Regilding the Gilded Lily: Heath's W-2M

dcgillespie

Fisher SA-100 Clone
Subscriber
INTRODUCTION

Immediately following WW II, there was very little in the way of truly high fidelity amplification equipment available to audio enthusiasts. But then came D.T.N. Williamson's Landmark design that introduced the Williamson amplifier to the world in the May, 1947 edition of Wireless World. He revisited his creation in a followup article in August of 1949, and by that time, the British design caught the attention of Mr. David Sarser and Melvin Sprinkle, who were so impressed with it that they introduced an Americanized version in their article which appeared in Audio Engineering in November of 1949. Dubbed the "Musician's Amplifier”, it sported a large (for its power level) Peerless OPT, and made a splash in the audio world that is still felt to this day.

In 1952, Heathkit, who already had audio amplifier kits on the market, decided to really up their game with their audio products, and introduced the W1-A1 two chassis basic amplifier kit, which was a near exact copy of the Sarser and Sprinkle version of the Williamson amplifier. The OPT that Heath used was notably smaller than that used by Sarser and Sprinkle, but it was still a genuine Peerless unit none the less. Independent of these developments, were David Hafler and Herbert Keroes of Acrosound Transformer fame, who released their original Ultra-Linear OPT in November of the year prior. Powering the screen grids from taps on the OPT primary winding (an idea originally suggested but never exploited by the Brits), it too made an equally big splash as that of Mr. Williamson's design a few years earlier. With both the Williamson amplifier and Ultra-Linear output stage seen as monumentally huge advancements in the audio scene, the idea of marrying the two concepts together was hatched in short order.

By June of 1952 the wedding had occurred, with Hafler and Keroes publishing their article on converting the original Williamson amplifier to Ultra-Linear operation. The Williamson amplifier originally employed a triode output stage, but by changing the OPT to the (now) famous Acro TO-300 Ultra-Linear OPT, power output was increased, yet with no deterioration of the Williamson's other fine attributes. Sarser and Sprinkle took note of this development, and immediately started experimenting with using the (originally unused) intermediate primary winding taps provided on the Peerless OPT they specified for their version of the Williamson amplifier to power the screen grids of the output tubes. While the taps were not located exactly where Hafler and Keroes specified for true Ultra-Linear operation, they were close enough to make the experiments worthwhile. The results were conclusive enough that this change, along with some other less significant changes, formed the basis of their followup article the very next month in July of 1952 entitled "Gilding the Lily" -- the lily being the original Musician's Amplifier they had introduced less than three years earlier. It was a period of rapid change in the fledgling early days of high fidelity, where you had better not blink, or you'd miss the next great advancement coming along.

Now Heathkit, not wanting to miss out on all of this advancement, immediately sourced a new and larger OPT from that offered in the W1-A1, since that transformer had no intermediate primary winding taps. The new transformer they chose was larger, but still not as large as the original piece specified for the Musician’s Amplifier (Peerless S-265Q). However, as an Altec Lansing 20-20 Peerless model 16277, it had intermediate primary winding taps, and they were located at the same position as those on the S-265Q (50% of the winding). With this new transformer, the use of its primary taps to power the screen grids, and a change to new, tight tolerance 5881 output tubes and 5V4G rectifier tube, the W-2M was born in January of 1953. In the owner's manual for the W-2M, Heath references Sarser's and Sprinkle's Gilding The Lily article, and how the improvements outlined in it with tapped screen grid operation were incorporated into the new W-2M amplifier. It is that amplifier that is the subject of this thread.

However gilded as it might have been over Heath's original Williamson offering, the design of the W-2M still leaves a lot of sonic performance on the table versus what it is capable of delivering. This is not meant to slam Heath, nor its efforts with the W-2M. After all, they were simply following the efforts of Sarser and Sprinkle, who themselves were following the efforts of Williamson. Most of the items to be discussed relate to the (soon to become) low and high frequency stability issues of the design, which when unchecked, have a marked influence on the audible presentation of any amplifier. At the time of the W-2M, much was still being learned regarding the performance criteria of a global NFB design, let alone even defining what good stability was. So Heath was hardly alone in this concern.

Additionally, in executing the W-2M, Heath strangely (or wisely depending on your point of view) did not follow some of the Gilded Lily's recommendations -- which was certainly their choice, but the lack of any discussion regarding the issues involved leaves its users at a disadvantage. On the other hand, there was an improvement that Heath made of their own accord in the W-2M, that Sarser and Sprinkle likely would have discussed in their Gilded article, had their test procedure not masked the concern addressed. And, there were even two versions of the W-2M as well. But all of this will be discussed in due course.

One last point I'll mention in this opening salvo: This re-gilding effort was not born out of any desire to re-engineer a piece of audio history simply because it could be done, or for want of something to do. Neither was this an effort to make it into something it is not. Rather, every effort has been made to preserve what it is, retaining the best of its performance attributes, while making stability improvements to the design that really should be considered as mandatory: Depending on how a loudspeaker load is or is not connected to the W-2M, it is at best a marginally stable design, displaying either sustained LF or HF oscillation under certain conditions, with the ever present possibility of damage to HF driver elements always being high. At the very least, such instabilities will really muddy any LF sonic presentation, and degrade the clarity of HF transient information as well. This is only made worse then by today's highly dynamic sources, that provide significant energy in frequencies close to where these instabilities reside. Therefore, the goal was not to chase better sound for the sake of the chase. However by properly addressing the stability issues of the design -- if only to provide for safer operation of the amplifier and the HF drivers it powers -- better sound will definitely be the result.

For now however, a few pics are presented of my W-2M shortly after I received it, and which became the target of this re-gilding effort. Whoa boy! The adventure begins. The person who constructed this unit clearly had little experience building electronic kits, as the picture shows. Besides the creative lead dress, none of the retaining rings for mounting the tube sockets were ever properly snapped into place, so that in an upright position, they could easily fall down around a socket's terminals, shorting them together, to ground, or both. It definitely made it interesting trying to remove any tubes from their sockets. In fact, the resistor providing B+ power to the preamp socket was almost certainly burnt to a crisp due to this issue. It’s a wonder that the transformers weren't fatally damaged with such lack of construction detail. Incredibly however, they were all good, and all of the connections were correct as well — mounting rings not withstanding. At one time then, this model of kit construction may have actually operated -- or not. The history of a kit piece can be interesting to say the least. The pic shows the unit after I had snapped all of the socket retaining rings into their properly seated position, and first started the process of completely rebuilding it (input cap has been removed).

