KGSS Mod(s)

[bobkatz]

Schematics and photos attached!

I got my belated Christmas present from Justin the other day, a lovely KGSS amp. Finally I got to hear my new Stax 007 Mk2 (Omega 2) in all their glory. Knowing that I was going to install my own Goldpoint 1 dB/step attenuators, and a Left to mono switch (to check headphone balance), I asked Justin to agree to construct a KGSS without wiring the input connections to the back panel or installing a volume control (yes, he did test and burn in the board). Leaving me with some fun to do on my own! The first thing I discovered is that the KGSS is extremely sensitive. I judged it has at least 15 dB too much extra gain, even with consumer (nominal -10 dBu) sources. Without an additional attenuator, I even found it easy to incite HF oscillation by opening my Goldpoints fully, with an open unbalanced input. This is not a good idea! So I designed an additional 15 dB attenuator circuit, installed it at the KGSS input. Then I proceeded to listen to a wide variety of reference material through my Benchmark DAC-1. This includes much of my own material which I have mastered and which I know intimately monitored through the Revel loudspeakers in my mastering room. To be honest, I was disappointed with the high end, it was just not open enough nor did the "air" frequencies come close to that of my Revels. Others have reported the Omega 2's to be a little "dark" or closed in and I was prepared for that. This is partly due to the extended bass response of these wonderful headphones, which may be as much as 1 dB too much below about 60 Hz.

So, I decided to design a simple passive equalizer to compensate for the Stax response. This took advantage of the 15 dB pad I had already installed. I figured I might need between 1/2 and 1 dB rise from 10K to 20K, and in Spice I designed the filter. Then I went to the bench, and proceeded to make a baseline frequency response measurement before installing the EQ (which consists of two added capacitors per channel, one on each phase of the balanced line). Much to my dismay, I discovered that this KGSS is rolled off. Flat at 1 kHz, it was -.13 dB down at 5k, -.48 dB down at 10k, -1.1 dB down at 16K, and -1.61 dB down at 20kHz!!! This was the explanation for the loss of air and high end, not the Stax phones at all. I measured this rolloff at the headphone outputs, the same rolloff whether balanced or single ended (relative to ground). I made measurements at various gains and pot positions, and nothing significant changed. Left and right channel measured identically (the same rolloff) within 0.1 dB. I conclude that some internal capacitive losses are occurring, despite Kevin's current-mode circuit. Or some unknown. Clearly, whatever is causing the rolloff is fully symmetrical as the plus and minus outputs of both channels measure identically. It sure sounds like the rollof is within the amp itself. I think that Spritzer has been working on a new version of the KGSS, and I'd like to ask him to perform frequency response measurements on a stock unit and a new or modified unit and report here. I leave the possibility of capacitive losses in my input wiring, but it's doubtful, considering the cable lengths and positions of the resistors. I should have measured the response at the input terminals before I installed the amp in the mastering room, but that is not possible without too much effort on my part. I will do that measurement if someone else reports flat response with his KGSS.

There was full agreement in response between my high-end Fluke meter, my Audio Toolbox, and a classic Ballantine TVM, so it's not my measurement equipment, and I DID measure the test oscillator in the Audio Toolbox with the same tools, and I compensated for the generator's own slight irregularity (it has to be turned down +0.5 dB at 20kHz, but not at 16 kHz and below, strangely).

I am pleased to report that after the installation of the EQ mod, the amp is flat to less than 0.1 dB from 20 through 20 kHz, and it sounds marvelous now! So, what gives? Bottom line: I highly recommend this mod; the one-pole curve of the eq exactly mirrors the losses in the amp, and I no longer sense a loss of air or presence. I'm a happy camper now! I can give anyone who wishes a spreadsheet or fixed calculations for the 1 dB/step attenuator if you wish. Note that the Goldpoint wafer is normally designed for a ladder attenuator, but I didn't like the idea of sending the signal through all those resistors in series, so I designed it as a fixed series resistor with a switched single variable resistor. It's a fully-balanced variable U pad.

KGSS mod input circuit.pdf

Edited January 24, 2011 by bobkatz

[luvdunhill]

isn't this the part when you complain that Justin's specs of 5Hz-45KHz +/-0.1dB aren't accurate and demand your money back? I guess that's more HF, but if you post it over there be sure to update us with the ensuing reaction.

