Photos and description - Marcin Sławicz

The beginnings of the project

  The idea of ​​building my own tube amplifier has been bothering me for the last two years. I am not a maniacal audiophile and using "ordinary" solid state equipment was enough for me (I always preferred to listen to music than equipment). Now, however, my worn-out amplifier is starting to suffer from the ailments of old age, and although I could regenerate it, there is a great opportunity to implement a tube venture.

  At the beginning I was thinking about a design based only on triodes, but rejecting the SE circuits burdened with too many inconveniences. A very interesting description of the push-pull amplifier with direct filament triodes can be found on the Lynn Olson website. It is worth taking a look there because of the extremely interesting solutions used in his projects. The described amplifiers, however, have a major disadvantage - cost (mainly due to the 300B or 2A3 tubes and interstage transformers). So I had to look further.

  My attention was drawn to indirectly heated double 6AS7 power triodes, once used mainly in power supply systems, but also great as electron tubes in the output stage of audio amplifiers. The cost of electron tubes would be much lower, but due to the low voltage gain factor, in this case, expensive and difficult to obtain interstage transformers or two or more triodes in parallel connection would have to be used. Mr. Russ Sadd described on his website a push-pull amplifier with 6AS7 triodes.

  My project took a few more months, during which I slowly became convinced that a successful power amplifier does not have to have triodes in the output stage. I began to consider the use of beam tetrodes working in the gain stage in an ultra-linear configuration. Such a circuit combines the advantages of triode sound (low distortion) with high efficiency and stability of tetrodes and pentodes. I had a choice of 6L6 / 5881, KT66, KT88 / 6550 tubes, commonly used both in guitar amplifiers and in Hi-Fi designs.

  Another period of my project is searching the net in order to select the basic amplifier circuit. The amplifier should not be complicated, because a complex circuit does not guarantee high-quality sound, and with limited measurement possibilities, it can be difficult to start up. Mass-produced devices must ensure the repeatability of production and the relative stability of parameters during subsequent operation. When designing an amplifier for yourself, you can often take shortcuts without worrying about the subsequent service.

  My choice fell on a well-known layout that has been tested in thousands of homes around the world. It will be the next version of the D.T. N. Williamson. Almost every company that used to produce tube amplifiers had a product to a greater or lesser extent based on this famous circuit. You can find hundreds of articles on the Internet describing different varieties of Williamson amplifiers. So let's take advantage of these rich experiences today.

Design assumptions

  In 1947, Mr. Williamson introduced an amplifier circuit that was a real breakthrough in the pursuit of high-quality sound reproduction. The most characteristic elements of this amplifier are the split load phase splitter and the use of a transformer transmitting the signal in the range of 2Hz ÷ 60,000Hz (a necessary condition for achieving the stability of the amplifier with a closed feedback loop).

  All stages of the Williamson amplifier are, in fact, extremely simple, but at the same time they perfectly cooperate with each other, ensuring relatively low signal distortion. Nevertheless, the system has several drawbacks, which efforts were made to improve in the following years. The figure below shows the 1949 version of the amplifier with the component values marked.

  The input stage, phase inverter and control stage were usually built on 6SN7, 6CG7 or 12AU7 (ECC82) electron tubes. Their operating point was selected incorrectly, which resulted in harmonic distortions of 2% already at half the nominal power. Appropriate modifications to the system allow to obtain distortions not exceeding 0.5% up to the level of waveform clipping.

  The first versions of the amplifiers sounded too soft when trying to play strong bass. This was due to poor power filtering. Increasing the capacity of the filtering and decoupling capacitors not only improved the transmission of pulses, but also improved the stability of the amplifier.

  Another not the best idea was to use a pair of tetrodes in the output stage in a triode connection with a common cathode resistor and without capacitive decoupling. This arrangement significantly limited the output power and good high frequency transmission. A much better circuit was the ultralinear amplifier, proposed in 1951 by David Hafler and Herbert Keroes, allowing to obtain significantly lower distortions with similar output power and global feedback parameters. Fortunately, the ultra-linear power stage can be perfectly matched with the rest of the Williamson amplifier stages.

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

  • Topology based on modified versions of Williamson amplifiers.
  • Load split phase inverter.
  • Power stage in an ultralinear system.
  • A simple power supply that ensures a "soft start" of the amplifier.

In the further part of the description, individual parts of the Concertino amplifier will be discussed.

Amplifier input circuit

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

  In the original Williamson amplifier diagram and practically all its later versions, the input stage uses an electron tube in a common cathode system with local feedback and a global feedback signal fed to the cathode. The output signal from the anode goes directly to the input of the phase divider in the distributed load system. In order for the phase divider to fulfill its function properly, its grid must have a relatively low voltage (25 ÷ 35% of the supply voltage of this stage, i.e. usually 90 ÷ 100V). This causes very unfavorable working conditions for the first vacuum tube, especially if it is the 6SN7 tube, which works much better with anode voltages in the range of 150 ÷ ​​250V. The article "Williamson Tube Amplifier Modifications" gives a way to improve this situation by changing the operating points and supply voltages of both tubes.

