The JLH Gateway to Class-A :: 10

The Master's Voice ...  and More

        Building the JLH Classic amplifier takes quite a bit of artistry in that one must judge the 'musicality' of the amplifier -- a factor that its designer had set much store by. For JLH the designer, engineering concepts were only the means to achieve 'subjective truth in amplification'. But of course, he was not a pure subjectivist; nor was he fully in the 'objective/measurements' camp. As he easily walked the middle ground, we could see that  his genius and artistry lay when, with aplomb, he blended the art and science of amplifier design. Half a century (and more!) has passed since he had given the world his 'simple' amplifier, but still it continues to exude audio truth as he had envisaged.

        Over the past decades, especially after the passing of the great man, questions and doubts of the JLH enthusiasts and builders were answered largely by a dedicated few, who themselves were designers of accepted stature. My initial interest in the JLH which had petered out after a couple of builds and auditions back in the 1970s somehow re-surfaced when JLH had released some high powered and highly acclaimed amplifier designs, but which were beyond my reach in those days for many reasons. But somehow the JLH 'simple amplifier' managed to stay at the back of my mind.

Notes of Wisdom

In the meantime the one good thing that I had kept at regularly was my "trawling" of the JLH thread in Diyaudio. I thank Providence that I had taken notes from the thread off and on. When the idea of again building the classic amplifier enthused me, my notes were already voluminous. These notes, amply supplemented by "fishing expeditions" in the covid times and later have been 'distilled' into what could be called the best collection of peer help for the DIYer and JLH enthusiast. I am presenting them here, again in no strict chronological order, with the hope that it will benefit the community. I am sure a read will enable you to pick the brains of many stalwarts in the field who hand-held many a builder in the Forum threads. 

        I am starting off the Notes with some of the instructions from the Master himself. 

John Linsley Hood: 

"Relatively few design decisions now remained. Of these, the most significant was whether to give the second, class-A amplifier stage a bootstrapped collector load or to use a constant current source. The merits of these two arrangements are fairly well balanced. For example, the bootstrapped driver will give a slightly larger output power for the same HT voltage, and is simpler, while the constant current source arrangement will give a slightly lower second harmonic distortion figure which allows the same overall THD figure to be obtained with a lower open-loop gain - which makes the preceding circuitry simpler"
-----

" In order to minimise the effect of common mode distortion at the input, while preserving the facility of a direct coupled structure, the customary ' long- tailed pair' has been replaced by a single transistor, in which the offset which would otherwise arise due to its base-emitter voltage and the emitter circuit current flow through the feedback resistor, is removed by an additional current source employed to inject a sufficient quantity of current into the emitter of the input transistor to offset this voltage. Again, since the input transistor remains at ambient temperature, the constant current source can be arranged to track this thermally, and remove errors in dc offset due to ambient temperature changes." 

-----

The constructor of the circuit should make adjustments to the valure of R2 (not R1, which is part of the bootstrap circuit) to obtain the correct standing current... (P-149) "the resistor R2 can be used to set the static current of the output stage." The use of a variable resistor, in series with some suitable fixed value, would have facilitated the setting of this; the potentiometer should also be at the end of R2 nearest to R1. 
-----

# The calculation of the output voltage/current usually done is valid for symmetrical waveforms --speech and music waveforms are often not so, and so allowance should be made for this.

# Real loudspeaker loads are not purely resistive, but reactive, and their impedance might fall to much lower than the nominal value.

# For optimum performance of the output stage, current swings should not take either transistor into current cutoff.

This necessitates a quiescent current in excess of the theoretical minimum "sine wave into resistive load" value.
-----

" It is possible to reduce the r.f. response of the amplifier to give a smooth 6dB roll-off beyond 50kHz – which removes much of the need for care in the layout of components, without detriment to the harmonic distortion in the audible range, and without any audible alteration to the performance – by connecting a 1,000pF capacitor between the collector of Tr3 and the emitter of Tr4; a 1,000pF in series with 100 ohms between the collector of Tr4 and earth; and a 0.01mF in series with 8 to 10 ohms between the output (‘X’) and earth. (It should be noted that either all of these components should be added or none at all, they are not alternatives."
-----
"Listener fatigue" is often associated with solid state amplifiers, because most of these are able to deliver large amounts of power at supersonic frequencies, which the speakers in a high quality system will endeavour to deliver to the listener. In this context it should be remembered that in an amplifier which has a flat power response from 30 Hz to 180 kHz, 90% of this power spectrum will be supersonic.

