Thank you, I think this may have been the root of my problem. I have the day off so I will look around for parts and see what I can find now that I have this important tip. There are other caps in there that I will likely need to replace. Is there a chart somewhere that shows the old value and the new equivalent? I have read that some people say the resistors drift in value over time as well, would you recommend replacing all of those and or anything else (selenium rectifier)? Or should I just start with the caps and go from there. Obviously just doing the capacitors would be a much easier job.
Heathkit W4-AM modifications
- Thread starter magnetmaz
- Start date
s-petersen
Scott
The idea is to find the closest match, .5 is commonly replaced with .47 etc..
The .25 could also be replaced with a .27
The tolerance on the old caps were commonly +/- 20% The resistors mostly can be measured in circuit,if they measure low,they may be in parallel with something else,if high,they are likely bad. Tolerance was commonly 10% on resistors (silver 4th band). Most new resistors are 2%
The electrolytic multi can caps, can be replaced with cheaper single caps under the chassis. it would be best to increase the voltage rating of the can caps and others,because of today's higher line voltages.
600-630v for the small caps. The can cap has (4) 20mfd@450v caps inside (this should be 500v) The (2) 20mfd@350v (should be 450v)
I have an electronic assembly manual for this amp,pm me with your e-mail if you would like a copy.Is yours gold or grey?
The .25 could also be replaced with a .27
The tolerance on the old caps were commonly +/- 20% The resistors mostly can be measured in circuit,if they measure low,they may be in parallel with something else,if high,they are likely bad. Tolerance was commonly 10% on resistors (silver 4th band). Most new resistors are 2%
The electrolytic multi can caps, can be replaced with cheaper single caps under the chassis. it would be best to increase the voltage rating of the can caps and others,because of today's higher line voltages.
600-630v for the small caps. The can cap has (4) 20mfd@450v caps inside (this should be 500v) The (2) 20mfd@350v (should be 450v)
I have an electronic assembly manual for this amp,pm me with your e-mail if you would like a copy.Is yours gold or grey?
Last edited:
Mog -- The most important caps to replace are:
1. The power supply can cap.
2. The two discrete power supply caps up under the choke (the transformer on the left when the unit is upright and viewing from the front)
3. The four coupling caps (these are the ones you originally questioned -- replace with .22 uF 630 volt units). Modern caps are physically smaller today for a given capacitance and voltage rating than those of yesteryear, therefore, the 630 volt caps offer greater safety than the old 400 volt caps, and are still smaller than the old 400 volt units were, as well.
4. The discrete electrolytic cap in the output tube area (use 22 uF for the old 20 uF there).
By replacing all of these caps, the unit will be safe to run, without fear of doing any collateral damage should one of the old caps really let go.
There are a couple of small rectangular "mica" caps as well that may have their value printed on them. These caps will last your lifetime and beyond. Therefore, replace them only if you are bent on everything being new.
Resistors do drift with age -- particularly the cheaper ones used in the kits like this unit was. It is a good idea to replace them as part of your rebuild.
Once you get it up and running, you will no doubt be very impressed with its performance -- and should be. These are great amps with wonderful output transformers. Be sure to give it plenty of ventilation -- and most particularly, the center transformer (the power transformer). It has a history of running hot and ultimately failing over time as a result. Keeping it cool -- even to the point of using forced air -- will prevent that problem.
Once you've gotten comfortable with this unit, consider the modifications presented in this thread. It will make for a notably better unit yet. Mates are rather routinely available at auction.
Good luck with your W4-AM!
Dave
1. The power supply can cap.
2. The two discrete power supply caps up under the choke (the transformer on the left when the unit is upright and viewing from the front)
3. The four coupling caps (these are the ones you originally questioned -- replace with .22 uF 630 volt units). Modern caps are physically smaller today for a given capacitance and voltage rating than those of yesteryear, therefore, the 630 volt caps offer greater safety than the old 400 volt caps, and are still smaller than the old 400 volt units were, as well.
4. The discrete electrolytic cap in the output tube area (use 22 uF for the old 20 uF there).
By replacing all of these caps, the unit will be safe to run, without fear of doing any collateral damage should one of the old caps really let go.
There are a couple of small rectangular "mica" caps as well that may have their value printed on them. These caps will last your lifetime and beyond. Therefore, replace them only if you are bent on everything being new.
Resistors do drift with age -- particularly the cheaper ones used in the kits like this unit was. It is a good idea to replace them as part of your rebuild.
Once you get it up and running, you will no doubt be very impressed with its performance -- and should be. These are great amps with wonderful output transformers. Be sure to give it plenty of ventilation -- and most particularly, the center transformer (the power transformer). It has a history of running hot and ultimately failing over time as a result. Keeping it cool -- even to the point of using forced air -- will prevent that problem.
Once you've gotten comfortable with this unit, consider the modifications presented in this thread. It will make for a notably better unit yet. Mates are rather routinely available at auction.
Good luck with your W4-AM!
Dave
Has anyone done the mods in this article?
http://www.stiftsbogtrykkeriet.dk/~mcs/Williamson/index.html
Some of the changes look to be in Dave's schematic.
There are some further thoughts in this link too.
http://www.chimeralabs.com/williamsmod.html
http://www.stiftsbogtrykkeriet.dk/~mcs/Williamson/index.html
Some of the changes look to be in Dave's schematic.
There are some further thoughts in this link too.
http://www.chimeralabs.com/williamsmod.html
Hi, Dave, a question please. In the first mod schematic that is in red, that is the mods you have recommended as I recall. The second schematic with both red and blue is a different mod that is not quite what you had recommended but included your mods, correct? As I see it, there are some changes in the coupling caps in red on the blue/red schematic that were not included in your original schematic, correct? And would those values, which if I am reading them correctly are 1 mfd from the phase splitter to the voltage/driver tubes, and .15 mfd to the output tubes. This is a big change in value from the .25 of the original values. Thanks, Randy
trobbins
Super Member
I'll add some queries in for Dave too 
(just joshing)
Wright in 1961 (http://dalmura.com.au/projects/williamson_verstaerker.pdf ) recommended increasing the driver stage common cathode value up to 1k. Increasing that resistor should allow more overload margin, a titch better inherent gain matching of the triodes, and drop the idle current by a mA or two. Of interest was your comment about the L63 perhaps not being too equivalent to the 6SN7 - Wright also seems to imply that there was a noticeable difference - however the Marconi datasheet curves for the L63 are pretty much an exact replica of any generic 6SN7 curves I've come across. Did you come across such a difference ?
