The SOHA's input stage is a simple grounded cathode amplifer with a CCS plate load. We designed three different CCSs so that at least one of them could be built by anyone around the globe from available parts. Even so, there were some problems finding components. Since the CCS sets the plate current in the tube and since we wanted 40V on the plate, it was necessary to provide a cathode bias trimpot for adjustment to set the plate voltage. This trimpot is necessary because tubes vary and because they seem to vary more at low voltages.
The goal for the SOAH II input stage is to eliminate both the CCS complexity and the trimpot. I tried a number of designs using the grounded cathode configuration (single triode). They all involved an opamp servo and I didn't like any of them because they were too complicated. One of the other ideas for the SOHA II had used a differential pair for the input stage. I decided to have a look a this again. After some thought and a few sims, the input stage took shape.
Using A Common Cathode Amp
The first thing to do with a pair of triodes with common cathodes is to put a CCS load on the tail. This reduces distortion. It also tends to make the voltage gain of each half equal, but in this case this doesn't matter because the input stage is not really a differential pair. It's actually a common cathode amplifier where the first triode operates as a cathode follower driving the second triode in grounded grid mode. The basic pair looks like this:
Notice that the first triode has no plate load, forcing it to operate like a follower. There are still two important problems to solve with this configuration: 1) How to set the plate voltage of the second triode to 40V with a B+ of 60V and 2) How to provide a very high RL when there is only 20V drop across it. These problems necessitate some kind of active solution.
Adding A Necessary Current Mirror
The first step is to solve the RL problem. A simple CCS for RL leave us with the same problems as the SOHA. No good. To make this work we use another technique, fairly commonly known, of using a current mirror to load the pair of triodes. The first triode gets the control transistor. Like this:
The current mirror forces the currents in the triodes to be the same and provides a high RL for the second triode. Unfortunately, it doesn't completely solve the problem because the plate voltage of the second triode falls close to the B+ and, thus, the triode can't swing positive voltage. We still have to get the plate voltage down to 40V. The quietest way to do this is to insert a resistor between the mirror and the first plate. Like this:
The resistor value is chosen to drop about 20V (including the BE diode drop and the emitter resistor drop) down from the B+. If the tail current is set to 2mA then the triode current is 1mA and the closest standard resistor value is 18k which will drop about 19V from the B+. This is close enough for tubes.
Now how does this work? One step at a time.
- The tail CCS sets the total current flowing through the triode pair
- The current mirror forces each triode to have the same current, each half the total set by the tail CCS
- The resistor, BE junction of the BJT, and emitter resistor force the plate of the first triode to be, approximately, 19V less than the B+, about 41V
- The tail CCS adjusts the cathode voltage of the first triode so that, given its plate current forced by the CCS and mirror, its plate voltage will be 41V. The is no other option providing that the triode is conducting and not in positive grid mode
- The second triode sees exactly the same cathode bias as the first triode. And because of the mirror, it also has exactly (or almost exactly) the same plate current
- Since the second triode has the same grid bias and plate current as the first triode, it MUST set its plate voltage to be the same as the first triode - 41V
How close the second triode's plate is to 41V depends on how closely the triodes match. Generally, good working triodes in the same envelope are reasonably well matched. But, beyond this, we don't need exactly 41V at the plate. We just need something close to this. A few volts either way won't matter. A 20% mismatch in the triodes tends to be a few volts either way.
Note that because the tail CCS maintains a constant total current and because the mirror equally divides this current between the two triodes, the current through the plate resistor is constant for any type of tube. Thus, the first triode will always sit at about +41VDC no matter which twin triode is rolled in.
Finishing the Tail CCS
Now, how to do the tail CCS? Once again the simplest and most widely available CCS will be the trusty ring-of-two BJTs. The final input stage, then, is this:
The trimpot (which we worked so hard to eliminate) is needed only to set the reference triode's plate voltage one time. No adjustment is needed when tubes are changed. A fixed 2mA is achieved with a single 330Ω resistor in place of the fixed resistor/trimpot combination, but having the flexibility of an adjustment to get 2mA is a better option. See Amplifier Schematic section.