  The phase inverter with split load, although it maintains the perfect symmetry of both output waveforms, has a very low coefficient of supply ripple suppression and a significant difference in the output impedances of both circuits. The second disadvantage is effectively eliminated in the third stage of the amplifier (power stage driver), while the low PSRR factor can be corrected by appropriate modification of the first two stages of the amplifier.

  The solution to the problem was described by John Broskie in the TubeCad Journal (April 1999) and used in his "Aikido" amp series. The low PSRR of the phase separator was treated here as an advantage and contributed to a very good suppression of the ripple of the power supply of the first two stages of the amplifier. The idea is to deliberately introduce disturbances into the circuit in order to obtain an undisturbed waveform at the output.

  If the operating point of the first ellectron tube is set so that the DC voltage at the anode is half of the supply voltage of this stage, the network ripple suppressed by 6dB (half amplitude) will appear at the output. The ripple will transfer through the capacitor C1 to the grid of the second electron tube and will appear in coincident phase at the second stage cathode and in opposite phase at the second stage anode. But on the anode of this system there will also appear ripples flowing directly from the power supply in the compatible phase. The two waveforms, when summed, will give the anode a waveform similar in phase and amplitude to the signal on the second stage cathode. The result is not a differential waveform, but a common waveform that will be suppressed in the next amplifier stages. Of course, the system suppresses not only ripple, but also any interference from the power circuit.

  The figure above shows a simulation of the network ripple suppression of the first two amplifier stages (all waveforms are shown without the DC component). In order for the final result to give a measurable waveform, the ripple amplitude of the power supply was exaggerated (60V) and the filtering capacities were radically reduced. The figure shows the first stage power supply ripple amplitude 0.95V. The differential signal at the output of the second stage does not exceed 37mV, so the network ripple suppression reaches the value of 28dB. For comparison, the damping of the ripple of the input stages of the traditional Williamson circuit is only a dozen or so dB.

  The system described above has two further advantages. Resistors R4 and R5 allow you to conveniently set the operating point of the second vacuum tube, which will make it possible to obtain the maximum output amplitude of the phase inverter (usually R4 = R5 is set). These resistors can be very large (in the order of a few megohms).

  If the values ​​of R2 and R6 are the same, the first two stages will form a circuit with a constant current consumption (the current flowing through the resistor R9 does not change). This condition will guarantee the maximum stability of the power supply of 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. The potentiometer in the cathode circuit of the first stage of the amplifier allows to minimize the differential mains ripple signal measured between the outputs of the second stage (which should occur at the anode potential of the V1B tube equal to half the supply voltage of the first two stages. The gain of the first stage is about 10 without closed feedback loop and just over 2 with feedback A double ECC82 triode is used in the first two stages.

Amplifier control stage

   In the third stage of the amplifier - the driver of the final stage - there are no major surprises. It is a differential amplifier with a common cathode resistor. Adding a capacitor parallel to the cathode resistor will not substantially change the operation of the system, but will increase the stage distortion unnecessarily (deterioration of the stage symmetry).

    The choice of a tube for this stage may be a problem. The 6SN7 electron tube proposed by Williamson loads the previous stage with a relatively large input capacity (on the order of 70pF). For a low-impedance phase inverter cathode circuit, this is not a problem, but for an anode circuit, it reduces the high-frequency gain. Although the driver copes with this problem quite well, it is worth considering the use of a vacuum tube with a smaller grid-anode capacity.

    The ECC82 is such an electron tube. However, it has another major disadvantage. The control stage must be able to amplify the waveform to the value of 30V ÷ 35V. At such amplitudes, the ECC82 electron tube generates considerable distortions, about 2.5 times greater than that of 6SN7. Therefore, a poorly designed Williamson amplifier can achieve distortions of 5% ÷ 10% without a feedback loop before clipping the waveform (and most of these distortions come from the driver stage).

   The 6SN7 electron tube shows its advantage even for much smaller amplitudes. In the end, I decided to use the 6SN7EH electron tubes (the equivalent of the old RCA 5692 with a red plinth). The control stage built on these electron tubes has a gain of about 16..

   In many common schematic diagrams the value of the cathode resistor R16 is too small (e.g. 220Ω or 390Ω), which unfavorably determines the operating point of the electron tube (Ugk = -2V ÷ -3V). Since the control stage must be able to supply the tubes with a voltage power of about 70Vpp, the grid voltage of the 6SN7 tubes will vary by more than 4Vpp. In order for the grid of these tubes not to fall within the range of positive voltages, it is safer to set Ugk = -4V or less..

