I'm using a 24-bit sound card, and frequently measuring the JLH's around -90dB (signal to highest distortion peak) on low power levels and 8ohm load. Distortion will go up getting close to clipping. The TIP41C boards with toshiba outputs are close to -90dB for 2nd and 3rd harmonics at the moment. IMD (13&14kHz) is a little bit higher, which is a little bit disappointing (compared to PNP boards from zerozone). However I have not been able to run them 'high power' due to 1A max on the lab supply. I have another supply with high current, but only 14V. Will try them with a 19V laptop brick soon.
For measurements I use a (under 100EUR) usb sound card to generate the signals and to read them back in. I use the ARTA demo version for this (free). I don't remember exactly, but distortion readings in 'loopback' (output to input on sound card) are at least 100dB down. Some small spikes from sampling or other stuff are visible in some areas of the FFT window, but from loopback tests I know they are there, and can disregard them.
Here the input signals can be combinations of many frequencies, and I find the IMD interesting. Sometimes there is not a clear correlation to the single tone tests. FFT in ARTA can be running continuously, and distortion spectrum displayed real time as trimmers and voltages are tweaked.
A cheap and powerful measurement tool, I see no reason to go 'old school' with oscillators etc.
I'm only thinking I should get one sound card only for measurements, and use the one I have for playback. It happens quite often that it's not where I need it.
For measurements I use a (under 100EUR) usb sound card to generate the signals and to read them back in. I use the ARTA demo version for this (free). I don't remember exactly, but distortion readings in 'loopback' (output to input on sound card) are at least 100dB down. Some small spikes from sampling or other stuff are visible in some areas of the FFT window, but from loopback tests I know they are there, and can disregard them.
Here the input signals can be combinations of many frequencies, and I find the IMD interesting. Sometimes there is not a clear correlation to the single tone tests. FFT in ARTA can be running continuously, and distortion spectrum displayed real time as trimmers and voltages are tweaked.
A cheap and powerful measurement tool, I see no reason to go 'old school' with oscillators etc.
I'm only thinking I should get one sound card only for measurements, and use the one I have for playback. It happens quite often that it's not where I need it.
Last edited:
That's the ideal way to do it. Old school does surprisingly well. As said Elektor got - 96 dB and -108 dB from a very unusual State Variable filter using a lamp and TL074. Looking at a measurement of .TL074 0 001% the 1 kHz or -100 dB is possible into 10 kohms The NE5532 is even lower. One of the more interesting tests is at 0dBu 600R at just under 0.8 Vrms. After that -20 and +20dBu. My design might have a tiny bit of performance available from gain tweaking. Old school like the JLH still can do the job. On paper the JLH can not work. The most amazing thing is that the distortion is in theory below any humans ability to hear it and it's output impedance quite low. It's bandwidth exceptionally extended by standards of the day and even now. Hiss might be lower with circuit adjustment. Three input transistors could be used. Although not recommended they could be linked together. This is due to statistical noise cancelling. Noise can be seen as random collections of sine and cosine waves. Being random many cancel. The difference is larger than that. The noise becomes less jagged. What you hear is warmer and more natural. The open loop gain of the JLH is less than 60 dB from memory. This suggests the open loop performance is very good. Most of that performance comes from the cheap bootstrap capacitor. Useful that no crossover distortion injected into the bootstrap loop. Anyone up for measuring the bootstrap. I could imagine 0.25% thd at 10 kHz without it.
For fun I changed the TL084 to LM 2902 a less good LM324N. I love these op amps as comparators that can source and sink current. As audio devices they are hopeless. As an unmodified version of the circuit -55dB and plenty of random harmonics.Pulling the opamps into class A using 10K to the negative rail is a transformation. Quite a reasonable result. Trying the TL084 in the same arrangement was disappointing as 4 th harmonic is a byproduct. What putting the inferior op amp in proves is that the filter stages are excellent . Lm324N has is an underbiased class B output stage. It is class AB up to about 1 mA. It exactly demonstrates my theory of survival of the unfitest. The 2902 has a lower voltage limit. Doubtless fall out spec samples. Made a surprisingly good amplifier in the Linn Valhalla.
