Ever since people started playing with electronic equipment in the early 20th century, it’s been known that electronic equipment runs on smoke. If the smoke escapes, the equipment stops working.
Sometimes, the manner of the smoke escaping is discreet and quiet, other times it escapes forcefully.
Whilst it is said that any machine can be a smoke machine if you operate it wrong enough, it is none the less desirable for machines to refrain from letting their magic smoke out (either explosively or otherwise) in normal operation.
Clearly this principle has resonated with other manufacturers, as a quick look on Aliexpress reveals among the list of features for this equipment
Not Exploding is an attribute we look for in electronic equipment. This manufacturer seems proud to advertise the fact
Accordingly we introduce the DNE certification mark: DNE for Does Not Explode.
Henceforth all our amplifiers will be carrying this mark.
The other day I got an email from Spotify which very neatly illustrated the problems that classical music present to streaming platforms or personal media servers:
Last I knew, Chopin has been dead for at least 170 years. Very clever of him to be releasing fresh tracks while decomposing.
In Classical music we tend to focus more on “Works” not Tracks.
A Work is created by a composer. It may consist of one or more tracks. A work may stand on its own, or be part of a collection.
A work is different from an Album – albums are not really a thing in classical music. Back in the days when CDs were king, one CD could contain more than one work, if they were short. Usually works for such compilation CDs were selected with some theme in common, such as same performer, same composer or similar.
Conversely, one long work could also span several CDs, such as the Verdi Requiem, for example.
Non-classical is easy. For example, there is only one “No Jacket Required” album and it’s by Phil Collins. Very easy to categorise and search. Conversely for any one particular classical Work there are usually multiple recordings available, and for the more popular works like Vivaldi’s Four Seasons or Beethoven’s 5th symphony, there are hundreds of different recordings.
Each work is of a particular type: Symphony, choral symphony, opera, concerto, to name a few.
Each different recording of a work has different performers. Typically the credits will include the conductor, the orchestra (if it’s an orchestral work), the soloist (if it’s a concerto) and the vocalist(s) if it’s operatic or choral, and all the performers if it’s chamber music or a solo piece (like the Bach cello suites). Also the venue in which it was recorded.
On top of that each work has a genre – naive music libraries tend to lump them all into “classical” but as with all types of music, there are many sub and sub-sub genres.
Most classical music – especially symphonies and concertos – contain individual movements (aka tracks) which have the same name as tracks in other works, because the naming convention tends to be the italian name for the tempo of the track. (“Presto”, “Allegro Assai” etc) – so the individual track name is meaningless without knowing which work it belongs to, and moreover which performance of which work.
From all this it should be clear that putting together a well organised classical music library is a somewhat more complex undertaking than simply putting a bunch of folders together, ripping a bunch of CDs and putting each CD into its own folder, and hoping for the best!
These concerns are all equally valid for streaming services like Spotify and Tidal and others. Clearly in the example above, someone has released a new recording of a Chopin work, and the “album artist” tag is set to “Frederic Chopin” – hence the email.
Clever Chopin, dead for 170 years, buried in Paris and his heart buried in Poland, yet still churning out the hits!
With non-classical music it’s way easier. Enter “Phil Collins” into the search, and you get a list of all Phil Collins’ albums (sorted by year if you’re lucky!)
Classical? Not so much.
Questions I might want to ask my classical music library:
Show me all the Beethoven symphonies in my library (I’m querying Works by composer and Work Type) – I expect to see a list of works that meet the criteria, with information about each work that makes it unique (eg year of recording, conductor/orchestra, venue, etc). Clicking on each work should then explode it into its individual tracks/movements (and these need to be in the right order too, shuffle is not a thing for classical!)
Or I could perhaps have been taken with a particular singer’s voice. Carolyn Sampson did a magical recording of the Bach secular cantatas and perhaps I might want to see what else she’s done. This is a search by performer and once again I want a list of works, this time where she’s credited as the singer.
You can think up any type of query based on work, composer, work type, genre, performer, collection etc, and you should be able to find it easily. The key here is that the defining classification in all these cases is the work, not the album. Yet the metadata tags for music files don’t even contain a “work” field!