Next up, we’ll start getting into the regilding discussion. Next time.

Dave

Below: Could use just a little tidying up!
SAM_1919.JPG

Below: Check out the interesting matched set of tubes this baby came with. They both tested just fine!
SAM_1929.JPG
 
DISCUSSION OF THE SARSER AND SPRINKLE GILDED LILY

So just how well does the stock W-2M emulate Sarser and Sprinkles’ Gilded Lily? By physical appearances, the OPT of the Gilded Lily (S-265Q) appears to be a 40 watt transformer, while Heath's 16277 is smaller, and would appear to be a 20 watt transformer. Right off the bat then, the W-2 would appear to come to the party disadvantaged. I have never had an S-265Q to test in circuit against the Gilded Lily article — but I do have the Sarser and Sprinkle article, and the measured data published in it. So that’s the reference standard we’ll have to use for comparison purposes.

For starters then, the published performance specifications offered for the Gilded Lily in the article bear some discussion:


POWER OUTPUT
In the early 50’s, rated power output was often stated in terms of “IM Power Output”, along with an attendant level of IM Distortion produced at the rated IM power level given. Other times, power was given as simply sine wave power output, with both rating systems routinely used within a given article, but possibly varying back and forth from one article to the next within a common series of articles. For example, Sarser and Sprinkle seem to favor sine wave power output in the original Musicians article, but then favored IM power in the followup Gilded article. Frankly, it can get hard to keep track of which system is being used at what point in text, making performance comparisons tedious and much harder to make than they need to be.

Sidebar: IM Power is based on a 4:1 two-tone test signal. Average-responding AC voltmeters calibrated to indicate the RMS value of a single frequency sinusoidal waveform will not accurately register the RMS value of a two-tone test signal, and gives rise to the term IM Power Output, or statements like “power as read on the IM Meter”. While the frequencies used for IM test tones varied over the years, they were always based on a 4:1 ratio of low to high test frequencies. As long as the 4:1 ratio is maintained, any power output level stated in terms of IM power can be converted to equivalent RMS power output (regardless of the test frequencies used), by multiplying the IM power level by a factor of 1.47.

DISTORTION
The IM Distortion figures quoted with IM power output ratings were often at stratospheric levels based on today’s standards, indicating the distortion norms that were considered customary at that time. For example, Sarser and Sprinkle use an upper limit of 8% IMD (a point at which they say “serious distortion” begins) in the spread of distortion levels used to define IM power output. At this level of distortion, compression of the waveform is well underway, making the power output produced at such a distortion level of dubious value — either then, or now. But back in the day, anything of about half this amount would be considered rather ordinary.

Assigning a modern, accurate power output rating to the Gilded Lily then would be helpful. This first entails converting the IM power given to equivalent RMS power, and then specifying that power as the level that can consistently be achieved which resides safely before but on the threshold of the point where the IM distortion curve makes its upward turn for a moon-shot run. The closest we can get to obtaining that information from data provided in the Gilded article comes from the statement “….while the equivalent sine wave power at 1.5% IM is 22 watts”. Since IMD starts to rise quite rapidly once maximum power output is reached, an honest RMS power rating of the Gilded Lily can realistically be set at 20 watts RMS.

POWER RESPONSE
From the graph provided in the Gilded article, power output at mid-band frequencies is shown to be down 1 db at 20 Hz, and down almost 2 db at 20 kHz. With mid-band power output now established at 20 watts RMS, that means that at 20 Hz there is a power output of 15.9 watts RMS available, and at 20 kHz there is about 13 watts RMS available. Midband power output is shown to extend from 60 Hz to 7 kHz, with a smooth reduction on either side of this spectrum down to the levels given for the extremes of the audio bandwidth.

STABILITY
The authors state that “With most of the amplifiers converted (to the Gilded Lily) by the authors, there is no tendency towards instability either at sub-audible or supersonic frequencies.” They indicate however that it is also possible for the loudspeaker to have a tendency to “breathe” or oscillate slowly at 1 Hz. This is because with tapped screen operation, the gain of the output stage is increased by about 4 db, nearly all of which is converted into increased NFB. As a result of this increase, LF stability can become upset, causing the breathing condition noted. Therefore, they recommend that the NFB resistor in the Gilded version be increased from its original value of 5000 Ohms, to 6800 Ohms, which then maintains the original 20 db of NFB in the loop when the tapped screen connection is used.

OTHER PERFORMANCE CHARACTERISTICS
The Gilded article indicates that checks were made on response, square wave performance, and source impedance, and “were found to be affected very little” from that of the original Musician’s Amplifier. For the record then, the performance of those characteristics as given in the Musician’s Amplifier article are as follows:

1. Response: Under the test conditions given (amplifier fed from oscillator through a 500K series resistor), the response was said to be flat from 20 Hz to 80 kHz, with a 3.4 db rise at 96 kHz, with the output beginning to fall again at 100 kHz.

2. Square Wave Performance: From the author’s statements, “The amplifier performs exceptionally well when tested with square waves. … it was found that this amplifier passed without distortion square waves having fundamental frequencies from 20 Hz to 5 kHz.”

3. Source Impedance: Given as .67 Ohm on the 16 Ohm tap.

These are the performance results given then by Sarser and Sprinkle to define the performance of their Gilded Lily version of the Musician’s Amplifier. Next time, we’ll look at the specifications as given by Heath for their model W-2M, and let the comparisons begin. Since I’ve got measured data from my own Heath unit that is the subject of this exercise, any holes in Heath’s data can get filled in real quick. We’ll get started on that next time.

Dave
 
DISCUSSION OF THE HEATH W-2M - PART I - (supporting pics to follow at the end)

The specifications of interest as given by Heath don’t exactly match up to those given for the Gilded Lily, but are offered here for the record:

FREQUENCY RESPONSE: within 1 db, 10 Hz to 100 kHz.

HARMONIC DISTORTION: At 5 watts: < .5% between 20 Hz and 20 kHz.

IM DISTORTION: At 5 watts equivalent power output: .5% using 60 and 3 kHz.

INPUT VOLTAGE: .75 vac for 5 watts power output.

