I’ve done several tests using a simple gyrator PCB test mule. It was time to build a proper and flexible test mule for extreme abuse:

1. 2 Boards for current flexibility
2. Board 1: BF862
3. Board 2: J310
4. Top FET is IXTP08N100D2 for 1000V operation
5. ZIF socket pins for CCS reference resistor and RMu. This will give the necessary flexibility to try any combination in the gyrator depending on the triode and or the power supply
6. External pots for ease regulation of anode voltage

The top MOSFETs are bolted on the aluminium case which will act as heatsink. For tests this should be sufficient.

The top plate of the case looks like this:

There are 4mm posts are for HT supply, GND, mu-output, Anode. There are also a pair of 2mm posts for current sensing per board.

Some further soldering to do and job done!

On my previous post, I covered my initial build work on the gyrator test mule using the gyrator PCB. I did all the lovely soldering work (which I do enjoy not like milling or drilling) and proceeded to do several tests.

Some interesting observations based on my abuse of the gyrator which yielded on several MOSFETs and JFET damaged as a result:

• CCS reference: I used an external multi turn 5KΩ potentiometer via lead cables. I wired it incorrectly and that contributed to one of the initial faults. Be sure you look carefully on this if you use an external pot. if you use the on-board trimpot, this is not an issue.
• JFET: this is the interesting one. If you want to run the lower JFET at very low biasing current for a larger jFET (e.g. J310) you will find that the JFET needs to operate close to cut-off voltage (somewhere between -2 and -6V). This VGS required will definitely forward bias the Zener protection diode D1 and prevent from reaching lower bias current (I found it about 10mA for J310). To resolve this you just need to add a back to back zener as shown below. This isn’t a problem for an BF862 or a 2SK170 as their cut-off voltages are quite small.
• Failure: if you abuse the FETs, they will die. And if they die you will get a nice short across them and you will measure nearly HT at the mu output. Just replace the MOSFET and JFET (probably both are damaged)

Back to back 15V protection zeners hacked in the PCB

Here is the implemented J310 gyrator:

J310 Gyrator with back to back zeners

Final Build

The build itself is quite packed. Thanks to the 2mm connectors I can easily service the boards. The fiddling part is replacement of FET and the MOSFET. My word of advice here is to cut the device leads and leave the leads soldered in the PCB. You then solder the new FET/MOSFET on top of the old leads. It’s not neat, but it’s much easier than removing the previous solder and re-soldering the components.

Final view before closing the top cover

Some interesting tests

First test I’d like to share is the famous 4P1L. As a driver it’s superbly linear, however due to its low mu (8-9) we need a step up transformer or another stage. Anyhow, here is a simple test with filament bias. Without being pretentious, the operating point isn’t optimal and it can be improved should you need 200Vpp or more:

50 Vrms test on 4P1L

The harmonic profile is very nice and 0.12% @50Vrms (140Vpp) is very good.

Now let’s look at the secret valve which has been talked about in the forums for some time. The 12HL7 frame-grid pentode (Va=400V, Vs=330V – Pa=10W). This is a beauty of a pentode which is very linear. It performs as well as the 6e6p-e, 4P1L in terms of linearity but has a nice mu of about 30 which makes it an attractive option for a driver:

210Vpp drive of 12HL7 in triode mode!

As you can see on the above, the 12HL7 is biased at about 30mA, 200V at the anode and the distortion is very low (0.2%), with predominant H2 profile.

Now I can do some more interesting tests quicker with this mule, including Parafeed drive. I hope the mule doesn’t give up and die eventually

Jose Martins sent me an email with his recent built on the 27/56 preamplifier using the gyrator load and these PCBs. I recently posted an idea using the lovely 27 IHT valve here.

Here is a picture of the finished preamplifier:

Hello Ale,

Here is some photos of my new creation where I used your Gyrators boards.

This is a line 27/56 preamplifier. Working great with your Gyrator boards. I’m planning to use also in near future a second output with Magnequest parafeed transformers B7/15K.

I built  this project in old Antique Sound Lab AQ2001 inox steel chassis that I have for some years in my garage.

The filament supply for tubes are regulated and noise of this preamplifier is very low even with old Arcturus 127 triodes from  1923. It is very important to connect the cathode of these old tubes to the filament.

I can also use 56 triodes to get more gain with no modification. Now I must test all my stock of 27 and 56 triodes. I’m very happy with the sound even compared with others DHT tube preamplifiers I have (26, 4P1L, VT-25, 1G4, RE084, Ba, Ca, TM2).

