simmconn

Thanks. I guess nobody likes to receive fake parts. But if it’s a low-current, 1200V or better grade IGBT, its characteristics can come very close to the SiCFET we use in threshold voltage, transconductance and even input/output capacitance. It might actually work, at least in the GRHV. I’d be interested to try it out. Well, we should probably stop talking about SicFET or IGBTin this thread, as they are irrelevant to T2.

You haven’t answered my question how you determined that the part is a fake (other than from the appearance).

If you use them in a constant current source, the Vbe-Ic curve (figure 10) is probably more important than hFE. If you don’t have ways to measure, pick two that are closest in hFE for each channel and hopefully they come close in other parameters as well. If you ask me, I always order at least 2x more than the BOM qty and match them on a curve tracer at or near the actual operating point. The bigger the pool, the more likely you’ll end up with well matched pairs.

If I were to choose from a well executed raster image logo in the marking and a badly burned vector one, I’d opt for the latter in a heartbeat. I’ve explained the reason and would not repeat here. For the hFE, you need to know the Ic and Vce your testing was performed at, in order to make a meaningful comparison with the numbers in the datasheet. If you don’t know, measure them. The datasheet says minimum hFE is 170 at 25 degree C, Ic=-1ma and Vce=-5V. From the chart you can see hFE goes slightly up from Ic=-1ma to -10ma and then to -20ma before it starts to drop after Ic=100ma (the red line). If your sample tests at hFE=150 between Ic=-1ma and -10ma, it’s below spec and would be a reject by the factory. You can see the red line is well above 200 and approaching the 300 line, so 340 is not a surprise. +/-20% is considered normal variation. Why would they specify min hFE at only 120 at Ic=-20ma? It could be a simple mistake, or indicating that the hFE could drop as early as Ic=-20ma, unlike what the typical curve suggests, which is a bad news for circuit designers (not applicable for our applications). Nevertheless, sandbagging would not get them into trouble anyways, if you know what I mean. I could go on and on and brag about my affiliation with the semiconductor industry, but let’s keep the personal information out of this discussion. If you think my comments make sense, think about it. Otherwise, just take it with a grain of salt.

How do you determine if a part is a fake? If a part with a poorly done or a suspiciously looking marking meets all the specifications of the genuine part that you can verify, would you still declare it a fake?

Okay. Please do report back when you hear from Mouser, then I’ll explain why I think hFE of 341 is more reasonable than 150 for an STN9360.

You offered your opinion and analysis, I offered mine. Although we disagree, I hope we can respectfully disagree.

Mouser is legit, ‘Mauser’ is probably not. The one in the first picture is more likely a fake. Even without a known genuine sample for comparison, you can tell by at least two factors: The four top corners of the plastic molding are rounded and inconsistent, suggesting that the package may have been sanded. The ST logo is a raster image composed of parallel horizontal lines, suggesting that it is a scanned reproduction. A genuine product would have a vector artwork since it is the manufacturer’s original design. Other factors such as the laser marking font or etching depth can vary from manufacturer from manufacturer and even from factory to factory. You would need a known genuine sample for comparison. A low cost transistor tester such as DY294 can test breakdown voltages up to 1kV and measure hFE at different collector current settings. You can choose one that’s close to the transistors’ actual operating point. DY294 Digital Transistor DC Parameter Tester Field Effect Tube Tester Multifunction Semiconductor Tester https://a.aliexpress.com/_mqGVojt

Picture 2: An “audio grade transformer “ didn’t care to color code the wires properly, such that the assembly tech has to attach hand-written labels to tell them apart. Maybe only the decal is custom made? Nonetheless it shows the production volume of this kit. Picture 3 and 4: It’s funny they try to use different colored quick disconnect to do dummy-proofing. Did they not know that those quick disconnects are colored differently for a reason? Pink/red for 18GA or smaller, blue for 14/16GA and yellow for 12/10GA. They used similar wire sizes regardless of the requirement of the quick disconnect, and the yellow one is apparently oversized (loose crimping). So much for a “professionally put together” kit.

It’s not hard to do a 1:1 clone of the PCB. It’s also a great interview question for a junior electrical engineer to come up with a circuit that does this. However not a lot of them would probably know how a vacuum tube filament/heater behaves.