With these values and devices the current mirror load on the second triode is about 3MΩ. This resistance will be in parallel with the load provided by the buffer.
Any respectable successor to the SOHA must replace the opamp with a good discrete buffer. There are quite a few discrete buffers to select from. These are a few:
- Diamond Buffer
- PPA Buffer
- JISBOS Buffer
- Stacker Buffer
The main feature that any hybrid buffer must have is a high input impedance so that the buffer doesn't load the tube. The raw PPA buffer is pretty good. The unmodified JISBOS is aroung 1MΩ and the Stacker is more than 3MΩ. The Diamond buffer, which will not have high enough Zi for this amp.
I thought about using the Stacker buffer, but it's too complicated for this amp (good for the Stacker though because the Stacker is a high-end amplifier). The JISBOS Zi falls off too fast with frequency because of the GD and GS capacitances of the input JFETS. The PPA buffer must be servoed to work properly which leads directly to the Stacker buffer design. I also designed a single ended buffer and with some encouragement from the Stacker team, I decided to use this in the SOHA II. Everybody likes SE buffers. There seems to be something magical about fully class A operation, even if the amp heats your house in winter.
A Cool (or maybe hot) SE Buffer
The SOHA II SE Buffer is a CCS loaded emitter follower. Nothing new, except that the design is targeted at creating a high Zi. To get a high Zi we must have current gain in the buffer which requires at least one Darlington pair. Like this:
The 820Ω resistor keeps the input BJT in class A for all signal conditions. The “C” class BC550 is selected because it has a high HFE, which increases the current gain and, thus, increases the Zi. The O/P BD139 is intended to run at about 100mA to give the buffer the ability to drive very demanding loads without cutting off. This means that the CCS has to handle 100mA too. Which means that the CCS will also have a power transistor in it.
The High Current CCS
The simplest BJT CCS would be a power BJT with an LED from its base to the negative rail and an emitter resistor to set the bias. But, a single power BJT has poor characteristics for a really good CCS. We need to partner the power BJT (needed to handle the current) with a high gain BJT to improve CCS behavior. There are several ways to do this, such as a cascode arrangement. However, the simplest CCS and the one with the lowest voltage drop is a ring-of-two pair of BJTs. Like this:
The CCS has a theoretical dynamic resistance of about 37kΩ, staying flat from 1Hz to about 100kHz and dropping to about 25kΩ at 1MHz. This dynamic resistance would be low for, say, a plate load, but is good enough for a follower whose Zo is only a few ohms. It is more than 100 times larger than the resistance of typical high Z headphones.
The trimpot should adjust the CCS from about 60mA to over 100mA so that the builder can set this for the headphones in use. Some will only need 60mA idle current and this will reduce heat generation in the O/P stage.
This buffer topology will get very close to 1V from the rails, in this case, about 14V peak.
A Servo that Stays out of the Signal Path
What's missing from the buffer now is a way to bias the base of the input BJT. How ever this is done, it must be with a high value resistor to preserve a high Zi. The buffer also needs a servo to zero the DC at the output. We can solve two problems if we can return the servo to the input with a high value resistor. This is where the Darlington stage helps us. Because of the high current gain in the Darlington, the base current in the BC550C is very low, typically just a few μA. This low current means that we can use a 1MΩ from the servo to the base with just a volt or two drop across the resistor. An opamp servo's range will only be from rail to rail so the nominal voltage drop across the return resistor has to be much smaller than this to give the opamp room to account for variations in components. There is nothing special about the servo. It looks like this:
The servo sets the base bias of the input BJT to set the DC offset at the output to zero. Because of the 1MΩ resistor and the high Zi of the Darlington pair, the Zi of this buffer is about 950kΩ to above the audio band. This value is high enough for any tube that will be used in the front end. The 18kΩ resistor forces the opamp to always draw current, keeping it in class A mode even though the base current is very small.
