The plot, and the marker values, are absolute values. The vertical range is as labeled, from +10dBV to -110dBV. That's 12dB per division, necessary to get the whole spectra visible.
The fundamental is at +10dBV; the marker values are also absolute values. So to get them relative to the fundamental, you have to add +10dB to the marker readouts to get the H2 and H3 relative number.
For the JLH, that puts H2 at -80 and H3 at -94.
For the 3886, they're both right at -100.
Last edited:
Aspirin. So we will win!
3886 according to some reviews works better at a load of 8 ohms and above.
3886 according to some reviews works better at a load of 8 ohms and above.
Last edited:
This is probably obvious to most, but for the sake of correctness, it should read "you have to add -10dB to the marker readouts..."
Basically in this use the power amp is working like a big op amp, a very happy one at that. I often noticed quiet modest 1970s receivers like my Akai 4040 were transformed into headphones. Sennheiser 414 that ABBA seemed to use were 600 ohms in the old days and could run direct. The big deal was they sounded very high end even by today standards. The speakers Dynaco A25 helped bridge the gap.
So I finally put together a rig to test the Hfe at working currents, following @kozard's setup in post #7211. For most 3055's tested, it told quite a different story than the little transistor testers did.
I tested a couple dozen of the 2N3055K's I asked about in a separate thread. Remarkably consistent 95-96 Hfe from unit-to-unit. In the JLH, slew rate was about half of the MJ15022's, so probably a 2.5MHz device. And distortion at 1kHz was back up to -74.
So I'm getting the picture that, to get 1kHz harmonics down at least 80dB takes good Hfe matching and linearity, AND faster transistors.
Has anyone been able to get better than H2 @-80dB? If so, how?
I tested a couple dozen of the 2N3055K's I asked about in a separate thread. Remarkably consistent 95-96 Hfe from unit-to-unit. In the JLH, slew rate was about half of the MJ15022's, so probably a 2.5MHz device. And distortion at 1kHz was back up to -74.
So I'm getting the picture that, to get 1kHz harmonics down at least 80dB takes good Hfe matching and linearity, AND faster transistors.
Has anyone been able to get better than H2 @-80dB? If so, how?
I posted some FFT plots earlier in the thread, and if I remember correctly H2 & H3 were around -90dB at 1W. This is with a PNP version from ZeroZone. Main thing to reduce the distortion was to find the optimum bias current, and adjusting the balance between the two resistors to the base of the 'current source' output transistor to optimize (AC) current sharing between the output transistors.
I have written about it before, and even posted a sim of the amp. The problem is to find something in this long thread..
I'm sure that I've posted something in this thread along the lines of reducing distortion. General points:
1. The PNP input device needs to run at a higher gm.
To do this the feedback resistor and its "grounding" resistor need to be reduced.
In the original design the PNP runs at about 300uA. Its gm is only ~12mA/V. At 1mA this increases to 40mA/V. But the 220 ohm "grounding" resistor would prevent the higher gm - so needs to be reduced. Suggest 22 ohms and 270 for the feedback. The 22 ohms will however reduce the input gm back to 20mA/V but that will still reduce the distortion compared with the original.
2. As the current is now higher, the 8.2k base resistor on the driver is too high.
But reducing it does nothing much because of the voltage non-linearity on the driver and output bases. THerefore, a current mirror needs to be added as Kokorianz did. Add two NPN's for a current mirror in place of the 8.2k, with 100 ohm emitter resistors, and take a 27k (or whatever is appropriate to your power supply voltage) to the Vcc rail, decoupled for filtering for the reference mirror.
The base resistor of the lower output transistor (2.2k in the original) can be reduced to improve (slightly) the linearity of the driver, but that needs the current supply to be adjusted. Nominally it would be the same value as the lower bootstrap resistor for symmetry, but Kokorianz also suggested 1k was nearer optimum.
Probably all the centering and quiescent current adjustments will need readjusting after these changes.
