I commented in the video that this was a true DC amplifier. As a few of you pointed out it's NOT. What I meant to say is that the frequency response goes down to almost DC. FLAT. It's quite hard to measure the last few Hz but as far as I can measure it simply does not roll off at all. Sorry If I miss led you. I will show this on the next video.
Hello Michael... Cheers for your video and just as well for your alleged "silliness"... It makes the subject lighter! Back on the topic of the amplifier being a "DC" amplifier. Well, it's true that - as you already pointed out yourself - it's not actually a DC amplifier... but your cautionary advice is still correct. As it stands, at lower frequencies it is indeed a DC amplifier with a gain of 1:1 so, it will not turn a 1V DC input voltage into a 31.3V at the output (as its gain setting resistors would mandate) but it will STILL turn a, say, 6V DC input into 6V DC output which is just as awful and will probably burn your speakers.
Here's a couple of suggestions to be implemeted in the mentioned order until the problem is "hopefully" solved.: 1. Replace the R1 ( 2.2K ) input resistor with a 0.47uF MKT capacitor ( This is not a DC coupled amplifier anyway. ) 2. Q9 & Q10 are serving as capacitor-multipliers stabilizing the voltage to the differential-amp & voltage driver stage. Adding 100nF MKT capacitors parallel with C5 & C8 + C6 & C9 will not hurt. 😉 3. The voltage across Q12 is very small ~1.2V This is mainly due to the Quasi config used by Q11 & Q13. This Vce voltage of Q12 could in fact be too small to be stable. 🧐Start removing C7 from the VBE multiplier. The current in R14, R15, R16 is far too low in my opinion. I suggest replacing these resistors with ones that are 1/10 the value => R14: 560 ohm, R15: 470 ohm, R16: 220 ohm trim. ( This will allow the current in the resistor network to be much higher than I-base @ Q12 thus enhancing stability. ) Adjust the trim-pot in order to obtain ~20mA thru the final power stage without an input signal. 4. The circuit is using Quasi-coupled transistors in both differential amp and the power section. 🤨This structure can cause oscillations. ( try to replace Q1 & Q2+ Q3 & Q4 with 2 normal NPN Darlington transistors. ) 5. For some strange reason the high side current source to the differential stage and the "voltage driver stage" current source shares Q6, R12 & C2 as stabilizing elements with only 1.2V across Q6. This provides a constant current of 4mA to the diff-stage & around +30V on the high side of R11. This may work as intended, but the usage of "shared regulators" could potentially create an unwanted signal path.🤔 Start with a 100nF MKT parallel with C2. If this doesn't help proceed changing R11 to 220ohm and connecting it to a separate adjustable voltage of ~ 1.2 Volt lower than the emitter of Q9. ( F.x. a 3.3V zener parallel with a 100nF MKT and 5K trim + 22K to ground. ) Connect the center-pin of the trim to the modified R11= 220 ohm and adjust the current thru R13 to ~10… 15mA. IMPORTANT. It's crucial that you have completed "Step 3" so that current in your drivers Q11, Q13, Q14, Q15, Q16, Q17 is less than ~20mA ! Use a current limited power supply set to max 50mA ! ) 6. The gain is controlled by [ R17 / R18 ] with C3 fixating the DC gain = 1 and -3dB @ ( 2*pi*330 ohm*1000uF)^ -1 = 0.5Hz. Changing C3 to 47uF bringing the -3dB up to 10Hz, would be a more conservative choice. ( The gain should be less than 30 with the input being maximum 2 Vpp to avoid clipping. ) While tinkering you could replace R18 with a 560 ohm resistor to "play safe". ( I guess you could make this into a real DC coupled amp by shorting C3, but you will have to add "offset adjust" to get 0V output at idle, which this circuit doesn't have. ) 7. Q18A + Q18B is a Widlar current mirror. ( no remarks here. ) This is a very simple amplifier yet with some strange approaches and also cutting corners. 😬 IMHO. Don't buy this design. If you want something that is more thought thru, get the open-source Beta22 either as a kit or semi-finished product that only needs a power supply. 👍
I would also suggest an output inductor before the R30/C10 stabilization circuit. A few microhenries in parallel with a 10 ohm resistor should suffice.
Hi Michael, try installing a small capacitor around 10pF across R17 to limit out of band HF gain. You could also try a small capacitor around 68pF from Base to Collector of Q7. The suggestions below are excellent, the Quasi-coupled transistors in the differential input stage is unusual and could be problematic. The addition of a choke in the output as suggested is a good practus.
