simmconn

I’m not sure what point you try to make. Isn’t 101dB/V at 8 Ohms equivalent to 80dB/mW? How can the same spec on one hand ‘low efficiency’ and at the same time ‘high sensitivity’?

@jokerman777 I use a 10K trimmer in place of R73 for a forum member's DIY T2 as well, to trim the ~10V offset down to near zero. I know the transistors are most likely not matched on that unit and it doesn't make sense to remove them and match again. In your case the offset is significantly more. I would try to root cause it before replacing R73 with a trimmer. FYI the Vbe of Q33 and Q32 measured on that unit are 0.5xV and 0.6xV respectively. It's definitely a problem if Vbe measures negative. To troubleshoot point C and D, you can temporarily disconnect the 2.2M resistors R80 thru R83. Measuring across R89/91 with a meter having 10M Ohm input impedance only adds about 1% error, so no worries. Check the polarities and short on D29 thru D32. They could pull the voltage at C/D through R80 thru R83 if in a wrong polarity. Once the voltage on C/D correctly reflects O+/O-, the balance servo can still bottom out trying to adjust. Measure the output voltage of the opamps. If they come within 3V of its power rails, I'd trim the battery voltage gently so the opamps come back to the 'comfort zone'.

@jokerman777 I'm not sure what problem you have with the battery now, especially with the working channel. Can you still reproduce the problem in your post? It's normal to see both top and bottom voltage changes as your battery voltage 'contracts' or 'relaxes', since multiple feedback loops are in action trying to compensate the changes. I'm not too worried about Hfe of Q23. You probably already did the math. At 740V battery voltage, the current through Q23 is about 3.23mA. Since R39 passes about 0.1mA, Hfe of Q23 should be at least ~33 in order not to starve the Q16/19/20 branch even in the extreme situation. The original design seems so marginal, since the 2SC3675 is spec'd for Hfe>30. The good thing is that in reality, the C3675 has a quite slanted output curves unlike the datasheet suggests. Its effective DC gain increases as Vce increases. One of my samples measured at Hfe=38 with Ib=10uA at low voltage on a tester only needs about Ib=40uA to get Ic=3.2mA at Vce=740V. Essentially you get doubled Hfe in the real circuit compared to on a transistor tester. Sorry my CRT curve tracer has a broken flood gun so I can't show you the complete picture. Ib=40uA is still a large chunk of the 100uA from R39, which seems to be different from one would think, that the load current is usually a small fraction of the DC operating current of either arm in a diff amp. I think your way of reducing R39 is in a right direction. If I were to take a wild guess, the R39 could have been 6.2k (Blue-Red-black-Brown) instead of 62k (Blue-Red-black-Red) in the original T2 design. You see that besides the tiny decimal point, the color coding is also very close to each other. It's not uncommon that such mix-up could happen in any stage of manufacturing, and copying. With R39=6.2k it'll make the A1486s operate at a comfortable 500uA each and still within 200mW which is okay without a heat sink. I would expect the adjustment to be less finicky now that the main amp has a lot more juice and the output device has 30% less load. More importantly, some of my genuine C3675 show onset of breakdown at Ic=3.2mA, Vce=740V. Getting the current down would also help on that. By the way the A1486s has excellent low current linearity that's probably why the battery had worked at all under such a low current. Regarding the battery current, the R62/63/74 also drain some current so it needs to be added to what you measured on R58/59. Let's get your math straightened out. From -561V to -541V it is going UP and not down, this is important especially when most of the circuit is referenced to B- (-560V). For the Q32 and Q33 in the channel with large offset, have you measured their Vbe separately, and compare with the good channel? I hate to repeat myself, but you've got to measure all nodes. When something isn't working, every component and every connection is a suspect until proven otherwise.