The higher open loop gain may cause greater risk of HF oscillation. For compensation, don't use a Miller but use a phase lead connected 220pf (maybe with additional series resistor) between the driver collector and PNP emitter.
I've not optimised these arrangements for high speed transistors (still).
1. The PNP input device needs to run at a higher gm.
To do this the feedback resistor and its "grounding" resistor need to be reduced.
In the original design the PNP runs at about 300uA. Its gm is only ~12mA/V. At 1mA this increases to 40mA/V. But the 220 ohm "grounding" resistor would prevent the higher gm - so needs to be reduced. Suggest 22 ohms and 270 for the feedback. The 22 ohms will however reduce the input gm back to 20mA/V but that will still reduce the distortion compared with the original.
2. As the current is now higher, the 8.2k base resistor on the driver is too high.
But reducing it does nothing much because of the voltage non-linearity on the driver and output bases. THerefore, a current mirror needs to be added as Kokorianz did. Add two NPN's for a current mirror in place of the 8.2k, with 100 ohm emitter resistors, and take a 27k (or whatever is appropriate to your power supply voltage) to the Vcc rail, decoupled for filtering for the reference mirror.
The base resistor of the lower output transistor (2.2k in the original) can be reduced to improve (slightly) the linearity of the driver, but that needs the current supply to be adjusted. Nominally it would be the same value as the lower bootstrap resistor for symmetry, but Kokorianz also suggested 1k was nearer optimum.
Probably all the centering and quiescent current adjustments will need readjusting after these changes.
The higher open loop gain may cause greater risk of HF oscillation. For compensation, don't use a Miller but use a phase lead connected 220pf (maybe with additional series resistor) between the driver collector and PNP emitter.
I've not optimised these arrangements for high speed transistors (still).
Last edited:
It should be impossible to hear at this level. If a class AB we could infer the machine might have traits which give a sonic signature. When class A harder to believe. Tube devotees seem to prefer old school power supplies. At 100 mA that's easier. If you draw how a PSU is part of the output stage it looks very messy and yet it works reasonably well when JLH. Perhaps being a constant voltage source helps. Nearly all regulators are inferior to the JLH. I did wonder if a LM1875 would make a regulator. Probably one of my worse ideas? LM317 likely much better.
The Stax headphones most recently looked at are likely an ideal load for class AB. For a power supply class AB is an ideal load. It looks likely in this unique case class AB is unsubtly better. I would reduce the JLH current to the minimum if using the Stax. They say you can't have too much of a good thing. When current be certain. I have a hunch 250 mA is enough for the sta. IM distortion might reduce.
The Stax headphones most recently looked at are likely an ideal load for class AB. For a power supply class AB is an ideal load. It looks likely in this unique case class AB is unsubtly better. I would reduce the JLH current to the minimum if using the Stax. They say you can't have too much of a good thing. When current be certain. I have a hunch 250 mA is enough for the sta. IM distortion might reduce.
JLH 1969 is an (almost) push-pull OTL amplifier operating in class A. The current consumption is constant in the absence of a signal. When working to the bus (+), the output capacitor is charged. The current from the power supply flows through the upper transistor, capacitor and speaker to ground (-) of the power supply. The current through the upper transistor increases, and through the lower one decreases. I think that the current consumed from the source is increasing.
In the next half cycle, the capacitor is discharged through the speaker and the lower transistor to ground. The current drawn from the power supply through the upper transistor decreases.
There is some similarity in the nature of current consumption with an OTL class AB amplifier. The difference is that the change in current consumption in class A occurs relative to the quiescent current (more or less). In OTL class AB, the output stage consumes current during capacitor charging (change from 0 to maximum). When the capacitor is discharged, the consumption is minimal.
In this reasoning, I did not take into account the work of the bootstrap and the charge-discharge of the capacitor in the power supply after the SMPS or linear regulator.
In the next half cycle, the capacitor is discharged through the speaker and the lower transistor to ground. The current drawn from the power supply through the upper transistor decreases.
