Feedback was that more pictures were preferred. So here they are. I have little time, but slowly I will make progress I hope. The main chopping board is what is left now. Layout is tricky as have not enough space given the size of the OPTs

I really love the Ba DHT preamp, if you need the gain in your system, is likely to be one of the best sounding DHT preamps in my experience. As received many requests for the SPICE model for the Ba DHT, here it is:

**** Ba TRIODE Composite DHT *****************************************
* Created on 10/13/2017 18:33 using paint_kit.jar 2.9
* www.dmitrynizh.com/tubeparams_image.htm
*
* Traced and model by Ale Moglia valves@bartola.co.uk
* (c) 2017 Ale Moglia and Bartola Ltd. UK
* www.bartola.co.uk/valves
*———————————————————————————-
.SUBCKT DHT_Ba 1 2 3 4 ; P G K1 K2
+ PARAMS: CCG=1P CGP=3.8P CCP=1P RFIL=7
+ MU=14 KG1=8940 KP=84 KVB=5232 VCT=-3.5 EX=1.47 RGI=2000
* Vp_MAX=350 Ip_MAX=10 Vg_step=1 Vg_start=0 Vg_count=11
* Rp=4000 Vg_ac=55 P_max=1.5 Vg_qui=-48 Vp_qui=300
* X_MIN=75 Y_MIN=51 X_SIZE=492 Y_SIZE=530 FSZ_X=1192 FSZ_Y=679 XYGrid=false
* showLoadLine=n showIp=y isDHT=y isPP=n isAsymPP=n showDissipLimit=y
* showIg1=n gridLevel2=n isInputSnapped=n
* XYProjections=n harmonicPlot=n harmonics=y
*———————————————————————————-
RFIL_LEFT 3 31 {RFIL/4}
RFIL_RIGHT 4 41 {RFIL/4}
RFIL_MIDDLE1 31 34 {RFIL/4}
RFIL_MIDDLE2 34 41 {RFIL/4}
E11 32 0 VALUE={V(1,31)/KP*LOG(1+EXP(KP*(1/MU+V(2,31)/SQRT(KVB+V(1,31)*V(1,31)))))}
E12 42 0 VALUE={V(1,41)/KP*LOG(1+EXP(KP*(1/MU+V(2,41)/SQRT(KVB+V(1,41)*V(1,41)))))}
RE11 32 0 1G
RE12 42 0 1G
G11 1 31 VALUE={(PWR(V(32),EX)+PWRS(V(32),EX))/(2*KG1)}
G12 1 41 VALUE={(PWR(V(42),EX)+PWRS(V(42),EX))/(2*KG1)}
RCP1 1 34 1G
C1 2 34 {CCG} ; CATHODE-GRID
C2 2 1 {CGP} ; GRID=PLATE
C3 1 34 {CCP} ; CATHODE-PLATE
D3 5 3 DX ; FOR GRID CURRENT
D4 6 4 DX ; FOR GRID CURRENT
RG1 2 5 {2*RGI} ; FOR GRID CURRENT
RG2 2 6 {2*RGI} ; FOR GRID CURRENT
.MODEL DX D(IS=1N RS=1 CJO=10PF TT=1N)
.ENDS
*$

You can download the file here: Ba spice triode model

When everything was going to plan…

This build became one of the quickest and eventually the most painful from all, perhaps not really. However, it was very challenging in the end. I will tell you why in more detail. Yet, it has been a fantastic learning experience.

The amplifier building experience was going swiftly. For someone who struggles to find some free time to work on the projects, the lockdown gave several nights of work. I rigourosly made progress and everything was working to plan, until the final test was made. I individually tested every stage and part of the amplifier and everything worked just fine.

However, when I made my final test and brought up the output HT via a VARIAC, I found that the amplifier driver stage / source follower oscillated! Argh, the worst nightmare came true. Very unpredictable behaviour and hard to pin down. Spent few nights without success. After several attempts to find the root cause, I reached out for help to Rod Coleman. Rod has always provided tons of advice and his experience is great. He’s also very helpful and ready to support where he can. That makes him so valuable in these situations.

The real culprit

Although there was a LF oscillation which made the -200V bias raw supply go crazy. It was puzzling to see the circuit to operate well when operated under a bench supply. We suspected on stray inductance and capacitances playing around and turning the SF into an oscillator.

So ended up adding extra bypass capacitors, changed the Source Follower tail CCS for a resistor load instead. Still after desperate changes and modifications, I was unable to isolate the source of oscillation.

Looking at the passive -200V supply. I ended up testing leakage currents to earth and found that I had the -200V at +80V from earth! Dissecting the supply, I sadly found the Mains transformer with a leakage short into the copper screen between secondary and primary which is connected to earth. Phew!

Luckily I had a similar transformer around, which made me compromise a bit the supply and voltage levels, which still deliver what I want, and voila. The amplifier was now working like a charm!

The amplifier is dead quiet, amazing. Flat response up to 80kHz which is hard to achieve without these OPTs. The hybrid mu-follower plus the source followers have a bandwidth of 3Mhz. At 37dB gain is a fantastic stage, but a devil one if you don’t take care of taming any source of potential wild oscillation due to the high gain and bandwidth.

I will post diagrams soon, for now will enjoy the sound of it after so much work.

I’m delighted with the sound so far. Strong bass, detail and transparency of treble and overall sound image is perfect. Glad to see the 300B back on my system.

This isn’t a build I’d recommend to get into, unless you’re prepared to live with fixed bias and ton of iron, etc. However, you need to take care of driving a 300B if you want to get the best sound out of it. That doesn’t come cheap and there is no free lunch of course!

I’ve been enjoying and carefully listening my new 300B amplifier. I have to say that I love every bit of its sound, treble detail and strong bass. The amplifier is fast and can drive very well my speakers. I only discovered that due to my low level DAC, the gain of the D3a in triode is yet not enough to get it to maximum power. So, I hooked in my beloved 01a preamp. The overall gain is too much of course so had to place the volume control at the output of the 01a stage.