With the exception of the OPT and a few other less obvious but equally important details, the W-2M is faithful copy of Sarser and Sprinkle’s Gilded Lily. However, due to the spareness of Heath’s specifications and their misalignment with those given for the Gilded Lily, and because a greater scope of performance measurements are now available for the W-2M, it suggests that some aspects of the design bear further discussion:


POWER OUTPUT/POWER RESPONSE
At the onset of clipping (no visible waveform deformity), midband power output at 1 kHz was 22.1, 21.1, and 22.1 watts RMS for the 4Ω, 8Ω, and 16Ω output configurations respectively. Full midband power was basically available from 20Hz up to 6 kHz in all configurations (based on a 20 watt power output rating), with each end of this band then beginning its slope down to the following power output levels noted at each end of the audio spectrum:

1. At 20 Hz, the power levels were 20.7, 19.22, and 20.25 watts respectively. The drop off slope from midband power to these power levels was gradual and even.

2. At 20 kHz, power output fell to 16.81, 15.82, and 16.81 watts respectively. As power output slopes to these power levels however, the drop off from midband power is not a gradually reducing slope, but rather, falls slightly below 20 kHz power levels relatively quickly, before rising to meet the power level shown at 20 kHz. Heath noted a similar effect with the Peerless transformer of the W-5M as well, although it started at a slightly higher frequency. In the W-2M, power output at any frequency between 6 kHz and 20 kHz never falls below a -1 db power level based on a 20 watt power output rating.

Back in the golden years of vacuum tube audio, these power levels would qualify the W-2M to be marketed as a 20 watt power amplifier, since the power output at any point within the audio spectrum allows the amplifier to be rated within the standard distortion specification verbiage of the day, given as: “ xx% THD within 1 db of xx watts RMS power output”. The measured power output then supports and is consistent with the RMS power output rating assigned based on data from the Gilded article. Heath did not provide a maximum power output specification for the W-2M.

These power measurements were made with the amplifier operating directly from my local power line which hovered at 121 vac during testing. The decision was made to operate the amplifier this way for three reasons: (1) It is within the rated operating voltage published for the amplifier (105 - 125 vac), (2) It produced a heater voltage at the output tube sockets of 6.50 vac, which well within acceptable guidelines (there is voltage drop through the umbilical cord), and (3) It caused the output tubes to draw 63 mA each as specified in the Gilded article.

In comparison with the published power output response curve for the Gilded Lily, the W-2M compares very well, even besting their curve at the frequency extremes: The Gilded Lily is shown to be down in power 1 db and nearly 2 db at 20 Hz and 20 kHz respectively from midband power output, while with Heath’s smaller transformer, the W-2M’s power response is down 0 db at 20 Hz, and down only 1 db at 20 kHz based on a 20 watt power output rating.

From the perspective of power output and power response then, Heath’s OPT in the W-2M is an excellent performer for its day, and therefore an excellent choice on which to base their version of the Gilded Lily.

DISTORTION
Distortion of the W-2M was checked using the 16Ω configuration, so as to conform with that of Sarser and Sprinkle’s Gilded Lily. Understand that this represents best possible performance in Heath’s version, since distortion will rise accordingly at lower impedance settings due to the reduced NFB they employ. Both THD and IMD were checked. Sarser and Sprinkle did not provide any THD specifications for the Gilded Lily. THD figures are provided at the 5 watt RMS power level (Heath specification power point), and customary 1 db down from 20 watts RMS power level as well. IMD figures are provided at the 5 watt equivalent power level (Heath specification power point), and at the equivalent full power rating of 20 watts. The results are as follows:

@ 20Hz: 5W= .11%, -1db= .45%

@ 1kHz: 5W= .07%, -1db= .09%

@ 20kHz: 5W= .16%, 1db= .44%

IMD: 5 Watt equivalent = .14%, 20 Watt equivalent = .40%

The distortion performance results easily meet Heath’s specifications for the W-2M across the full audio spectrum, and bested those of Sarser and Sprinkle for IMD in the Gilded Lily as well. This is in keeping with the fact that low distortion is a hallmark of the Williamson design. This particular example of the W-2M produced distortion levels that are quite typical of those I’ve observed from other properly operating Williamson amplifiers in the past. No particular effort was made to achieve the results shown, other than ensuring that those components that are called out to be matched in fact are. The amplifier was almost completely rebuilt from the state in which it was received, but afterwards still included many of the original components to produce the performance results shown.

LOW FREQUENCY STABILITY
An advantage that the Gilded Lily has over the W2-M is its S-265Q OPT. As a well oversized unit, the advantage it potentially enjoys over Heath’s smaller transformer is that of increased primary inductance. With a notably larger core and all else being equal, primary inductance will be greater in the S-265Q than in the transformer Heath used — assuming that the same core material was used in the construction of both transformers (I have no information regarding this point). Greater inductance will provide an obvious aid to LF performance, but even more importantly for a Williamson amplifier, it will help to achieve an acceptable level of LF stability.

You get a sense of how Sarser and Sprinkle defined the idea of “acceptable” (LF stability), with their instruction to increase the value of the original NFB resistor to 6800Ω when making the Gilded Lily conversion. As previously stated, the adjustment acts to maintain the original Musician’s Amplifier NFB level of 20 db, which if not otherwise adjusted, would increase to nearly 24 db due to the higher gain of the tapped output stage connection over that of the original triode connection. The adjustment then prevents possible loudspeaker breathing at around 1 Hz or less (LF instability) were the adjustment not made. Apparently then, acceptable LF stability back in the day was defined in terms of absolutes, as in the fact of either outright oscillation or not, as you see how small the margin of stability could be for it to still be considered acceptable.

Today we know that even marginally stable NFB amplifiers can color the sound in so many ways (none good), with the proper stabilization of marginally stable designs representing a huge opportunity for improvement. The W2-M is a marginally stable design: When configured for a 16Ω load (which produces 20 db of NFB), the W-2M will in fact routinely display the loudspeaker breathing warned of in the Gilded article — even with the recommended 6800Ω NFB resistor (standard with the kit) installed. This NFB resistor value may have kept the Gilded Lily out of LF trouble with its larger S-265Q OPT. But with the smaller Heath transformer, it does not when configured for 16 Ohm loads. Therefore, plenty of room for improvement exists on this end of the audio spectrum in the W-2M.
 