-----
"Because the transition frequency of the output transistors is of the order of 4MHz, whereas those of Q3 and Q4 are in the 400MHz range, the circuit has an in-built dominant lag in its loop NFB characteristics. This ensures that the loop gain has fallen below unity before the loop phase angle reaches 180 degrees. No additional HF compensation networks are therefore necessary to ensure complete stability, even with reactive loads." 

* * * * *

Class-A versus Class-B: "... the 'slight edginess' in music reproduction often observed with Class-B amplifiers is the subjective effect of crossover distortion which tends to rise with reducing power output and with increasing frequency. High string notes, whose harmonics run up to quite high orders, would be more affeted by this than LF of greater sound intensity. The crossover effect produces a whole series of odd harmonics extending to high orders, which to the ear are inharmonious, to say the least.

In many modern amplifier designs, power bandwidth extends to very high frequencies. While a passband in reasonable excess of the audible spectrum is desirable for  maintaining waveform shape, it is my preference to have a slow roll-off at a rate of about 6 dB/octave, so that disturbing harmonics are eliminated. Although such harmonics might fall above audibility, they can certainly interfere with the complex music waveforms, creating "listener fatigue". -- Gordon J King
-----
The 1969 version, single rail with output capacitor, sounds plenty superb.

I would recommend you consider the addition of an R-C filter to the supply rail feeding the phase splitter, or if no large C available, use a capacitance multiplier in the rail since it acts to improve the PSRR of the amplifier without any obvious downside (needs a large value C). You don't need a cap-multi on the main power rail-- its a waste of power. In my version of this amp ( TGM9 thread) I found that you get most of the PSRR benefits simply by isolating the front end. You can put a low power cap multiplier on the rail to the front end-- I found it sufficient to use a resistor and large cap, and it is silent. -- Bigun
-----
Volt Reg for input/driver - With the addition of a a capacitance multiplier/regulator in the power rail, between the front end and the output stage we achieve three things:

i) no voltage limitation on the output stage which can access the full power supply voltage - the input stage can swing high because of the bootstrap

ii) drastically improved PSRR performance of the amplifier

iii) slow turn-on to eliminate the switch-on thump -- Bigun   

-----

Another consideration is how the hfe of your output power transistor varies with current. In my version of this amp I chose to use parallel devices (x4 + x4 for one ch) and set the current so that the DC operating point was near peak hfe and unlikely to be pushed into beta droop. Hiraga was my inspiration for this consideration-- he, in one of his articles, recommends setting current for the small signal transistors at a point where hfe varies symmetrically around the dc operating point. -- Bigun
----
CFP VAS option: I modified the amp by installing a Sziklai pair in place of TR3, the phase splitter/VAS. I used a 1k collector load for the master device across which the slave device derives its control voltage. The higher Hfe of this pair required some adjustment of the input device bias to keep the output at the mid point of the power rails.

I installed a 33p np0 cap across the b-c of the master device to ensure similar phase margin to the version I had before. Simulations say this will provide a similar UGF but with a good bump in OLG and a corresponding decrease in harmonic distortion, especially 3rd order and above.

A 1kHz square wave produced a clean output waveform without ringing.

Sound: I was not happy with this change. The magic seemed to be gone and the sound became a bit sterile. It was a cleaner sound, but bass was not as deep even if still well defined. The mids were 'flat' and overall I felt some irritation although I would not describe it as harsh. I would be happy with a cleaner sound but not an irritating sound. Interesting that my TGM2 gave me a similar feeling, it was an experiment with using a CFP LTP input stage compared with a regular LTP. Chances are that I'll remove the CFP and go back to a single device for TR3.

