| Home |
| Search |
| Today's Posts |
|
#41
Posted to rec.audio.tubes
|
|||
|
|||
12AX7 grid current
Ian Bell wrote:
To tackle the "mystery" of the distortion on the grid I suggested to temporarily block the plate of 12AX7 to ground by an electrolytic, say 10uF. This will reduce the gain to practically zero. After that remeasure distortion on the grid. If it reduces to nearly zero (as with the tube pulled out or cold) then it will show that the Miller effect creates (or rather magnifies) this virtual distortion. If the distortion is still high, then it is some non-linear conductance of the grid current or nonliarity of the grid capacitance due to pulsating electron cloud (space charge). Apart from a scientific curiosity you can derive something result from this experiment. For instance, it it proves Miller effect cause, then it will further inspire you to try EF86, which will not have any Miller effect at all. I would be surprised if it were a Miller effect but I will do the experiment you suggest and let you know the results. Cheers Ian The trouble with experiments is you get results and this time they are not as I expected, nor indeed I suspect as anyone expected. To recap with 380mV on the grid at 2KHz I got the following distortion figures at the grid: 2H -49, 3H -62, 4H -70, 5H -77 I added a 22uF 400V electrolytic from plate to ground and the figures became: 2H -60, 3H -70, 4H -77 So the distortion is reduced but by no means to zero. 2H is reduced by 11dB, 3H by 8dB and 4H by 7dB. I am not sure what the mechanism is - all I feel safe in saying is that adding a very low value ac plate load appears to reduce the distortion measured at the grid. I keep making my point about grid current rising as anode voltage falls but no-one's picked up on it. A grid is a grid, be it input or screen. Doesn't it have a knee? Where's the knee on an EF86? As usual, I could be totally wrong... Listen to Alex more. His idea may not have solved your problem, but it's made a key contribution to analysis. Ian |
|
#42
Posted to rec.audio.tubes
|
|||
|
|||
|
12AX7 grid current
"Ian Bell" wrote in message ... Patrick Turner wrote: On Sep 24, 6:34 pm, wrote: "Ian wrote in message ... I just spent some time perusing the 12AX7 data sheets in more detail, particularly the part about RC coupled amplifiers which gives the maximum output before the onset of grid current under various conditions. Based on that data I calculated the cathode voltage and also the peak input voltage before grid current for several values of Ra, Rk and Va. Subtracting the two gives the grid bias at the onset of grid current and this seems to be pretty constant at about -0.6V. It also seems that with the 300V HT I have there is really only one operating point that will give me the 102V peak output swing I need without reaching grid current and that is with a 220K anode load, 2.2K up the cathode leg and a quiescent current of close to 0.6mA. I will try those values and see what I get. That's very close to what the "Muiderkring Electronic Tubes Handbook" (1968/p152) specifies for an ECC83 @ 300Vb and a Vo of 36Veff/102Vpp. Ra 220K R'g1 680K (load) Rk 2K2 Ia 0.63mA Vout 36Veff A 44x d 3.6% Rg 1M The same page specifies 11 other operating points for the ECC83. Can't find the book online but linked below is an equivalent.http://www.mif.pg.gda.pl/homepages/f.../054/2/288.pdf Rgds, Gio From the above list of performance figures we can see that there is about 1% THD per 10Vrms of output. But what is the series input resistance and to what extent would series source resistance of 100k contribute to the distortion figure of 3.6% at 36vrms output? And the answer is as follows: I used 200K in the anode and 2K2 in the cathode. HT = 313V, Plate = 170V, cathode = 1.407V With 125mV rms on the grid (2KHz) via 100K I get: 2H -59dB, 3H -77dB 4H unmeasurable The signal voltage at the plate was 10V rms which means the stage gain was 80. At the plate the measured distortion was: 2H -49dB, 3H -75dB, 4H unmeasurable So the tube itself mainly contributes 10dB of 2H The input signal was then changed to give 30V rms at the plate. Plate signal distortion was then: 2H -34dB, 3H -43dB, 4H -49dB, 5H -50dB, 6H -55dB, 7H -58dB And under the same conditions looking at the transformer secondary into 100R we get: 2H -34dB, 3H -41dB, 4H -48dB, 5H -49dB, 6H -53dB, 7H -56dB Which means the white follower plus the transformer is only contributing a couple of dB to the distortion. Also, at this point the signal voltage on the grid was just under 380mV rms and the distortion at the grid was: 2H -49, 3H -62, 4H -70, 5H -77, 6H unmeasurable which shows we are approaching grid distortion but the tube distortion is still dominant. As I mentioned before, the reason for the 100K series input resistor so I can apply shunt/shunt NFB. Connecting 2Meg from the white output back the the 12AX7 grid drops the output from 2.5V rms to 0.6V rms i.e. there's about 12dB of NFB. I increased the input to give 2.5V rms at the transformer secondary and the distortion at the transformer secondary was then: 2H-45dB, 3H -50dB, 4H -55dB, 6H -58dB, 7H -61dB which is about 11dB lower overall as expected. So, this basically seems to work. At 30V rms the 12AX7 produces about 2% distortion (-34dB) which compares well with the data sheet figure of 2.5% at 367V rms. The White follower and transformer do not seem to add significantly to this. NFB gives the expected overall reduction to just under 0.6%. I would still like the distortion to be lower so I may well try the EF86. I am not sure how much there is to be gained. From the data sheet I have, with 220K in the anode, 300V supply and 1Meg to the screen grid, this will give a stage gain of 188 times or about 7.5dB more open loop gain. Distortion is 5% at 56V rms output so if this is proportional to level, then at 30V rms output I would expect it to be about 2.7% which is close to what the 12AX7 produces. However, the extra 6dB of open loop gain means the NFB is increased by the same amount so the overall distortion should be about 6dB lower so I hope it will get below the 0.3%. Cheers Ian Hello Ian, To tackle the "mystery" of the distortion on the grid I suggested to temporarily block the plate of 12AX7 to ground by an electrolytic, say 10uF. This will reduce the gain to practically zero. After that remeasure distortion on the grid. If it reduces to nearly zero (as with the tube pulled out or cold) then it will show that the Miller effect creates (or rather magnifies) this virtual distortion. If the distortion is still high, then it is some non-linear conductance of the grid current or nonliarity of the grid capacitance due to pulsating electron cloud (space charge). Apart from a scientific curiosity you can derive something result from this experiment. For instance, it it proves Miller effect cause, then it will further inspire you to try EF86, which will not have any Miller effect at all. If it proves to be grid current nonlinearity, then little can be done other than trying different tubes and possibly reducing the heater voltage... Alex |
|
#43
Posted to rec.audio.tubes
|
|||
|
|||
|
12AX7 grid current
On Sep 26, 1:23*am, Ian Bell wrote:
Patrick Turner wrote: On Sep 24, 6:34 pm, *wrote: "Ian *wrote in message ... I just spent some time perusing the 12AX7 data sheets in more detail, particularly the part about RC coupled amplifiers which gives the maximum output before the onset of grid current under various conditions. Based on that data I calculated the cathode voltage and also the peak input voltage before grid current for several values of Ra, Rk and Va. Subtracting the two gives the grid bias at the onset of grid current and this seems to be pretty constant at about -0.6V. It also seems that with the 300V HT I have there is really only one operating point that will give me the 102V peak output swing I need without reaching grid current and that is with a 220K anode load, 2.2K up the cathode leg and a quiescent current of close to 0.6mA. I will try those values and see what I get. That's very close to what the "Muiderkring Electronic Tubes Handbook" (1968/p152) specifies for an ECC83 @ 300Vb and a Vo of 36Veff/102Vpp. Ra * 220K R'g1 680K (load) Rk * 2K2 Ia * 0.63mA Vout 36Veff A * *44x d * *3.6% Rg * 1M The same page specifies 11 other operating points for the ECC83. Can't find the book online but linked below is an equivalent.http://www.mif.pg.gda.pl/homepages/f.../054/2/288.pdf Rgds, Gio *From the above list of performance figures we can see that there is about 1% THD per 10Vrms of output. But what is the series input resistance and to what extent would series source resistance of 100k contribute to the distortion figure of 3.6% at 36vrms output? And the answer is as follows: I used 200K in the anode and 2K2 in the cathode. HT = 313V, Plate = 170V, cathode = 1.407V With 125mV rms on the grid (2KHz) via 100K I get: 2H -59dB, 3H -77dB 4H *unmeasurable Let us remind readers that THD = 20 x log ( THD voltage / output signal voltage ) And % THD = 100 x ( THD voltage / output signal voltage ) So THD of -20dB is 10% THD, -30dB = 3.16%, -40dB = 1.0%, -50dB = 0.316%, and -60dB = 0.1%, and -70dB = 0.032%. So with Rsource = 100k, and 10Vrms anode output, THD = 0.1% approx. The signal voltage at the plate was 10V rms which means the stage gain was 80. At the plate the measured distortion was: 2H -49dB, 3H -75dB, 4H unmeasurable But now anode THD = 0.3% approx..... So the tube itself mainly contributes 10dB of 2H I'm having trouble following you. I would have thought that **without** the 100k Rsource the THD is 0.1% and **with** 100k it increases to to 0.3% and all this when you have only 0.125Vrms grid signal applied to produce 10Vrms at the anode in both cases. The input signal was then changed to give 30V rms at the plate. Plate signal distortion was then: 2H -34dB, 3H -43dB, 4H -49dB, 5H -50dB, 6H -55dB, 7H -58dB So THD is still dominated by the 2H. you could take all other H away from THD and the measurement of THD won't change much because the THD = square root of ( sum of the of each harmonic voltage squared ). Anyway, THD would be about -30dB from what you say, or about 3%, so 1% per 10Vrms. But **at 10Vrms**, the THD could be expected to be 1%, but with/ without 100k Rsource it measured much lower, which suggests a very non- linear increase in THD, and that at low outputs, the 12AX7 is fairly blameless, and so mayve 99% of manufacturers never cared a hoot about the source resistance in front of any amp stage using a 12AX7. It would seem there is a considerable gain in the form of much lower THD to be had at lowish levels by keeping Rsource nice and low, and so a graph could be with THD versus Rsource where Vout = 10Vrms with Rsource values ranging from say 10k to 1M. I suspect maybe not much fall in THD may occur where Rsource was lower than 10k. This may mean that where you have a CD player feeding the 12AX7, then the gain pot would best be a 20k type which is very easily driven by a 1Vrms signal from a CD player or a preamp of some kind with typical follower output. The 20k pot offers a max Rsource of only 5k if the input to the pot is say a low 600 ohm source. I might add that I have offered a few customers a 20k log DACT switched attenuator in a box to drive a pair of power amps from a CD player and I have never heard the complaint that "it doesn't sound as good as a preamp". And under the same conditions looking at the transformer secondary into 100R we get: 2H -34dB, 3H -41dB, 4H -48dB, 5H -49dB, 6H -53dB, 7H -56dB Which means the white follower plus the transformer is only contributing a couple of dB to the distortion. The White follower IS GOOD, at low levels. In most tube amps driving headphones or speakers, most THD is generated in the OP tubes amd least in IP tubes. But where you have a follower OP stage and the IP stage has to make a high signal voltage, bingo, you get more Dn than is wanted. In many "conventional" tube amps for speakers there is often a resistance divider for