You've intentionally unbalanced the anode resistors in that driver. Your added comment points to the Heathkit OT (part # 51-17) having a slight behavioural quirk to it - is that the correct interpretation?
With the coupling cap corner frequency splitting, you've lowered the capacitance used for the driver output, and increased the splitter output values. Most changes in the 1950's increased the driver output coupling capacitance to achieve a better split, and lower corner frequencies. Was your intent (apart from splitting the poles further apart) to aim at improving output stage blocking distortion recovery timing, rather than apparently to make it worse (as per Heathkit's W-5M change to 1uF on the driver output)?
Your step network on the first stage is much more 'severe', in terms of lowering the frequency where loop gain starts to drop, and continuing that dominant pole most likely through 0dB crossover. Did the OT need that level of stability bullet-proofing?
Ciao, Tim
(just joshing)Wright in 1961 (http://dalmura.com.au/projects/williamson_verstaerker.pdf ) recommended increasing the driver stage common cathode value up to 1k. Increasing that resistor should allow more overload margin, a titch better inherent gain matching of the triodes, and drop the idle current by a mA or two. Of interest was your comment about the L63 perhaps not being too equivalent to the 6SN7 - Wright also seems to imply that there was a noticeable difference - however the Marconi datasheet curves for the L63 are pretty much an exact replica of any generic 6SN7 curves I've come across. Did you come across such a difference ?
You've intentionally unbalanced the anode resistors in that driver. Your added comment points to the Heathkit OT (part # 51-17) having a slight behavioural quirk to it - is that the correct interpretation?
With the coupling cap corner frequency splitting, you've lowered the capacitance used for the driver output, and increased the splitter output values. Most changes in the 1950's increased the driver output coupling capacitance to achieve a better split, and lower corner frequencies. Was your intent (apart from splitting the poles further apart) to aim at improving output stage blocking distortion recovery timing, rather than apparently to make it worse (as per Heathkit's W-5M change to 1uF on the driver output)?
Your step network on the first stage is much more 'severe', in terms of lowering the frequency where loop gain starts to drop, and continuing that dominant pole most likely through 0dB crossover. Did the OT need that level of stability bullet-proofing?
Ciao, Tim
Wow - All great questions! You guys are really making me have to dig back in my memory, as the work on that amp was done so long ago.
First, I apologize for the hurky-jurky approach in which this modification was presented -- which is no ones particular fault. The backstory is that the presentation began life as private communication when the OP contacted me regarding HF stability concerns he had with his W4-AM, to which I then supplied him with the modifications I had developed for that model, which altered the NFB network, step network, and input network to address those issues. The success of that then led to giving him the rest of the modifications developed, that address LF stability issues and distortion reduction. The OP then approached me about making all of the modifications available to others by way of this thead, which resulted in the double schematic presentation, and posting my quotes from the private communication we had. So, the presentation was different, but the final (second) schematic does represent the accurate total modification developed for the unit. As to the specific questions presented then:
1. Randy -- The values you cite from the modified schematic are correct, and provide very good LF stability.
In general, the LF stability of an R/C coupled NFB amplifier suffers as the time constant (or poles) between the various coupling networks become closer together -- exactly as Tim eluded to. In the original design, there was a factor of just under 5 times between the constants of the two interstage coupling networks, which resulted in the very poor LF stability displayed by that design. With the modified design, there is now a factor of over 31 times separating the constants of the two networks, with the resulting improvement in LF stability that change provides.
As to why the spread in time constants was achieved as it was, it became merely matter relating to ease of execution. In any design using R/C coupling, blocking distortion becomes a possibility any point in the design if any grid reaches Eg = 0, and at any frequency. In any practical design of quality, this will almost always occur at the output stage, as it does in the W4-AM as well. In the W4-AM, when it does occurs, it particularly manifested itself at lower frequencies because the LF stability of the original design is so poor. As a basic point then, blocking distortion in an R/C coupled NFB amplifier cannot be avoided, but its effects can be greatly minimized (or nearly eliminated for all practical purposes) by the use of cathode bias at the stage where blocking can occur, and by designing (or improving in this case) LF stability to be of a very high order. The former effort acts like a shock absorber to limit the harshness of blocking distortion, while the latter seeks to prevent any uncontrolled LF instability oscillations from piling on to any blocking events that occur. Since the W4-AM employs cathode bias in the output stage, it merely became a matter of improving LF stability not only to directly improve that performance characteristic, but also to improve the effects of any blocking distortion events as an added bonus.
Since the effort then was to create the greatest possible spread between the time constant of the two coupling networks, this goal can be most easily effected by raising the value of the coupling cap at the stage employing the greatest grid return resistance value, and reducing the capacitance of the network employing the smallest grid return value, which is what was in fact done. With the first network's time constant being raised to the point of effectively being direct coupled in this exercise, it then became a matter of simply choosing a time constant for coupling into the output stage that worked well in conjunction with the other two significant LF time constants in the design (the output stage cathode bypass cap, and that of the output transformer), relative to achieving good combination of LF response, LF distortion performance, and LF stability. In the end, achieving a good combination of these (and other) performance characteristics is what separates a really great design, from the mediocre or poor designs, that seek to achieve one or two great specifications for the sake of the marketing department -- which is exactly what got the design of the original W-4 in trouble in the first place.
2. Tim -- Thank-you for your analysis of the modified design (always welcomed). To your points:
A. I've never worked with a real L63, but simply noted that as used in the original Williamson design, the 6SN7 is biased to draw needlessly excessive quiescent current, and so reduced it to a level that is more appropriate for the application, frankly as an aid to tube life and general heating of the amplifier and related components. That was my primary goal with the move, although it likely did produced a minor improvement in the distortion characteristics OF THAT STAGE. However, this is really a moot point, as in the original design, the driver stage is capable of delivering basically double the amount of drive needed for the output stage to develop full power output, so the major effect this move had on distortion reduction is basically limited to a portion of the driver output capability that was never used in the first place. As to adding a level of overload protection to the stage, the output stage is the first to overload in this design, with the driver grids not even needing 3 volts peak each to produce that condition. This is well under the bias level of that provided by even that of the original cathode resistor value for the driver stage, so yes, increasing the driver stage cathode bias resistor did improve overload protection -- but it was hardly needed in the first place. Rather, it was again, an added bonus as a result of addressing other aspects of the design.