   In some amplifiers (also in the original Williamson amplifier schematic), the supply voltage is applied to the anode resistors R17 and R18 through an additional potentiometer, which helps to adjust the symmetry of the circuit for the AC waveform. In practice, the circuit copes well with slight waveform asymmetry and the use of two equal anode resistors is perfectly sufficient.

The output stage

  The power stage of the Concertino amplifier will be made in an ultralinear system according to the recommendations of Hafler and Keroes from 1951.

  Power electron tubes in the ultralinear system will achieve about 75% of the power compared to the tetrode / pentode system and at least twice as much power as the triode system with comparable operating parameters. The sound obtained in ultralinear systems, due to the low distortion and better damping coefficient, resembles the sound obtained from triode amplifiers. The ultralinear circuit is perfect for Hi-Fi amplifiers, and less often used in guitar amplifiers.

  The recommended electron tubes for ultralinear systems are usually PL36, 6V6GT, 6973, 5881 / 6L6GC / 7591, KT66, KT88 / KT90, 813. EL34, 6CA7 and EL84 pentodes give a slightly less clean sound, although they are also often used. My choice fell on the 6L6GC beam tetrodes currently produced by JJ Electronic.

  Another issue was the choice of the polarization system of the power tubes. Fixed bias would improve the efficiency of the amplifier and make it easier to adjust the quiescent current of the power stage. The economy of the system is practically non-existent, because a negative bias voltage generation and regulation system should be built. The auto bias circuit is easier to implement and is considered by many to sound better. Electron tubes with automatic polarization more gently enter the clipping range, which may be important in loud listening sessions and in ineffective loudspeakers.

  The bias circuit of the output stage tubes that I used has a simple but effective mechanism of the symmetrization of the quiescent current, found both in the original Williamson and in later designs, e.g. by Heathkit. The electron tubes work in class AB1 (up to about 10W in class A). The value of the resistor R25 is used to set the quiescent current of both electron tubes (the sum of the cathode currents is about 120mA, which at an anode voltage of 430V will generate about 23.5W of rest power in each electron tube). The R32 wire potentiometer equalizes the quiescent currents of both electron tubes (the cathodes should have the same potential, which is a sufficient approximation of the condition of the same anode current of the tubes). Blocking capacitors in the cathode circuit of the power tubes may slightly affect the level of distortion of the output stage (which way - 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 carry the band 2Hz ÷ 60,000Hz. It is not easy to get a transformer with the right parameters (many of the audio transformers sold have been designed with guitar amplifiers in mind). In the end, I decided not to risk and buy the long-proven TG36 toroidal transformers, sold by Amplifon and used in their WL36 and WL25 amplifiers in a very similar arrangement. The transformers have multi-section windings, ultralinear taps and a nominal impedance between the anodes of 6.6kΩ (therefore they are suitable for cooperation with 6L6GC electron tubes).

  The amplification of this stage in the ultralinear system (tap for 43% of the primary winding) is about 9. The amplifier should easily deliver about 25W of power per channel. So there should be no problem with driving my small monitors with an efficiency of around 87dB/W/m.

Power Supply

  Currently, making the power supply with the use of a simple mains transformer, a semiconductor bridge and a bank of filtering capacitors seems to be the simplest and the most appropriate solution. However, such a system generates high switching noise and requires the use of an additional delayed anode voltage connection. So why not turn to old proven techniques and solve several problems at once.

  So I will use a tube rectifier with an LC filter. Directly heated rectifiers are not suitable - here, too, a delayed start system would be needed (the lack of such a system was a serious drawback of Williamson's original design). Among the indirectly heated tubes one can choose, for example, 5AR4 / GZ34. However, one electron tube will not handle both channels - two must be used so that the permissible operating parameters are not exceeded (270mA of average current consumption and about 1A of peak consumption).

  Reading Lynn Olson's pages came to the rescue again. Like Olson, I will use two 6D22S suppression diodes to rectify the current. Each has only one anode, so two tubes are needed for a full-wave rectifier. They have their drawbacks: Magnoval pins (less accessible sockets) and cathode connection with a cap in the upper part of the lamp. Instead, they offer very low switching noise, low forward voltage (15V), high peak current (2A) and very long warm-up time (30s), thus solving the problem of the amplifier's soft start. The 6D22S electron tubes require a filament voltage of 6.3V. It is permissible to use the common 6.3V winding for all tubes in the system, but in my amplifier, the rectifying tubes will receive a filament current from a separate winding (with the possibility of forcing the cathode potential of these tubes on the winding).

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

  The mains transformer was made to order by the MKPT company (Telto version). It is a 250VA toroid giving the output 2 * 390V for the anode circuits, 6.3V for the glow of 6D22S electron tubes and 2 * 3.15V for the incandescence of the remaining tubes (center tap on the ground potential to minimize mains hum). If my estimates are correct, I should get an anode voltage of about 430V under load.

