Concertino tube amplifier

Details: Written by: Marcin Sławicz; Category: Tube equipment; Last Updated: 15 December 2025; Hits: 979

Photographs and description - Marcin Sławicz (2005)

The beginnings of the project

The idea of building my own tube amplifier has been nagging at me for the past two years. I'm not a die-hard audiophile, and using "ordinary" solid-state equipment was perfectly adequate for me (I've always preferred listening to music, not equipment). Now, however, my worn-out amplifier is beginning to show signs of age, and while I could refurbish it, a perfect opportunity presents itself to pursue a tube-based endeavor.

Initially, I considered a design based solely on triodes, but rejected the inherent disadvantages of SE circuits. I found a very interesting description of a push-pull amplifier with directly heated triodes on Lynn Olson's website. It's worth checking out for the incredibly interesting solutions used in his designs. However, these amplifiers have a major drawback – cost (mainly due to the use of 300B or 2A3 tubes and interstage transformers). So I had to look further.

My attention was drawn to the indirectly heated 6AS7 dual power triodes, once used primarily in power supplies but also performing admirably as output stage tubes in audio amplifiers. The tubes would have been significantly cheaper, but their low voltage gain would have required expensive and difficult-to-find interstage transformers or two or more triodes connected in parallel. Russ Sadd has described a push-pull amplifier with 6AS7 triodes on his website.

The project languished for several more months, during which I slowly realized that a successful power amplifier didn't necessarily need triodes in the output stage. I began to consider using beam tetrodes in an ultralinear configuration. This configuration combined the sonic advantages of triodes (low distortion) with the high efficiency and stability of tetrodes and pentodes. My choices included the 6L6/5881, KT66, and KT88/6550 tubes, commonly used in both guitar amplifiers and hi-fi designs.

The next step is to search the web to select a basic amplifier circuit. The amplifier shouldn't be complex, as an overly complex circuit doesn't guarantee high sound quality, and with limited measurement capabilities, it can make commissioning unnecessarily difficult. Serial designs must ensure repeatable production and relatively consistent parameters during subsequent use. When designing an amplifier for yourself, you can often take shortcuts, without worrying about future service.

The choice fell on a well-known circuit, proven in thousands of homes worldwide. This will be the next version of the D.T.N. Williamson amplifier. Almost every company that once manufactured tube amplifiers offered a product based, to one degree or another, on this famous circuit. Hundreds of articles describing various variations of Williamson amplifiers can be found online. It's worth taking advantage of this wealth of experience today.

Design assumptions

In 1947, D.T.N. Williamson presented an amplifier circuit that represented a true breakthrough in the pursuit of high-quality sound reproduction. Its most distinctive features included a split-load phase splitter and a transformer capable of transmitting a signal in the 2–60,000 Hz range (a necessary condition for amplifier stability in a closed-loop system).

All stages of the Williamson amplifier are essentially extremely simple, yet they work together remarkably well, ensuring relatively low signal distortion. However, the circuit suffers from several shortcomings, which were improved over the years. The drawing below shows the 1949 version of the amplifier, with component values indicated.

The input stage, phase inverter, and driver stage were typically built around 6SN7, 6CG7, or 12AU7 (ECC82) tubes. Their operating point was incorrectly selected, resulting in harmonic distortion of around 2% at half the rated power. Appropriate circuit modifications allow for distortion levels no greater than 0.5%, down to clipping.

Early versions of the amplifiers sounded too soft when attempting to reproduce strong bass. This was due to poor power supply filtering. Increasing the filtering and decoupling capacitances not only improved impulse handling but also improved the amplifier's stability.

Another less-than-ideal idea was to use a pair of tetrodes in the output stage, connected in a triode configuration with a common cathode resistor and without capacitive decoupling. This arrangement significantly reduced output power and good high-frequency response. A much better design was the ultralinear amplifier, proposed in 1951 by David Hafler and Herbert Keroes, which achieved significantly lower distortion with similar output power and global feedback parameters. Fortunately, the ultralinear output stage can be perfectly combined with the other stages of the Williamson amplifier.

After reading many (fortunately easily accessible) articles and analyzing even more schematics of similar amplifiers, the basic design assumptions crystallized:

Topology based on modified versions of Williamson amplifiers.
Split-load phase inverter.
Power stage in an ultra-linear system.
A simple power supply that provides a "soft start" for the amplifier.

The following sections will discuss individual parts of the Concertino amplifier.

Input stage

The first two stages of the amplifier – the input amplifier and the phase splitter – should be considered together.

In Williamson's original circuit, and virtually all of its later variations, the input stage utilizes a common-cathode tube with local feedback and a global feedback signal applied to the cathode. The output signal from the anode is fed directly to the input of the phase splitter in a split-load configuration. For the phase splitter to function properly, its grid voltage must be relatively low (on the order of 25–35% of the stage's supply voltage, typically 90–100V). This creates very unfavorable operating conditions for the first tube, especially if it is a 6SN7 tube, which operates much better at anode voltages in the 150–250V range. The article "Williamson Tube Amplifier Modifications" describes a method for improving this situation by modifying the operating points and supply voltages of both tubes.