As to Wright's work, I believe his findings are flawed, as the one huge element he did not address with his re-biasing work, is that by increasing the bias on all the "front end" stages, this significantly reduced the open loop gain (OLG) of the design. This has the effect of significantly reducing the amount of NFB applied after the loop is closed. Granted, he did readjust the value of the feedback resistor proportionately when he increased the value of the input stage cathode resistance, but the overall effect was one of notably reducing the amount of NFB applied.
It is important to understand that when incorrectly applied, NFB can be come quite ineffective, and even detrimental to the performance of an amplifier. Such was the case of the W4-AM that is the subject of this discussion. Note that by the modifications made -- which includes now bringing the NFB from the 8 Ohm tap -- the amplifier is light years more stable than it was, retains the vast majority of the frequency response displayed by the EARLY W-4 amplifiers, while also retaining most of the low distortion characteristics of the original design as well. I would invite you to check out what happened to the distortion and response characteristics of the late version W4 amplifier series, where Heath was forced to address the poor stability of the original design. To achieve a respectable level of HF stability, the distortion and response characteristics went right out the window compared to that of the original offering. It was a case of NFB not being correctly applied -- in either version. As well, note that the NFB coming from the 8 OHM tap has nothing to do with the popular notion that using the tap that is actually loaded to supply the NFB provides the best performance. In this case, using the 8 Ohm tap, supplemented by the HF characteristics available at the 4 Ohm tap, provides the greatest HF stability, regardless of which tap is actually loaded. I have found that properly compensated, it makes little if any difference which output tap is used for the NFB connection, and which is actually loaded. Relative to Wright's work then, my work strongly suggests that what Wright was actually experiencing was the effects of reducing the NFB (by way of reducing OLG), in a design in which it was poorly applied. In doing so, performance then actually improved, which he (wrongly in my opinion) attributed to the increased bias he applied to the front end stages. Interestingly, by his own words, the graphs Wright provides to support the improvement in performance his efforts produced do not even relate to the amplifier his text addresses at all, but merely "suggests" the level of improvement achieved! The performance data I supplied are all real numbers from my own Heath units modified as discussed.
B. The step work did in fact need to be that "severe", but with the new NFB loop, still produced a response that was perfectly flat to 20 kHz (down 1 db at 60 kHz), a very nice 10 kHz square wave form with good rise time and no ringing, and absolute stability: No amount of capacitance only loading would promote sustained HF oscillation. Importantly, the HF networks shown work equally well with both the Chicago OPT (as typically used on the early gray chassis units), and the Stancor transformer supplied with later units. Which leads to the final point, being:
C. The 220K shunt resistor shown across the plate load resistor of the "bottom" driver tube should only be installed in units with the Chicago OPT. For whatever the reason, there is an imbalance in these transformers requiring slightly unequal drive to each primary half to obtain minimum THD and IM distortion. I have noted this with a number of Chicago OPTs, as used in both Heath and Eico pieces. Most notably, the Chicago OPT used in the Eico HF-20/22/35 amplifiers displays the same exact imbalance, requiring the same amount of correction as well. This observation has been made in not just one or two of these units, but a few dozen anyway over the years -- each time requiring the same compensation -- such that I am entirely satisfied that it is a transformer related issue. In any event, the resistor should be installed as the OPT supplied so dictates.
I hope this covers it all!
Dave
First, I apologize for the hurky-jurky approach in which this modification was presented -- which is no ones particular fault. The backstory is that the presentation began life as private communication when the OP contacted me regarding HF stability concerns he had with his W4-AM, to which I then supplied him with the modifications I had developed for that model, which altered the NFB network, step network, and input network to address those issues. The success of that then led to giving him the rest of the modifications developed, that address LF stability issues and distortion reduction. The OP then approached me about making all of the modifications available to others by way of this thead, which resulted in the double schematic presentation, and posting my quotes from the private communication we had. So, the presentation was different, but the final (second) schematic does represent the accurate total modification developed for the unit. As to the specific questions presented then:
1. Randy -- The values you cite from the modified schematic are correct, and provide very good LF stability.
In general, the LF stability of an R/C coupled NFB amplifier suffers as the time constant (or poles) between the various coupling networks become closer together -- exactly as Tim eluded to. In the original design, there was a factor of just under 5 times between the constants of the two interstage coupling networks, which resulted in the very poor LF stability displayed by that design. With the modified design, there is now a factor of over 31 times separating the constants of the two networks, with the resulting improvement in LF stability that change provides.
As to why the spread in time constants was achieved as it was, it became merely matter relating to ease of execution. In any design using R/C coupling, blocking distortion becomes a possibility any point in the design if any grid reaches Eg = 0, and at any frequency. In any practical design of quality, this will almost always occur at the output stage, as it does in the W4-AM as well. In the W4-AM, when it does occurs, it particularly manifested itself at lower frequencies because the LF stability of the original design is so poor. As a basic point then, blocking distortion in an R/C coupled NFB amplifier cannot be avoided, but its effects can be greatly minimized (or nearly eliminated for all practical purposes) by the use of cathode bias at the stage where blocking can occur, and by designing (or improving in this case) LF stability to be of a very high order. The former effort acts like a shock absorber to limit the harshness of blocking distortion, while the latter seeks to prevent any uncontrolled LF instability oscillations from piling on to any blocking events that occur. Since the W4-AM employs cathode bias in the output stage, it merely became a matter of improving LF stability not only to directly improve that performance characteristic, but also to improve the effects of any blocking distortion events as an added bonus.