There are plenty of suitable candidates for the SiCFET. I count 9 different parts just by running a simple search on Digikey. Look for 1700V rated Vdss, IDmax less than 10A, Ciss less than 240pf, TO-247-3 package. I would prefer ones that are DC-SoA rated, with a moderate transconductance, low and stable Crss across the entire VDS range. Avoid those that are not characterized for linear operation. The vendor doesn’t want to guarantee those use case, and you will be on your own. I would try the onsemi NVHL1000N170M1 if I were to build another KGSSHV Carbon, although there are cheaper options that may be as good. The closest sub for LT1021 is the LT1236. As far as I know they only differ in long-term drift specs. There are other shunt mode 10V references but none come close to the noise performance of LT1021/LT1236. Of course you can also use second-hand, recycled ones. Those are cheap and already aged to perfection (in terms of long term drift).

Ok, I stand corrected. The Megatron XL uses a 'hybrid' biasing scheme for the 300B. The grid is adjustable between 0 and 0.99% of B- ('fixed'-biasing), which cancels out less than 10% of the bias generated by the cathode resistor (self-biasing). The voltage between the cathode (K) and the grid (G) is the actual grid bias (Vgk), regardless of self-bias or fixed-bias. In a fixed bias setup, you’d measure between GND and G when K is grounded.

The Megatron final stage is CCS-loaded single-ended output working in class-A. There is no problem selecting operating point that way. However the final stage is self-biased. The grid bias eats up part of the B-, in other words B- is not equal to Eb in the tube datasheet. 20 to 30mA is more than enough for an estat amp. Since 300B has a low mu (2.85), which is about 1/3 to 1/4 of a triode-strapped EL34, the undistorted output voltage will be less than with EL34. But the DHT fame and the aesthetics probably more than compensate for that.

Not exactly the same part but being the same package they are more in common that they are different: https://fscdn.rohm.com/en/products/databook/applinote/ic/power/linear_regulator/to252_thermal_resistance_information_an-e.pdf In the final stage CCS, the 10M90S is dissipating 8W to 9W. The theta-JA needs to get down to single digit °C/W in order for the junction to stay comfortably within spec, not to mention that the output DC offset thermal drift has always been our enemy. TO-252 is not going to get you there. Also, even if the copper is well coupled with the aluminum angle, the tiny cross section would still give considerable thermal resistance. You are right in that it is just like the Ohm's law. With theta-JC being 3.1 already, there is not a lot of slack to play with. I think IXYS is being sloppy when it comes to thermal specs in the 10M90S datasheet, by specifying only one number for both packages. With some other parts having about 2x theta-JC in TO-252 than in TO-220, I doubt IXYS can do that much better.

Yep, if you had checked the theata-jc of those packages you wouldn’t have bothered with such “test”.

I’m surprised to see the series pot only gets 16dB CMRR at min position. The attenuation at that position would have been much higher than 16dB already, for both common mode and differential mode signals. Did you use the same schematic in your earlier post to connect the series pot when using the APx CMRR measurement? I think the source XLR pin1 should be connected to input XLR pin 1, and series pots’ pin 1,3 should be between the source XLR pin 1 and 2/3 respectively, like how you would connect in a real system.

I think you are not measuring CMRR, but rather the common mode to differential mode conversion ratio. With the ‘conventional’ dual pot connection, the common mode signal at the input is attenuated at the same ratio as the differential signal; while with the shunt connection (without the help of a line isolation transformer), the common mode signal is not attenuated at all. Apx has common mode drive, why not use it to evaluate CMRR?

I like the idea with the 211. It also helps keeping the noise from the filament supply at bay with the common cathode configuration. If matching tubes for the CCS sounds too much, we can always consider the 1700V depletion mode SiC JFET. Thermal management is going to be a challenge though at 35ma.

Nice two-piece design! Actually TE and Molex both have 0.093” (2.36mm) crimp contacts in their portfolio. One can make real cheap Stax sockets using those and a 3-D printed shell. The contacts are so cheap that I think not being able to extract them after assembly would be okay. The gold-plated varieties are more expensive, but I’d feel less guilty than cannibalizing Neutrik jacks. Below is one of the one-piece shells I printed using PETG. However I was not quite happy with the surface finish of the 3-D printed shells. Machined shells using acetal and PEI both give nice surface finishes and are plenty rigid. Perhaps too rigid that the RR1 plug starts to show fitting difficulties. 😅

I posted the pictures to show how I derived the 10.459mm pin center-to-center dimension. The center-to-center distance is critical in designing the plug or socket, because it is independent from the tolerance of the individual pin diameter. 0.4mm error is quite a bit when put in perspective with regard to the nominal dimension. Your own measurement also shows that the pin distance of the RR1 plug is smaller than the Stax plug. If I were you, I'd go back to the drawing board and find out if the target dimension was incorrect to begin with, or wasn't well controlled in production, rather than sweeping the dust under the rug with 'no disruption in functionality'. A smaller plug can force into the socket thanks to the flexibility of the socket contact and/or the plastic shell. If the socket is made of hard material (such as G10 or phenolic resin) and the contacts are held to high tolerance, a smaller plug would have a hard time fully plugged in. There hasn't been a wide-spread problem because most of the sockets either use soft material (such as Teflon) or contacts that are not held to high tolerance (such as the tuning fork style contact used in the Stax sockets). I'm not saying that every RR1 plug has as large tolerance as mine. But if not, you may have a product consistency issue. Maybe hand-soldering the pins on an acrylic retainer (melting point 160°C) wasn't a good idea after all. "ensure everything is built with precision from the start", "adhering to the strict Production and QC directives". Those are easier said than done.

