Single Ended ECC99  6C33C Amplifier with regulated tube PSAuthor: Dmitry Nizhegorodov (dmitrynizh@hotmail.com). My other projects and articles
2. Design 3. Schematic 4. Construction 5. SPICE check 6. Adding Negative FeedBack
6.2 6db feedback 6.3 12db feedback 6.4 5x (14db) feedback 6.5 MultiLevel NFB Amp 8. Practical implementation 9. References
1. IntroA friend of mine, Orest Baidan, has obtained vintage tube regulated power supplies Lambda C881M and C481M. He asked me to outline a singleended tube amp design best utilizing one of these PSs. Lambda C481M offers 125325V of B+ at up to 400mA DC, plus 20A of 6.3 VAC. Lambda C881M can provide up to 800mA. After looking at these parameters, I suggested a stereo amp with two 6C33C [1] as the output tubes, each loaded with OneElectron UBT1 [2] transformer, running in class A driven by a nice, linear driver tube. Thus this project was born. I designed the schematic, Orest designed the layout, the chassis, and built the amp. It turned out to be a pleasant project, hence here's a short writeup.
2. DesignThe design described here is done backwards, from the load to the output stage to the driver and so on. The design is done using graphical methods only (plate curves, loadlines), for pedagogical reasons and just for fun of staying away from SPICE.6C33C is a twoplate power triode. It can continuously dissipate up to 60W on the plates and needs 6.3V, 6.6A for its heater. 6c33c is a currentsource tube, originally designed for regulated power supplies. It is best utilized with idle current of 200250mA. However, this would require custom output transformers, as a SE OPT rated for 250mA is hard to find. Also, this would rule out C481M and put way too much strain even on C881M; regulated supplies tend to freak when the load draws above the limit even momentarily, I did not want to stress that too much. OneElectron UBT1 fits nicely, since it is rated at 160mA max. Hence, drawing 120150mA per channel at idle would be reasonable, and even C481M can be used if the load is no too steep and the output of C481M is shunted with a cap. With that, we can guesstimate to dissipate up to 310 * .15 = 46W on each tube, assuming fixed bias. Since it is known 6C33C tubes are sometimes finicky in fixed bias mode, I decided to start this project from simpler and more reliable autobias configuration. This can deliver less power than fixed bias mode would, but also opens some possibilities for experiments with bypassing versus nobypassing, etc.
2.1 Output stage designThe plot on the right shows the plate curves of 6c33c. 60W dissipation zone is outlined with red dots. The horizontal red line is 150mA current line. The lowleft corner triangle is the slope of 1.6k load (100mA up, 160v left) and another one at 200mA, 320V repeats that slope. The actual target loadline is parallel to them  it's the blue line. It crosses the 150ma, 220V idle point (a.k.a. quiescent point). This loadline delivers voltage swing of 330V PP, and current swing of 220mA PP, and needs ~83V of grid swing. The bias is ~83V. The value of the autobias resistor can be calculated as bias_voltage/idle_current = 83V/150mA = 560 ohm. An autobias resistor of 600 ohm would dial into 82v, 135mA region. This setup requires 170VPP amplitude from the driver. A good driver should be able to deliver at least 200VPP and have fairly low output impedance, as 6C33C may start drawing some noticeable grid current when grid/cathode voltage is approaching zero.