I ripped out the CFP VAS and put back the single non transistor. The readjusted the dc-offset to half the rail voltage.

The sound is back ! - I don't think I feel any more temptation to mess with it. Now it's time to sit back and enjoy! This version of the JLH69 is now highly recommended! -- Bigun
-----
Whether floating ground or not, there are going to be at least two capacitors in the speaker current path. With floating ground there's a capacitor from each rail to the speaker return point (like a split rail design), and with the non-floating there's the output cap and the rail-to-ground cap. In both cases you have two caps, both outside of the feedback loop. So why would they sound different ?

The floating ground puts power supply ripple and speaker return current in the same pair of caps. The non-floating version doesn't mix signal and power supply ripple in the main output cap. I'm surprised people can hear a difference and would like to know why.

Perhaps it matters where you 'ground' the feedback shunt capacitor - ideally it should be connected to the return side of the speaker not the input signal ground because the series - shunt feedback resistors are a potential divider connected across the speaker. -- Bigun
-----
Completely removing hum from a JLH class-A is not difficult.

A simple series 1 ohm resistance filter with additional 10,000uF power rail capacitor will reduce 100Hz power rail ripple by approx 16dB or a factor of 6.

With a single power rail version the induced hum arises as it comes through the bootstrap and into the output stage w.r.t. (limited) NFB loop generated output terminal impedance.

Another way of reducing this hum at much less cost than using an additional electrolytic, is to optimally minimise it by connecting a high value resistor directly between the +ve power rail and the emitter of the first transistor.

The value I used was 820k, but it could be anywhere between 220k and 2.2M ohms depeding on devices, input filter/impedance etc. There is nothing to lose by trying this; it does not upset the biasing in any way, nor feed other noise into the amplifier, simply try different values until the best quietening can be acheived. This might be hard to believe, but it can work, and even with a simple bridge plus single electrolytic psu.

A third way involves phase nulling and this can acheive perfect hum quietening. Try say 27k connected to the +ve power rail; 270k connected to the emitter of the first transistor; a 5 to 10% tolerance 22nF capacitor connected to signal ground. When all three component leads are brought together the hum should be tuned completely away, though slight component value adjustment might be necessary for perfect hum nulling.

If your JLH amplifier is additionally driven by a low impedance source or pre-amplifier, as was recommended by JLH himself, then your amplifier should be virtually silent, even with your 105dB sensitive loudspeaker.
-- G. Maynard # 989
-----

The distortion is lowest when the devices are closely matched; that's one of the fundamental principles of class-A push-pull, also why the JLH is so clean at lower reproduction levels. Can I add that distortion is lower again when the bootstrapped collector resistor and the emitter resistor at the current splitter are also equal, as Thijs has done with his 470 Rs. -- G. Maynard

-----
The present design investigations are still being completed using the 30yr old specified 2,200uF series output capacitor. This value was not high enough, even in 1969, because when I paralleled it with 10,000uF I could hear a distinct improvement in bass response and loudspeaker damping, 2.2mF + 10mF made it equal the valve amplifiers I was using for comparison. A minimum of 10mF (10,000uF) should be used, and preferably 20 to 22mF when the single power rail version is being constructed. -- G. Maynard # 791
----
Output current: The JLH would normally provide only about 1.35 times quiescent. This was the case using original transistors of the 2N3055 type. If you use linear gain transistors like the MLJ3281A, (be warned of stability issues, needing pulldown compensation caps) then  it is possible to do much better- perhaps almost 2.

Measuring the gain linearity of the MJL3281A's I'd wired into my bench JLH-10W circuit.Within the very rough accuracy of the resistors, the gain varied from 90 to 96 to 94 over the range 80 mA to 4A. This is similar to the published typical gain (100 flat from 5 mA to 5A).-- John Ellis
----
Regarding the standing current, the old type of transistors certainly had considerable gain fall-off, and in the original circuit the only place one output transistor could get current was from the other, so they both have to conduct enough current that the required peak output is available. This, it seems to me, is the main reason JLH had to use a higher current than the theoretical value. With the modern transistors the gain is far more linear to higher currents, so in principle the quiescent current might be reduced a little, but the main reason to recommend them is that the faster frequency response gives a better performance. The original circuit had a bootstrap load for the voltage amplifier stage and consequently could only rise as fast as the output transistor would let it. 

