The D.T.N. Williamson Amp.
downloaded from http://www.xs4all/~ideas/amps
Introduction The Williamson amplifier is a design by the British engineer D. T. N. Williamson and was first published in the spring off 1947 in "Wireless World". Williamson was employed by "The M.O. Valve Company" (later with Ferranti ) as an engineer. The M.O. Valve Comp. was one of the constructors of the famous KT66 tube. The amplifier Williamson designed employed a pair of those tetrodes connected as triodes in a push-pull class A configuration and had a max. output power of aprox. 16 Watt. What made this design famous was its very low distortion. After war Europe had other things to do at that time and High Fidelity seemed a luxury. When the circuit design reached the mainland years later constructors reacted disappointed because the circuit was so simple. The high quality was the mere fruit of careful design and a very complicated outputtransformer. After a short period of popularity may constructors turned to the so called "Ultra-Linear" amplifiers which were invented a few years later. These gave more power with less tubes and seemed to have the same sound quality as the Williamson design at that time. We must consider that parts for tube amplifiers were very expensive at that time. A good outputtransformer for the Williamson Amplifier would cost a weeks wages or more and one must not be surprised that constructors would carefully compare every design in the financial aspect rather then looking at the last little bit of joy a triode class A amp. would bring.
Nowadays building tubeamps is a costly undertaking anyway and the reason people do this is merely because of the last little bit of joy that can be achieved when one looks carefully at every aspect of the amplifier, never mind the extra tube or that rare transformer. When one doesn't need an awful lot of output power the Williamson design is a construction one must consider since it is sounding very good and at the low levels of listening at home ( at night ) certainly better as comparable constructions of the Ultra Linear type. When one however wants to build a tubeamp. on a very tight budget one must be very patient and spend a year or so collecting the parts, when ones main diet is vinyl one can better start by building a good tubepre-amp. since that gives the greatest improvement for the money.
Is this amp still any good? Is it worth to build one? Are more recent circuits not much better? Williamson designed his amp for mono reproduction with a fairly high amount of negative feedback. That is ideal for mono. All designs in that time were mono-amps. Later, when stereo was invented, they just build two monoamps on the same chassis and claimed that it was a stereo amp. Unfortunately stereo amps must sound roomy and it is just not good enough to copy mono amps for stereo use. In most cases however decreasing the amount of feedback makes an amp better for stereo use. The Williamson design is still working perfectly with very little feedback because of its careful design. Later commercial designs are often developed for maximum power output with minimum parts.
It is often claimed that the ultra lineaer circuit gives the powerstage the quality of a triode-connected powerstage without the disadvantages of the triodes. In fact this is not really the case. The internal resistance remains twice as high at the same level of negative feedback, the low power distortion is much higher but it saves one tube since it is easier to drive. The typical advantage of a triode-connected powerstage such as the Williamson is the low internal resistance that provides with low negative feedback enough damping for the loudspeakers. Striking results can be expected with speakers that require not that much damping from the amp. such as transmissionlines. Enclosures with the size of shoeboxes require in most cases a lot of damping from the amp. The result of high negative feedback levels can be described as the typical "Hear me sounding good and not forgetting any detail" sound of many amps that makes listening tiring.