2.2 Driver stage designWhat driver tube fits the bill? Can a 2stage amplifer be easily costructed? Orest requested a 2stage design. If the input is 2VPP, it's a plank that's too high for any triodes and even many pentodes (look at the pentode data in [3] for inspiration) because the driver stage gain must be around 100. Anyways, Orest wanted a design that uses ECC99 as a driver. ECC99 is a capable and popular mediummu 5W twin triode, which is reported to be very linear, Quick estimates shows that a 2stage ecc99/6c33c amp needs over 10vPP of input signal and therefore the amp with ecc99 as a driver must have at least 3 stages. The gain stage need not provide a lot of amplification, though, and can be the other half of a twintriode ECC99. Thus we settled on ECC99  ECC99  6C33C topology, or a 4tube (2 tube per channel) stereo amp.The plot above shows that ECC99 needs around 500V B+ with resistive load to deliver clean, lowdistortion 200VPP into a reasonable load. Since we're limited by C*81M's B+, I opted for chokeloaded driver topology. As it can be seen from the blue triangle (which has a 20k slope), an ECC99 loaded with a 20k resistor and fed with 500V B+ underperforms a chokeloaded ECC99 seeing 200250v of B+. What choke can be use? This question really boils down to: what is the minimal inductance that can be used, and what DC the choke should be able to sustain. ECC99's Rp is ~4k. If our target lowfrequency 3Db limit is, say, 20Hz, the choke needs to have at least that much (4k) reactance at 20Hz, which means at least 30H. The plot shows the idle point of 18mA, hence a choke withstanding 20mA will suffice. It helped us that Orest had purchased a spare pair of ElectraPrint DRD driver chokes, these are >>30hs and can withstand 20ma easily. The plot shows the idle point of 18mA, 220V will require ~ 7V bias (this value found as an interpolation of grid line voltages alone the load line) and, therefore, an autobias resistor of 7V/18mA = 390 ohm. 380 or 400 ohm will do, too. The triode will dissipate ~ 4W under these conditions. Assuming the choke will eat ~ 20V of B+, bias eats 7, the RC filter to feed this stage can consume the rest  320207220 V  and its R needs be (320207220)V/18mA = 4k. No huge reason to give more current or B+ to the driver stage, yet if we wanted to dial into hottermost region, at the limit of dissipation, we'd need the same bias at 225V, 23mA. This is only 5V differences in B+, and since tubes may vary, it is unreasonable to let the R in the filter be less than 4k or have the autobias resistor less than 400 ohm. Design of the input stage is very straightforward (the loadline is seen in the left corner. The output swing will be less than 20VPP and with 200V B+, a 20K load gives 67mA of idle current. This needs 22.2V bias, and, therefore, 300 ohm as the value for the bias resistor. The value for the R in the input stage's RC filter can be 2 to 3k. The value for the shunt caps in the filters can be anywhere above 6uF. Nonelectrolytics can be used, as 10uf seems OK. Read more on this in the NFB/SPICE sections, as the interaction between the choke, the transformer and various caps in the circuit turned out to be a lot of fun to study. Finally, the cathode bypass caps: 100uf is 100 ohm at 17hz (160 ohm at 10 hz) hence 100uf is a good guesstimate for all 3 stages; the output stage may be Ok even with as low as 60uf (hint: this is a paper&oil territory). Read more on the values of these in the SPICE sections below.
3. SchematicThe 10uF caps are 400V, paper and oil (PIO). In case of C481M the rightmost 10uF cap must be increased to at least 6090uF, more is better. This is so because if both channels are loaded with fullswing signals, current draw may approach 520mA. The cap must be able to help the PS to withstand that. No such danger for C881M. The 0.1uF signal (coupling) caps are Teflon, paper&oil or P416 orange drops. The bypass 100uF caps are 5V, 15V, 150V left to right, and can be quality electrolytic or PIO. The output stage cap can be decreased to as low as 30uF. Resistor wattage is as follows. For the 600 ohm in cathode of 6c33c  14W dissipation, 25W recommended; for the 400 ohm in the driver's cathode, 20k in input triode's plate and 2k in input filter  0.5W is safe enough; for the 4k resistor in driver's filter  1.6W dissipation, a 5W resistor is required.
4. ConstructionThe output tubes and the autobias resistors generate lots of heat. Orest had to mount two radiators (shown) on the sides of the aluminum chassis, and add better ventillation around the tubes too cool down the chassis.