Perhaps this gives a clue as to why classical music tends to confound streaming services and bedevil people who curate their own media collection. My media server software (Plex) doesn’t stand a ghost of a chance against this complexity. It’s firmly rooted in the “one artist per album… list the albums per artist” philosophy. It doesn’t even support displaying or searching by Composer, and has no clue what a “work” is.
Where to from here? Perhaps Roon might be worth a second look… I had a trial, but was discouraged by the cost.
What do other classical music fans do, who want to manage their library electronically, I wonder?
This is a project that was born from a visit to a fellow amp-building enthusiast. There was a small output transformer sitting in a pile of wires and upon closer inspection it was connected to a breadboarded pushpull stage, with the driver stage sitting across the room connected by a long cable.
The size of the output transformer and the diminutive size of the tubes struck me, particularly because they were at odds with the full, rich sound I was hearing through the Klipsch speaker.
The driver tubes were EL95 and the output transformers were custom-wound, 9.1K Ra-a with 20% ultralinear taps.
I quickly acquired a set of EL95s and a pair of the output transformers. Then I returned home and sat on them for a while before deciding on a circuit topology. I opted for a preamp stage based around the 6N1 tube since I had plenty in stock. So I designed a simple topology based around a gain stage followed by a long-tail pair phase splitter. I calculated this would give me around 15dB more open-loop gain than I needed to get full output at 1v P-P input. Thus allowing 15dB global NFB, which is a good amount for a push-pull design.
(click to enlarge)
The blue numbers in the schematic represent actual measured values in the built circuit.
The next step after designing the circuit was to bread-board the initial stages, and fine-tune the resistor values with experimentation. Note especially the Long-Tail Pair, the first half has 14K load resistor while the second half has 15K. This was calculated using Blencowe’s method to provide the best balance, which was validated by measurement. To make 14K I parallelled a 39K and a 22K resistor.
After breadboard validation, it was time to lay out the PCB. I had one very specific consideration for this amplifier: I wanted to fit it into a small form-factor size case, the same size as my previous 3-ch preamp project, in fact. This necessitated a limit on the PCB size, so I opted for 160 X 100mm. Therein lay the challenge: To fit the whole amp, power supply included, onto this size board.
In the end, I was successful, and after review I sent the board out to be manufactured. When it came back, I built it:
After building, it was time for some testing. This was accomplished using a 8R resistive dummyload and Room EQ Wizard, using a 100R : 20R voltage divider on the speaker output to allow accurate measurements at 1W and at full rated power (7W)
This shows freq response at 1W and full power, noise floor at 1W and full power, and THD again at 1W and Full Power. (Noise Floor levels as references to output level). (This was with the amp totally unshielded sitting on the bench with long flying wires everywhere, so I expect the noise floor to improve when it’s built into a chassis)
Now, it’s just a case of putting it into a chassis. The chassis design has been completed – the CNC pattern files in QCad and the PDF of the laser engraving in LibreOffice Draw. Once the chassis is back from the sheetmetal workshop, I’ll complete the assembly. Stay Tuned…
Largely owing to convenience of supply, I’ve made my last few builds with Russian triodes instead of Western ones. These are in plentiful supply as NOS from the 70s and 80s, and available cheaply on eBay and other sources. Typically these are military-spec and quite consistent in quality, so they are fantastic for audio.
Some of them are pretty much direct equivalents (eg. 6N2 for 12AX7, 6N6 for ECC99) while others are close but not quite the same, meaning any circuit would need to be recalculated if subbing in a Russian tube.
This is relevant because most tube audio schematics I’ve found on the internet are specified with Western types, some of which have no direct equivalent.
So to bring several sources of information together in one handy reference, I made a spreadsheet which lists the different tube types and their vital parameters.
Red are Russian types, blue are Western. Click to see larger
To make things easier I made it into a table so you can sort and filter it in Excel. You can download the original file here
One thing worth remembering. The pinouts are different, specifically, for the filaments (heaters) so you can’t just drop a Russian tube in place of an equivalent as shown in the table above. Some re-configuring would be necessary.