For fun I changed the TL084 to LM 2902 a less good LM324N. I love these op amps as comparators that can source and sink current. As audio devices they are hopeless. As an unmodified version of the circuit -55dB and plenty of random harmonics.Pulling the opamps into class A using 10K to the negative rail is a transformation. Quite a reasonable result. Trying the TL084 in the same arrangement was disappointing as 4 th harmonic is a byproduct. What putting the inferior op amp in proves is that the filter stages are excellent . Lm324N has is an underbiased class B output stage. It is class AB up to about 1 mA. It exactly demonstrates my theory of survival of the unfitest. The 2902 has a lower voltage limit. Doubtless fall out spec samples. Made a surprisingly good amplifier in the Linn Valhalla.
The measurements improved on the Zerozone NPN board with 2SC5242 (originally with TIP41C) when I was able to increase Iq. They seemed happpiest between 1,5-1,8A with the 19V brick. That is a little high though, just generating more heat without extra output..
I used 5,6 & 1,8ohm resistors in series as load, and jumpered the 1,8ohm during measurements, and distortion jumped up abt 10dB with that small change. However, that is when the Iq is adjusted for minimum distortion in the 7,4ohm load. With some small adjustment in Iq, a few dB's decrease in distortion could be gained back with 5,6ohm. It is now around -90db for single sine and IMD with 7,4ohm.
Just for fun/reference I measured some tweeters distortion today, and the best I could get was abt -80dB (using same test signals, and FFT). I was surprised to see that. But the measurement mic&mic amp I borrowed is really good (B&K -expensive) and I had to place it close to the tweeter for optimal S/N ratio. Actually, on one tweeter, the 1kHz IM could never be seen above noise floor when measuring 13+14kHz, the S/N was close to 90dB with averaging. Only IM I could see was some 'grass' around the 13&14kHz. This was using a cheap receiver as measurement amp, so I'm thinking that the amp may even influence the measurement result a bit. It also has some hiss, so that might be a factor for S/N too.
I used 5,6 & 1,8ohm resistors in series as load, and jumpered the 1,8ohm during measurements, and distortion jumped up abt 10dB with that small change. However, that is when the Iq is adjusted for minimum distortion in the 7,4ohm load. With some small adjustment in Iq, a few dB's decrease in distortion could be gained back with 5,6ohm. It is now around -90db for single sine and IMD with 7,4ohm.
Just for fun/reference I measured some tweeters distortion today, and the best I could get was abt -80dB (using same test signals, and FFT). I was surprised to see that. But the measurement mic&mic amp I borrowed is really good (B&K -expensive) and I had to place it close to the tweeter for optimal S/N ratio. Actually, on one tweeter, the 1kHz IM could never be seen above noise floor when measuring 13+14kHz, the S/N was close to 90dB with averaging. Only IM I could see was some 'grass' around the 13&14kHz. This was using a cheap receiver as measurement amp, so I'm thinking that the amp may even influence the measurement result a bit. It also has some hiss, so that might be a factor for S/N too.
Last edited:
In the 1950s when the Quad electrostatic speakers arrived people made a serious attempt to see what we can hear. It was agreed that an amplifier of 1% thd was seemingly perfect if it had an exponential decay of harmonics of distortion. For safety 0.1% was considered perfect. This was before primitive class B design's came along. I remember one called Dulici which really was far from soft. Sonically like a worn out colour TV with cathode ray tube. Make no mistake, some of the best recordings were from this period. Audio Fidelity Louis Armstrong plays King Oliver 1957 is an example. One must also remember JLH is much more like a valve design. It should not produce anything we can hear if not defective. Blind tested many can't tell between very good and seemingly better.
IM distortion seldom is mentioned. My brother was given the job to upgrade a Goodman's 80 receiver. In his oppion the IM distortion was audible. He traced it to the VAS emitter diode. Whilst this worked in DC terms it just was wrong. The diode was replaced be a 56R emitter resistor.The result was a completely new amplifier. The DC points unchanged. Apparently the owner took it with him to hi fi shops to annoy them. The VAS was horribly willing to go unstable. I was surprised he managed the transformation. The rest was extensive capacitor upgrades.
