DHT Folded Cascode Experiments

Here we are on holidays and surprisingly I struggle to find any free time in which I can sit down in front of a tablet and write this blog entry. This is actually quite good as it’s a sign of me being disconnected and looking after my daughters. Life is good.

As said earlier on my previous post, I have a pile of experiments and stuff to share which I’ll hopefully do in due course. The first one which sprang to mind was the folded cascode experiments which I conducted last year. This came around due to two main quests I was working on. First one, my phono stage which actually uses one 50dB gain stage made up of a folded cascode E810F triode-strapped. Marvelous results and hats off to Rod Coleman for building a small PCB kit for his circuit. Second quest, was to optimise my pre-amp/driver stages in my DHT system. Well, it seems like a never ending story. I kind of like it that way as it pushes me constantly to find new avenues and circuits. This was what ignited me on looking at combining the higher gain and benefits of the folded cascode topology, with the unique sound of the DHTs.

I’m not going to cover the folded cascode topology nor its circuit details. I’ve done that to death in previous posts. You can search around and will get a lot of information about it. So please, don’t send me an email asking question about how the circuit works, etc. You can do a bit of research yourself.

For obvious reasons due to my personal preferences, I picked up on the VT-25/10Y as my first DHT victim on this circuit. The VT-25 has an average transconductance for a DHT: 1.5mS (a 300B had 5mS for comparison purposes). In perspective, the gm is low for this circuit and we will see soon why.

I wanted to combine the sound of the VT25 with a higher gain which I wanted to get from end to end on my system. I targeted 30dB which obviously is beyond what you can do with a normal hybrid mu-follower stage which is mu the maximum gain possible. With a DHT normally this is around 8 and you could push it to 15-20 with some specific triodes.

When I built the first prototype test, I resourced the available PCBs I have to build this circuit like a Lego game. I used a CCS PCB, a pair of Source Followers (SFs) for output buffer and for voltage references and then a simple shunt PNP transistor as the pass element in a PCB prototype. You can avoid all this complexity and get a simple and compact PCB from Rod. I did this so I could play around and adjust the circuit and learn through experimentation.

First test was made with a VT25 DHT. I used the following configuration:

CCS load:

1. IXTP08N100D / DN2540 cascode

2. Trimpot 100R in parallel with 270R

3. Padding resistor = 10R

4. Folded Cascode device: ZTX951

5. Rc = 33K wirewound 7W

6. Bias supply for shunt cascoded: Gyrator PCB hacked for bias supply. This was later replaced with a SF board with a reference voltage generated by a CCS.

Here is the initial test using fixed bias, you can see the reduction in gain to about x40 and HF response limited to 23kHz when loaded with 100K:

Given the higher output impedance, the frequency response and overall gain was compromised with a 100K load. So I added a source follower: DC-coupled with the Shunt Cascode and an output capacitor of 220nF to the load.

VT25 Shunt Cascode. Experiment 2 (addition of SF output buffer)

As you can see in the above plot, the frequency response is now much better as the SF isolates the load from the high-impedance output of the shunt cascode stage.

Well, all this looks nice and rosy as we are having a nice DHT stage with lots of gain. So, how this stage actually performs?

Although distortion is acceptable at 4Vrms (0.05%) it increases up to 0.52% for 20V output. What I then tried was to implement some cathode degeneration to help with the linearisation of this stage. With a resistor of 300R as Rk distortion results where much better and the frequency response wasn’t impacted as I was using a low impedance voltage regulator for the filament circuit, not CCS based. Otherwise the circuit using a CCS-type filament regulators (e.g. Rod Coleman’s) will lead to a lot of hum picked up by the DHT unfortunately. The gain is reduced to 35 with same Rc and this is due to the effective gm is reduced due to the cathode resistor. This is the level of gain I’d like to have in my system, so it was great. Frequency response was good up to 86kHz with the source follower as previous test.

The current of the Shunt cascode was adjusted (CCS) to minimise distortion at maximum level. 144V at the Rc is high and limits the output swing (ideally 100V should have provided more headroom), however the distortion is minimised for a voltage across Rc about 130-145V:

This is the final stage I ended up playing with. It sounds very nice but not as good as the 01a-VT25 dual stage I have. Likely being the high harmonics introduced at louder volume levels. It doesn’t have the same headroom as the other 2 valve setup. So I wouldn’t use it unless is for small signal amplification, not to drive +100Vpp. Here is the circuit I implemented. Again, if you want to build this, ask Rod for his kit. It’s way more simpler than the below and is all included in a tiny PCB:

If you were to implement this circuit with Rod’s board, it would look like this:

01a DHT Folded Cascode

Of course after the relatively success of the VT-25 folded cascode stage, I couldn’t resist the temptation of building a similar one with the 01a. Knowing that this topology relies on good gm on the lower device, I knew I was trending waters which likely were going to conduce me to a nice outcome. Nevertheless it was worth trying it. Implementing the same circuit with 01a, I struggled with the following:

1. With cathode degeneration (Rk=1k6). The filament acts as an antenna and you need to be careful to pick up noise. With a low voltage filament regulator works well. However, the gm is reduced to a level that overall gain is just 10 with the Rc of 39k. I will probably need a resistor of 78k or 80k to deliver a gain of 20.