I think a gain of about 130-140 should be ok. Perhaps if I get around in adding the 6SF5 stage then it may be good enough.

So this got me thinking. Of course I have on my list 2 driver tests:

1. 4P1L pentode – yes, pentode!
2. 46/47 DHT with a SUT. I could do a 1:4 gain step up and drive it with the 01a. Need to double check the capacitive load if it’s too much or not for the 01a. The resulting gain will be around 150.

I’d like to move back to 2 stages though. So will try the above anyway, that’s the purpose of this latest modular amplifier build.

I immediately turned into the folded cascode (aka “shunt cascode”) topology which Rod Coleman introduced me to years ago. If you haven’t read about it, please read this post, this and this one.

Challenge is for me that if I’d want to add a 6e5p in here, will mean about 40mA or more of additional current, which would require a rework of the PS. This means more work.

I looked at the D3a or even the E810F instead. Challenge is that this topology is able to deliver higher gain than μ (which is what I need), however this is only possible at low distortion levels when the valve is operated at high current / gm. This means you need to operate the valve preferably at lower anode voltages and higher currents. This is counterproductive when you want many volts to swing. Therefore some of the frame-grid pentodes with good dissipation can work well. That is why the 6e5p is a good candidate.

I recently built a PCB with folded cascode and other small circuits for pentode drivers which I’m intending to use in phono stage experiments. The PCB is the same size as my HV LED voltage reference so can be stacked on top as per below:

Swinging 200Vpp at very low distortion is not easy. The D3a was a bit dissapointing on my tests so I moved onto the E810F. It has a higher gm so works better here. I tried the first setup like this:

E810F folded cascode. Not that great performance at 100Vpp

And the results aren’t as great as expected. Gain is great however, the 0.4% THD can be easily improved by many simple triode-connected contenders at 100Vpp.

If you want to do better, we need to operate the E810F at higher current, which turns out lower voltage to keep within the safe power dissipation levels. The below is a good compromise:

49dB driver stage! This can do 0.2% at 150Vpp but that’s about its maximum swing.

Gain turns out to be too much. If you want to lower R1, you need more current. Again means you need to watch for the ZTX558 dissipation limits or change the transistor. All of these means more current in the CCS which defeat my objective. Mind you that the attached is very similar to my phono stage and performs amazingly at low level signals. I’m currently working on a new phono stage and likely reuse this topology.

Nice to have been diverted my attention on this. You learn a lot from experimentation after simulating in Spice. No doubt. I will go back to my original plans now.

And yes, will make the time to do a proper writeup of the 300B amp. I need to get around to measure it in more detail. Even I still haven’t put in the EML300B valves! Still running with the chinese ones for testing.

My latest 300B amplifier brought me again closer to the D3a. I have to say it’s an exceptional driver for this amp as well as it can perform in a phono stage at same excellent level.

I took the opportunity to trace again one Siemens D3a NOS boxed as “POS 1.176 Q 31-X 601”. This one was handpicked as it measured at 102% (31mA) in triode mode with a Gm of 39mS.

I wanted to develop a pentode model for phono experiments so put this lovely valve back in the eTracer and used the Extract Model tool from Derk Reefman to develop the model below.

You can see current saturation effects clearly on the pentode traces when approaching to 0V, that is to say below 0.5V:

Here is the result after a few tweaks:

You can see the slight divergence of the model due to the current saturation experienced at very low grid voltages

Here are the 2 models and would be great to have your feedback after using them. You can download the files from here: D3a-Pentode-100V and here:D3a-Triode.

Pentode Model:

****************************************************

.SUBCKT D3a-Pentode-100V 1 2 3 4 ; A G2 G1 C;

*      Extract V3.000

* Model created: 10-Jun-2020

*

* Curves traced and model developed by Ale Moglia

* (c) 2020 Bartola Valves

* www.bartola.co.uk/valves email: valves@bartola.co.uk

*

*

X1 1 2 3 4 PenthodeDE  MU= 77.9 EX=1.433 kG1=  25.7 KP= 587.2 kVB =  2940.0 kG2=  106.5

+ Ookg1mOokG2=.29E-01 Aokg1=.13E-04 alkg1palskg2=.29E-01 be=  .049 als=  2.57 RGI=2000

+ CCG1=6.8P  CCG2 = 0.0p CPG1 = 0.04p  CG1G2 = 9.5p CCP=0.05P  ;

.ENDS

****************************************************

.SUBCKT PenthodeDE 1 2 3 4; A G2 G1 C

*

* NOTE: LOG(x) is base e LOG or natural logarithm.

* For some Spice versions, e.g. MicroCap, this has to be changed to LN(x).

*

RE1  7 0  1MEG    ; DUMMY SO NODE 7 HAS 2 CONNECTIONS

E1 7 0 VALUE=

+{V(2,4)/KP*LOG(1+EXP(KP*(1/MU+V(3,4)/SQRT(KVB+V(2,4)*V(2,4)))))}

E2   8 0 VALUE = {Ookg1mOokG2 + Aokg1*V(1,4) – alkg1palskg2*Exp(-be*V(1,4)*SQRT(be*V(1,4)))}

G1   1 4  VALUE = {0.5*(PWR(V(7),EX)+PWRS(V(7),EX))*V(8)}

G2   2 4 VALUE = {0.5*(PWR(V(7),EX)+PWRS(V(7),EX))/KG2 *(1+als*Exp(-be*V(1,4) * SQRT(be*V(1,4))))}

RCP  1 4  1G      ; FOR CONVERGENCE A  – C

C1   3 4  {CCG1}   ; CATHODE-GRID 1 C  – G1

C4   2 4  {CCG2}   ; CATHODE-GRID 2 C  – G2

C5   2 3  {CG1G2}   ; GRID 1 -GRID 2 G1  – G2

C2   1 3  {CPG1}  ; GRID 1-PLATE G1 – A

C3   1 4  {CCP}   ; CATHODE-PLATE A  – C

R1   3 5  {RGI}   ; FOR GRID CURRENT G1 – 5

D3   5 4  DX      ; FOR GRID CURRENT 5  – C

.MODEL DX D(IS=1N RS=1 CJO=10PF TT=1N)