The Circuit

The picture at the top of this page shows the overall circuit of the revised version as published in 1949 in "Wireless World" and indeed it is simple! It consists of three stages, Note that the values of C3 and C4 are the values from the "Ultra Linear Williamson" that was published much later.
The input stage combined with the "concertina" phasesplitter using tubes as 6SN7, 12AU7 orECC82.
The driver stage. In the original version a 6SN7 was used. For those who are interested in the experiment: 12BH7 and 12AU7 or ECC82 .
The power stage using in the original version a pair of KT66 that was impossible to obtain. Rumors are that it is in the Far East in production again. Alternatively try the latest version of "Sovtek's" 6L6WXG or 5881 that are capable of handling the dissipation. The slim 6L6GC versions that are in fact 6L6GB wont handle the current! The EL34 or its US counterpart 6CA7 can also be used. American constructors tend to prefer the foreign EL34 and the Europeans vice versa.
More important then looking at the three individual stages that are quite simple is the fact that those stages together make the amplifier that became famous! Designers took ample efforts to compose an amplifier out of succeeding stages so that the distortion in one stage reduced distortion in the next stage. That is why some circuits became classics!
Every change in the circuit can turn a masterpiece into a disaster if you don't know what you are doing. On paper is exchanging the 6SN7 for the more modern 12AU7 modest surgery. In reality you increase the bandwidth of the phasesplitter because the capacitve load of the 12AU7 on the phasesplitter is only half. This alone can cause instability!
The main reason the Williamson amplifier became famous was the output transformer. This was a piece of workmanship that was not seen before. Williamson had calculated that its bandwidth should range straight from 2Hz to 60.000Hz (being the third harmonic of 20.000Hz.)! Only the best transformer possible would allow the amp to operate stable with a strong negative feedback loop. Nowadays it is a very common to design amplifiers so that they are straight to 20kHz without feedback.
The Input stage

The input stage looks simple but was fairly new when presented by Williamson. The easy to obtain 12AU7 does the job just as good as the 6SN7 when the current is adjusted by a somewhat increased R4. The absence of an input capacitor means that the input signal must have no dc component. Check that. On many amps an suppresser resistor is connected between R1 and the grid of V1a to avoid instability (Typical value 10K) when however stability is not a problem such a suppresser should be avoided. R4 is not partially decoupled by a capacitor like in other amps and enjoys a lot of current feedback through R4 and as a result not much amplification (aprox.12X without the feedback loop but very low distortion.) On the plate side of the tube there is the direct connection to the grid of V1b which dictates the tensions and currents of the latter. C10 and R26 play an important role in controlling the supersonic characteristics and are discussed in the feedback loop section. V1b is connected as a cathodefollower because of the unusual high value of R5 it has full current feedback and therefore no amplification, no distortion and hardly any grid/plate capacitance.
A cathode follower could better be called a grid follower since the cathode follows the grid as long there is any positive tension on the plate side. The voltage swing on the plate side of the tube is the mirror image of that gallant knight of the grid, the cathode since R5 and R7 are twins. That means as long as capacitance plays no role. The true fact is that the impedance on the cathode side of the phase splitter is only 1kOhms and on the plate side it is 22kOhms, which is still very low! Around 100k cycles ( depending on the grid/plate capacity of the following tube ) the output on the plate starts to drop.
Unfortunately all phase-splitters have their disadvantages. Some engineers suggest that a 22K resistor on the cathode side is a good idea, fortunately the rest of the circuit can cope with the imperfections of the phase splitter in a very forgiving manner. At normal frequencies the phase splitter is, unlike many other circuits, very symmetrical.
It is sometimes discussed that feeding the two tube systems in this stage from two HT sources is wrong and can cause low frequency oscillation. Feed them from one HT source is also wrong since the two systems are directly coupled and must therefore be allowed to adjust themselves to their own tensions and currents. Perhaps it is the best two allow the two tube systems to adjust themselves to 320v by adjusting R4 and R6 slightly and then, when voltages are identical, connect the hot sides of the two smoothing caps. Remember that altering R25 may alter all the tensions and currents in the circuit.
The Driver stage
The driver stage seems pretty uninteresting, tough at a second glance, you may be tempted to think that decoupling R10 with a capacitor of say 100mfd is a good idea! Or maybe that cap is not there because of current feedback? Two times no! This peculiar construction is the backbone of the Williamson amplifier. There is no feedback through this resistor and amplification will not be significantly increased by decoupling it. In fact there is very little ac activity over this resistor and yet it decreases the distortion in this circuit enormously
The two tube systems seem to correct their unlinearities and the shortcomings of the previous stage. When you look at it any longer, this stage has some similarities with the so called "long tailed pair" phase splitter. Still it has its own short comings! One of them is the grid/plate capacitance of the 6SN7. At an amplification rate of aprox. 12 to 14 thanks to the Miller Effect it augments to 70pF on both inputs. 70pF on the cathode follower output doesn't mean a lot but on the plate side of the phase splitter it is significant.
Though this stage tends to correct this mistake it is worth considering a tube with less grid/plate capacitance! The 12AU7 for example does the same job at less then half the capacitance! But the question is then, "is my Williamson still a Williamson?" On the output side of this circuit the same problem arises, the Miller Effect of the output stage which is as large ( or larger ) as at the input side but the output impedance of this stage with R11 and R13 at 47kOhms it influence on the amplifier is much more significant and makes it hardly possible to make the amp run straight to 20.000cycles when no feedback is applied. Since the input capacitance of a 12AU7 is less then half of a 6SN7 it seems a good idea to double the tubes like in the drawing left, at the same time reducing R11 and R13 by half ( think about their dissipation !) and enjoy the double bandwidth ! The good news is, "It works!" This way you can drive two pairs of power tubes, reducing the necessary plate to plate impedance of the output transformer to 5000 Ohms, having twice the output power and still be close to the original Williamson time constants in the circuit. As long as you pay for the electrical energy it requires, you have my blessing.