Exactly ... Looking at the scope traces we can see very high frequency oscillations on the test signals. These fuzzies are high frequency oscillations under load. Bypassing R17 with a small value cap will increase feedback, reducing gain at higher frequencies which, fuzzies or no, should improve that amplifier's stability quite noticeably. As a suggestion, they could look into various values, with an eye to limiting the bandwidth to about 30khz. It might also help to bypass C3 with a ceramic cap to compensate for value reduction at higher frequencies. I also agree that a more or less standard Zobel filter at the output would probably help.
Chokes will be added in due course; however, it will not help in this case. At present I'm only feeding resistors as loads. This would only help when feeding real loudspeakers. PS we are almost there!! watch this space.
@@Douglas_Blake Indeed. 👍It's a good idea to limit the bandwidth both up and down. 🤗For audio purposes nobody needs anything outside audible range. In order to eliminate unwanted oscillations you will foremost need to place some "HF limiting elements" in the feed-forward loop like C4 on Q8. You could apply the same method ( just with with bigger values ) to Q11 and Q13. If HF-limiting elements is placed only in the feed-back loop, the amplifier could become incapable of handling "out of band" noise because this will be undetected. 😕
I am a rf circuit guy so think differently but I would definitely get a sensitive probe on a SA you might see something or not but hf oscillating in tight circuits is definitely a thing I watch for. After getting bit a few times. The small handheld SA are brilliant at bringing higher testing facilities to us mortals. Have several bigger SA but the little one is always my go to. Good luck with finding the problem.
Thank you for posting the new schematic! I noticed these differences with V4: - The base of Q5 is now decoupled by C2, fixing a mistake in V4. - The clamping diodes on the collectors of Q18A and Q18B have been removed. - R9 is now 150Ω (was 100Ω in V4) so the current in the differential input-stage is lower. - Q9 and Q10 in the supply rails are a valid way to filter out noise, in my opinion. - Not decoupling the main power rails is outrageous! - R21 and R26 are now 33Ω (was 100Ω in V4), so the current in Q11 and Q13 is higher (ok, I guess). - Q11 and Q13 used to have a base resistor of 100Ω, for high frequency stability! - The input transistors Q1 ~ Q4 are now in a very different configuration. In V4, there were two single input transistors with two cascode transistors. This forms a pure current output, feeding into a current mirror. In V5 however, Q1 and Q3 are kept at a constant current, increasing the input impedance. Q2 and Q4 are emitter followers, feeding into a current mirror! (Q18A and Q18B) That seems pretty bad to me, the emitter of Q2 looks straight into the miller capacitance of C4/Q8. It would make much more sense to replace Q1 ~ Q4 with a pair of PNP darlingtons. Then you would have pure current sources feeding into the current mirror again. ======= Lowering the current in the input-stage may make the clamping diodes obsolete. But where does the amplifier clip now? R5 and R6? That would limit the output swing and increase distortion prematurely. ======= You're right, version 5 seems to be a failed experiment.
Hi Michael Just before guests arrive for Christmas I did a simulation on best quiescent current per output transistor. CFP output stages are prone to sensitivety to bias vs crossover distortion, so this is important. The best perfomance at 20KHz is 22mA or 44mA for the pair. You should the measure 44mV over one common output resistor. 14mA (Guess that was recommended) will yield more distortion. More current beyond 22mA will also increase distortion. Merry Christmas Claus
Hi Claus, yes, your numbers seem to make sense. We have found that a higher QI does indeed reduce the crossover spike to zero. This board does have quite odd characteristics. I will show this in the next video, to long winded to talk about it here. Still on final tests, now it's Christmas. I do hope to film a video on the topic very soon. Looks like the other issue has been sorted also.
Greetings from Serbia. I see that this v.5 doesn't have 100R base stoppers at the driver transistors. In the earlier versions, which were more like straight copies of Doug Self's Load Invariant amp, these resistors provided some kind of stability. In my opinion this is not ideal solution depending on combination of used semis and does not provide unconditional stability. These Sziklai (CFP) amps are very tricky concerning stability. As I explained on other forums, all you need to do is to put 220pF cap between B and C of PNP driver transistor (in this case 2SA1837). Everything will work just fine.