The split-load phase inverter, while maintaining excellent symmetry of both output waveforms, has a very low power supply ripple rejection ratio and a significant difference in the output impedances of both branches. The latter drawback is effectively eliminated in the third amplifier stage (the power stage driver), while the low PSRR can be improved by appropriate modification of the first two amplifier stages.

The solution to this problem was described by John Broskie in TubeCad Journal (April 1999) and implemented in his Aikido series of amplifiers. The low PSRR of the phase splitter was considered an advantage and contributed to very good power supply ripple suppression in the first two amplifier stages. The idea is to deliberately introduce noise into the circuit to obtain a clean output.

If the operating point of the first tube is set so that the DC voltage at the anode is half the supply voltage of this stage, the output will show grid ripple attenuated by 6 dB (half the amplitude). The ripple will transfer through capacitor C₁ to the grid of the second tube and appear in phase at the cathode of the second stage and in opposite phase at the anode of the second stage. However, ripple coming directly from the power supply will also appear in phase at the anode of this stage. Both waveforms, when added together, will produce a waveform at the anode similar in phase and amplitude to the signal at the cathode of the second stage. This creates a common waveform, not a differential one, which will be attenuated in the subsequent amplifier stages. The circuit, of course, suppresses not only the ripple but also all interference from the power supply.

The figure above shows a simulation of mains ripple suppression for the first two amplifier stages (all waveforms shown without DC bias). To achieve a measurable result, the power supply ripple amplitude was exaggerated (60V) and the filtering capacitances were drastically reduced. The figure shows a first-stage power supply ripple amplitude of 0.95V. The differential signal at the second-stage output does not exceed 37mV, so mains ripple suppression reaches 28dB. For comparison, the ripple suppression of the input stages of a traditional Williamson circuit is only a dozen or so dB.

The circuit described above has two further advantages. Resistors R₄ and R₅ allow for conveniently setting the operating point of the second tube, which allows for maximum output amplitude of the phase inverter (usually set to R₄ = R₅). These resistors can have very large values (on the order of several megaohms).

If the values of R₂ and R₆ are the same, the first two stages will form a circuit with constant current consumption (the current flowing through resistor R₉ does not change). This condition will guarantee maximum power supply stability for the first two stages when the amplifier is driven by the input waveform.

The last figure shows the layout of the first two stages of the Concertino amplifier. A trim potentiometer in the first stage cathode circuit allows for minimizing the differential mains ripple signal measured between the second stage outputs (which should occur when the anode potential of the V_1B tube is equal to half the supply voltage of the first two stages). The gain of the first stage is about 10 without closed feedback and slightly over 2 with feedback. The first two stages use an ECC82 double triode.

Driver stage

The third stage of the amplifier – the output stage driver – contains no major surprises. It is a differential amplifier with a common, non-blocking cathode resistor. Adding a capacitor in parallel with the cathode resistor will not fundamentally change the circuit's operation, but it will unnecessarily increase stage distortion (deteriorating stage symmetry).

Choosing the right tube for this stage can be a challenge. Williamson's proposed 6SN7 tube loads the previous stage with a relatively large input capacitance (around 70 pF). This isn't a problem for the low-impedance cathode phase inverter circuit, but for the anode circuit, it reduces the gain at high frequencies. Although the driver copes with this problem quite well, it's worth considering a tube with a smaller grid-to-anode capacitance.

The ECC82 is such a tube. However, it has another fundamental drawback: the driver stage must be able to amplify the waveform to 30-35V. At such amplitudes, the ECC82 tube generates significant distortion, about 2.5 times greater than the 6SN7. Therefore, a poorly designed Williamson amplifier, without a feedback loop, can achieve distortion of 5-10% before clipping occurs (and most of this distortion comes from the driver stage).

The 6SN7 tube shows its advantage even at much lower amplitudes. Ultimately, I decided to use 6SN7EH tubes (the equivalent of the old RCA 5692 with a red base). The driver stage built around these tubes has a gain of about 16.

In many common circuits, the value of the cathode resistor R₁₆ is too low (e.g., 220Ω or 390Ω), which unfavorably sets the tube operating point (U_gk = -2÷-3V). Since the driver stage must be able to supply the power tubes with a voltage of about 70V_pp, the voltage on the grid of the 6SN7 tubes will vary by more than 4V_pp. To keep the grid of these tubes out of the positive voltage range, it is safer to set U_gk = -4V or lower.

In some amplifiers (including Williamson's original circuit), the supply voltage is applied to the anode resistors R₁₇ and R₁₈ via an additional trimmer potentiometer, which helps adjust the circuit's symmetry for the AC waveform. In practice, the circuit handles slight waveform asymmetry well, and using two equal anode resistors is perfectly sufficient.