Since the effort then was to create the greatest possible spread between the time constant of the two coupling networks, this goal can be most easily effected by raising the value of the coupling cap at the stage employing the greatest grid return resistance value, and reducing the capacitance of the network employing the smallest grid return value, which is what was in fact done. With the first network's time constant being raised to the point of effectively being direct coupled in this exercise, it then became a matter of simply choosing a time constant for coupling into the output stage that worked well in conjunction with the other two significant LF time constants in the design (the output stage cathode bypass cap, and that of the output transformer), relative to achieving good combination of LF response, LF distortion performance, and LF stability. In the end, achieving a good combination of these (and other) performance characteristics is what separates a really great design, from the mediocre or poor designs, that seek to achieve one or two great specifications for the sake of the marketing department -- which is exactly what got the design of the original W-4 in trouble in the first place.
2. Tim -- Thank-you for your analysis of the modified design (always welcomed). To your points:
A. I've never worked with a real L63, but simply noted that as used in the original Williamson design, the 6SN7 is biased to draw needlessly excessive quiescent current, and so reduced it to a level that is more appropriate for the application, frankly as an aid to tube life and general heating of the amplifier and related components. That was my primary goal with the move, although it likely did produced a minor improvement in the distortion characteristics OF THAT STAGE. However, this is really a moot point, as in the original design, the driver stage is capable of delivering basically double the amount of drive needed for the output stage to develop full power output, so the major effect this move had on distortion reduction is basically limited to a portion of the driver output capability that was never used in the first place. As to adding a level of overload protection to the stage, the output stage is the first to overload in this design, with the driver grids not even needing 3 volts peak each to produce that condition. This is well under the bias level of that provided by even that of the original cathode resistor value for the driver stage, so yes, increasing the driver stage cathode bias resistor did improve overload protection -- but it was hardly needed in the first place. Rather, it was again, an added bonus as a result of addressing other aspects of the design.
As to Wright's work, I believe his findings are flawed, as the one huge element he did not address with his re-biasing work, is that by increasing the bias on all the "front end" stages, this significantly reduced the open loop gain (OLG) of the design. This has the effect of significantly reducing the amount of NFB applied after the loop is closed. Granted, he did readjust the value of the feedback resistor proportionately when he increased the value of the input stage cathode resistance, but the overall effect was one of notably reducing the amount of NFB applied.
It is important to understand that when incorrectly applied, NFB can be come quite ineffective, and even detrimental to the performance of an amplifier. Such was the case of the W4-AM that is the subject of this discussion. Note that by the modifications made -- which includes now bringing the NFB from the 8 Ohm tap -- the amplifier is light years more stable than it was, retains the vast majority of the frequency response displayed by the EARLY W-4 amplifiers, while also retaining most of the low distortion characteristics of the original design as well. I would invite you to check out what happened to the distortion and response characteristics of the late version W4 amplifier series, where Heath was forced to address the poor stability of the original design. To achieve a respectable level of HF stability, the distortion and response characteristics went right out the window compared to that of the original offering. It was a case of NFB not being correctly applied -- in either version. As well, note that the NFB coming from the 8 OHM tap has nothing to do with the popular notion that using the tap that is actually loaded to supply the NFB provides the best performance. In this case, using the 8 Ohm tap, supplemented by the HF characteristics available at the 4 Ohm tap, provides the greatest HF stability, regardless of which tap is actually loaded. I have found that properly compensated, it makes little if any difference which output tap is used for the NFB connection, and which is actually loaded. Relative to Wright's work then, my work strongly suggests that what Wright was actually experiencing was the effects of reducing the NFB (by way of reducing OLG), in a design in which it was poorly applied. In doing so, performance then actually improved, which he (wrongly in my opinion) attributed to the increased bias he applied to the front end stages. Interestingly, by his own words, the graphs Wright provides to support the improvement in performance his efforts produced do not even relate to the amplifier his text addresses at all, but merely "suggests" the level of improvement achieved! The performance data I supplied are all real numbers from my own Heath units modified as discussed.
B. The step work did in fact need to be that "severe", but with the new NFB loop, still produced a response that was perfectly flat to 20 kHz (down 1 db at 60 kHz), a very nice 10 kHz square wave form with good rise time and no ringing, and absolute stability: No amount of capacitance only loading would promote sustained HF oscillation. Importantly, the HF networks shown work equally well with both the Chicago OPT (as typically used on the early gray chassis units), and the Stancor transformer supplied with later units. Which leads to the final point, being:
C. The 220K shunt resistor shown across the plate load resistor of the "bottom" driver tube should only be installed in units with the Chicago OPT. For whatever the reason, there is an imbalance in these transformers requiring slightly unequal drive to each primary half to obtain minimum THD and IM distortion. I have noted this with a number of Chicago OPTs, as used in both Heath and Eico pieces. Most notably, the Chicago OPT used in the Eico HF-20/22/35 amplifiers displays the same exact imbalance, requiring the same amount of correction as well. This observation has been made in not just one or two of these units, but a few dozen anyway over the years -- each time requiring the same compensation -- such that I am entirely satisfied that it is a transformer related issue. In any event, the resistor should be installed as the OPT supplied so dictates.
I hope this covers it all!
Dave
trobbins
Super Member
Good on ya Dave - very generous of you to provide that descriptive insight.
That issue with the Chicago OPT's is certainly interesting. Do you think the amp was moving in to class B for the higher power levels where distortions were being reported? The required symmetry within the OPT goes to another level of complexity when trying to design for good distortion with class AB1 operation, compared to just class A. I also expect that the common modification of bypassing the output stage common cathode resistance was aimed at constraining distortion rise as class B operation was encroached upon in the endeavour to raise power ratings.
I don't think Wright's rebiasing of the input stage does significantly change open-loop gain. The 6SN7 operating point Ri and u changes from rebiasing are pretty minor. The input stage's stage gain (Vo1/Vgk) is not affected, nor is the gain with feedback connected. The feedback resistance has to be lowered to restore the same operating conditions, with Morgan Jones in his 3rd Ed book showing how R4 influences the situation. The gain of the first stage from the amp input (Vo1/Vin) with feedback disconnected (or at very low or high frequency operation) only drops from 23 to 22dB (based on u=20, Ri=9k2), although the effective Ri increases from 19k to 28k due to the influence of (u+1).R4 (R4 from Williamson's circuit). The phase splitter stage gain would only have negligible change.