[mypasswordis]

I would say the HF roll-off is a combination of the two, and frankly I'm pretty sure the O2MK2 rolls off faster than ~1.5dB at 20kHZ.(I'm in the O2-is-dark camp) Nice job on the (thankfully) single pole filter though, I'm sure it helped.

Perhaps check the Stax thread for the bass port mod on the O2Mk2, it should help the boomy bass and clean up the midrange a tad.

[kevin gilmore]

The kgss was designed to european inputs which is .6 vrms input which makes 600 vrms output.

A gain of 60db. Or a voltage gain of 1000.

3 people in all of the last 10 years said that the gain was too low, and i told them how

to get an extra 6db by swapping resistors. One person said the gain was too high as he

listens very quietly and i told him how to change the input fet source resistors to

reduce the gain 10db.

Sounds like your professional source is way high in output voltage, something like 10vrms

and even with resistor changes, this would clip the input stage. So an input attenuator

is required.

With a shunt attenuator, and the attenuation at minimum, and no input load resistors

(depends on which of the many versions of circuit board you have) i'm sure its going

to oscillate. The 500k resistors on the new kgsshv board take care of that issue.

Not sure about the frequency response, i'm going to have to look up my data, i did not

have time to do that last night. If the lack of frequency response bothers you, you

certainly could build a T2 which is way flat to well over 100khz.

Edited January 24, 2011 by kevin gilmore

[Craig Sawyers]

The kgss was designed to european inputs which is .6 vrms input which makes 600 vrms output.

A gain of 60db. Or a voltage gain of 1000.

FWIW the T2-clone also has a gain of 1000. My D-A has an output of 3Vrms and that puts the DACT at pretty much centre travel at the typical loud level at which I listen. Takes maybe three clicks each way to compensate for recording level differences. Half way on a DACT is -28dB, and the steps are 2dB over most of the range.

[kevin gilmore]

Actually the gain of the T2 is 1001. 10100/100

[spritzer]

I've never heard about any issues with excessive gain on the KGSS. It is in the same ballpark as the Stax amps and the pot is always in the 11 to 1 o'clock position for me.

As for the FR response of the amp, you are really starting on the wrong end since the SR-007Mk2 is the culprit here. They are very non linear but read the last few pages of the Stax thread about how you can fix that.

[bobkatz]

isn't this the part when you complain that Justin's specs of 5Hz-45KHz +/-0.1dB aren't accurate and demand your money back? I guess that's more HF, but if you post it over there be sure to update us with the ensuing reaction.

Post it over "there"? Where's "there"? I did alert Justin to this post by email. I'm very happy with this amp, and basically I'm curious what mechanism is causing the non-flat response. It would be a perfect complement to Lambas, for example. But I know that Kevin designed it for use with O2s. Can some of you who are capable please measure the frequency response of your units? I've not seen a frequency response or distortion spec anywhere, to be honest. But again, this is not an issue. It's an immaculately-designed PC board and circuit, and who knows, maybe the "Katz mod" will become a standard ).

BK

[nikongod]

I am pretty sure that the OP's "excessive gain" comment comes from the fact that he is using a stepper with 1db steps=24db of range. Nobody notices that they have too much gain when their volume controls are set to -39db 24db+15db=39db Considering that he only needed to knock off another 15db id say the gain is pretty well set.

Why not change the gain with the feedback resistors?

I am curious which inputs were used to test. I mean SE or balanced? I have seen the frequency response described in SIMULATIONS of the BH using SE inputs. I know, I know, its a weak comparison but the topology of the input stage and how feedback is applied are similar.

[bobkatz]

I've never heard about any issues with excessive gain on the KGSS. It is in the same ballpark as the Stax amps and the pot is always in the 11 to 1 o'clock position for me.

As for the FR response of the amp, you are really starting on the wrong end since the SR-007Mk2 is the culprit here. They are very non linear but read the last few pages of the Stax thread about how you can fix that.

Let's face it, Spritzer, you like to listen loud. ;-). We could get into measurements of input level versus SPL, etc., but it's not important. I strongly feel the KGSS has excessive gain with U.S. or Professional level outputs. Either way we can agree to disagree on that. And clearly, for those who wish to try the Katz "eq mod", you have to attenuate to create the RC network. As for whether the problem with the Stax is due to the Stax or the KGSS (or the particular KGSS that I have) why not say that it can be due to both! If the amp does not measure flat, that is a problem and as far as I'm concerned the first place to tackle any perception of HF loss. We all need one or two of you to please carefully measure and either confirm or refute my measurements before I/we do further investigation. As an experienced audio engineer I know how to measure frequency response and I would not have posted if I was not 100% sure of my measurements. The only thing I did not do was measure frequency response at the input terminals to see if there is a loss in the cabling or a capacitive rolloff due to the input impedances involved.