Output stage

The output stage of the Concertino amplifier will be designed in an ultralinear configuration according to the recommendations of Hafler and Keroes from 1951.

Ultralinear output tubes can achieve approximately 75% of the power of a tetrode/pentode configuration and at least twice the power of a triode configuration, with comparable operating parameters. The sound produced by ultralinear circuits, due to their low distortion and improved damping factor, resembles that of triode output stages. Therefore, ultralinear circuits are ideal for hi-fi amplifiers, but are less commonly used in guitar amplifiers.

Recommended tubes for ultralinear circuits are usually PL36, 6V6GT, 6973, 5881/6L6GC/7591, KT66, KT88/KT90, 813, EL34, 6CA7, and EL84 pentodes provide a slightly less pure sound, although they are equally common. My choice fell on the 6L6GC beam tetrodes currently manufactured by JJ Electronic.

Another issue was the selection of a bias circuit for the output tubes. Fixed bias would improve the amplifier's efficiency and facilitate the regulation of the output stage's quiescent current. The cost savings are practically nonexistent, as a circuit must be built to generate and regulate a negative bias voltage. An automatic bias circuit is simpler to implement and is considered by many to sound better. Tubes with automatic bias tend to clip more gently, which can be important for loud listening and inefficient speaker systems.

The output tube biasing system I used has a simple but effective quiescent current symmetrization mechanism, found both in Williamson's original and in later designs, such as those from Heathkit. The tubes operate in class AB1 (up to about 10W in class A). The value of resistor R₂₅ sets the quiescent current of both tubes (the sum of the cathode currents is approximately 120mA, which, at an anode voltage of 430V, will result in approximately 23.5W of quiescent power dissipation in each tube). The wire-wound potentiometer R₃₂ equalizes the quiescent currents of both tubes (the cathodes should have the same potential, which is a sufficient approximation of the condition for equal anode current). Blocking capacitors in the cathode circuit of the power tubes can slightly affect the level of output stage distortion (in which direction depends on the circuit and the power tubes used). In some designs, these capacitors are omitted, but in class AB they must be used.

According to Williamson's recommendations, the output transformer should cover a frequency range of 2-60,000Hz. It's not easy to obtain a transformer with the appropriate parameters (many commercial audio transformers were designed with guitar amplifiers in mind). Ultimately, I decided not to risk it and purchased the proven TG36 toroidal transformers, sold by Amplifon and used in their WL36 and WL25 amplifiers in a very similar configuration. The transformers have multi-section windings, ultra-linear taps, and a nominal plate-to-plate impedance of 6.6kΩ (making them suitable for use with 6L6GC tubes).

The gain of this stage in an ultralinear configuration (43% primary winding tap) is around 9. The amplifier should easily deliver around 25W of power per channel. So, there shouldn't be any problem driving my small monitors with an efficiency of around 87dB/W/m.

Power supply

Currently, building a power supply using a simple mains transformer, a semiconductor bridge, and a bank of filtering capacitors seems to be the simplest and most appropriate solution. However, such a circuit generates significant switching noise and requires an additional circuit for delayed anode voltage application. So why not resort to old, proven techniques and solve several problems at once?

So I'll use a tube rectifier with an LC filter. Directly heated rectifier tubes are not suitable – a delayed start circuit would be necessary here too (the lack of such a circuit was a serious flaw in Williamson's original design). Indirectly heated tubes, such as the 5AR4/GZ34, are a good choice. However, one tube won't power both channels – two tubes are needed to ensure the operating limits (270mA average current draw and approximately 1A peak current draw) aren't exceeded.

Lynn Olson's pages came to the rescue again. Like Olson, I'll be using two 6D22S snubber diodes for rectification. Each has only a single anode, so a full-wave rectifier requires two tubes. These have their drawbacks: Magnoval pinouts (harder-to-reach sockets) and a capped cathode connection on top of the tube. In return, they offer very low switching noise, low forward voltage (15V), high peak current (2A), and a very long warm-up time (30s), thus solving the amplifier's soft-start problem. 6D22S tubes require a 6.3V filament voltage. It's acceptable to use a common 6.3V winding for all tubes in the circuit, but in my amplifier, the rectifier tubes will receive filament current from a separate winding (with the option of forcing the winding to the potential of the tubes' cathodes).

I will build the filter as a double П: CLCLC. I will use two air-gap chokes (fluorescent lamp ballasts) with an inductance of 1.56H, a resistance of 48Ω, and a maximum current of 0.37A. The power supply simulation performed with PSU Designer II shows an anode voltage of 428V with 2.46mV ripple. In the case of a single П-type filter, I would need to use a choke with an inductance of over 100H to achieve a similar effect.

The mains transformer was custom-made by MKPT (made by Telto). It's a 250VA toroid, providing 2x390V output for the anode circuits, 6.3V for the 6D22S tube filaments, and 2x3.15V for the remaining tube filaments (center tapped to ground to minimize mains hum). If my estimates are correct, I should get an anode voltage of around 430V under load.

