kgst

[luvdunhill]

I would try and rotate the toroid so where all the wires exit doesn't point towards the inputs.

[gepardcv]

Thanks for the suggestions. I rotated the toroid to point directly at the back panel, but it didn't help much. Anyway, this is more of a theoretical problem than a real one. The hum disappears once I plug in a source, and the amp sounds terrific.

 

Many thanks to Kevin and Birgir for sharing this design and helping with the build. Thanks to everyone who helped me with parts — @headinclouds for the transformer and matched 6S4A quads, @luvdunhill for input JFETs. Thanks to everyone else who answered my endless questions.

 

If the Carbon can improve on this, I'll be exceedingly impressed. Looking forward to building one.

[luvdunhill]

What is between signal ground and the chassis, electrically / circuit-wise?

[gepardcv]

Full list of connections (other than the obvious ones without which the amp wouldn't work at all):

 

- XLR input: pin 1 to chassis

- XLR output: pin 1 to chassis

- Star ground (SG) to chassis

- IEC ground to SG

- Transformer to SG

- PSU ground to SG

- Pot input and output grounds to SG

- PSU ground to amp ground (both HV and LV lines)

 

I also daisy-chained my power front switch LED from one of the amp boards +15V (through a 3k resistor) and ground lines. I thought that might be bad, and tried disconnecting it, but it made no difference.

[Earspeakers]

Let it burn in and check bias over several days, and try swapping tubes. Also use a HV probe to test the bias, not a straight DMM which will be about 100VDC too high. 

 

I had something like that too from these issues. One of my tubes died (early failure) so I swapped a new one and did a thorough checkup, now its ready to play after a minute and is dead silent. 

[luvdunhill]

IEC earth must go to chassis. You might need a ground loop breaker between star ground and chassis (anti-parallel diodes, RC, etc.)

Edited July 13, 2015 by luvdunhill

[gepardcv]

@luvdunhill: The IEC module I use (Schaffner FN1394-10-05-11) has its earth line connected to the chassis through its body, according to my multimeter.

 

@Earspeakers: You mean something like this? I have plenty of matched tubes, it's easy enough to swap them and see what happens. But is that probe really necessary? It seems to be designed for thousands of volts, not just 580.

[kevin gilmore]

. Also use a HV probe to test the bias, not a straight DMM which will be about 100VDC too high. 

 

 

nope, a standard 10meg input impedance dmm is going to read about 30% to low.

 

the high voltage probes have a 1Gohm impedance, so far more accurate.

[kevin gilmore]

IEC earth must go to chassis.

 

absolutely true for safety reasons. AND FOR NO OTHER REASON!

 

but for minimum hum, everything inside the chassis including power supply ground, rca ground, and any ground referenced

outputs should float from the chassis. works way better that way.

[gepardcv]

nope, a standard 10meg input impedance dmm is going to read about 30% to low.

 

the high voltage probes have a 1Gohm impedance, so far more accurate.

 

Seriously?! So after painstakingly setting my bias to 580V from the test point, it turns out I'd biased for ~400V instead? Live and learn!

Edited July 14, 2015 by gepardcv

[kevin gilmore]

The test point on the power supply is low impedance! So 580 there is 580 no matter what kind of meter you use.

 

most of the other amps don't have test points, and the only thing you can do is measure at

the stax jack which is 4.7M away from a low impedance voltage source.

 

for all of my newer power supplies, the 580v is generated with zener diodes, and is not

adjustable like the much older supplies.

Edited July 14, 2015 by kevin gilmore

[nopants]

maybe there's some merit in jury-rigging the zener string to the old power supply's doubler

[Earspeakers]

The test point on the power supply is low impedance! So 580 there is 580 no matter what kind of meter you use.

 

Good to know thanks, I had heard third hand to always use the HV probe. But for example, for some setting the HV probe shows 680 (or something like that) while the regular shows 580, whereas if low impedance they should be the same?

Edited July 14, 2015 by Earspeakers

[sorenb]

Is this BIAS measuring more complicated than the DMM forming a voltage divider when doing the measurement?

Isn't that always the case and a 'problem' when measure something with high output impedance like the BIAS supply?

 

... but is it more complicated than 'guessing' that the output impedance  is 5Meg (if that is what your string adds up to) and compare it against the 10Meg of the DMM ...so one would have 2/3 over the DMM and needs to multiply the measured value with 3/2 to get the real value ?

While this might not be strictly correct in an academic sense, I guess it is good enough for any pratical purposes.