Schematic diagram of the amplifier

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

  Marking the signal mass depends on the place of its occurrence. This corresponds to the principle of mass distribution in a star system. All local grounds are connected at one point near the capacitor C36. Also at this point, the signal ground connects to the amplifier housing and the protective ground of the power supply.

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

  The Zobel circuit between the pins of the secondary windings of the output transformers improves the amplifier's stability at high frequencies (especially important when the load is disconnected).

  The amplifier circuit includes elements ensuring the stability of the amplifier in the supraacoustic range. These are the capacitors C3, C4, C7 and C8, and resistors R13 and R14. The value of the C7 and C8 capacitance should be selected experimentally when starting the amplifier (the criterion of minimizing the shifts and oscillations in the reproduction of the square wave).

  At the amplifier input I used a double 100kΩ potentiometer with linear characteristics. Together with the resistors R7 and R8, the approximation of the exponential characteristic is obtained (much better than in most so-called logarithmic potentiometers). The operation of such a potentiometer was described by Rod Elliott in the article "Better Volume Control". The figures below show the error of concurrency of both potentiometers measured by me and their characteristics. The error is minimized by selecting the values ​​of the resistors R7 and R8. At commonly used loudness levels (attenuation 60dB ÷ 20dB) the error of the modified potentiometer does not exceed 0.15dB. I have never had similarly good results when measuring factory logarithmic potentiometers. Potentiometers of the reputable ALPS company allow up to 3dB of the value concurrency error.

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


  This is where the real problems begin. Not every amateur electronics engineer has a sufficient mechanical workshop. I am able to do simple things like drilling, sawing and grinding myself at home. I will have to outsource the more complicated ones, such as sheet bending or punching holes for electron tubes.

  In my case, the housing will have to be adapted to the rest of the audio equipment, which means an integrated black construction with a width of 43 cm. At the same time, the housing must be simple and allow for convenient spatial assembly. Moreover, it cannot disfigure and cost a fortune.

  I decided to make the housing shown schematically in the drawing. The basic chassis will consist of a 2mm thick bent steel plate. The folds will form the front and rear panels. The sides will be made of varnished wood and permanently attached to the sheet. Access to the inside of the housing is possible thanks to a screwed bottom - a steel plate with a thickness of 1 mm. Tubes and toroidal transformers will be placed on top of the device and will require additional protective and masking elements. The remaining elements will be installed inside the housing.

  The 2mm thick steel plate forms the essential chassis measuring 398mm x 360mm and only 50mm high. The deflection of the sheet was not made exactly at right angles and has a relatively large radius, which, however, is not a defect with the assumed housing structure.

  Chassis after drilling and sawing. More than 100 holes have been drilled, although these are only the necessary ones (some of the elements will be glued).

  Chassis after powder coating and screen printing.

  Side panels in preparation. After a few attempts, I decided to use an "ebony" stain varnish.

  The assembled chassis creates a rigid and durable box.

Mounting the amplifier

  For the purpose of assembly, I built a special mount made of wooden elements, on which the chassis rests firmly, and which will allow me to run the amplifier upside down.

   First, I installed a steel angle inside the housing for additional stiffening of the structure. In its vicinity, the housing has to support about 8 kg of weight, which consists of two chokes and three toroidal transformers. Then I installed all the external elements (vacuum tube sockets, sockets and switches) and the power filter chokes. The above photo also shows most of the soldering connectors that allow for convenient spatial assembly of electronic components. The connectors with distance sleeves were glued to the housing with epoxy glue.

   After installing the transformers, the amplifier weighs almost 10 kg. Moving, lifting or turning such a heavy structure has become quite a difficult task from then on.

  "Stuffing" numerous transformer leads is the first serious assembly task. There is not much space at all and it was necessary to shorten the leads significantly.

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

  The power supply is assembled and ready for testing. In the center you can see the opposite ends of the aluminum tubes shown in the previous photo. Slightly to the left (near the choke) there are a few solder lugs. This is the central point of the amplifier ground - hence the ground will be distributed to other circuits. At this stage, the filament circuits for all tubes were also established.

  The temporary load of the power supply was 5 thirty-watt resistors with a total resistance of 1800 Ω. During the test, they should generate over 100W of power.

  Fortunately, the first power-up took place without any unwanted pyrotechnic effects. As expected, the use of 6D22S diodes ensures a long and smooth start of the power supply. The first volts on load appear approximately 15 seconds after the system is turned on. The voltage then rises gently until it reaches the target value after approximately 35 seconds.

  Two chokes and two large-sized capacitors, each with a capacity of 500 µF, constitute a very effective main power filter. Under test conditions at the filter input, a direct voltage of 462V is measured at 26Vrms ripple. At the output of the filter, at 439V DC, the mains ripple drops to a level lower than 0.5mVrms. These values ​​are fully consistent with the results of the simulation performed with the PSU Designer II program.

