English only please.
I forgot to say. LM317 usually works very well. It's Darlington output is coupled to an op amp of the LM741 era. It is effectively an improved 7805/9/12 etc regulator. The improvement being current and a chance to reduce output noise. It would be a 1.25V regulator if no gain used. Current is about 50% better. The Darlington side steps stability problems as long as the date sheet is followed about input and output decoupling. The 1.25V comes from a band gap reference. I sometimes use LEDs as a fake band gap, Real ones are doubtless far better. LED's seem better than diodes. Oscillator limiters come to mind.
The LD1084 is slightly more complex. It uses a compound output a bit like Quad 303 output stage. Being NPN out it is reasonably stable.
The PNP pass transistor seen in data sheets ( 7805 etc ) should be treated with caution. TIP2955 being an ideal choice if tempted. Many other lowest drop out regulators also have PNP outputs. Again caution is wise.
The 1 MHz power bandwidth is likely to be OK. The output capacitor is quite helpful in optimising that. It must be remembered the LM317 is an additional device in the chain. Unlike a class AB amplifier it isn't being asked to fluctuate much which helps greatly. It's if you like an additional Darlington on the JLH NPN totem-pole.
The LD1084 is slightly more complex. It uses a compound output a bit like Quad 303 output stage. Being NPN out it is reasonably stable.
The PNP pass transistor seen in data sheets ( 7805 etc ) should be treated with caution. TIP2955 being an ideal choice if tempted. Many other lowest drop out regulators also have PNP outputs. Again caution is wise.
The 1 MHz power bandwidth is likely to be OK. The output capacitor is quite helpful in optimising that. It must be remembered the LM317 is an additional device in the chain. Unlike a class AB amplifier it isn't being asked to fluctuate much which helps greatly. It's if you like an additional Darlington on the JLH NPN totem-pole.
The one thing LM317 gives you is massive ripple rejection for pennies. I would say it easily changes ripple from 4V rms down to 4 mV rms. Most likely 0.4mV. Adjust by ear to get it right by finding the dropout voltage If you have 25 VDC likely 22 VDC is possible. You should always aim for about 1 Vrms as initial ripple. It helps everything if you do. Below that isn't really required. If you need 1.2 amps ripple current spec needs to be much higher ( 5A? ).
I would build this alternative part on 1940s style tag-strrip. A TIP142 can replace LM317. The bias might be a 24 V zener diode ( 1.3 or 2 watt ). 20 mA might be enough to power the base. I suspect you would need 30 VDC initial to make i work ( 27 on a good day ). Knowing the JLH has some ability to work with some PSU ripple this likely to work and fit the same space as a LM317. a 220 uF cap to the zener is a good cheap upgrade. This is not unlike a JLH PSU.
Some would add a TL431 active zener ( I think ON semiconductors best on noise ). If you do you must read all of the design notes. Although this is fine what you have done is build a LM317 type device. TL431 has the option for feedback to make it as ridged as a LM317. You might consider not using that ( simple zener ). If so you very likely have the best device. I found I could get it to drive 2SC5200 up to about 1.2 amps. I suspect 80 mA was used. I have used a LM317 on a higher voltage and NPN power transistor`with no feedback,.This was first class.
TL431 - Wikipedia
I would build this alternative part on 1940s style tag-strrip. A TIP142 can replace LM317. The bias might be a 24 V zener diode ( 1.3 or 2 watt ). 20 mA might be enough to power the base. I suspect you would need 30 VDC initial to make i work ( 27 on a good day ). Knowing the JLH has some ability to work with some PSU ripple this likely to work and fit the same space as a LM317. a 220 uF cap to the zener is a good cheap upgrade. This is not unlike a JLH PSU.
Some would add a TL431 active zener ( I think ON semiconductors best on noise ). If you do you must read all of the design notes. Although this is fine what you have done is build a LM317 type device. TL431 has the option for feedback to make it as ridged as a LM317. You might consider not using that ( simple zener ). If so you very likely have the best device. I found I could get it to drive 2SC5200 up to about 1.2 amps. I suspect 80 mA was used. I have used a LM317 on a higher voltage and NPN power transistor`with no feedback,.This was first class.
TL431 - Wikipedia
Remembering that idea. I changed a MJ3000 Darlington for a 2SC5200 fast NPN power transistor. The TL431 to my surprise could drive it. I suspect the gain of the transistor at only 1 amp was very high. Thinking back it was a fluke. The TL431 must have been running at 1 watt which is too high. It was zener type circuit without output feedback. If a PNP transistor added in a feedback pair it could be ideal. An oscilloscope required. BC327 perhaps with 2SC5200 The current the TL431 needs coming through the BC327 base. Still quite simple. I should correcct that. 2SA1943 and BC337.