I recommend using modern linear gain transistors such as MJL3281A.They have high a gain (not necessarily higher than the original MJ481's) but the linear gain will reduce distortion and they give a better frequency response. In my set-up I needed 33pF capacitor, wired in parallel with the feedback resistor, to stabilize the amp. You may also need a small RC-LR (Zobel network) in the output to stabilize against RF pickup or capacitive loading in the speaker leads.

You will almost certainly have to set the quiescent current up by trial and error. I would start by using a 1k driver load (no bootstrap) just to measure the quiescent current, then determine what load resistor to use for a current of approx. 1A (30V supply), which, having a linear gain, will be very nearly linear proportional. Then divide into two and connect the tap to the bootstrap capacitor. Finally I would use larger bootstrap and output capacitors for a better bass response. -- John Ellis
----
Amp instability: High capacitance introduced by braided speaker cables etc are likely to make the amp unstable. An output coil and zobel network might be worth considering. A Pi type Thiele Network between the amp PCB and the output terminals will make the amp immune to speaker cable capacitance variations. The output Zobel must be returned to the PCB Power Ground as close to the output devices as possible.In an addendum to his original article JLH says that a 12" length of wire on the o/p before introducing any capacitance will prevent any oscillation - that's about 300nH.

MJL3281A transistors can be used in the JLH amplifier as long as you add a small compensation capacitor across the feedback resistor -- the value needed is somewhere around 33pF (start with the minimum) plus.

Another point regarding the JLH: if the input current is only about 250 microamps, the emitter resistor (the one on the ground side of the feedback network) should not be greater than 100 ohms. I'm using a rule of thumb that says the external resistance ought to be no greater than the internal Re. If the 270 ohms is reduced to 100 and the feedback resistor to 1.2k (maybe 1W too) then the distortion should be halved . 

I also have a concern with VR1 and VR2 in this circuit (Geoff dual supply with CCS). There is no limiting resistor, unlike VR3 which has R11 in series. At 0 ohms Q6 and Q8 (CCS trs) might fry, and so then cause Q2 and/or Q1 (output devices) to expire. Resistor R10 seems to have no purpose other than measuring the current in Q2. I'd put a 47 ohm in series with VR1 and a 22 ohm in series with VR2. You can always reduce these later if necessary. I would also make sure they are at max resistance to start with, and not rely on the power supply to limit the current when you switch on next time. -- John Ellis   

-----
Regarding the output stage, the upper output device might look like an emitter follower but as it is current driven by the bootstrap (think high impedance) it is really only a gain stage. The output transistors therefore don't really offer a composite low+high impedance. The output pair together simply provide current gain with one acting in opposite phase to the other. So the output impedance is only controlled by the inherent OP impedance lowered by negative feedback as with most power amp circuits.

The way to think of the JLH is to consider each output transistor as a current amplifier. The driver transistor load resistors provide a constant current due to the bootstrap (which I must say is another feature of the design: the absence of emitter resistors means the bootstrap is quite effective) which splits two ways: part into the upper base and part through the driver transistor and (mostly) into the lower base. Any current which the upper transistor needs to turn on more (in a positive output swing (for the NPN Output verstion)) is sucked out of the bias resistors and that robs the lower base, so that transistor turns off. Conversely if the lower transistor needs more amps it sucks it through the driver transistor and robs the upper device. IN reality it is the signal controlling these currents, rather than the tail wagging the dog, but that is the gist of what happens.

I think the beauty of this design is what made it iconic, despite the shortcomings.

The distortion will be lower if the output transistors are more gain-linear, as the cancellation in the currents is not perfect. The upper output current is as you mentioned from an emitter but the lower is collector, so there is at least a base current offset difference. -- John Ellis

I would imagine the JLH to have more effective use of feedback in that the loop never is broken. First of all, negative feedback applied to a voltage output will reduce the output impedance (or raise the impedance if current output). So that is closed-loop output impedance. That can easily be demonstrated and calculated using simple feedback theory.