I’m very happy with your boards. Thank you for your great products.

Best Regards

José Martins

Ps: please excuse my bad English.

This preamp looks great, well don Jose. BTW: your English is really good, so nothing to apologise of. I hope to get some circuit diagrams and description from Jose to update the post. In the mean time, here are some detailed pictures of this fantastic build:

I went to see my friend Tony today and helped him to fix his 01a preamp implementation. Time ago Tony used a prototype version of my gyrator PCB to build the Gen2 preamp with the addition of an output follower to address the slew rate limitations he had on his system due to the larger capacitive load.

Luckily we found the fault easily and it was a bad solder in one of the smoothing HT chokes. Once fault was rectified, we proceeded to take some measurements of this preamp.

The preamp circuit diagram is below. Is the classic 01a preamp Gen2 with the addition of a basic source follower to drive Tony’s amp:

Tony’s 01a preamp implementation

Looking at the frequency sweep the preamp shows about a gain of x8 (18dB) flat from 3Hz up to about 180kHz. This is impressive and is thanks to the bootstrapped input capacitance of the MOSFET which provides a gentle load to the 01a:

FR of Tony’s preamp

I don’t think you can get better than this

Here is a plot of the harmonics and THD at a signal level of 2Vrms. Distortion is extremely low and with a double output at 4Vrms increases only to 0.008%:

The build is really neat. The LT supply is on a separate chassis and HT is on same preamp chassis (see TXs on the right). The pair of sockets has rubber suspensions, the Rod Coleman boards are at the bottom and you can see the output caps at the top as well as the volume control and input selector on the left:

Testing the preamp

Least to say: his system sounds impressive. This is one of the best 01a preamps I’ve heard myself. Well done Tony!

I’ve been on some business travel so haven’t had much time to work on stuff, however I did get a set of gyrator boards for a friend and a customer:

1. BF862 configured for 4P1L preamp
2. 2SK170 configured for 01a preamp

4P1L preamp with BF862 gyrator

Many have asked me about this preamp with gyrator load. Here is the latest implementation which I preferred most in terms of sound. The mu resistor is 470Ω which is a nice compromise between BF862 transconductance and distortion. I adjusted it on test. I use a 100nF for C1 so R6 is 10MΩ. R4 can be either 300KΩ, 330KΩ or even 390KΩ. Difference would be only on the voltage range for the CCS. I found running it at 25mA to be perfectly fine, some BF862 can even do J310. I prefer this SMD compared to the J310. It performs much better even at high frequency:

4P1L gyrator boards

Back to the 01a Preamp

The finished boards looked like this:

01a Gyrator boards ready for customer

Here is a quick measurement with a lovely UV201a (circa 1922-1924) with brass base:

The FR is not as great as the BF862 with 4P1L, however it reaches out to 100kHz which is fantastic compared to other preamps using different loads.

Here is the harmonic profile for a 2Vrms signal showing the low distortion and the cadence of harmonics characteristic of this valve:

How many more times

Led Zeppelin wrote a fantastic song on their first album: how many more times. You may not be a rock fan, but hey: what a great song. How many more times do I want to get back to this “slew rate” theme? I don’t know, as much as I have to. Plenty of comments out there of bad designs with wimpy drivers attempting to take the 300B/2A3 or even 45 valves to full tilt with disappointing results. Either way, they always blame the valves.

I came back to revisit the driving of capacitive loads effectively as I’m working on a new 4P1L PSE amplifier. Slowly, but getting there. Previously I looked at adding a buffer to the 01a preamp as a result of slew rate limitations found in Tony’s implementation of this preamp.

The circuit design

Well, nothing new here. There are plenty of follower designs out there. From a basic MOSFET with a source resistor like the one implemented by Tony to more elaborate ones which improves the circuit by reducing the output impedance and stabilising the quiescent current of the CCS.