Maybe I’ve got an outlier, but the specimen I have measures only 10.459mm between L- and R+, center-to-center. The distance between L+ and R-/BIAS has similar error that I can visually see it when placed head-to-head with a Stax plug. Without tapping into the proprietary design data of your plug, could you tell us how the Stax plug measures on your end?

Anything new about the 5-pin plug? I hope the pin pitch is fixed this time.

It’s likely that large amount of AC could mess up with DC measurements. The T2 amp stability is a bit tricky. If my simulation is correct, its loop gain 0dB crossing happens with a -40dB/Dec which is unusual. Taking out the 5pf would most likely cause the amp to oscillate. The question is whether your amp is/was oscillating with them in place. An oscilloscope would be needed to make sure of that. With a scope you can also trace back and see where the oscillation originated. Well, the last T2 builder whom I criticized for not having proper equipment shied away, so I’ll refrain from saying anything further. Also in my simulation the C1/R5/R92 compensation network doesn’t have much effect to the stability since their corner frequency is quite low. Even lower is the corner frequency of the balance servo. So they probably don’t have as much effect as the battery voltage. It may have just happened to me that the unit I worked on has 740V well in the middle of the ‘stable zone’. You could test the CCS as individual blocks. Remove the EL34 and connect the anode pin to ground at the socket, then you can test the CCS at equivalent to idle condition. If you’re less confident about your CCS, connect output+ instead of o+ to ground and the 5.1k output resistor will offer some protection. The 5pf cap and the 100k resistor form a zero at about 318kHz. If you increase the cap slightly it may give you a bit more phase margin. But if everyone else’s amp work fine with 5pf, you should not need more.

To get the balance servo out of the way I would disconnect R84 and R85 from the opamps and then tie their loose ends to ground. This mimics both opamps' outputs at the center level. Disconnecting R88 and R90 alone may not be ideal because the opamps have input offset voltage and can still integrate against it over time. However, if you measure near zero at the opamp outputs you can leave it as is and not worry about R84 and R85. Check to see if the J79 is shot or fake. That enhancement mode PMOS FET takes about -0.45V Vgs to conduct 10mA. Note that the mu of EL34 in triode connection is about 10. So under the same plate current, 5-6V of Vg difference would cause 50-60V difference on the plate. Do you measure and match the parts before putting them in the amp? The voltage drop on the LED depends on the current and part-to-part variations. I see some people like to push them all the way down on the PCB. Not only you can't measure the voltage drop easily on the component side, but it's also bad due to thermal stress. I remember when I first used LEDs they were pretty fragile, soldering needs to happen at least 5mm from the body with tweezers to help dissipate the heat. D7 and D8 are in very different locations. You probably mistake one for another. True. The balance between O+/O- can affect the CCS a little bit but not enough to throw the LED off that much. Your starting point is that V(R17)=V(R27). Going down the chain, the cause for D10/D11 (I assume you meant those) to drop less than their counter parts are: 0) poor LEDs, 1) Q11 or Q12 stole too much current (easy to verify by measuring across R21 thru R26) , or 2) Q13 thru Q15 have low Hfe (harder to verify but unlikely). There are two feedback paths that maintain the battery negative voltage, the front-end servo and the global NFB. For the front-end servo, Q26 & Q27 Vg↑ → Q34 Vb↑ → I(R9 & R10)↑ → U1 Vk↓ → U1 Va↓ → U2 Va↓ → Q4 & Q5 Vs↓ → Q26 & Q27 Vg↓ For the global NFB, Q26 & Q27 Vg↑ → Q26 & Q27 Vd↓ → U3 & U4 Vk ↓ → U3 & U4 Va ↓ → U1 Vk↓ (see above for the rest of the loop) It looks like there are multiple problems. So bad parts or questionable PCB connectivity?

I did not say dB/V and dB/mW are the same spec. I said specifically 101dB/V at 8 Ohms is the same as 80dB/mW. If you know why dB/V at 1000 Ohm is the same as dB/mW, you'll understand the math behind my statement.