More of a in between solution seems the use of a 12BH7 for this stage, reducing R11 and R13 to 33kOhms and adjusting the current through the circuit by R10 to 8mA. This solution sounds best.
The Powersstage
The drawing below reassembles the power stage. It works with better quality 6L6, 5881, El34 and KT66. It doesn't work with cheap phony imitations which are made by the same companies as the better ones that can often be recognized by their impressive vacuum envelopes. It has as little parts as possible. R21 regulates the dc current and R17 the dc balance. It is in important that all components connected between the grids 1 and the ground are first quality and checked thoroughly since malfunction draws to tube (s) past its limits.To measure the current through each tube, first check the dc resistance of each primary coils, then apply Ohm's law to calculate the tension needed over these coils for correct adjustment. For the resistors in the cathode circuit, take types that can handle 4 or 5 watts! The drawing below shows why a plate to plate impedance is your best choice.
It also shows that if your speaker impedance is low, decoupling the cathode circuit with a capacitor of 100 mfd is a good idea at parties! These lines are drawn for a pair of EL34 but the constructor of those tubes made an effort making the tube interchangeable with the KT66 and friends. The resistance to the cathode must be chosen a bit lower, round 235 Ohms. For EL34 change R16 and R18 to 47 Ohms, R23 and R24 can also be decreased to 47 Ohms with the EL34 but that doesn't make that much difference. Leaving those resistors out however causes the tubes to oscillate at RF. Note that the base connections of the EL34 have a separate connection for grid 3 where the KT66 and friends have a non connected (NC) connection.
The good thing about a decoupling capacitor that is not there is that you can't hear it!