Thank you for your work Ron. I didn't build my kit since I can't match with my signal tracer the transistor he came with. Now if Michael thinks the result is atrocious and the PCB layout is bad there is no magic solution for this. Spending money on a piece of "Turd" for fun? Maybe, but I'm not in a hurry...R.I.P LJM L12-2 Ver 5
Well, since Mike released this video we may have discovered a "mod" or two that can make it tolerable (not wonderful... merely *tolerable*). However, that being said it is a very unstable design, and definitely NOT an improvement over the earlier v3 or v4 boards. More on this soon. Thanks.
Merry Christmas ! in case this your last upload before the holidays. I have often wondered about those pre packaged current mirrors, I mean it should be a great idea as the two devices are matched (trimmed at manufacture) and of course as thermally bonded as can be, I just haven't heard people say much about them. I will try them in a front end one day ! I will look forward to the next part but now I will go watch the first again for a refresh ! Oh yeah, have you ever worked on really expensive kit and I mean £25,000.00p and up, if yes was it interesting please :) !
Yes, they are a good idea but tend to be quite expensive. They were used all the time in the 70s. They are just NOT fets lol Regarding expensive amplifiers, no I don't offer a repair service at all. All my friends are poor, like me!
The "FET1" position on the PC board appears to have been laid out to accommodate a genuine dual FET, or at least one or two particular part numbers of them. Then it didn't work out quite as they had hoped (but I am purely guessing at this..!). Fortunately (for them) if you take a commonly available commercial dual NPN transistor and flip it upside down ("dead bug" mounting) the connections work out OK without changing the board layout. The connections to make the pair into a current mirror are done externally to the device package in the board layout itself. To be honest, it appears to us that the real reason it was done was simply to save board space. However no matter how you look at it, there are so many design flaws on this PC board that we haven't even begun to describe them all.
Hi Michael, I have now received two in kit form and will hold off building them up until I hear more, hopefully some good news. Thanks for your efforts.
*Thanks very much Mike.* 👍👍 Reducing the gain from its "as-shipped" 31X down to around 11X by means of some simple resistor changes decreases the distortion levels at ANY output amplitude (even just short of clipping levels) by around 70%. More on that later. The instability of the QI circuit (and it is *very* unstable indeed) can be improved with a few changes. Unfortunately, although they appear simple on paper they are somewhat complex to perform on the actual PC board. This is because of the convoluted board layout. There is yet another relatively simple change that appears to get rid of MOST (but not all) of the general board instabilities. However this needs quite a bit more testing. Finally, even more reductions in distortion can be had by another handful of simple component value adjustments. More on this later as well. Please stay tuned... Oh, and *IF* you can find version 3 or version 4 boards, our best advice at this point would be to purchase them rather than this version. Be very careful though because many sellers (such as on eBay) are showing version 3 or version 4 boards in their listings. However, if you then order them you are very likely to receive version 5 boards instead. So be certain to ask the seller what version they will ACTUALLY be shipping before hitting the "buy" button. *Buyer beware..!*
No, the amplifier, out of the box does not have chokes on the output, it should and will on my board. This is not the issue as the test results are on a resistive load, so not applicable in this case. Progress is being made, too early in the day to publish, but we do have a working sample that is stable. I will update as soon as we are 100% happy with the result. Just got the QI to get stable now!
I don't like the The Sziklai Darlington Transistors Q1, Q2 and Q3, Q4. Cascodes are often tricky to stabilize in my experience! Try changing Q1 and Q3 to BC546B (or equals). Get the legs corect. You don't need the Sziklai coupling. My 5 cents! Happy Xmas!!
It does not look good that the emitters of driver transistors in Sziklai go directly to the output, instead through the resistors. Also, the higher resistor value here would increase temperature stability.
Indeed... Even though I never tried this approach myself, just by the looks of it, it seems this "deviation" from what an actual Sziklai pair should be, only detracts from the "standard" approach. It makes the thermal stability poorer and it dampens the local negative feedback that is intrinsic to the Sziklai pair, stealing from it one of its qualities in comparison to Darlington pairs which is of greater linearity.
I think they try to make as high gain as possible in order to make low distortion. But they are, probably, not high learned engineers, and doctors, like them who, for example, created the LM3886.