If you use a thick PCB, be aware that the through hole pads can fail during component removal (disconnection between the annular ring and the hole barrel or inside the barrel) that is difficult to detect. A simple workaround is to always solder both sides of the through hole pad after replacing a component. 'De-rating' may not be what you think it is. De-rating means reduction of a rating based on certain operating conditions. For example, the LTL4213 LED has max If of 15mA at TA=25C, "Derating Linearly From 50℃ at 0.2mA/C" means at 60C, the max If de-rates to 15-(60-50)*0.2=13mA. You are operating the LED at way below its max If, de-rating doesn't apply here. What you worried about is perhaps degradation, that the If/Vf relationship changes after long-term exposure to high temperature. We've discussed it somewhere in this forum. If I remember correctly Vf increases for the same If due to degradation and not the other way around. Yes D2 and D3 are for protection purpose only. Previously both the LED string and the K246 are suspicious based on your voltage measurement. Fixing either of them would make the battery work more stably, but not completely right until you fix both of them.

Voltage across R73 and R64 should differ by 2x Vbe and not the same. That probably indicates that your output offset servo Q33/Q32 isn’t working properly. If you measure any Vgs at more than 1.5V, chances are that the MOSFET is fake. Not a big deal since they are used as source followers except Q26/27 where linearity matters. R99 and R100 are isolation resistors meant to reduce the effect of parasitics when you measure the battery with a multimeter. They don’t carry any current.

I bought this finished protector board on Taobao for about $12. The parts look reasonably good. It could use better cleaning off the flux residue on the bottom side. Totally worth the time I would otherwise have to spend soldering it.

Be aware that the JST XH series is only rated for 250V AC/DC.

What about the A1943/C5200 or A1302/C3281 variants? They are more linear in the medium current range (1A to 5A) than the 21193/94 but are also faster.

Congratulations! I can’t say that I fully agree with your analysis, but I guess getting the amp working is what matters.

LED: The numbers are all over the place: Your IV curve shows the LTL4213 should have Vf between 1.725V and 1.750V with 0.5mA of current. You've got three different voltage ranges for the LED. Which one is correct? You need to find out where the dependencies come from. Your previous LED chain voltage measurement shows 12.1-10.7=1.4V of voltage variations with not much If change. Is the 1.4V evenly distributed across 7 LEDs? If you suspect Vf change due to the temperature coefficient, that's easy to verify. Just keep monitoring the LED voltage with the amp powered on from cold. 2SK246: Your test indicates that the 'backup ones from ebay' are probably good. Unless the ones on the board are from the same seller and same batch, the test result probably doesn't help much more than that. Rather than replacing parts shotgun style. I'd take voltage measurements shotgun style, then sit down and analyze the data. Replacing parts without knowing why and what would be a very inefficient way of troubleshooting. For example, when looking into the active battery, measure voltage on all nodes. Beware of the burden by the DMM input impedance (usually 10M Ohms but YMMV), so plan carefully when you measure across high-value resistors. When looking into the final stage, collect enough voltage data within the CCS so you can calculate the current supplied by the CCS. Measure the voltage at Output+/Output-, then determine if the tubes are in the right operating point. If you know very well how the circuit works, you can strategically take only a few measurement at key spots; otherwise just do more leg work and collect all data. Don't worry if some could be redundant. They often end up helping you where least expected.

You can’t control the ganged mechanical contacts to open or close at the exact same moment. So technically when two or more contacts are connected in parallel, the interrupting current rating won’t double. However, paralleling the contacts does reduce the contact resistance, if that’s your goal.

Yes the 2SK246 looks suspicious. Its |Yfs| should be around 1mS around Id=300uA, so changes in Vgs of 1.51V-1.43V=80mV should generate 80uA of change in Id. Apparently that's not the case here. You could use a 2SK373 or even a 2SK117 as a substitute. The 2SK208 is also a close sub, albeit in SOT-23 package. Also, Q16 Ib should not be that high. The current thru R39 is about 100uA, split between Q16 and Q17. If Q16 has an hFE >100, it's Ib should not be over 1uA even in the extreme case.