The bass issue intrigues me! I did read about the port mod but concluded that people were saying the port was to fix the "Stax Fart" and it looked to me that sealing the port was not to alleviate any sonic disorders. I'll certainly investigate and try the port mod when I get a chance. I'm quite worried about the procedure to put the earpads back due to that spring, never having tried opening these phones. Does anyone have a link to the instructions for earpad takeapart? Will Mortite work to seal this port? http://www.amazon.com/Frost-B2-Mortite-Caulking-19-ounce/dp/B000LNODSQ

Thanks, all,

Bob

[justin]

Bob,

Did you measure the FR after lowering the gain? i would guess variations in the FR from KG's original amp, which was built at least a decade ago, would be due to changes in open loop gain from use of slightly different transistors, or slightly different voltages of the LEDs

KG's article says -3dB @ 45KHz into a Omega II load. you are right that i do have +/- 0.1dB on my website. that is likely a copy & paste error thats been sitting there for 6 or more years

Regarding gain levels, 60dB is the same as Stax uses in their amplifiers. On the BHSE I did lower it to 54dB because most people are using high output level sources and I thought it was more than sufficient. For most it is, but the occasional user pairs it with a low-output phono stage and runs into problems. This is especially true with the DACT stepped attenuators which save huge dB-steps for the end of the rotation.

Edited January 24, 2011 by justin

[bobkatz]

I am pretty sure that the OP's "excessive gain" comment comes from the fact that he is using a stepper with 1db steps=24db of range. Nobody notices that they have too much gain when their volume controls are set to -39db 24db+15db=39db Considering that he only needed to knock off another 15db id say the gain is pretty well set.

Why not change the gain with the feedback resistors?

I am curious which inputs were used to test. I mean SE or balanced? I have seen the frequency response described in SIMULATIONS of the BH using SE inputs. I know, I know, its a weak comparison but the topology of the input stage and how feedback is applied are similar.

I used the balanced inputs for my listening and measurements. The way I have it set, the unbalanced inputs simply ground the low side of the line going to the selector switch which then goes into the balanced Goldpoint attenuator.

As for the feedback resistor, I'd rather leave Kevin's judgment of how much negative feedback he wants. Good description of what's going on with the stepper pot. To emphasize, holding your attenuators at -39 dB for normal playback seems to me to be a dangerous condition. You could easily blow your ears out by accident! I've got sufficient output level from the DAC to need to hold the attenuator anywhere from about -16 through -3 dB for an extremely wide range of playback sources and music, with the average of good sources falling at about -9 dB on the attenuator. That's an optimum but not excessive setting for me.

Edited January 24, 2011 by bobkatz

[bobkatz]

Bob,

Did you measure the FR after lowering the gain? i would guess variations in the FR from KG's original amp, which was built at least a decade ago, would be due to changes in open loop gain from use of slightly different transistors, or slightly different voltages of the LEDs

Yes, I did.

Makes sense. Justin, can you measure the frequency response of the next amp you build (be sure to test at various potentiometer settings to see if it's affected by the pot). It would be useful for Kevin to step in and talk about this. No one (least of all me) is upset about this turn of events. To repeat, I'm happy with this amplifier! I found a problem, made a solution and perhaps an even better solution can be found.

Here's the exact order of events:

1) I listened and found the amp to have excessive gain, especially for use with 24-step 1 dB/step pots. I was running too close to the bottom of the steppers. At this point I did not listen too carefully to the sound as I did not have good range on the pots.

2) I did NOT measure the response of the amp at this time. I immediately installed the 15 dB attenuator circuit, consisting of two 20K and 1 10 K resistor in a balanced U pad attached directly to the input terminals at the PC board. (Justin, this replaced the 200 Ohm resistors you had installed).

3) I listened, and detected a loss in the high end while listening.

4) I attributed the problems to the reknowned issue with the Stax O2 Mk2. It seemed logical.

5) I designed a circuit to boost the high end to compensate for the perceived loss.

