unconditional stability with 0.22uF used on any tap you'll have to experiment with the Zobel R+C damping networks across the V1 anode RL and maybe across the OPT sec, say 0.05uF + 27 ohms across 16 ohm winding. Just what response you get with each tap at say 4 watts may vary with different HF response peaks with say 0.22uF with an R load in parallel. I like to see 14Hz to 65kHz at full power and slightly wider BW at lower powers and no chance of oscillations with a pure C load. But I bet you find at 18 W the power bandwidth is much less than 14Hz to 65kHz if you have made the amp unconditionally stable. Look to see if the amp is stable without any load. Often Wiliamsons won't be, and the CRO will show a trace bouncing up an down at some low F. You may need a step network after the 0.47uF coupling caps to the balanced driver grids, 1M + 0.05uF in parallel, then 220k bias R. Quad-II take their NFB from what is a 4 ohm outlet point not meant for a speaker. Their 16 ohm setting is more stable than the 8 ohm setting. But changing the turns for 16 or 8 ohms means the amount of NFB changes so Quad went to a point where the voltage change didn't occur. With the tapped sec there is no change to the turns used for the sec so you could uses 16, 8 or 4, depending on which gives best overall stability. Patrick Turner. |
|
#5
Posted to rec.audio.tubes
|
|||
|
|||
|
Best tap for NFB?
sounds good to me. *You're limited by output Z matching probably. *Maybe putting speakers BETWEEN taps will get you a few more watts? *Depends on how they wound the secondaries, etc.. The OP has not mentioned the OPT Z ratio P to S. Say its 6k: 16, 8 or 4. A 4 ohm speaker should be able to be used between the 16 and 4 terminals and get the same PO as for 4 to Com. The use of a speaker taken to between 16 and 8 terminals is a match for 6k : 1.44 ohms but a speaker of 4 ohms used there will raise the RLa-a to 15k0 and PO will be all pure class A if it is not now. Power will be less than 18W but THD/IMD will be low. The use between 8 and 4 gives a load match of 6k0 : 0.6 ohms approx and it is useless. Winding losses will be high. Fisher may have wound two identical tapped 16 ohm sec sections with taps all paralleled, but I doubt it. Patrick Turner. |
|
#6
Posted to rec.audio.tubes
|
|||
|
|||
|
Best tap for NFB?
On Jun 4, 9:39*am, Patrick Turner wrote:
sounds good to me. *You're limited by output Z matching probably. *Maybe putting speakers BETWEEN taps will get you a few more watts? *Depends on how they wound the secondaries, etc.. The OP has not mentioned the OPT Z ratio P to S. Say its 6k: 16, 8 or 4. *A 4 ohm speaker should be able to be used between the 16 and 4 terminals and get the same PO as for 4 to Com. The use of a speaker taken to between 16 and 8 terminals is a match for 6k : 1.44 ohms but a speaker of 4 ohms used there will raise the RLa-a to 15k0 and PO will be all pure class A if it is not now. Power will be less than 18W but THD/IMD will be low. The use between 8 and 4 gives a load match of 6k0 : 0.6 ohms approx and it is useless. Winding losses will be high. Fisher may have wound two identical tapped 16 ohm sec sections with taps all paralleled, but I doubt it. Patrick Turner. Patrick, The p-p impedance is 6.4 Kohms to nominal 0, 4, 8, 16 taps. My instinct was to use the 16 ohm tap for NFB for the reason you state (most secondary turns around the iron!) I did have that "slow bounce" instability you mention but got rid of it by reducing NFB. Unfortunately, that leaves the amplifier with too much gain for the Heathkit WA-P2 preamp I am using - noise from preamp too high. Do you have a reference to your "... CRO will show a trace bouncing up an down at some low F. You may need a step network after the 0.47uF coupling caps to the balanced driver grids, 1M + 0.05uF in parallel, then 220k bias R." The o/p tubes are pentode connected. I tried triode connection (same bias) but the power then maxed out at barely 6 watts! Presumably all class A. So I put the pentodes back!. Then I removed the individual (or course!) cathode bias caps to lower the gain but the power then maxed out at barely 12 watts... not good! Perhaps I need to rethink the bias... presently individual 390 ohms per cathode. Still thinking on the gain reduction fix... Cheers, Roger |
|
#7
Posted to rec.audio.tubes
|
|||
|
|||
|
Best tap for NFB?
On Jun 4, 1:24*pm, flipper wrote:
On Sat, 4 Jun 2011 09:57:51 -0700 (PDT), Engineer wrote: On Jun 4, 9:39*am, Patrick Turner wrote: sounds good to me. *You're limited by output Z matching probably. *Maybe putting speakers BETWEEN taps will get you a few more watts? *Depends on how they wound the secondaries, etc.. The OP has not mentioned the OPT Z ratio P to S. Say its 6k: 16, 8 or 4. *A 4 ohm speaker should be able to be used between the 16 and 4 terminals and get the same PO as for 4 to Com. The use of a speaker taken to between 16 and 8 terminals is a match for 6k : 1.44 ohms but a speaker of 4 ohms used there will raise the RLa-a to 15k0 and PO will be all pure class A if it is not now. Power will be less than 18W but THD/IMD will be low. The use between 8 and 4 gives a load match of 6k0 : 0.6 ohms approx and it is useless. Winding losses will be high. Fisher may have wound two identical tapped 16 ohm sec sections with taps all paralleled, but I doubt it. Patrick Turner. Patrick, The p-p impedance is 6.4 Kohms to nominal 0, 4, 8, 16 taps. *My instinct was to use the 16 ohm tap for NFB for the reason you state (most secondary turns around the iron!) I did have that "slow bounce" instability you mention but got rid of it by reducing NFB. *Unfortunately, that leaves the amplifier with too much gain for the Heathkit WA-P2 preamp I am using - noise from preamp too high. *Do you have a reference to your *"... CRO will show a trace bouncing up an down at some low F. *You may need a step network after the 0.47uF coupling caps to the balanced driver grids, 1M + 0.05uF in parallel, then 220k bias R." You normally have blocking cap (the .47uF) and a grid bias resistor to ground (or bias V), right? He's talking about putting a resistor in series with the blocking cap, on the grid resistor side, so you then have a voltage divider. That cuts gain by the resistor ratio. Now, bypass (parallel across it) the added resistor with the .05uF. That cuts the divider out above the 3dB point so you have a low F cut, from the divider, and no cut above the 3dB point. The o/p tubes are pentode connected. *I tried triode connection (same bias) but the power then maxed out at barely 6 watts! *Presumably all class A. So I put the pentodes back!. *Then I removed the individual (or course!) cathode bias caps to lower the gain but the power then maxed out at barely 12 watts... not good! Perhaps I need to rethink the bias... presently individual 390 ohms per cathode. Still thinking on the gain reduction fix... Cheers, Roger Ah, ha... got it! I presume this is the final RC coupling to the o/p tubes. Actually, I only used 0.05 uF anyway with 470 Kohm 6L6 grid resistors to ground. I figured the -3dB point was low enough at 6.8 Hz! The Williamson used 100K grid resistors (on the KT66's) and 0.25 uF coupling caps. If I drop my 6L6 grid resistors to 100K (and used a larger coupling cap for say, -3 dB at 10 Hz), then they load the stage driver output impedance a bit more and reduce the gain a bit, as I want. The I could do what Patrick said (and you clarified, thanks) and stablize the VLF, then add even more NFB to drop the gain further to where I need it for the WA-P2. Back to the schematic, pencil and calculator... Cheers, Roger |
|
#8
Posted to rec.audio.tubes
|
|||
|
|||
|
Best tap for NFB?