First up – the review. Around Feb 2019 I was contacted by Richard Varey, from Witchdoctor magazine – about supplying an amplifier for review. He’d see the old ATRAD-Audio website and wanted to know more.
During the month he had it, a buyer expressed interest in taking it after the review was finished. So it was sold to him and he is reportedly happy with it.
The end of ATRAD Audio(well, not quite, but a bit of a change) Shortly after this point, it was brought to my attention that my operation looked too much like a commercial endeavour, and taking commissioned builds meant I’d need to comply with various regulations both for the product and my own qualifications. Additionally, to comply with these electrical regulations, I’d need to sumbit a sample to a laboratory for expensive destructive testing.
Clearly this was out of reach for a hobbyist so I decided reluctantly to cease operation. The ATRAD Audio site was repurposed into what you see now.
However this is not quite the end. It turns out if I build things for myself and then sell them as used, the restrictions don’t apply, because when buying a used appliance, all risk passes to the buyer.
So this is how I will continue.
Which is the way most vacuum tube amp hobbyists run anyway.
Since I started building amplifiers, I’ve been reliant on one supplier in particular, for my chassis and casework.
Most suppliers of parts can be replaced – however when you’re sourcing outwork services you enter into a relationship with your supplier that goes beyond a straight customer / supplier dynamic.
Introducing Embrace Design who have provided all the CNC and laser cutting and engraving services for the cases for my amplifiers. The service doesn’t stop there however. Henri has been a rich source of suggestions and ideas and has been completely cooperative in working with me to get exactly the look and aesthetics I have been looking for, as well as a good emergency weekend source of fasteners and other necessities. Much of the exterior appearance of my amplifiers is down to Henri’s skill with translating the designs into reality. Worth acknowledging here.
… in which I make myself unwelcome at a high-end hi-fi store. This is a personal experience which happened over a decade ago, but I decided to put it here because it’s still relevant. These are the two products in question
Audio Alchemy Digital Transmission Interface, US$ 1600
Generic 10/100 8-port auto-sensing ethernet hub, around US$ 40 – 60 (at the time) depending on brand.
On the day in question, I walked into a hi-fi shop and the proprietor greeted me with a wide smile. This was a real high-end place where the demo systems are set up with speaker cables as thick as fire hoses, and just the rack that the system is sitting on has a five-figure price ticket. We got talking (this was years before I started designing and building gear myself, I was just looking at buying a new pair of speakers which is what drew me into the shop). Pretty soon he had figured out my system and had decided that it would be improved by the addition of the top product, above. While extolling its many and varied virtues, he inadvertently tripped himself up by completely inaccurately describing the phenomenon of jitter … seems he hadn’t read up enough on the technical manual that accompanied the product. After a little while, I attempted to summarise my understanding of this product back to him, to show him I’d been listening. He was all ready to sell me one until I started asking some questions which led me to explain the function of the second pictured device…
Conversation went something like this: “So, this machine will take a PCM data stream at 1.4 megabits per second (Red Book CD standard), store and buffer it, then output an identical data stream according to its own internal clock, which is carefully designed to be high-accuracy and not susceptible to disruption, correct?”
“Yes, and [long spiel about how that makes it sound better, yada yada]”
“And it’s $2200.” [That was the $NZ price at the time]
“Yes, possibly the best value enhancement you can make to a digital system…..”
“OK, so what then would you say about a device that does this at around 70 times the data rate, and not for one but eight separate inputs?”
“Well the DTI represents the cutting edge of digital transmission design, so perhaps in the future something might be designed that could do what you say, but it would be a very high-end piece of equipment, so only the most serious audiophiles would require it, and that’s assuming there’d ever be a digital recording standard that would utilise such a bit rate”
“So you’re saying it’d be expensive then?”
“Something with over 500 times the processing capacity of this machine? Very!”
At which point I explained the functionality of the 10/100 ethernet hub, and then its price. He wasn’t smiling any more.
In fact, I got the distinct sense I’d outstayed my welcome in that establishment.
– This is why I don’t go into high-end hi-fi stores any more. –
The push-pull design for output stages has persisted since times of antiquity. It was one of the very earliest circuit designs, and has persisted until the present day, with modern solid-state linear amplifiers still overwhelmingly using it.