  After starting the power supply, it's time to install the power stage. There were few components to mount, but some of them have large dimensions, such as 0.47 µF blocking capacitors, 5-watt cathode circuit resistors and a 100 Ω wire potentiometer (in the upper part of the photo).

  The first turning on of the amplifier's output stage went without any problems. After a dozen or so seconds from switching on the power supply, the anode voltage slowly increases (the power tubes are already warmed up and load the power supply) and stabilizes after about 40 seconds. Due to the incomplete load of the power supply (no first stages of the amplifier), the anode voltage is a bit too high (ultimately it is to be around 430V). The value of the resistor R56 is selected to obtain the appropriate quiescent current of the power tubes. Two 470 Ω resistors connected in parallel (cathode current of each electron tube approx. 57 mA) proved to be suitable. The R5 wire potentiometer allows you to effectively equalize the quiescent current of both power tubes (equal cathode potential of both tubes).

  After connecting the loudspeaker, I could hear a slight hum from the mains. Measurement of the signal at the output of the amplifier gave a value of 0.8mVrms at a fundamental frequency of 100Hz. The photo on the right shows the waveform at the transformer output. Further tests confirmed that the cause of the hum does not lie in the arrangement of the amplifier components. The change in the configuration of the filament power supply circuits (including various ways of symmetrization of the circuit) and the change of the ground routing did not affect the level of interference at the output.

  After removing the output tubes and starting the amplifier with an artificial load, it turned out that the signal at the output was still inducing (the effective value decreased to 0.5mV). This clearly shows the magnetic coupling between the output transformers and the mains transformer. Placing a simple steel sheet partition between the transformers significantly reduced the mains hum. Also, changing the mutual position of the transformers significantly reduced the hum, but ultimately I would prefer to avoid this method of noise elimination. The solution to the problem will be the use of magnetic shielding of transformers (probable hum reduction by about 10dB) and the use of global negative feedback (hum reduction by a dozen or so dB). Then the hum level should not be a problem even when using loudspeakers with high efficiency.

  In the third stage of construction, the control stage was assembled. The photo shows just a few resistors and capacitors that make up this stage. On the left side you can also see the power filter of this stage (4.7kΩ resistor and 56µF capacitor). The launch of this amplifier stage did not bring any surprises.

  As you can see in the attached drawing, the supply voltage of the stage is slightly higher than the nominal one (350 ... 360V) due to the lack of load of the RC filter with the voltage stage (not yet assembled). Hence, the quiescent current of the control stage is slightly higher than assumed, which, however, does not affect the proper operation of the system. The quiescent current and the anode voltage of both branches are not the same due to the divergence of the parameters of both triodes. Since this is a differential amplifier configuration, it is not possible to equalize these currents without breaking the symmetry for the variable component. Step operation for an alternating signal is normal (perfect symmetry). The measured voltage gain is 17 (slightly more than previous estimates showed).

  In the last stage of starting the amplifier, the voltage stage and the phase splitter were assembled and started. At the top of the photo you can see a potentiometer for adjusting the quiescent current of the first triode (to obtain an anode voltage of exactly half the value of the first stage supply voltage).

  The figure shows the voltages and currents at individual points in the system. The elements compensating the amplitude-phase characteristic (C3 and R13) will be selected only after closing the negative feedback loop. The operation of the system for the variable component of the signal is correct. The measured voltage gain of the first stage is 9.78 and the phase splitter is 0.87 (in each branch). Due to the very good filtering of the supply voltage (mains ripple is not measurable), it is difficult to observe and measure the desired performance of the Aikido amplifier (operating principles can be found on the "Input stage" page). Perhaps the circuit will later be simplified to the traditional Williamson design (comparative testing will be needed). The whole circuit tested in open loop generates low noise at the output and a mains hum of 1.25Vrms (audible in the loudspeaker from a distance of about 30cm). Both types of distortion will be reduced by applying global feedback. It is worth noting that the mains hum remained at the level measured immediately after the activation of the power stage. This indicates that there are no additional hum sources in the input and control stage.

  The figure below shows the voltage values ​​of the test signal in various places of the amplifier circuit (green), the gain values ​​of individual stages (blue) and the voltage values ​​supplying individual stages (red). The gain of the open loop amplifier is 92.5.

Amplifier startup

  It is often assumed that for ultra-linear systems it is sufficient to cover the amplifier with a global feedback loop with a depth of several dB. It is worth remembering that the implemented system has a number of local feedbacks (in the input stage, phase divider and in the circuit of the shielding grid of the power stage), which reduce signal distortions even without the use of global feedback. However, global coupling is needed, for example, to lower the output impedance of the amplifier.

  At this stage, I assumed a global feedback with a depth of 16dB, which with an open loop gain of 92.5 will ensure full control of the amplifier after giving the input signal with an amplitude of about 1.35V. With a resistance in the first stage cathode circuit of about 600 ohms, a 10k ohm feedback resistor will be needed. Frequency compensation elements are also included, the target values ​​of which will be selected at a later stage of starting the amplifier.