My oscillator 5777 was upgraded in a most unlikely way. By slightly tweaking the negative feedback the output increases from 3.6 to 4.6 Vrms. A change of 360 to 390R. As the design has filtering 2.22 Vrms results. First stage distortion was -58dB and now -66 dB. It's certain each would need tweaking. Being cautious the ultimate distortion could be -98 dB. That is if common mode problems don't appear. Big thick ground wires help. I saw a shoot out between a Wien oscillator and DSP. Not much difference except the DSP was a little bit untidy. I made a simple version of this using a Johnson counter with Mobius loop feedback. I think the more successful version wass 3 bit times two. More steps made it worse! With simple filtering 0.1 % thd resulted. It probably was better than filtering a square wave. It made me realize digital needs everything right to work well. Analogue from first principles easier. Distortion from the test oscillator is a unhelpful. Distortion from an amplifier beyond a certain point might be a quest rather than anything a fellow human might hear. Most excellent speakers produce 1% the. Some think distortion reduced by loop feedback can measure better and sound worse. However better distortion due to better parts sounds better. I suspect the measurement wasn't the reason. More likely distortions like IM that have no simple name. What people forget is the loudspeaker often is nothing like the test loads. The speaker is an energy store.
A friend was given a pair of Quad ESL 63. I had £200 for an amplifier. Yamaha A500. The system sounded dreadful until the Rega Planar 3 was connected. The sound was stunning and one of the best considering the modest amplifier. I knew Yamaha could cope with complex loads. It also proved how mediocre CD was at the time. The Rega had wow, for all that it sounded high end.
IM distortion seldom is mentioned. My brother was given the job to upgrade a Goodman's 80 receiver. In his oppion the IM distortion was audible. He traced it to the VAS emitter diode. Whilst this worked in DC terms it just was wrong. The diode was replaced be a 56R emitter resistor.The result was a completely new amplifier. The DC points unchanged. Apparently the owner took it with him to hi fi shops to annoy them. The VAS was horribly willing to go unstable. I was surprised he managed the transformation. The rest was extensive capacitor upgrades.
My oscillator 5777 was upgraded in a most unlikely way. By slightly tweaking the negative feedback the output increases from 3.6 to 4.6 Vrms. A change of 360 to 390R. As the design has filtering 2.22 Vrms results. First stage distortion was -58dB and now -66 dB. It's certain each would need tweaking. Being cautious the ultimate distortion could be -98 dB. That is if common mode problems don't appear. Big thick ground wires help. I saw a shoot out between a Wien oscillator and DSP. Not much difference except the DSP was a little bit untidy. I made a simple version of this using a Johnson counter with Mobius loop feedback. I think the more successful version wass 3 bit times two. More steps made it worse! With simple filtering 0.1 % thd resulted. It probably was better than filtering a square wave. It made me realize digital needs everything right to work well. Analogue from first principles easier. Distortion from the test oscillator is a unhelpful. Distortion from an amplifier beyond a certain point might be a quest rather than anything a fellow human might hear. Most excellent speakers produce 1% the. Some think distortion reduced by loop feedback can measure better and sound worse. However better distortion due to better parts sounds better. I suspect the measurement wasn't the reason. More likely distortions like IM that have no simple name. What people forget is the loudspeaker often is nothing like the test loads. The speaker is an energy store.
A friend was given a pair of Quad ESL 63. I had £200 for an amplifier. Yamaha A500. The system sounded dreadful until the Rega Planar 3 was connected. The sound was stunning and one of the best considering the modest amplifier. I knew Yamaha could cope with complex loads. It also proved how mediocre CD was at the time. The Rega had wow, for all that it sounded high end.
What I should have said yesterday was that the majority of amplifiers that were disliked were or are class AB. For more than 1 volt of the output there would be be no connection between the speaker and the power supply. Using negative feedback we ask the amplifier to leap the divide. Obviously this is a horrible procress. Optimum bias would help. The JLH never has this problem. Thus any similarities with class AB are suspect. If we measure the current waveform into a speaker we might know more. In my case my speaker cable would have enough resistance to offer a measurement. I like 0.5 diameter solid core.