 

Kevin has added a test point on some of the PSUs that connects before the 4,7Meg or what ever one has choosen for the balast resistor, there you only have 200kohms or there abouts ....and the DMM will measure something which is like 2% off .... since the BIAS voltage is set by a series of zeners seeing a value in the neigborhood of the expected should be good enough ... not much to do about it other than changing some zeners ...

 

How a high voltage probe with high impedance apparently can bump up the voltage I do not understand

Edited July 14, 2015 by sorenb

[spritzer]

Measuring before the ballast resistor will give the correct value unless the meter is seriously fucked.  This is high voltage so using a quality meter is the rule here. 

[kevin gilmore]

this is all about the real impedance of the dmm. Almost all of which are specified as 10M.

But when you actually measure them, all over the place. especially if you have over-voltaged

the thing a few times. So its a random number generator.

 

The real ultra high impedance meters are expensive, and the high voltage probes are +/-10%

 

it would be nice to get rid of all the zeners in the bias supply with some ultra accurately

regulated power supply. it would  be much lower noise, but way more complicated.

pretty much another hv supply with smaller caps

Edited July 14, 2015 by kevin gilmore

[chinsettawong]

I really like the adjustable bias voltage.  It's really useful for testing my DIY headphones.

[nopants]

How do you determine the ideal bias voltages for your diy headphones? Sound? 

[nopants]

I've been playing with the servo in my KGST build.  I had a couple of problems. 

 

First, at very high volumes, the servo would fail.  I fixed that by moving away from the LT-1021-10 which requires about 1.3mA to run properly.  I moved over to the LM4040-10 which wants about 150uA.  I set it just a bit higher at 170uA.  Problem solved

 

Second, I have a small oscillation of about 100mV around 6.7MHz.  This one has been a bit trickier for me.  I figured I'd post the schematic here.  This schematic has the updated cascaded current source, which is currently not on my board.  The model runs fine.

 

The schematic uses different values for the sand, but they are closely matched to the actual parts used.  Here's where I'm at:

 

 

The way it works is that I built a rough op amp (similar to the T2 batteries) which functions as an error amp to hold the input to Q1 at 10V (voltage divider via R9, R10, R13, R14 against R11) same as the input at Q2.  It uses Q3/Q4 (I've used a single Darlington transistor) to create the proper current (about 20mA) to hold the output at 0V.  This is similar in principle to the T2 servo, but need to push/pull in the opposite direction (why I used the error op amp approach).

 

It works great (barring the oscillation) and holds the output within a volt or so from cold to fully warmed up.  It is a simple average of the +/- legs.  Since the oscillation is common mode, outside the audible range and relatively small, I can't hear it. 

 

Any thoughts would be appreciated.

 

Hey Kerry, are you willing to share any of the models you've been using for LTspisce?  I'm interested in messing around with this to test out some modifications to the newer KG amplifiers.

[sorenb]

I really like the adjustable bias voltage.  It's really useful for testing my DIY headphones.

I second that 

I guess there are more people interested in a variable BIAS design ...

[gepardcv]

but for minimum hum, everything inside the chassis including power supply ground, rca ground, and any ground referenced

outputs should float from the chassis. works way better that way.

 

To make sure I understand the exact connections required:

 

- IEC earth to chassis

- Transformer to chassis

- XLR input grounds (pin 1) to XLR output grounds (pin 1)

- Pot input grounds disconnected

- Pot output grounds to amp boards

- PSU grounds to amp boards

 

The KGST has two ground points on its two terminals — I assume only one of these needs to be connected to the PSU, and one to the pot output?

[Kerry]

Kevin has a new version of the servo that he's using with the carbon. I can publish a link later

[Earspeakers]

A generic Chinese meter reads

• 577 VDC LV probe
• 631 VDC HV probe

Keithley 2015 audio bench top DMM

• 590 VDC LV probe
• 645 VDC HV probe

I haven't run a calibration on the Keithley. I could take it into work to get cal'd but probably not necessary, I think the measurement can be trusted, and it does correlate with the cheap Chinese. 

 

Taking the Keithley data the difference is 55VDC which is 9% of the LV value, but Kevin meant +- 10% of the HV probe resistance. According to the data sheet for mine the spec is +-10%, interestingly the data sheet for the 40kV version has a spec at +- 1%. Further, the data sheet for the HV probe specifies it needs to be used with a meter with a 10 Mohm, but the input impedance of the Keithley is 1MOhm. Not sure about the Chinese meter. 

 

Have to think about this

Edited July 14, 2015 by Earspeakers

[kevin gilmore]

unscrew the red barrel on the hv probe and notice the accuracy of the resistor.

[Earspeakers]

looks like one of those Caddock type resistors - an etch on a ceramic substrate or some such

Edited July 14, 2015 by Earspeakers