  After connecting the feedback loop, I did not find any oscillations in the acoustic or supraacoustic band. Mains noise and hum decreased to a barely audible level, which, however, was impossible to measure. It turned out that the amplifier is unstable below the acoustic band. The output level waved irregularly in the range of about 200mV with a peak frequency of 1Hz ÷ 2Hz. This "undulation" did not affect the ability to transmit the acoustic signal, and at the same time controlling the amplifier with the signal did not affect the amplitude or frequency of the waveform. Before further measurements, the cause of this instability had to be eliminated.

  The lower limit frequency of the transformer measured in the system with the output power of 1W is about 5Hz (during tests below the frequency of 6Hz, the output waveform already had visible distortions resulting from saturation of the transformer core). It is the dominant pole of the system. The next three poles of the circuit come from the RC elements coupling the amplifier stages and lie around 1.5Hz. Around 1Hz, the phase difference between the amplifier output and input reaches 180 degrees with still high open loop gain. This causes the system to oscillate around the frequency of 1 Hz irregularly. The solution to the problem will be the mutual separation of the poles and the reduction of the open loop gain for frequencies less than 16Hz. I made the following modifications to the amplifier:

1. Change in the value of the control and power coupling elements (R42 = R43 = 220kΩ, C17 = C18 = 0.047uF). This establishes a new dominant pole for 16 Hz.

2. Adding a 10uF capacitor to the input circuit. This allows for efficient operation of the Aikido system for frequencies lower than 1Hz (cut-off frequency 0.016Hz) and makes the pole of this degree irrelevant.

3. Changing the value of the capacitance of the C25 capacitor in the first stage power filter from 22uF to 100uF. This reduces the slow-varying power float around 1Hz (new filter cutoff frequency 0.16Hz).

  The change no.1 is essential to ensure the stability of the system. The reduction of the time constant of RC elements coupling the output stage also has another desired effect - faster recovery of the amplifier from the overload state (when the voltage on the control grid exceeds the cathode potential and the capacitor C17, charged with the instantaneous grid current, must then discharge through resistor R42).

  In many series-produced tube amplifiers, the cut-off frequency of the last RC unit was set quite high: 7Hz (Altec Lensing, Audio Innovations, Heathkit, Jolida) or 16Hz (Eico, Grommes). This provided sufficient low frequency stability for systems with two or more capacitive coupling stages. On the other hand, many Williamson circuits published on the Internet (including the well-known scheme from Practical Electronics) certainly do not provide sufficient stability below the acoustic band (at least if the output transformer used has slightly worse parameters than the original produced by Partridge).

  The figure above shows the amplitude and phase characteristics of the Concertino amplifier in an open loop (for frequencies <30Hz). Using 16dB deep global feedback, I got about 45 ° phase margin and 8dB gain margin. After stabilizing the circuit, I was finally able to measure the noise level at the output of the amplifier. The meter showed about 0.2mVrms. After performing the test of shielding the loudspeaker transformer, the noise level decreased to 0.1 mV. The mains hum was practically inaudible even with the ear placed directly against the loudspeaker.

  The time has come to judge whether the Aikido input circuit has a practical advantage over the typical Williamson input stage. To reduce errors, I took measurements simultaneously (Aikido circuit in one channel, Williamson circuit in the other, switching channels during the tests).

  In all tests, the Aikido circuit showed its superiority by giving a measurement result in the range of 0.15 ÷ 0.22mV, while the Williamson system in the range of 0.24 ÷ 0.50mV (the result was always 2 ÷ 8dB worse than that obtained in the adjacent channel). When using transformer shielding, the noise value decreased to about 0.115mV for the Aikido system and 0.175mV for the Williamson system. These differences unequivocally decide on the sense of using the Aikido system.

  The Williamson amplifier will probably also require appropriate compensation in the high frequency range. I estimate the limit frequency of the loudspeaker transformer used by me at around 70kHz. This is the lowest pole above the acoustic band. The next poles come from the "upper" half of the control stage (110kHz), the input stage (800kHz), the power stage (1.5MHz) and the "lower" half of the control stage (2MHz). With proper compensation, the last three should not be important for the stability of the system.

  Without compensating elements, a loop gain A * b of 1 can be expected somewhere around 200kHz and a phase shift of about 150 °. This should ensure the stability of the amplifier with an attached resistor load and perhaps also without an attached load (then the role of the load for high frequencies is played by the Zobel circuit connected to the output). Tests showed the true stability of the amplifier with a resistor load and relative stability without the load connected (the amplifier oscillated when driven by a signal).