.ENDS PenthodeDE

****************************************************
.SUBCKT D3a-Triode 1 2 3; A G C;
* Extract V3.000
* Model created: 10-Jun-2020
*
*
* Curves traced and model developed by Ale Moglia
* (c) 2020 Bartola Valves
* www.bartola.co.uk/valves email: valves@bartola.co.uk
*
*
X1 1 2 3 TriodeK MU= 77.92 EX=1.515 KG1= 25.7 KP= 587.2 KVB= 2940. RGI=2000
+ CCG=6.7P CGP=3.3P CCP=1.0P ;
.ENDS

****************************************************
.SUBCKT TriodeK 1 2 3; A G C
*
* NOTE: LOG(x) is base e LOG or natural logarithm.
* For some Spice versions, e.g. MicroCap, this has to be changed to LN(x).
*
E1 7 0 VALUE=
+{V(1,3)/KP*LOG(1+EXP(KP*(1/MU+V(2,3)/SQRT(KVB+V(1,3)*V(1,3)))))}
RE1 7 0 1G
G1 1 3 VALUE={0.5*(PWR(V(7),EX)+PWRS(V(7),EX))/KG1}
RCP 1 3 1G ; TO AVOID FLOATING NODES IN MU-FOLLOWER
C1 2 3 {CCG} ; CATHODE-GRID
C2 2 1 {CGP} ; GRID-PLATE
C3 1 3 {CCP} ; CATHODE-PLATE
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)
.ENDS TriodeK

My friend Rob was right. The UX120 sound is unique. Perhaps is due to its cylindrical anode, thoriated-tungsten filament. Anyhow, he was also right that they pick hum and are extremely microphonic!

Well, not a candidate for a preamp indeed. However, with a low mu of over 3, it just added an extra gain kick into my 300B system:

The modification of the Mule was quick an easy. Fitted back in the UX4 sockets, adjusted the SiC array to 6V (5.8V drop) on filament bias. The hybrid mu-follower boards were adjusted down to 85V and 7-8mA. They are DC coupled into a pair of Source Follower boards biased at 15mA

Also I fitted the copper screen board on the back as I had in my Ba/Aa preamps. This works perfectly and when grounded there is no hum picked by the valve. No need to cover the valve at all.

The sound is amazingly good. Level of clarity and detail which is unique of an DHT. Very similar to the 01a sound in my view. Just great. I had to put the volume pot at the output as the microphonic noise is terrible. Yet, level is set to be usable (very usable) in my system. Love this sound indeed.

The 300B still is driven by the D3a. I will experiment with them in the future as have few things to do before changing them.

And time arrived to start experimenting with the different drivers. Why? Well, the whole point of the latest modular design is that I can easily (I hope) make quick changes and experiment.

The D3a driver board can be replaced with a new one. In fact I will be building a pair of stacked boards (which are in essence ground planes) to hold the SUT as well as a pentode screen bias regulator and the driver board.

First incarnation will be on the 46 DHT. However, drilled the board to fit either loctal (e.g. 4P1L, C3g, etc.) as well as noval for other 9-pin drivers I have in mind.

The build process is fast. I have the ground plane PCBs, which I drill (M3) to fit multiple plastic hex standoff to hold either PCBs or different turret/2mm connectors to plug in/out the board as needed. This LEGO-like approach is amazing, very happy with the flexibility and speed of work I can get with this.

Stay tuned.

Finally I managed to get the time to write this up. Hope you enjoy it, you can find it here.

The 300B amplifier journey continues as planned. A bit lower than expected, however small steps being made. Recently I mentioned about the 46 driver.

I made some progress on the LL7903 step-up transformer which is wired on 1:8.

With 1:8 wiring, don’t expect same performance level as in 1:2. However, I carefully trimmed the Zobel network in the secondary to reduce the resonant peak. The suggested network obviously isn’t fit for purpose in this mode and even tuning the resistor with the suggested 400pF isn’t good enough to tame the peak. I resourced my box of Russian NOS mica capacitors and pulled out a 4,700pF one. By playing with the frequency response test y tuned it manually with a potentiometer and found that 4K7 was right resistor:

You will see on the picture below, the driver will have 2 boards. One mounted on top of the other. The lower PCB holds the LL7903 pair and there is further room for a couple of fixed bias regulators. When I get to experiment again with pentode drivers, I will use this space to place the screen regulator:

Here is the frequency response for the LL7903 in 1:8 with the zobel network. The secondary is unloaded as it goes straight into the 46 grid. The source impedance in this case is 50Ω. Shouldn’t be massive difference in response with the DAC’s impedance which is lower. I guess you can’t expect the 70kHz response of the 1:2 wiring and the LF pole will be dominated by the 46 hybrid mu-follower. The 27kHz is good enough for me:

The driver circuit remains the same. I will play with the operating point so Rk may change. As am running the 300B from a single supply, I may run the 46 cooler somewhere between 20 and 30mA depending on the current balance needed with the output stage:

With a mu of 5.6 and 1:8 SUT we should get a gain close to 44 (32.8dB). With high efficiency speakers it will do fine with a single stage for the 300B. Promising. As said before, my previous 46 driver incarnation was a delight and sounded fantastic. Yet, I will never implement the crazy idea again of filament bias with these bias voltage and current levels needed!