phone use. The R divider is often say 56R + 8R so that when phones are used the phone loading does not change the amp load much at all and the load the OP tubes see is about 7 times higher than what the tubes and OPT were mainly meant for, ie, an 8 ohm speaker. So with a much higher tube load, the OP stage gain becomes high and the total amount of GNFB becomes close to maximal and the amp works in pure class A so THD of the OP tubes then drops to much lower levels with phone compared to when speakers are used, if the comparison is made at the speaker terminals and the voltage is the same speaker use or phone use. In fact it is fairly close to being the same because if you have 1V output for a speaker and plug in the phones, they will usually switch out the speakers and the phone may get say 1/7 Vrms, which may be plenty for the same percieved volume level to the listener. The function of the 56R+8R divider also offers a great betterment to SNR. Also, at this point the signal voltage on the grid was just under 380mV rms and the distortion at the grid was: 2H -49, 3H -62, 4H -70, 5H -77, 6H unmeasurable which shows we are approaching grid distortion but the tube distortion is still dominant. As I mentioned before, the reason for the 100K series input resistor so I can apply shunt/shunt NFB. Connecting 2Meg from *the white output back the the 12AX7 grid drops the output from 2.5V rms to 0.6V rms i.e. there's about 12dB of NFB. I increased the input to give 2.5V rms at the transformer secondary and the distortion at the transformer secondary was then: 2H-45dB, 3H -50dB, 4H -55dB, 6H -58dB, 7H -61dB which is about 11dB lower overall as expected. So, this basically seems to work. At 30V rms the 12AX7 produces about 2% distortion (-34dB) which compares well with the data sheet figure of 2.5% at 367V rms. The White follower and transformer do not seem to add significantly to this. NFB gives the expected overall reduction to just under 0.6%. I would still like the distortion to be lower so I may well try the EF86. I am not sure how much there is to be gained. From RDH4, and their pages on comparisons between signal amps using triode or pentode connected tubes, the triode connection gives more THD at lower levels up to about 14vrms when T or P THD becomes equal and above 14vrms the triode connection is superior. Unfortunately, RDH4 isn't being very fair at all because they should have really compared say a 12AX7 with specified Rsource with a pentode such as EF86 with the same Rsource and with both set up to give approximately the same gain. I have a feeling RCA who sponsored the writing of RDH4 wanted folks to buy more of the many pentodes they made rather than stay with the triodes which were regarded as the audio gold standard before RDH4 was written. I suggest the EF86 works with the least change of gm for a +/- Ia swing with Ea at about 150Vdc and Ia at 3mA. If the B+ is +300Vdc, then the VdcRL = 150V so RLdc must be 50k and with following Rg = 470k the total RL = 45k approx so gain will be about 90, or very similar to a 12AX7. But just what sort of harmonic mix you may measure is anyone's guess because pentodes are renowned for higher % of odd number H products. And trying to get say 36Vrms at low THD with any pentode to drive a low THD output stage may become much more difficult than say using a 6CG7 triode, especially if the triode had a CCS RLdc so it only has to produce ac power into the Rg of say 470k. From the data sheet I have, with 220K in the anode, 300V supply and 1Meg to the screen grid, this will give a stage gain of 188 times or about 7.5dB more open loop gain. Distortion is 5% at 56V rms output so if this is proportional to level, then at 30V rms output I would expect it to be about 2.7% which is close to what the 12AX7 produces. However, the extra 6dB of open loop gain means the NFB is increased by the same amount so the overall distortion should be about 6dB lower so I hope it will get below the 0.3%. Proof of the pudding is in the eating. So far only marginal benefits might be had with a pentode. But maybe you could try say a 6BX6 which has twice the gm and gain of the EF86 . The use of 12dB global NFB probably isn't enough to "dull the sound" which audiophiles may like to think always happens when NFB is used. I often think audiophiles are wrong about NFB, if the listening tests I've witnessed are any guide. Methinks a two stage amp with say 12AX7 IP tube and SE EL34 ( yep, EL34 !) in triode and normal speaker OPT and with 12dB GNFB does give rather glorious sound and I use such a beast in my kitchen radio using a lousy old OPT from an old tape recorder/player from the 1950s and a Deluxe Rola 12" speaker plus dome tweeter for a fairly flat response and sensitivity of about 95dB/W. With headphones, there is far more ability than is required. With a cascade circuit where you have a 12AX7 driving an SET OP stage the 2H of the IP tube cancels the 2H produced by the OP tube. Its not a huge cancelation, and the 3H is additive, but 3H is damn low anyway. With a very high RL on the EL34 OP tube it then produces very little 2H and more cancelation of 2H becomes possible. I would have thought a pair of EL84 or even better, a pair of EL86 in a White follower stage would produce far less THD than your 12BH7. Maybe then a loop of series voltage GNFB only around the driver cascade pair of triodes such as 6CG7 would then give you a very linear 36Vrms drive to a White OPstage. Then I thought of using a 6080 or 6AS7G with each half rigged to make a White follower with Ia = 60mA, Ea = 100Vdc for both triodes and without an OPT at all but when you draw the load lines it looks real bad for THD because each tube may see 64 ohms with 32 ohm phones and a 64 ohm RL is less than 1/2 Ra and a poor load match. Its doable, but gain is rather low, and you'd need lots of NFB to do as well as with a normal anode loaded SET OP stage with OPT. Enter 6C33C onto the Stage. The less said the better :-) Of course with a suitable OPT with primary having a CT, you could always rig the two driver triodes you do have in the Circlotron connection, and with a PP LTP driver stage with CCS cathode sink. The amount of NFB in the OP stage is high, and both triodes always see equal load conditions because there is none of that "current regulation signal" from the top R of a White follower, and load change to the White causes a different loading condition in each triode. The same happens in a SRPP gain circuit for any OP stage. The Circlotron requires 4 very quiet B+ rails for a stereo amp, with good electrostatic shielding between mains and HT windings, so it is a challenge. If not a Circlotron, you are left with the other option of using a McIntosh style OP stage which needs two equal primary windings each with a CT, but only ONE B+ rail is needed for 2 channels. Again, a special OPT may be needed. The other simpler alternative to a White follower driver to an OPT is to simply use both triodes in the 12BH7 paralleled and with a common CCS to a -200V, and with grid bias of 0V, so that the Vswing can be quite a lot higher than when you have two triodes in series between say a rail of +250V. The OPT can be cap coupled to the commoned cathodes and the THD will be mainly 2H and 3H maybe lower than the White. The mainly 2H of the 12AX7 should more likely cancel the 2H of the OP stage and maybe the outcome could be better than a White. With Ea at say 200V, there is a limit for Ia lest Pda get a bit big for 12BH7 and so I would be inclined to head for an EL86 or EL84 where Ia could happily be 40mA for Ea at 200V, giving a Pda of 8W which is well within Pda ratings for the tubes. Some folks would just use a 6BM8 or 6GW8 with a normal OPT, with pentode section trioded. Patrick Turner. Cheers Ian- Hide quoted text - - Show quoted text - |
|
#44
Posted to rec.audio.tubes
|
|||
|
|||
|
12AX7 grid current
On Sep 27, 1:24*am, Ian Bell wrote:
Ian Bell wrote: Alex wrote: "Ian wrote in message ... Patrick Turner wrote: On Sep 24, 6:34 pm, wrote: "Ian wrote in message ... I just spent some time perusing the 12AX7 data sheets in more detail, particularly the part about RC coupled amplifiers which gives the maximum output before the onset of grid current under various conditions. Based on that data I calculated the cathode voltage and also the peak input voltage before grid current for several values of Ra, Rk and Va. Subtracting the two gives the grid bias at the onset of grid current and this seems to be pretty constant at about -0.6V. It also seems that with the 300V HT I have there is really only one operating point that will give me the 102V peak output swing I need without reaching grid current and that is with a 220K anode load, 2.2K up the cathode leg and a quiescent current of close to 0.6mA. I will try those values and see what I get. That's very close to what the "Muiderkring Electronic Tubes Handbook" (1968/p152) specifies for an ECC83 @ 300Vb and a Vo of 36Veff/102Vpp. Ra 220K R'g1 680K (load) Rk 2K2 Ia 0.63mA Vout 36Veff A 44x d 3.6% Rg 1M The same page specifies 11 other operating points for the ECC83. Can't find the book online but linked below is an equivalent.http://www.mif.pg.gda.pl/homepages/f...054/2/288..pdf Rgds, Gio From the above list of performance figures we can see that there is about 1% THD per 10Vrms of output. But what is the series input resistance and to what extent would series source resistance of 100k contribute to the distortion figure of 3.6% at 36vrms output? And the answer is as follows: I used 200K in the anode and 2K2 in the cathode. HT = 313V, Plate = 170V, cathode = 1.407V With 125mV rms on the grid (2KHz) via 100K I get: 2H -59dB, 3H -77dB 4H unmeasurable The signal voltage at the plate was 10V rms which means the stage gain was 80. At the plate the measured distortion was: 2H -49dB, 3H -75dB, 4H unmeasurable So the tube itself mainly contributes 10dB of 2H The input signal was then changed to give 30V rms at the plate. Plate signal distortion was then: 2H -34dB, 3H -43dB, 4H -49dB, 5H -50dB, 6H -55dB, 7H -58dB And under the same conditions looking at the transformer secondary into 100R we get: 2H -34dB, 3H -41dB, 4H -48dB, 5H -49dB, 6H -53dB, 7H -56dB Which means the white follower plus the transformer is only contributing a couple of dB to the distortion. Also, at this point the signal voltage on the grid was just under 380mV rms and the distortion at the grid was: 2H -49, 3H -62, 4H -70, 5H -77, 6H unmeasurable which shows we are approaching grid distortion but the tube distortion is still dominant. As I mentioned before, the reason for the 100K series input resistor so I can apply shunt/shunt NFB. Connecting 2Meg from the white output back the the 12AX7 grid drops the output from 2.5V rms to 0.6V rms i.e. there's about 12dB of NFB. I increased the input to give 2.5V rms at the transformer secondary and the distortion at the transformer secondary was then: 2H-45dB, 3H -50dB, 4H -55dB, 6H -58dB, 7H -61dB which is about 11dB lower overall as expected. So, this basically seems to work. At 30V rms the 12AX7 produces about 2% distortion (-34dB) which compares well with the data sheet figure of 2.5% at 367V rms. The White follower and transformer do not seem to add significantly to this. NFB gives the expected overall reduction to just under 0.6%. I would still like the distortion to be lower so I may well try the EF86. I am not sure how much there is to be gained. From the data sheet I have, with 220K in the anode, 300V supply and 1Meg to the screen grid, this will give a stage gain of 188 times or about 7.5dB more open loop gain. Distortion is 5% at 56V rms output so if this is proportional to level, then at 30V rms output I would expect it to be about 2.7% which is close to what the 12AX7 produces. However, the extra 6dB of open loop gain means the NFB is increased by the same amount so the overall distortion should be about 6dB lower so I hope it will get below the 0.3%. Cheers Ian Hello Ian, To tackle the "mystery" of the