  However, it should be expected that the target load (loudspeakers with an electric crossover at the input) will require a much larger margin of stability. Connecting a loudspeaker to the amplifier's output resulted in oscillations with a frequency of less than 200kHz. An equally disturbing effect was created after connecting the load in the form of a 0.22uF capacitor - fourteen-millisecond oscillations of 185kHz frequency with a nine-millisecond gap between them. The amplifier undoubtedly requires frequency compensation for stability regardless of the type of load connected.

  Elements R13 and C3 introduce delay compensation in the range above the acoustic band. With the values shown in the figure, the circuit creates a new dominant pole for f = 23kHz and zero for f = 110kHz. The next pole is at f = 70kHz where the open loop gain drops to about 20 (26dB) and the phase shift is around 120 °. The third pole, at f = 110kHz, is eliminated by the zero coming from the compensation circuit. In this way, the drop in characteristics maintains a steepness of 12 dB / oct all the way to the fourth pole, around f = 800kHz. Slightly above the frequency of the other pole (about 90kHz) is the point for which loop gain A * b = 1. The phase shift at this point is approximately 130 °. A phase margin of 50 ° should ensure the amplifier's unconditional stability.

  The accelerating compensation shown in the figure influences the transmittance [b] function of the feedback circuit. The value of C7 = 56pF puts the pole for f = 130kHz into this function and zero for f = 11MHz. This compensation is not necessary as the lag compensation has already ensured sufficient system stability. However, it is indicated due to the "acceleration" of the feedback loop for higher frequencies and the limitation of metastases and oscillations in the reproduced pulses. However, too high a capacitor value may destabilize the amplifier.

  In practice, because the phenomena occurring in the system are a bit more complex, the values ​​of the amplifier's compensation elements should be experimentally verified to obtain the required system stability. I used the method described many times by Patrick Turner on the newsgroup.

  In the first step, with the assumed value of the C7 capacitor in the feedback circuit (currently selected value of 47pF), the C3 capacitance of the lagging compensation circuit should be selected. When the amplifier is loaded only with a capacitance in the range of 10nF ÷ 4.7uF, a peak frequency depending on the connected load is obtained on the amplifier's transfer characteristics. The capacitance of C3 must be large enough to ensure that this peak never exceeds + 6dB in relation to the nominal level (measured at 1kHz), and that in the acoustic range (f <20kHz) the transmission characteristic does not differ by more than 1.5dB from the nominal. These conditions were met by the C3 capacitor with a capacity of 680pF (the maximum measured peak equal to + 4.77dB at f = 71kHz and the 1uF capacitor connected to the output). Using a value of C3 = 680 pF limits the open loop bandwidth of the amplifier to f = 17kHz (measured value). Higher capacitance will contribute to even better stability of the amplifier circuit, but it will reduce the feedback effect where it is still needed (below 10kHz).

  In the second step, the resistance value R13 of the lag compensation circuit is selected. Find the maximum resistance value at which the amplifier does not oscillate regardless of the load attached (nominal resistive, capacitive, inductive, no load). The tests should be performed in the absence of a signal and by controlling the amplifier with a square wave signal of different amplitude. In my case, the maximum value of R13 is 4kΩ.

  When looking for the maximum value, it is also checked for which R13 value the optimal shape of the transferred waveform occurs (minimum pulse transfer, minimum oscillations, maximum slope). Ultimately I decided to use R13 = 3kΩ.


8Ω load; f = 4800 Hz; 1V / div; 50us

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

  In the last step, it is checked for which value of the C7 compensating capacitor a good damping of oscillations in the envelope of rectangular pulses (so-called hum) is obtained. One should be careful here, because excessive increase of this capacity may cause the loss of the amplifier's stability under certain operating conditions. If the specified C7 value is significantly different from the previously assumed one, the permissible value of the R13 resistor should be verified again. All of this is easier to describe than to do. The entire procedure is laborious but leads to a good margin of stability in the amplifier. We get an unconditionally stable amplifier that:

  • does not oscillate without connected load,
  • does not oscillate with the load in the form of a coil of any value,
  • does not oscillate with the load in the form of a capacitor with any value in the range of 0.01 ÷ 10uF,
  • does not oscillate with any of the above loads, driven by a rectangular signal,
  • when driven by a sinusoidal signal with a frequency of several Hz and an amplitude sufficient to saturate the output transformer, it 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 is still stable. In my case, the feedback resistor can be reduced down to 1.6kΩ with no signs of oscillation at the amplifier output. This gives a feedback loop depth of 28.3dB. It can therefore be assumed that the amplifier with a resistor load has a sufficient gain margin of 12.4 dB.

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

Preamplifier buffer

  The preamplifier buffer plays an additional role, not related directly to the tube amplifier circuit, and therefore it is not included in the main diagram. The function of the buffer is to separate the regulated signal output intended for the external subwoofer from the input circuit of the first amplifier vacuum tube. It is the only piece of the circuit that contains semiconductor components (and while I see nothing wrong with using them, the main amplifier circuit remains free of them, to be as close as possible to the circuits used half a century ago).