Going DHT end to end

As previously mentioned, I played around with the 46 driver.  I love it sound and is a great valve. However, there were 2 reasons that pushed me to switch to the 47. Firstly, I wanted an extra tad of gain. Secondly, I have a nice stash of RCA 247 globe which measure extremely well. I’ve been reserving it for a 47 PP amp with local feedback (a la Pimm) and hopefully will get to in the future. Anyhow, the 47 in triode mode has a mu of about 8 which in combination with the SUT, gives me good gain to drive my 300B. After tweaking on the bench the stage for optimal swing and distortion performance, I ended up with the following circuit:

Ok, there is a catch. I use a low output impedance DAC which I then control via software the volume. If you want to add a volume potentiometer then you have an issue as will impact the frequency response. If you want to implement this, you will have to use a cathode follower to drive the LL7903, something like this will work perfectly. An AVC would be an option. I’m using Slagle’s AVC which is superb.

Taking the 47 (or 46) around 200V and 30mA gets the best out of it. You also need to watch out for some old valves as they may struggle to provide full swing at low distortion and (more importantly) generate too many unwanted odd harmonics at high voltage swings. You can check this with an FFT view at the mu output of the hybrid mu-follower.

I will not explain again why I personally use the mu-follower, however there is a great advantage here. You can use avoid a bypass  capacitor at the cathode resistor which is great for this higher Vg valves. As filament bias becomes unpractical here (I’ve been there done that), the hybrid mu-follower wins here, and sounds great!

There are a few compromise adjustments made here as have maximum 200mA DC capacity at the HT, so dialled the 300B down to safe levels. Also I’d reduced about 1V the filament regulator raw voltage to help with the heatsink dissipation. Other than that, the amp is the same as previously shared:

If you look at the driver performance, it’s great to see that it can do about 0.12% for 200Vpp output. Mainly H2 and the relative H2 level is -58.1dB. Not many drivers can do this. Only few DHTs. Best I could get with IHT is about 0.2%. Anyhow this is great performance. Same flat distortion measured up to 30kHz, the driver has enough drive current. At LF (below 100Hz), the distortion of the LL7903 dominates and brings it up to 0.3-0.5% at 20Hz. Still very good:

Ignore LF noise as my bench is noisy and get 50Hz plus byproduct creeping in as induced noise into the system.

Looking at the frequency response you will see a very nice performance up to 26kHz. I tested with a 220nF/510K output network for coupling out the mu output:

There is a resonance  above 100kHz which will be tamed by the output stage, so I won’t bother on it.

Here is a 247 in action:

Ok. The modular build of the 300B allowed me to swap easily the D3a driver board for the 47 one. Only a few changes needed to adjust the filament raw supplies and I was ready to test:

Sound impressions

Ok, I’m a DHT lover and was right to get back to 100% DHT. The D3a is amazing driver, don’t get me wrong. However, the 46 and 47 brings a level of extra detail I was missing slightly. Also, removing the resistive volume control and getting back to Dave Slagle’s AVC was a perfect match. I think having only 2 valves in the signal path pays off big time. The dynamics are there, strong, fast amplifier. the detail is unique, as you can only get with DHTs. Clearly, I’ve tried many option and keep going back to DHTs.  Bass is deep and powerful. Man, I love this amp and can get loud if you push it!

Going full DHT has its toll. The power supply complexity can be daunting but worth it. A lot of iron for such a simple signal path. Well, that’s the DHT reality. I highly recommend this setup. If you can’t get the 47, there are plenty other DHTs to consider. I’d go for a VT25/10Y or 801a. Mainly a lot of the mid-mu DHT can work here. The circuit is very easy to adapt. You only need to change the cathode resistor and dial the right anode voltage on the hybrid mu-follower board. That’s it. Nothing more.

I have some EML20a valves around which I didn’t like much as preamps. they pick up him easily but they are design for this job. Also the EML30a can be a good option. I would also test my ER801a as am loving to get back so thoriated-tungsten filament sound back into my system.

Anyhow, I’m enjoying thoroughly the sound of this amp. Will keep this for a long while.

My friend Mirek sent me from Czech Republic a few valves for testing, including a pair of precious EML300B Mesh valves. I managed to slot them into my system before departing for holidays. I only listened to them for a few hours, so these are only initial impressions.

I like the sound and was expecting the additional level of detail and sound of the mesh plate. Worth addition clearly. I wouldn’t say it’s a significant step improvement, just minor, subtle details are clearer. You won’t go wrong with the standard EML300B. The Mesh is a nice upgrade but you will need to think (as always) where it best to put the money on. I’d invest in the iron and overall circuit before you get to throw more money on the output valves. That’s my view.

Anyhow, great job Emission Labs for this superb valve.

It was about time to get my hands on this driver experiment. I’ve been trying to find the time for a while and could only make it due to the obliged COVID-19 isolation upon return from holidays.

The idea is simple. I wanted to use a pentode driver to swing large volts (e.g. 200Vpp) whilst retaining the triode-like characteristic from harmonic perspective and low distortion. A nice challenge and fun to work on.

Have to say that the parallel/parallel feedback (also referred to as “Schade feedback” by some in audio) when applied locally in the output stage, does sound very nice and is a very nice way of implementing high-gm pentodes used for vertical service in TV. They can produce very low distortion and sound amazing when implemented correctly. I’m not covering this now, as it has been dwelled on for some time by many good people out there.

The circuit topology is as follows:

Instead of driving the grid, we fix the grid and we drive the cathode. This has a benefit of avoiding signal loss when the PMOS (Q1) is driving the feedback resistor array (R1 and R2). However, the screen to cathode voltage isn’t constant so distortion may increase. The benefit also is to avoid loading the anode and have flexibility on the values of R1 and R2. This can be done also by placing a NMOS transistor between the divider and the grid. This is good technique when operating in A2 and grid current is needed. C1 and also potentially another Capacitor between R2 and ground may be needed to equalise the HF response depending on the anode load and the output valve, etc.

It’s a rather complex arrangement but does work. The hybrid mu-follower is formed with the pentode and the active load (aka gyrator) which provides the maximum output swing at very low distortion.

The gain of the stage is derived by the voltage divider R1 and R2. Ideally if the valve had infinite gain, it would approximate to R1/R2. However, gain is limited on the pentode so we need to apply the formula but measuring (or estimating) the open loop gain of V1.