I remembered your suggestion and did a quick and dirty test and replaced the current source mosfet with a bipolar today. Still with a 220ohm 'gate resistor' on the base of the bipolar.
Distortion increased quite a bit, and distortion spectrum now looks like a standard JLH with a little bit high distortion, second harmonic at maybe -56dB, and the others falling from this. With dual FETs, it had almost only 2nd and 3rd harmonics, both in the -80dB range. Iq seemed more stable with the bipolar.
It does have an overshoot on the negative flank with square input with both versions, but the overshoot was more narrow and possibly a bit smaller with one bipolar.
In favor of the fet-version is the ability to drive low impedance, very little increase in distortion with lower impedance.
Distortion increased quite a bit, and distortion spectrum now looks like a standard JLH with a little bit high distortion, second harmonic at maybe -56dB, and the others falling from this. With dual FETs, it had almost only 2nd and 3rd harmonics, both in the -80dB range. Iq seemed more stable with the bipolar.
It does have an overshoot on the negative flank with square input with both versions, but the overshoot was more narrow and possibly a bit smaller with one bipolar.
In favor of the fet-version is the ability to drive low impedance, very little increase in distortion with lower impedance.
Optimum bias would eliminate the gap. Without it, what you are describing is not a class AB amplifier at all, it is class C.
Have you ever managed to set up a class B optimum bias amplifier? Douglas Self has shown how hard it is. I find even he overly simplifies the solution. I built his exact solution and it didn't quite work. Blomley I think had a patent on this. The Quad 303 does well by understanding the problem. Better than the 405 which is sometimes described as class C which it isn't. Quad amps invert outside of a Quad combination. That is different in sound as transients and kick drums have phase even if the engineer got it wrong.
More experts than just DS will concur that optimum bias is achieved at a 26 mV drop across Re. You can argue about the relationship of that Vre to optimum sound quality, optimum THD, IMD or whatever you believe is more important or historically significant but optimum bias level is simply defined as a DC voltage.
Rotel I suspect was the source of that thinking or more accurately Stan Curtis. Quad 303 was best at 5 to 10 mA.
I think it's too easy to talk about simple measurements that serves well enough to set up an amplifier. The point I was making was that a genuine class A amplifier compared with a class AB or optimum class B are very different animals. If the AB or B is exceptionally well designed it might behave as nicely as a class A. The JLH considering it's simplicity is near enough a reference design. It does very little wrong except use a bit of power. It's constantly conducting output stage won't have the problems that designers of class AB are obsessed by. The JLH has remarkable damping factor considering it has lowish open loop gain. As I said before like a bumble bee it seems it shouldn't work. As I also said before the key component of the JLH is the driver transistor bootstrap capacitor. If ever a old school circuit was better it's here. Fit a high grade. Panasonic FC series is cheap also. In some ways that capacitor makes the driver transistor think it has a 300 V supply without risk of anything. The circuit is a compromise as it is also is doing current sink modulation. The JLH might get twice as hot if it didn't do this. It's a type of feed forward. It doesn't need to be sophisticated as it only needs supply a small excess of current.
Class D is interesting. Things that might sometimes get exaggerated when class AB become serious issues when class D. The topology of Hypex looks very like AB. To get very low distortion one transistor or more likely FET must switch off nano seconds before the following FET switches on. Too long a delay means mediocre distortion and too tight results in flames. As far as I know quite high distortion when class D can sound totally acceptable. As Douglas Self wrote" It still has crossover distortion except it's shifted to a higher frequency ".
I use to visit Quad. I always remember Ross Walker saying " There are only three ways to hook up a transistor and one of those is wrong " . He was saying it could always be argued that anything that can be invented when transistors would be obvious. This was the 405 whoes patents were almost impossible to enforce. Even though transistors didn't exist it had been described in 1929. Ross pointed out it's not class A plus C as C implies a bias mechanism as in radio transmitters. It's underbiased class B. I always speculated that BUZ900 type FET would suit the 405 better as unlike a slow NPN it isn't a switch. It's not as linear which I doubt is a problem. The speed is mostly irrelevant when the 405 as it uses feed forward error correction. An ft of 300 khz is enough.