Amplifier diagram

The drawing shows the complete schematic of the amplifier (clicking opens a detailed drawing).

The signal ground marking depends on its location. This corresponds to the principle of ground distribution in a star configuration. All local grounds converge at a single point near capacitor C₃₆. This point also connects the signal ground to the amplifier chassis and the power supply's protective ground.

Resistors R₂₂ and R₂₃ apply a global negative feedback signal. The amplifier has an open-loop gain of approximately 90. With the feedback loop closed, the gain drops to less than 20. The loop depth (13dB for the component values shown) is set by the value of resistors R₂₂ and R₂₃ (the setting of potentiometers R₃ and R₄ also has an effect). Full amplifier drive (25W) is achieved with an input signal amplitude of approximately 1V, so there will be no problem driving the amplifier from typical audio sources (CD, tuner, tape recorder).

The Zobel circuit between the secondary winding leads of the output transformers improves the amplifier's high-frequency stability (especially important when the load is disconnected).

The amplifier circuit includes components that ensure amplifier stability in the supra-acoustic range. These include capacitors C₃, C₄, C₇, and C₈, and resistors R₁₃ and R₁₄. The values of capacitances C₇ and C₈ should be selected experimentally during amplifier startup (a criterion for minimizing overshoots and oscillations in square wave reproduction).

At the amplifier input, I used a dual 100kΩ linear potentiometer. Together with resistors R₇ and R₈, I obtained an approximation of the exponential characteristic (significantly better than most so-called logarithmic potentiometers). The operation of such a potentiometer was described by Rod Elliott in his article "Better Volume Control." The figures below show the measured common-mode error of both potentiometers and their characteristic curve. This error is minimized by selecting the values of resistors R₇ and R₈. At typical volume levels (60-20dB attenuation), the error of the modified potentiometer does not exceed 0.15dB. I have never obtained similarly good results when measuring factory-made logarithmic potentiometers. Potentiometers from the renowned company ALPS allow for a common-mode error of as much as 3dB.

A drawback of the volume control shown here is that the load on the signal source changes with the slider position. In the far right position, the input resistance drops to approximately 13kΩ.

The input signal from the volume slider also goes to the buffer circuit. This circuit is not shown in the diagram because it serves an auxiliary role (providing a signal for an active subwoofer).

Amplifier housing

This is where the real problems begin. Not every amateur electronics technician has a sufficient mechanical workshop. I can do simple tasks like drilling, sawing, and sanding myself at home. More complex tasks, like bending sheet metal or cutting holes for vacuum tubes, I'll have to entrust to specialists.

In my case, the case will have to match the rest of my audio equipment, which means an integrated black design with a width of 43cm. The case must also be simple and allow for easy installation. Furthermore, it can't detract from the interior design or cost a fortune.

I decided to build the enclosure schematically shown in the drawing. The basic chassis will consist of 2mm thick bent steel sheet. The folds will form the front and rear panels. The sides will be made of varnished wood and permanently attached to the sheet metal. Access to the interior of the enclosure will be provided by a screw-down bottom – a 1mm thick steel plate. The tubes and toroidal transformers will be located on the top of the unit and will require additional protective and masking elements. The remaining components will be mounted inside the enclosure.

A 2mm thick steel plate forms the main chassis, measuring 398mm x 360mm and only 50mm high. The sheet metal bend is not perfectly perpendicular and has a relatively large radius, but this is not a disadvantage given the intended housing design.

The chassis after drilling and sawing. Over 100 holes were drilled, though these are only the necessary ones (some components will be glued).

Chassis after powder coating and screen printing.

The side panels are in preparation. After several trials, I decided to finish the surface with an "ebony" varnish stain.

The assembled chassis creates a rigid and durable box.

Amplifier assembly

For assembly purposes, I built a special bed from wooden parts, on which the chassis rests securely, and which will later enable me to run the amplifier in an upside-down position.

First, I installed a steel angle bracket inside the enclosure to further stiffen the structure. The enclosure must support approximately 8kg of weight, which consists of two chokes and three toroidal transformers.

Next, I installed all the external components (tube sockets, sockets, and switches) and the power supply filter chokes. The photo above also shows most of the solder terminals, which allow for easy spatial assembly of the electronic components. The terminals and spacers were glued to the housing with epoxy glue.

After installing the transformers, the amplifier weighs almost 10kg. Moving, lifting, or turning such a heavy structure became quite a challenge from that point on.

"Cramming" the transformer's numerous leads was the first major assembly task. There wasn't much space, and the leads had to be shortened significantly.

The 6D22S diodes are now in place. The wires leading to the cathode caps are hidden within the bent aluminum tubes.

The power supply is assembled and ready for testing. Inside, you can see the opposite ends of the aluminum tubes shown in the previous photo. Slightly to the left (near the choke) are several solder lugs. This is the amplifier's central ground point – this will be the ground connection for subsequent circuits. At this stage, the filament circuits for all the tubes were also installed.