On Jun 5, 2:57*am, Engineer wrote:
On Jun 4, 9:39*am, Patrick Turner wrote: sounds good to me. *You're limited by output Z matching probably. *Maybe putting speakers BETWEEN taps will get you a few more watts? *Depends on how they wound the secondaries, etc.. The OP has not mentioned the OPT Z ratio P to S. Say its 6k: 16, 8 or 4. *A 4 ohm speaker should be able to be used between the 16 and 4 terminals and get the same PO as for 4 to Com. The use of a speaker taken to between 16 and 8 terminals is a match for 6k : 1.44 ohms but a speaker of 4 ohms used there will raise the RLa-a to 15k0 and PO will be all pure class A if it is not now. Power will be less than 18W but THD/IMD will be low. The use between 8 and 4 gives a load match of 6k0 : 0.6 ohms approx and it is useless. Winding losses will be high. Fisher may have wound two identical tapped 16 ohm sec sections with taps all paralleled, but I doubt it. Patrick Turner. Patrick, The p-p impedance is 6.4 Kohms to nominal 0, 4, 8, 16 taps. *My instinct was to use the 16 ohm tap for NFB for the reason you state (most secondary turns around the iron!) I did have that "slow bounce" instability you mention but got rid of it by reducing NFB. *Unfortunately, that leaves the amplifier with too much gain for the Heathkit WA-P2 preamp I am using - noise from preamp too high. *Do you have a reference to your *"... CRO will show a trace bouncing up an down at some low F. *You may need a step network after the 0.47uF coupling caps to the balanced driver grids, 1M + 0.05uF in parallel, then 220k bias R." The o/p tubes are pentode connected. *I tried triode connection (same bias) but the power then maxed out at barely 6 watts! *Presumably all class A. So I put the pentodes back!. *Then I removed the individual (or course!) cathode bias caps to lower the gain but the power then maxed out at barely 12 watts... not good! Perhaps I need to rethink the bias... presently individual 390 ohms per cathode. Still thinking on the gain reduction fix... Cheers, Roger- Hide quoted text - - Show quoted text - Try reading my schematics at my website, in particular, try the principles embodied in the 5050 amp design at http://www.turneraudio.com.au/Integrated5050.htm You'll find the networks used between input SET stage and following driver LTP will tame all HF and LF oscillations. But the values I show won't suit what you have. The principles behind networks need to be learnt so you can apply them. Its called "critical damping". Patrick Turner. |
|
#9
Posted to rec.audio.tubes
|
|||
|
|||
|
Best tap for NFB?
On Jun 4, 3:58*pm, flipper wrote:
On Sat, 4 Jun 2011 12:20:30 -0700 (PDT), Engineer wrote: On Jun 4, 1:24*pm, flipper wrote: On Sat, 4 Jun 2011 09:57:51 -0700 (PDT), Engineer wrote: On Jun 4, 9:39*am, Patrick Turner wrote: sounds good to me. *You're limited by output Z matching probably. *Maybe putting speakers BETWEEN taps will get you a few more watts? *Depends on how they wound the secondaries, etc.. The OP has not mentioned the OPT Z ratio P to S. Say its 6k: 16, 8 or 4. *A 4 ohm speaker should be able to be used between the 16 and 4 terminals and get the same PO as for 4 to Com. The use of a speaker taken to between 16 and 8 terminals is a match for 6k : 1.44 ohms but a speaker of 4 ohms used there will raise the RLa-a to 15k0 and PO will be all pure class A if it is not now. Power will be less than 18W but THD/IMD will be low. The use between 8 and 4 gives a load match of 6k0 : 0.6 ohms approx and it is useless. Winding losses will be high. Fisher may have wound two identical tapped 16 ohm sec sections with taps all paralleled, but I doubt it. Patrick Turner. Patrick, The p-p impedance is 6.4 Kohms to nominal 0, 4, 8, 16 taps. *My instinct was to use the 16 ohm tap for NFB for the reason you state (most secondary turns around the iron!) I did have that "slow bounce" instability you mention but got rid of it by reducing NFB. *Unfortunately, that leaves the amplifier with too much gain for the Heathkit WA-P2 preamp I am using - noise from preamp too high. *Do you have a reference to your *"... CRO will show a trace bouncing up an down at some low F. *You may need a step network after the 0.47uF coupling caps to the balanced driver grids, 1M + 0.05uF in parallel, then 220k bias R." You normally have blocking cap (the .47uF) and a grid bias resistor to ground (or bias V), right? He's talking about putting a resistor in series with the blocking cap, on the grid resistor side, so you then have a voltage divider. That cuts gain by the resistor ratio. Now, bypass (parallel across it) the added resistor with the .05uF. That cuts the divider out above the 3dB point so you have a low F cut, from the divider, and no cut above the 3dB point. The o/p tubes are pentode connected. *I tried triode connection (same bias) but the power then maxed out at barely 6 watts! *Presumably all class A. So I put the pentodes back!. *Then I removed the individual (or course!) cathode bias caps to lower the gain but the power then maxed out at barely 12 watts... not good! Perhaps I need to rethink the bias... presently individual 390 ohms per cathode. Still thinking on the gain reduction fix... Cheers, Roger Ah, ha... got it! *I presume this is the final RC coupling to the o/p tubes. The driver tubes grids. Actually, I only used 0.05 uF anyway with 470 Kohm 6L6 grid resistors to ground. *I figured the -3dB point was low enough at 6.8 Hz! I know the datasheet says 'up to' 500k for grid leaks with cathode bias but I'd lower them to at least 220k for better bias stability. It's not quite so critical with driver tubes, because of the plate resistor, but output tube run away makes for bad juju. The Williamson used 100K grid resistors (on the KT66's) and 0.25 uF coupling caps. If I drop my 6L6 grid resistors to 100K (and used a larger coupling cap for say, -3 dB at 10 Hz), then they load the stage driver output impedance a bit more and reduce the gain a bit, as I want. *The I could do what Patrick said (and you clarified, thanks) and stablize the VLF, then add even more NFB to drop the gain further to where I need it for the WA-P2. Back to the schematic, pencil and calculator... Cheers, Roger Quick update: I replaced the "470K" (actually they were 400K) 6L6 grid leaks with 100K and left the 0.05 uF coupling caps in (-3 dB point now at 32 Hz.) Result: no sign of VLF instability. I can now increase the global NFB from a near unstable 33K resistor to about 15K with no signs of VLF surging. The larger NFB has reduced the overall gain to about what I need for the WA-P2 preamp, so I don't think I'll need a special LF roll-off coupling to the 6L6's. So far win-win. I still have to fine tune the parallel NFB cap to cut square wave ringing (there's quite a lot!) and reconfirm all performance figures. Cheers, Roger |
|
#10
Posted to rec.audio.tubes
|
|||
|
|||
|
Best tap for NFB?