Agree! I would put 10 ohm from the emitter on the drivers to the output. Perhaps there is room for a small resistor from the PCB mounting hole, to the transistor emitter pin. This might be enough to ensure stabillity, but I would try. BR Claus
Hi Michael and Ron. I did some simulation on loop stability using the schematic you have shown in the video. I have used LTSpice and the Tian probe method. Usually close to reality. My initial investigation have shown that the amplifier without an output coil (I chose 800nH) can be or are close to unstable. Without the coil the amplifiers gain margin (8ohm) is 11dB and phase margin is 85 degree. With the coil gain margin is 20dB and phase margin is 74 degree. If you load the amplifier with 8ohm/1uF the amplifier reacts wildly. This amplifier badly need an output coil in order to garantee stability. There could of course be other problems like paracitic oscillations, but at least put an 800nH output coil//3.3ohm eg. on the speaker terminal. If I can help with info on the open loop issue, please say so. BR/Claus
Hi Claus, thank you for your time on this problem. We think we have solved the instability issue and hope to publish here in the next few days. So far, we have only been looking at instability on resistive loads, or NO loads. I think, the amplifier, like most do need a choke/resister network on the output as you suggested. I will incorporate one into the tests.
@@MichaelBeeny Hi Michael. My bet will be that the upper CFP outpair is unstable. CFP is known for unstability if there is inductance to the base leads or capacitive loads. a snubber can usually cure this. I did not test the board, so this is only a guess. BR/Claus
Hi Mike. I have bought the “ old” l12 amplifier. Thought I give it an go. I have an 40 vac transformer with is good size. Is 40 volt ac to high voltage for the board.? -
2x40vac will give you around 2x57vdc. That is a bit on the high side for the L12, I think they officially state that 2x55vdc is the absolute maximum. But as usual, they like to live on the edge of what is safe. I would recommend you stay around 30 to 33vac on the transformer. It will give you between 42 and 47vdc after rectification and smothing caps. That also leaves you with a little room for mains voltage variation.
@RandomUser6947 I tend to agree, Because I wanted max power, I used a DC of 50 volts. I would tend to regard this as a maximum. The main thing is to use large heatsinks, the cooler your electronics is the longer it will last. Don't use a higher voltage unless you really need it.
Can anyone recommend a nice easy ,ish, good sounding amplifier board- preferably with an 55 v ish rail. The L9 accuphase E405-mod. Has anyone tried this board?
For me also Sziklai looks dangerous. I would increase the emmiter resistors to 0.22 ohms for starter, Also, maybe to try to play with Miller capacitor.
@@vladnurk4710 Just type L12/2 ver 5 on Ebay or Ali, lots to choose from. Confirm with the seller EXACTLY which boards the have before you part with any money.
Well to my untrained eye there is nothing I see that is 'special' I would note the absence of a 'Zobel network' on the output, my next thought would be 'Miller' compensation but 100pF is reasonable value which sadly does start to point at layout. Anyway, again I look forward to what trained eyes can see !!
We thought the same thing. Mike and I have both tried increasing *and* decreasing the 100pF CDOM capacitor's value. Either way we went things got worse in regard to the output instabilities, although he saw a ray of hope at 47pF. However, even that did not work when I tried it, it merely moved the parasitic "fuzz" to a different spot in the waveforms. The instability is up into the multiple megahertz region, by the way. In the end, sadly, it apppears very likely that this will prove to be a horrendously bad board layout.
My comment with a link to TI seems to have disappeared? Well! I just wanted to say that I have had good use of a small loop antenna (TI has a video on youtube) in search of a culprit transistor that oscillates. There are a number of things in the design that could cause this, but it is easier to measure it. Solder a coil, eg. 6-8 mill diameter and 5-6 windings to a peace of coax with a BNC connector in the other end. Connect it to the scope and probe on the board. It should show you which part of the amplifier is bad. It could of course be other problems, but then you have ruled out local problems. BR/Claus
I commented in the video that this was a true DC amplifier. As a few of you pointed out it's NOT. What I meant to say is that the frequency response goes down to almost DC. FLAT. It's quite hard to measure the last few Hz but as far as I can measure it simply does not roll off at all. Sorry If I miss led you. I will show this on the next video.
Hello Michael... Cheers for your video and just as well for your alleged "silliness"...
It makes the subject lighter!
Back on the topic of the amplifier being a "DC" amplifier.
Well, it's true that - as you already pointed out yourself - it's not actually a DC amplifier... but your cautionary advice is still correct.
As it stands, at lower frequencies it is indeed a DC amplifier with a gain of 1:1 so, it will not turn a 1V DC input voltage into a 31.3V at the output (as its gain setting resistors would mandate) but it will STILL turn a, say, 6V DC input into 6V DC output which is just as awful and will probably burn your speakers.
Well done Michael. Whilst frustrating, attempting to solve a problem is a joy in itself. Even if the result is not what one wants.