The point is is that the snap shot thd only says working to prototype or not. It will say approximately how good. What it won't do is say which amplifier sounds best. The jlh doesn't have any major engineering problem. If correctly made and not like I doubt it's the amplifier at fault.
A few people doubt I'm distortion is a totally reliable measurement of goodness. In highly defective amplifiers it can be gross. Douglas Self wrote that it can be sidestepped as bad amplifiers are universally bad. He implied that if distortion can be held low to 10 kHz plus these problems don't exist. I suspect he is right. JLH is OK on that.
I think it's too easy to talk about simple measurements that serves well enough to set up an amplifier. The point I was making was that a genuine class A amplifier compared with a class AB or optimum class B are very different animals. If the AB or B is exceptionally well designed it might behave as nicely as a class A. The JLH considering it's simplicity is near enough a reference design. It does very little wrong except use a bit of power. It's constantly conducting output stage won't have the problems that designers of class AB are obsessed by. The JLH has remarkable damping factor considering it has lowish open loop gain. As I said before like a bumble bee it seems it shouldn't work. As I also said before the key component of the JLH is the driver transistor bootstrap capacitor. If ever a old school circuit was better it's here. Fit a high grade. Panasonic FC series is cheap also. In some ways that capacitor makes the driver transistor think it has a 300 V supply without risk of anything. The circuit is a compromise as it is also is doing current sink modulation. The JLH might get twice as hot if it didn't do this. It's a type of feed forward. It doesn't need to be sophisticated as it only needs supply a small excess of current.
Class D is interesting. Things that might sometimes get exaggerated when class AB become serious issues when class D. The topology of Hypex looks very like AB. To get very low distortion one transistor or more likely FET must switch off nano seconds before the following FET switches on. Too long a delay means mediocre distortion and too tight results in flames. As far as I know quite high distortion when class D can sound totally acceptable. As Douglas Self wrote" It still has crossover distortion except it's shifted to a higher frequency ".
I use to visit Quad. I always remember Ross Walker saying " There are only three ways to hook up a transistor and one of those is wrong " . He was saying it could always be argued that anything that can be invented when transistors would be obvious. This was the 405 whoes patents were almost impossible to enforce. Even though transistors didn't exist it had been described in 1929. Ross pointed out it's not class A plus C as C implies a bias mechanism as in radio transmitters. It's underbiased class B. I always speculated that BUZ900 type FET would suit the 405 better as unlike a slow NPN it isn't a switch. It's not as linear which I doubt is a problem. The speed is mostly irrelevant when the 405 as it uses feed forward error correction. An ft of 300 khz is enough.
The point is is that the snap shot thd only says working to prototype or not. It will say approximately how good. What it won't do is say which amplifier sounds best. The jlh doesn't have any major engineering problem. If correctly made and not like I doubt it's the amplifier at fault.
A few people doubt I'm distortion is a totally reliable measurement of goodness. In highly defective amplifiers it can be gross. Douglas Self wrote that it can be sidestepped as bad amplifiers are universally bad. He implied that if distortion can be held low to 10 kHz plus these problems don't exist. I suspect he is right. JLH is OK on that.
I'm listening to the FET-version (mentioned earlier) of the JLH at the moment, so I thought I would add some subjective first impressions. I'm running them from the same 19V laptop supplies, and current is set pretty low around 0,6A. Standard single supply, and no floating ground this time. I replaced the output capacitors with some salvaged 10mF that measured good. The ones with the kit measured poorly in ESR, and and I think higher capacitance was needed too. Everything else is as the kit was delivered.
These sound more detailed and precise than the other JLH's. Not as laid back and relaxing as the other JLH's I listened to, but more 'fast' and forward in the sound. Sounds firm and good in bass too. This sound kind of 'demands more attention'.
I have not listened long enough to comment on fatigue, but no headaches so far.
There are issues with the circuit. As I mentioned before with the overshoot on square wave negative flank, but they don't sound too bad when I just connect them and listen. I'm thinking the cause could be some capacitive coupling to the lower FET, but I have no idea how to solve it without creating other issues, so I just decided to let them play.
These sound more detailed and precise than the other JLH's. Not as laid back and relaxing as the other JLH's I listened to, but more 'fast' and forward in the sound. Sounds firm and good in bass too. This sound kind of 'demands more attention'.