To simulate OL conditions, the classical approach is to split the feedback resistor into two: in the case of the JLH this means making the 2700 ohm 2480 and 220 (which mimics the feedback grounding resistor) then join the junction to ground with a large capacitor.

The D.C. conditions are identical to what they were. The A.C. mid-band conditions (audio band, 1kHz) are almost identical except for a minor difference is that the load (typically 8 ohms) is now shunted by 2480 ohms instead of 2700, (but excludes the emitter impedance of the PNP, which strictly speaking makes the classic approach invalid, however, the emitter impedance to ground (assuming a low impedance source) is in the order of 100 ohms, but in this case as the feedback resistor is shunted by an 8 ohm load makes this error small).

"Measuring" the output (in simulation) gives an open loop impedance of 15.5 ohms while the closed loop is 0.39 ohms for the two conditions I used (which is probably not a true comparison as the output voltages were different, but gives some idea of NFB).

However, as the small signal conditions inside the loop will be much the same for a given output current, the "loop output impedance" is going to be the same too. But the concept of a "loop output impedance" I suggest is somewhat nebulous as actual OP impedance has an effect on the sound.

The rather high OP impedance of the JLH might account for some differences in sound between speakers. It is due to the relatively low OLG which is due to the use of a rather high 220 ohm feedback "grounding" resistor.
-- John Ellis
-----
With a bootstrap, the drive voltage is "booted" higher than the rail, and in principle, the output can reach the rail voltage minus Vce(sat) of the OPT, rather than rail minus Vbe(OPT)+Vce(sat) (CCS) +Vbe(regulator).  I kept the bootstrap for maximum positive output. This certainly gave the best sound (and lowest distortion) from the JLH's I have built.-- John Ellis
-----
In a class AB deign it is usually the Miller capacitor which limits the feedback response to cause slew rate induced distortion. In the JLH, it is only the output transistors which have a slow response (unless high speed devices) and as mentioned with an fhfe of about 50kHz the JLH will work to 50kHz without TID. Or if at lower input levels than needed for full output, say half voltage (25% power) to 100kHz or more. Which is probably where most listeners operate the JLH as 1W is quite loud. Alternatively as long as the signal source does not have these high frequencies the JLH should be free of TID. -- John Ellis
-----

Build by Tiago Sierra, Brazil

fredbloggstwo: How I got on experimenting with the JLH Class A amplifier and using Darlingtons in the output stage. (Circuit diagrams derived from Geoff Moss's modifications.)

My day-to-day amp is a Geoff Moss version of the JLH with a high current capability (the ESL version). I use this one because my monitors (B&W Matrix 80 Mk3) can dip down to about 4 ohms in their impedance curve, which means that a normal JLH will have problems driving it at high levels as it not the most efficient of speakers. This amp has 2 sets of output transistors (MK15003) with 0.1 ohm current balancing resistors. 

Having experimented with that quite a bit, I replaced the driver/phase splitter with a Szikla pair having a gain of about 3. The measured distortion dropped by about the same value and to my ears the mid range cleaned up: things like the trailing edges of piano notes seem to just be there. You can 'hear the room' on some recordings which all adds up to me of the amp really controlling the speakers. The bass also was something to die for -- gut wrenching. The only stability prop was a 47pF from the collector to base of the NPN on the Sziklai Pair.

By upping the power rails to +/- 30 volts and increasing the standing current, it gives a comfortable 40 watt and get nice and warm. Distortion is down at about 0.0035% or so. Although I have to admit, I cannot hear the difference below 0.1 % and I think that most people would be the same in a double blind test.

I built two mono blocks with separate power supplies in another box. You can see some pictures in post 3497 of this thread.