Good on you and Ron
Here's a couple of suggestions to be implemeted in the mentioned order until the problem is "hopefully" solved.:
1. Replace the R1 ( 2.2K ) input resistor with a 0.47uF MKT capacitor ( This is not a DC coupled amplifier anyway. )
2. Q9 & Q10 are serving as capacitor-multipliers stabilizing the voltage to the differential-amp & voltage driver stage. Adding 100nF MKT capacitors parallel with C5 & C8 + C6 & C9 will not hurt. 😉
3. The voltage across Q12 is very small ~1.2V This is mainly due to the Quasi config used by Q11 & Q13. This Vce voltage of Q12 could in fact be too small to be stable. 🧐Start removing C7 from the VBE multiplier. The current in R14, R15, R16 is far too low in my opinion. I suggest replacing these resistors with ones that are 1/10 the value => R14: 560 ohm, R15: 470 ohm, R16: 220 ohm trim.
( This will allow the current in the resistor network to be much higher than I-base @ Q12 thus enhancing stability. )
Adjust the trim-pot in order to obtain ~20mA thru the final power stage without an input signal.
4. The circuit is using Quasi-coupled transistors in both differential amp and the power section. 🤨This structure can cause oscillations. ( try to replace Q1 & Q2+ Q3 & Q4 with 2 normal NPN Darlington transistors. )
5. For some strange reason the high side current source to the differential stage and the "voltage driver stage" current source shares Q6, R12 & C2 as stabilizing elements with only 1.2V across Q6. This provides a constant current of 4mA to the diff-stage & around +30V on the high side of R11. This may work as intended, but the usage of "shared regulators" could potentially create an unwanted signal path.🤔
Start with a 100nF MKT parallel with C2. If this doesn't help proceed changing R11 to 220ohm and connecting it to a separate adjustable voltage of ~ 1.2 Volt lower than the emitter of Q9.
( F.x. a 3.3V zener parallel with a 100nF MKT and 5K trim + 22K to ground. ) Connect the center-pin of the trim to the modified R11= 220 ohm and adjust the current thru R13 to ~10… 15mA. IMPORTANT. It's crucial that you have completed "Step 3" so that current in your drivers Q11, Q13, Q14, Q15, Q16, Q17 is less than ~20mA ! Use a current limited power supply set to max 50mA ! )
6. The gain is controlled by [ R17 / R18 ] with C3 fixating the DC gain = 1 and -3dB @ ( 2*pi*330 ohm*1000uF)^ -1 = 0.5Hz. Changing C3 to 47uF bringing the -3dB up to 10Hz, would be a more conservative choice.
( The gain should be less than 30 with the input being maximum 2 Vpp to avoid clipping. ) While tinkering you could replace R18 with a 560 ohm resistor to "play safe".
( I guess you could make this into a real DC coupled amp by shorting C3, but you will have to add "offset adjust" to get 0V output at idle, which this circuit doesn't have. )
7. Q18A + Q18B is a Widlar current mirror. ( no remarks here. )
This is a very simple amplifier yet with some strange approaches and also cutting corners. 😬
IMHO. Don't buy this design. If you want something that is more thought thru, get the open-source Beta22 either as a kit or semi-finished product that only needs a power supply. 👍
I would also suggest an output inductor before the R30/C10 stabilization circuit. A few microhenries in parallel with a 10 ohm resistor should suffice.
Hi Michael, try installing a small capacitor around 10pF across R17 to limit out of band HF gain. You could also try a small capacitor around 68pF from Base to Collector of Q7. The suggestions below are excellent, the Quasi-coupled transistors in the differential input stage is unusual and could be problematic. The addition of a choke in the output as suggested is a good practus.
Exactly ...
Looking at the scope traces we can see very high frequency oscillations on the test signals. These fuzzies are high frequency oscillations under load. Bypassing R17 with a small value cap will increase feedback, reducing gain at higher frequencies which, fuzzies or no, should improve that amplifier's stability quite noticeably. As a suggestion, they could look into various values, with an eye to limiting the bandwidth to about 30khz.
It might also help to bypass C3 with a ceramic cap to compensate for value reduction at higher frequencies.
I also agree that a more or less standard Zobel filter at the output would probably help.
Chokes will be added in due course; however, it will not help in this case. At present I'm only feeding resistors as loads. This would only help when feeding real loudspeakers.
PS we are almost there!! watch this space.
@@Douglas_Blake Indeed. 👍It's a good idea to limit the bandwidth both up and down. 🤗For audio purposes nobody needs anything outside audible range.