DISCUSSION OF THE HEATH W2-M - PART II -

HIGH FREQUENCY STABILITY
Dr. Williamson initially suggested that for units built to his original design that displayed HF stability issues, a small step network could be added across the plate resistor of the AF amplifier stage as a temporary measure in resolving any HF transient stability issues until the real culprit (usually the OPT) could be identified and corrected. However, in his later August, 1949 article describing improvements to the design, he made a practical decision to make the suggested step network a permanent part of his design. He further went on record stating that in addressing HF stability concerns, use of a customary phase advance cap connected across the NFB resistor would not help to stabilize his design if it was built to his specifications — but that it might help control instability when using OPTs with different HF characteristics.

Strangely, three months later when Sarser and Sprinkle introduce the Musician’s Amplifier to America, they do not include, recommend, or even comment on either of these expedients, and of course since Heath was simply following Sarser and Sprinkle’s work, they didn’t address them in the W1-A1, either. When the Gilded article was published, they were still MIA — with Heath almost following along accordingly; while the W-2M does not employ a step network, it does include a phase advance cap.

Sidebar: Why many (but not all) American versions of the Williamson amplifier were slow to include an appropriate step network can only be speculated. At the time, there was heavy emphasis here on providing a frequency response that could reach the moon (flat to 100 kHz seemed to be the hot ticket) — with the emphasis apparently being more than just marketing ploy; it truly seemed to be the accepted engineering practice of the day. Use of a phase advance cap or step network would act to reduce such a high wire response act, and so was apparently dismissed (or deemed not needed?) accordingly in both of Sarser and Sprinkles’ articles. This is unfortunate, as it is the proper application of such networks that can really tame the HF instability tendencies of the Williamson design, without unduly negatively impacting performance in the audio spectrum. In fact, the installation of appropriate tailoring networks will actually improve performance in the audio band, by eliminating the impact that supersonic instabilities can have on it.

Continuing on, Heath decided to include a phase advance cap across the NFB resistor in the W-2M, as a departure from the Gilded modification. The effect of this cap is two fold in that: (1) it helps to reduce ringing that occurs on HF square waves in a loaded condition for any output configuration, and (2) it allows the amplifier to remain stable in the 16Ω configuration when unloaded, but actually causes the amplifier to become unstable and oscillate in the 8Ω configuration when unloaded. This is important, because at at certain frequencies, many speaker systems can represent many times their nominal impedance, which effectively unloads the amplifier. Therefore, just because you're careful and never operate the amplifier without a load connected doesn’t mean that the amplifier always “sees” that load during actual use. It is these types of conditions that cause marginally stable NFB amplifiers to have a definite impact on sound quality. In the 4Ω configuration, the NFB level has been cut in half, so the amplifier does not break into oscillation in an unloaded condition, whether the cap is in place, or not.

Ultimately, a direct comparison between the HF stability of the W-2M and either version of Williamson’s or Sarser and Sprinkle’s amplifiers cannot be made, since the actual HF stability they achieved in their amplifiers is unknown — although it is a virtually certainty that all of their testing was conducted with a proper resistive load connected at all times. There can equally be little doubt that passing a 10 kHz square wave through any of them will show the usual spikes, bumps, and ripples across the wave top, indicating their marginal stability condition, even if outright oscillation is not present. The W-2M, being cut from the same amplifier cloth, in fact does produce 10 kHz square waveforms of this type, with the opportunity for improvement at this end of the spectrum then being ripe as well.

NEGATIVE FEEDBACK LEVEL
While Williamson and Sarser and Sprinkle both provided formulas to determine alternate values for the NFB resistor to be used in their amplifiers when anything other than the designated 16Ω output configuration is used (so as to maintain the target 20 db of NFB regardless of output impedance configuration), Heath chose to ignore this point and instead, used the 16Ω value of 6800Ω (as recommended by Sarser and Sprinkle) for all output configurations. Therefore, while the 16Ω configuration of the W-2M produces the same NFB level as that of the Gilded Lily, configuring Heath’s amplifier for lower output impedances causes it to operate with reduced amounts of feedback, as the output impedance is reduced accordingly. The measured NFB levels when loaded with a proper resistive load for each output configuration then are as follows:

@16Ω = 21.1 db
@8Ω = 18.8 db
@4Ω = 15.9 db

While the reduced feedback at lower impedance levels does act to improve stability issues at both ends of the audio spectrum, it also causes other performance parameters to deteriorate. Most notably, distortion and internal output impedance will both increase as feedback is reduced, and the gain of the amplifier will increase as well. Consider that:

@16Ω: Measured Sensitivity = 1.50 vac for full power output (corresponds/w Heath 5 watt spec)
@8Ω: Measured Sensitivity = 1.00 vac for full power output.
@4Ω: Measured Sensitivity = 0.79 vac for full power output.

Distortion and internal impedance levels are acceptably low for any of the possible output configurations, but these performance parameters will still as much as double from the 16Ω to 4Ω configurations. All else being equal, the volume control will also appear to be more responsive at lower output impedance configurations. Stability is impacted accordingly as well. With the 16Ω configuration, loudspeaker breathing is noted as mentioned earlier, while absent on lower impedance configurations. On the high frequency end, the maximum level of capacitance that the amplifier can tolerate before becoming unstable (sustained oscillation) for both loaded and unloaded conditions is as follows:

@16Ω: Loaded = .03 uF
Unloaded = .005 uF
@8Ω: Loaded = .10 uF
Unloaded = 0.00 (unstable w/o load)
@4Ω: Loaded = .33 uF
Unloaded = 300pF. After 1000 uF, stable to .012 uF

The amount of capacitance the amplifier can tolerate in a loaded condition increases as the load impedance is configured for lower settings, which is typically the case. In the W-2M however, these values are further magnified at the lower impedance settings by the reduced NFB in those configurations as well. However, the amount of capacitance that the amplifier can tolerate with no load in any configuration changes very little, and is in fact, quite small. So while the amount of capacitance the amplifier can tolerate when properly loaded is adequate, the amount it can handle unloaded is woefully inadequate. As a result, varying impedance loads like that of a typical speaker system can make the possibility of triggering instability — if only for very brief periods — very high.
Having so many performance parameters change between the various output configurations is a cumulative effect. As a result, even if three otherwise identical speakers could be found except for their rated impedance, the audible presentation will be different with each speaker due to the changing performance parameters within the amplifier for each configuration. Part of the re-gilding process then will include resolving this issue, to help bring the W-2M into the 21st century.