1. With filament bias, Rfil = 27R (which is biased a bit too high) the distortion is high:

I could only minimise the distortion by reducing the collector voltage to about 30V:

The 01a stage is a no go. Not enough transconductance is available to provide a linear stage. Here is the oposite example with a 300B:

Clearly the higher the Gm (degenerated) the more linear the stage is. 28dB gain with 22K Rc impacted slightly by 100K output load. There is no SF buffer and the frequency response is very good.

300B folded cascode frequency response

If you want to implement a DHT in a stage like this, the 300B is a winner given its gm. I’d likely try replacing the cathode resistor with an array of SiC diodes. However, would need to check distortion and optimal point. I ran out of testing time when was playing with this breadboard and should probably get back to it.

When I was exchanging emails with Rod about my findings here’s what Rod commented on the gm impact:

Yes, higher gm, lower distortion is exactly what I found. In almost every case, taking a Shunt Cascode circuit and replacing the valve with one of higher gm lowers distortion, and this action has a good record of improving the sound. High gm lowers distortion, and allows lower values of load resistor – without affecting the anode current in the valve.

For fun, I went back to my old lab book from May 2005 to check what I did when I developed the circuit: The first round of tests of the Shunt Cascode designs were used to compare EL84, EF80, PC86, EF95, Western Electric 717A [all IDHTs] and the DHT LP2, Marconi Osram manufacture.

They were measured for gain, and then tested for sound with the help of a Classical musician. The DHT came last in those tests, the record shows.

It’s not too hard to see why. DHTs owe their superior qualities to the lack of curve-compression in the curves’ high voltage, low-current segment. This gives measurably better results with wide swings, especially with reactive loads (i.e. speakers or long cables), which draw the load line out into an ellipse.

As you know, Shunt cascode applies a vertical load-line to the triode. In most cases, the DHT advantage is not helpful, because the worst-distorting zone of indirectly-heated triode curves is not covered in the signal swing.

DHTs in shunt cascode may not work best with the usual operating points: in many cases, lower anode voltage and higher current will be better. But still, a high-gm frame-grid 1950s/60s designed valve will usually be better, in the same position. Even the (maybe still plentiful?) EF80 is very fine-sounding.

For a preamp, the 6Э5П as shunt cascode will give the same outstanding performance that it does as a driver. It can be optimised by running a lower anode voltage and higher current, if desired. A lower value of load resistor will also help, and you can get usefully low drive impedance this way.

The 6Э5П is a superb Shunt Cascode valve, partly for its property of gm-saturation, at currents over 30mA. So running lower voltage, and higher current (35-38mA or even more) with this one (and also, but to a lesser extent – the 6Э6П) may give even better results!

More generally, low anode voltage (80-150V) in Shunt Cascode stages permits the use of many different kinds of low-cost frame-grid pentodes & triodes for a preamp – including many that are not well-thought of, for conventional amps.

Rod Coleman (2018)

And indeed, a 6Э5П / 6Э6П makes a much better driver in shunt cascode. I posted before about this and would recommend this to anyone who wants to build a higher gain driver for a 300B or a transmitting valve output stage.

Folded Cascode in Phono Stages

I intend to cover this in detail at some point. However, due to its subject complexity, it will take me a while to summarise my work on this.

Nevertheless, I wanted to show an example of how this folded cascode circuit can perform brilliantly when married with a high-gm valve. I’m using the E810F in triode mode as part of my phono stage. This valve is configured as second stage of amplification with a whooping 50dB gain and superb linearity. It’s situated in between a split RIAA network which leverages the advantage of the output Rc impedance of the folded cascode as part of the RIAA network.

I built and tested the FC PCBs from Rod. Stuffed it with an IXT08N100D top MOSFET and used 27K wirewound RC resistor. The E810F valves are biased at about 160V and have a 47R cathode resistor.

Really nice performance with 60dB gain, 45kHz bandwidth and distortion is very low. Nice harmonic profile. Good job on this circuit from Rod!

If you look at the above, you will see that the stage has an amazing low distortion for 60Vpp / 20Vrms and 50dB of gain! Difficult to achieve whilst providing low input capacitance thanks to the cascode configuration. Also the harmonic profile is very nice. And it sounds great too!

Here is a view of my initial phono prototype with Rod’s board:

Hope you have enjoyed this post and got the bug of the folded cascode!