In order to eliminate unwanted oscillations you will foremost need to place some "HF limiting elements" in the feed-forward loop like C4 on Q8.
You could apply the same method ( just with with bigger values ) to Q11 and Q13.
If HF-limiting elements is placed only in the feed-back loop, the amplifier could become incapable of handling "out of band" noise because this will be undetected. 😕
Bypassing the R17 may increase transforming the negative into positive feedback. On the other hand, adding the miller capacitor might work.
@@mp3mag
The last thing you want is positive feedback ... that's how you build oscillators.
It will be interesting to scope around the circuit to see if the source of the instability can be found. Watching with interest!
In process... 😁
Coming soon..!
I am a rf circuit guy so think differently but I would definitely get a sensitive probe on a SA you might see something or not but hf oscillating in tight circuits is definitely a thing I watch for. After getting bit a few times. The small handheld SA are brilliant at bringing higher testing facilities to us mortals. Have several bigger SA but the little one is always my go to. Good luck with finding the problem.
Thank you for posting the new schematic!
I noticed these differences with V4:
- The base of Q5 is now decoupled by C2, fixing a mistake in V4.
- The clamping diodes on the collectors of Q18A and Q18B have been removed.
- R9 is now 150Ω (was 100Ω in V4) so the current in the differential input-stage is lower.
- Q9 and Q10 in the supply rails are a valid way to filter out noise, in my opinion.
- Not decoupling the main power rails is outrageous!
- R21 and R26 are now 33Ω (was 100Ω in V4), so the current in Q11 and Q13 is higher (ok, I guess).
- Q11 and Q13 used to have a base resistor of 100Ω, for high frequency stability!
- The input transistors Q1 ~ Q4 are now in a very different configuration.
In V4, there were two single input transistors with two cascode transistors.
This forms a pure current output, feeding into a current mirror.
In V5 however, Q1 and Q3 are kept at a constant current, increasing the input impedance.
Q2 and Q4 are emitter followers, feeding into a current mirror! (Q18A and Q18B)
That seems pretty bad to me, the emitter of Q2 looks straight into the miller capacitance of C4/Q8.
It would make much more sense to replace Q1 ~ Q4 with a pair of PNP darlingtons.
Then you would have pure current sources feeding into the current mirror again.
=======
Lowering the current in the input-stage may make the clamping diodes obsolete.
But where does the amplifier clip now? R5 and R6?
That would limit the output swing and increase distortion prematurely.
=======
You're right, version 5 seems to be a failed experiment.
Hi Michael
Just before guests arrive for Christmas I did a simulation on best quiescent current per output transistor. CFP output stages are prone to sensitivety to bias vs crossover distortion, so this is important. The best perfomance at 20KHz is 22mA or 44mA for the pair. You should the measure 44mV over one common output resistor. 14mA (Guess that was recommended) will yield more distortion. More current beyond 22mA will also increase distortion.
Merry Christmas
Claus
Hi Claus, yes, your numbers seem to make sense. We have found that a higher QI does indeed reduce the crossover spike to zero. This board does have quite odd characteristics. I will show this in the next video, to long winded to talk about it here. Still on final tests, now it's Christmas. I do hope to film a video on the topic very soon. Looks like the other issue has been sorted also.
Greetings from Serbia. I see that this v.5 doesn't have 100R base stoppers at the driver transistors. In the earlier versions, which were more like straight copies of Doug Self's Load Invariant amp, these resistors provided some kind of stability. In my opinion this is not ideal solution depending on combination of used semis and does not provide unconditional stability. These Sziklai (CFP) amps are very tricky concerning stability. As I explained on other forums, all you need to do is to put 220pF cap between B and C of PNP driver transistor (in this case 2SA1837). Everything will work just fine.
Thank you for your work Ron. I didn't build my kit since I can't match with my signal tracer the transistor he came with. Now if Michael thinks the result is atrocious and the PCB layout is bad there is no magic solution for this. Spending money on a piece of "Turd" for fun? Maybe, but I'm not in a hurry...R.I.P LJM L12-2 Ver 5
Well, since Mike released this video we may have discovered a "mod" or two that can make it tolerable (not wonderful... merely *tolerable*).
However, that being said it is a very unstable design, and definitely NOT an improvement over the earlier v3 or v4 boards.
More on this soon.
Thanks.