The Negative Feedback Loop
The amount of overall feedback depends on the value of one resistor and determines almost all the amplifiers characteristics. This resistor has the shirt number 25. The amplifier is born without this resistor and may function very good without it tough bass reproduction may be far from ideal and some distortion and hum may be heard. Most engineers talk about feedback in decibels. The authors found it more practical to speak about the amount of times it reduces the sensitivity of the amplifier. In the initial sensitivity of the amplifier without feedback is 200mV and after applying feedback it is reduced to 600mV the feedback is factor 3. Williamson specifies 20dB feedback which means nothing else as 10 times.
The famous number 25 resistor returns the signal from the output of the amp to the input where it is compared with the input signal by subtraction. Any part of a signal that wasn't in the input is immediately corrected. By applying negative feedback the input stage has a reservoir of amplification it can draw from if anything goes wrong.
The factor of feedback determines how big that reservoir is and how much we allow the amplifier to correct things. We must not forget that the system has no build in intelligence and it will also try to correct things that cannot be corrected causing even more distortion. Apart from reducing sensitivity, it reduces distortion, increases the damping factor and widens the bandwidth of the amplifier.
It does all these things in more or less the same ratio as it reduces sensitivity . Sensitivity however is the easiest thing to measure! If you start experiment with feedback, make it a habit to measure always at the same frequency and to the same outputlevel. ( 400cycles and 3volt output are my favorites) Well if feedback is such a good thing, why not tons of it? Because all that correcting makes the amp nervous. Its is better to leave it to do its job at some extend the way it wants to do it. That sounds better. The 10 times Williamson used is in my humble opinion way to much.
A feedback factor of 3 ( or about 10dB ) is more to my taste. Distortion is low enough to be inaudible and the bandwidth sufficient. The only thing at feedback levels this low that can cause problems is the damping factor to the loudspeakers. Well, try any level of feedback by exchanging the horrible number 25 resistor, try values between 5kOhms and 100kOhms. If your amps starts oscillating at feedback factors under 20 stay away from high feedback levels till you found out why. If you are measuring in your shack with a dummy load resistor on the output, try every now and then what happens if you remove the dummyload, leaving the output only connected to the oscilloscope with some square waves fed in the amp.
If it starts oscillating when you remove the dummyload, chances are that it also oscillates when it is connected to your loudspeakers since the coils in your speakers have at an infinite frequency an infinite impedance, just like your dummyload when you remove it. Experimenting with feedback means you also have to keep a constant eye on the capacitor with the shirtnumber 10. Its value mentioned in the original circuit is only the correct value when everything including the output transformer are the same Williamson used! To adjust it to your amplifier, apply square waves to the input at 10Kcycles and adjust the value of C10 till your waves are almost square. A little overshoot may remain. If you alter the amount of feedback you must check C10 again. Some amplifiers have a kind of C10 like capacitor parallel to R25 with the same purpose, adjusting the square waves. ( In fact adjusting the phase shift ) Some other have both. If you want to spend a few years or so in your shack finding a way to make such a construction stable please do so! E-mail me sometime!
The Output transformer
The output transformer Williamson proposed was a bulky one. Its core was equal to the DIN E-I 150 that is used nowadays. The core is 150mm High (6'') and with the copper weighted 6kg. (13lbs). The core material was at that time a premium quality but to modern standards it is only standard grade. The reason Williamson used such a core was that he wanted to reach an initial inductance of 100H and a max. inductance of 600H. This can only be reached with a enormous core and 4400 turns with a standard grade core. To keep the leakage inductance within acceptable margins the primary is divided in two mirror-identical sections that are each divided in 5 sections of 4 layers of 88 turns of 0.3mm copper. In between those 5 primary sections are 4 secondary sections consisting of 2 layers of 44 turns of 1.2mm. ( 1''=25mm). All layer are insulated with 0.05mm paper and all sections are interleaved with 0.4mm insulating linen tape.
The drawing gives an impression of how things are connected.All primary coils are in series and the secondary can be connected in series and parallel thus giving the loudspeaker impedance's of 1.7, 6.8 and 15.2 Ohms. This transformer works! For those who want to try it themselves to make such a transformer the author found out that it is not entirely impossible with a winding machine. The most important feature of such a machine is a reliable mechanical counter. The transformer, although it is very good is somewhat out of time. The dimensions are not very practical, the copper losses in the primary are with 250 Ohms between the plates somewhat high and a leakage induction with 30mH somewhat high and last but not least, the output impedance's are impractical. And yet, the author has a pair of transformers that are wound exactly to the Williamson specs, it sounds wonderful.
The first thing you must know about output transformers is that the square root of the impedance ratio in the turn ratio. (the quadrate of the turn ratio is….) The second thing, leakage inductance can easily be measured with a LCR meter. Short the secondary and measure the inductance that is left over in the primary. Don't be surprised when it in much lower as 30mH. The author reached values as low as 3.8 mH. The selfinductance of the entire transformer is not so easy to measure since it depends on the excitation of the iron core. Therefore it is more practical to measure the initial selfinductance at 5 volts. If one connects the primary to a 5 V ac supply with a ammeter in series one must measure less then 100microamps for an acceptable transformer. Since the quality of core material has improved enormously over the years it is not impossible to use a smaller core of grain oriented material with somewhat less turns and still reach a higher primary inductance.
Although it is possible to calculate everything of a transformer it is hard to say how a transformer performs once it is wound. The best guaranty that a do it yourself transformer works is sticking to the original Williamson design and diminish the diameter so that you reach a turn ratio of 35 when the entire secondary is connected in two parallel sections that are put in series. You can also develop a transformer yourself and make use of the modern materials. Be prepared for dozens of transformers that look nice on paper but for reasons that will remain unknown (till you are wiser) don't sound any good. Reject the transformers that don't sound at all and improve the ones that seem better. It is learning the hard way as the author experienced.
