Sound Quality Vs. Measurements

Status
Not open for further replies.
Dejan,

The limited choice of locally available FETs is dictated by the requirements of Japanese TV sets and some audio, I think.

As said, 2SK246 (BL preferred) and matching 2SJ103 can be used. It is a tradeoff. Transconductance is very low (next to 2SK170/2SJ74), but so are parasitic capacitances and pinchoff voltage are high, so they are easier to bias, more resistant to RF, plus they have their zero tempco at around 1mA and at higher currents can be used offset Bipolar transistors drift diamond transistor style, so there is some mileage there.

They are mostly in the trip of "if it ain't a FET of some kind, it can't play music well". Well, maybe a tube or two, but that's it.

Tubes, Fets and BJT's are all devices that have uses. What they precisely do best varies. I like my tubes and like J-Fets for small signal, but I never really got on well with Mosfet output Amp's.

For solid state I feel compound structures that mix low capacitance lateral or J-Fets as "drivers" to BJT's allow a more design freedom and remove some of the probelms in designing SS Amp's rather neatly.

For a real extreme VAS, how about 2SK246 & PNP Rush cascode (and same on the other rail)? Now that should be sufficiently freaky to make everyone scratch their heads... You could do the same Fet Follower drives common base BJT trick in the input stage too... Many interesting options if you combine Bipolar and Fet...

Ciao T
 
DVV please send me your email address and I will send you what I am about to use from Parasound, which is the A21. The JC-1 is bigger and better, but it is TOO powerful for my needs. You can learn and compare from what I send you. Mostly you will find more fets where you put bipolar transistors.
This design, derived from the Ampzilla, is pretty good, but it is about 40 years behind the times. Yes we were making as good, for all practical purposes, 40 years ago, as prototypes, just like you are doing, in fact, I had been using the comp-diff bipolar input stage since 1968, when I worked at Ampex. BUT about 1972, I shifted to the complementary differential jfet input stage that I developed independently as well, and since then, I have never gone back to bipolar input stages, for all the reasons that T describes so well.
 
That seems eminently plausible --- somehow the footnote got lost. As far as which is superior, it depends on whether you want a negative tempco of collector current or not. I recall an opamp using standard two-diode bias
which did so partly to make the transconductance of the nondegenerated input pair more constant with temperature.

To clarify: the LED bias will correspond to a low-tempco, the two diodes to a negative tempco, the latter being appropriate to some extent for stabilizing a diff pair gm.

Brad

Actuallay, back-to-back, with some thermal compound, two transistors make a great current source - extremely thermally stable.

The problem with LEDs is that they are rather different between them, depenind which you happen to have. Much like ground - theoretically, ground is ground is ground, but we all know what happens if you carry that view into a design - ground loops, osciallations, niose, distortion.

I just went through a cycle of very poor quality LEDs. I use them as visual indicators of the filter status. In their day, in the 70ies, LEDs were sad to have a life expectancy of 80 years, and I lived to see that shrunk down to 14 days in some cases. Chinese products of godawful quality, don't last longer than about a month. Back to Germany for my material sourcing.

Classic diodes, like 1N4148, are FAR more reliable and last forever. As do the ancient ZPD (0.6W) and ZY (1.3 W) series of zener diodes.
 
T is correct DVV. The 2SK246-j103 combination should be adequate in this case, just buy a number of them, maybe 50 or more each for easy matching. They are CHEAP!

Muchas gracias for the confirmation, John. As you may remember, I mentioned the fact that Thorsten and I have "known" (via the Internet) each other for some 11 or 12 years now, from other forums. This is why I tend to heed his advice - experience has taught me that I'll be the better for it.

And if bought Thorsten just one beer for every piece of advice he gave either me or someone else over time, Thorsten would be in a contual stupor. :D:D:D But, I am forever the optimist, I still believe that some day, somewhere, somehow, I will get to buy him beer in person. Even if sometimes I don't quite sound the optimist.

Just one note - 50 is NOT a safe number. I could get lucky, true, but I hate haphazard attempts, so it's more like 100. Usually, out of a population of 100, one can count on getting 4, perhaps 5, really well matched pairs. To me, "really well matched" means +/- 2% or better.

I still grieve the day Analog Devices discontinued their MAT series - those were down to 1% easy. Costly, but well worth the money if you wanted outstanding.
 