Challenge here is introducing some level of DC feedback to avoid drift of the valve.

The added complexity is a negative supply (in most of the cases) to feed the PMOS driving the cathode. There is a way of avoiding this at the expense of extra complexity, and this is what I did to test the stage:

The added complexity is a voltage follower and DC shifter (Q1) for the feedback divider with some element of DC feedback from the anode. In the above diagram, R1 and R2 will derive a DC voltage which will be followed by Q1 and feed the end of the feedback divider (R5 and R6). This way the negative supply is avoided, and some level of DC feedback is introduced.

 There is alternatively a way of introducing some level of DC feedback via the screen as well. The above circuit, shows Q2 providing the screen voltage. In this case, the screen voltage is obtained from the mu-output via the divider R9 and R10.

D3a Pentode Driver

So, I couldn’t wait longer and with the limited time available I put together a rat’s nest testing jig with the following implementation of the D3a pentode as driver:

A few additions. Firstly, I derived the DC feedback from the mu-output. It varies more than the anode with changes in anode current as there is a mu-resistor in series, although is rather small in the grand scheme of things.

I did a few simulations in Spice and found that reducing the screen voltage to about 70V and anode to 200V provided the best performance. There is a first board which Has a simple voltage reference (T1) which takes the input from the Screen Supply. This voltage reference provides the DC reference to Q2 and sets the cathode DC voltage. The signal input comes via C5. The feedback divider (R1 and R3) is DC-shifted by Q1 which takes the reference from the mu-follower output via R2-P1-R6. P1 is the only one I ended up using to do the DC adjustment.

The measured open-loop gain (A0) is about 280 given the low anode and screen currents, transconductance is rather low. With R1=68KΩ and R3=470Ω, I measured closed-loop gain of 98. This is perfect for a  one-stage  driver in a 300B SE Amp for example.

The D3a operates at a very low current, about 5-6mA.

The above measurement shows a good performance of 0.53% for 180Vpp and a 100KΩ.

You can see that H3 dominates from the low feedback ratio provided so there is more of a pentode-like behaviour. Yet, for a pentode stage is good performance for 40dB of gain.

The circuit drifts a bit and takes some time to stabilise. I left rather too-long  time constants (e.g. R4/C4) which should be halved or more. That will work better.

If we apply more feedback, as expected the gain is reduced. Here is an example of the same circuit with gain of 27 (29dB):

You can see that distortion is now reduced due to greater feedback to 0.22% for same output level of 185Vpp. This is very good performance, albeit the harmonic profile still has the H3 domination as before.

Interesting experiment, learned a lot from it as usual. There are ways of improving this stage to get better stability for sure.

Will I build it? Not sure, this is a much more complex circuit. Despite having a great performance, it still needs a lot of sand around it to work. It does excel in avoiding the Miller problem of the high-gain triode stages, so the source will drive without a problem the PMOS.

A similar topology is very very useful for output stages and that is where you can obtain a big improvement in implementing the traditional “plate to grid” feedback in a clever and effective way.

I’ve been doing some push-pull output stage experiments with either cathode feedback (CFB) via the output transformer (Toroidy custom-made OPT) and parallel-parallel feedback (aka plate-to-grid or a la “Schade”) with PMOS driving the cathode. See some funny experiments here on the driver side.

I breadboarded multiple auxiliary circuits to supply the Screen, elevate grid DC and drive the cathode with great results. I decided to produce a series of modular prototype PCBs to assist with multiple designs. They all have the same size and mounting holes so can stack up one over the other. I have used the same board size of the HV OSRAM LED Voltage Reference I use on my 300B SE Amp.

The boars go like this:

The circuit is out of the book standard. A voltage divider with bias and balance trimpots to adjust both valves. They are buffered with a PNP source follower which has the option of either an emitter resistor or a CCS formed by a simple LND150 which should give about 1 to 2mA stable current depending on the device IDSS:

The second PCB is a bit more complex, albeit it takes 15min to solder it through. This is a flexible screen bias PCB for a pair of valves. Also it has some flexibility to use as screen DC feedback stabilisation from the output. See the example here.

The actual circuit is a tad more complex. There is a current limiter circuit added to T1. You can use it as a source follower if you want as well. R6 is needed to avoid oscillations on T1 should it get to cut-off. You want at least 1-2mA dragging out of T1 at all times. In most of the cases you have a jumper on J1. If you want to use the circuit to stabilise screen with feedback from an output stage, then you will use B1 separate from B2:

Also made another PCB (the one in black on the right) which decouple +B with an electrolytic, a PIO cap and has a bleeder resistor plus an INS-1 neon bulb on board. It can connect 4 daughter boards.

The PMOS driver looks like this:

Similar to the screen circuit. If you’re using to drive the cathode of an output stage, you won’t use the bias voltage divider and cap-couple the input signal to the PMOS gate instead. No C1 of course.

Output stages

So playing with some nice output stages with GU-50, RL12P35, 6P36s and other lovely TV vertical pentodes here are some of the configurations you may find some use to these circuits:

I’m not a fan of gNFB and will never be. However, the PP bias circuit and the screen bias circuit in 2 boards can bias any pentode pair like the above.

I have the OPTs with CFB winding (10%) so the above is a good option to apply the local feedback to the output stage.

If you don’t want to cap-couple the driver (which should be an LTP) you can use an IT configured for push-pull operation. I love this option, however you will have to bias both grids with same regulator. You can balance the anode currents with the screen regulators which are still independent, so this is handy.

Finally, if you want to implement the parallel-parallel local feedback arrangement with the PMOS driver, here is how you will do so:

The feedback is set by the pair R1-R3 and R2-R4. DC balance can be adjusted by both the screen bias as well as the PMOS drivers. Each PMOS driver (there are 2 in a PCB) will drive the cathode. You will need likely a bigger heatsink as you will have about 10-12W in many cases dissipated at the PMOS.

Hopefully will get some more time to play with them soon.