With tubes, a typical topology is given by the circuit below. (Click to magnify). The anode (plate) voltage on either side comes through the primary of the output transformer. This design uses a pentode tube, which has a screen grid. This is attached to taps on the output transformer to run in Ultra-Linear mode, increasing efficiency and reducing distortion.
In this design, the cathodes are tied to ground through a very low value shunt resistor. The resistor is simply there to provide a small voltage drop from which the current through the tube can be measured. It plays no other role in the circuit, other than being a fuse if the tube red-plates.
Values of one ohm or ten ohms are typical of this arrangement.
Typical implementation of a Fixed-bias, ultralinear Push-Pull output stage with a pair of pentode tubes (EL84 / 6BQ5)
Because the cathode is at (or very close to) ground potential, this requires the control grid to have a negative DC bias voltage applied to it, to regulate the flow of current through the tube.
If there was no negative bias applied to the tube, it would go into full conduction, the plate would glow red hot, fireworks will happen and that would be Bad, mmkay? So we need to contrive to feed a constant negative voltage into the grid, along with the signal, to achieve the desired regulation.
The voltage required depends on several factors… as a very rough rule-of-thumb, take the screen voltage and divide it by the tube’s mu (gain) to get the maximum negative bias voltage likely to be needed
Looking at the circuit above… An EL84 has a mu of 20 and in this implementation the screen voltage is 350 (in Ultra linear we take the screen voltage as the same as the anode/plate voltage), which gives us 17.5 volts. Multiply by -1 since we’re dealing with negative volts. So we’re likely to need around -17.5 volts.In this case, our adjustment range is from -12.5 to -21.5 volts.
How to set up the bias adjustment resistor values The bias voltage needs to be adjustable. Both tubes need to be drawing the same current, otherwise the net current through the transformer will not be zero, which will lead to magnetisation of the transformer core. This is a most undesirable situation and left unchecked, it will cause quantum fluctuations in the space-time continuum. Well ok maybe not that bad, but the transformer will saturate unevenly and distort the sound.
The usual approach is to use a voltage divider network with a potentiometer, as above. Couple of points about this design.
The more negative the voltage goes, the lower the current through the tube
In this implementation, if the potentiometer fails, it will fail safe. The most common mode of failure with potentiometers is the wiper lifting off the track. If this happens, effectively the voltage at the grid of the tube will go full negative, reducing the current through the tube to (almost) zero. This is far more desirable than the voltage reaching zero and the tube immediately red-plating.
Expanding on (2) – please don’t ever build this circuit with just the pot wiper connected to the grid. When the pot fails (and it will, eventually) it’ll likely take the tube with it.
If you’re going to build this circuit, it’s intuitive to set it up so that clockwise rotation of the pot increases the current through the tube (ie brings the biasing voltage closer to zero)
The next question is – what value resistors will be needed? This is where some trial and error in the calculations is needed. Using Ohms’ law, these are the variables:
The desired bias voltage adjustment range
The input negative voltage from the power supply
From there, you can calculate the values for the resistors and potentiometer to give you the range you need.
Put in the numbers in the red. Experiment with the values for R1 and R3 and the potentiometer, until you get the desired voltage range in the “Output V” column.
The “Build-Out R” represents the load seen by the preceding driver or phase-splitter stage, so watch the maximum “Rg-k” from the tube’s datasheet isn’t exceeded. In the case of an EL84, that value is 300K. (In the datasheet there will be a specification for this, and it’s diffrent depending on the mode of use of the tube. So look for the “two tubes, class AB” which is in most pentode/tetrode datasheets)
The columns of this spreadsheet:
Step – the setting on the potentiometer
Total R – the total resistance from the bias voltage to ground
Output V – the negative voltage as fed to the control grid
P(R1) – the amount of power dissipated by R1
P(R3) – power dissipated by R3
P(Pot) – the power dissipated by the potentiometer
Pot pwr % – the power dissipated by the pot expressed as a percentage of the pot’s total power rating AND the amount of track being used to conduct.