It looks like most of the voltage adjustment comes from the wandering reference, no wonder you said RV1 doesn't have much effect. Assuming your Q16 path works okay, comparing the two diagrams, V(R42) dropped (6.22-6.56)/6.56=-5.18%, almost the same as the LED string voltage drop (11.4-12)/12=-5%, which means the 2SK246 ccs didn't have much effect at all. To give you a data point, my mocked-up 2SK246 ccs changes V(22k) from 6.5496V to 6.5446V (less than -0.1%) when the supply changes from 12V to 11.4V. Your LED string seems not great, either. When the current drops from 703uA to 663uA (-5.7%, ignoring the Q16 base current), the LED string voltage drops -5% as well. Not much of regulation, isn't it? I grabbed a random red LED from my stash and in similar situations its forward voltage changes from 1.7914V to 1.7881V, only -0.18%. Your LED probably has 700uA right around the knee on its IV curve. It may not be a bad part, just not suitable for this location. The above assumes the Q16 and the rest of the battery circuit works okay. Oh well, let's fix one problem at a time.

The difficulty setting the V(R42) to 6.55V could be due to something else. I assume that you observed different pin out when substituting 2SA1486 with 2SA1413. As long as the LED strings are lit stably and not flickering, the RV2 should adjust smoothly for 6.55V across R42. The idea is to set the 2SK246 ccs to about 297uA with the majority of that going through R42 and only a tiny bit from the base of Q16. If you have some spare 2SK246 from the same batch, resistors and pot, breadboard a circuit and test with a bench DC power supply. See if the pot hits the bottom or something wrong with your 2SK246. It’s a good idea to measure around the active battery circuit, using the top node as the reference point, mark voltages and currents on the schematic. The active battery does not take rocket science to troubleshoot. You just need to know 1) Ohm’s law, 2) Transistors have hfe, and 3) A forward-biased silicon junction has a voltage drop about 0.6V. There is no need to be paranoid about the battery not being spot-on at 740V. Since it is transparent to the audio signal, they can even be purposely set slightly differently from each other to compensate unbalances elsewhere in the circuit. But if your amp don’t like them being near 740V at all, that means some other parts of the amp is quite different from what the designer has envisioned. There are quite a few feedback loops in the works that maintain a near-zero offset. Let go of the active battery voltage temporarily in exchange for zero offset can set those loops at their normal operating point and help you find out who’s not behaving correctly. Mark any voltage reading with a ‘*’ if you hear oscillation when you get the reading, as it may not be reliable.

Watch the power rail voltages as you see strange things happen, such as a low battery bottom voltage (-549V) coincide with negative output offset. Depending on your 10M90s in the PSU, you may need to reduce the current setting resistor values if they get into current limiting condition. It’s a good practice to load-test every power rails at 20% more than the rated current. Spend some time doing measurements across the board, mark the voltages at the key nodes in the schematic and verify the voltage and current that are marked in the pdf. Those are very valuable info.

No grid resistor R53 for the 300B? I wonder why the original design uses such a high value 22k. It’s going to interact with the Cga and cause some interesting negative feedback at high frequencies. Also, with the 300B, the max output amplitude could be quite limited. You’ll need about -100V bias to get the 300B to 400V @ 10mA on the plate. But the max negative grid swing is limited to about -160V due to the 2Sk216 Vds max. The result is limited positive output swing, or operating the 2Sk216 at risk.

Going two digits after the decimal point probably not going to be very accurate (check the datasheet of your meter for AC accuracy specified in % of reading + % of range and/or number of digits). I normally use a simpler way when calculating the SPL. Sr-007 has a sensitivity of 100dBSPL/100V, so 50mV of pure tone gives you 100-20log(100/0.05)=34dBSPL. The 007 has a f0 around 45Hz (a bit more gain compared to 1KHz) but audibility is also affected by the equal loudness curves, so YMMV. Without an FFT it’s hard to tell the harmonic composition of the noise, which affects its audibility. I guess you are not inclined to feed this into your sound card…

I’d measure the output AC voltage across +/-, provide that your meter is up to snuff (has a 200mV AC range for instance). If there is indeed audible hum from the amp, you would get a reading least in double digit mV.