I certainly found Wright's article to only show sparse results to support his proposed changes, and results were just subjectively presented. Maybe it was a limit on article size.
Ciao, Tim
That issue with the Chicago OPT's is certainly interesting. Do you think the amp was moving in to class B for the higher power levels where distortions were being reported? The required symmetry within the OPT goes to another level of complexity when trying to design for good distortion with class AB1 operation, compared to just class A. I also expect that the common modification of bypassing the output stage common cathode resistance was aimed at constraining distortion rise as class B operation was encroached upon in the endeavour to raise power ratings.
I don't think Wright's rebiasing of the input stage does significantly change open-loop gain. The 6SN7 operating point Ri and u changes from rebiasing are pretty minor. The input stage's stage gain (Vo1/Vgk) is not affected, nor is the gain with feedback connected. The feedback resistance has to be lowered to restore the same operating conditions, with Morgan Jones in his 3rd Ed book showing how R4 influences the situation. The gain of the first stage from the amp input (Vo1/Vin) with feedback disconnected (or at very low or high frequency operation) only drops from 23 to 22dB (based on u=20, Ri=9k2), although the effective Ri increases from 19k to 28k due to the influence of (u+1).R4 (R4 from Williamson's circuit). The phase splitter stage gain would only have negligible change.
I certainly found Wright's article to only show sparse results to support his proposed changes, and results were just subjectively presented. Maybe it was a limit on article size.
Ciao, Tim
Hi Tim -- Thanks again for your analysis. I agree that the changes in gain produced at each stage from Wright's modifications will be small. However, with the stages all being cascaded, it means that the effects of gain reduction in each stage will be accumulative. A quick exercise I ran then suggests that implementing all of his biasing suggestions results in reducing OLG to about 75% of the original value, with a near similar reduction produced in the FB factor as a result. The majority of OLG reduction is produced by the increased current FB produced across the (basically) doubled value of the input stage cathode bias resistor. Against this reduction in OLG, Wright simply made the new FB resistor value (required as a result of his changes) proportional to the new input stage cathode bias resistance value, in the same way that these components were originally related before his modifications were applied. Therefore, his modifications do result in a degree of FB reduction.
However, since no real results were provided for the modifications suggested, the original Williamson circuit would need to be mocked up to do some before and after tests, as well as open and closed loop testing for verification of Wright's claims. It is an interesting discussion, so maybe I will do that at some later date. But for now, life has my plate over flowing. In any event, I again thank-you for your analysis, as engineering for good sound starts with good engineering to begin with!!
The W4 amplifier series is ultimately a Class AB1 design. But due to the use of a 10K Ohm plate-to-plate impedance OPT and the use of 50% UL taps, it extends the Class A portion of operation notably further over that of more conventional 6L6 based UL designs.
Dave
However, since no real results were provided for the modifications suggested, the original Williamson circuit would need to be mocked up to do some before and after tests, as well as open and closed loop testing for verification of Wright's claims. It is an interesting discussion, so maybe I will do that at some later date. But for now, life has my plate over flowing. In any event, I again thank-you for your analysis, as engineering for good sound starts with good engineering to begin with!!
The W4 amplifier series is ultimately a Class AB1 design. But due to the use of a 10K Ohm plate-to-plate impedance OPT and the use of 50% UL taps, it extends the Class A portion of operation notably further over that of more conventional 6L6 based UL designs.
Dave
Last edited:
Dave, first of all, thanks so much for taking the time to share your thoughts and advice. It is appreciated!
I noticed in this thread that someone reported you are a fan of EICO amps (specifically including the HF-22 and -35) and that you have worked up some mods for them. I am familiar with your mods for the HF-60 but I don't recall seeing anything about the HF-22 or HF-35. I realize the circuits are all very similar but I am curious whether you have any specific comments or recommendations for those amps. Thanks in advance.
I noticed in this thread that someone reported you are a fan of EICO amps (specifically including the HF-22 and -35) and that you have worked up some mods for them. I am familiar with your mods for the HF-60 but I don't recall seeing anything about the HF-22 or HF-35. I realize the circuits are all very similar but I am curious whether you have any specific comments or recommendations for those amps. Thanks in advance.
I've always had a lot of problems with the HF-22 and HF-35 amplifiers, for various individual as well as common reasons:
1. The HF-22 has very marginal drive capabilities into the output stage, which is particularly pronounced at low frequencies due to the roll off produced by the time constant of the coupling networks.
2. The OPT that is shared between these two models -- while a superb device -- is really out of its league for use in a 35 watt amplifier. To that point, the HF-35 can just manage 35 watts at mid frequencies, but at 20 Hz, 27 watts is about all she wrote. Now I realize that this power level basically complies with the "within 1 db of 35 watts" language that Eico uses to specify their power output and distortion specifications for this model. But in this case, I believe that very common power/distortion language is being used to cover for a transformer that is hopelessly overpowered at 20 Hz above 27.0 watts of power output. And the idea of the amplifier meeting this power level while producing no more than .5% THD at 20 Hz as specified is the stuff that the Sunday comics are made of. The intent of the language is to ensure that measurements are not made with the amplifier clipping, and to account for general component tolerance as well -- NOT to make cover for components of inadequate specification for the design. Note that the OPTs in the HF-87 are much more appropriate for the application they serve in.
3. As with other Eico designs, the NFB/HF stability network design for both units favors frequency response specifications (the darling of the day), at the expense of producing notably poor stability performance. Not to solely blame Eico, but Heath, Fisher, and others had the very same problem, with all of them having to make mid production corrections to add stability to their units, and Heath even having to come out with new models to because their prior units gained such...er.....notoriety.
For all of the above reasons, I have always felt that the very best performance from these units is produced by using a combination of their attributes:
A. Use an HF-35 chassis and transformer set, with 5V4 GA rectifier tube and a decent choke (a 3 Hy 200 mA 60 Ohm choke is about the biggest you can physically mount next to the rectifier tube), but use 6L6 class output tubes with a Williamson style DC balance control. This makes for an oversized power supply compared to the power transformer offered with the HF-20/22 amplifiers, and provides for delayed, slow rise B+ as well.