distortion on the grid I suggested to temporarily block the plate of 12AX7 to ground by an electrolytic, say 10uF. This will reduce the gain to practically zero. After that remeasure distortion on the grid. If it reduces to nearly zero (as with the tube pulled out or cold) then it will show that the Miller effect creates (or rather magnifies) this virtual distortion. If the distortion is still high, then it is some non-linear conductance of the grid current or nonliarity of the grid capacitance due to pulsating electron cloud (space charge). Apart from a scientific curiosity you can derive something result from this experiment. For instance, it it proves Miller effect cause, then it will further inspire you to try EF86, which will not have any Miller effect at all. I would be surprised if it were a Miller effect but I will do the experiment you suggest and let you know the results. Cheers Ian The trouble with experiments is you get results and this time they are not as I expected, nor indeed I suspect as anyone expected. To recap with 380mV on the grid at 2KHz I got the following distortion figures at the grid: 2H -49, 3H -62, 4H -70, 5H -77 I added a 22uF 400V electrolytic from plate to ground and the figures became: 2H -60, 3H -70, 4H -77 So the distortion is reduced but by no means to zero. 2H is reduced by 11dB, 3H by 8dB and 4H by 7dB. I am not sure what the mechanism is - all I feel safe in saying is that adding a very low value ac plate load appears to reduce the distortion measured at the grid. With an anode load of say 100ohms only and electrolytic bypassing any other RL to 0V, then the Ia can be measured for THD in the signal across the 100ohms. Its not very easy to do with low Ia changes of say 1mA which produces such a very low signal across 100R, only 0.1mV, and if THD in the current flow as 1%, then THD = 0.01mV, and so plotting a graph of THD against tube current increase with various Rsource values at the grid becomes as difficult as chasing a butterfly ina Cessna. But you maybe can just measure dc changes at low levels if you have a good dc amp and meter. and dc linearity can thus be measured as can the dc voltage across the Rsource to grid, if any. With very low RL values of say 100ohms, the load line one would see drawn on the tube data Ea / Ia curves is almost a vertical line. One will find Ia varying with Eg1 at a very non linear fashion, or to a "2/3 rule" as ppl say, something involving the cube root of something squared. The THD generated with a very low RL is the maximum you get and is the same for where have a real triode or pentode and there is no Miller and no internal triode NFB due to the triode gain when RL is a high value. Pentodes operate with high current THD regardless of their RL for a given same Ia swing because there is very low internal NFB because it is prevented by the action of the screen if it is held at a constant B+ relative to the cathode. Usually I have found pentodes to always give higher THD than triodes where one wants to use a pentode in a similar situation and with similar loading and gain as for the triode case. With pentodes in signal amps, the lower the Ia change the less the THD -generally. And lower gm pentodes are often more linear than high gm types and just how one sets them up can vary the outcome enormously. High gm pentodes can have enormous gain with a moderate RL value. One might reduce this excessive gain if one uses an unbypassed Rk, and reduce the THD. But to get enough "degenerative" local current FB with a pentode the RL must be kept low enough to get current in Rk and hence a FB voltage at the k. The lower the RL the greater the THD, so only so much can be achieved. Some wicked reductions of gain and THD may be had where you make a voltage amp using a mu-follower where the top tube is a triode and with say 15k Rk between its k to the bottom tube anode and where the bottom tube is a high gm pentode. One may get open loop gain of over 700. With shunt FB it can be reduced to say 70, and THD is literally decimated, and Rout from the top triode cathode becomes much lower than if that triode were just a normal cathode follower. Anyway, from the anode curves for a triode, we can see that where the operating Ia dc point is as high as possible the changes to gm for +/- changes to Eg1 become lower as the Ra lines become more parallel to each other. Or put another way, as the Ia dc point is lowered down into where the Ra curves for values of Eg1 are well curved, the gm variations and THD become much higher. Not only that, from the triode curves with RL = to a vertical line, ie, RL = a very low value, the Ia change for Eg1 change are little different to a typical pentode, and not very linear. Where Ia change is minimised, so to will be THD for both pentodes and triodes for where voltage change is not excessive, but where Ia change is very low with a load close to a CCS, the triode become superior because of its internal NFB. I recently replaced EF86 in a pair of Quad-II amps with Polish made EF80. I increased Ia in these driver pentodes by about 2.5 times that of EF86 and arranged the EF80 as a real LTP rather than the original quasi LTP with paraphase positive FB to increase gain. The PSU was completely revised. The THD went lower for the same amount of applied NFB. The sound was exceptional with either KT66, EL34, KT88, 6550 etc. Despite EF80 maybe not looking quite so linear as EF86 from the curves, they work well if set up well with just the right value of Eg2. From pentode curves it is very difficult to SEE just what harmonic products might be expected. The 2H is always the most easy to see and estimate from load lines drawn, but other odd H harmonics can really only be seen when you measure a particular tube with a certain load and Ia Ea condition. What you measure is what you get, and low Rsource and high Rin should always dominate any amp design, and thus help the music to survive, IMHO. Patrick Turner. Cheers Ian- Hide quoted text - - Show quoted text - |
|
#45
Posted to rec.audio.tubes
|
|||
|
|||
|
12AX7 grid current
On Sep 27, 1:53*am, Ian Bell wrote:
Don Pearce wrote: On Sun, 26 Sep 2010 10:36:07 +0100, Ian wrote: I would be surprised if it were a Miller effect but I will do the experiment you suggest and let you know the results. If it is Miller it will be strongly frequency dependent. That should be easy to check. d Looks like there is some frequency dependent. I made additional tests at 100Hz, 500Hz and 4KHz as follows (grid distortion in original circuit): 100Hz * 2H -63dB, 3H -65dB 500Hz * 2H -58dB, 3H -61dB 2KHz * *2H -49dB, 3H -62dB 4KHz * *2H -46dB, 3H -65dB Looks to me like the second harmonic rises with frequency but the third varies little so I would guess the 2H is coupled somehow from the plate and the 3H is probably the onset of real grid current distortion. It might be interesting to repeat this with a grounded grid configuration.. But then the CD player source or other source has to work into the low Rin of a cathode circuit. Grounded grid can be good for MC input but then you have problems with the Ek and dc blocking, so a cap coupling is needed and the cap value is large, and usually must be an electrolytic, maybe a tantalum for small size bypassed with plastic C and I still don't like it. But then there is the use of a 2SK369 in CASCODE with the triode which can be 1/2 6DJ8, 6CG7, 12AT7 or 12AU7 or a trioded 6EJ7, 6BX6 etc, because Ia needs to be 5mA. 12AX7 cannot be used in such a situation. Miller may be made low, gain controlled nicely with local current FB and THD not too bad. Noise for headphones will be lower than any 12AX7 might give. AFAIK, the j-fet has no detrimental issues when using high Rsource values and shunt NFB could be used OK. J-fet input impedance does have some C component but is otherwise a very high impedance like most medium µ triodes. J-fet THD is like a triode THD with mainly 2H but it does produce about 1% per volt of output, so that 5Vrms at a drain gives 5%. This is to be expected from such a miniscule device working with an Ed of only about 10Vdc. Open loop gain with j-fet = fet gm x tube RL, and if tube RL = 20k and fet gm is 40mA/V, then OLG = 800. If current FB in the fet source circuit reduces this to say 80 and equal to a 12AX7, and where you have 1V at the drain driving cathode with 6DJ8, then you may find the closed loop THD with a cascode to be less than any 12AX7 on its own, and being forced to produce 36Vrms under duress. Last time I bought about 20 x 2SK369 they cost me $1.20 each and all had closely matched gm, so close that I could get good channel balance without any global NFB and very little local current FB. Patrick Turner. Cheers Ian |
|
#46
Posted to rec.audio.tubes
|
|||
|
|||
|
12AX7 grid current
On Sep 27, 5:02*am, "Ian Iveson"
wrote: Ian Bell wrote: To tackle the "mystery" of the distortion on the grid I suggested to temporarily block the plate of 12AX7 to ground by an electrolytic, say 10uF. This will reduce the gain to practically zero. After that remeasure distortion on the grid. If it reduces to nearly zero (as with the tube pulled out or cold) then it will show that the Miller effect creates (or rather magnifies) this virtual distortion. If the distortion is still high, then it is some non-linear conductance of the grid current or nonliarity of the grid capacitance due to pulsating electron cloud (space charge). Apart from a scientific curiosity you can derive something result from this experiment. For instance, it it proves Miller effect cause, then it will further inspire you to try EF86, which will not have any Miller effect at all. I would be surprised if it were a Miller effect but I will do the experiment you suggest and let you know the results. Cheers Ian The trouble with experiments is you get results and this time they are not as I expected, nor indeed I suspect as anyone expected. To recap with 380mV on the grid at 2KHz I got the following distortion figures at the grid: 2H -49, 3H -62, 4H -70, 5H -77 I added a 22uF 400V electrolytic from plate to ground and the figures became: 2H -60, 3H -70, 4H -77 So the distortion is reduced but by no means to zero. 2H is reduced by 11dB, 3H by 8dB and 4H by 7dB. I am not sure what the mechanism is - all I feel safe in saying is that adding a very low value ac plate load appears to reduce the distortion measured at the grid. I keep making my point about grid current rising as anode voltage falls but no-one's picked up on it. A grid is a grid, be it input or screen. Doesn't it have a knee? Where's the knee on an EF86? As usual, I could be totally wrong... Listen to Alex more. His idea may not have solved your problem, but it's made a key contribution to analysis. Ian- Hide quoted text - - Show quoted text - Measure, measure, measure, measure, and the truth will out. But it ain't always easy. Patrick Turner. |
|
#47
Posted to rec.audio.tubes
|
|||
|
|||
|
12AX7 grid current