DVV please send me your email address and I will send you what I am about to use from Parasound, which is the A21. The JC-1 is bigger and better, but it is TOO powerful for my needs. You can learn and compare from what I send you. Mostly you will find more fets where you put bipolar transistors.
This design, derived from the Ampzilla, is pretty good, but it is about 40 years behind the times. Yes we were making as good, for all practical purposes, 40 years ago, as prototypes, just like you are doing, in fact, I had been using the comp-diff bipolar input stage since 1968, when I worked at Ampex. BUT about 1972, I shifted to the complementary differential jfet input stage that I developed independently as well, and since then, I have never gone back to bipolar input stages, for all the reasons that T describes so well.

I'm at dvv@beograd.com , with an (almost) unlimited capacity to send or receive.

Ampex, huh? I loved those 70ies machines, they actually left tanks envious of their build quality. But over here, Studer/reVox was King, Ampex machines were rare, and it was local nuke explosion when Radio Belgrade's most prestigious Studio 5 purchased an MCI machine in the late 70ies. They must have had a thousand Studers and reVoxes, with some Nagra thrown in for good measure, and then, shock! horror! not a Studer, but an MCI. Wow!

I do miss open reel machines. They were such FUN!
 
Dejan,



As said, 2SK246 (BL preferred) and matching 2SJ103 can be used. It is a tradeoff. Transconductance is very low (next to 2SK170/2SJ74), but so are parasitic capacitances and pinchoff voltage are high, so they are easier to bias, more resistant to RF, plus they have their zero tempco at around 1mA and at higher currents can be used offset Bipolar transistors drift diamond transistor style, so there is some mileage there.



Tubes, Fets and BJT's are all devices that have uses. What they precisely do best varies. I like my tubes and like J-Fets for small signal, but I never really got on well with Mosfet output Amp's.

For solid state I feel compound structures that mix low capacitance lateral or J-Fets as "drivers" to BJT's allow a more design freedom and remove some of the probelms in designing SS Amp's rather neatly.

For a real extreme VAS, how about 2SK246 & PNP Rush cascode (and same on the other rail)? Now that should be sufficiently freaky to make everyone scratch their heads... You could do the same Fet Follower drives common base BJT trick in the input stage too... Many interesting options if you combine Bipolar and Fet...

Ciao T

Now you REALLY have me shifting gears into 6th.

Some time ago, I did a constant current source by cascoding two bipolars, with LEDs on their bases, and, as you quite rightly pointed out, it had people scratiching their heads. Now, combining FET and bipolar, that will produce some headaches. :D Good idea, though. That I have to try.

Completely agreed on MOSFET output stages. They (almost) always leave me with a feeling of unfinished business, somehow I always feel something has been left out, though I can't put my finger on it.

Only two amps with MOSFET outputs I remember I really loved. One is an old Perraux model from 1980 (I think?), rated at 100 wpc, using the old Hitachi devices; somehow, that did it for me, it sounded memorable through all these years. The other is a copy of Goldmund's Mimesis 3 (I think?), a nominally 50 wpc model - knowing the guy who did it, I believe it is a very faithful copy, illegal as it may be. That one also sounded memorable. But in general, I am not amused. :D

In your last paragraph above, you mention a "Rush cascode"; I am not sure I know what that is, perhaps I do but not under that name, could you elaborate a bit, perhaps with a diagram, if you have the time?
 
Stick with Studer, I switched to Studer analog tape recorders in 1974. In 1979, I 'improved' Studer by throwing away the electronics and making my own for Mobile Fidelity, and later for Wilson Audio in 1983. Studer made great transports, but relatively lousy electronics, getting worst with each succeeding generation. Amazing, but true.
 
I routinely use LEDs from Chicago Lamps for bias purposes. A bag of ~200 left.

Bought a grab bag of 200 to build a "Red Light District" voltage bias tube amp. Man were they all over the place. I think I paid $1.50 for all, so it was worth sorting them to get strings of 11 Volts. Got sidetracked and did not complete it. What I have learned, I probably will go a different direction. It is a "Mengus" 6P1 amp I hacked up. Cheap as they come. I learned a lot about tubes. IIt is about time to just sell it and my HK.
 