B. Use the Williamson front end as used in the HF-20, but with greatly improved NFB and HF/LF stability modifications, similar to those I developed for the amplifier of this thread. Use a 12AU7 where the EF86 once was, while the 6SN7 can remain in place, now acting as the driver tube of the Williamson front end.
I have two HF-35 amplifiers that are built with these modifications, and their performance is really superb, providing 22 watts RMS from 20 Hz to 20 kHz, with all distortions below 1% at this power level (except for rising to about 1.15% at 20 kHz), with a frequency response that is +0/-.5 db to 70 kHz, displaying an excellent 10 kHz square wave, and absolute stability. It really makes for the best possible performance of all of the HF-20/22/35 circuit possibilities.
I have found that the best sound from a NFB amplifer is always produced when both frequency response, AND good transient stability are achieved together, along with low distortion and adequate power output. Designs that champion frequency response at the expense of good stability will always have a degree of muddiness and/or harshness depending on how bad or at what end of the spectrum the instability occurs. The modifications I developed for both the W4-AM and the Eico amplifiers first and foremost champion producing a much better combination of performance characteristics as described above, while correcting other basic issues along the way.
Dave
1. The HF-22 has very marginal drive capabilities into the output stage, which is particularly pronounced at low frequencies due to the roll off produced by the time constant of the coupling networks.
2. The OPT that is shared between these two models -- while a superb device -- is really out of its league for use in a 35 watt amplifier. To that point, the HF-35 can just manage 35 watts at mid frequencies, but at 20 Hz, 27 watts is about all she wrote. Now I realize that this power level basically complies with the "within 1 db of 35 watts" language that Eico uses to specify their power output and distortion specifications for this model. But in this case, I believe that very common power/distortion language is being used to cover for a transformer that is hopelessly overpowered at 20 Hz above 27.0 watts of power output. And the idea of the amplifier meeting this power level while producing no more than .5% THD at 20 Hz as specified is the stuff that the Sunday comics are made of. The intent of the language is to ensure that measurements are not made with the amplifier clipping, and to account for general component tolerance as well -- NOT to make cover for components of inadequate specification for the design. Note that the OPTs in the HF-87 are much more appropriate for the application they serve in.
3. As with other Eico designs, the NFB/HF stability network design for both units favors frequency response specifications (the darling of the day), at the expense of producing notably poor stability performance. Not to solely blame Eico, but Heath, Fisher, and others had the very same problem, with all of them having to make mid production corrections to add stability to their units, and Heath even having to come out with new models to because their prior units gained such...er.....notoriety.
For all of the above reasons, I have always felt that the very best performance from these units is produced by using a combination of their attributes:
A. Use an HF-35 chassis and transformer set, with 5V4 GA rectifier tube and a decent choke (a 3 Hy 200 mA 60 Ohm choke is about the biggest you can physically mount next to the rectifier tube), but use 6L6 class output tubes with a Williamson style DC balance control. This makes for an oversized power supply compared to the power transformer offered with the HF-20/22 amplifiers, and provides for delayed, slow rise B+ as well.
B. Use the Williamson front end as used in the HF-20, but with greatly improved NFB and HF/LF stability modifications, similar to those I developed for the amplifier of this thread. Use a 12AU7 where the EF86 once was, while the 6SN7 can remain in place, now acting as the driver tube of the Williamson front end.
I have two HF-35 amplifiers that are built with these modifications, and their performance is really superb, providing 22 watts RMS from 20 Hz to 20 kHz, with all distortions below 1% at this power level (except for rising to about 1.15% at 20 kHz), with a frequency response that is +0/-.5 db to 70 kHz, displaying an excellent 10 kHz square wave, and absolute stability. It really makes for the best possible performance of all of the HF-20/22/35 circuit possibilities.
I have found that the best sound from a NFB amplifer is always produced when both frequency response, AND good transient stability are achieved together, along with low distortion and adequate power output. Designs that champion frequency response at the expense of good stability will always have a degree of muddiness and/or harshness depending on how bad or at what end of the spectrum the instability occurs. The modifications I developed for both the W4-AM and the Eico amplifiers first and foremost champion producing a much better combination of performance characteristics as described above, while correcting other basic issues along the way.
Dave
trobbins
Super Member
I'm interested to check out the mid-band gain changes, and should get my working W on the bench during October to do some simple comparison tests. The W is the 1949 'new version' circuit but with screen regulated 807's in to a WWFB. I basically restored it 2 years ago but had to put it away before any stability tests could be done, so checking out the Wright changes will be a good catalyst.
Your results will be interesting to see -- although any stability assessments made will be for that unit only. Since FB and stability networks are so OPT dependent, the modifications I posted for this thread relate only to the Heath W4-AM and its UL output stage. Still, using the stock configuration of your W to establish baseline information and then installing Wright's changes will certainly shed some light on the effectiveness of his changes. Changing your W to operate with regulated screen grids no doubt significantly altered the stability performance of the original W design, due to much higher plate resistance the output tubes display when operating in pentode mode.
Dave
Dave
Dave, I still have a couple questions after reading this thread and looking at your mods for the HF-60. While not technically part of the Heathkit discussion, it is related so I hope the other readers won't mind.
1. I am puzzled why the driver stage in the HF-22 and HF-35 would have "marginal drive capabilities." I would have expected a long-tail pair 6SN7 to be able to drive 6L6GC or EL-34 outputs without breaking a sweat. Is the B+ for the driver stage not high enough? Would a negative voltage reference for the cathode (like your mod to the HF-60) cure the problem?
2. Do you prefer the Williamson input/driver such as the Heathkit mod described here over any form of the long-tail pair driver?
3. If so, would an even simpler driver be even better, namely, just a split-load inverter driving the outputs directly like the Dyna designs?
I am trying to decide which topology to use for my next DIY amps. They will use KT-66 outputs and Chicago OPTs from Eico HF-22 amps. Everything else is open for discussion. The input/driver could be your modified Heathkit circuit, or the Eico EF-86/6SN7 LTP, or EF-86/12AU7 split-load inverter, or something else. I am looking for ideas and recommendations. Thanks!
1. I am puzzled why the driver stage in the HF-22 and HF-35 would have "marginal drive capabilities." I would have expected a long-tail pair 6SN7 to be able to drive 6L6GC or EL-34 outputs without breaking a sweat. Is the B+ for the driver stage not high enough? Would a negative voltage reference for the cathode (like your mod to the HF-60) cure the problem?
