The Shunt Cascode Driver

A heavy-weight driver

IMG_0320Rod Coleman came up with a brilliant design recently which baptised as “shunt cascode” driver. For those who cannot stand a pinch of sand in their circuits, I suggest you skip this post now. This hybrid circuit is actually a folded cascode if we consider the book terminology. What makes attractive of this design is its outstanding performance against the classic multistage designs aimed at achieving a large drive signal for output stages such as 300B, 6C4C/2A3, etc. I personally haven’t build it yet but according to Rod the sound is superb.

Before building a stage which will replace my current 45 SE driver, I thought it made sense to analyse the circuit and understand why is claimed to be such a great alternative for today’s designs.

First of all, let’s present the “shunt cascode” circuit:

shunt cascode article 1The circuit is formed by a triode in common cathode mode loaded by a CCS and the output is feeding a common-base bipolar stage (yes, a bipolar transistor!). The purpose of the BJT transistor (Q1) is to fix the anode voltage and force the triode to operate in a (nearly) vertical load line. The triode is then operating as a transconductance amplifier (or current pump). As the input voltage varies (grid-cathode) this is translated into a change in the anode current. If we look at the anode which has a fixed voltage, the current variation will be reflected at the Q1 emitter as the CCS current is fixed (and it has a very large impedance). Q1 presents a very low input impedance as it is connected in a common-base mode. The current variation at the emitter is “buffered” and reflected at RL. Here is where the “amplification” is manifested as the current is forced through RL which creates an output voltage change (Vo). The BJT does not convert current to voltage, you can easily demonstrate this by setting RL to zero. Then, the voltage out is zero, but the current swing is almost entirely unchanged.

So you will be probably asking yourself: so what? What is the benefit of this configuration? Well, it happens to be that the gain of this design is dependent on the triode transconductance (gm1) and RL only. In fact, the gain is:

 A_{V}=g_{m}\cdot R_{L}

IMG_0318What makes it really interesting is that the gain can be much higher than the valve’s mu. You can easily achieve 200-300 gain with only one stage. The gain is dependant on the valve’s transconductance and RL only. So any high transconductance valve is a good match here. In fact, most of the TV high-frequency pentodes triode-strapped (e.g. EF80, E180F, EF86, etc.) are great candidates.

Some key points from the DIYAudio thread to highlight from this topology are:

  1. The amplification is entirely determined by the triode
  2. The shunt cascode operates with a static anode voltage, therefore Mr Miller is not crushing you with his habitual burden. The input capacitance is just the Cgf + Cga of the triode, without upscaling from the gain. The is the killer for CCS-loaded high-gain triode stages – and demands very low impedance preamp drive.
  3. The quality of the cascode voltage-source is critically important.
  4. The base drive to the cascode has a strong influence on the sound.
  5. Looking at the hfe as a function of collector current and that the base current adds to (subtracts from) the collector current, any variation in base current will be reflected at the output to the valve.With a carefully chosen transistor, and operating conditions, the base current robs a minor linear fraction from the output current. A FET would fix that, naturally, but the FET’s gm is too low, and capacitance too high (in most circuits).
  6. Compare previously stated flaw in point 5 with other means of getting the same gain. With 2 triode stages, you multiply the nonlinearities of the first stage in the second stage. These nonlinearities include the variation of gm across the large-signal swing (this alone is worse than the base current error). The power-supply rejection (lack of it!) is also amplified by stage 2.
  7. Since the music current only circulates between the triode and the load, the supply current is constant instantaneously (not just on average, like a normal SE class A stage). This means that the power supply capacitor carries no Music signal – a huge advantage in itself. And the gain is so high, that the triode’s cathode resistor can be unbypassed – eliminating another perennial nasty. It can also run into the following stage with dc-coupling (lose the coupling capacitor, too), so you can have a capacitor-free driver stage!
  8. The BJT is not acting in an amplifying capacity at all – its one and only function is to fix the anode voltage to the value set by the (fixed) cascode voltage. If you prefer circuit analogies, think of it as an emitter-follower – with only a dc input. The emitter has very low output impedance, driving a very high output impedance (the triode’s anode). Therefore the emitter totally defines the voltage. An emitter values holds its properties well, regardless of collector current – when correctly designed.
  9. The BJT’s contribution should be very linear, so whatever the triode expresses with be transmitted to the output grid. Keep in mind that the amp’s output will be the combined contributions of the input and output tube. To the extent that it’s 2nd harmonic, they will cancel or partly cancel each other, as they are in antiphase. So a very horizontal load line for the input tube (less 2nd) could actually mean more 2nd at the output.
  10. Comparing its performance to a CCS driver stage, the CCS loaded amplifier would only have a gain of about mu – meaning you would need to swing more volts as grid voltage to get the same voltage output. In almost every case, this would take the voltage into the high-voltage region of the triode, where the curves begin to compress, and the low-voltage region, where they expand. This distortion would be many times worse than the effects we are witnessing from Vbe changes. Add to that the serious problems of input-loading caused by the high Miller capacitance, and it becomes clear that the horizontal load-line stage has major distortion effects to contend with, that are almost completely avoided in shunt-cascode.
  11. A darlington BJT pair instead of Q1 is better because there is effectively no base current error, and the capacitances are lower with BJTs (compared FETs), and the gm higher.
  12. The high R-load works better, because the grid has to swing less for the same output. This reduces triode distortion.
  13. A resistor is better in cathode because:
    – it degenerates the gm, but linearises it (with zero phase error,unlike loop feedback)
    – it holds the anode current steady (important for dc stability)