I have not listened long enough to comment on fatigue, but no headaches so far.
There are issues with the circuit. As I mentioned before with the overshoot on square wave negative flank, but they don't sound too bad when I just connect them and listen. I'm thinking the cause could be some capacitive coupling to the lower FET, but I have no idea how to solve it without creating other issues, so I just decided to let them play.
Last edited:
The FET-version has some resistors already, but maybe your idea is a series resistor after the feedback point?
Sketch of the circuit from post 5776:
https://www.diyaudio.com/forums/att...h-10-watt-class-amplifier-20200108_133906-jpg
Talking about series resistors.. the series resistor for the tweeter is several ohms, and the resistance in the coil to the woofer is probably in the 0,2ohm range too.. I'm testing amps on passive speakers (not my active setup) to be able to evaluate the complete range.
Sketch of the circuit from post 5776:
https://www.diyaudio.com/forums/att...h-10-watt-class-amplifier-20200108_133906-jpg
Talking about series resistors.. the series resistor for the tweeter is several ohms, and the resistance in the coil to the woofer is probably in the 0,2ohm range too.. I'm testing amps on passive speakers (not my active setup) to be able to evaluate the complete range.
Last edited:
Member
Joined 2009
Paid Member
amen, can we finally stop bichin' about output capacitors - they are not a limitation for great sound.
Bigun: In my case I agree, but also knowing that I did not believe it myself before, I think others will have to discover for themselves too. In this case I definitely agree. These are 8ohm Vifa P17 drivers, so they are not reaching very low, and cones are small and light. If it had been 4ohm drivers with heavy cones and working range down to 20Hz, maybe things would be different. But, with the right capacitor (enough capacitance and ESR I guess), the result should be fine with those too.
I actually simulated capacitor sizes in Xsim with my speakers in advance to get a feeling for what capacitor size would be needed. As you probably know, there is a 'trick' to reinforce/tune bass with a large series capacitor for high Qt woofers in sealed boxes. I wanted to avoid this effect by using a large enough capacitor.
I actually simulated capacitor sizes in Xsim with my speakers in advance to get a feeling for what capacitor size would be needed. As you probably know, there is a 'trick' to reinforce/tune bass with a large series capacitor for high Qt woofers in sealed boxes. I wanted to avoid this effect by using a large enough capacitor.
Last edited:
Everything in audio is contradiction. As long as you know that you are safe. Usually something slightly outside of your scientific focus is the real factor. My brother was very good at making amplifiers sound good. He use to repair stuff for me when he was unemployed. He built up an understanding from repair work. His number one idea was to ensure capacitor time constants were not too similar in the amplifier. Douglas Self has his own take on this. Make the feedback arm capacitor get close to DC and tune the bass at the input. Remember the PSU capacitors count. The JLH distortion curve suggest a Douglas Self avoidable distortion is there. Some silly or jealous people say the NAD 3020 is no good because it uses these tricks. They would rather have a less nice sound because it suits their views of science. Science in this case is helpful friend and not a judge of taste. Hi fi is a defence of Ying Yang.
Today I finally finished all four regulators for my JLH69. Two for positive and two for negative voltage. Now the power supplies of both channels are separated by regulators. The voltage drop is about 3V, the current through each is 2A and the output voltage from the regulators is 19V. The scheme remained the same except for changing R1 to a higher value (18.2 to 24.9 ohms). In ltspice I have an improvement of almost 20dB, the difference is only in 2A less current through the regulator.
I haven't done any big measurements yet, but the first test results are better than the 4A version.
Also, hearing tests are better, even my wife hears the difference.
I haven't done any big measurements yet, but the first test results are better than the 4A version.
Also, hearing tests are better, even my wife hears the difference.
Attachments
In the days before JLH a regulator wasn't possible. At great expense a choke and capacitor power supply would have been used. The regulator is an additional amplifier. The JLH regulator was very simple. If using faster transistors I suspect it benign. One I built for a valve amplifier gave a little second harmonic distortion. It used a 500 volt FET.
For my valve DAC I use this shematic, orginal is from ARC LS25MKII.
This is the best HV reg. I ever made.