Merry Christmas ! in case this your last upload before the holidays. I have often wondered about those pre packaged current mirrors, I mean it should be a great idea as the two devices are matched (trimmed at manufacture) and of course as thermally bonded as can be, I just haven't heard people say much about them. I will try them in a front end one day ! I will look forward to the next part but now I will go watch the first again for a refresh ! Oh yeah, have you ever worked on really expensive kit and I mean £25,000.00p and up, if yes was it interesting please :) !
Yes, they are a good idea but tend to be quite expensive. They were used all the time in the 70s. They are just NOT fets lol
Regarding expensive amplifiers, no I don't offer a repair service at all. All my friends are poor, like me!
The "FET1" position on the PC board appears to have been laid out to accommodate a genuine dual FET, or at least one or two particular part numbers of them.
Then it didn't work out quite as they had hoped (but I am purely guessing at this..!).
Fortunately (for them) if you take a commonly available commercial dual NPN transistor and flip it upside down ("dead bug" mounting) the connections work out OK without changing the board layout.
The connections to make the pair into a current mirror are done externally to the device package in the board layout itself.
To be honest, it appears to us that the real reason it was done was simply to save board space.
However no matter how you look at it, there are so many design flaws on this PC board that we haven't even begun to describe them all.
@@MichaelBeeny and me :)
Hi Michael, I have now received two in kit form and will hold off building them up until I hear more, hopefully some good news. Thanks for your efforts.
We are getting there Stuart, I don't think you will have long to wait.
*Thanks very much Mike.* 👍👍
Reducing the gain from its "as-shipped" 31X down to around 11X by means of some simple resistor changes decreases the distortion levels at ANY output amplitude (even just short of clipping levels) by around 70%.
More on that later.
The instability of the QI circuit (and it is *very* unstable indeed) can be improved with a few changes.
Unfortunately, although they appear simple on paper they are somewhat complex to perform on the actual PC board.
This is because of the convoluted board layout.
There is yet another relatively simple change that appears to get rid of MOST (but not all) of the general board instabilities.
However this needs quite a bit more testing.
Finally, even more reductions in distortion can be had by another handful of simple component value adjustments.
More on this later as well.
Please stay tuned...
Oh, and *IF* you can find version 3 or version 4 boards, our best advice at this point would be to purchase them rather than this version.
Be very careful though because many sellers (such as on eBay) are showing version 3 or version 4 boards in their listings.
However, if you then order them you are very likely to receive version 5 boards instead.
So be certain to ask the seller what version they will ACTUALLY be shipping before hitting the "buy" button.
*Buyer beware..!*
Is there an instability throughout the amplifier ? Does it show up at or past a certain point? No output chokes?
No, the amplifier, out of the box does not have chokes on the output, it should and will on my board. This is not the issue as the test results are on a resistive load, so not applicable in this case. Progress is being made, too early in the day to publish, but we do have a working sample that is stable. I will update as soon as we are 100% happy with the result. Just got the QI to get stable now!
I don't like the The Sziklai Darlington Transistors Q1, Q2 and Q3, Q4. Cascodes are often tricky to stabilize in my experience! Try changing Q1 and Q3 to BC546B (or equals). Get the legs corect. You don't need the Sziklai coupling.
My 5 cents!
Happy Xmas!!
It does not look good that the emitters of driver transistors in Sziklai go directly to the output, instead through the resistors. Also, the higher resistor value here would increase temperature stability.
Indeed... Even though I never tried this approach myself, just by the looks of it, it seems this "deviation" from what an actual Sziklai pair should be, only detracts from the "standard" approach.
It makes the thermal stability poorer and it dampens the local negative feedback that is intrinsic to the Sziklai pair, stealing from it one of its qualities in comparison to Darlington pairs which is of greater linearity.
I think they try to make as high gain as possible in order to make low distortion. But they are, probably, not high learned engineers, and doctors, like them who, for example, created the LM3886.
Agree! I would put 10 ohm from the emitter on the drivers to the output. Perhaps there is room for a small resistor from the PCB mounting hole, to the transistor emitter pin. This might be enough to ensure stabillity, but I would try.
BR
Claus
Hi Michael and Ron.
I did some simulation on loop stability using the schematic you have shown in the video. I have used LTSpice and the Tian probe method. Usually close to reality.
My initial investigation have shown that the amplifier without an output coil (I chose 800nH) can be or are close to unstable. Without the coil the amplifiers gain margin (8ohm) is 11dB and phase margin is 85 degree. With the coil gain margin is 20dB and phase margin is 74 degree.