I played with many designs and settled with a simple one based in MOSFETs. You can argue that a pair of cascoded bipolar can improve the performance, but I love the simplicity and effectiveness of this design. The follower is based around the MOSFET M1. In my case, I used a Cree SiC MOSFET C3M0280090D as the capacitances are very low (i.e. Crss=5pF @VDS=50V), has relatively high transconductance (Gfs=3S) and performed well. You can use whichever enhancement MOSFET you like here (even the same STP3NK60ZFP used as M2 will do really well), and you should try your choice and listen. The circuit is DC coupled to the driver stage, however if you’re using this circuit to driver an output stage in DC with grid bias (see below diagram), then the output capacitor C2 is omitted (as the output of the follower is connected directly to the valve grid) and then C2 is moved to the input between the driver and buffer (as Cin):

In between the two you will add your bias voltage (Vbias via Rbias) and instead of referencing the circuit to ground it will go to the negative bias supply (VB). If you move C2 from the output to the input (Cin) then its value can be reduced significantly (the LF pole is formed by Cin and Rbias). If you bias network to the gate of M1 is high impedance (e.g. 1MΩ) then you can reduce the value of Cin to 10-20nF which its great.

You need VB to allow for enough headroom of your driving signal. however the lower, the more dissipation will take place across M2.

R2 is the classic gate stopper (as so it’s R3), but I also added ferrite beads (f1 and f2).  D1 and D2 are key to protect the maximum VGS of M1 whilst they are not needed for M2 as they are built-in on the MOSFET. M1 is arranged in source follower mode but instead of having a source resistor a CCS formed by M2 is in place as it will minimise the output impedance and maximise the performance of M1. M2 gate has negative feedback from Q1. Q1 senses the current flowing throw R4 which stabilise the current on the M2 FET. If more current flows it will increase conduction of Q1 and therefore turn off M2. I set the buffer to 10mA quiescent current on M1 by setting R4 to Vbe/Iq. Vbe is 0.7V typically  and Iq is the quiescent follower current. 10mA should be plenty unless you want to drive heavy loads.

I initially omitted C1, but my tests proved that you need a decoupling cap there. Good to filter unwanted RF noise. I had it at least in my workbench.

For an HT of 200V, both M1 and M2 are dissipating about 1-1.2W. M2 can better live with a clip heatsink and M1 since its size, can work without. If you are increasing HT then you need to look at proper heatsinks as needed.

Tests

Here is a simple test which shows the slew rate effect. Firstly, we have the output of the SiC MOSFET follower in the below FR diagram. You can see that with a heavy load like 2,200pF the response is good up to 200-210kHz for an output level of 10Vrms. This is what you would get out of the BF862-based gyrator without a demanding load, however, you will not get this FR response without the buffer (as we will see later below). The LF difference is due to the 220nF output capacitor interacting with the 1/gm output resistance of the SiC MOSFET whereas I have a 1μF output capacitor on the Gyrator Test Mule:

The second FR test shows the response of the gyrator without the output buffer. This is just a plain UX-201A valve biased with filament bias and Ia=3ma. The load is same aggressive 2,220pF//100kΩ. The HF response is impacted due to the slew rate effect and is reduced to 90kHz. Still pretty good but if you have an output stage (in particular PSE) you would expect potentially higher capacitance and of course a large output signal level. Remember, slew rate is directly proportional to the signal level.

The above THD plot shows the harmonic response of the UX-201a BF862 gyrator stage DC-coupled to the SiC follower. The output level is 10Vrms. Distortion is lower than 0.001% Any unwanted gremlins below -90dB are the noise of my workbench setup. Even the 32kH and 40kHz spikes which I suspect are my LED lightning or similar.

I ran the same tests at 10kHz and 20kHz and you can see the 2-4dB distortion difference due to slew rate.

Doggy bag

1. SiC follower performs really well. Watch out for oscillation, beads and gate stoppers are a must. Either good MOSFET with low Crss (e.g. 8pF) and high transconductance should work here as well.
2. Distortion of the SiC MOSFET follower is really low. And it sounds nice!Slew rate is minimized and maximum performance of driver is achieved with the follower.
3. However, if you  are using the gyrator for a DHT preamp, unless is paramount and the load is really demanding (e.g. >1nF) then you don’t need a buffer stage as the slew rate would be ok and FR of the preamp would be high enough.

Looking to build a new 01a Gen2 preamp shortly. This time will be for the rare UV-201a valves. They need a special socket UV4 rather than the usual UX4. See width difference of both filament pins. In the UV4 all pins have same size as below:

Weather they have brass or bakelite bases, the new UV4 sockets produced by Luciano Bandozzi (Jakeband) are of supreme quality. I highly recommend them:

I already reviewed these and you can see more info here. I hope to post some pictures of the new preamp shortly

A common theme of our age, run. We keep running, but not physically only. Our minds are constantly interrupted, like an CPU IO interrupt. Well, sort of. You get the gist. We run from home to work, from there to here, from here to nowhere. The reality is though, we are always on the move. Do we really like that? Hey, probably no, but that’s the way things work these days.