"Ian Bell" wrote in message ... Ian Bell wrote: Alex wrote: "Ian wrote in message ... Patrick Turner wrote: On Sep 24, 6:34 pm, wrote: "Ian wrote in message ... I just spent some time perusing the 12AX7 data sheets in more detail, particularly the part about RC coupled amplifiers which gives the maximum output before the onset of grid current under various conditions. Based on that data I calculated the cathode voltage and also the peak input voltage before grid current for several values of Ra, Rk and Va. Subtracting the two gives the grid bias at the onset of grid current and this seems to be pretty constant at about -0.6V. It also seems that with the 300V HT I have there is really only one operating point that will give me the 102V peak output swing I need without reaching grid current and that is with a 220K anode load, 2.2K up the cathode leg and a quiescent current of close to 0.6mA. I will try those values and see what I get. That's very close to what the "Muiderkring Electronic Tubes Handbook" (1968/p152) specifies for an ECC83 @ 300Vb and a Vo of 36Veff/102Vpp. Ra 220K R'g1 680K (load) Rk 2K2 Ia 0.63mA Vout 36Veff A 44x d 3.6% Rg 1M The same page specifies 11 other operating points for the ECC83. Can't find the book online but linked below is an equivalent.http://www.mif.pg.gda.pl/homepages/f.../054/2/288.pdf Rgds, Gio From the above list of performance figures we can see that there is about 1% THD per 10Vrms of output. But what is the series input resistance and to what extent would series source resistance of 100k contribute to the distortion figure of 3.6% at 36vrms output? And the answer is as follows: I used 200K in the anode and 2K2 in the cathode. HT = 313V, Plate = 170V, cathode = 1.407V With 125mV rms on the grid (2KHz) via 100K I get: 2H -59dB, 3H -77dB 4H unmeasurable The signal voltage at the plate was 10V rms which means the stage gain was 80. At the plate the measured distortion was: 2H -49dB, 3H -75dB, 4H unmeasurable So the tube itself mainly contributes 10dB of 2H The input signal was then changed to give 30V rms at the plate. Plate signal distortion was then: 2H -34dB, 3H -43dB, 4H -49dB, 5H -50dB, 6H -55dB, 7H -58dB And under the same conditions looking at the transformer secondary into 100R we get: 2H -34dB, 3H -41dB, 4H -48dB, 5H -49dB, 6H -53dB, 7H -56dB Which means the white follower plus the transformer is only contributing a couple of dB to the distortion. Also, at this point the signal voltage on the grid was just under 380mV rms and the distortion at the grid was: 2H -49, 3H -62, 4H -70, 5H -77, 6H unmeasurable which shows we are approaching grid distortion but the tube distortion is still dominant. As I mentioned before, the reason for the 100K series input resistor so I can apply shunt/shunt NFB. Connecting 2Meg from the white output back the the 12AX7 grid drops the output from 2.5V rms to 0.6V rms i.e. there's about 12dB of NFB. I increased the input to give 2.5V rms at the transformer secondary and the distortion at the transformer secondary was then: 2H-45dB, 3H -50dB, 4H -55dB, 6H -58dB, 7H -61dB which is about 11dB lower overall as expected. So, this basically seems to work. At 30V rms the 12AX7 produces about 2% distortion (-34dB) which compares well with the data sheet figure of 2.5% at 367V rms. The White follower and transformer do not seem to add significantly to this. NFB gives the expected overall reduction to just under 0.6%. I would still like the distortion to be lower so I may well try the EF86. I am not sure how much there is to be gained. From the data sheet I have, with 220K in the anode, 300V supply and 1Meg to the screen grid, this will give a stage gain of 188 times or about 7.5dB more open loop gain. Distortion is 5% at 56V rms output so if this is proportional to level, then at 30V rms output I would expect it to be about 2.7% which is close to what the 12AX7 produces. However, the extra 6dB of open loop gain means the NFB is increased by the same amount so the overall distortion should be about 6dB lower so I hope it will get below the 0.3%. Cheers Ian Hello Ian, To tackle the "mystery" of the distortion on the grid I suggested to temporarily block the plate of 12AX7 to ground by an electrolytic, say 10uF. This will reduce the gain to practically zero. After that remeasure distortion on the grid. If it reduces to nearly zero (as with the tube pulled out or cold) then it will show that the Miller effect creates (or rather magnifies) this virtual distortion. If the distortion is still high, then it is some non-linear conductance of the grid current or nonliarity of the grid capacitance due to pulsating electron cloud (space charge). Apart from a scientific curiosity you can derive something result from this experiment. For instance, it it proves Miller effect cause, then it will further inspire you to try EF86, which will not have any Miller effect at all. I would be surprised if it were a Miller effect but I will do the experiment you suggest and let you know the results. Cheers Ian The trouble with experiments is you get results and this time they are not as I expected, nor indeed I suspect as anyone expected. To recap with 380mV on the grid at 2KHz I got the following distortion figures at the grid: 2H -49, 3H -62, 4H -70, 5H -77 I added a 22uF 400V electrolytic from plate to ground and the figures became: 2H -60, 3H -70, 4H -77 So the distortion is reduced but by no means to zero. 2H is reduced by 11dB, 3H by 8dB and 4H by 7dB. I am not sure what the mechanism is - all I feel safe in saying is that adding a very low value ac plate load appears to reduce the distortion measured at the grid. Cheers Ian Thanks for the experiment Ian. It appears that there is indeed some non-linearity of grid current, though it is not the main contributor. Miller effect related distortion is several times larger. However it is not the Miller effect itself causes distortion. Simply harmonics from the plate leak onto the grid through the plate-grid capacitance. If the tube were ideally linear, then there would be no extra distortion on the grid. Similarly there would be no such effect in a pentode no matter how non-linear it can be on its own. Back to the grid current nonlinearity. Is it conductive current or capacitive current? As I wrote earlier, grid-cathode section works as a varicap. With the negative bias reducing, the conductive electron cloud (space charge) approaches grid wires closer, increasing capacitance. This mechanism explains non-linearity of the capacitance. Non-linearity of the conductive grid current is obvious. At about --1.3V bias grid current might reach 0.1...1uA. It reduces approximately 3 times per 100mV. At --1.9V it will reduce to 1nA (in theory). To find out whether it is capacitive or conductive grid current is non-linear, the following experiments could help: 1. With the plate decoupled to ground, vary grid bias by changing Rk (bypassed by a 100uF). The larger the Rk and respectively the bias the smaller the grid distortion should be. It should be the case for both capacitive and conductive grid current. 2. Measure frequency dependancy of the grid distortion. If it is proportional to frequecy -- it is capacitive. If independent -- conductive. 3. Reduce the heater voltage. Grid distortion should be reducing too, for both capacitive and conductive effects. It would be very interesting to see what results you get. Regards, Alex 1 |
|
#48
Posted to rec.audio.tubes
|
|||
|
|||
|
12AX7 grid current
"Patrick Turner" wrote in message ... On Sep 27, 1:24 am, Ian Bell wrote: I recently replaced EF86 in a pair of Quad-II amps with Polish made EF80. I increased Ia in these driver pentodes by about 2.5 times that of EF86 and arranged the EF80 as a real LTP rather than the original quasi LTP with paraphase positive FB to increase gain. The PSU was completely revised. The THD went lower for the same amount of applied NFB. The sound was exceptional with either KT66, EL34, KT88, 6550 etc. Despite EF80 maybe not looking quite so linear as EF86 from the curves, they work well if set up well with just the right value of Eg2. Alex: Is EF80 the one with the cathode offset off the centre? Is it the same as EBF80 (6N8) but without the two diodes? If yes, then I would be interested to know which Eg2, Ea and Ia you recommend for lower distortion. (I restore radios, and EBF80 is very common there driving either 6M5 (EL80) or a 6V6 clone of some sort. Possibly I could optimise the performance of the stage with your recommendations.) |
|
#49
Posted to rec.audio.tubes
|
|||
|
|||
|
12AX7 grid current
Ian Iveson wrote:
Ian Bell wrote: Looks like there is some frequency dependent. I made additional tests at 100Hz, 500Hz and 4KHz as follows (grid distortion in original circuit): 100Hz 2H -63dB, 3H -65dB 500Hz 2H -58dB, 3H -61dB 2KHz 2H -49dB, 3H -62dB 4KHz 2H -46dB, 3H -65dB Looks to me like the second harmonic rises with frequency but the third varies little so I would guess the 2H is coupled somehow from the plate and the 3H is probably the onset of real grid current distortion. Surely the grid current distortion is assymetric? That would make it substantially even H? It might be interesting to repeat this with a grounded grid configuration. Too many tests complicates confusion. Ian But grouded grid eliminates Miller effect. Cheers Ian |
|
#50
Posted to rec.audio.tubes
|
|||
|
|||
|
12AX7 grid current
Ian Iveson wrote:
Ian Bell wrote: To tackle the "mystery" of the distortion on the grid I suggested to temporarily block the plate of 12AX7 to ground by an electrolytic, say 10uF. This will reduce the gain to practically zero. After that remeasure distortion on the grid. If it reduces to nearly zero (as with the tube pulled out or cold) then it will show that the Miller effect creates (or rather magnifies) this virtual distortion. If the distortion is still high, then it is some non-linear conductance of the grid current or nonliarity of the grid capacitance due to pulsating electron cloud (space charge). Apart from a scientific curiosity you can derive something result from this experiment. For instance, it it proves Miller effect cause, then it will further inspire you to try EF86, which will not have any Miller effect at all. I would be surprised if it were a Miller effect but I will do the experiment you suggest and let you know the results. Cheers Ian The trouble with experiments is you get results and this time they are not as I expected, nor indeed I suspect as anyone expected. To recap with 380mV on the grid at 2KHz I got the following distortion figures at the grid: 2H -49, 3H -62, 4H -70, 5H -77 I added a 22uF 400V electrolytic from plate to ground and the figures became: 2H -60, 3H -70, 4H -77 So the distortion is reduced but by no means to zero. 2H is reduced by 11dB, 3H by 8dB and 4H by 7dB. I am not sure what the mechanism is - all I feel safe in saying is that adding a very low value ac plate load appears to reduce the distortion measured at the grid. I keep making my point about grid current rising as anode voltage falls but no-one's picked up on it. I thought I did in one of my earlier posts where I did tests at various operating point. The anode voltage varied by over 100V but there was no discernable change in grid distortion. Cheers Ian A grid is a grid, be it input or screen. Doesn't it have a knee? Where's the knee on an EF86? As usual, I could be totally wrong... Listen to Alex more. His idea may not have solved your problem, but it's made a key contribution to analysis. Ian |
|
#51
Posted to rec.audio.tubes
|
|||
|
|||
|
12AX7 grid current
Patrick Turner wrote:
On Sep 26, 1:23 am, Ian wrote: Patrick Turner wrote: On Sep 24, 6:34 pm, wrote: "Ian wrote in message ... I just spent some time perusing the 12AX7 data sheets in more detail, particularly the part about RC coupled amplifiers which gives the maximum output before the onset of grid current under various conditions. Based on that data I calculated the cathode voltage and also the peak input voltage before grid current for several values of Ra, Rk and Va. Subtracting the two gives the grid bias at the onset of grid current and this seems to be pretty constant at about -0.6V. It also seems that with the 300V HT I have there is really only one operating point that will give me the 102V peak output swing I need without reaching grid current and that is with a 220K anode load, 2.2K up the cathode leg and a quiescent current of close to 0.6mA. I will try those values and see what I get. That's very close to what the "Muiderkring Electronic Tubes Handbook" (1968/p152) specifies for an ECC83 @ 300Vb and a Vo of 36Veff/102Vpp. Ra 220K R'g1 680K (load) Rk 2K2 Ia 0.63mA Vout 36Veff A 44x d 3.6% Rg 1M The same page specifies 11 other operating points for the ECC83. Can't find the book online but linked below is an equivalent.http://www.mif.pg.gda.pl/homepages/f.../054/2/288.pdf Rgds, Gio From the above list of performance figures we can see that there is about 1% THD per 10Vrms of output. But what is the series input resistance and to what extent would series source resistance of 100k contribute to the distortion figure of 3.6% at 36vrms output? And the answer is as follows: I used 200K in the anode and 2K2 in the cathode. HT = 313V, Plate = 170V, cathode = 1.407V With 125mV rms on the grid (2KHz) via 100K I get: 2H -59dB, 3H -77dB 4H unmeasurable Let us remind readers that THD = 20 x log ( THD voltage / output signal voltage ) And % THD = 100 x ( THD voltage / output signal voltage ) So THD of -20dB is 10% THD, -30dB = 3.16%, -40dB = 1.0%, -50dB = 0.316%, and -60dB = 0.1%, and -70dB = 0.032%. So with Rsource = 100k, and 10Vrms anode output, THD = 0.1% approx. The signal voltage at the plate was 10V rms which means the stage gain was 80. At the plate the measured distortion was: 2H -49dB, 3H -75dB, 4H unmeasurable But now anode THD = 0.3% approx..... So the tube itself mainly contributes 10dB of 2H I'm having trouble following you. I would have thought that **without** the 100k Rsource the THD is 0.1% and **with** 100k it increases to to 0.3% and all this when you have only 0.125Vrms grid signal applied to produce 10Vrms at the anode in both cases. Perhaps I did not explain myself well enough. The 0.1% is measured at the GRID (when there is 10V rms on the plate) and the 0.3% is measured at the PLATE under the same conditions. Cheers Ian The input signal was then changed to give 30V rms at the plate. Plate signal distortion was then: 2H -34dB, 3H -43dB, 4H -49dB, 5H -50dB, 6H -55dB, 7H -58dB So THD is still dominated by the 2H. you could take all other H away from THD and the measurement of THD won't change much because the THD = square root of ( sum of the of each harmonic voltage squared ). Anyway, THD would be about -30dB from what you say, or about 3%, so 1% per 10Vrms. But **at 10Vrms**, the THD could be expected to be 1%, but with/ without 100k Rsource it measured much lower, which suggests a very non- linear increase in THD, and that at low outputs, the 12AX7 is fairly blameless, and so mayve 99% of manufacturers never cared a hoot about the source resistance in front of any amp stage using a 12AX7. It would seem there is a considerable gain in the form of much lower THD to be had at lowish levels by keeping Rsource nice and low, and so a graph could be with THD versus Rsource where Vout = 10Vrms with Rsource values ranging from say 10k to 1M. I suspect maybe not much fall in THD may occur where Rsource was lower than 10k. This may mean that where you have a CD player feeding the 12AX7, then the gain pot would best be a 20k type which is very easily driven by a 1Vrms signal from a CD player or a preamp of some kind with typical follower output. The 20k pot offers a max Rsource of only 5k if the input to the pot is say a low 600 ohm source. I might add that I have offered a few customers a 20k log DACT switched attenuator in a box to drive a pair of power amps from a CD player and I have never heard the complaint that "it doesn't sound as good as a preamp". And under the same conditions looking at the transformer secondary into 100R we get: 2H -34dB, 3H -41dB, 4H -48dB, 5H -49dB, 6H -53dB, 7H -56dB Which means the white follower plus the transformer is only contributing a couple of dB to the distortion. The White follower IS GOOD, at low levels. In most tube amps driving headphones or speakers, most THD is generated in the OP tubes amd least in IP tubes. But where you have a follower OP stage and the IP stage has to make a high signal voltage, bingo, you get more Dn than is wanted. In many "conventional" tube amps for speakers there is often a resistance divider for phone use. The R divider is often say 56R + 8R so that when phones are used the phone loading does not change the amp load much at all and the load the OP tubes see is about 7 times higher than what the tubes and OPT were mainly meant for, ie, an 8 ohm speaker. So with a much higher tube load, the OP stage gain becomes high and the total amount of GNFB becomes close to maximal and the amp works in pure class A so THD of the OP tubes then drops to much lower levels with phone compared to when speakers are used, if the comparison is made at the speaker terminals and the voltage is the same speaker use or phone use. In fact it is fairly close to being the same because if you have 1V output for a speaker and plug in the phones, they will usually switch out the speakers and the phone may get say 1/7 Vrms, which may be plenty for the same percieved volume level to the listener. The function of the 56R+8R divider also offers a great betterment to SNR. Also, at this point the signal voltage on the grid was just under 380mV rms and the distortion at the grid was: 2H -49, 3H -62, 4H -70, 5H -77, 6H unmeasurable which shows we are approaching grid distortion but the tube distortion is still dominant. As I mentioned before, the reason for the 100K series input resistor so I can apply shunt/shunt NFB. Connecting 2Meg from the white output back the the 12AX7 grid drops the output from 2.5V rms to 0.6V rms i.e. there's about 12dB of NFB. I increased the input to give 2.5V rms at the transformer secondary and the distortion at the transformer secondary was then: 2H-45dB, 3H -50dB, 4H -55dB, 6H -58dB, 7H -61dB which is about 11dB lower overall as expected. So, this basically seems to work. At 30V rms the 12AX7 produces about 2% distortion (-34dB) which compares well with the data sheet figure of 2.5% at 367V rms. The White follower and transformer do not seem to add significantly to this. NFB gives the expected overall reduction to just under 0.6%. I would still like the distortion to be lower so I may well try the EF86. I am not sure how much there is to be gained. From RDH4, and their pages on comparisons between signal amps using triode or pentode connected tubes, the triode connection gives more THD at lower levels up to about 14vrms when T or P THD becomes equal and above 14vrms the triode connection is superior. Unfortunately, RDH4 isn't being very fair at all because they should have really compared say a 12AX7 with specified Rsource with a pentode such as EF86 with the same Rsource and with both set up to give approximately the same gain. I have a feeling RCA who sponsored the writing of RDH4 wanted folks to buy more of the many pentodes they made rather than stay with the triodes which were regarded as the audio gold standard before RDH4 was written. I suggest the EF86 works with the least change of gm for a +/- Ia swing with Ea at about 150Vdc and Ia at 3mA. If the B+ is +300Vdc, then the VdcRL = 150V so RLdc must be 50k and with following Rg = 470k the total RL = 45k approx so gain will be about 90, or very similar to a 12AX7. But just what sort of harmonic mix you may measure is anyone's guess because pentodes are renowned for higher % of odd number H products. And trying to get say 36Vrms at low THD with any pentode to drive a low THD output stage may become much more difficult than say using a 6CG7 triode, especially if the triode had a CCS RLdc so it only has to produce ac power into the Rg of say 470k. From the data sheet I have, with 220K in the anode, 300V supply and 1Meg to the screen grid, this will give a stage gain of 188 times or about 7.5dB more open loop gain. Distortion is 5% at 56V rms output so if this is proportional to level, then at 30V rms output I would expect it to be about 2.7% which is close to what the 12AX7 produces. However, the extra 6dB of open loop gain means the NFB is increased by the same amount so the overall distortion should be about 6dB lower so I hope it will get below the 0.3%. Proof of the pudding is in the eating. So far only marginal benefits might be had with a pentode. But maybe you could try say a 6BX6 which has twice the gm and gain of the EF86 . The use of 12dB global NFB probably isn't enough to "dull the sound" which audiophiles may like to think always happens when NFB is used. I often think audiophiles are wrong about NFB, if the listening tests I've witnessed are any guide. Methinks a two stage amp with say 12AX7 IP tube and SE EL34 ( yep, EL34 !) in triode and normal speaker OPT and with 12dB GNFB does give rather glorious sound and I use such a beast in my kitchen radio using a lousy old OPT from an old tape recorder/player from the 1950s and a Deluxe Rola 12" speaker plus dome tweeter for a fairly flat response and sensitivity of about 95dB/W. With headphones, there is far more ability than is required. With a cascade circuit where you have a 12AX7 driving an SET OP stage the 2H of the IP tube cancels the 2H produced by the OP tube. Its not a huge cancelation, and the 3H is additive, but 3H is damn low anyway. With a very high RL on the EL34 OP tube it then produces very little 2H and more cancelation of 2H becomes possible. I would have thought a pair of EL84 or even better, a pair of EL86 in a White follower stage would produce far less THD than your 12BH7. Maybe then a loop of series voltage GNFB only around the driver cascade pair of triodes such as 6CG7 would then give you a very linear 36Vrms drive to a White OPstage. Then I thought of using a 6080 or 6AS7G with each half rigged to make a White follower with Ia = 60mA, Ea = 100Vdc for both triodes and without an OPT at all but when you draw the load lines it looks real bad for THD because each tube may see 64 ohms with 32 ohm phones and a 64 ohm RL is less than 1/2 Ra and a poor load match. Its doable, but gain is rather low, and you'd need lots of NFB to do as well as with a normal anode loaded SET OP stage with OPT. Enter 6C33C onto the Stage. The less said the better :-) Of course with a suitable OPT with primary having a CT, you could always rig the two driver triodes you do have in the Circlotron connection, and with a PP LTP driver stage with CCS cathode sink. The amount of NFB in the OP stage is high, and both triodes always see equal load conditions because there is none of that "current regulation signal" from the top R of a White follower, and load change to the White causes a different loading condition in each triode. The same happens in a SRPP gain circuit for any OP stage. The Circlotron requires 4 very quiet B+ rails for a stereo amp, with good electrostatic shielding between mains and HT windings, so it is a challenge. If not a Circlotron, you are left with the other option of using a McIntosh style OP stage which needs two equal primary windings each with a CT, but only ONE B+ rail is needed for 2 channels. Again, a special OPT may be needed. The other simpler alternative to a White follower driver to an OPT is to simply use both triodes in the 12BH7 paralleled and with a common CCS to a -200V, and with grid bias of 0V, so that the Vswing can be quite a lot higher than when you have two triodes in series between say a rail of +250V. The OPT can be cap coupled to the commoned cathodes and the THD will be mainly 2H and 3H maybe lower than the White. The mainly 2H of the 12AX7 should more likely cancel the 2H of the OP stage and maybe the outcome could be better than a White. With Ea at say 200V, there is a limit for Ia lest Pda get a bit big for 12BH7 and so I would be inclined to head for an EL86 or EL84 where Ia could happily be 40mA for Ea at 200V, giving a Pda of 8W which is well within Pda ratings for the tubes. Some folks would just use a 6BM8 or 6GW8 with a normal OPT, with pentode section trioded. Patrick Turner. Cheers Ian- Hide quoted text - - Show quoted text - |
|
#52
Posted to rec.audio.tubes
|
|||
|
|||
|
12AX7 grid current
Alex wrote:
"Ian wrote in message ... Ian Bell wrote: Alex wrote: "Ian wrote in message ... Patrick Turner wrote: On Sep 24, 6:34 pm, wrote: "Ian wrote in message ... I just spent some time perusing the 12AX7 data sheets in more detail, particularly the part about RC coupled amplifiers which gives the maximum output before the onset of grid current under various conditions. Based on that data I calculated the cathode voltage and also the peak input voltage before grid current for several values of Ra, Rk and Va. Subtracting the two gives the grid bias at the onset of grid current and this seems to be pretty constant at about -0.6V. It also seems that with the 300V HT I have there is really only one operating point that will give me the 102V peak output swing I need without reaching grid current and that is with a 220K anode load, 2.2K up the cathode leg and a quiescent current of close to 0.6mA. I will try those values and see what I get. That's very close to what the "Muiderkring Electronic Tubes Handbook" (1968/p152) specifies for an ECC83 @ 300Vb and a Vo of 36Veff/102Vpp. Ra 220K R'g1 680K (load) Rk 2K2 Ia 0.63mA Vout 36Veff A 44x d 3.6% Rg 1M The same page specifies 11 other operating points for the ECC83. Can't find the book online but linked below is an equivalent.http://www.mif.pg.gda.pl/homepages/f.../054/2/288.pdf Rgds, Gio From the above list of performance figures we can see that there is about 1% THD per 10Vrms of output. But what is the series input resistance and to what extent would series source resistance of 100k contribute to the distortion figure of 3.6% at 36vrms output? And the answer is as follows: I used 200K in the anode and 2K2 in the cathode. HT = 313V, Plate = 170V, cathode = 1.407V With 125mV rms on the grid (2KHz) via 100K I get: 2H -59dB, 3H -77dB 4H unmeasurable The signal voltage at the plate was 10V rms which means the stage gain was 80. At the plate the measured distortion was: 2H -49dB, 3H -75dB, 4H unmeasurable So the tube itself mainly contributes 10dB of 2H The input signal was then changed to give 30V rms at the plate. Plate signal distortion was then: 2H -34dB, 3H -43dB, 4H -49dB, 5H -50dB, 6H -55dB, 7H -58dB And under the same conditions looking at the transformer secondary into 100R we get: 2H -34dB, 3H -41dB, 4H -48dB, 5H -49dB, 6H -53dB, 7H -56dB Which means the white follower plus the transformer is only contributing a couple of dB to the distortion. Also, at this point the signal voltage on the grid was just under 380mV rms and the distortion at the grid was: 2H -49, 3H -62, 4H -70, 5H -77, 6H unmeasurable which shows we are approaching grid distortion but the tube distortion is still dominant. As I mentioned before, the reason for the 100K series input resistor so I can apply shunt/shunt NFB. Connecting 2Meg from the white output back the the 12AX7 grid drops the output from 2.5V rms to 0.6V rms i.e. there's about 12dB of NFB. I increased the input to give 2.5V rms at the transformer secondary and the distortion at the transformer secondary was then: 2H-45dB, 3H -50dB, 4H -55dB, 6H -58dB, 7H -61dB which is about 11dB lower overall as expected. So, this basically seems to work. At 30V rms the 12AX7 produces about 2% distortion (-34dB) which compares well with the data sheet figure of 2.5% at 367V rms. The White follower and transformer do not seem to add significantly to this. NFB gives the expected overall reduction to just under 0.6%. I would still like the distortion to be lower so I may well try the EF86. I am not sure how much there is to be gained. From the data sheet I have, with 220K in the anode, 300V supply and 1Meg to the screen grid, this will give a stage gain of 188 times or about 7.5dB more open loop gain. Distortion is 5% at 56V rms output so if this is proportional to level, then at 30V rms output I would expect it to be about 2.7% which is close to what the 12AX7 produces. However, the extra 6dB of open loop gain means the NFB is increased by the same amount so the overall distortion should be about 6dB lower so I hope it will get below the 0.3%. Cheers Ian Hello Ian, To tackle the "mystery" of the distortion on the grid I suggested to temporarily block the plate of 12AX7 to ground by an electrolytic, say 10uF. This will reduce the gain to practically zero. After that remeasure distortion on the grid. If it reduces to nearly zero (as with the tube pulled out or cold) then it will show that the Miller effect creates (or rather magnifies) this virtual distortion. If the distortion is still high, then it is some non-linear conductance of the grid current or nonliarity of the grid capacitance due to pulsating electron cloud (space charge). Apart from a scientific curiosity you can derive something result from this experiment. For instance, it it proves Miller effect cause, then it will further inspire you to try EF86, which will not have any Miller effect at all. I would be surprised if it were a Miller effect but I will do the experiment you suggest and let you know the results. Cheers Ian The trouble with experiments is you get results and this time they are not as I expected, nor indeed I suspect as anyone expected. To recap with 380mV on the grid at 2KHz I got the following distortion figures at the grid: 2H -49, 3H -62, 4H -70, 5H -77 I added a 22uF 400V electrolytic from plate to ground and the figures became: 2H -60, 3H -70, 4H -77 So the distortion is reduced but by no means to zero. 2H is reduced by 11dB, 3H by 8dB and 4H by 7dB. I am not sure what the mechanism is - all I feel safe in saying is that adding a very low value ac plate load appears to reduce the distortion measured at the grid. Cheers Ian Thanks for the experiment Ian. It appears that there is indeed some non-linearity of grid current, though it is not the main contributor. Miller effect related distortion is several times larger. However it is not the Miller effect itself causes distortion. Simply harmonics from the plate leak onto the grid through the plate-grid capacitance. If the tube were ideally linear, then there would be no extra distortion on the grid. Similarly there would be no such effect in a pentode no matter how non-linear it can be on its own. Back to the grid current nonlinearity. Is it conductive current or capacitive current? As I wrote earlier, grid-cathode section works as a varicap. With the negative bias reducing, the conductive electron cloud (space charge) approaches grid wires closer, increasing capacitance. This mechanism explains non-linearity of the capacitance. Non-linearity of the conductive grid current is obvious. At about --1.3V bias grid current might reach 0.1...1uA. It reduces approximately 3 times per 100mV. At --1.9V it will reduce to 1nA (in theory). To find out whether it is capacitive or conductive grid current is non-linear, the following experiments could help: 1. With the plate decoupled to ground, vary grid bias by changing Rk (bypassed by a 100uF). The larger the Rk and respectively the bias the smaller the grid distortion should be. It should be the case for both capacitive and conductive grid current. 2. Measure frequency dependancy of the grid distortion. If it is proportional to frequecy -- it is capacitive. If independent -- conductive. 3. Reduce the heater voltage. Grid distortion should be reducing too, for both capacitive and conductive effects. It would be very interesting to see what results you get. Regards, Alex 1 There are lots of experiments I COULD do, but my original purpose was to design and build a headphones amp not start a PhD! When I get time, I may well return to this problem and undertake more extensive experiments in an attempt to get to the bottom of it. In the meantime I am going to try the EF86. Of course, there is nothing to stop YOU doing some experiments of your own. Cheers Ian |
|
#53
Posted to rec.audio.tubes
|
|||
|
|||
|
12AX7 grid current
On Sep 28, 5:33*pm, "Alex" wrote:
"Patrick Turner" wrote in message ... On Sep 27, 1:24 am, Ian Bell wrote: I recently replaced EF86 in a pair of Quad-II amps with Polish made EF80. I increased Ia in these driver pentodes by about 2.5 times that of EF86 and arranged the EF80 as a real LTP rather than the original quasi LTP with paraphase positive FB to increase gain. The PSU was completely revised. The THD went lower for the same amount of applied NFB. The sound was exceptional with either KT66, EL34, KT88, 6550 etc. Despite EF80 maybe not looking quite so linear as EF86 from the curves, they work well if set up well with just the right value of Eg2. Alex: Is EF80 the one with the cathode offset off the centre? Is it the same as EBF80 (6N8) but without the two diodes? If yes, then I would be interested to know which Eg2, Ea and Ia you recommend for lower distortion. (I restore radios, and EBF80 is very common there driving either 6M5 (EL80) or a 6V6 clone of some sort. Possibly I could optimise the performance of the stage with your recommendations.) I have not noticed any off centreness in EF80. It looks like EF86 which is shorter. For EF80 in LTP in Quad-II I had B+ = +380V which is the screen supply for KT88. This B+ is filtered via the existing good Quad-II choke and C = 150uF. The B+ at the OPT CT is about +390Vdc and top of 470uF 450V rated with Si diodes through series 33R, so Vripple is negligible anywhere and OPT CT has nice low grounding impedance much lower than the silly 16uF value Quad used originally. RL to each EF80 anode is 100k 2W, Ea = 180Vdc.. Screen R to each screen is 390k and Eg2 = 160Vdc. 0.39uF is connected between the two screens. The 5Vac unused heater winding is in series with 1/2 the 6.3Vac heater winding to give an 8.3Vac winding which is rectified in a single wave doubler to make -18.8Vdc with good C values to reduce ripple. An MJE340 is used as to makea constant current sink to EF80 cathodes, 5.2mAdc. Each EF80 has 1k0 plus 220uF bypass to make a dc regulating RC cathode biasing circuit to ensure bias balance. The bottoms of each bias RC network are commoned and go to the collector of MJE340 CCS. The Rg for following KT88 = 270k each, taken to a -9V fixed bias. Cathode biasing is used on KT88 with 1,000uF plus 440 ohms. I used Dynamic Bias Stabilisation to ensure bias stabilty when working hard in class AB1 with low RL loads. See my website about Dynamic Bias Stabilisation at http://www.turneraudio.com.au/300w-5...stabilizer.htm GFB is from point Q on OPT to one grid of input LTP, from network 470R + 100R and with 1,000pF across 470R. At input there is passive filter 0.1uF plus 220k to EF80 input grid. HF stability was improved to very good with 100pF plus 10k between KT88 grids and with 0.22uF plus 10R across speaker output with OPT strapped for 8 ohms and which will never be changed because there is no good 4 ohm outlet and the 16 ohm strapping is useless. PO max at clipping was 28W with 0.4% THD but this is with B+ sag due to class AB op. B+ remains stable with music and peaks go to a max of 33Watts. At a watt the THD was about 6dB lower than the original amps = less than 0.05%, and Rout = 0.68 ohms. THD at 12W = 0.1%. Both amps showed very similar THD unlike originals which often have much different THD since 2H can be far higher than Walker ever mentioned and probably due to EF86 mis-matches and OP tube mismatches. The EF80 / 6BX6 is not to be confused with E80F which I found is exactly the same as EF86. E80F is a special quality pentode. EF80 has a different pin out to EF86 or E80F, so rewiring the socket is needed. But EF80 has the same pin out as 6EJ7 which can also be used but with different screen resistors, and unbypassed cathode R of about 220R because 6EJ7 has about twice gm and gain of EF80. Gain with EF80 is 210 approx for each tube of LTP but it can vary maybe 15% betwen EF80 samples but I found this didn't matter and 2H remained very low. The original EF86 circuit with the paraphase drive is a real compromised way of building a balanced amp with enough gain to allow enough GNFB. Its a fudge. It doubles its own THD. But Walker didn't have any MJE340 on hand to assist design strategy. But he would have done better with a 6CG7 LTP with 12AT7 SE input stage but I guess he has his reasons for not doing it that way. Patrick Turner. |
|
#54
Posted to rec.audio.tubes
|
|||
|
|||
|
12AX7 grid current
Ian Bell wrote:
I keep making my point about grid current rising as anode voltage falls but no-one's picked up on it. I thought I did in one of my earlier posts where I did tests at various operating point. The anode voltage varied by over 100V but there was no discernable change in grid distortion. Oh. Maybe I've lost track. I gave some thought to how such a test could be done with DC without introducing more complicating variables. Alex's idea of an AC short seemed like a reasonable half-way house, because it allows a varying Va to be compared with a constant Va. However, the results are open to several interpretations. Is it an AC-only effect, or does it also happen at DC? Is it the low AC load that makes the difference, or the constant Va, or are they two sides of the same coin? It might be useful to try to reproduce the same effect by altering the DC conditions, to freeze-frame the point of lowest Vak and highest Vgk reached in your pre-Alex test. Does this result in the same grid current as for the AC case? Is that what you've already done? This test, together with your pre- and post-Alex data, would distinguish between AC and DC effects. There are two issues that interest me here. One is the concept of a "non-linear capacitance" which makes no more sense to me than a non-linear inductance or resistance. What can be non-linear about a farad, a henry, or an ohm? The concept of linearity applies to relationships, not values. Is my log pot a non-linear resistance because it's log, or because it varies with current-induced temperature change, or because it slowly corrodes? If I put an SS rectifier accross a capacitor, does that make the capacitance non-linear? It could be characterised as such, and analysis would render a complicated equation to express that non-linearity. Rather like choosing to analyse the solar system from the point of view of earth, when standing on the sun would be so much simpler. There are several fixed electrodes in a vacuum. How can the capacitance between each pair vary? Seems to me a sensible starting point to stand on the fact it doesn't, and look at all other variables from the point of view of that assumption. Each interelectrode capacitance is perfect: two fixed plates in a vacuum. The only thing that could be responsible for any variation is the mass transport of charge between the electrodes. The rules governing this transport don't care about frequency: they apply at DC equally, so how can they be responsible for an effect that is best described as a varying capacitance? Surely a much simpler equation would result if the combination of capacitance and mass transport were disentangled and characterised as a DC, rectifying effect, in shunt with a constant capacitance? After all, that would result in the increase in 2H that you observe. The other matter of interest for me is to what extent grids can be universally characterised using the same equation. IIRC Duncan Munro's last triode model, including a full grid model instead of the common diode bodge, treated the grid in a similar way to the screen of a pentode. Alas, the explanatory document is no longer available at his site. My hypothesis, deliberately contrary to Alex's, is that there is a constant capacitance, in combination with a current arising from the transport of charge which varies in a way which is non-linear with respect to the grid voltage. The former is an AC effect, the latter DC. The DC non-linearity arises, my hypothesis continues, from the universal properties of a grid, and can be seen as an elbow in those curves labelled "anode characteristics" in datasheets. They often show screen grid current, but I haven't seen one that shows control grid current. The major problem with my contention is that it ignores the rectifying nature of a valve, in that my control grid characteristic curve would cross from negative to positive current, and valves don't behave backwards the same way as forwards. OK, so draw the control grid curve the same shape as a screen grid but further down, so that it crosses the voltage axis at the appropriate point. Now chop off the negative current part. Simple. Maybe. In that case you would expect, given constant Vgk, grid current to be zero at high Vak, remain so as Vak is reduced, and then shoot up steeply as Vak becomes very low. Where very low is, I would be interested to find out. I'd check this as a farewell tribute to my valve tester, but it works by mysterious AVO magic, and would only make everything much more complicated. It uses full-wave AC on everything except the control grid, which is rectified AC. Consequently, for my beautiful AVO, the distinction between AC and DC is always, sadly, moot. Ian |
|
#55
Posted to rec.audio.tubes
|
|||
|
|||
|
12AX7 grid current
Ian Iveson wrote:
Ian Bell wrote: I keep making my point about grid current rising as anode voltage falls but no-one's picked up on it. I thought I did in one of my earlier posts where I did tests at various operating point. The anode voltage varied by over 100V but there was no discernable change in grid distortion. Oh. Maybe I've lost track. I gave some thought to how such a test could be done with DC without introducing more complicating variables. Alex's idea of an AC short seemed like a reasonable half-way house, because it allows a varying Va to be compared with a constant Va. However, the results are open to several interpretations. Is it an AC-only effect, or does it also happen at DC? Is it the low AC load that makes the difference, or the constant Va, or are they two sides of the same coin? It might be useful to try to reproduce the same effect by altering the DC conditions, to freeze-frame the point of lowest Vak and highest Vgk reached in your pre-Alex test. Does this result in the same grid current as for the AC case? Is that what you've already done? This test, together with your pre- and post-Alex data, would distinguish between AC and DC effects. I'll check. I did the pre Alex tests at several dc conditions which resulted in a 100V variation in Va and no change is distortion measured at the grid. Not sure if any of these same conditions occurred when I did Alex's cap test. There are two issues that interest me here. One is the concept of a "non-linear capacitance" which makes no more sense to me than a non-linear inductance or resistance. What can be non-linear about a farad, a henry, or an ohm? Well Henries are well known to be non linear, just look at the B-H curve. It is well known that measured inductance varies with signal level in passive equalisers for instance. I am sure Patrick knows of similar effects in output transformers. A non linear capacitance might have a capacitance value that varies with applied signal lev el for example. I think someone suggested the tube capacitances might vary due to instantaneous bias changes altering the space charge. The concept of linearity applies to relationships, not values. Indeed. Is my log pot a non-linear resistance because it's log, or because it varies with current-induced temperature change, or because it slowly corrodes? If I put an SS rectifier accross a capacitor, does that make the capacitance non-linear? It could be characterised as such, and analysis would render a complicated equation to express that non-linearity. Rather like choosing to analyse the solar system from the point of view of earth, when standing on the sun would be so much simpler. There are several fixed electrodes in a vacuum. How can the capacitance between each pair vary? Seems to me a sensible starting point to stand on the fact it doesn't, and look at all other variables from the point of view of that assumption. Each interelectrode capacitance is perfect: two fixed plates in a vacuum. The only thing that could be responsible for any variation is the mass transport of charge between the electrodes. The rules governing this transport don't care about frequency: they apply at DC equally, so how can they be responsible for an effect that is best described as a varying capacitance? Surely a much simpler equation would result if the combination of capacitance and mass transport were disentangled and characterised as a DC, rectifying effect, in shunt with a constant capacitance? After all, that would result in the increase in 2H that you observe. The other matter of interest for me is to what extent grids can be universally characterised using the same equation. IIRC Duncan Munro's last triode model, including a full grid model instead of the common diode bodge, treated the grid in a similar way to the screen of a pentode. Alas, the explanatory document is no longer available at his site. My hypothesis, deliberately contrary to Alex's, is that there is a constant capacitance, in combination with a current arising from the transport of charge which varies in a way which is non-linear with respect to the grid voltage. The former is an AC effect, the latter DC. The DC non-linearity arises, my hypothesis continues, from the universal properties of a grid, and can be seen as an elbow in those curves labelled "anode characteristics" in datasheets. They often show screen grid current, but I haven't seen one that shows control grid current. The major problem with my contention is that it ignores the rectifying nature of a valve, in that my control grid characteristic curve would cross from negative to positive current, and valves don't behave backwards the same way as forwards. OK, so draw the control grid curve the same shape as a screen grid but further down, so that it crosses the voltage axis at the appropriate point. Now chop off the negative current part. Simple. Maybe. In that case you would expect, given constant Vgk, grid current to be zero at high Vak, remain so as Vak is reduced, and then shoot up steeply as Vak becomes very low. Where very low is, I would be interested to find out. I'd check this as a farewell tribute to my valve tester, but it works by mysterious AVO magic, and would only make everything much more complicated. It uses full-wave AC on everything except the control grid, which is rectified AC. Consequently, for my beautiful AVO, the distinction between AC and DC is always, sadly, moot. Ian Cheers ian |
|
#56
Posted to rec.audio.tubes
|
|||
|
|||
|
12AX7 grid current
flipper wrote:
On Fri, 17 Sep 2010 23:39:35 +0100, Ian wrote: flipper wrote: On Fri, 17 Sep 2010 10:48:50 +0100, Ian wrote: flipper wrote: On Thu, 16 Sep 2010 21:58:37 +0100, Ian wrote: snip Yes, I have come across the same effect in my work with mu followers. I was just surprised to see it acting so soon. generally you can allow the signal to take the grid up to -1V before worrying about grid current bu that seems not to bee the case with the 12AX7 - today I fed it with a mere 25mV and it produced 1% distortion. There's something funny going on here. You are chasing phantoms as it isn't 'positive grid drive'. It's called a "grid leak" resistor for a reason: there's grid (leak) current. (This, btw, is what determines the 'recommended maximum' grid leak resistor.) It's small but there nonetheless and, being roughly proportional to plate current, varies with signal as well. As for the size of it, RDH4 recommends that, for cathode bias 'high mu' triodes, the grid leak be no larger than 3x the (DC) plate load, and no larger than 4x the preceding stage's plate load, but that's a bias shift consideration and larger values were common practice. As for distortion, what you've got on the grid is a summing junction of your input and the 'grid leak' signal plus, of course, inter electrode capacitances so any measurement there will, of necessity, be 'distorted' relative to your source. Lower the source impedance and you increase it's proportion of the sum so the relative distortion decreases but it will never be '0' unless you turn off the tube (all of which you've observed). Why are you trying to measure distortion on the grid anyway? Only because I traced back to there starting from the ouptut. What distortion did you get on the plate and what were you expecting it to be? Exactly the same as at the grid - about 2H at -30dB or worse. If I recall, the original grid number you posted was -33dB. The reason I asked is, up to now, you hadn't mentioned plate distortion. If I short the input resistor 2H at the grid