On Jun 5, 7:34*pm, flipper wrote:
On Sun, 5 Jun 2011 13:21:37 -0700 (PDT), Engineer wrote: On Jun 4, 3:58*pm, flipper wrote: On Sat, 4 Jun 2011 12:20:30 -0700 (PDT), Engineer wrote: On Jun 4, 1:24*pm, flipper wrote: On Sat, 4 Jun 2011 09:57:51 -0700 (PDT), Engineer wrote: On Jun 4, 9:39*am, Patrick Turner wrote: sounds good to me. *You're limited by output Z matching probably. *Maybe putting speakers BETWEEN taps will get you a few more watts? *Depends on how they wound the secondaries, etc.. The OP has not mentioned the OPT Z ratio P to S. Say its 6k: 16, 8 or 4. *A 4 ohm speaker should be able to be used between the 16 and 4 terminals and get the same PO as for 4 to Com. The use of a speaker taken to between 16 and 8 terminals is a match for 6k : 1.44 ohms but a speaker of 4 ohms used there will raise the RLa-a to 15k0 and PO will be all pure class A if it is not now. Power will be less than 18W but THD/IMD will be low. The use between 8 and 4 gives a load match of 6k0 : 0.6 ohms approx and it is useless. Winding losses will be high. Fisher may have wound two identical tapped 16 ohm sec sections with taps all paralleled, but I doubt it. Patrick Turner. Patrick, The p-p impedance is 6.4 Kohms to nominal 0, 4, 8, 16 taps. *My instinct was to use the 16 ohm tap for NFB for the reason you state (most secondary turns around the iron!) I did have that "slow bounce" instability you mention but got rid of it by reducing NFB. *Unfortunately, that leaves the amplifier with too much gain for the Heathkit WA-P2 preamp I am using - noise from preamp too high. *Do you have a reference to your *"... CRO will show a trace bouncing up an down at some low F. *You may need a step network after the 0.47uF coupling caps to the balanced driver grids, 1M + 0.05uF in parallel, then 220k bias R." You normally have blocking cap (the .47uF) and a grid bias resistor to ground (or bias V), right? He's talking about putting a resistor in series with the blocking cap, on the grid resistor side, so you then have a voltage divider. That cuts gain by the resistor ratio. Now, bypass (parallel across it) the added resistor with the .05uF. That cuts the divider out above the 3dB point so you have a low F cut, from the divider, and no cut above the 3dB point. The o/p tubes are pentode connected. *I tried triode connection (same bias) but the power then maxed out at barely 6 watts! *Presumably all class A. So I put the pentodes back!. *Then I removed the individual (or course!) cathode bias caps to lower the gain but the power then maxed out at barely 12 watts... not good! Perhaps I need to rethink the bias... presently individual 390 ohms per cathode. Still thinking on the gain reduction fix... Cheers, Roger Ah, ha... got it! *I presume this is the final RC coupling to the o/p tubes. The driver tubes grids. Actually, I only used 0.05 uF anyway with 470 Kohm 6L6 grid resistors to ground. *I figured the -3dB point was low enough at 6.8 Hz! I know the datasheet says 'up to' 500k for grid leaks with cathode bias but I'd lower them to at least 220k for better bias stability. It's not quite so critical with driver tubes, because of the plate resistor, but output tube run away makes for bad juju. The Williamson used 100K grid resistors (on the KT66's) and 0.25 uF coupling caps. If I drop my 6L6 grid resistors to 100K (and used a larger coupling cap for say, -3 dB at 10 Hz), then they load the stage driver output impedance a bit more and reduce the gain a bit, as I want. *The I could do what Patrick said (and you clarified, thanks) and stablize the VLF, then add even more NFB to drop the gain further to where I need it for the WA-P2. Back to the schematic, pencil and calculator... Cheers, Roger Quick update: I replaced the "470K" (actually they were 400K) 6L6 grid leaks with 100K and left the 0.05 uF coupling caps in (-3 dB point now at 32 Hz.) Result: no sign of VLF instability. That's good news but I'd be tempted to bump those caps to .1uF, or even .22uF like in the Williamson, and lower the caps into the driver grids and here's why. NFB will try to make the output, times gain, equal the input which, of course, is why we're using NFB. But you're cutting the signal 3dB (or more, depending on the frequency) at the output tube grids with the coupling cap roll-off. That means the input gain stage is going to increase it's signal (at that frequency) by 3dB to compensate (NFB at work) and the same increase goes into the driver tubes. Now, a 3dB increase might not be a problem for your drivers but you can see that, sooner or later, depending on how much of a shelf you're doing, it can become a clipping problem because they have to swing a larger signal to get through that output tube coupling cap. And, in addition to potential clipping, increasing the output level, of everything before the roll-off, also increases distortion in those stages There's nothing you can do about the first gain stage and concertina (the signal level there is the smallest of the whole amp so there should be 'room' for it) but if you move the cut to the driver tube grids then they will be operating 'like normal'. *I can now increase the global NFB from a near unstable 33K resistor to about 15K with no signs of VLF surging. The larger NFB has reduced the overall gain to about what I need for the WA-P2 preamp, so I don't think I'll need a special LF roll-off coupling to the 6L6's. *So far win-win. I still have to fine tune the parallel NFB cap to cut square wave ringing (there's quite a lot!) and reconfirm all performance figures. You might want to consider shelving HF which, if you look at Patrick's schematic, is the cap, and it's series resistor, in parallel across the bottom resistor of the divider. I.E. that cap shunts HF, through the resistor in series with it, to ground making the divider ratio lower so it cuts HF into the LPT. Good point about the drivers, flipper. I had not thought of that. I'll see what I can do to roll off the bottom right after the phase splitter and add a couple of 0.1 uF's in parallel with the 0.05's to the 6L6 grids. Since posting above, I played with the NFB cap bypass using a radio tuning cap to see what value I need. There is a problem... as I increase the cap, just before the square wave ringing stops an HF instability pops up on the back of the square wave. I'm pretty sure I need the HF shelf you speak of. Cheers, Roger |
|
#11
Posted to rec.audio.tubes
|
|||
|
|||
|
Best tap for NFB?