The power supply's temporary load consisted of five 30-watt resistors with a total resistance of 1800Ω. During the test, they should have dissipated over 100W of power.

Fortunately, the first power-up was without any unwanted pyrotechnic effects.

As expected, the use of 6D22S diodes ensures a long and gentle start-up of the power supply. The first volts appear at the load approximately 15 seconds after the system is turned on. The voltage then gradually increases, reaching the target value after approximately 35 seconds.

Two chokes and two large capacitors, each with a capacitance of 500µF, constitute a very effective main power supply filter. Under test conditions, the filter input measured 462VDC with 26Vrms ripple. At the filter output, at 439VDC, the mains ripple dropped to less than 0.5mVrms. These values are fully consistent with the simulation results performed using PSU Designer II.

After powering up the power supply, it was time to assemble the output stage. There were only a few components to assemble, but some of them were quite large, such as the 0.47µF blocking capacitors, the five-watt cathode circuit resistors, and the 100Ω wirewound potentiometer (at the top of the photo).

The power stage booted up without any problems. A few seconds after powering up, the anode voltage slowly increases (the power tubes are already warmed up and putting a strain on the power supply) and stabilizes after about 40 seconds. Due to the power supply not being fully loaded (the first amplifier stages are missing), the anode voltage is slightly too high (the target is approximately 430V).

The value of resistor R₅₆ is selected to achieve the appropriate quiescent current for the power tubes. Two 470Ω resistors connected in parallel proved to be adequate (cathode current of each tube is approximately 57mA).

The wire-wound potentiometer R₅ effectively equalizes the quiescent current of both power tubes (equal potential for both cathodes).

After connecting the speaker, a slight mains hum could be heard. The signal measured at the amplifier's output was 0.8 mV_rms at a fundamental frequency of 100Hz. The photo on the right shows the transformer output waveform.

Further testing confirmed that the hum was not caused by the amplifier's component layout. Changing the filament circuit configuration (including various circuit symmetrization methods) and changing the grounding method did not affect the output noise level.

After removing the output tubes and running the amplifier with a dummy load, it turned out that the output signal was still induced (the effective value dropped to 0.5mV). This clearly indicates magnetic coupling between the output transformers and the mains transformer. Placing a simple sheet steel barrier between the transformers significantly reduced the mains hum. Changing the relative position of the transformers also significantly reduced the hum, but I would ultimately prefer to avoid this method of eliminating interference.

The solution to the problem would be to magnetically shield the transformers (probably reducing the hum by about 10dB) and to implement global negative feedback (reducing the hum by several dozen dB). In this case, the hum level should not be a problem even with high-efficiency loudspeakers.

In the third stage, the driver stage was assembled. The photo shows only a few of the resistors and capacitors that make up this stage. The power supply filter for this stage (a 4.7kΩ resistor and a 56µF capacitor) is also visible on the left.

Starting the stage didn't yield any surprises.

As you can see in the attached drawing, the stage's supply voltage is currently slightly higher than the nominal value (350-360V) due to the RC filter not being loaded by the voltage stage (not yet assembled). Therefore, the driver stage's quiescent current is slightly higher than expected, but this does not affect the correct operation of the circuit.

The quiescent current and anode voltage of both branches are not identical due to the discrepancy in the parameters of the two triodes. Because this is a differential amplifier configuration, it is impossible to equalize these currents without violating symmetry for the AC component.

The stage operates correctly for the AC signal (perfect symmetry). The measured voltage gain is 17 (slightly higher than the previous estimates).

In the final step, the voltage stage and phase splitter were assembled and powered up. At the top of the photo, you can see the first triode's quiescent current adjustment potentiometer (to achieve an anode voltage exactly half the first stage's supply voltage).

The voltages and currents at individual locations in the circuit are indicated in the diagram. The components compensating the amplitude-phase characteristic (C3 and R13) will be selected only after the negative feedback loop is closed.

The circuit's operation for the AC component is correct. The measured voltage gain of the first stage is 9.78, and the phase splitter gain is 0.87 (in each branch). Due to the very good filtering of the supply voltage (mains ripple is immeasurable), it is difficult to observe and measure the desired performance of the Aikido amplifier (the operating principles are provided on the "Input Stage" page). The circuit may later be simplified to a traditional Williamson circuit (comparison tests will be required).

The entire circuit, tested in open-loop mode, generates a small output noise and mains hum of 1.25V_rms (audible through a loudspeaker from about 30cm away). Both types of interference are reduced by the use of global feedback. It's worth noting that the mains hum remained at the level measured immediately after powering up the output stage. This indicates the absence of additional sources of hum in the input and driver stages.

The figure below shows the test signal voltage values at various locations in the circuit (green), the gain values of the individual stages (blue), and the supply voltages of the individual stages (red). The amplifier's open-loop gain is 92.5.