Sziklai pair - Wikipedia
Sziklai pair - Wikipedia
Following on from where I left off (post #7831), I made the following changes:
Output transistors to matched MJ15022.
Output capacitor to NOS Panasonic 4700uF/25V. Much lower ESR at 20kHz and 120Hz.
Experimented with three different power supplies: the Lambda switcher shown in the previous pics; a Boschert SL30-3612R 4A switcher; and a Lambda LCS-series (linear regulator w/ pass transistor). In each case, capacitance was added to give the flattest impedance/phase at the boards' power terminals.
The 'constant' in the listening sessions were the LM3886-based amps I've been working on in parallel to the JLH.
Just like any other amp, the JLH is *very* sensitive to the quality of the power supply. All three of these units result in a remarkably different sound.
The Boschert switcher was easily the worst. Fairly noisy (700uV in a 40kHz BW). No way to make the impedance flat. It was around 130mOhms up to 1kHz or so, then the output cap dominates and it goes down to it's ESR at higher freqs. As I see it, this is the main problem with switchers. It made the JLH's sound like a cheap chipamp, thin and edgy, not at all musical.
The Lambda switcher was much better. Noise about 300uV, impedance fairly flat around 40mOhms after adding 68uF to each board power terminals. (The switchers do not seem to have the rock-steady-flat output impedance characteristic that linears with big output caps have; they're generally level but with unevenness.) The sound was much improved over the Boschert, much more neutral in the upper mids and highs.
The Lambda linear was easily the best. This supply is from the early 80's (yes, I recapped it...). It's pass transistor is an old 200kHz Ft part which only gives a flat output Z up to 1kHz. I added 4700uF/50V/R012 ESR at the output to flatten the Z, perhaps could/should have gone with 6800u. Noise was at the meter's residual. Cleaner sound across the entire spectrum. Not entirely neutral, though. Main complaint was a forward-sounding upper midrange, and bottom could be more solid. (It's quite possible that a regulator with more bandwidth would fix this.)
So, overall I thought the amps with the Lambda linear sounded pretty good until I switched to the 3886-based amps, and everything snapped into focus. What stood out most was the overal clean-ness, the clarity, richness, and definition of the fundamental tones - the woody-ness of violins/cello/bass, warmth, body, and clarity of voices, tone and "slam" of drums, deeper/clearer ambience, etc. The entire presentation suddenly came together and made sense. I had expected the Class A design to excel in this respect, but it was easily beaten by a high-feedback AB chipamp on a good board. At it's best, the JLH was "whitening" the sound. So much for "optimum" harmonic distortion profiles...
I will continue to tweak on the JHL's, but it has an uphill climb, for sure.
Output transistors to matched MJ15022.
Output capacitor to NOS Panasonic 4700uF/25V. Much lower ESR at 20kHz and 120Hz.
Experimented with three different power supplies: the Lambda switcher shown in the previous pics; a Boschert SL30-3612R 4A switcher; and a Lambda LCS-series (linear regulator w/ pass transistor). In each case, capacitance was added to give the flattest impedance/phase at the boards' power terminals.
The 'constant' in the listening sessions were the LM3886-based amps I've been working on in parallel to the JLH.
Just like any other amp, the JLH is *very* sensitive to the quality of the power supply. All three of these units result in a remarkably different sound.
The Boschert switcher was easily the worst. Fairly noisy (700uV in a 40kHz BW). No way to make the impedance flat. It was around 130mOhms up to 1kHz or so, then the output cap dominates and it goes down to it's ESR at higher freqs. As I see it, this is the main problem with switchers. It made the JLH's sound like a cheap chipamp, thin and edgy, not at all musical.
The Lambda switcher was much better. Noise about 300uV, impedance fairly flat around 40mOhms after adding 68uF to each board power terminals. (The switchers do not seem to have the rock-steady-flat output impedance characteristic that linears with big output caps have; they're generally level but with unevenness.) The sound was much improved over the Boschert, much more neutral in the upper mids and highs.
The Lambda linear was easily the best. This supply is from the early 80's (yes, I recapped it...). It's pass transistor is an old 200kHz Ft part which only gives a flat output Z up to 1kHz. I added 4700uF/50V/R012 ESR at the output to flatten the Z, perhaps could/should have gone with 6800u. Noise was at the meter's residual. Cleaner sound across the entire spectrum. Not entirely neutral, though. Main complaint was a forward-sounding upper midrange, and bottom could be more solid. (It's quite possible that a regulator with more bandwidth would fix this.)