FREQUENCY RESPONSE
Sarser and Sprinkle measured frequency response by first passing the oscillator test signal through a resistance equal the input impedance of their amplifier — in their case, a 500K resistor. Remarkably, they report that using this technique, the Musician’s Amplifier (of which the Gilded Lily version was found to be affected very little from in this regard) was flat from 20 Hz up to 80 kHz, showing a 3.4 db peak at 96 kHz — all as given earlier. They state that this is the customary way of measuring amplifiers professionally. True — for industrial equipment. But this hardly represents how the amplifier will be used in practice (or any home high fidelity amplifier for that matter), where it will be directly connected to the output of a preamplifier, who’s output impedance will intentionally be quite low. I can only imagine how strong their peak at 96 kHz would be were the amplifier tested when directly driven from the oscillator’s output!

In any event, using their method to test the frequency response of the W-2M produced the following (expected) results (ref: 1 kHz), using the 16Ω output configuration:

@10 kHz: -.1 db
@20 kHz: -.6 db
@30 kHz: -1.1 db with a smooth reduction from there that culminates in being down 8.5 db @100 kHz (no sub peaks)

The above response was obtained with the 47 pF phase advance cap installed. When this cap was removed (since the Gilded Lily used no such cap), response at 100 kHz was down 7.5 db.

This response curve is starkly different from that reported by Sarser and Sprinkle. I can only say that the response obtained is quite within expectations according to theory when using this method of testing. In contrast, with all else equal except removing the series resistor, driving the W-2M directly from the oscillator output produced the following results (phase advance cap in place):

@20 kHz: +/- 0 db
@100 kHz: +1.1 db produced as a gradual rise from 20 kHz (no intervening sub peaks)

This result would be considered in line with Heath specifications.

INTERNAL OUTPUT IMPEDANCE
This was found to be 0.75Ω on the 16Ω output configuration. This is insignificantly different from the value of 0.67Ω as given for the Musician’s Amplifier. The difference can easily be attributed to the former’s tapped screen grid operation, versus the latter’s triode operation of the output stage. Heath did not provide a figure of internal output impedance for the W-2M.

VERSION
The earliest circuit, build instructions, and pictorials of the W-2M show it does not include the 20 uF/150 volt output stage cathode bypass cap that later builds include. Sarser and Sprinkle were basically noncommittal about the inclusion of such a cap, suggesting that for those concerned that its omission would degrade performance install a 50 uF/50 volt cap in that location to address their concerns. At some point, Heath sought to address those same potential concerns as well, and decided to start including a cap of the value noted.

Testing shows that inclusion of the cap produces no impact on the NFB factor, or on HF square wave presentations. It also has a basically unmeasurable impact on 20 Hz power output and THD at this frequency as well. As mentioned in the Gilded article, this is in fact due to the amplifier operating in Class A throughout virtually all of it power output range. However, the cap has a rather dramatic effect on IMD, with it’s inclusion dropping IMD over three fold versus that produced without it. No doubt, this is the reason that Heath starting including the cap in the W-2M. Any units that do not have this cap included should install it for best possible performance.

Next up, making simple changes that make for major improvements — next time.

Dave
 
PERFORMANCE MODIFICATIONS FOR THE HEATH W-2M - PART I -

From the previous discussion, the areas for opportunity include:

Improving Low Frequency stability.
Improving High Frequency stability.
Provide consistent 20 db of NFB regardless of output configuration used.
Address misc areas of opportunity.


IMPROVING LOW FREQUENCY STABILITY
David Hafler and Herbert Keroes were the first to address the LF (and HF) instabilities of the Williamson amplifier by pointing out that the time constants of the two R/C coupling networks on either side of the driver stage are the same. As they state, this is not desirable in a NFB amplifier, because the identical time constants of the two networks maximizes the phase shift produced for a given frequency loss. Their solution was to up the value of the first pair of coupling caps by a factor of 5, providing a 5:1 ratio between the two time constants, which allows for the same tailored loss of LF response to be had, but with less phase shift — all of which improves the LF stability of the design.

The expedient works, but only to an extent because it was limited by the physical size of the components of that day. Today, much larger capacitance can be had in a much small package. Because of that, greater advantage can then be taken of increasing the ratio between the time constants of the two coupling circuits. The practical limits of this approach will be found by increasing the value of the coupling caps into the driver stage to 2.2 uF each, and decreasing the value of the coupling caps into the output stage to .1 uF each. This creates a ratio of over 100:1 between the time constants to maximize the benefit of this approach.

This move effectively makes the coupling circuits between the phase inverter and driver stages all but direct, making them basically invisible to the circuit relative to LF phase shift. To improve upon this effort would require making the coupling between the driver and output stage partially direct coupled (adjusting the phase inverter coupling caps accordingly), which would require a major redesign of the amplifier. Short of that, this approach produces very high LF stability with rapid settling under pulsed conditions — even in an unloaded condition, which is an extremely tough test for any Williamson amplifier to pass. Copies of the amplifier that achieve good LF stability under loaded conditions are often still at best marginally stable under unloaded conditions. This remedy puts a fork in the whole LF issue of the W-2M.

There is one other area however that also needs to be addressed to achieve the maximum improvement in LF stability: Power supply decoupling. One of the cost saving measures that Heath did in their version of the Gilded Lily was to dispense with the power supply decoupling network between the B+ to the output stage, and that of the driver stage. As a result, the driver stage runs directly off of the same B+ that supplies the output stage, which is poor practice. Both Williamson and Sarser and Sprinkle employed a decoupling network in this position, and the W-2M deserves no less if the best LF stability is to be achieved.

The easiest way to implement this into the W-2M is to utilize the extra unused lead within the umbilical cord. The original B+ lead can serve to power the output stage, while the unused lead can be pressed into service to power the small signal stages. This will allow the new dropping resistor and decoupling cap to be placed under the power supply chassis, where there is plenty of room to accommodate them. Use a 3000Ω 2 watt resistor and 33 uF 500 volt cap to decouple the driver stage (and earlier stages) from the output stage B+ source.

As a final note, unless you are planing to use the W-2M with a WA-P1 preamplifier, the input .05 uF coupling cap can be removed (shorted out) and dispensed with. The 2.2M grid resistor should be replace with a 470K resistor, and should be mounted across the two terminals of the input jack.