I have a hectic life myself, no doubt. As so probably you. Either way, the most precious moments in life – at least in the XXI century – is to unwind, stop and enjoy a bit of the slow movement. Slowly put the needle on that record, slowly sit down on your comfy sofa and slowly pour that single malt.

The rest is your imagination that I will so much struggle to put into words

Ale

I have a new microscope for SMD soldering. If you want your gyrator PCB with a presoldered BF862, now it’s possible!

Fall 2016

End of summer is here, and for some the beginning of the building season. Well, not for me am afraid. My parental duties and work are keeping me very busy these days. I don’t have the free time I used to have before (I guess I’m not the only one on this so won’t rant on it). Today, building DIY audio gear is  a matter of  a well planned and negotiated  free-time that worths more than gold to me. Well, that’s the way it goes. Anyhow, I picked up my daughter from nursery yesterday and on the way back I was faced with this beautiful landscape. I guess nature give us some gifts from time to time, you just happen to be on the right place at the right time:

Standing on the middle of the street with the pram was a bit dangerous so had to park my daughter on the side whilst I managed to take this picture. Time ago, I’d have taken probably a long time to take this snapshot, but now it was as quick as a bank robbery. Just take the phone out and shoot – you can’t take your time when you have a crying toddler on the pram!

A tail of buffers

I think I have spent far too much time designing, building and testing preamplifier, perhaps more than amplifiers lately. I don’t know why. I guess I fell in love with the preamps and their contribution to sound overall. Who knows, who cares.

A common challenge most of the owners of SS amps is: input impedance is too low. Hey, do you remember I had to put out all of the dangerous HT equipment when my daughter was born? So I’m one of those who own an SS amp (despite I keep building a new 4P1L PSE and 300B amps at this moment). Mine sounds very nice and not like a traditional SS amp, though. The main disadvantage is its low input impedance (below 10K). There are many out there with SS amps with low input impedance so if you are one of them, you’d find this post interesting (if not, just keep reading it for fun!). Also, I have seen too many cases (or comments around the audio forums) about the poor performance of an AVC or preamp which in cases was attributed to the demands of a low input impedance or high capacitive loads. They are hard to drive, so well. What do you expect?

The circuit design

For once, I think I will surprise you. I’m not using any hollow device here. Pure sand instead.

What?? Yes, you read well: sand, sand, sand!!!

JFET devices perform really well (within certain designs and conditions) and they do sound nice. I love them. I have used them in phono stages as well as the lower FET of my gyrator boards.

My friend Tim kindly sugested reading this post (which I highly encourage to devour its site as it has plenty of fantastic info and experience) whilst I was experimenting with SiC MOSFET followers as driver of output stages. I played with a similar design when I was building a SLCF stage. Adding a servo is very useful to avoid cap coupling.  Despite you all purists, having a servo is ok. It sounds great, just chill out. The beauty of this circuit is the fact that there is no cap on the signal path. The cap on the servo is clearly on the feedback loop and has limited involvement as provides zero gain in AC to keep the bias point where is needed.  I still use a nice film part here anyhow.

The trick of this circuit is matching the FETs. You will have to play with LTSpice if you want to optimise this circuit depending on the FET IDSS you have. I carefully measured and traced several FETs to match pairs and get the right ones using Locky’s tracer. The FETs are NOS but still able to get hold of. Probably not for long.
The circuit is very simple. A buffer stage formed by a push pull pair of jFETs. In the post above, a nice combination of valve +pFET was used. In this case, we drop the valve, simplified the power supply and instead we use the complementary pair of 2SK170 + 2SJ74 instead. The supply can be as low as a pair of 9V batteries or a quiet supply ±12V.

The key thing on this circuit design is the matching of the jFETs (between channels)  as well as balancing the degeneration across the two jFETs (P and N types) to minimize distortion. Low degeneration will lead to higher distortion, whereas the opossite to higher impedance but lower distortion. Unfortunately, I had to manually select the FET with the tracer and then look at the THD response whilst replacing R2 and R3 with 100Ω trimmers. I manually balanced the circuit.