5. SPICE check6c33c was modelled using the plots from the audiomatica website [4]. The plot is done using Sofia tracer, which is in a way more "real" than [1]. The tube model was found using my "paint" java tool [5]:
** 6C33CCRV ************************************************************ * Created on Wed Oct 12 16:43:48 PDT 2005 using tube.model.finder.PaintKIT * URL: http://www.mclink.it/com/audiomatica/tubes/6c33c.htm * .SUBCKT TRIODE_6C33CCRV 1 2 3 ; P G K ; + PARAMS: CCG=3P CGP=1.4P CCP=1.9P RGI=2000 + MU=2.7090 EX=1.4629 KG1=406.875 KP=14.8125 KVB=11.4375 VCT=0.8320 ; Vp_MAX=400.0 Ip_MAX=0.6 Vg_step=30.0 * E1 7 0 VALUE={V(1,3)/KP*LOG(1+EXP(KP*(1/MU+(VCT+V(2,3))/SQRT(KVB+V(1,3)*V(1,3)))))} RE1 7 0 1G G1 1 3 VALUE={(PWR(V(7),EX)+PWRS(V(7),EX))/KG1} RCP 1 3 1G ; TO AVOID FLOATING NODES C1 2 3 {CCG} ; CATHODEGRID C2 2 1 {CGP} ; GRID=PLATE C3 1 3 {CCP} ; CATHODEPLATE D3 5 3 DX ; FOR GRID CURRENT R1 2 5 {RGI} ; FOR GRID CURRENT .MODEL DX D(IS=1N RS=1 CJO=10PF TT=1N) .ENDSUBT1, see the spec at [2], was modelled as:
.SUBCKT XFRM_UBT1 1 2 3 4 ; +PARAMS: LPRIM=8 LLKG=2mH RPRIM=165 CPRIM=2nf LRATIO={16/1600} * LPRIM IS THE TOTAL PRIMARY L (VARIES WITH MEASUREMENT). * LLKG IS THE LEAKAGE L (MEASURABLE: CONSISTENT). * RPRIM IS THE TOTAL PRIMARY R. * CPRIM IS THE MEASURED PRIMARY CAPACITANCE. * LRATIO IS THE INDUCTANCE RATIO: (8 OHMS)/(PRIMARY Z). .PARAM QFCTR={LPRIM/LLKG} ; QFACTOR. RS1 1 2 10000K RP1 1 11 {RPRIM} LPleak 11 12 {LLKG} LP1 12 2 {LPRIM} CS1 12 2 {CPRIM} LP2 41 3 {LPRIM*LRATIO} RSEC1 4 41 .01 ; secondary resistance KALL LP1 LP2 {11/(QFCTR)} ; COUPLING .ENDSwith this model, the frequency responce of the amp shows all the properties of a "real", not ideal, transformer:
The distortion data is typical for a SET. No harmonic cancellation is in sight, meaning the driver stage is clean. 6c33cc is expected to produce up to 6% THD fullswing, mainly 2nd harmonic.
This plot show how the value of the power stage cathode bypass cap afftects lowfrequency responce. The boost is due to LC resonance between UBT1's inductance, and the capacitance. If you need a boost, it can be as low as 30uf, otherwise make it > 100uf.
This plot is variability of transformer's leakage inductance by stray capacitance. the value of 36 is expected for a UBT1, while it is shown that even a much larger value does not spoil the picture, and a larger value may actually deliver some HF boost to those neededing it. An equivalent result can be obtained with a capacitive shunt of the primary with a 510nF value.
This plot is a sweep of the load capacitance. As shown, it reflects back into the amplifier as yet another HF pole. Nothing bad happens until 300500nF.
6. Adding Negative FeedBackThe amp can run into very high efficiency speakers (> 100 dbwm) straight, with no NFB. If ran at 3+ watts, it starts sounding progressively more euphonic (which may be considered "Sweet", pleasant on "sparce" jazzy material) yet inducing too much IMD (which may be considered really bad on full orchestra or multitrack pop/rock material). One way to improve things is to add a small degree of negative feedback. The best it to have a way to select an appropriate level of NFB.It is a mitstake to believe that any amplifier with sufficient gain can utilize a feedback; there are factors not affecting a zerofeedback design that influence stability and frequency responce liearity once a feedback is added. In particular, multistage amplifiers with lots of LC poles do present problems on both ends of frequency range when NFB of sufficient depth is introduced; the amp described here is not an exception; spikes in the response at ~ 10..20hz and >50khz showup; the way to deal with that is to introduce appropriate dampening.