In article ,
Engineer wrote: On Jun 4, 3:58*pm, flipper wrote: On Sat, 4 Jun 2011 12:20:30 -0700 (PDT), Engineer wrote: On Jun 4, 1:24*pm, flipper wrote: On Sat, 4 Jun 2011 09:57:51 -0700 (PDT), Engineer wrote: On Jun 4, 9:39*am, Patrick Turner wrote: sounds good to me. *You're limited by output Z matching probably. *Maybe putting speakers BETWEEN taps will get you a few more watts? *Depends on how they wound the secondaries, etc.. The OP has not mentioned the OPT Z ratio P to S. Say its 6k: 16, 8 or 4. *A 4 ohm speaker should be able to be used between the 16 and 4 terminals and get the same PO as for 4 to Com. The use of a speaker taken to between 16 and 8 terminals is a match for 6k : 1.44 ohms but a speaker of 4 ohms used there will raise the RLa-a to 15k0 and PO will be all pure class A if it is not now. Power will be less than 18W but THD/IMD will be low. The use between 8 and 4 gives a load match of 6k0 : 0.6 ohms approx and it is useless. Winding losses will be high. Fisher may have wound two identical tapped 16 ohm sec sections with taps all paralleled, but I doubt it. Patrick Turner. Patrick, The p-p impedance is 6.4 Kohms to nominal 0, 4, 8, 16 taps. *My instinct was to use the 16 ohm tap for NFB for the reason you state (most secondary turns around the iron!) I did have that "slow bounce" instability you mention but got rid of it by reducing NFB. *Unfortunately, that leaves the amplifier with too much gain for the Heathkit WA-P2 preamp I am using - noise from preamp too high. *Do you have a reference to your *"... CRO will show a trace bouncing up an down at some low F. *You may need a step network after the 0.47uF coupling caps to the balanced driver grids, 1M + 0.05uF in parallel, then 220k bias R." You normally have blocking cap (the .47uF) and a grid bias resistor to ground (or bias V), right? He's talking about putting a resistor in series with the blocking cap, on the grid resistor side, so you then have a voltage divider. That cuts gain by the resistor ratio. Now, bypass (parallel across it) the added resistor with the .05uF. That cuts the divider out above the 3dB point so you have a low F cut, from the divider, and no cut above the 3dB point. The o/p tubes are pentode connected. *I tried triode connection (same bias) but the power then maxed out at barely 6 watts! *Presumably all class A. So I put the pentodes back!. *Then I removed the individual (or course!) cathode bias caps to lower the gain but the power then maxed out at barely 12 watts... not good! Perhaps I need to rethink the bias... presently individual 390 ohms per cathode. Still thinking on the gain reduction fix... Cheers, Roger Ah, ha... got it! *I presume this is the final RC coupling to the o/p tubes. The driver tubes grids. Actually, I only used 0.05 uF anyway with 470 Kohm 6L6 grid resistors to ground. *I figured the -3dB point was low enough at 6.8 Hz! I know the datasheet says 'up to' 500k for grid leaks with cathode bias but I'd lower them to at least 220k for better bias stability. It's not quite so critical with driver tubes, because of the plate resistor, but output tube run away makes for bad juju. The Williamson used 100K grid resistors (on the KT66's) and 0.25 uF coupling caps. If I drop my 6L6 grid resistors to 100K (and used a larger coupling cap for say, -3 dB at 10 Hz), then they load the stage driver output impedance a bit more and reduce the gain a bit, as I want. *The I could do what Patrick said (and you clarified, thanks) and stablize the VLF, then add even more NFB to drop the gain further to where I need it for the WA-P2. Back to the schematic, pencil and calculator... Cheers, Roger Quick update: I replaced the "470K" (actually they were 400K) 6L6 grid leaks with 100K and left the 0.05 uF coupling caps in (-3 dB point now at 32 Hz.) Result: no sign of VLF instability. I can now increase the global NFB from a near unstable 33K resistor to about 15K with no signs of VLF surging. The larger NFB has reduced the overall gain to about what I need for the WA-P2 preamp, so I don't think I'll need a special LF roll-off coupling to the 6L6's. So far win-win. Have you checked to be sure the amplifier remains stable at low frequencies without a load? With pentode connected output tubes the output stage low frequency pole can shift radically higher in frequency when the load is disconnected, possibly screwing up your careful staggering of pole frequencies and causing instability. Regards, John Byrns -- Regards, John Byrns Surf my web pages at, http://fmamradios.com/ |
|
#12
Posted to rec.audio.tubes
|
|||
|
|||
|
Best tap for NFB?
On Jun 6, 9:34*am, flipper wrote:
On Sun, 5 Jun 2011 13:21:37 -0700 (PDT), Engineer wrote: On Jun 4, 3:58*pm, flipper wrote: On Sat, 4 Jun 2011 12:20:30 -0700 (PDT), Engineer wrote: On Jun 4, 1:24*pm, flipper wrote: On Sat, 4 Jun 2011 09:57:51 -0700 (PDT), Engineer wrote: On Jun 4, 9:39*am, Patrick Turner wrote: sounds good to me. *You're limited by output Z matching probably. *Maybe putting speakers BETWEEN taps will get you a few more watts? *Depends on how they wound the secondaries, etc.. The OP has not mentioned the OPT Z ratio P to S. Say its 6k: 16, 8 or 4. *A 4 ohm speaker should be able to be used between the 16 and 4 terminals and get the same PO as for 4 to Com. The use of a speaker taken to between 16 and 8 terminals is a match for 6k : 1.44 ohms but a speaker of 4 ohms used there will raise the RLa-a to 15k0 and PO will be all pure class A if it is not now. Power will be less than 18W but THD/IMD will be low. The use between 8 and 4 gives a load match of 6k0 : 0.6 ohms approx and it is useless. Winding losses will be high. Fisher may have wound two identical tapped 16 ohm sec sections with taps all paralleled, but I doubt it. Patrick Turner. Patrick, The p-p impedance is 6.4 Kohms to nominal 0, 4, 8, 16 taps. *My instinct was to use the 16 ohm tap for NFB for the reason you state (most secondary turns around the iron!) I did have that "slow bounce" instability you mention but got rid of it by reducing NFB. *Unfortunately, that leaves the amplifier with too much gain for the Heathkit WA-P2 preamp I am using - noise from preamp too high. *Do you have a reference to your *"... CRO will show a trace bouncing up an down at some low F. *You may need a step network after the 0.47uF coupling caps to the balanced driver grids, 1M + 0.05uF in parallel, then 220k bias R." You normally have blocking cap (the .47uF) and a grid bias resistor to ground (or bias V), right? He's talking about putting a resistor in series with the blocking cap, on the grid resistor side, so you then have a voltage divider. That cuts gain by the resistor ratio. Now, bypass (parallel across it) the added resistor with the .05uF. That cuts the divider out above the 3dB point so you have a low F cut, from the divider, and no cut above the 3dB point. The o/p tubes are pentode connected. *I tried triode connection (same bias) but the power then maxed out at barely 6 watts! *Presumably all class A. So I put the pentodes back!. *Then I removed the individual (or course!) cathode bias caps to lower the gain but the power then maxed out at barely 12 watts... not good! Perhaps I need to rethink the bias... presently individual 390 ohms per cathode. Still thinking on the gain reduction fix... Cheers, Roger Ah, ha... got it! *I presume this is the final RC coupling to the o/p tubes. The driver tubes grids. Actually, I only used 0.05 uF anyway with 470 Kohm 6L6 grid resistors to ground. *I figured the -3dB point was low enough at 6.8 Hz! I know the datasheet says 'up to' 500k for grid leaks with cathode bias but I'd lower them to at least 220k for better bias stability. It's not quite so critical with driver tubes, because of the plate resistor, but output tube run away makes for bad juju. The Williamson used 100K grid resistors (on the KT66's) and 0.25 uF coupling caps. If I drop my 6L6 grid resistors to 100K (and used a larger coupling cap for say, -3 dB at 10 Hz), then they load the stage driver output impedance a bit more and reduce the gain a bit, as I want. *The I could do what Patrick said (and you clarified, thanks) and stablize the VLF, then add even more NFB to drop the gain further to where I need