In my case I matched a pair of 2SK170 with IDSS of 8mA @12V as well as a pair of 2SJ74 with an IDSS of 7.6mA @12V.  Here is an example of the 2SK170:

You could argue that R1Ω could be increased to 100KΩ or more. However, you will impact the noise performance of the stage. If signal to noise ratio is high, then so be it. You may get better distortion performance overall if the previous stage distortion is very sensitive to the load (R1)

Now, let’s talk about the servo. Yay! Many will dread the excessive sand involved in this circuit. Well, I have to say that the servo circuit sounds really nice. I’m not the first one saying this (check Meno on this). Ok, have you accepted the sand on this design? Let’s move forward. The servo is a simple design. U1 is operating at open loop at DC so given its high gain it’s forcing both input pins to match their levels so to speak. In other words, the op amp will try to make the negative input as close as it can to the positive input. The positive input is grounded. Hence, the negative feedback loop via R6 will draw current down from J1 and change it bias level through R6 to keep the output at 0V (or at least as close as it can be). C1 forms a low pass filter with R7 to ensure that under AC conditions the gain of the op amp is unity so the op amp doesn’t interfere with the circuit performance. The positive input is at ground, remember. The pair of diodes limits the input levels in the servo to ensure a smoother response on transients (e.g. Power up).

The circuit (in my case given IDSS of FETs) bias at about 7-8mA. This ensure it can drive heavy loads without Slew Rate distortion issues. The measurements below are a proof of this.

Why would you bother with this circuit? Well, you may not need it and it that case forget it and don’t build it. Don’t add an extra stage to your system unless you really need it in view. Less is more.

However, if you’re looking to drive an AVC/TVC or SS amp with low input impedance or a valve amp with an input transformer, then you may want to consider this circuit.

Surprised? Well, I’m a valve fan, however I appreciate where the sand performs at it best. And we don’t have P-type valves anyhow! I’m very fond of jFETs and their sound in general. As I said before, I use them in RIAA stages as well as gyrator loads. Sand can bring an interesting twist in our hollow designs. In this case, I ended up gobsmacked with the results myself (see below)

In my view, the downsides of this circuit are:

The 2SK170 and 2SJ74 are discontinued devices.

FET’s IDSS parameters are all over the place. So be prepared to buy a handful of them and match them manually. Be patient

If you don’t match them, then the circuit will be unbalanced and higher distortion as well as an offset voltage will be at the output.

Good news is that Linear Technologies have revived these FETs (Thanks to Rick for reminding me of this!). You can use the LSK170 and LSJ74 instead. You can get them already matched by different sellers. Check the DIYaudio store for example.

Building the circuit

It took me a day to build this completely. For once, I took a day of from work and locked myself up in the workshop to get this done. A very pleasant experience as it all run smoothly as planned (at least for once!).

For simplicity and speed I used prototype PCB boards (one for each channel). Here are some pictures which may help you inspire in your version:

I firstly build a mule test circuit with a protoboard to ensure I could test the pair of jFETs and minimise the distortion. Then, I proceeded with each board build and test independently.

Initial testing of the jFET buffer

The entire circuit can be fit in a small aluminium box which you can drill easily and fast enough. The circuit also has an LED power indicator as well as a ground lift circuit comprised of a 10Ω 3W resistor in parallel with a 100nF ceramic cap and a reversed dual Schottky diodes for high current faults through ground. Given is a line stage, you may not need this and a 10Ω resistor will suffice. However, I had the set already soldered point-to-point so I used it as-is.

How well it measures?

The measures below speak for themselves. The circuit is dead quiet and has great PSRR given the nature of the follower topology. The distortion is minimal and in the case below, my new sound card has a higher H3 component and distortion level can get lower than 0.0048%-0.005%:

Well, the interesting thing is when you put the stage to drive a heave load: 4.7kΩ Most preamps will struggle. Here is the output at 1kHz and 1V, which shows that distortion has increased only 0.002% approx due to H2 raise by 8dB:

Of course at higher frequency and output levels, distortion will increase due to slew rate. A

What is also interesting to show here is the frequency response with same heavy load (4.7kΩ) and 2,200pF capacitance to drive. The output can do very well up until 2Mhz:

How does it sounds?

Very clean, as it should do. In fact, I can’t hear any change other than the improvement on response. This buffer is feeding my subwoofer amp as well as the SS main amp. I can notice a more solid and sharper bass as well as clearer treble. Subtle difference, but worth to my ears. I have to keep running with this setup for some time to get further impressions.