The 470μ capacitor on the positive rail gives the follower a reservoir to draw from during high current swings. We don't need one on the negative rail because the CCS, obviously, keeps the current through the negative rail constant.
High Voltage Multiplier
The SOHA II retains the "only one transformer" design of the original SOHA. The SOAH II uses a variant of a Cockroft Walton (CW) voltage multiplier. The basic CW multiplier consists of a stacked series of diodes and capacitors as shown here:
The CW voltage multiplier is a stacked set of capacitors with steering diodes. Each capacitor is charged to the peak of the AC voltage on the transformer, Vp = 1.414 x VAC. Where VAC is the full secondary voltage. Each step up the ladder increases the voltage by one Vp. The right side has the even multiples and the left side the odd multiples.
However, the diagram above has no reference. In order to use a CW multiplier in an amp we have to ground reference it. But, what the CW does depends on where you put the ground. There are two logical places, either side of the AC secondary. These are:
In the “even” multiplier, going up the right side ladder, the DC (with no load) is a constant voltage. But on the left side, the odd side, the DC has the full AC secondary voltage imposed upon it. With this ground reference, only the even multiples are useful. The “odd” mutiplier does just the opposite with the odd multiples having the constant DC. But, we can always select the grounding scheme we need to get the voltage multiple that we want.
For the SOHA II, however, we have a third grounding scheme. In this case the transformer has a center tap and the CT is the ground reference point. Like this:
Unfortunately, with this scheme all of the DC voltages have half of the full secondary AC imposed on them. This is because none of the capacitors in the ladder are ground referenced. Thus, we need one more modification.
The SOHA II Voltage Multiplier
For the SOHA II we want to achieve around 100V B+. The transformer is 30VCT, which makes the Vp of the secondary about 42V. To get over 100V we need a 2.5X multiplier (2.5 x 42V = 105V). For this we can use the second step on the ladder (5Vp/2) on the odd side. To get rid of the large AC component we simply remove the last capacitor from the stack and reference it to ground, like this:
With this scheme we get a constant DC at the top of the multiplier. Of course, when current is drawn the DC will show aome ripple due to the charging/discharging of the last capacitor. And this leads to the last modification that makes the SOHA II HV multiplier.
Half-wave to Full-wave
Looking at the basic CW multipliers you can see that they are half wave rectifiers. To see this just look at only the bottom section. It has one diode and one capacitor, a classic half-wave configuration. This half-wave property extends all the way up the stack. When load is applied, the multiplier misses half of the cycle, requires more current from the transformer, and generates more noise. But, we can make a full-wave multiplier by replicating the full 2.5X stack and then connecting it 180° out of phase to the transformer. Like this:
The two steering diodes at the top prevent the separate multipliers from discharging into each other. The mulitpliers operate on opposite halves of the AC cycle, improving current sourcing capability, reducing voltage drop under load, and reducing ripple noise. The final capacitor acts as the first filter capacitor in the HV power supply.
This diagam, slightly modified, is on the SOHA II Power Supply schematic.
Alternative Zener Input Stage
The 18k resistor at the plate of the input cathode follower reduces the bandwith of the amp. Even though the current mirror keeps the current in the triodes reasonably constant, it is not perfect in doing its job. Thus, any small currents that do flow in the first plate see an 18k load and this decreases bandwidth and speed. We can eliminate this problem by using a zener diode in place of R4. An 18V zener will drop the right amount of voltage but it will have a dynamic impedance of less than 1Ω, significantly inproving the bandwith of the front end. This schematic looks like this:
There is one additional positive and one negative aspect of this modification.
The positive feature is that now you can adjust the tail current without affecting the voltages on the plates of the tubes. Some builders may hear audible differences at these low voltage, low current operating points.
The negative feature is that zeners are generally noisier than resistors and this noise will be transmitted to the plate of the second triode through the mirror. You may hear a higher noise floor with a zener in place of R4.