If you load the amplifier with 8ohm/1uF the amplifier reacts wildly. This amplifier badly need an output coil in order to garantee stability.
There could of course be other problems like paracitic oscillations, but at least put an 800nH output coil//3.3ohm eg. on the speaker terminal.
If I can help with info on the open loop issue, please say so.
BR/Claus
Hi Claus, thank you for your time on this problem. We think we have solved the instability issue and hope to publish here in the next few days. So far, we have only been looking at instability on resistive loads, or NO loads. I think, the amplifier, like most do need a choke/resister network on the output as you suggested. I will incorporate one into the tests.
@@MichaelBeeny Hi Michael. My bet will be that the upper CFP outpair is unstable. CFP is known for unstability if there is inductance to the base leads or capacitive loads. a snubber can usually cure this. I did not test the board, so this is only a guess.
BR/Claus
Hi Mike. I have bought the “ old” l12 amplifier. Thought I give it an go. I have an 40 vac transformer with is good size. Is 40 volt ac to high voltage for the board.? -
2x40vac will give you around 2x57vdc. That is a bit on the high side for the L12, I think they officially state that 2x55vdc is the absolute maximum. But as usual, they like to live on the edge of what is safe.
I would recommend you stay around 30 to 33vac on the transformer. It will give you between 42 and 47vdc after rectification and smothing caps. That also leaves you with a little room for mains voltage variation.
@RandomUser6947 Solid advice. Recommended voltage is around 45VDC.
Thank you for the advice. It is appreciated 👍⭐️
@RandomUser6947 I tend to agree, Because I wanted max power, I used a DC of 50 volts. I would tend to regard this as a maximum. The main thing is to use large heatsinks, the cooler your electronics is the longer it will last. Don't use a higher voltage unless you really need it.
Can anyone recommend a nice easy ,ish, good sounding amplifier board- preferably with an 55 v ish rail. The L9 accuphase E405-mod. Has anyone tried this board?
For me also Sziklai looks dangerous. I would increase the emmiter resistors to 0.22 ohms for starter, Also, maybe to try to play with Miller capacitor.
Even with this issue can I get a board to play with?
OF course, I just want people to understand the issues as of now.
@@MichaelBeeny yes i could help used to design analog & digital circuitry
Sure, but buyer beware... 🙂
@@ronschauer839 I am across the ditch so where to purchase
@@vladnurk4710 Just type L12/2 ver 5 on Ebay or Ali, lots to choose from. Confirm with the seller EXACTLY which boards the have before you part with any money.
Thanks for video mister Beeny. You make me an unbearable suspens! I can't wait to see what happens next.
I can hardly wait myself Cyril lol
@MichaelBeeny 🤣
Hello, try to lower R11. Best regards !!!
Well to my untrained eye there is nothing I see that is 'special' I would note the absence of a 'Zobel network' on the output, my next thought would be 'Miller' compensation but 100pF is reasonable value which sadly does start to point at layout. Anyway, again I look forward to what trained eyes can see !!
We thought the same thing.
Mike and I have both tried increasing *and* decreasing the 100pF CDOM capacitor's value.
Either way we went things got worse in regard to the output instabilities, although he saw a ray of hope at 47pF.
However, even that did not work when I tried it, it merely moved the parasitic "fuzz" to a different spot in the waveforms.
The instability is up into the multiple megahertz region, by the way.
In the end, sadly, it apppears very likely that this will prove to be a horrendously bad board layout.
@@ronschauer839 Thanks for the info ! and we shall see :)
The FET looks like 2SK2145
Of course the amplifier is unfit for commercial sale as is, but if you want to find the culprit, a loop antenna can be a good help.
BR Claus
sir pls make a circit of high level input for active subwoofer throu speaker output from old amplifier since some amplifier do not hv sub out
The amplifier should not have a sub output, it's the preamp that should provide this. Or the sub it's self.
My comment with a link to TI seems to have disappeared? Well!
I just wanted to say that I have had good use of a small loop antenna (TI has a video on youtube) in search of a culprit transistor that oscillates. There are a number of things in the design that could cause this, but it is easier to measure it. Solder a coil, eg. 6-8 mill diameter and 5-6 windings to a peace of coax with a BNC connector in the other end. Connect it to the scope and probe on the board. It should show you which part of the amplifier is bad.
It could of course be other problems, but then you have ruled out local problems.
BR/Claus
Oh dear is all I can say.
It's not the end of the world Conor, things are looking up and we are making progress.
Thanks