Starting the system

It's often assumed that for ultralinear systems, it's sufficient to include a global feedback loop of several dB within the amplifier. It's worth remembering that the implemented circuit incorporates a number of local feedback loops (in the input stage, the phase splitter, and the output stage's screen grid), which reduce signal distortion even without the use of global feedback. However, global feedback is necessary, if only to lower the amplifier's output impedance.

At this stage, I assumed a global feedback depth of 16dB, which, with an open-loop gain of 92.5, would ensure full amplifier drive with an input signal amplitude of approximately 1.35V. With a first-stage cathode resistance of approximately 600Ω, a 10kΩ feedback resistor would be required. Frequency compensation elements were also included, the target values of which would be selected later in the amplifier's commissioning process.

After connecting the feedback loop, I detected no oscillations in or above the audio frequency range. The mains noise and hum decreased to barely audible levels, but this was unmeasurable. It turned out that the amplifier was unstable below the audio frequency range. The output level oscillated irregularly in the range of approximately 200mV with a peak frequency of 1-2Hz. This oscillation did not affect the audio signal's transferability, and driving the amplifier with a signal did not affect the amplitude or frequency of the oscillation. Before further measurements, the cause of this instability had to be eliminated.

The transformer's lower limit frequency, measured in the circuit at 1W output power, is approximately 5Hz (during tests below 6Hz, the output waveform already exhibited visible distortion due to transformer core saturation). This is the dominant pole of the circuit. The next three poles originate from the RC elements coupling the amplifier stages and lie around 1.5Hz. Around 1Hz, the phase difference between the amplifier's output and input reaches 180 degrees, with the open-loop gain remaining high. This causes irregular oscillations in the circuit around 1Hz.

The solution to this problem is to move the poles apart and reduce the open-loop gain for frequencies below 16Hz. I made the following modifications to the circuit:

1. Changing the values of the coupling elements between the driver and power stages (R₄₂ = R₄₃ = 220kΩ, C₁₇ = C₁₈ = 0.047uF). This establishes a new dominant pole for a frequency of 16Hz.

2. Adding a 10uF capacitor to the input circuit. This allows the Aikido circuit to operate efficiently at frequencies below 1Hz (cutoff frequency 0.016Hz) and makes the polarity of this stage irrelevant.

3. Changing capacitor C₂₅ in the first-stage power supply filter from 22uF to 100uF. This reduces the slow-changing power supply drift around 1Hz (new filter cutoff frequency 0.16Hz).

Change No. 1 is crucial for ensuring circuit stability. Reducing the time constant of the RC coupling elements in the output stage also has another desirable effect – faster recovery from clipping (when the voltage on the control grid momentarily exceeds the cathode potential, capacitor C₁₇, charged by the momentary grid current, must then discharge through resistor R₄₂).

In many production tube amplifiers, the cutoff frequency of the last RC element was set quite high: 7Hz (Altec Lensing, Audio Innovations, Heathkit, Jolida) or 16Hz (Eico, Grommes). This ensured sufficient low-frequency stability for circuits with two or more capacitive coupling stages. However, many Williamson circuits published online (including the well-known circuit from Practical Electronics) certainly do not provide sufficient stability below the audio frequency range (at least not if the output transformer used has any worse parameters than the original Partridge design).

The figure above shows the open-loop amplitude and phase response of the Concertino amplifier (for frequencies < 30Hz). Using a 16dB global feedback depth, I achieved approximately 45° phase margin and 8dB gain margin.

After stabilizing the system, I was finally able to measure the amplifier's output noise. The meter showed approximately 0.2mV_rms. After temporarily shielding the output transformer, the noise level dropped to 0.1mV. Mains hum was practically inaudible, even with my ear pressed directly against the speaker.

The time was ripe to determine whether the Aikido input stage offered any practical advantages over a typical Williamson input stage. To minimize errors, I performed the measurements simultaneously (using the Aikido circuit in one channel and the Williamson in the other, swapping channels during the tests).

In all tests, the Aikido circuit demonstrated its superiority, producing measured results within the 0.15-0.22mV range, while the Williamson circuit produced results within the 0.24-0.50mV range (always 2-8dB worse than the adjacent channel).

Using temporary transformer shielding, the interference dropped to approximately 0.115mV for the Aikido circuit and 0.175mV for the Williamson circuit. These differences clearly demonstrate the viability of using the Aikido circuit.

The Williamson amplifier will likely also require appropriate high-frequency compensation. I estimate the cutoff frequency of the output transformer I'm using to be around 70kHz. This represents the lowest pole positioned above the audio frequency range. The subsequent poles come from the "upper" half of the driver stage (110kHz), the input stage (800kHz), the output stage (1.5MHz), and the "lower" half of the driver stage (2MHz). With appropriate compensation, the latter three should not affect the system's stability.

Without compensation, we can expect an A*b loop gain of 1 somewhere around 200kHz and a phase shift of approximately 150°. This should ensure the amplifier's stability with a resistive load connected, and perhaps even without it (in which case the Zobel circuit connected to the output acts as the high-frequency load).