About Pot pwr % In the specifications for the potentiometer, there will be a power rating. However that power rating is across the entire length of the track. If the pot is set to half-position (assuming it’s a linear taper which in this design it is) then the power handling drops to half. So, this Pot pwr % column shows how much power the pot is dissipating as a percentage of its maximum taking the wiper position into consideration.
Using the circuit Putting all this together, it’s easy to see how it works. Adjust the potentiometer for maximum negative voltage (wiper closest to the left, in this schematic). Power on the circuit and let it stabilize. Measure the voltage across the cathode resistor. Then adjust the potentiometer until the desired current is flowing through the tube.
What is the desired current through the tube? Glad you asked. This depends on the tube itself, and the B+ voltage, and your preference regarding bias.
As a rule of thumb, biasing at around 70% of maximum is the sweet spot with most types when using this conifguration. If you bias low, that’s called “cold” biasing. It means the current through the tube will be low, and the sound may take on a thin, glassy, brittle aesthetic, and the distortion will increase. Bias too hot and you’ll shorten the life of the tubes for no real benefit. 70% is the goldilocks zone. So. Look at the tube’s datasheet. For an EL84 we see the maximum plate dissipation is 12 watts.
70% of 12 watts is 8.4 watts
B+ is 350 Volts
So, using Ohms Law:
we solve for I at 24mA
Across 10 ohms (cathode resistor) our 24mA will give 0.24V
So we want to see 0.24V across the cathode resistor. Adjust the potentiometer until that’s the value shown. Then repeat for the other tube in the circuit, then do a final check that they’re both the same (or as close as you can get)
Please feel free to use my spreadsheet – I developed it to assist in choosing the resistor network values, and also to ensure the power rating of the potentiometer wasn’t being inadvertently exceeded at any setting.
The amplifiers I’ve built so far have all incorporated a delayed turn-on circuit for the high voltage supply. The intention is to allow the 6 volt supply to turn on first and allow the valves to reach operating temperature before turning on the high voltage supply.
This is accomplished with a simple circuit based around a 555 timer IC in monostable mode, set up to a delay of around 25 seconds.
The circuit I’ve been using, while functioning, had a few problems. Driving a relay directly from the output of a 555 IC resulted in a lot of voltage drop through the IC and the relay coil voltage being low, for a 5V relay it was getting around 3.5 volts, fortunately this is still enough to trigger it, but less than ideal.
My re-design of the circuit was prompted by my addition of a 2-colour LED to the design, to glow red at initial turn-on but change to green when the timer activates and the HT voltage is turned on.
These LEDs are 2-pin, they work by reversing the polarity into them. So they’re 2 LEDs in one envelope, and depending on the polarity of the applied voltage, one will be forward biased and glowing, the other reverse biased and dark.
After breadboarding it and measuring carefully, this is the circuit I designed:
Note in this diagram my symbol library for the MOSFET is wrong… if you’re gonna use this same MOSFET be very aware its pinout (viewed from top) is S-G-D instead of G-D-S. So my pin numbers are wrong. Sorry about that.
The 7805 voltage regulator is not strictly necessary but it does result in a nice 4.9V across the relay coil.
The 330K and 68µF cap provide the time constant for the timer IC. The formula in this mode is:
T = 1.1 x R x C
The MOSFET Q1 buffers the output of the IC switching the negative on or off to the relay based on the voltage at the gate, which comes from the output of the IC at pin 3. This starts low until 25sec elapses then goes high and stays high until power down.
The two 330R resistors form a voltage divider, at the mid-point the voltage is 2.5V. When the relay is off, the + voltage will flow through the coil (which is around 62 Ohms) and then into the LED, then to ground through the lower 330R resistor. This results in a voltage drop of around 0.2 volts across the relay coil, not enough to turn it on.
When the IC turns on, the voltage appears at the gate of the MOSFET, switching the transistor on. This effectively shorts the Drain and Source, causing the negative to connect to the relay and the LED. At which point the return path for the LED is through the top 330R resistor, so this reverses the polarity across the LED causing it to change colour.
The reverse-biased diode across the relay is for flyback suppression.
After breadboarding, I’ve designed a single-layer PCB layout for this circuit which is 35mm x 35mm utilising a W02 rectifier.