Let’s look now at the gain of this stage in more detail:

shunt cascode article 2To simplify the analysis, I will use a fixed-bias stage and ignore the CCS output impedance. So if we replace the triode and the BJT with their models like in the diagram on the left, we can see then that the following expressions can be derived from this circuit:

V_{O}=-\alpha \cdot i_{e}

g_{m1}\cdot V_{i}+\frac{V_{a}}{R_{a}}=i_{e}

V_{a}=-i_{e}\cdot r_{e}

Doing a bit of algebra crunching with the three expressions above, we can arrive to the following gain expression considering that:
\alpha =\frac{\beta }{\beta +1}\approx 1
 r_{e}=\frac{\alpha }{g_{m2}}\approx \frac{1}{g_{m2}}
The gain of the shunt cascode is then:
A_{V}=g_{m1}\cdot R_{L}\cdot \left ( \frac{r_{a}}{r_{e}+r_{a}} \right )
So considering that:
r_{a}\gg re \Rightarrow \frac{r_{a}}{r_{e}+r_{a}}\approx 1
Then we arrive to the simplified gain formula:
A_{V}\simeq g_{m1}\cdot R_{L}

The driver in practice

So, how will this circuit perform to drive a 45 SE stage where at least 100Vpp is required?
I could use any of these Russian pentode drivers, but I thought about testing the lovely 6e5P first. Rod has kindly helped me in refining a 6e5P driver with this topology. Here is the initial take on it:
6e5p shunt cascode example
A simple DN2540 CCS set to about 42mA and a darlington BJT formed by a pair of PBHV9040Z set to 250V at its base by a shunt regulator to ensure stability of such a high gain driver. The 6e5P (triode-strapped) is biased at about 37mA and 251V. RL is a 27K which gives about 135.
Looking at the THD performance we can see that its lower than 0.22% for about 160Vpp. The predominant H2 component is then reduced and odd harmonics increase significantly:
6e5p shunt cascode THD exampleA DC coupled version can be achieved by placing a gyrator in parallel with RL to set the output voltage to the bias requirements of the output valve and provide a high impedance to let RL set the gain of the shunt cascode driver. This will allow removing the coupling cap (C4 in the diagram above) and provide an end-to-end cap-free system 🙂
Well, enough for today, I will probably look at refining this driver for a 6C4C or 4P1L PSE output…
Now it’s time to build this driver and judge its sound!
Ale

Author: Ale Moglia

"A mistake is always forgivable, rarely excusable and always unacceptable. " (Robert Fripp)

13 thoughts on “The Shunt Cascode Driver”

  1. Another great idea by Rod Coleman!
    I followed the discussion on diyaudio and I can’t wait to find out what you think of its sound! This looks very promising!

    I’m sure a cascoded CCS for the first stage would be so much better than the single DN2540… I usually use IXTP08N100D2 + IXTP01N100D depletion MOSFETs.

    Note: using a gyrator does not make the system cap-free: there is one in the gyrator and the signal is going through it… What do you think?

    1. Hi Vincent.
      A cascoded CCS is always good for better PSRR. In this case, just a simple MOSFET is performing really well. You can change if you want to.

      Regarding the cap-free comment, you are right that there is a very small signal is shunted through the capacitor into the high-impedance reference voltage (e.g. an LND150 CCS and a 220K resistor through a 2Meg-4Meg resistor). This is also applied to the base of the gyrator FET which is then amplified modulating the output CCS current in a very small way. I’m inclined to say that the impact is meaningless compared to a capacitor being in the signal loop and the actual current going through it which is amplified by the following stage.
      Not a perfect world, but definitely the gyrator’s capacitor is one that my ears can tolerate very well… 🙂
      Ale

  2. I’m building this circuit now, and I don’t have amazing ripple free 300V. Theoretically, what s preventing us from putting a capacitor to ground directly at the anode? It is going to create an extra LP filter, I realize this, but is it the tube’s anode resistance or should I calculate the Rload in parallel with it? Or am I on the wrong track?

    1. Not really, I’d be a short in AC. If you supply is not that clean, then build a cap multiplier with a good size cap to drive the base of the PNP transistor. The CCS will also do a good job in rejecting the ripple due to its high AC impedance.
      Cheers

      1. I’ve been unable to avoid oscillation (around 1MHz) so far with just increasing the capacitance on the base supply. Any quick tips?

          1. It’s a small fabricated board with mostly SMDs. It’s just that I keep blowing the mosfets with strategically placed tiny caps that have worked in the past in those spots.

          2. The CCS depletion FET? What caps are u referring to? Drop me an email with your diagram and pictures as is hard to follow what you’re doing!

  3. Åle, what would happen if one uses a choke (eg. 100-200 H capable of 10 mA dc) instead of R6?
    Then the topside of the choke is about 5-10 volt DC, and stable in DC at that; this means the grid of the next stage can be connected directly, suppressing C4 and R3 while the cathode resistor of the driven tube needs to be increased a bit to accomodate for the higher grid DC voltage.
    [I should simulate but would only have my own circuit to view.]
    albert

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.