  The signal from the volume potentiometer slider is fed to the input of the non-inverting amplifier, which, due to the high input resistance in this configuration (about 1e12Ω), guarantees that the buffer does not affect the input signal of the tube amplifier. The buffer gain is around 16dB. With the input signal with an amplitude of 1.6V (the maximum signal not causing the tube amplifier to overload) we get a signal with an amplitude of 10.3V at the output of the buffer, so it is within the operating range of the operational amplifier with a supply of ± 12V.

  The system is powered by an additional small network transformer with a secondary voltage of 2 x 12 V. As the system has only a few elements, it was assembled on a piece of universal PCB.


Unless otherwise noted, measurements were made with an 8 W resistive load without transformer shielding.

  • Układ: Williamson; Aikido front end; ultralinear push-pull; klasa AB1
  • Nominalna moc wyjściowa: 2 x 25W (f=1kHz sinus; THD=0.21%)
  • Maksymalna moc wyjściowa: 2 x 32W (f=1kHz sinus; THD=1%)
  • Pasmo mocy:
    • 7Hz÷78kHz (P=0.2W sinus; ±3 dB; 0dB dla f=1kHz)
    • 7Hz÷75kHz (P=1W sinus; ±3 dB; 0dB dla f=1kHz)
    • 10Hz÷68kHz (P=5W sinus; ±3 dB; 0dB dla f=1kHz)
    • 17Hz÷60kHz (P=25W sinus; ±3 dB; 0dB dla f=1kHz)
  • Nierównomierność charakterystyki przenoszenia: ±0.1 dB (f=20Hz÷20kHz; P=1W)
  • THD dla f=1 kHz
    • 0.03% (P=0.2W sinus; f=1kHz)
    • 0.03% (P=1W sinus; f=1kHz)
    • 0.08% (P=5W sinus; f=1kHz)
    • 0.21% (P=25W sinus; f=1kHz)
  • THD dla f=20Hz÷10kHz
    • <0.05% (P=0.2W sinus; f=20Hz÷10kHz)
    • <0.1% (P=1W sinus; f=20Hz÷10kHz)
    • <0.2% (P=5W sinus; f=20Hz÷10kHz)
    • <0.6% (P=25W sinus; f=20Hz÷10kHz)
  • Poziom szumu i przydźwięku sieciowego na wyjściu
    • <0.2mV (97dB poniżej poziomu nominalnego; bez ekranowania transformatorów; odczep 8W)
    • <0.1mV (103dB poniżej poziomu nominalnego; z ekranowaniem transformatorów; odczep 8W)
  • Impedancja wejściowa: 47kΩ (f=20Hz÷20kHz)
  • Nominalna impedancja obciążenia: 4Ω lub 8Ω
  • Czułość wejść: 0.95Vrms sinus (P=25W)
  • Wzmocnienie napięciowe: 14.83 (odczep 8Ω)
  • Współczynnik tłumienia: 3.3 (oszacowany)
  • Globalne sprzężenie zwrotne: 15.9dB

Power bandwidth measured at 4 different levels. The asterisk in the graph indicates the point at which the distortion level due to saturation of the output transformer increases sharply.

Unevenness of the transmission characteristics in the acoustic band. The 0dB level corresponds to a power of 1W on an 8Ω resistive load.

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

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

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

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

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

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

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

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

The spectrum of intermodulation distortions (f1 = 1kHz, f2 = 1.1kHz).

The spectrum of the signal at the amplifier's output when driven with a sine wave (f = 1kHz, P = 1W).

The spectrum of noise and distortions at the output of the unregulated amplifier (unweighted measurement).

The spectrum of noise and distortions at the output of the unregulated amplifier (weighted measurement - ANSI A).

The effect of a slight distortion of the amplifier when driven with a sinusoidal waveform with a frequency of 20Hz and a large amplitude and a waveform with a frequency of 1kHz and a small amplitude. The amplifier shows no signs of clogging. The input level is 113% of its maximum value not to distort the amplifier.

The effect of a strong distortion of the amplifier with the input waveform as in the picture above. The amplifier shows signs of clogging for no more than half the period of the signal waveform. Input level is 145% of maximum value not to distort the amplifier.

Links and gallery

Main resources used in the design of the amplifier:

Equipment and software used during the measurements:

  • multimetr cyfrowy
  • oscyloskop 2-kanałowy 50 MHz
  • generator przebiegu sinusoidalnego i prostokątnego 1 Hz - 200 kHz
  • Yoshimasa Electronic Inc. - DSSF3 Realtime Analyzer
  • Audua - Speaker Workshop
  • Sintrillium - Cool Edit Pro (obecnie: Adobe Systems Incorporated - Adobe Audition)

Marcin Sławicz

(Materiał opublikowany na w 2005r .)