6) Before installing the EQ circuit, I took a baseline frequency response measurement of the amp. It was rolled off 1.6 dB at 20 kHz compared to 1 kHz. Solution: I should attack the amp problem first, before the phones! I redesigned the EQ to exactly compensate for the measured loss.

7) I installed the EQ circuit and remeasured was happy to discover it made the amp flat to within 0.1 dB from 20 Hz to 20 kHz.

I listened! I'm a happy camper. (Well, would be nice to tame the below 60 Hz resonance of the Stax, but it's a pleasant sound nevertheless, will investigate the port mod).

[bobkatz]

I would say the HF roll-off is a combination of the two, and frankly I'm pretty sure the O2MK2 rolls off faster than ~1.5dB at 20kHZ.(I'm in the O2-is-dark camp) Nice job on the (thankfully) single pole filter though, I'm sure it helped.

Perhaps check the Stax thread for the bass port mod on the O2Mk2, it should help the boomy bass and clean up the midrange a tad.

Based on the exact EQ that I applied, and careful listening/comparing with my reference speakers, I would say the O2MK2 (on its own) is no more than -2.5 at 20 kHz, subject to personal taste. Combine that with my measured response of my KGSS before EQ, I'd say that's a deadly combination. I'm not currently tempted to boost my amp beyond its current flat state, it sounds good. However, bright recordings are not excessively bright, and I could see another listener tilting it up another 1/2 to 1 dB at 20K depending on personal taste. This will also change if I do the port mod.

BK

[justin]

when are you going to implement a version of the stax diffuser circuit?? you should do this all passively, and hand wind the inductors

Edited January 24, 2011 by justin

[luvdunhill]

Justin, don't forget that the original did not need any compensation in the output stage as well.

[justin]

Justin, don't forget that the original did not need any compensation in the output stage as well.

forgot about that, KG's ancient pictures are too fuzzy to see if he snuck in any compensation caps

also, the 2SA1156 was changed to 2SA1486 because the 1156 was running too close to its 400V limit, though that was really more of an issue on the Blue Hawaii.

[Craig Sawyers]

1) I listened and found the amp to have excessive gain, especially for use with 24-step 1 dB/step pots. I was running too close to the bottom of the steppers. At this point I did not listen too carefully to the sound as I did not have good range on the pots.

Ah - there's the problem. For your preferred listening levels, and source output voltage you definitely need attenuation to centre the -12dB point of your Goldpoint.

And yes, I guess that the gain of the amp could be changed. But going to say 40dB from 60dB has to done with a little care to make sure it is still stable at the lower gain, or at least has not picked up any overshoot.

[bobkatz]

forgot about that, KG's ancient pictures are too fuzzy to see if he snuck in any compensation caps

also, the 2SA1156 was changed to 2SA1486 because the 1156 was running too close to its 400V limit, though that was really more of an issue on the Blue Hawaii.

Hmmmm, we may be getting somewhere. Can KG take a measurement of the original circuit for reference (or, I assume, your original specs on the Headamp site referred to the original circuit). But you don't have any compensation caps, do you? We could spice this to the end of days, but if you want to measure first, and find which stage is the source of the rolloff, one way would be to remove the negative feedback and (very carefully) measure frequency response at each stage---with a DC isolated, battery-operated voltmeter with adequate bandwidth. Not being very familiar with the fancy current mode topology, I'm not sure 100% if this is the correct procedure. It may be over my head.

BK

Edited January 24, 2011 by bobkatz

[grawk]

I can't imagine a situation where I'd want to have a max attenuation of only 24dB on any amp.

[Craig Sawyers]

I can't imagine a situation where I'd want to have a max attenuation of only 24dB on any amp.

Although Bob did his own attenuator using the Goldpoint roll-your-own version (and so deliberately chose 1dB per step), other commercial switched attenuators like the DACT, and attenuator transformers like Sowter http://www.sowter.co.uk/transformer-attenuators.php are available in fine step versions. The DACT fine is 0.5dB per step(!)

But I'd certainly go for the more standard Goldpoint or DACT set up, with big steps for the first few, then 2dB per step thereafter. I find 2dB per step works well for me.

[luvdunhill]

one more point. The T2 has an input filter that is used to protect the input from unwanted RF signals. However, what many people don't typically realize (as they leave this filter off) is that this filter has a side benefit of improving the bandwidth of the feedback signal in various topologies, for example a differential pair input. It's a subtle point, but one worth mentioning here.