The jFET buffer in action, resting on top of the LME amp

It took me some time to write up this blog post as I struggled to find the time after building this piece of equipment. Hope you enjoy it (at least reading it)

Using the Gyrator PCB board:

Just playing with the layout a bit. The gyrator boards, the UV4 sockets and the Rod Coleman regulators. All in a tiny aluminium box:

This is what I managed to do with a couple of hours at the workshop. Time is gold for me these days. Quick drilling and fitting the main components. You can’t get a smaller preamp than this one. It’s quite packed:

Next, soldering. Yay, just looking forward to my favourite part of the build process

Adding filament bias and output caps

Just a quote from a friend who inspired the cartoon below

“With all that is happening in the world, I’m just liking being able to come home and put a record on. There’s something honest about it”. – Tom Browne

First listening test with these beauties, it sounds really nice!

I usually don’t write about music, mainly because this blog isn’t about my music preference and I respect that people may have different tastes.  Either way, I couldnt’ resist in posting about this great compilation of Led’s BBC sessions. If you don’t like them, avoid reading this. If you do, it’s worth exploring this 3 CD compilation. 

It’s rock at its best in my view. Very powerful, specially outside the protection remit of the recording studio which exposes the essence of any band in my view.

Don’t expect any hi-if recording here, just pure magic and music, that’s all about

Enjoy the power of Zeppelin!

Happy Thanksgiving. 

Ale

Introduction

The main challenge when implementing valve amplifiers using transmitting valves or valves which require a significant voltage swing (e.g. 300B, 45, etc.) is the driver. Getting the driver right is not easy. You’re asking for a single stage to swing 150 to 200Vpp at minimum distortion. There are some ways you can achieve this:

1. Implementing 2 stage voltage amplification. Here is where we find a lot of bad designs and poor results. Sometimes the 300B gets a bad reputation due to a wimpy or poor driver. Many designs out there use 2 stages of 6SN7 for example. Nothing wrong about using the 6SN7, however when you cascade 2 stages the sound is muddled at low levels. Harmonic profiles may be encouraging but they simply don’t sound great.
2. Implementing a high-mu driver stage. There are several high-mu drivers out there than can swing plenty of volts. 6Э5П, 6Э6П, 6j52P, 6j49p-DR, E280F, C3g, etc. They work well, specially if you couple them with a gyrator, you can achieve hi gain. If you opt for degenerating the cathode resistor, the gyrator still provides a low output impedance to avoid degrading it due to the degeneration resistor. I’m a big fan of this approach. The only disadvantage is that you need a buffer/line-stage capable of driving the Miller capacitance. I have a nice preamp/line stage so this isn’t a problem to me.
3. Implementing a pentode driver. Pentode don’t suffer from Miller capacitance. However, you need to find the right driver, not all sound well in my experience. I like the 4P1L and C3g. You can use a gyrator load with pentodes as well. Some folks complain about the pentode harmonic signature. I think this is a question of personal taste. 
4. Implementing a shunt cascode driver. Hey, this is what this post is about! There are several benefits already discussed at length on this topology.  If you need high gain and minimum capacitance load (e.g. Miller) as you have a DAC output for example, this is what you should look into. The Shunt Cascode operates the triode in a vertical load line (not horizontal like the CCS or gyrator).

Design

You should start by reading this extensive blog post. That will provide you with a lot of information around the shunt cascode and how it works. Back in 2013 I started playing with the 6Э5П in this topology. It was quite promising. Now, I have revisited and built this driver to see how it really performed.

The design is very similar to what we discussed back then. I shall proceed in describing the circuit, in particular the changes made. The driver is still the marvellous 6Э5П. There are few valves out there that I don’t like as much as I do with the 6Э5П. I measured the curves long time ago when I started with the curve tracer project. I also tested the 6Э5П and 6Э6П extensively. I do love the 6Э6П as well, it’s one of my favourite drivers.

The 6Э5П is biased at about 200V/30mA with a degeneration cathode resistor of 120Ω. As the gain of this stage isn’t dependent on the μ of the valve, then is good to do this to improve the linearity of the driver. M2 forms a CCS with Rmu. It provides the current to the 6Э5П as well as the current to the common base stage formed by Q1 and Q2. The gain of this stage is gm times R5. The gm is the valve’s transconductance The collector current of the MPSA92 is kept low to ensure distortion is minimised as well as its operated under SOA. D3 provides a protection to the darlington pair when is reversed biased. 

The gain of this stage was measured to be x140 (or 43dB). That equals to a degenerated transconductance of 5mA/V with a cathode resistor of 120Ω and a gain resistor for 27kΩ. 