drops below -70dB and at the anode falls to -40dB which is about what I would expect. Seems reasonable. I note you said, up at the top of this post, it was 1% (-40dB) with 25mV too. So I was expecting -40dB 2H at the anode and got a lot higher so traced back to the grid and found it was the same but that it wnet away with near zero source impedance. Well, that's still consistent with what I described. If the culprit were positive grid drive one would expect the plate distortion to go up roughly the same as the grid but, while it certainly goes up, it doesn't go up nearly as much. We should note that 'positive grid' drive is not the only place where grid induced distortion occurs. For one, it occurs as I described and secondly.... which takes more explanation so... When the grid is 'very negative' reverse grid current is roughly proportional to plate voltage but as grid V approaches 0, but before 'positive grid drive', there is an inflection point where it sharply curves upward toward positive and, then, eventually goes positive. So, from the inflection point to the positive point, reverse grid current is no longer linear (in that region) with respect to plate voltage. (see RDH4 Chapter 2, figure 2.10, if you have a copy). yes, I am very familiar with that section!! That's not nearly as noticeable with a low source impedance but with a high source impedance you have the same situation I described earlier except the reverse plate--grid feedback into your source impedance is no longer linear. I.E. The tube introduces distortion in amplifying from grid to plate and, rather than a 'linear' reverse FB, the plate to grid feedback further distorts because your grid signal (peaks) traverses the inflection (I mean, if it does). I concur that grid distortion as I understand it begins at the inflection which is some smallish negative value of grid bias. By looking at the data shhets for output level and gain at the onset of grid current distortion I was able to calculate the expected value of the inflection for the 12AX7. My problem was that using signals below this level still seemed to cause measurable distortion in the grid circuit, in other words the inflection was not where I expected it to be. Now, the mechanism that causes distortion with high source resistances once the inflection is passed is due, according to RDH and others, to the effective reduction in the grid to cathode impedance from that point onwards. Personally I am not sure and for my purposes I want to avoid this region anyway so the exact mechanism is not important to me. I would have thought your grid bias was low enough but the 25mV number suggests maybe not. That was what I thought. The calculated inflection point ISTR was around -0.6V and with over -1V of bias I should have been well away from this region with a 25mV signal. Hence my puzzlement and I wondered if I had made a stupid mistake in wiring it up. On reflection, after leaving it a couple of days then trying the 2K2 Rk and 220K anode resistor version from the datasheet I noticed I have left a 560R Rk in circuit when I had tride other values so my biasses may n ot have been what I thought. Secondly, The probe I was using had times 10 switch which I used when looking at the anode and switched t 1X for the grid. It is possible I inadvertently had is switched to x10 when I made the 25mV reading which would have made it 250mV or 0.35V peak which with a biad of say -0.6V would have resulted in grid distortion. However, another thing to consider: You explain that the purpose of the 100k is for feedback. Well, for one, when the loop is closed your effective grid resistance will be lower than 100k because of the FB resistor. Second, your grid signal will also be of smaller magnitude, being the sum of the source minus FB, than what you see open loop. The net result is you may not have as much of a problem open loop measurements suggest, depending on how well you're simulating the closed loop conditions. Yes and no. The grid resistance will indeed be lower but the signal level at the grid will be unchanged for the same output level. With NFB the input level needs to be increased to obtain the original output level and of course the amount it needs to be raised IS the amount of NFB. Cheers Ian Cheers ian Cheers Ian |
|
#57
Posted to rec.audio.tubes
|
|||
|
|||
|
12AX7 grid current
flipper wrote:
On Thu, 30 Sep 2010 11:05:24 +0100, Ian wrote: flipper wrote: On Fri, 17 Sep 2010 23:39:35 +0100, Ian wrote: flipper wrote: On Fri, 17 Sep 2010 10:48:50 +0100, Ian wrote: flipper wrote: On Thu, 16 Sep 2010 21:58:37 +0100, Ian wrote: snip Yes, I have come across the same effect in my work with mu followers. I was just surprised to see it acting so soon. generally you can allow the signal to take the grid up to -1V before worrying about grid current bu that seems not to bee the case with the 12AX7 - today I fed it with a mere 25mV and it produced 1% distortion. There's something funny going on here. You are chasing phantoms as it isn't 'positive grid drive'. It's called a "grid leak" resistor for a reason: there's grid (leak) current. (This, btw, is what determines the 'recommended maximum' grid leak resistor.) It's small but there nonetheless and, being roughly proportional to plate current, varies with signal as well. As for the size of it, RDH4 recommends that, for cathode bias 'high mu' triodes, the grid leak be no larger than 3x the (DC) plate load, and no larger than 4x the preceding stage's plate load, but that's a bias shift consideration and larger values were common practice. As for distortion, what you've got on the grid is a summing junction of your input and the 'grid leak' signal plus, of course, inter electrode capacitances so any measurement there will, of necessity, be 'distorted' relative to your source. Lower the source impedance and you increase it's proportion of the sum so the relative distortion decreases but it will never be '0' unless you turn off the tube (all of which you've observed). Why are you trying to measure distortion on the grid anyway? Only because I traced back to there starting from the ouptut. What distortion did you get on the plate and what were you expecting it to be? Exactly the same as at the grid - about 2H at -30dB or worse. If I recall, the original grid number you posted was -33dB. The reason I asked is, up to now, you hadn't mentioned plate distortion. If I short the input resistor 2H at the grid drops below -70dB and at the anode falls to -40dB which is about what I would expect. Seems reasonable. I note you said, up at the top of this post, it was 1% (-40dB) with 25mV too. So I was expecting -40dB 2H at the anode and got a lot higher so traced back to the grid and found it was the same but that it wnet away with near zero source impedance. Well, that's still consistent with what I described. If the culprit were positive grid drive one would expect the plate distortion to go up roughly the same as the grid but, while it certainly goes up, it doesn't go up nearly as much. We should note that 'positive grid' drive is not the only place where grid induced distortion occurs. For one, it occurs as I described and secondly.... which takes more explanation so... When the grid is 'very negative' reverse grid current is roughly proportional to plate voltage but as grid V approaches 0, but before 'positive grid drive', there is an inflection point where it sharply curves upward toward positive and, then, eventually goes positive. So, from the inflection point to the positive point, reverse grid current is no longer linear (in that region) with respect to plate voltage. (see RDH4 Chapter 2, figure 2.10, if you have a copy). yes, I am very familiar with that section!! That's not nearly as noticeable with a low source impedance but with a high source impedance you have the same situation I described earlier except the reverse plate--grid feedback into your source impedance is no longer linear. I.E. The tube introduces distortion in amplifying from grid to plate and, rather than a 'linear' reverse FB, the plate to grid feedback further distorts because your grid signal (peaks) traverses the inflection (I mean, if it does). I concur that grid distortion as I understand it begins at the inflection which is some smallish negative value of grid bias. By looking at the data shhets for output level and gain at the onset of grid current distortion I was able to calculate the expected value of the inflection for the 12AX7. My problem was that using signals below this level still seemed to cause measurable distortion in the grid circuit, in other words the inflection was not where I expected it to be. You've mentioned this 'deduction' from the data sheets before but I'm not clear on what you're using or how you're determining it but if it's from the class A RC amp list then I'm not sure it's the inflection you're calculating because source impedance is low for those. That is what I am using and the data sheet specifically says distortion and output level measured at the onset of positive grid current. Now, the mechanism that causes distortion with high source resistances once the inflection is passed is due, according to RDH and others, to the effective reduction in the grid to cathode impedance from that point onwards. Personally I am not sure and for my purposes I want to avoid this region anyway so the exact mechanism is not important to me. Off hand I don't recall that explanation in RDH4 but, in any case, I was giving my understanding of it and not trying to mimic RDH4. I would have thought your grid bias was low enough but the 25mV number suggests maybe not. That was what I thought. The calculated inflection point ISTR was around -0.6V and with over -1V of bias I should have been well away from this region with a 25mV signal. Hence my puzzlement and I wondered if I had made a stupid mistake in wiring it up. Well, it happens and I can recall staring at the scope myself saying "THIS IS NOT POSSIBLE." hehe On reflection, after leaving it a couple of days then trying the 2K2 Rk and 220K anode resistor version from the datasheet I noticed I have left a 560R Rk in circuit when I had tride other values so my biasses may n ot have been what I thought. Secondly, The probe I was using had times 10 switch which I used when looking at the anode and switched t 1X for the grid. It is possible I inadvertently had is switched to x10 when I made the 25mV reading which would have made it 250mV or 0.35V peak which with a biad of say -0.6V would have resulted in grid distortion. However, another thing to consider: You explain that the purpose of the 100k is for feedback. Well, for one, when the loop is closed your effective grid resistance will be lower than 100k because of the FB resistor. Second, your grid signal will also be of smaller magnitude, being the sum of the source minus FB, than what you see open loop. The net result is you may not have as much of a problem open loop measurements suggest, depending on how well you're simulating the closed loop conditions. Yes and no. The grid resistance will indeed be lower but the signal level at the grid will be unchanged for the same output level. With NFB the input level needs to be increased to obtain the original output level and of course the amount it needs to be raised IS the amount of NFB. That's why I said it depended on how well you were simulating closed loops conditions and for all I knew the '100k' might have been a substitute reflecting the effective grid impedance with everything closed up. The 100K is the input arm of shunt.shunt NFB. The feedback are is 2Meg Cheers ian Cheers Ian Cheers ian Cheers Ian |
| Reply |
|
| Thread Tools | |
| Display Modes | |
|
|
Similar Threads
|
||||
| Thread | Forum | |||
| Screen grid mu | Vacuum Tubes | |||
| Grid Resistors | Vacuum Tubes | |||
| Grid current limit specs for "good" output tubes? | Vacuum Tubes | |||
| DC on the grid | Vacuum Tubes | |||
| Grid current in 6C33C | Vacuum Tubes | |||
Linear Mode