I have plenty of these German WWII pentodes. I’ve been saving them for a few projects. Quite likely for a push-pull amp given the OB speaker project I’m cooking. Either way, I have a nice set of CFB SE as well as PP OPTs from Toroidy custom made. They are very good transformers.

So, if you want a simple SE amplifier that can do 8W, here is a good contraption to go after:

The OPT has a couple of feedback windings to provide 10% CFB. I set the valve rather hot, I know. Probably a tad too much. Should be 35W maximum. You can do GU-50 here and get more power, cheap and easy.

The driver is a gem. The EF184 is a superb pentode when triode-strapped. Can do very well on the same levels of a D3a and is (yet) somehow still cheaper. Cathode degeneration plays well and easy here in the hybrid mu-follower.

Some people would say: “pentode and cathode feedback, arghh!!”. Trust me, good path down this way when feedback is local of course.

Although the stage can do 8W and THD will jump up to 2.9% is nice to see the triode-like harmonic distribution. Here is the view at 1W:

If you push the output stage to its limits, you will get this:

Bandwidth is over 65kHz. A simple and nice amplifier. With different arrangements on feedback like plate to grid with cathode driven via a PMOS and increasing the FB ratio to 20%, you can get less than 1% at 7W output. But that’s a rather more complex amplifier. This is simpler.

Long time ago I built a series variable voltage regulator for 600V. It worked flawlessly and survived all sorts of abuses as is on my bench HT supply.

With the same circuit design, I developed the final stackable PCB (see previous post here) with this regulator:

Looks more complicated than it is. The single-supply Op Amp (LM358) needs a low voltage supply. I derived this from a simple CCS (DN2540) and a pair of 12V Zener diodes. The voltage reference is the famous TL431A and with P1 you can adjust the output voltage. The feedback resistor pair (R14 and R15) senses the output. C6 is for frequency compensation. The MOSFETs used are ST3LN80K5 which have built in protection Zener diodes, so none of the ones shown in the diagram are actually needed. T4 provides current protection to the pass device T2.

According to the LT Spice simulations, it should have about 80dB at 100Hz. I could only measure 50dB as my bench noise was about 300μV and input ripple was low already. I should test it further.

Anyhow, the other interesting test I made was to develop a 60-90V bias circuit from a 12V input. I used the cheap SMPS Chinese modules fed from 12V (100mA) into an RC circuit at the output (470R/100μF) before feeding the voltage regulator. Was nice to see the rock solid output at 70V and same 300μV noise.  I was thinking of using this approach to generate the fixed bias of the RL12P35 SE Amp – part I amplifier. In this way I could get away from a negative supply and generate the bias from the heater supply. Not sure how reliable this would be, perhaps won’t do it at all.

Some time ago I got hold of a nice stash of Telefunken EL152. These German pentodes are amazing. After playing with the RL12P35P and then obviously GU-50 (which is a copy of the LS-50), the EL152 was a nice valve to have at hand as it’s actually the LS-50 in a different bottle.

The B-10V socket is quite tricky as it seems like it was designed for the EL/FL-152 and similar Telefunken valves. Anyhow, managed to get some new ceramic ones to trace the curves and generate a Spice model. Hope you find this useful.

Here’s the model using Derk’s tool:

LT Spice models

Pentode model is below and also here: EL152-pentode-model:

****************************************************
.SUBCKT EL152-pentode 1 2 3 4 ; A G2 G1 C;
*      Extract V3.000
* Model created: 23-Oct-2020
*
* Curves traced and Model developed by Alejandro Moglia
* (c) 2020 by Bartola Ltd. UK 
* For DIY Audio Use Only
* www.bartola.co.uk/valves email: valves@bartola.co.uk
*
X1 1 2 3 4 PenthodeD  MU=  5.4 EX=1.390 kG1= 707.4 KP=  32.3 kVB =  5122.1 kG2=46214.5
+ Ookg1mOokG2=.14E-02 Aokg1=.45E-06 alkg1palskg2=.14E-02 be=   .07 als= 50.58 RGI=2000
+ CCG1=10.0P  CCG2 = 0.0p CPG1 = 0.11p CG1G2 = 0.0p CCP=14.5P  ;
.ENDS

****************************************************

.SUBCKT PenthodeD 1 2 3 4; A G2 G1 C
*
* NOTE: LOG(x) is base e LOG or natural logarithm.
* For some Spice versions, e.g. MicroCap, this has to be changed to LN(x).
*
RE1  7 0  1MEG    ; DUMMY SO NODE 7 HAS 2 CONNECTIONS
E1 7 0 VALUE=
+{V(2,4)/KP*LOG(1+EXP(KP*(1/MU+V(3,4)/SQRT(KVB+V(2,4)*V(2,4)))))}
E2   8 0 VALUE = {Ookg1mOokG2 + Aokg1*V(1,4) - alkg1palskg2/(1 + be*V(1,4))}
G1   1 4  VALUE = {0.5*(PWR(V(7),EX)+PWRS(V(7),EX))*V(8)}
G2   2 4 VALUE = {0.5*(PWR(V(7),EX)+PWRS(V(7),EX))/KG2 * (1+ als/(1+be*V(1,4)))}

RCP  1 4  1G      ; FOR CONVERGENCE A  - C
C1   3 4  {CCG1}   ; CATHODE-GRID 1 C  - G1
C4   2 4  {CCG2}   ; CATHODE-GRID 2 C  - G2
C5   2 3  {CG1G2}   ; GRID 1 -GRID 2 G1  - G2
C2   1 3  {CPG1}  ; GRID 1-PLATE G1 - A
C3   1 4  {CCP}   ; CATHODE-PLATE A  - C
R1   3 5  {RGI}   ; FOR GRID CURRENT G1 - 5
D3   5 4  DX      ; FOR GRID CURRENT 5  - C
.MODEL DX D(IS=1N RS=1 CJO=10PF TT=1N)
.ENDS PenthodeD