it for the WA-P2. Back to the schematic, pencil and calculator... Cheers, Roger Quick update: I replaced the "470K" (actually they were 400K) 6L6 grid leaks with 100K and left the 0.05 uF coupling caps in (-3 dB point now at 32 Hz.) Result: no sign of VLF instability. That's good news but I'd be tempted to bump those caps to .1uF, or even .22uF like in the Williamson, and lower the caps into the driver grids and here's why. NFB will try to make the output, times gain, equal the input which, of course, is why we're using NFB. But you're cutting the signal 3dB (or more, depending on the frequency) at the output tube grids with the coupling cap roll-off. That means the input gain stage is going to increase it's signal (at that frequency) by 3dB to compensate (NFB at work) and the same increase goes into the driver tubes. Now, a 3dB increase might not be a problem for your drivers but you can see that, sooner or later, depending on how much of a shelf you're doing, it can become a clipping problem because they have to swing a larger signal to get through that output tube coupling cap. And, in addition to potential clipping, increasing the output level, of everything before the roll-off, also increases distortion in those stages There's nothing you can do about the first gain stage and concertina (the signal level there is the smallest of the whole amp so there should be 'room' for it) but if you move the cut to the driver tube grids then they will be operating 'like normal'. *I can now increase the global NFB from a near unstable 33K resistor to about 15K with no signs of VLF surging. The larger NFB has reduced the overall gain to about what I need for the WA-P2 preamp, so I don't think I'll need a special LF roll-off coupling to the 6L6's. *So far win-win. I still have to fine tune the parallel NFB cap to cut square wave ringing (there's quite a lot!) and reconfirm all performance figures. You might want to consider shelving HF which, if you look at Patrick's schematic, is the cap, and it's series resistor, in parallel across the bottom resistor of the divider. I.E. that cap shunts HF, through the resistor in series with it, to ground making the divider ratio lower so it cuts HF into the LPT. For a Williamson the problems of overdrive at LF due to FB action boosting the error signal in input & concertina stages may be overcome by using the shelving network before the balanced amp, and where there is a huge amount of headroom. But you need to use TWO networks each 0.47uF then 1M//0.05uF, then 220k series grid bias R. The highest pole in this network is determimed by the 0.05uF and 220k which gives -3dB at 14Hz, then the there is a second pole at about 2.7Hz when the response levels out at -14.9dB, then a third pole between 0.47uF and 1,220k at 0.28Hz. Phase shift between 14Hz and 1Hz is never more than +45 degrees and only increases to +90d at 0.28Hz when gain loss from other networks in the amp have reduced it to keep Nyquist happy. It is possible to put such shelf networks between balanced amp and OP tubes - it works OK, but that's where the balanced amp might become over driven as you point out. But fact is there is SFA signal in most music below 20Hz. I've sometimes used TWO pairs of shelf networks where OPT inductance is too small as in the case of a Ming Da amp I'm re-engineering here for a customer. It has SET + LTP input stages followed by balanced amp with 300B driving 845 in PP, so there are 4 stages and 3 lots of C&R coupling between stages. Shelve networks were found to vastly improve bass and stability compared to the Chinese Dumbo Design. The shelf networks reduce gain where you just do not want much open loop gain nor do you want a large amount of NFB applied, ie, below 14Hz and above 20kHz. Patrick Turner. Cheers, Roger- Hide quoted text - - Show quoted text -- Hide quoted text - - Show quoted text - |
|
#13
Posted to rec.audio.tubes
|
|||
|
|||
|
Best tap for NFB?
On Jun 6, 8:42*pm, flipper wrote:
On Sun, 5 Jun 2011 23:50:22 -0700 (PDT), Patrick Turner wrote: On Jun 6, 9:34*am, flipper wrote: On Sun, 5 Jun 2011 13:21:37 -0700 (PDT), Engineer wrote: On Jun 4, 3:58*pm, flipper wrote: On Sat, 4 Jun 2011 12:20:30 -0700 (PDT), Engineer wrote: On Jun 4, 1:24*pm, flipper wrote: On Sat, 4 Jun 2011 09:57:51 -0700 (PDT), Engineer wrote: On Jun 4, 9:39*am, Patrick Turner wrote: sounds good to me. *You're limited by output Z matching probably. *Maybe putting speakers BETWEEN taps will get you a few more watts? *Depends on how they wound the secondaries, etc.. The OP has not mentioned the OPT Z ratio P to S. Say its 6k: 16, 8 or 4. *A 4 ohm speaker should be able to be used between the 16 and 4 terminals and get the same PO as for 4 to Com. The use of a speaker taken to between 16 and 8 terminals is a match for 6k : 1.44 ohms but a speaker of 4 ohms used there will raise the RLa-a to 15k0 and PO will be all pure class A if it is not now. Power will be less than 18W but THD/IMD will be low. The use between 8 and 4 gives a load match of 6k0 : 0.6 ohms approx and it is useless. Winding losses will be high. Fisher may have wound two identical tapped 16 ohm sec sections with taps all paralleled, but I doubt it. Patrick Turner. Patrick, The p-p impedance is 6.4 Kohms to nominal 0, 4, 8, 16 taps. *My instinct was to use the 16 ohm tap for NFB for the reason you state (most secondary turns around the iron!) I did have that "slow bounce" instability you mention but got rid of it by reducing NFB. *Unfortunately, that leaves the amplifier with too much gain for the Heathkit WA-P2 preamp I am using - noise from preamp too high. *Do you have a reference to your *"... CRO will show a trace bouncing up an down at some low F. *You may need a step network after the 0.47uF coupling caps to the balanced driver grids, 1M + 0.05uF in parallel, then 220k bias R." You normally have blocking cap (the .47uF) and a grid bias resistor to ground (or bias V), right? He's talking about putting a resistor in series with the blocking cap, on the grid resistor side, so you then have a voltage divider. That cuts gain by the resistor ratio. Now, bypass (parallel across it) the added resistor with the .05uF. That cuts the divider out above the 3dB point so you have a low F cut, from the divider, and no cut above the 3dB point. The o/p tubes are pentode connected. *I tried triode connection (same bias) but the power then maxed out at barely 6 watts! *Presumably all class A. So I put the pentodes back!. *Then I removed the individual (or course!) cathode bias caps to lower the gain but the power then maxed out at barely 12 watts... not good! Perhaps I need to rethink the bias... presently individual 390 ohms per cathode. Still thinking on the gain reduction fix... Cheers, Roger Ah, ha... got it! *I presume this is the final RC coupling to the o/p tubes. The driver tubes grids. Actually, I only used 0.05 uF anyway with 470 Kohm 6L6 grid resistors to ground. *I figured the -3dB point was low enough at 6.8 Hz! I know the datasheet says 'up to' 500k for grid leaks with cathode bias but I'd lower them to at least 220k for better bias stability. It's not quite so critical with driver tubes, because of the plate resistor, but output tube run away makes for bad juju. The Williamson used 100K grid resistors (on the KT66's) and 0.25 uF coupling caps. If I drop my 6L6 grid resistors to 100K (and used a larger coupling cap for say, -3 dB at 10 Hz), then they load the stage driver output impedance a bit more and reduce the gain a bit, as I want. *The I could do what Patrick said (and you clarified, thanks) and stablize the VLF, then add even more NFB to drop the gain further to where I need it for the WA-P2. Back to the schematic, pencil and calculator... Cheers, Roger Quick update: I replaced the "470K" (actually they were 400K) 6L6 grid leaks with 100K and left the 0.05 uF coupling caps in (-3 dB point now at 32 Hz.) Result: no sign of VLF instability. That's good news but I'd be tempted to bump those caps to .1uF, or even .22uF like in the Williamson, and lower the caps into the driver grids and here's why. NFB will try to make the output, times gain, equal the input which, of course, is why we're using NFB. But you're cutting the signal 