IMPROVING HIGH FREQUENCY STABILITY
Achieving good HF stability in a Williamson design is best achieved by putting together the individual approaches previously proposed by others. The trick is to implement them so that they each compliment one another to create maximum stability, with a minimum of performance loss in the process. For the W2-M, there are three steps to take to address the HF stability issue:


1. Install a 10K .5 watt resistor between the input jack and terminal #1 of the first 6SN7. This assumes that the grid resistor has been moved to the input jack as previously mentioned.

2. Install a step network consisting of a 1.3K resistor in series with a 1200 pF cap across the 47K plate load resistor connected to terminal #4 of the of the first 6SN7. The network can be installed either way across the plate load resistor. A mylar or mica cap is fine.

3. Install one of the new NFB networks offered to suit your purposes.

4. Optional: Install a .022 uF 100Ω Zobel network across the speaker output terminals. It does not matter which way it is installed across the terminals, and it remains connected across the speaker terminals regardless of the output configuration used. This network does not draw any notable power from the amplifier (it does not reduce 20 kHz power output), and the resistor stays plenty cool during sustained full power testing at 20 kHz. The amplifier does not need this network to remain stable under any condition of loading (speaker or otherwise). Its sole function is to clean up a small instability noted on square waves when there is no load on the amplifier. It is entirely optional.

NEW FEEDBACK NETWORKS FOR CONSTANT 20 DB NFB
Six new compensated NFB networks are provided for you to use depending on the output configuration and the output stage connection you wish to use. Heath gave instructions allowing the user to operate the W-2M with a triode output stage as found in the original Musician’s Amplifier. The problem is, as designed, just as the NFB level is reduced when using lower output impedance configurations, it is further reduced if choosing to use the triode output stage option as well. Therefore, a separate network is required not only for the impedance configuration you use, but also based on the output stage connection you use. Chose the network that is correct for your application. All networks maintain 20 db of NFB for the chosen output stage/output impedance configuration chosen. When changing the amplifier between the various options, the network, and the wiring of the options themselves is the only thing that is changed. All other modifications remain in place regardless of the options chosen in this section. The new networks are as follows:

For tapped screen grid operation:
@16Ω = 7.5K in parallel with 140 pF
@8Ω = 5.1K in parallel with 200 pF
@4Ω = 3.6K in parallel with 300 pF.

For triode operation:
@16Ω = 4.7K in parallel with 100 pF
@8Ω = 3.3K in parallel with 150 pF
@4Ω = 2.2K in parallel with 220 pF

MISCELLANEOUS MODIFICATIONS

1. The amplifier should be fused with a regular 2A fuse. The 3A fuse rating was sized to cover the amplifier and also the Auxiliary AC receptacle since the fuse powered it as well.

2. The series power supply electrolytic caps (4 total) should each have a 220K 1W resistor wired in parallel with it. This will enforce an equalized operating voltage drop across each of these caps.

3. The driver stage cathode resistor should be replaced with a 680Ω .5 watt resistor. This value biases the tube to produce the greatest amount of linear output for the B+ supply it operates from. With this change, the driver stage is capable of providing over twice the drive signal required by the output stage to produce full power output, and at very low distortion. The new NFB networks and performance figures of the modified amplifier are based on the new resistor value in place. The original value of 390Ω limits the drive capability of the stage, and causes the tube to draw needlessly high current flow, shortening the life of the tube.
 
PERFORMANCE MODIFICATIONS FOR THE HEATH W-2M - PART II -

MEASURED PERFORMANCE DATA FROM THE MODIFIED W-2M

Power Output: RMS
@20 Hz = 20.7 watts
@1 kHz = 23.0 watts
@20 kHz = 18.3 watts (with a similar slope as described before from midband power output)

Distortion: THD at 1 db below 20 watts RMS
@20Hz = .55%
@1 kHz= .125%
@20kHz = 2.0%
IMD = .62% at 20 watts equivalent power output.

Frequency Response: (@1 watt) Flat from 10 Hz to -0.15 db at 20 kHz, -1 db @ 60 kHz, -3db @ 80 kHz

Stability:
Loaded: No amount of capacitance will cause the amplifier to break into oscillation in a loaded condition with any output configuration. Rapid settling under pulsed conditions.

Unloaded: Rapid settling under pulsed conditions with any output configuration.
Maximum capacitance allowable:
@16Ω = .05 uF
@8Ω = .05 uF
@4Ω = .15 uF

COMMENT
In comparing the results of the modified amplifier to that of the stock design, there are four stand out points of interest:

1. 20 kHz power
2 .20 kHz THD
3. Frequency Response
4. Stability

In regilding Heath’s Gilded Lily, the goal of improving HF stability was achieved by exchanging excessive HF response for improve HF stability. In the stock design, HF response is needlessly excessive, while HF stability is woefully inadequate. By creating a controlled HF response, the response of the amplifier can be limited to less than the natural resonate frequency of the OPT, which then improves HF stability. Due to the quality of the OPT used, it’s resonate frequency is so high that even though amplifier response is tailored to less than that frequency, the frequency response within the audio bandpass remains flat. Therefore, a better balance between these two performance parameters has been achieved. It is not a zero sum gain however between just those two performance parameters. 20 kHz THD increases in the process, which is typical. A THD of 2.0+% is exceedingly common in many well stabilized NFB designs. Any of the Dynaco designs would do well to reach just 2% at near full power output at 20 kHz. Ditto for many Eico, Pilot, and other Heath designs as well. In fact, with few exceptions, the designs that don’t reach to near the 2% mark are usually insufficiently stable at high frequencies. On the other hand, note that because the new step network addresses not only frequency response at supersonic frequencies, but also phase shift at those frequencies as well, the amplifier is actually able to produce more usable power output at 20 kHz now. Ultimately, the tradeoffs and compromises make for a greatly improved amplifier. Frequency response within the audio bandwidth is still all but ruler flat, with supersonic response down just 1 db at 60 kHz. In return, the amplifier can now handle 10X the capacitance across its unloaded output terminals in all impedance configurations than it could before. This is with the 4 and 8Ω output configurations operating with a full 20 db of NFB now — with the 8Ω configuration now remaining stable with
a .05 uF capacitance only load. By comparison, the stock amplifier was completely unstable in an unloaded condition at this impedance setting.