****************************************************
.SUBCKT EL152-triode 1 2 3; A G C;
* Extract V3.000
* Model created: 23-Oct-2020
*
*
* Curves traced and Model developed by Alejandro Moglia
* (c) 2020 by Bartola Ltd. UK 
* For DIY Audio Use Only
* www.bartola.co.uk/valves email: valves@bartola.co.uk
*
X1 1 2 3 TriodeK MU= 5.43 EX=1.423 KG1= 707.4 KP= 32.3 KVB= 5122. RGI=2000
+ CCG=0.0P CGP=0.0P CCP=0.0P ;
.ENDS

****************************************************
.SUBCKT TriodeK 1 2 3; A G C
*
* NOTE: LOG(x) is base e LOG or natural logarithm.
* For some Spice versions, e.g. MicroCap, this has to be changed to LN(x).
*
E1 7 0 VALUE=
+{V(1,3)/KP*LOG(1+EXP(KP*(1/MU+V(2,3)/SQRT(KVB+V(1,3)*V(1,3)))))}
RE1 7 0 1G
G1 1 3 VALUE={0.5*(PWR(V(7),EX)+PWRS(V(7),EX))/KG1}
RCP 1 3 1G ; TO AVOID FLOATING NODES IN MU-FOLLOWER
C1 2 3 {CCG} ; CATHODE-GRID
C2 2 1 {CGP} ; GRID-PLATE
C3 1 3 {CCP} ; CATHODE-PLATE
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)
.ENDS TriodeK

It’s been a long and weird year with COVID-19. This evidently has kept me a bit distracted and changed priorities (as probably has done to you as well). Anyhow, here are the belated curves I was asked to publish.

The SRS-551 is a lovely powerful transmitting pentode from RTF. Similar (or close enough) to RS1003 and F3a. Much more accessible in price is the SRS-551. I have a nice stash of NOS valves somewhere lost in my attic storage. Either way, they sounded nice and could definitely live in my setup. The valve was intended for audio modulation in FM transmitters so it’s very reliable and linear device. Distortion was extremely low when I measured it but most importantly it could work really well with local feedback to squeeze out 14W in Single-Ended at nearly 490V/100mA bias point with an LL1623 OPT. The distortion was below 0.7% predominantly H2.

Here are the curves traced with uTracer back in 2016 and model developed using Derk’s tool:

LT Spice Models

Below is the pentode model, you can download it from here as well:SRS551-VS100V

****************************************************
.SUBCKT SRS551-VS100V 1 2 3 4 ; A G2 G1 C;
*      Extract V3.000
*
* Model created:  3-Apr-2016
*
* Curves traced and model developed by Ale Moglia valves@bartola.co.uk
* www.bartola.co.uk/valves
*
X1 1 2 3 4 BTetrodeDE  MU= 24.02 EX=1.281 kG1=  89.0 KP= 136.4 kVB = 2598.7 kG2=  739.5
+Sc=.31E+00 ap=  .039 w=    15. nu=   .00 lam=    11.6
+ Ookg1mOokG2=.988E-02 Aokg1=.12E-04 alkg1palskg2=.988E-02 be=  .138 als=  5.27 RGI=2000
+ CCG1=27.0P  CCG2 = 0.0p CPG1 = 0.24p  CG1G2 = 0.0p CCP=15.0P  ;
.ENDS
****************************************************
.SUBCKT BTetrodeDE 1 2 3 4; A G2 G1 C

*
* NOTE: LOG(x) is base e LOG or natural logarithm.
* For some Spice versions, e.g. MicroCap, this has to be changed to LN(x).
*
RE1  7 0  1MEG    ; DUMMY SO NODE 7 HAS 2 CONNECTIONS
E1 7 0 VALUE=
+{V(2,4)/KP*LOG(1+EXP(KP*(1/MU+V(3,4)/SQRT(KVB+V(2,4)*V(2,4)))))}
E2   8 0 VALUE = {Ookg1mOokG2 + Aokg1*V(1,4) - alkg1palskg2*Exp(-be*V(1,4)*SQRT(be*V(1,4)))}
E3   9 0 VALUE = {Sc/kG2*V(1,4)*(1+tanh(-ap*(V(1,4)-V(2,4)/lam+w+nu*V(3,4))))}
G1   1 4 VALUE = {0.5*(PWR(V(7),EX)+PWRS(V(7),EX))*(V(8)-V(9))}
G2   2 4 VALUE = {0.5*(PWR(V(7),EX)+PWRS(V(7),EX))/KG2 *(1+als*Exp(-be*V(1,4) * SQRT(be*V(1,4))))}
RCP  1 4  1G      ; FOR CONVERGENCE A  - C
C1   3 4  {CCG1}   ; CATHODE-GRID 1 C  - G1
C4   2 4  {CCG2}   ; CATHODE-GRID 2 C  - G2
C5   2 3  {CG1G2}   ; GRID 1 -GRID 2 G1  - G2
C2   1 3  {CPG1}  ; GRID 1-PLATE G1 - A
C3   1 4  {CCP}   ; CATHODE-PLATE A  - C
R1   3 5  {RGI}   ; FOR GRID CURRENT G1 - 5
D3   5 4  DX      ; FOR GRID CURRENT 5  - C
.MODEL DX D(IS=1N RS=1 CJO=10PF TT=1N)
.ENDS BTetrodeDE

And here is the triode model I developed manually using Dmitry’s tool. You can also triode-wire the pentode model from Derk if you want to. You can also download it from here: SRS551-triode-model