It depends. Usually you want the attenuation to be at a later stage of the signal chain when possible, since the noise of the previous stage (DAC) will be amplified by the gain of the later stage (headphone amp). You can argue that the DAC's intrinsic noise is low enough that even if amplified by the full gain of the later stage (headphone amp), it is still well below the intrinsic noise of the later stage, in which case it doesn't make much of difference implementing the attenuation at either places. There are also practical concerns. Using a passive attenuator such as a volume pot increases the source impedance of the later stage, which often increases its noise and in some cases, affects its frequency response and linearity (more distortion). Many DACs with brilliant SINAD numbers employ distortion compensation. They are tuned for the lowest measured distortion at 0dBFS input. As soon as you reduce the (digital) input level, the harmonic profile changes quite significantly. The result of the above is that the system may 'sound' different with digital volume control vs analog volume control. Last but not least, one's ears are the most valuable device in the signal chain. Personally I'd rather have an old-school volume pot as close to the headphone amp/power amp as possible as the 'emergency shutoff valve'.

May I ask the reason for reducing R39? Is it because of the STN9360 HFE drop under very low collector current? I did find that STN9360 and 2SA1486 response differently to the compensations I added to improve stability, but that’s probably more due to their different Cob. Thanks!

That sounds normal. Probing the R42 (-) injects noise to the diff amp. Connecting R42 (+) to the same meter reduces the injected noise. Also the active batteries will not be *exactly* the same in term of stability even if you match the parts. To what level did the voltage of the two batteries of the R channel decrease? What are the voltages at the (+) and (-) of the two batteries when that happens? Remember I suggested you to check 1) the K246 CCS 2) whether the active battery still maintains the gain when the output decreases [V(+)-V(-)]/V(R42). If you only have one multimeter, how did you notice the voltage of the battery change when you probe R42? Both the input and output stage offset servo and the global NFB affects the DC offset. So it's important to see what causes the offset to be off beyond their correction capabilities. To be safe, also check the input impedance of your meter. The STN9360 generally have higher HFE than the 2SA1486. Everything else being equal, the stability will be a little worse. Using Q23 with lower HFE may help with stability, but I feel that the oscillation and the thermal-related stability may be two separate issues here. I don't quite understand that some people would spend thousand of $ on parts, building a unit that has a fair market value of close to 10k, while not willing to invest a few hundred $ on quality tools and equipment. If we know Stax only uses a multimeter to build and test T2, we'll probably avoid their amps like plague. How did that become ok when some lucky individuals do the same?

Please try to describe the procedure with more details. 1. Did you use a bench multimeter, or a portable one? The bench meter has slightly different characteristics between pos and neg leads, the portable one barely has any. 2. Did you put the pos lead on the pos node of R42 or the other way around? Did you connect the pos lead first or neg lead first? 3. when you mention “R42+ and R42-“, did you mean the R42s in the two active batteries of the same channel? Chances are the reference chain on the active battery is marginal, such that the input impedance of your meter would throw it off. Check your meter, also check the K246 CCS. The voltage across the LED string and the voltage across RV2 shouldn’t change. If the active battery oscillates, the DC voltage across R42s in the +/- arms of the same channel could be different. Do you have more than one meter? You can monitor the output of the active battery (across the test points) when you measure the voltage on R42.

Use your most sensitive organs. Do not connect the headphones. The sound emanating from the tubes when T2 oscillates a good hearing test, in terms of both sensitivity and frequency response 😆. Seriously, you could use the AC function of your multimeter. Most of them are good to tens of kHz, some good to a few hundred kHz. If something’s going on, there would be a good chance to see it on the AC meter once the reading settles. The active battery circuit can be marginally stable. The capacitance of the test lead can tip the balance and cause it to oscillate. But R42 should usually be safe to probe on. If you see a lower voltage, you might want to check if the output voltage of the active battery (+/-) has dropped proportionally, i.e it still maintains its gain. From there you’ll know whether the reference chain or the amplifier part of the active battery should be checked next.

Yep. Oppo is primarily considered as a cellphone company. The other CE products are their “side business”, at least from shipping volume point of view.

I wish you doubled success with those. Oops, squared!😆