Ever the experimenter, I decided to try and up the output power a bit -- aiming for about 60 to 80 watts and see if the distortion could drop down a bit more, so this is why I had a play with the Darlington output stage, but I made a discrete version as I don't have TO-3 Darlingtons around. I used the MJ15003 with a BD139-16 in front of it and a 100 ohm resistor across the MJ15003 base to emitter.

The PCB I made can be used for the standard Moss configuration or the S-Pair mod so I tried four combinations, all with a single output pair and I used the same output transistors for all the tests.

The upper MJ15003 has a hfe of 66 and the lower a hfe of 73 and the power rails are+/- 22volts from a stabilized bench supply (my trusty Skytronics). The heatsinks are really big and get up to about 24 degrees above ambient.

The variants I tried are as follows:

(a) The standard Moss version - baseline configuration for reference

(b) The Moss version with a S-pair driver

(c) Version (a) with a Darlington output

(d) Version (b) with a Darlington output

I measured the distortion for a couple of output voltages: 2.8, which equates to about 1 watt into 8 ohms, and 6.3 volts, which is about 5 watts. Note that these are all measured values into an 8 ohm dummy load.

The results look very promising:

(a) Standard Moss config:

O/P volts, Iq, Distortion(%)

2.8, 1.9A, 0.004

6.3, 1.9A, 0.01

(b) Moss version with a S-Pair having a gain of 3X:

Vout, Iq, Distortion (%)

2.8, 1.9A, 0.0013

6.3, 1.9A , 0.003

These figures seem about right since we are increasing the open loop gain by a factor of 3 and I would expect the distortion to drop by about the same.

Now the good stuff.

To add the Darlington stage I obviously had to make some adjustment to the current setting circuit as it now has to provide less current to the output stage due to their enormous hfe values, which means that it is under less stress. Interestingly, I did not have to retune the notch filter in the Radford DMS-4, which implies that there was no discernible phase shift by adding the BD139.

The Darlingtons had a hfe of 7070 for the upper and 9460 for the lower.

(c) Standard Moss config with Darlington output:

Vout, Iq, Distortion (%)

2.8, 1.9A, 0.0016

6.3, 1.9A, 0.0036

(d) Moss Config with S-pair and Darlington output:

Vout Iq Distortion (%)

2.8 (1 Watt), 2.0A, 0.0007

6.3 (5 Watt), 2.0A, 0.001

This seems good so I carried on a bit --

10V (12.5Watt), 0.0018

11V (15 Watt), 0.0022

12V (18 Watt), 0.0024

13V (16.9 Watt), 0.003

14V (24.5 Watt), 0.008 - onset of clipping

The low power distortion figures for (d) are approaching the limits my equipment which is Viktors low distortion oscillator and a Radford DMS-4. The lower limit is about 0.0003% distortion reading. Viktor's oscillator is at least a decade down on this.

Audition: So what about listening: I tried the (d) config and could not hear any difference against my day-to-day amp I mentioned above. The main thing however, is that it opens the door to some high power outputs since it puts less load onto the driver stage.

The figures seem to imply that the S-Pair, acting as a voltage amp of 3x, drives the output more linearly and the Darlington supplies more grunt to the output stage reducing distortion even more.

Build details : The exercise was one of experimentation and wanting to try different driver and different output stages.

Consequently, I split the amp into two physical parts/boards that are linked by a short (3cm) umbilical cable allowing me to mix n' match easily.

(a) a driver board that can accommodate a standard configuration or a S-Pair driver.

(b) an output board that can accommodate standard single output pair, multiple output pairs (up to 4) or Darlington output pairs with their current sharing resistors.

Here are the circuit diagrams. (See Part-9) Some additional components were added to make Geoff Moss's driver board more acceptable - some stopper resistors on the presets and an input shunt resistor to keep the input cap charged.

I have to admit that I do like the single input transistor geometry: LTPs bring a plethora of problems that have to be got around. -- Mike   

-----

        Guess that is more than enough to keep you glued to the screen for quite some time. We will have to have a few more pages to do any justice to the Notes of Wisdom from the past.

        Kindly note that this is also work in progress as additions are to be expected soon.

        Do read and be guided by the words of wisdom from the past.

* * * * * * * * * * * *