On the low end, speaker breathing was typical with the 16Ω configuration, while at the 4 and 8Ω settings, pulse conditions produced plenty of waveform bounce that were slow to settle. Revising the LF time constant ratio between the two interstage coupling networks sufficiently reduced the phase shift at these frequencies to make for a marked improvement in LF stability on all impedance settings.

As for the distortion figures in general, remember that the results presented for the stock design were based on the 16Ω configuration, and deteriorate at lower impedance levels due to the reduced NFB generated in those settings. The modified design maintains its performance characteristics with all output configurations now. Also, the amplifier now operates with tight conformity to the target 20 db NFB level. In contrast, the stock design was found to operate at just over 21 db NFB.

Audibly, this all works to produce a notably more solid bass presentation (typically being muddy before), and much smoother HF presentation that tended towards strident. The midrange still maintains all the presence that makes the design so attractive to begin with. The modified amplifier shows greatly improved manners then in both the lab and the listening room.

FINAL THOUGHTS
The Williamson amplifier was the first design to use a global feedback loop around the entire amplifier, including the output transformer. This approach, coupled with using 20 db of NFB in the loop, was quite radical at the time. It being a new concept then, all of the important performance parameters regarding this type of amplifier design had not yet been identified at the time of the W-2M. As a result, extending frequency response and reducing distortion (the most “sellable” specifications) continued to rule the day, since those were the most important performance criteria at the time. The idea of needing a controlled sub and supersonic frequency response, transient response, and stability margin — in addition to low distortion — was simply not on the radar, let alone having any practical standards for these performance parameters, or knowledge of the relationship they must maintain with one another to achieve excellence in overall performance. It was an issue that dogged many if not most manufacturers of the amplifier during the early years of its production in America. Heath was right in the middle of all this with its W-2M. But with a little juggling to rebalance and realign the emphasis of the individual performance characteristics, the W2-M becomes a truly stellar performer. For those of you who are into the vintage-vintage that this amplifier represents, I think you will find these modifications to be well worth your time to install.

Happy listening!

Dave
 
Some pics taken during the regilding process:

Below: These are of the stock design, normally loaded:
SAM_1920.JPG

SAM_1921.JPG

SAM_1922.JPG

Below: These are of the modified design with tapped screen operation, normally loaded:
(The last pic in this series is for the 16Ω configuration of this series. I was trying different lighting conditions at the time.)
SAM_1923.JPG

SAM_1924.JPG

SAM_1927.JPG

Below: Triode operation at 4Ω. Other impedance configurations present the same way.
SAM_1926.JPG
 
Some final pics:

This first series represents different loading conditions than that of just a normal resistive load. For each loading condition shown, each pic represents a very high level of stability, being nowhere close to spilling over into full sustained oscillation. Ringing is identical on the top and bottom of the waveform as a further indication of the stability level achieved. No pics are possible of the stock design in these configurations, since the 8Ω configuration produces full sustained oscillation with no load attached.

Below: Modified amplifier, with 8Ω configuration and no load.
SAM_1933.JPG

Below: Modified amplifier, 8Ω configuration, properly loaded, with .25 uF cap across speaker terminals.
SAM_1932.JPG

Below: Modified amplifier, 8Ω configuration, no load, with .05 uF cap across speaker terminals.
SAM_1934.JPG

Below: Some of the other NFB networks for different output and output stage configurations. Also, the resistors required for triode operation of the output stage.
SAM_1931.JPG

Below: An underside shot of the modified amplifier. Still waiting on the final coupling caps between the phase inverter and driver stage at this point. Yellow cap explained in a moment.
SAM_1930.JPG

Below: An underside shot of the power supply chassis. Discussion of the relay will follow.
SAM_1928.JPG

The relay was installed early on to provide an instant off disconnect for the B+ connection to the output stage, this to prevent a large turn off spike noted in the output tubes and OPT. This is due to the poor LF stability of the stock design, and is very hard on both the output tubes, and the output transformer. The relay prevents the spike from happening, protecting these items. However, with the new LF stability modification, the relay is really not required. It can still be used if desired, but becomes icing on the icing on the cake. The orange drop cap is connected across the relay contacts, while in the amplifier chassis, the yellow cap is connected between the output tube B+ connection and ground. Together, they eliminate all switching transients that might be produced by the relay contacts.

NOTE: Heath W-5M owner with small OPT: Potentially large turnoff spikes in the W-5M may be one of the reasons that this particular transformer is so prone to failure. Watch the speaker cone and output tubes in dim light at shut down. If a significant LF pulse is produced, the speaker cone will display to it (best seen with large diameter speakers), and the tubes will usually show the pulse.

Dave
 
Looks like the 8 ohm tap may not be quite at the right position. Just looking at your power and distortion numbers in stock form, 4 and 16 show up equal with 8 being a little worse.

Dumb question from someone who doesn't own one, when you say "configured for 8 ohm operation" should I take that to mean this has only 2 speaker output terminals on the back, and the idea was to connect the terminal to the appropriate tap on the transformer? I would further guess that since NFB in stock form changes depending which is selected that the feedback resistor comes off the speaker output terminal rather than off a tap directly on the trafo ?
 
Gadget -- In my experience, the 8 Ohm tap is located more accurately than most, with very close performance produced -- but yeah, as always, it's not exact. I don't know of any transformer that is.........

The output connections for the Peerless transformer in the W-2M consists of two 4Ω windings that have taps on the windings. The windings terminate at terminal lugs to make the connections. Both full secondary windings are seriesed for 16Ω operation, paralleled for 4Ω, and the two taps are connected to together to make the 8Ω connection (using the top portion of one winding, and the bottom portionof the other).

6 -- The scope displays are simply set for a convenient size, and are not representative of a measured level. As for adjusting the vertical gain from loaded to unloaded conditions, the height of the waveform changes very little due to the very low output impedance produced.

Dave
 
Dave: Nice write up!

FWIW, my somewhat simplistic take regarding the 8 ohm vs 4/16 ohm performance anomalies is that both the 4 ohm and 16 ohm strappings utilize 100% of the secondary windings while the 8 ohm strapping does not. Net effect is that 8 ohm strapping is not as well coupled to the primary as are the other strappings.