[bobkatz]

I can't imagine a situation where I'd want to have a max attenuation of only 24dB on any amp.

I've been running an optimized 1 dB/step monitor gain control in various rooms and studios around the house for about 8-10 years now. By optimized I mean that the monitor gain has been calibrated to produce 83 dB SPL with -20 dBFS RMS pink noise per channel. After optimizing the power amp gain this way, the control typically runs in a range of 4 to 6 dB around a -9 center, for the majority of music that's ever been recorded, and as high as 0 dB in rare cases (certain audiophile recordings) and as low as -16 dB in rare cases (certain egregiously-distorted and overmodulated hypercompressed recordings). The music can be made as loud as the most insensitive ears can take and as soft as anyone would wish. 24 steps are definitely enough if you optimize the gain structure versus SPL. This includes a range including surround reproduction of wide-range theatrical films (which can be made to sound annoyingly "too loud" at the top of the pot), stereo reproduction of audiophile material, stereo reproduction of well-recorded music, and stereo reproduction of hypercompressed material as hot as "Black Eyed Peas", or "Green Day", for example. Once the gain range has been optimized to this SPL, even the most sensitive listener on the planet will play Green Day at perhaps -18, which still leaves 6 dB at the bottom for extreme circumstances.

I once ran a TKD 1 dB/step attenuator with as many as 48 steps if I recall correctly, and never went below about -16 and found the rest of the steps to be unnecessary, once the gains were optimized. If they're not optimized, then you find the attenuator just runs in a different average position but still within a range of about 6 dB or less on an average day, or 16 dB on the extremes. If after this optimization, anyone who finds they need more than 24 dB max attenuation for normal music listening, AND can find music that they can tolerate near 0 dB on the same attenuator, then I swear they must be deaf.

I also have installed a dim switch on my KGSS to take care of phone calls or in-room conversation if you want the amp to be quiet. That's the only legitimate need for an extra low step below -24 ). You could also make the last step be -40 instead of -24, but a dim switch is much faster, more convenient and more ergonomic. Conclusion: 24 1 dB steps are enough, when the system is optimized.

BK

[bobkatz]

Thanks Kevin. Your contributions to the headphone and audio community are legendary! Your quote below, my reply above.

The reason why I felt the gain was too much was a combination of using a professional level source (which I can adjust, however) and a 1 dB/step attenuator that has to be optimized as it only has 24 steps. This amp has the two 500 K resistors to ground installed on the input side but it is not the HV model. The 500K resistors are included on your schematic at http://gilmore2.chem.northwestern.edu/projects/showfile.php?file=gilmore2_prj.htm.

Who knows how I got it to oscillate (it was painful but I had my hands on the pot, so it was momentary) and it was only with an open input and only the unbalanced input. I don't think I have any ground loops, but the unshielded twisted pair input cables are not that far from the outputs. Let's leave it, the additional input resistors have removed the oscillation (or, simply, plugging in any source) and it's not my concern.

As for the freq. response, it would be educational to see data. What do you think of my suggestion of removing the feedback and measuring the response at each stage in the currently-made unit (Justin says the only changes have been a couple of transistors)?

Thanks again,

Bob

The kgss was designed to european inputs which is .6 vrms input which makes 600 vrms output.

A gain of 60db. Or a voltage gain of 1000.

3 people in all of the last 10 years said that the gain was too low, and i told them how

to get an extra 6db by swapping resistors. One person said the gain was too high as he

listens very quietly and i told him how to change the input fet source resistors to

reduce the gain 10db.

Sounds like your professional source is way high in output voltage, something like 10vrms

and even with resistor changes, this would clip the input stage. So an input attenuator

is required.

With a shunt attenuator, and the attenuation at minimum, and no input load resistors

(depends on which of the many versions of circuit board you have) i'm sure its going

to oscillate. The 500k resistors on the new kgsshv board take care of that issue.

Not sure about the frequency response, i'm going to have to look up my data, i did not

have time to do that last night. If the lack of frequency response bothers you, you

certainly could build a T2 which is way flat to well over 100khz.

Edited January 24, 2011 by bobkatz

[grawk]

It seems to me adjusting the range on the attenuator to match what need makes more sense than changing the gain on an amp that isn't readily set up for that.