The author has after years finally developed a transformer that is better then the original Williamson on a E-I 130B core of grain oriented material. It has sectional windings, has a leakage inductance of 8mH, 120 Ohms dc between the plates, 0.14 Ohm dc secondary and has a secondary impedance of 8 Ohms to a primary of 9000 Ohms. It has connections for ultra linear use with other constructions. Its still weighs 5Kg (11lbs). The author has not decided yet to produce it himself or to license a workshop to do it. If coming winter brings nice weather for indoor jobs some extra examples for evaluation may be produced! For information on this subject please mail awvdkerk@worldaccess.nl mentioning OPT as subject.
The Power Supply
The power supply is as important as the rest of the amplifier since it supplies the power you are listening to. If the power is bad, the sound is bad. There are people who insist to build the power supply the same way Williamson did because it sounds better they say. If audible at all, it is a matter of taste. The supply with a rectifier tube has a higher inner resistance since the power transformers that are made to work with rectifier tubes are made with a fairly high dc resistance in the copper of the secondary because the specs of the rectifier tubes clearly states they expect that resistance in the order of 100 Ohms in order to survive. Transformers made for solid state rectifiers have a dc resistance in the secondary as low as 2 or 3 Ohms. It is not impossible this difference in dc resistance results in the difference in sound reproduction that some claim to hear. So if you use solid state, try some resistance in the positive lead especially if the plate voltage is a bit high. The 450volts Williamson uses is a bit to high for electrolyte caps that are rated 450WV. 425 volts is saver.
One must expect the mains voltage to swing 5% from the rated value! For 425v using a bridge rectifier you need a transformer the has a secondary voltage of about 310v. Using a rectifier tube you may need two times 350v to its plates. If one can afford an extra regulated power supply for the input stage this is recommended! In a stereo amp two of those stages draw only 19 mA. so a small extra transformer will do.
Chokes are rather difficult to get nowadays and when you can buy them in specialized audio shops they are incredibly expensive as if those things were something special. Fortunately you can make them yourself from ordinary transformers with a E-I core lamination. Remove the lamination, put all the coils in series ( watch the direction of the windings !) and replace the core material so that the E 's are on one side and the I 's on the other. Allow an airgap between the legs of the E 's an the I 's by putting a thin strip of polyester in between. Test your choke with 50 or 60 cycles from a mains transformer at the current you want to use it for and refine it. Compare it with a choke that you know is good. Put it in a neat metal enclosure to make it look expensive. In the Williamson design the reservoir caps are rated only 8 mfd. Still he claims his amp to have a signal to noise ratio of 100dB. In his days electrolytes were very expensive and chokes are a great aid in smoothing H.T.!
Nowadays we rather buy a couple of 100mfd"s in reservoir and smoothing caps. Replace any cap that you think is to small for any value you think is better but remember that rectifier tubes don't like to see a capacity over 32mfd! Some people will heat the tubes with ac others with dc. It doesn't make that much difference. Williamson used ac and achieved the aforementioned 100dB. If you use dc consider putting the heaters in series for 12.6v, you need only half the current and the loss in the rectifier is much less. In terms of ac you need aprox 10 volts. If you end up a little higher then 12.6 ac, build in some resistance, this prolonques the live of the heaters. The exact 6.3 or 12.6 volts at the rated mains voltage is very important for the livecycle of the tubes. Since the cathode voltage of the phase splitter tube is about 100 volts you may consider to superimpose some 50 volts to the heater circuit to reduce the chances of a breakdown of the tube. If you heat with ac this also reduces the heater hum since the heater at this voltagelevel will not emit electrons. You can take the 50 volts from resistors between ground and HT that total some 200kOhms that you need anyway to discharge the reservoir caps. Try 22K and 200K. When wiring the heaters don't forget that the current becomes enormous if you connect all the heaters in paralel. Especialy if you plan to use the EL34 with 1,6amps heater current each!
Time Delay H.T. Switch
It is very important to delay the high tension to come in if you want to save the live of your electrolytic caps and the power tubes. There are many circuits in the market, most of them using solid state. This is one with a tube. It has the benefits of the memory of the heater of the tube. If the mains is interrupted only for a sec or two, it switches h.t. on immediately. If the tube is cold it takes as long as it takes to heat it up. I don't know what you have in your junkbox so you have to experiment yourself on the exact values. The resistor in the heater circuit mentioned below is to make the tube heat up a little slower then the other tubes rather then to lower the tension. If your power transformer has connections for fixed bias that you don't use for that purpose try those for this purpose.
General Construction
If you are familiar with the construction of tube amps, please skip this page! For those who are still there, a list of do's and don't 's: Do make a metal chassis that is big and strong enough to last a century! Do make a plan for the layout of all parts avoiding inputs and outputs to "see" each other! Do take care take the cores of transformers and chokes are not in the same direction. Do keep transformers away from inputs and input circuits. Don't remove tubes from equipment that is in use. Do make an earth bushbar, a thick copper wire following the circuit picking up all the earth bound leads and connect it to the chassis at the input. Do consider that you encounter problems when connecting a headphone with a stereo headphone jack. Sometimes these problems are mere theoretical. Do keep the leads in the grid circuit short. Don't use resistors below 1 watt. Do calculate the dissipation of each resistor in advance. Don't use capacitors above their rated voltage. Don't expect old capacitor to survive the voltage that is stamped on them. Do expect the circuit to explode the first time you turn it on. Do expect the negative feedback loop to be connected reversed the first time you turn your amp on causing severe oscillation due to positive feedback. Do keep unshielded tubes / amps away from children. Do save energy somewhere else in your life before you deserve a tube amp. Do respect the energy that can be stored in electrolyte's. Do discharge them with a resistor, not by short circuit. Do keep an eye open for old equipment that is hardly used and has perfect parts e.g. silver plated tube sockets in old army transmitters, as good as new tubes in tape recorders on attics since tape recording soon became boring to most people. Do use fuses, for powertransformers take slow blow fuses, do expect them to blow every now and then and be rather save then sorry. Don't mix circuit plans, choose one and stick to that one, e.g. don't try fixed bias with a Williamson. Do have fun, be patient, if tired from constructing all day, save testing to the next day after going over the construction once again. Don't use multimeters with frequencies over 400 cycles or with strange wave forms. Don't believe everything you read in the magazines on this subject.