Tests demonstrated the amplifier's actual stability with a resistive load and relative stability without a load (the amplifier oscillated when driven by a signal).

However, the target load (a loudspeaker with an electrical input crossover) will require a significantly greater margin of stability. Connecting the loudspeaker to the amplifier's output resulted in oscillations with a frequency of just under 200kHz. An equally disturbing effect arose when a 0.22uF capacitor was added to the load – fourteen-millisecond bursts of oscillations at 185kHz with a nine-millisecond break between bursts. The amplifier undoubtedly requires frequency compensation to achieve stability regardless of the type of load connected.

Elements R₁₃ and C₃ introduce delay compensation in the transient range. With the values shown in the figure, the circuit creates a new dominant pole at f=23kHz and a zero at f=110kHz. The next pole is at f=70kHz, where the open-loop gain drops to approximately 20dB (26dB), and the phase shift is approximately 120°. The third pole, at f=110kHz, is nullified by the zero from the compensation circuit. This maintains a steepness of 12dB/octave until the fourth pole, located around f=800kHz.

Slightly above the second pole (around 90kHz) is the point where the loop gain A*b=1. The phase shift at this point reaches approximately 130°. A phase margin of 50° should ensure absolute amplifier stability.

The acceleration compensation shown in the figure below affects the transfer function b of the feedback circuit. The value of C₇ = 56 pF introduces a pole at f=130kHz and a zero at f=11MHz. This compensation is not necessary, as the delay compensation already provides sufficient system stability. However, it is recommended to "accelerate" the feedback loop at higher frequencies and reduce overshoots and oscillations in the reproduced pulses. However, too large a capacitor value can destabilize the amplifier.

In practice, because the phenomena occurring in the circuit are somewhat more complex, the values of the amplifier's compensation elements should be verified experimentally to achieve the required system stability. I used the method described many times by Patrick Turner on the rec.audio.tubes discussion group.

In the first step, with the assumed value of capacitor C₇ in the feedback circuit (current value 47pF), capacitance C₃ must be selected for the delay compensation circuit. Loading the amplifier solely with capacitances in the range 10nF to 4.7uF will result in a peak in the frequency response characteristic that depends on the connected load. Capacitance C₃ must be large enough so that this peak never exceeds +6dB from the nominal level (measured at 1kHz), and that in the acoustic range (f<20kHz), the response characteristic does not deviate by more than 1.5dB from the nominal level. These conditions were met by capacitor C₃ with a capacitance of 680pF (maximum measured peak of +4.77dB at f=71kHz with a 1uF capacitor connected to the output). Using a value of C₃=680pF limits the amplifier's open-loop bandwidth to f=17kHz (measured value). While a larger capacitance will improve the system's stability, it will also reduce feedback where it is still needed (below 10kHz).

The second step involves selecting the value of the R₁₃ resistance in the delay compensation circuit. Find the maximum resistance value at which the amplifier will not oscillate regardless of the connected load (nominal resistive, capacitive, inductive, or no load). Tests should be performed with no signal present and with the amplifier driven by a square-wave signal of varying amplitude. In my case, the maximum value of R₁₃ is 4kΩ.

While searching for the maximum value, I also check the value of R₁₃ that provides the optimal waveform (minimum pulse overshoot, minimum oscillation, maximum slope). Ultimately, I decided to use a value of R₁₃=3kΩ.

Load 8Ω; f=4800Hz; 1V/div; 50us/div

Load 1uF; f=4800Hz; 1V/div; 50us/div

The final step is to check the value of the compensating capacitor C₇ that provides good damping of oscillations on the envelope of rectangular pulses (so-called ringing). Caution should be exercised here, as excessively increasing this capacitance can lead to amplifier instability under certain operating conditions. If the specified value of C₇ differs significantly from the previously assumed value, the allowable value of resistor R₁₃ should be re-verified.

All this is easier said than done. The entire procedure is time-consuming, but it leads to achieving a good margin of amplifier stability. The result is an unconditionally stable amplifier that:

does not oscillate without a connected load,
does not oscillate with a coil load of any value,
does not oscillate with a load in the form of a capacitor of any value in the range of 0.01÷10uF,
does not oscillate with any of the above loads, driven by a square wave signal,
driven by a sinusoidal signal with a frequency of several Hz and an amplitude sufficient to saturate the output transformer does not induce oscillation packets at the moments of saturation of the transformer core.

A good test is to find the maximum feedback value at which the amplifier remains stable. In my case, the feedback resistor can be reduced to 1.6kΩ without any trace of oscillation at the amplifier output. This results in a feedback loop depth of 28.3dB. Therefore, we can assume that the amplifier with a resistor load has a sufficient gain margin of 12.4dB.

Clicking on the image will open the final Concertino amplifier schematic with all the corrections described above.

Preamplifier buffer

The preamplifier buffer serves an additional purpose, unrelated to the actual tube amplifier circuit, and therefore isn't included in the main schematic. The buffer's purpose is to separate the regulated output signal intended for the external subwoofer from the input circuit of the first amplifier tube. It's the only part of the circuitry that contains solid-state components (and while I see nothing wrong with their use, the core amplifier circuitry remains free of them to best resemble the circuits used half a century ago).