There are a few considerations to bear in mind when you set yourself to implement this circuit. This is a high-gain stage, so you need short and tight connections. Ferrite beads as well as stopper resistors. Otherwise this thing will oscillate like in a rave party.  You can see the build pictures for reference further down.

As you can see there is a delicate balance between the collector current and the output bias point to ensure maximum swing at large output levels. The stage was designed to

The cascode reference voltage is crucial. You ought to get this as stable as possible. The preferred option is a shunt regulator. The current consumption is minimal so it’s easy to implement. 

I opted for a quick and dirty approach to test the circuit using components I had at hand. This isn’t a regulator, it’s a voltage reference without feedback or regulation.

Let me explain briefly how it works. The reference voltage is provided through a resistor formed by R9 and R14. The CCS formed by M1 and RSET provides a stable current to generate the reference voltage. This voltage is fed into a PNP emitter follower (Q4) which has a cascoded configuration with Q5 to avoid stressing the follower. Q5 is fed from a 100V reference formed by R14. The emitter follower has a CCS on the tail (instead of a resistor) formed by Q3 and R12. R12 sets the CCS current which is about 4mA. Rset sets the reference voltage for the shunt cascode stage. It provides a stable voltage reference with minimum ripple thanks to the filtering of C3 and the capacitance multiplier effect of Q4. PSRR is about 100dB according to the simulation. 

If you want a more stable and sophisticated approach, you should opt for a shunt regulator or a series feedback regulator as well. 

The stability of this stage is dependent on both the CCS as well as the reference voltage on the shunt darlington pair. 

Build

Again, I was lucky enough to get 3 hours free of work on Friday and build this circuit on a prototype PCB. Here is how it looks like:

As said earlier, with a high-gain circuit like this, it’s imperative that connections are short, layout is well planned. I added ferrite beads to grids, anodes and collectors. This helps preventing any unwanted oscillations. 

If your target gain is considerable, then avoid point to point wiring. It will be a recipe for failure. Layout of this stage is critical, so a PCB is recommended.

Testing

It’s always a real challenge to test a high gain stage breadboard. Building this cascode stage isn’t straight forward and I wouldn’t recommend it to any beginner. Avoid this if you don’t know really what you’re doing.

You will have to set the reference voltage to 200V and then proceed to trim the CCS current whilst looking at the oscilloscope. Drive the stage hard to make sure you’re not clipping the output. Given the high gain, it’s crucial to set the collector voltage to the right level to avoid clipping when driving it at 150 or 200Vpp. 

Here’s a taste of the stage at 55V/155Vpp. Distortion is really low. The artefacts between 15kHz and 20kHz are from my bench. Probably the LED lighting system which is amplified greatly with the high-gain stage:

Funny enough, comparing results with Rod Coleman, it seems we both arrived to same distortion performance from different approaches to the same stage. I like that.

Should I go down this route?

In my opinion, unless you’re not looking to drive your amp with a single stage without having any buffer, then probably don’t go this route. The driver is more complex than other ones and requires some skills to build this (and get it stabilised as well). If you are flexible to accommodate a line-stage / buffer then Miller isn’t a problem to you. In that case, you can opt for some other alternatives.

Here is what I have tried myself with success.

My first and personally preferred option is the lovely 6j49p-DR.  

Look at the performance of this valve at 200Vpp:

Amazing, and it sounds brilliant too. I have recommended it to a friend who used it as fist stage of a guitar amp and he was delighted with. 

Secondly, albeit with lower gain, the triode-strapped 12HL7. This valve has generated a lot of noise in the DIYAudio community, and its reputation is well justified. Again, the performance at 200Vpp is better than the shunt cascode with a simple gyrator load:
One of my other favourite drivers it the brother of the 6Э5П. The 6Э6П has a nice H2 profile when driven at a full tilt:

As you can see, carefully chosen triode-strapped pentodes can do this job. You need to consider the Miller capacitance though, so the ability to drive the stage is crucial to avoid HF rolloff. 

What’s next?

I was abusing the stage until I accidentally touched the DN2540 tab with a crocodile clip and fried the FET. Well, shit happens as they say. I need to replace the CCS MOSFET before I can do further work. 

Surely this driver is on my list of test with the 300B SE amp.

Hope you enjoyed this post.

cheers

ale

Introduction

Did you ever dream about having a nice valve headphone amp? I did. Several times in fact. I’m not keen on the OTL designs and yet, owning a pair of Grado’s SR80i 32Ω headphones, I’ve been looking at different topologies and designs for headphone stages. 