Test results

After building a number of transformers and amplifiers we tested these monsters. They sounded very good but some people like numbers and facts.

The transformers gave away the following facts: Leakage inductance is:
prim to sec: 8.3mH
½ prim to ½ prim 7.8mH
½ prim to sec 4.05mH

The starting inductance was over 100H at 5 volts, The maximum inductance is expected to be over 600H Core material was VM111-35 Din with Carlite insulation, EI 130B weighing 3.1Kg.

Impedance= 9000 to 8 ohms Winding is sectional with 6 primary coils interleaved by 5 secondary coils in each section. DC resistance over the total primary is between 120 and 130 ohms (varies with each roll of copperwire). DC resistance secondary is (calculated) 0.14 ohms.

With a generator with Ri=0 ohms the frequency response was flat from 7Hz to 100KHz, with a generator with Ri=2000 ohms freqencyresponce fell 3dB at 100KHz..

The amplifier we tested was equipped with the double driver stage as proposed in the driver stage chapter and a pair of EL34 in the powerstage. The overall feedback was 9.5 dB. The value of the capacitor C10 was 200pF. Frequency response feel 3dB at 95KHz. The square waves showed a risetime of 4 microseconds and an overshoot that lasted 4 microseconds. The overshoot ripple was 1/12 of the preceding rising line high. The rising line was very straight. The square wave at 40Hz was not so straight as the one of the "Conrad Johnson Premier 11A" but much better as many of the very expensive commercial constructions as the "Luxman 38S".

Harmonic distortion remained till 8 watt output power as low as 0.025% rising to 0.25% at 14watts. Clipping point was between 11 and 12 volts at 8 ohms as was to be expected.