So, overall I thought the amps with the Lambda linear sounded pretty good until I switched to the 3886-based amps, and everything snapped into focus. What stood out most was the overal clean-ness, the clarity, richness, and definition of the fundamental tones - the woody-ness of violins/cello/bass, warmth, body, and clarity of voices, tone and "slam" of drums, deeper/clearer ambience, etc. The entire presentation suddenly came together and made sense. I had expected the Class A design to excel in this respect, but it was easily beaten by a high-feedback AB chipamp on a good board. At it's best, the JLH was "whitening" the sound. So much for "optimum" harmonic distortion profiles...
I will continue to tweak on the JHL's, but it has an uphill climb, for sure.
Last edited:
A good 3886 chip, connected according to a good circuit, on a good PCB, assembled with good parts, having a good power supply unit, it works well.
Then there's something drastically wrong with your JLH implementation. Speaker mismatch, most likely. What is the actual minimum impedence (& efficiency) of your speakers?
Yes, a good LM3886 can sound very good. But it's still a class AB op-amp, with everything that implies. The JLH isn't. But it is current limited and has a large capacitor in the output.
So if you need another 6db of power or more current than a JLH can deliver or need to drive DC, then perhaps an LM3886 based amp is good choice.
Power supplies matter. As Julian Vereker originator of Naim Audio said " An amplifier is a signal modulated power supply". Class AB uses the power supply in a better way. Class A could be said to have a saturated power supply and AB not. Not my words, I understood the concept.
I have to agree. And I'm quite surprised by it.
You asked this same question a few weeks ago. They're not speakers, they are transformer-coupled electrostatic headphones, nominal load around 20 Ohms. I recently got another set (Stax SR-X with SR-7 adaptor) and they are an even easier load, around 27 Ohms nominal.
No offense, but I am agnostic to Class A vs AB dogma. I trust my measurements and listening. There is nothing "drastically wrong" with these JLH69 amps. Parts quality is high. Small-signal transistors selected via curve tracer. Distortion measurements are as good or better than anything else I've seen posted here. They don't sound "bad". Problem is, they are simply outperformed by the 3886 at gain of 9 on hifiocean's pcb (now discontinued, btw...).
The 1kHz distortion curves tell some of the tale. Measurement conditions are identical (3.1VRMS into 18 ohms). The 3886 board:
And the JLH board:
Freq scale is linear to highlight the harmonics. The 3886 has 19dB less H2 and 6dB less H3. And has less noise below say 500Hz. It also has lower output impedance (around 20mOhms vs 250mOhms for the JLH). There is every reason for it to sound better. And it does.
Last edited:
> hifiocean's pcb now discontinued
Looks very much like the original one from Brian GT:
https://enjon.uk/wp-content/uploads/2019/05/amp-top.png
Chipamp.com LM3886 PCB – (mis) Adventures in Hi-Fi
Patrick
Looks very much like the original one from Brian GT:
https://enjon.uk/wp-content/uploads/2019/05/amp-top.png
Chipamp.com LM3886 PCB – (mis) Adventures in Hi-Fi
Patrick
let's be honest - gain 9x is not serious for home use - will need another stage of amplification.with so much negative feedback, anything can reach such a thd level ..but will it sound better?
It would be interesting to see JLH vs 3886 starting at 10Khz and up.
JLH with matched outputs @ 1.3 A you can get 2H -80db...
It would be interesting to see JLH vs 3886 starting at 10Khz and up.
JLH with matched outputs @ 1.3 A you can get 2H -80db...
Going from DAC to amp, 9dB can be plenty. The days of -10dBV consumer output levels are long gone...
My current THD measurement rig has issues (mostly its DAC) at 10kHz. But I would put my money on the 3886. Personally, I find THD to be less meaningful at 10k and up.
This is the Class A dogma...
This setup is already H2 at -80dB... you have to add the +10 signal level to the marker readout value. That's why I say, this JLH amp measures as good as any I've seen posted here.
Last edited:
> It would be interesting to see JLH vs 3886 starting at 10Khz and up.
Just look at Fig 49 of LM3886 datasheet.
Patrick
Just look at Fig 49 of LM3886 datasheet.
Patrick
Tnx for replay.
I know it - but it only shows at 30W and 60W as far as I can remember. If we compare with JLH then it seems to me correct to compare 1W-2W -3W area.I like the upper frequencies better on Class A amps ...but it is a matter of taste ...
I know it - but it only shows at 30W and 60W as far as I can remember. If we compare with JLH then it seems to me correct to compare 1W-2W -3W area.I like the upper frequencies better on Class A amps ...but it is a matter of taste ...