The signal from the volume slider is fed to the input of the non-inverting amplifier. This configuration's high input resistance (around 1e12Ω) ensures that the buffer has no effect on the tube amplifier's input signal. The buffer's gain is approximately 16dB. With an input signal amplitude of 1.6V (the maximum signal that will not overdrive the tube amplifier), the buffer output signal has an amplitude of 10.3V, which is within the operating range of an operational amplifier with a ±12V power supply.

The circuit is powered by an additional small mains transformer with a secondary voltage of 2x12V. Since the circuit has only a few components, it was assembled on a piece of universal board.

Parameters

Unless otherwise noted, measurements were performed with an 8Ohms resistive load and without transformer shielding.

Layout: Williamson; Aikido front end; ultralinear push-pull; AB1 class
Nominal output power: 2x25W (f=1kHz sine; THD=0.21%)
Maximum output power: 2x32W (f=1kHz sine; THD=1%)
Power band:
- 7Hz÷78kHz (P=0.2W sine; ±3 dB; 0dB for f=1kHz)
- 7Hz÷75kHz (P=1W sine; ±3 dB; 0dB for f=1kHz)
- 10Hz÷68kHz (P=5W sine; ±3 dB; 0dB for f=1kHz)
- 17Hz÷60kHz (P=25W sine; ±3 dB; 0dB for f=1kHz)
Frequency Response Uniformity: ±0.1 dB (f=20Hz÷20kHz; P=1W)
THD for f=1 kHz
- 0.03% (P=0.2W sine; f=1kHz)
- 0.03% (P=1W sine; f=1kHz)
- 0.08% (P=5W sine; f=1kHz)
- 0.21% (P=25W sine; f=1kHz)
THD for f=20Hz÷10kHz
- <0.05% (P=0.2W sine; f=20Hz÷10kHz)
- <0.1% (P=1W sine; f=20Hz÷10kHz)
- <0.2% (P=5W sine; f=20Hz÷10kHz)
- <0.6% (P=25W sine; f=20Hz÷10kHz)
Output Mains Noise and Hum Level
- <0.2mV (97dB below nominal level; without transformer shielding; 8W tap)
- <0.1mV (103dB below nominal level; with transformer shielding; 8W tap)
Input Impedance: 47kΩ (f=20Hz÷20kHz)
Nominal Load Impedance: 4Ω or 8Ω
Input Sensitivity: 0.95Vrms sine (P=25W)
Gain Voltage: 14.83 (8Ω tap)
Damping factor: 3.3 (estimated)
Global feedback: 15.9 dB

Power bandwidth measured at 4 different levels. The asterisk on the graph indicates the point where distortion levels increase dramatically due to output transformer saturation.

Uneven frequency response in the audio band. 0dB corresponds to 1W of power into an 8Ω resistive load.

Input-output phase characteristic (P=1W).

Harmonic distribution at the amplifier output for a sinusoidal signal f=1kHz, P=1W. Total harmonic distortion THD=0.025%.

Total harmonic distortion as a function of output power (f=1kHz).

Total harmonic distortion as a function of frequency (P=0.2W).

Total harmonic distortion as a function of frequency (P=1W).

Total harmonic distortion as a function of frequency (P=5W).

Total harmonic distortion as a function of frequency (P=25W).

Intermodulation distortion spectrum (f1=17kHz, f2=18 kHz).

Intermodulation distortion spectrum (f1=1kHz, f2=1.1kHz).

The spectrum of the signal at the amplifier output when driven by a sinusoidal waveform (f=1kHz, P=1W).

Noise and interference spectrum at the output of an undriven amplifier (unweighted measurement).

Noise and interference spectrum at the output of an undriven amplifier (weighted measurement - ANSI A).

The amplifier experienced slight clipping when driven by a high-amplitude 20Hz sine wave and a low-amplitude 1kHz wave. The amplifier showed no signs of clipping. The input level was 113% of the maximum level that would not cause clipping.

The amplifier experiences severe clipping with the input waveform shown in the photo above. The amplifier shows signs of clipping for no longer than half the waveform's period. The input level is 145% of the maximum value that does not cause the amplifier to clip.

Links and gallery

Main resources used in the amplifier design:

Lynn Olson - The Amity, Raven, and Aurora
Russ Sadd - Griffon Amplifiers
Bert van der Kerk - The Williamson Tube Amplifier
Chimera Laboratories - Williamson Tube Amplifier Modifications
John Broskie - Tube CAD Journal
Randall Aiken - Aiken Amplification
Patrick Turner - Turner Audio
Duncan Munro - Duncan's Amp Pages
Rod Elliott - Elliott Sound Products
rec.audio.tubes newsgroup (authors: Patrick Turner, Henry Pasternack, Dan Marshall, Chris Hornbeck and others)