To cope with that, the first coupling cap is reduced to several dozen nF (depends on the level of feedback); the grid shunt resistor of the 2nd stage was reduced to 150k. The interaction between the second and the 3rd stages is complex due to several poles that show up when NFB level goes up; the coupling cap there is not reduced, but increased to .25uF; it is a standart practice in 3stage amp design to limit bandwith of the input stage while opening up the follwing stage to the maximum. Highfrequency correction (dampening) can be done in two ways: (1) by shunting R3 with a capacitor (few hundred pF to several nF), or (2) by shunting the grid of VL2. Both techniques are NFBleveldependent. The first method is common in many commercial amplifiers; the second method is more appropriate for multilevel NFB switch idea, see below. N.B: this configuarion also allows to keep convenient value of bypass caps. It is sufficient to vary NFB by altering the value of r20. Thus, r20 = 0 will mean no nfb. Thus, this is nonNFB setup with new values of some of the parts. Distortion data and frequency plots are the same as above. Note that c1/r6 slight;y compensates for the hump at ~ 18hz due to the resonant property of L1C4. Thus, increasing R6 makes the hum visible in spice. YMMW, especially considering OPT's tendenc to rolloff, this may not be an issue when feddback is zero, yet will bite when NFB is deep and may cause motor boating. C9 is set at 500pf to compensate ringing caused by transformer's leakage inducatce and sray capacitance; in reality this may not be needed at zero NFB. Below are the data for several levels of NFB, with optimal values of c9 and c1 and corresponding frequency plots.
6.1 3db feedback.R20=8, C1=47n, R3 shunt = 2n (or VL2 shunt = 1.2n)
6.2 6db feedbackR20=18, C1=30n R3 shunt = 3n (or VL2 shunt = 2n)
6.3 12db feedbackR20=50, C9=500p, C1=100nbad damping, we show it as an example! see the frequency plot:
same but tunedin damping: C1=14n R3 shunt = 5.6n (or VL2 shunt = 4.7n)
6.4 5x (14db) feedbackR20=80, C1=10n R3 shunt = 7.2n
this is as much NFB as practical, as sensitivity drops to 2vRMS at 6W out. here is corresponding square signal responce:
if you find this too much of ringing (I do not), increasing C9 to ~10n will reduce the overshot yet will introduce ~.5..1db rolloff at 20khz (it must be remembered that the overload is due to ~100khz ringing). Another good idea is to use *both* R3 shunt and VL2 shunt, which will result in steeper rolloff after ~20khz. In addition, a ~ 1nf can be placed in parallel to R21.
6.5 MultiLevel NFB AmpThese ideas were prototyped in SPICE but not [yet] tried as mods to the amp; the following is the same amp with a 3pole, 4way switch allowing to have. say, 0db, 3db, 6db, 12db levels of feedback. Notice that all 3 banks have at least one alwaysconnected value and therefore one pin is not used. This helps reducing switching pops. For the same reason there are leakage resistors between the caps in each bank. R3 shunting is shown yet VL2 shunting would be more elegant as it does not require leakage resistors.
7. Max OutFinally, a thoughtexperiment  what if we could run more DC through the OPTs, were willing to add negative bias supply and run 6c33c at max power  would this topology match, at least in SPICE, the specs of Lamm ML2.1 amps? (18W @ 3% THD, 0.775Vin)Changing R12 to 100, having a negative 80V in series with R10, and with R21=700, R20=120 (~20db NFB) we run 6C33Cs with 205mA DC and 270V cathodetoplate, thus dissipating 55.3W. In this regime, SPICE reports 1% THD at 18W. Under 10W, It is < .5%, mostly 2nd harmonic. Sensitivity is 3v. This actually "beats" ML2.1, except for sensitivity. With R21=1k, R20=50 (12dB nfb) we get 18W out at 1.2VRMS in, with 2 % THD:
HARMONIC FREQUENCY FOURIER NORMALIZED PHASE NORMALIZED NO (HZ) COMPONENT COMPONENT (DEG) PHASE (DEG) 1 1.000E+03 1.682E+01 1.000E+00 1.293E+00 0.000E+00 2 2.000E+03 2.158E01 1.283E02 7.604E+01 7.345E+01 3 3.000E+03 2.027E01 1.205E02 2.216E+00 6.094E+00 4 4.000E+03 1.165E01 6.925E03 7.985E+01 7.468E+01 5 5.000E+03 1.004E01 5.968E03 1.724E+02 1.660E+02 6 6.000E+03 6.623E02 3.938E03 9.481E+01 1.026E+02 7 7.000E+03 4.130E02 2.456E03 9.517E+00 1.856E+01 TOTAL HARMONIC DISTORTION = 2.037093E+00 PERCENT
8. Practical implementationWhere to get OPTs tolerating 200 to 300 mA DC? One option rarely discussed is running two OPTs sideby side, in parallel, thus splitting the DC in half. This section contains various SPICE simulations proving that this is a very attractive solution when if such highcurrent OPTs are not available. Abstracting what follows below, we can say that two transformers run in parallel extremely smoothly if DC resistances, primary inducatance and winding ratios match closely. By closely we mean less than 10% mismatch. This should be a very easily atteinable target; One should expect that from a batch of several, maybe as little as four UBT1 transformers two pairs can found that match DCR, Lprim and turns within 10%.Hence the proposed solution is to simply match two pairs of UBT1, wiring the primaries in parallel. The follwing plots summarize distortion figures for the same configuarion as described above  R20=50, R21=1K, I=205mA, Upc = 270V (55.3W) but this time with two UBT1 transformers, with primary and secondary coils in parallel.