**** SRS551 TRIODE ******************************************
* Created on 04/03/2016 19:56 using paint_kit.jar 2.9 
* www.dmitrynizh.com/tubeparams_image.htm
* Plate Curves image file: SRS551 triode.png
* Data source link: www.bartola.co.uk/valves 
*
* Traced and model developed by Ale Moglia valves@bartola.co.uk
*
*----------------------------------------------------------------------------------
.SUBCKT SRS551-T 1 2 3 ; Plate Grid Cathode
+ PARAMS: CCG=3P CGP=1.4P CCP=1.9P RGI=2000
+ MU=21.6 KG1=240 KP=140 KVB=300 VCT=0.132 EX=1.4 
* Vp_MAX=350 Ip_MAX=200 Vg_step=2 Vg_start=0 Vg_count=8
* Rp=4000 Vg_ac=55 P_max=60 Vg_qui=-48 Vp_qui=300
* X_MIN=81 Y_MIN=52 X_SIZE=481 Y_SIZE=465 FSZ_X=1117 FSZ_Y=651 XYGrid=false
* showLoadLine=n showIp=y isDHT=n isPP=n isAsymPP=n showDissipLimit=y 
* showIg1=n gridLevel2=n isInputSnapped=n 
* XYProjections=n harmonicPlot=n harmonics=y
*----------------------------------------------------------------------------------
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 ; TO AVOID FLOATING NODES
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} ; CATHODE-GRID 
C2 2 1 {CGP} ; GRID=PLATE 
C3 1 3 {CCP} ; CATHODE-PLATE 
D3 5 3 DX ; POSITIVE GRID CURRENT 
R1 2 5 {RGI} ; POSITIVE GRID CURRENT 
.MODEL DX D(IS=1N RS=1 CJO=10PF TT=1N) 
.ENDS 
*$

It’s been a while since I last share some of the experiments. I’ve done a lot with different drivers topologies and pentodes lately, yet not much time to sit down and do a proper write-up.

I have a project in hands which is a small Mono amplifier to drive a Celestion guitar speaker I have in my Marshall clone. I will use this to play my synthesiser gear. Low volume levels needed in my workshop so can do with a nice small Single-Ended Amp.  For now, I focused on playing with different driver topologies and settled with the pentode with local feedback as a result of my evolving experiments and the guidance provided by JC Morrison on this topic:

JC Morrison has covered this circuit in detail so I suggest you read his blog post which are fantastic.  Therefore I will only add a few notes about the circuit. I want to get a good gain level which can be done without sacrificing too much the input impedance. About 30KΩ-ish would be ok without pushing R2 to be too big. I’m trying to keep cathode bias with diodes to maximise the transconductance of the valve and therefore the open loop gain. The screen voltage stability will be derived from the mu-output of the CCS. In practice will have R6 and R7 with a trimpot in between for fine adjustment.

Why the EF37a? I have a few of them NOS Mullard and have yet a fascination of pre-WWII valves. It sounds and looks great as well!

Here is the nice small modular approach. Both boards are mounted on top of each other. The PCBs have been tested before, so it’s a question of soldering on now:

I hope to report further progress during the Holiday season.

Stay tuned

On the last post, I shared this great circuit. Now, with the Holiday season and being locked down I somehow find the time to build it. A very quick process as it’s all modular. I’ve got all PCBs that can be interconnected like LEGO, that’s what I have always in mind when I design a new PCB.

I had already built a Screen Bias PCB, which was fully populated for a stereo setup. I’m only using one side of the board. A bit of a waste, but the board was ready available. On the back you can see the Source Follower. On top of the Screen Bias board there is the output coupling capacitor. I added two 4-pin turret boards on each side to facilitate the connections between the two daughter PCBs. The one on the top (right) holds the EF37a and under the octal socket you have 2 4-pin turret boards with the SiC diodes and the feedback resistors. On the back there is space for a filament regulator (although I’m planning to run this one with AC), the mini CCS board and connecting turrets:

The previous picture I think explains it all. Here is the assembled unit:

Let’s see how this little unit behaves. I was surprised how stable it is given it has 2 CCS fighting with each other (the CCS anode load and the pentode). The screen feedback works really well to stabilise the unit and loading the mu-output is pretty handy.

Here is a snapshot on its performance:

The stage can do below 0.3% for 100Vpp. Mainly H2 dominated and the higher harmonics only appear when you’re pushing it to over 150Vpp.

Nice to see it can do nearly 50kHz with a gain of slightly over 40dB. I think it’s a very nice stage.

Yes, there are better pentodes than the EF37a, but I love this one!

More to be done…

After a few exchanges with different people, it was time to make a few tweaks to the EF37a driver.  Increasing the G3 voltage helps in squaring up the pentode curves as shown by few before. By increasing G3 to +12V or up to 20-25V can be beneficial in the long voltage swing of the pentode. There are a couple of interesting threads in DIYaudio to check out if you’re interested.

I should have fired up the eTracer and play with G3 biasing before modding the driver, but couldn’t help myself from doing this mod as it was very simple. The Screen voltage circuit I use has 2 identical versions of it in the PCB as it was designed for a stereo setup in mind. As I reused a PCB I built for other tests, I had a readily spare screen voltage regulator which could be easily tweaked to supply G3 at the levels needed:

To ensure I could deliver 10-20V range at G3 with same PCB, I operated the G3 grid from the G2 grid bias circuit which provides a 112V output. This actually gives me good adjust range on G3. Now the fun part was to experiment different G3 voltages looking at the FFT plot at maximum voltage swing (i.e. 200Vpp in this case).

It  was interesting to see the following:

1. Above 14-15V, the distortion decay profile was the best. Progressive reduction from H2 to H6. However, the distortion was higher. Overall THD up to 0.58%
2. Reducing from 14V to 13V, you can get a predominantly higher H2 & H3, whilst H4-H6 reduces down to an even level. Overall THD is lowered
3. Finally when you get down to 12V, things look probably the best in terms of overall THD. Odd harmonics get slightly lowered when even are a tad higher. You can see the profile below. Overall THD comes down to 0.47%. Nearly 0.1% improvement thanks to 1.3dB reduction in H2 and about 7-8dB on the rest of the harmonics. Good stuff.

This is now looking better. Mind you that the performance of this stage between 100Vpp and 150Vpp is extremely good with overall distortion below 0.2%. I’m just pushing it to the limit to see how it can be optimised.

Now need to move on to the rest of the mono amplifier. It will be fun to listen to this as well.