2. Do you prefer the Williamson input/driver such as the Heathkit mod described here over any form of the long-tail pair driver?
3. If so, would an even simpler driver be even better, namely, just a split-load inverter driving the outputs directly like the Dyna designs?
I am trying to decide which topology to use for my next DIY amps. They will use KT-66 outputs and Chicago OPTs from Eico HF-22 amps. Everything else is open for discussion. The input/driver could be your modified Heathkit circuit, or the Eico EF-86/6SN7 LTP, or EF-86/12AU7 split-load inverter, or something else. I am looking for ideas and recommendations. Thanks!
Addressing your questions:
1. The drive problem is with the HF-22 -- not the HF-35, and as I indicated, primarily at the lowest frequencies.
As designed, the 6SN7 long tailed pair can in fact provide adequate drive for the HF-50/60 designs -- but it is already very near it's limits even in that application, being barely (and I mean barely) capable of 40 volts peak per side at 20 Hz at eh output tube grids under full power conditions. This is why the stage had to be completely reworked to properly drive 6550 tubes with low distortion, as the original driver design gives out of clean drive well before the output stage is driven to full power with those tubes.
In down sizing from these models, the HF-35 still works well because the B+ supply to the stage is unchanged from the higher output units, and drive requirements are nearly the same as well. Even though output stage bias voltage is slightly lower in the HF-35 under quiescent conditions, its average cathode bias voltage rises to effectively the same level of bias as used in the higher output fixed bias models, when full power output is achieved. Therefore, drive conditions remain basically the same. The time constant of the output stage R/C coupling network is reduced which works against the inverter at the lowest frequencies, but it none the less seems to just squeak by in this model.
In the HF-22 however, two minor changes are enough to cause the driver to misstep notably under full power low frequency conditions. First, the driver stage B+ is about 30 volts lower -- certainly what you don't need in that model, and second, the average cathode bias voltage on the 6L6s rises to about 41 vdc under full power conditions at 20 Hz. Oops. The slight reduction in B+ coupled with the slight increase in drive requirement is enough for the driver stage to cry uncle in the HF-22. It is also hampered with the same reduced R/C time constant it shares with the HF-35 versus that used in the bigger brothers, which just adds insult to injury.
In any proper design, the driver stage should have ample (reserve) drive into the output stage at any frequency within the audio spectrum, so that distortion from the stage is minimized. But that just isn't the case with the HF-22.
2. I love the performance of the long tailed pair. But to directly drive the higher bias tubes (no intervening stages) with enough reserve capability so as to effectively remove the driver from the distortion picture means that you've got to juice it up pretty high. In the modified HF-60 design I published, the inverter is effectively operating from what, about 550 vdc under quiescent conditions -- this so that it will still have plenty of reserve capability under worst case conditions -- that being at 20 Hz, under full power conditions, when the B+ has sagged about 35 volts or so. These scenarios are really helped if a really stiff or regulated B+ supply is used to eliminate the sag factor. Ultimately however, the Williamson driver stage is the way to go for worry free drive capability. Getting the stability right is a bigger effort with the Williamson front end, but drive will never be a problem in traditional designs.
3. Your talking about a Williamson "type" driver stage, where the actual driver stage is eliminated. This of course was the effort that Laurent developed with the original Dynaco amplifiers. In the MK II, MK IV, MK VI, and Stereo 70 designs (as well as in their smaller 6BQ5 amplifiers), this approach works well enough, with the 6AN8 getting top billing for drive capability. However, in the MK III, the switch was made to high bias tubes, taxing the 6AN8 driver stage to the max requiring it to produce 110 volts pk-pk under full power conditions. It takes a very healthy 6AN8 to be able to do that at 20 Hz, and once again, introducing driver stage distortion into the mix.
For your new project, the Williamson front end would be a slam dunk as long as stability at both ends of the spectrum are adequately addressed. If the long tail pair is considered, much would depend on the available B+ voltage for the stage, and whether your output stage operates in pentode, triode, or UL mode.
Dave
1. The drive problem is with the HF-22 -- not the HF-35, and as I indicated, primarily at the lowest frequencies.
As designed, the 6SN7 long tailed pair can in fact provide adequate drive for the HF-50/60 designs -- but it is already very near it's limits even in that application, being barely (and I mean barely) capable of 40 volts peak per side at 20 Hz at eh output tube grids under full power conditions. This is why the stage had to be completely reworked to properly drive 6550 tubes with low distortion, as the original driver design gives out of clean drive well before the output stage is driven to full power with those tubes.
In down sizing from these models, the HF-35 still works well because the B+ supply to the stage is unchanged from the higher output units, and drive requirements are nearly the same as well. Even though output stage bias voltage is slightly lower in the HF-35 under quiescent conditions, its average cathode bias voltage rises to effectively the same level of bias as used in the higher output fixed bias models, when full power output is achieved. Therefore, drive conditions remain basically the same. The time constant of the output stage R/C coupling network is reduced which works against the inverter at the lowest frequencies, but it none the less seems to just squeak by in this model.
In the HF-22 however, two minor changes are enough to cause the driver to misstep notably under full power low frequency conditions. First, the driver stage B+ is about 30 volts lower -- certainly what you don't need in that model, and second, the average cathode bias voltage on the 6L6s rises to about 41 vdc under full power conditions at 20 Hz. Oops. The slight reduction in B+ coupled with the slight increase in drive requirement is enough for the driver stage to cry uncle in the HF-22. It is also hampered with the same reduced R/C time constant it shares with the HF-35 versus that used in the bigger brothers, which just adds insult to injury.
In any proper design, the driver stage should have ample (reserve) drive into the output stage at any frequency within the audio spectrum, so that distortion from the stage is minimized. But that just isn't the case with the HF-22.