3dB (or more, depending on the frequency) at the output tube grids with the coupling cap roll-off. That means the input gain stage is going to increase it's signal (at that frequency) by 3dB to compensate (NFB at work) and the same increase goes into the driver tubes. Now, a 3dB increase might not be a problem for your drivers but you can see that, sooner or later, depending on how much of a shelf you're doing, it can become a clipping problem because they have to swing a larger signal to get through that output tube coupling cap. And, in addition to potential clipping, increasing the output level, of everything before the roll-off, also increases distortion in those stages There's nothing you can do about the first gain stage and concertina (the signal level there is the smallest of the whole amp so there should be 'room' for it) but if you move the cut to the driver tube grids then they will be operating 'like normal'. *I can now increase the global NFB from a near unstable 33K resistor to about 15K with no signs of VLF surging. The larger NFB has reduced the overall gain to about what I need for the WA-P2 preamp, so I don't think I'll need a special LF roll-off coupling to the 6L6's. *So far win-win. I still have to fine tune the parallel NFB cap to cut square wave ringing (there's quite a lot!) and reconfirm all performance figures. You might want to consider shelving HF which, if you look at Patrick's schematic, is the cap, and it's series resistor, in parallel across the bottom resistor of the divider. I.E. that cap shunts HF, through the resistor in series with it, to ground making the divider ratio lower so it cuts HF into the LPT. For a Williamson the problems of overdrive at LF due to FB action boosting the error signal in input & concertina stages may be overcome by using the shelving network before the balanced amp, and where there is a huge amount of headroom. That's what I said. But you need to use TWO networks each 0.47uF then 1M//0.05uF, then 220k series grid bias R. Same as doing it at the output tubes. The highest pole in this network is determimed by the 0.05uF and 220k which gives -3dB at 14Hz, then the there is a second pole at about 2.7Hz when the response levels out at -14.9dB, then a third pole between 0.47uF and 1,220k at 0.28Hz. Phase shift between 14Hz and 1Hz is never more than +45 degrees and only increases to +90d at 0.28Hz when gain loss from other networks in the amp have reduced it to keep Nyquist happy. That may be fine for the amp in your example but that doesn't mean those poles are right for his. It is possible to put such shelf networks between balanced amp and OP tubes - it works OK, but that's where the balanced amp might become over driven as you point out. But fact is there is SFA signal in most music below 20Hz. He implied the pole needed was closer to 32Hz. There very well might not be a 'problem' shelving at the output grids but I see no good reason to put it there either. I've sometimes used TWO pairs of shelf networks where OPT inductance is too small as in the case of a Ming Da amp I'm re-engineering here for a customer. It has SET + LTP input stages followed by balanced amp with 300B driving 845 in PP, so there are 4 stages and 3 lots of C&R coupling between stages. Shelve networks were found to vastly improve bass and stability compared to the Chinese Dumbo Design. The shelf networks reduce gain where you just do not want much open loop gain nor do you want a large amount of NFB applied, ie, below 14Hz and above 20kHz. It should be noted that 'the purpose' given reduces NFB in that region and, as a result, increases distortion so you don't want to do 'more' than needed and the ideal case would be none at all, assuming you could make the amp stable with none at all. Well, I'm confident the OP will work it out for himself. The LF shelving networks I suggest, and where I put them, ie, never right before the OP tubes do work in 90% of cases where otherwise an amp oscillates at LF, especially when there isn't any load connected when OLG goes high, amount of NFB increases and, oops, the darn thing wants to oscillate at 0.5Hz. Patrick Turner. |
|
#14
Posted to rec.audio.tubes
|
|||
|
|||
|
Best tap for NFB?
On Jun 7, 3:42*am, Patrick Turner wrote:
On Jun 6, 8:42*pm, flipper wrote: On Sun, 5 Jun 2011 23:50:22 -0700 (PDT), Patrick Turner wrote: On Jun 6, 9:34*am, flipper wrote: On Sun, 5 Jun 2011 13:21:37 -0700 (PDT), Engineer wrote: On Jun 4, 3:58*pm, flipper wrote: On Sat, 4 Jun 2011 12:20:30 -0700 (PDT), Engineer wrote: On Jun 4, 1:24*pm, flipper wrote: On Sat, 4 Jun 2011 09:57:51 -0700 (PDT), Engineer wrote: On Jun 4, 9:39*am, Patrick Turner wrote: sounds good to me. *You're limited by output Z matching probably. *Maybe putting speakers BETWEEN taps will get you a few more watts? *Depends on how they wound the secondaries, etc.. The OP has not mentioned the OPT Z ratio P to S. Say its 6k: 16, 8 or 4. *A 4 ohm speaker should be able to be used between the 16 and 4 terminals and get the same PO as for 4 to Com. The use of a speaker taken to between 16 and 8 terminals is a match for 6k : 1.44 ohms but a speaker of 4 ohms used there will raise the RLa-a to 15k0 and PO will be all pure class A if it is not now. Power will be less than 18W but THD/IMD will be low. The use between 8 and 4 gives a load match of 6k0 : 0.6 ohms approx and it is useless. Winding losses will be high. Fisher may have wound two identical tapped 16 ohm sec sections with taps all paralleled, but I doubt it. Patrick Turner. Patrick, The p-p impedance is 6.4 Kohms to nominal 0, 4, 8, 16 taps. *My instinct was to use the 16 ohm tap for NFB for the reason you state (most secondary turns around the iron!) I did have that "slow bounce" instability you mention but got rid of it by reducing NFB. *Unfortunately, that leaves the amplifier with too much gain for the Heathkit WA-P2 preamp I am using - noise from preamp too high. *Do you have a reference to your *"... CRO will show a trace bouncing up an down at some low F. *You may need a step network after the 0.47uF coupling caps to the balanced driver grids, 1M + 0..05uF in parallel, then 220k bias R." You normally have blocking cap (the .47uF) and a grid bias resistor to ground (or bias V), right? He's talking about putting a resistor in series with the blocking cap, on the grid resistor side, so you then have a voltage divider. That cuts gain by the resistor ratio. Now, bypass (parallel across it) the added resistor with the .05uF. That cuts the divider out above the 3dB point so you have a low F cut, from the divider, and no cut above the 3dB point. The o/p tubes are pentode connected. *I tried triode connection (same bias) but the power then maxed out at barely 6 watts! *Presumably all class A. So I put the pentodes back!. *Then I removed the individual (or course!) cathode bias caps to lower the gain but the power then maxed out at barely 12 watts... not good! Perhaps I need to rethink the bias... presently individual 390 ohms per cathode. Still thinking on the gain reduction fix... Cheers, Roger Ah, ha... got it! *I presume this is the final RC coupling to the o/p tubes. The driver tubes grids. Actually, I only used 0.05 uF anyway with 470 Kohm 6L6 grid resistors to ground. *I figured the -3dB point was low enough at 6.8 Hz! I know the datasheet says 'up to' 500k for grid leaks with cathode bias but I'd lower them to at least 220k for better bias stability. It's not quite so critical with driver tubes, because of the plate resistor, but output tube run away makes for bad juju. The Williamson used 100K grid resistors (on the KT66's) and 0.25 uF coupling caps. If I drop my 6L6 grid resistors to 100K (and used a larger coupling cap for say, -3 dB at 10 Hz), then they load the stage driver output impedance a bit more and reduce the gain a bit, as I want. *The I could do what Patrick said (and you clarified, thanks) and stablize the VLF, then add even more NFB to drop the gain further to where I need it for the WA-P2. Back to the schematic, pencil and calculator... Cheers, Roger Quick update: I replaced the "470K" (actually they were 400K) 6L6 grid leaks with 100K and left the 0.05 uF coupling caps in (-3 dB point now at 32 Hz.) Result: no sign of VLF instability. That's good news but I'd be tempted to bump those caps to .1uF, or even .22uF like in the Williamson, and lower the caps into the driver grids and here's why. NFB will try to make the output, times gain, equal the input which, of course, is why we're using NFB. But you're cutting the signal 3dB (or more, depending on the frequency) at the output tube grids with the coupling cap roll-off. That means the input gain stage is going to increase it's signal (at that frequency) by 3dB to compensate (NFB at work) and the same increase goes into the driver tubes. Now, a 3dB increase might not be a problem for your drivers but you can see that, sooner or later, depending on how much of a shelf you're doing, it can become a clipping problem because they have to swing a larger signal to get through that output tube coupling cap. And, in addition to potential clipping, increasing the output level, of everything before the roll-off, also increases distortion in those stages There's nothing you can do about the first gain stage and concertina (the signal level there is the smallest of the whole amp so there should be 'room' for it) but if you move the cut to the driver tube grids then they will be operating 'like normal'. *I can now increase the global NFB from a near unstable 33K resistor to about 15K with no signs of VLF surging. The larger NFB has reduced the overall gain to about what I need for the WA-P2 preamp, so I don't think I'll need a special LF roll-off coupling to the 6L6's. *So far win-win. I still have to fine tune the parallel NFB cap to cut square wave ringing (there's quite a lot!) and reconfirm all performance figures. You might want to consider shelving HF which, if you look at Patrick's schematic, is the cap, and it's series resistor, in parallel across the bottom resistor of the divider. I.E. that cap shunts HF, through the resistor in series with it, to ground making the divider ratio lower so it cuts HF into the LPT. For a Williamson the problems of overdrive at LF due to FB action boosting the error signal in input & concertina stages may be overcome by using the shelving network before the balanced amp, and where there is a huge amount of headroom. That's what I said. But you need to use TWO networks each 0.47uF then 1M//0.05uF, then 220k series grid bias R. Same as doing it at the output tubes. The highest pole in this network is determimed by the 0.05uF and 220k which gives -3dB at 14Hz, then the there is a second pole at about 2.7Hz when the response levels out at -14.9dB, then a third pole between 0.47uF and 1,220k at 0.28Hz. Phase shift between 14Hz and 1Hz is never more than +45 degrees and only increases to +90d at 0.28Hz when gain loss from other networks in the amp have reduced it to keep Nyquist happy. That may be fine for the amp in your example but that doesn't mean those poles are right for his. It is possible to put such shelf networks between balanced amp and OP tubes - it works OK, but that's where the balanced amp might become over driven as you point out. But fact is there is SFA signal in most music below 20Hz. He implied the pole needed was closer to 32Hz. There very well might not be a 'problem' shelving at the output grids but I see no good reason to put it there either. I've sometimes used TWO pairs of shelf networks where OPT inductance is too small as in the case of a Ming Da amp I'm re-engineering here for a customer. It has SET + LTP input stages followed by balanced amp with 300B driving 845 in PP, so there are 4 stages and 3 lots of C&R coupling between stages. Shelve networks were found to vastly improve bass and stability compared to the Chinese Dumbo Design. The shelf networks reduce gain where you just do not want much open loop gain nor do you want a large amount of NFB applied, ie, below 14Hz and above 20kHz. It should be noted that 'the purpose' given reduces NFB in that region and, as a result, increases distortion so you don't want to do 'more' than needed and the ideal case would be none at all, assuming you could make the amp stable with none at all. Well, I'm confident the OP will work it out for himself. The LF shelving networks I suggest, and where I put them, ie, never right before the OP tubes do work in 90% of cases where otherwise an amp oscillates at LF, especially when there isn't any load connected when OLG goes high, amount of NFB increases and, oops, the darn thing wants to oscillate at 0.5Hz. Patrick Turner. I've moved the LP roll-off filter to the grids of the driver pair (both 0.05 uF and 100K for -3dB at about 32 Hz ignoring phase splitter source impedance) and put the coupling to the 6L6 grids at 0.18 uF and 100K grid leak for a -3dB point at about 9 Hz (ignoring the about 8K source impedance from the driver plates.) No VLF instability seen. I plan to put the HF shelf network in the 1st stage plate circuit (still calculating it... the present roll off at about 64 KHz is too high), then tune the NFB cap for no oscillation on square waves (I hope!) Cheers, Roger |
|
#15
Posted to rec.audio.tubes
|
|||
|
|||
|
Best tap for NFB?
snip to appease the bean counters at Google.....
Well, I'm confident the OP will work it out for himself. The LF shelving networks I suggest, and where I put them, ie, never right before the OP tubes do work in 90% of cases where otherwise an amp oscillates at LF, especially when there isn't any load connected when OLG goes high, amount of NFB increases and, oops, the darn thing wants to oscillate at 0.5Hz. Patrick Turner. I've moved the LP roll-off filter to the grids of the driver pair (both 0.05 uF and 100K for -3dB at about 32 Hz ignoring phase splitter source impedance) and put the coupling to the 6L6 grids at 0.18 uF and 100K grid leak for a -3dB point at about 9 Hz (ignoring the about 8K source impedance from the driver plates.) * What exactly do you have between concertina a and k to grids of balanced driver? I use 0.47, then 1M // 0.047uF, then Rg biasing R = 220k. This gives low poles and good bass without instability. It works in all amps I've tried. No VLF instability seen. I plan to put the HF shelf network in the 1st stage plate circuit (still calculating it... the present roll off at about 64 KHz is too high), then tune the NFB cap for no oscillation on square waves (I hope!) Cheers, Roger I use a 50k pot in series with a double gang tuning cap from an old radio, 40pF to 700pF available. I run a 5kHz square wave at 1Vo without any load, assuming the amp won't oscillate at HF without a load and with say 15dB global NFB. Usually some *minimum* amount of CpF across the FB R may be needed to stop HF oscilations at least. The tuning gang may be used to determine this C value as well before other things. you may also need 22 ohms plus 0.047uF in series across the 16 ohm sec to give an R load at HF. But with the FBR C and zobel at output, amp should not oscillate at HF without a load. The best place for the R&C zobel to shelve HF gain is from V1 anode to 0V, or across the anode RLdc resistance, either way the tuning cap sees over 100Vdc across its plates so I use a 0.1uF also in series to block the Vdc, and 0.1 is a very low Z compared to the CpF you will want. With a 5kHz square wave you'll see overshoots. If you connect a 0.047uF across the OP maybe the amp oscillates maybe not, but if not the overshoots will increase in amplitude and ringing frequency will reduce. Adjust put and tuning C to minimise the ring, then try 0.1uF and 0.22uF and 0.47uF as loads Usually if there is some ringing but settling within 1/2 the width of the square wave flats and ringing amplitude is not more than 6 dB above the sq.wave flats the amp will remain stable with any other C value. With resistance load and C in parallel, ringing will reduce. With only R load, BW should extend to at least 35kHz, -3dB, and with a C added maybe a slight 3dB peak in response above 30kHz is seen but without a load it won't ever oscilate. The C&R act to lower the RL of V1 to perhaps 1/5 of normal value at 1kHz down to say 10k at 100kHz. So the Miller effect is pushed higher. I don't calculate this network; I tune it to be good. Patrick Turner. .. |
|
#16
Posted to rec.audio.tubes
|
|||
|
|||
|
Best tap for NFB?