the left plot is for the 8 ohm load connected to the 4 ohm taps, second one  to the 8ohm taps, the right one  to 16 ohm taps. The 8 ohm plot looks almost exactly identical to the 18W, 2% plot from the previous section. The values of the input voltages corresponding to the dot marks are the same. One might expect this to be the case for the 16 ohm connection; yet SPICE suggests 8 ohm taps, not 16 ohm taps must be used. On the right I plotted the singleOPT plot data from the previous section and the 8 ohm doubleOPT data from above on the same plot, using the same axis scaling. SOPT harmonics are for singleopt. DOPT harmonics are for doubleOPT. Some of DOPT are slightly higher than corresponding SOPT, but still very close. The above was for 2 identical OPTs. How about some mismatch? I conducted parametric runs in SPICE, deviating all basic parameters of UBT1s pairwise, in opposite directions, by as much as 2x and saw only modest change in power and distortion data. As an example, on the right is a rather crowded plot of distortion data of the same configuration as above, with 900mVamp 1khz on input, with winding ration of the transformers as indicated by green (0% deviation), red (+10,10% deviation), blue (+20,20% deviation), yellow (+30,30% deviation). Do not confuse these colors with the colors of the harmonic products, though. This plot shows that up to 20% deviation very slightly reduces the power, from 9.885W to 9.750W, and very slightly increases all harmonics but the third, which actually drops down 5 times. this is probably a local minimum; the 3rd rases again when deviations are 30%. It should be stressed, though, that even +10% is very artificial example. One should expect to match winding ratios very closely, within 1% with no difficulty. Here is a comparison plot for ideal match against 2% mismatch  one UBT1 has 16 extra turns, another  16 turns less. [tbd, but expect a close match] One related effect is current through mismatched turns. What happens with the windings when ratio mismatch takes place? The windings heat up, as there is significant current through. Thus, suppose there is 20% deviation (+10, 10) and each secondary up to the point of the parallel joint is 100mOhm. then the current through that, at 10W RMS into 8 ohm load is 60% of the current that runs when the transformers match. Still this accounts for only ~ 5mW change of dissipated heat. So much for the power and distortion analysis, how about frequency range and stability analysis? No stability issues SPICE concerns with; small deviations in primary inductance or resistance are negligible. One note regards primary inductance: the combined may be significantly higher than 4H because each UBT1 sees only 100mA. This way saturation occurs much lower in frequency. This maybe a good point to emphasize that 6c33c with its very low output impedance (80..150 ohm) drives 46 H loads incredibly linearly into singlehertz region: a 4.7H load is 600 ohm at 20hz, 150 ohm at 5 (!) hz.
9. References[1] http://www.mif.pg.gda.pl/homepages/frank/sheets/113/6/6S33S.pdf[2] http://www.oneelectron.com/Trans/UBT1_20.pdf [3] http://www.pmillett.com/pentodes.htm [4] http://www.mclink.it/com/audiomatica/tubes/6c33c.htm [5] http://www.dmitrynizh.com/tubeparams_image.htm
Author: Dmitry Nizhegorodov (dmitrynizh@hotmail.com). My other projects and articles