2. I love the performance of the long tailed pair. But to directly drive the higher bias tubes (no intervening stages) with enough reserve capability so as to effectively remove the driver from the distortion picture means that you've got to juice it up pretty high. In the modified HF-60 design I published, the inverter is effectively operating from what, about 550 vdc under quiescent conditions -- this so that it will still have plenty of reserve capability under worst case conditions -- that being at 20 Hz, under full power conditions, when the B+ has sagged about 35 volts or so. These scenarios are really helped if a really stiff or regulated B+ supply is used to eliminate the sag factor. Ultimately however, the Williamson driver stage is the way to go for worry free drive capability. Getting the stability right is a bigger effort with the Williamson front end, but drive will never be a problem in traditional designs.
3. Your talking about a Williamson "type" driver stage, where the actual driver stage is eliminated. This of course was the effort that Laurent developed with the original Dynaco amplifiers. In the MK II, MK IV, MK VI, and Stereo 70 designs (as well as in their smaller 6BQ5 amplifiers), this approach works well enough, with the 6AN8 getting top billing for drive capability. However, in the MK III, the switch was made to high bias tubes, taxing the 6AN8 driver stage to the max requiring it to produce 110 volts pk-pk under full power conditions. It takes a very healthy 6AN8 to be able to do that at 20 Hz, and once again, introducing driver stage distortion into the mix.
For your new project, the Williamson front end would be a slam dunk as long as stability at both ends of the spectrum are adequately addressed. If the long tail pair is considered, much would depend on the available B+ voltage for the stage, and whether your output stage operates in pentode, triode, or UL mode.
Dave
trobbins
Super Member
I just spent some time checking basic distortion performance of the input stages of a Williamson, and also introduced some of Wright's 1961 changes for comparison. The assessments were mainly of 2nd harmonic distortion changes (3rd and higher were well down), with open-loop operation (feedback disconnected), and probing with a 1 Megohm loading.
The Wright change (cathode resistance increased from 470 to 940 ohm, and power supply dropped) to the input stage drops stage gain, from 23.1 to 21.5dB (textbook drop was calculated as 23 to 22dB). For the same signal output level, distortion had increased a bit. Retaining an unbypassed 470 ohm, but bypassing the additional 470 ohm that was added (so as to retain the feedback ratio but still move the bias up), increased distortion further.
The Wright change to the driver stage (increasing the common cathode resistance from 390 to 860 ohm) increased distortion by about 10%.
The driver stage distortion is dominant, and tube rolling was well worth it to identify a reasonable DC balance, AC gain balance, distortion balance (to allow even orders to cancel later on), and low distortion. The best tube in my stash of 14 was a blackened glass Brimar 6SN7GT. 12AU7 certainly shows up with typically over twice the distortion.
I'm just using a low distortion sinewave source and REW5.1 for distortion measurement, so can't easily check out IM distortion. Now to include the output stage.
The Wright change (cathode resistance increased from 470 to 940 ohm, and power supply dropped) to the input stage drops stage gain, from 23.1 to 21.5dB (textbook drop was calculated as 23 to 22dB). For the same signal output level, distortion had increased a bit. Retaining an unbypassed 470 ohm, but bypassing the additional 470 ohm that was added (so as to retain the feedback ratio but still move the bias up), increased distortion further.
The Wright change to the driver stage (increasing the common cathode resistance from 390 to 860 ohm) increased distortion by about 10%.
The driver stage distortion is dominant, and tube rolling was well worth it to identify a reasonable DC balance, AC gain balance, distortion balance (to allow even orders to cancel later on), and low distortion. The best tube in my stash of 14 was a blackened glass Brimar 6SN7GT. 12AU7 certainly shows up with typically over twice the distortion.
I'm just using a low distortion sinewave source and REW5.1 for distortion measurement, so can't easily check out IM distortion. Now to include the output stage.
trobbins
Super Member
Another batch of testing on a Williamson - this time Intermodulation SMPTE measurements using REW 5.16's easy generator and spectrum analyser via a $1 usb soundcard.
The comparison was stock driver stage versus Wright 1961 modification. The amp didn't have feedback connected, and was a stock Williamson circuit except for 807's running pentode mode. I couldn't identify any change in IM results for output levels from 0.7W to 5W.
IM testing appears to me to be a much more sensitive method than simple harmonic or THD testing for final tuning of static and AC balance of a Williamson circuit - and provides a high level of confidence that the amp is well set up distortion wise.
The comparison was stock driver stage versus Wright 1961 modification. The amp didn't have feedback connected, and was a stock Williamson circuit except for 807's running pentode mode. I couldn't identify any change in IM results for output levels from 0.7W to 5W.
IM testing appears to me to be a much more sensitive method than simple harmonic or THD testing for final tuning of static and AC balance of a Williamson circuit - and provides a high level of confidence that the amp is well set up distortion wise.
I just stumbled upon this thread as a result of a friend of mine (Derekva) mentioning that he's getting into a pair of W4s and doing some of this work. As I too have a set of these awaiting some attention, I think I will now 'have at it' and see if I can't get them up and running. I still have a LOT of reading to do here, and am running out of time at the moment......so I wanted to 'post'...'subscribe'....and just ask: Did I miss the posting of the EL34 mod? I went back twice and never saw it. Just wondering if I missed it, or if it never got posted.
Also..... to the OP (Mag).... GREAT JOB!! I've gotta ask though; when you did your 'complete' resto to the amps, they seem to appear as having gone from the gray chassis versions to the 'gold' chassis versions. I didn't see any comments about this either....... but I'm wondering if there were actually 4 different amps here, if you actually swapped chassis with some 'nos' ones you might have had, or if you actually refinished your old gray chassis and re-stenciled them? In any event, the final versions in the wood bases look great.....but I had to ask about the chassis changes as well as mine are rust-buckets and will need to either be painted or replaced.
Dave......as always: WoW!!!
Tom D.
Also..... to the OP (Mag).... GREAT JOB!! I've gotta ask though; when you did your 'complete' resto to the amps, they seem to appear as having gone from the gray chassis versions to the 'gold' chassis versions. I didn't see any comments about this either....... but I'm wondering if there were actually 4 different amps here, if you actually swapped chassis with some 'nos' ones you might have had, or if you actually refinished your old gray chassis and re-stenciled them? In any event, the final versions in the wood bases look great.....but I had to ask about the chassis changes as well as mine are rust-buckets and will need to either be painted or replaced.
Dave......as always: WoW!!!
Tom D.