On Jun 8, 3:09*am, Patrick Turner wrote:
snip to appease the bean counters at Google..... See below in your text after ***************** Cheers, Roger Well, I'm confident the OP will work it out for himself. *********** Thanks, will do! The LF shelving networks I suggest, and where I put them, ie, never right before the OP tubes do work in 90% of cases where otherwise an amp oscillates at LF, especially when there isn't any load connected when OLG goes high, amount of NFB increases and, oops, the darn thing wants to oscillate at 0.5Hz. Patrick Turner. I've moved the LP roll-off filter to the grids of the driver pair (both 0.05 uF and 100K for -3dB at about 32 Hz ignoring phase splitter source impedance) and put the coupling to the 6L6 grids at 0.18 uF and 100K grid leak for a -3dB point at about 9 Hz (ignoring the about 8K source impedance from the driver plates.) * What exactly do you have between concertina a and k to grids of balanced driver? I use 0.47, then 1M // 0.047uF, then Rg biasing R = 220k. This gives low poles and good bass without instability. It works in all amps I've tried. ************** The coupling from concertina phase splitter to drive grids is now 0.05 uF and 100K to ground. No "VLF shelf" in place. VLF stable, no 0.5 to 1 Hz surges. Driver to 6L6 grids is 0.18 uF and 100K. No VLF instability seen. I plan to put the HF shelf network in the 1st stage plate circuit (still calculating it... the present roll off at about 64 KHz is too high), then tune the NFB cap for no oscillation on square waves (I hope!) Cheers, Roger I use a 50k pot in series with a double gang tuning cap from an old radio, 40pF to 700pF available. I run a 5kHz square wave at 1Vo without any load, assuming the amp won't oscillate at HF without a load and with say 15dB global NFB. Usually some *minimum* amount of CpF across the FB R may be needed to stop HF oscilations at least. The tuning gang may be used to determine this C value as well before other things. you may also need 22 ohms plus 0.047uF in series across the 16 ohm sec to give an R load at HF. ***************** The trick with the tuning cap is what I do for the NFB resistor bypass cap. Good idea to do it for the HF shelf setting, too, with pot, so will do this. Since it is in the 1st audio plate circuit your 0.1 uF isolator is well noted (see below)... and keep fingers off the cap frame, hot! Will add 22 ohms and 0.05 uF across the 16 ohm sec, too. Presently I have just 1.5K as a token load. But with the FBR C and zobel at output, amp should not oscillate at HF without a load. The best place for the R&C zobel to shelve HF gain is from V1 anode to 0V, or across the anode RLdc resistance, either way the tuning cap sees over 100Vdc across its plates so I use a 0.1uF also in series to block the Vdc, and 0.1 is a very low Z compared to the CpF you will want. With a 5kHz square wave you'll see overshoots. If you connect a 0.047uF across the OP maybe the amp oscillates maybe not, but if not the overshoots will increase in amplitude and ringing frequency will reduce. Adjust put and tuning C to minimise the ring, then try 0.1uF and 0.22uF and 0.47uF as loads Usually if there is some ringing but settling within 1/2 the width of the square wave flats and ringing amplitude is not more than 6 dB above the sq.wave flats the amp will remain stable with any other C value. With resistance load and C in parallel, ringing will reduce. With only R load, BW should extend to at least 35kHz, -3dB, and with a C added maybe a slight 3dB peak in response above 30kHz is seen but without a load it won't ever oscilate. The C&R act to lower the RL of V1 to perhaps 1/5 of normal value at 1kHz down to say 10k at 100kHz. So the Miller effect is pushed higher. I don't calculate this network; I tune it to be good. Patrick Turner. .. *************** All noted. I'll see if I need the VLF shelf. Presently the 0.05 uF/100K combination in front to the driver is stable, but your bypassed 1M and 220K does seem like a more elegant idea... will consider. And thanks for the advice, much appreciated. "My Scratch 15" is not a "Turner 5050" but I want it to be at least decent. ************* Cheers, Roger |
|
#17
Posted to rec.audio.tubes
|
|||
|
|||
|
Best tap for NFB?
************** *The coupling from concertina phase splitter to drive grids is now 0.05 uF and 100K to ground. *No "VLF shelf" in place. VLF stable, no 0.5 to 1 Hz surges. Driver to 6L6 grids is 0.18 uF and 100K. Try to check LF response at low level, say 1Vrms into nominal load with input testing signal kept level down to 2Hz at least. With GNFB, often the response below 50Hz will have a peak before the response rolls off quickly at a rate exceeding 12dB/octave. This peak might be between 5Hz and 25Hz and if more than +3dB is undesirable. The LF shelving network should give you a flatter bass response. No VLF instability seen. ***************** *The trick with the tuning cap is what I do for the NFB resistor bypass cap. *Good idea to do it for the HF shelf setting, too, with pot, so will do this. * The series R is essential with the C, and square wave ringing is reduced to a minimum by adjustment of BOTH R and C. Since it is in the 1st audio plate circuit your 0.1 uF isolator is well noted (see below)... and keep fingers off the cap frame, hot! *Will add 22 ohms and 0.05 uF across the 16 ohm sec, too. *Presently I have just 1.5K as a token load. 0.1uF may be too low a C value. Try 0.47uF, and don't worry about letting the moths from your purse because they'll all find a good home elsewhere. the "token load" of 1k5 is a trivial and useless load at the OPT sec. The Zobel with 22 ohms plus 0.047 will present an increasingly resistive load above 310kHz. The Z of Zobel at its pole will be 31 ohms at 153kHz. The 0.047 + 22r is where I'd start, and then increase C. If C was 0.22uF, pole is at 33kHz, and Z becomes mainly resistive by 66kHz. Sometimes the output sec Zobel just cannot help to prevent oscillations. Be prepared for doing things which are renowned to help but don't. Sometimes OPTs people try to use are very very poor. Sometimes you have to place a Zobel across each 1/2 primary of the OPT. If nominal RLa-a = 6k, then one might use network of 3k9 + 0.001 uF across each 1/2 primary, ie, from CT/B+ to each anode But with the FBR C and zobel at output, amp should not oscillate at HF without a load. The best place for the R&C zobel to shelve HF gain is from V1 anode to 0V, or across the anode RLdc resistance, either way the tuning cap sees over 100Vdc across its plates so I use a 0.1uF also in series to block the Vdc, and 0.1 is a very low Z compared to the CpF you will want. With a 5kHz square wave you'll see overshoots. If you connect a 0.047uF across the OP maybe the amp oscillates maybe not, but if not the overshoots will increase in amplitude and ringing frequency will reduce. Adjust put and tuning C to minimise the ring, then try 0.1uF and 0.22uF and 0.47uF as loads Usually if there is some ringing but settling within 1/2 the width of the square wave flats and ringing amplitude is not more than 6 dB above the sq.wave flats the amp will remain stable with any other C value. With resistance load and C in parallel, ringing will reduce. With only R load, BW should extend to at least 35kHz, -3dB, and with a C added maybe a slight 3dB peak in response above 30kHz is seen but without a load it won't ever oscilate. The C&R act to lower the RL of V1 to perhaps 1/5 of normal value at 1kHz down to say 10k at 100kHz. So the Miller effect is pushed higher. I don't calculate this network; I tune it to be good. Patrick Turner. . *************** *All noted. I'll see if I need the VLF shelf. Presently the 0.05 uF/100K combination in front to the driver is stable, but your bypassed 1M and 220K does seem like a more elegant idea... will consider. *And thanks for the advice, much appreciated. "My Scratch 15" is not a "Turner 5050" but I want it to be at least decent. The first amps I made were pretty damn awful. I was trying to use really bad OPTs taken from junk I found at the local tip. After awhile, my amps improved, especially when I made a lathe to wind my own OPTs which were better than anything I could buy. But I've never attracted a willing apprentice - they realise they'd never make any money, and young blokes always want lotsa money because she who sits upon a treasure demands it or she won't display her jewels. ************* *Cheers, Roger- Hide quoted text - Patrick Turner. |
![Engineer[_2_] is offline](images/SultanThemeVB3R/statusicon/user_offline.gif)
Linear Mode