On my board design I’ve also added a header for a regulated 5V power supply, in case it’s needed elsewhere (such as a tone control bypass relay for example).
The current and dissipation is such that no heatsinks are necessary on either the voltage regulator or MOSFET.
Be sure to put the relay on the AC side of the rectifier diodes, relays have a much easier time switching AC than DC and this is reflected in the voltage rating on the datasheet.
The EL84 amp is completed and has been removed from the workbench and is now in the living room where it’s been entertaining us the past few days.
First, a few pretty pictures
This is the best looking amp I’ve made so far. Great care was taken with centering and spacing. The translucent hole to the left of the volume control covers the IR detector for the remote control.
About the name
The amplifier is named “Matariki” which is in the Maori language of New Zealand. Literally translated, it means either “Eyes of God” or alternatively “Little Eyes”.
In more common usage, it is the name given to the Pleiades star cluster, when it becomes visible (which is mid-year, mid winter here) and has traditionally become associated with renewal, the Europeans decided to call it the “Maori New Year”
There was also a rare southern right whale which made an unusual appearance in Wellington Harbour recently, during Matariki, and the whale was thus informally named Matariki.While all this was happening, I was designing this amplifier. Hence the name
The case is aluminium, sourced from AliExpress, of the type I usually use. The front panel is 8mm thick, brushed aluminium. It required pockets being milled on the CNC from the back to accommodate the controls mounted through it.
The lights are 3mm LEDs but I decided I don’t like the bulging appearance they give when pushed through the front panel, so we laser-cut some 2mm clear acrylic into 3mm circles, so the lights on the front could be flat and flush. They press-fitted perfectly and the look was 100% what I was wanting.
The STBY LED is red, and the PWR led is dual-colour, it starts red at power-up and then when the HT switches on after 30 sec, it turns green.
The power switch and input selector are a rotary encoder: push to toggle power, rotate to cycle through the inputs.
Inside the case
Inside the chassis there’s the amplifier mainboard, which contains 6 tubes and is the phono, tone, gain and phase splitter. To the left of that is the base of the output valves, the long thin board contains the bias and cathode shunt resistors, test points, and on the track side, the four trimmers for adjusting the bias voltage.
The green boards are bought-in components: input selector, mains switch/standby, remote volume, and microcontroller.
The power supply contains the usual array of resistors and capacitors needed to provide the various voltages, as well as the usual 30 sec startup delay timer relay circuit I always use.
The DC voltages provided by the power supply are:
+320V +300V +270V +250V +6.0V (DC heaters for Phono stage, rectified from the 5vac secondary with Schottky diodes) -27V for fixed bias
In addition there’s the standby transformer which provides 9vac at around 200mA to power the microcontroller and standby board.
This was the first all-on-one-board amp I’d made and it was successful. Everything worked exactly as expected on the first power-up. All the components fitted on the board, the board itself was a success (first project with the new temperature-controlled PCB etching tank) and the board looks fine (although there’s no soldermask or silkscreen on it, it really is just single-sided naked copper tracks on FR4)
Likewise for the power supply.
The level of tidiness inside the case is better than anything I’ve achieved before, although I don’t think I’ll ever get to the level I am looking for… which is OK, because when you shoot for the moon you’re not gonna hit it, but you will end up in the treetops, which is a whole lot better than being on the ground.
The level of aesthetic appeal on this one is better than any of my previous projects as well. I am completely happy with that aspect.
From a technical standpoint, on this build I’d designed the board to allow phase compensation into the NFB loop. This is because NFB produces high-frequency ringing which you can see on the oscilloscope if you put a 10KHz squarewave into the input. At the output you get something like this:
Nasty ringing through the NFB
The prescribed method to resolve this is to phase-compensate the NFB with resistors and capacitors, the values of which are determined by experimentation. After doing this, the 10KHz squarewave output now looks like this:
NFB after phase compensation added
Finally, one aspect I am well pleased with is the listening test. Subjectively this is the cleanest sounding amp I have built to date.
Schematic as built
Click to enlarge. Might need to download / Save-As, to be able to read it
Editorial Mar 2019: This is a schematic for the second Matariki I built, with a few improvements over the first. I felt it prudent to replace the old schematic with this one.