Building a good headphone amp is a real challenge. The low impedance of most of the headphones out there (with exception of 600Ω ones), makes the task a very hard one for any valve. A simple stage with a transformer is very attractive. I found this transformer from Sowter specially designed for headphones which attracted me at fist sight. Having a mu-metal core provides a great frequency response. It can easily achieve 30kHz bandwidth. Having a mu-metal core provides a low distortion which is very attractive. The transformer isn’t cheap but you can’t cut costs on this if you’re looking to implement a high-quality amp for your headphones. Even 600Ω are not easy to drive by valve circuits such as White Cathode Followers and SRPPs. There are plenty of circuits out there (some not good at all) but I decided this time to explore and experiment a bit (as usual).

Pushing the gyrator

I wanted to keep the design very simple. I could have gone down the route of a simple transformer loaded, but transformers which can have similar wide bandwidth, gapped for DC and have amorphous core, are more expensive. A classical approach is a choke loaded parafeed design. Again, high quality chokes are expensive as well. You could go with an CCS load, but you want to take advantage of a mu-follower setup so you can provide low output impedance and minimise impact of a lowish (e.g. 4KΩ) load to the driving valve. 

A gyrator load has proven to be outstanding. Both in terms of performance as well as sound. Rather than a CCS, I opted for my beloved gyrator.

So, how can we build a single stage (a.k.a “spud”) which can drive these headphones in parafeed mode?. there are several high-mu triodes (or triode-strapped pentodes) which can do the job here. 

Parafeed cap

The resistance of the windings reflected to the primary is about 576Ω. If the load is 32Ω, the transformer is configured in 12:1, so the total load presented to the stage is about 5,184Ω. The output impedance of the gyrator is 1/gm. Despite expected gm is about 40mS for VDS=8V as per the BF862 data sheet, I’d expect it to be not even half of that at low VDS. So If we consider a conservative 20mS, the output impedance of the gyrator stage is around 50Ω. So for a pole at about 10Hz, we need the capacitor to be about 3uF. A 2uF should work well and provide the -3dB point about 12Hz. On my tests I used a 1uF Oil cap I had at hand. I would invest in a good quality film or oil cap. A +300Vdc part is needed and the higher the better. 

If you are using high-efficiency headphones (e.g. SPL>90dB) you don’t need much gain on this stage so the spud design is easily achievable with a wider range of triodes (or triode-strapped pentodes). I found that my headphones are really loud with 70-80 mV. If you consider the peaks and headroom needed, you need to design for 150 to 200mV output at low distortion. A gain of about x30 will give full tilt with an input of 500mV. Plenty enough. You can easily go with a low-mu DHT (e.g. 4P1L) and have enough to drive this with a 2V source. 

I tried several valves here, and here is my top 3:

1. 6Z49P-DR: amongst my favourite Russian pentode. Triode strapped can do x45-x50 gain. it’s extremely linear and sounds amazing in my experience. 
2. 12HL7: It has a reputation and recently highlighted in the DIY Audio community. With lower gain, it is very linear. Needs some more current than the 6Z49P though
3. 6e5P/6e6P: my favourite ones. Both are great valves and have mid-mu (30-ish) so will work fine here.
4. 6J52P: I love it but the parameters are all over the place. You won’t need this much gain.

After several tests on the bench I settled for the 6Z49P-DR and the 12HL7. I prefer the 6Z49-DR though. The cathode bias is with SiC diodes. You can do what you like here. I even try a degen resistor of about 82Ω. Didn’t see much improvement in distortion and you trade off some gain. 

On the workbench I use the fix bias supply of the curve tracer to find the sweet spot and then I change it to the cathode bias arrangement I want. That’s the way generally I try different things. 

How does it measure?

Here is a sample of the great performance. The 12HL7 is a tad more linear, but not much. See how low the distortion is at ridiculous output levels. 2V represents 125mW which will leave you deaf:

At peak (e.g. input 1V and output of 4V (or 500mW) which is the maximum the transformer can handle, the distortion is just 0.32%

How does it sound?

I like the sound of my Grado’s. I played some of Trane’s “Blue Train”, and the powerful bass notes of the intro are very detailed. I generally test with horn playing as take me back to my musician days when I played the sax. I can recognise subtle differences in the reed’s timbre and I have to say the sound of this headphone amp is delightful. Of course, it will rely heavily on the headphone’s quality.

Next steps

Build it on a nice case, but I haven’t got the time. Hopefully soon…