I've also found similar impedance related performance inconsistencies in designs using the more common tapped secondary. Almost without exception, the 16 ohm tap performs better than the 8 ohm tap which in turn performs better than the 4 ohm tap, all else equal. Based on some recent experience, even Mac's primo vintage amp designs exhibit this characteristic. Seems that Heath and many other old time mfgrs were aware of this situation because specs typically were provided for 16 ohm tapping only.
 
Yes, most transformers have best coupling on the 16 Ohm tap. Magnetic flux that ISN'T coupled to the secondary creates inductance (leakage inductance) that will limit high frequency performance. You can measure leakage inductance at the primary by shorting the secondary (the lower the better!). Coupling can be improved by interleaving the windings (wind one layer of primary, one layer of secondary, another layer of primary, etc.). There's a physical limit to that and dividing into more layers adds more capacitance as it reduces leakage inductance, so there will be a point of diminishing returns.
 
Excellent presentation Dave - an absolute pleasure to read.

To dot the i's, can you indicate the IM test frequencies you used, given that nowadays test equipment is probably set up for SMPTE and DIN formats (and noting there is also the CCIF format using 1:1 ratio of 19 and 20kHz test tones). Could you also indicate what equipment/software you used to measure THD, given that some will have diy equipment that may have difficulty measuring 20kHz THD with sufficient bandwidth.

Were you able to measure the noise floor and hum of the amp, wrt full power output? That would likely not have changed much since first light, except if construction lead dress was poor (somewhat doubtful from your description of 'as found' condition and the photo).

Did you just use two random NOS 5881 for all the above?

Were you able to notice if there was any power supply voltage overshoot soon after turn on from the conducting time differences between the 5V4G and the 5881's, given that the preamp caps had/have a 450V rating?

Would you have any interest in testing the overload recovery characteristic of the amp - as an indication of what the amp may be exposed to now given that music content can be much more dynamic.

Ciao, Tim
 
Thanks for the kind comments all.

Steve -- I agree completely about 8Ω tap performance. While changing the output impedance is a nuisance requiring the re-strapping, it does allow for better 4Ω performance than so many more convenient transformers that don't require the re-strapping hassle. Honestly however, I've seen far worse 8Ω performance that this transformer provides from otherwise highly respected manufacturers.

Tom -- Thanks for the comments. I've always understood that a 60-70 watts power level is about the best you can do if you want optimized performance across the entire audio spectrum in a given transformer. Of course, the turns ratio used has a big impact on things as well, but as you start getting into high power levels than this, the winding capacitance just makes things unworkable at high frequencies.

Tim -- I did not measure the noise floor of the amp in the as found condition, as I felt that by the time I repaired it to make it operable, I might just as well go ahead and redo the whole thing to save time. I'll be happy to measure the hum and noise to provide a benchmark since none was provided by Heath. Thank-you for pointing that out.

I did use good vintage Tung-Sol 5881s for testing, since that is what was available at the time. I have found any two of these tubes to be matched well enough to each other, with it easy to produce tightly matched pairs from a given stash of the tubes -- but I did not go to that effort. The tubes I used did have a quiescent current and power output (both @400 vdc) that was matched to better than 5% of each other, which is a very good match.

I will also check the voltage overshoot. The rectifier tube used was a good, vintage 5V4GA, since the 5V4G that came with the unit had no current flow through one half of the tube.

I'll PM you about the overload recovery test.

Dave
 
Geez -- I completely forgot to address Tim's first question:

1. The IM tests were in fact conducted using the SMPTE test signal of 60 Hz and 7 kHz mixed 4:1. I use a trusty Heath IM-5248 I built for such measurements. It has served well for vintage vacuum tube applications.

2. My THD test set is an HP 339A Distortion Analyzer, with oscillator distortion and analyzer resolution performance that is well above the needs for this project.

To round out the equipment used, my scope is a Tek 475A, and dvm DVM is an HP 3466A. All pieces are workhorses that are in excellent condition, with the DVM having calibration that is traceable to the NIST.

I apologize for not including this in my last post!

Dave
 
I think I have a Heathkit amp like that, but I am not sure which model it is, I know it's a 2 chassis like that. I'll have to look for it.
 
On the topic of designers-of-the-day adapting the Williamson circuit and progressing amp performance, the addition of a bypass capacitor across the output stage common cathode resistance was seen to be a bit hit and miss. Distortion (for those who could measure it in that era) often didn't show any generic improvement for triode mode, and wasn't consistent for tube types. In hindsight, Williamson got his circuit right (especially as he didn't need to add such a costly part in the mid 1940's), but caveat emptor for those wanting to implement different tube types and output stage formats.

An interesting insight to the issue was aired by Robert Mitchell back in 1955 - luckily for us, he had the time, incentive and parts to look in to the issue:
http://americanradiohistory.com/Archive-Audio/50s/Audio-1955-Nov.pdf
 
Dave, it sounds like you have the original Heathkit build instructions and circuit - I don't think they are available on-line yet.

A quick calc from your operating values indicates the 5881's were operating near to or at the design centre max power dissipation rating. Nowadays, people are probably a little more circumspect about whether they need to operate expensive tubes at close to max ratings due to wear out - although the aesthetic need to show off valve amp gear is a big benefit for cooling!
 
Tim -- The early manual I have is in fact right from AK's database for Heathkit. It is for the W-2, and therefore includes instructions for not only the W-2M, but the WA-P1 as well. The cathode bypass cap is not to be found anywhere in that manual. Since you see so many with the cap installed, that's what clued me into the fact that there must have been an early and later version.

The 5881 is rated for 23 watts Pd and 3 watts Sd in pentode mode, with 26 watts being allowed for both elements together in triode operation. As you say, this is based on the Design Center rating system. In the Gilded article, Sarser and Sprinkle only mention the triode mode rating -- possibly viewing their tapped screen operation as being an extention of triode mode conditions. In any event, at the end of the article, they address the fact that the tubes will see about a 400 volt drop across them, with a total current flow of 63 mA through each. This pegs total P+S dissipation at 25.2 watts, so they are certainly running the tube right up there near its limit. However, since the limit based on the more conservative Design Center system, this level of operation was considered to be a safe and reliable operating point back in the day. Vintage 5881 tubes should have no problem dealing with this level of operation. Modern manufacture tubes however might be a different story......

Dave
 
Last edited:
great thread I have pair of these installed in my 2nd main system great sounding amps that I had totally restored .Mine are the version from 1954 with TO-300 output transformers
 
Back
Top Bottom