Studer made great transports, but relatively lousy electronics, getting worst with each succeeding generation. Amazing, but true.
I 100% agree with you, Mr.Curl. They threw baby out with the bath water, when trying to fix phase response and also making the machines more "user friendly" with the help of numerous op-amps and logdacs. It still amazes me how people sing praise songs to these machines based on the looks only...
By the way, Fred Thal would love to see you on the Studer mailing list ;)
Studer Info Page

Best,
 
So some of you lot have spent your entire lives designing and building audio amplifiers..? But never actually designed one that you could actually say was good enough? I'm just trying to imagine a different field of engineering where someone would spend decades re-hashing a basic design without ever getting it to work properly!

(slightly tongue-in-cheek, but not completely!)
 
Hi,



All of this shows you how a devices parasitics change if voltage cross or current through the device change.

They allow you to derive the various parasitic resistance (early effect induced limited collector impedance, Beta changes, Changes of the internal emitter and base impedance as well as the parasitic capacitances), so they allow to tell you what goes wrong and where if you change voltages and/or currents (which you will as a transistor that works only static is not useful in an amplifier)...

Of course you can just blissfully ignore all the non-linear parasitics and throw a ton of looped feedback around the Amp instead...



That could be related to the fact that Motorola Semiconductors was spun off and is now called "On-Semi", just as Philips Semiconductors is now NXP and Hitatchi Semiconductors is now Renesas...

Ciao T

Sanyo bought Motorola? Guess it is better than Wang-Chung, but it seems the end result is the same. I new NXP as they say so. I did not know Renesas either. Hitachi completed selling their disk division to WD this week. Interesting.

Not ready to plug and pray yet, as I think I can understand a little more. My basic instinct is that loop feedback is how you fix what you can't in each stage. It is easier not to make a mess than to clean it up, but it will still need a bit of polishing. My level of understanding this could really only suggest my instinct means I need a shower.

I was playing in Spice with CCS's. Wow, not quite as constant as I had been lead to believe and way off a Spice model CCS. A thought occurred to me. Should not the CCS, and for that matter CM, be several times faster than the signal circuit? The other thing that surprised me was when doing AC analysis in the CCS, how it varied across the bandwidth a lot more than I expected. What I think this is telling me is how the gain of the transistor varies over frequency. I am going to play with different transistors in the model ( great how they don't blow up) and see what this teaches me about the topics I don't understand, all in the H parameters I guess. Baby steps. Waiting on Amazon. Books shipped. I need to fix my friends Kenwood. What a PITA to get into it. The phone stage is in a box behind the front panel.
 
"All over the place" in what way?

For my last few RLD builds, I just used Avago HLMP6000s right out of the bag. Worked perfectly.

Voltage drop. I wanted to get each string within 50mV or so to prevent current hogging. I screened them at the current they would see and if I remember, as much as 30 mV difference. It was well over a year ago. I put the project aside when I remodeled by office. (Stickely style). Never pulled it back out.

You liked the result? Maybe I'll finish mine. I changed the inputs to 6ay7's, improved part quality and learned how to set the bias and balance with a spectrum analyzer. It is better, but was still by far the worst amp I owned for my wife's critical hearing test. Of course, now I understand the only thing worse that a BJT driven by a hi-z load is a tube. I should listen with a preamp and the input 50K pot all the way up, all screaming 6W of it. I bought different outputs, but I don't even remember what they are. I guess I got dissinterested in tubes. I have only heard one modern amp, a Cary. Highly distorted but in a very musical way. I can see why many would want to live with them.
 
So some of you lot have spent your entire lives designing and building audio amplifiers..? But never actually designed one that you could actually say was good enough? I'm just trying to imagine a different field of engineering where someone would spend decades re-hashing a basic design without ever getting it to work properly!

(slightly tongue-in-cheek, but not completely!)

I've designed several, but they did not go beyond prototypes. Now it is time to start some production.

"Time to throw stones ended, it is the time to collect stones".
 
Last edited:
Set up properly (and built as shown), the distortion is negligible- not in the triple-ought class of the best solid state designs, but low enough (double-ought). The raison d'etre is overload recovery- it's much harder to hear clipping when it doesn't choke things for a few hundred milliseconds after each event.

30mV in a string is not a lot of difference (a couple percent) and doesn't change the array performance significantly. If it bothers you, put a 4R (or so) resistor in series with each string to balance things out. That adds less than an ohm of impedance at the cathode, which is as close to nil as you need to be.

Quality of the output iron is everything in a good tube amp.
 
Status
Not open for further replies.