There is some similarity in the nature of current consumption with an OTL class AB amplifier. The difference is that the change in current consumption in class A occurs relative to the quiescent current (more or less). In OTL class AB, the output stage consumes current during capacitor charging (change from 0 to maximum). When the capacitor is discharged, the consumption is minimal.
In this reasoning, I did not take into account the work of the bootstrap and the charge-discharge of the capacitor in the power supply after the SMPS or linear regulator.
The problem with the original JLH is not so much THD but IMD, I suggest.
In a simulation (which is a lot quicker than the soldering iron) 10kHz and 11kHz with 2N3055H produces about 0.2% 1kHz. Using epi 2N3055's gives a reduction to 0.1%. (at nearly full power)
That is where some audible differences may be apparent.
The simulated results for the modifications mentioned above give .05% IMD.
BTW I think that Semelab/Magnatec 2N3055's were similar to the original RCA 2N3055, with an 800kHz minimum ft. Not sure what the "k" referred to, maybe a voltage selection as Ian mentioned. Semelab tended to offer replacement parts for hi-rel apps, so were unlikely to change the spec. - but not known for sure.
From my measurements the epi 2N3055 are similar in performance to the 2N3716 4MHz part, so I suspect the "K" versions are slow types.
In a simulation (which is a lot quicker than the soldering iron) 10kHz and 11kHz with 2N3055H produces about 0.2% 1kHz. Using epi 2N3055's gives a reduction to 0.1%. (at nearly full power)
That is where some audible differences may be apparent.
The simulated results for the modifications mentioned above give .05% IMD.
BTW I think that Semelab/Magnatec 2N3055's were similar to the original RCA 2N3055, with an 800kHz minimum ft. Not sure what the "k" referred to, maybe a voltage selection as Ian mentioned. Semelab tended to offer replacement parts for hi-rel apps, so were unlikely to change the spec. - but not known for sure.
From my measurements the epi 2N3055 are similar in performance to the 2N3716 4MHz part, so I suspect the "K" versions are slow types.
Not quite! A common description, but actually the capacitor should NOT charge during a signal excursion, or the voltage across it will change.
Of course, it will charge or discharge a little, but the voltage should be so low as not to affect the output.
At low frequencies, an output capacitor does charge/discharge significantly, and thus spoils the output.
Exactly John. If we reduce current the amplifier is easier to drive. The driver phase splitter should be of high gain. I used BC337-40 and hoped it could cope. A BD139 would be more suitable except the high gain types are rare. What is so good about the JLH is ultimately what is bad about it. It is a very simple design which is far from simple to understand.
I would contrast another 1969 design Sinclair Z30. In some ways a very bad design. What is bad about it is also very good. High speed undersized transistors. Using 2SC5200 we could solve that. Biasing would suit class A. The devices it uses are non critical. The Z30 is the definitive blameless amplifier taken back a step. I notice the man who uses the blameless concept refuses to mention it. Many say RCA which to me is less good as a topology. Before anyone says, yes the Z30 was hopelessly unreliable. That wasn't the topology. The Z30 would be ideal for teaching power amp design. The Z30 is the missing link which for some reason people like to ignore.
I would contrast another 1969 design Sinclair Z30. In some ways a very bad design. What is bad about it is also very good. High speed undersized transistors. Using 2SC5200 we could solve that. Biasing would suit class A. The devices it uses are non critical. The Z30 is the definitive blameless amplifier taken back a step. I notice the man who uses the blameless concept refuses to mention it. Many say RCA which to me is less good as a topology. Before anyone says, yes the Z30 was hopelessly unreliable. That wasn't the topology. The Z30 would be ideal for teaching power amp design. The Z30 is the missing link which for some reason people like to ignore.
In the JLH, the power supply current largely follows the current in the upper output transistor, so it will increase for a + output and decrease for a - output.
As the current in the upper output transistor is not compensated by an opposite current, the average PSU current changes slightly due to the asymmetry.
As the current in the upper output transistor is not compensated by an opposite current, the average PSU current changes slightly due to the asymmetry.
On a complex musical signal, we will see a low frequency envelope modulated by the mid and high frequency components.