Not knowing the particular transfomer and load charteristics to be able to model them and calculate the C and RC network for diode snubbing in a particular preamp or power amp. Is there a test procedure an amatuer could follow with a scope to tune the C and RC elements..?
And does the snubbing remain effective in a class AB power amp where the load current can vary significantly versus a preamp with a steady current draw.
And does the snubbing remain effective in a class AB power amp where the load current can vary significantly versus a preamp with a steady current draw.
Last edited:
The snubbing takes care of the resonances caused by the diodes switching on and off to charge the caps in the rhythm of the mains - so to a first approximation independent on whatever the amp needs.
It's the transformer, the diodes and the wiring that is of importance.
Did you read Mark's article?
Jan
It's the transformer, the diodes and the wiring that is of importance.
Did you read Mark's article?
Jan
I expect John was weeding out transistors according to 1/f noise and other noises.
Taking a listen to the noise and selecting accordingly would be another approach.
Keep in mind this was for his 'best' at the time effort.
Dan.
I have found that 0.1uF in parallel with 1uF + 68R across each secondary works well with every power transformer I have tried it on.
Actually, subjectively, just a 1uF film cap will have a very good effect. This does not stop the ringing but does reduce the frequency of it.
Using the network mentioned above adds some damping which reduces the ringing.
Fine tuning the value of R by calculation or experimentation can stop the ringing entirely but subjectively I have found this fine tuning to be un-necessary - anything between 47R & 100R seems to be good enough
mike
Yes. The "softer" the recovery (the greater the tb/ta ratio), the lower the dI/dt, the smaller the stimulus into the RLC resonant circuit. The small signal incremental model (Figure 1c) suggests an easy way to experiment with this analytically: take the inverse Laplace Transform of the secondary circuit's transfer function, stimulate it with a few different I(t) waveforms having different dI/dt, and look at the output waveforms. Or if your Laplace is a bit rusty, let SPICE do the grunt work for you.
The good news is: you have all the time in the world to deal with this. The secondary RLC circuit's natural oscillation frequency omega_naught ranges from tens of kilohertz to thousands of kilohertz (lower for split bobbin EI core transformers, higher for toroids). But the diode charging pulses occur at a very leisurely 120 Hertz. So you have all the time in the world to let the RLC circuit settle, after the dI/dt stimulus; critically damped (zeta=1.0) or even over-damped (zeta=1.5) settling is still several orders of magnitude faster than the input sinewave. You can have as much smoothing as you want; but the appendix shows that some component values grow as a quadratic function of zeta. Thus: you can have as much smoothing as you choose to pay for.
This isn't a 100 kHz SMPS; rules of thumb ripped out of SMPS design handbooks are not applicable here.
Last edited:
IME selecting for absence of noise has almost always been possible. That is flat (white) noise with a 1/f corner at a sub-audio frequency. I would not design with a device where all exhibited a flaw and sorting was necessary especially something unpredictable as GR noise where there is no guarantee that tomorrow it would be different.
BTW the head swelled again yesterday when I was told Rainier Weiss himself calls the 743/745 a favorite op-amp. When he gets the Nobel I'll have a real JC story to tell.
Last edited:
I forgot the punch line, the folks at LIGO asked me yesterday why did we discontinue the 8 legs versions of so many of their favorite amplifiers.
Sure.
I'm talking about the Vendetta Pre Pre.....I expect JC was selecting for total 'sameness'....any channel/channel difference in noise destroys imaging precision.
That's a good feeling.
Dan.
Bún bò Huế (beef, pig's trotters, and blood soup) for lunch today do you think if I ask for it not too hot I'll survive. I figure an American ordering this and wanting it not too hot will trigger a Norman coordinate.
Last edited:
It isn't any mystery what I measured for noise with the Vendetta Phono stage. I measured the low frequency noise (10Hz) with a Quan-Tech 2173 Noise Analyzer.
I selected input devices for noise below 1.2nV at 10 Hz, as measured by the Analyzer. Somewhat noisier devices I used for the second stage. Very high noise (up to 10nV at 10Hz) I sometimes used with the second gain module in place of the complementary mosfet that I originally used. (10 times lower distortion with this last change, and I don't completely know why---upgrade 2 )
Now why did I worry so much about the 1/f noise in these fets? When you think about it, the RIAA curve forces me to give more than 80dB of gain at 10Hz, which is 10 times the gain at 1KHz. This is audible when all the low frequency is much higher than normal, in fact, I have to put in a 400Hz hi pass filter when accurately measuring the midrange noise of the Vendetta, in order to remove the low noise contribution, even with all my pre-selecting. Today, bipolar input stages do a pretty good job in this area, but jfets still have a ways to go. This is why I insisted that Scott Wurcer measure the low frequency noise (even a few samples) of the Phillips fet that he introduced here with a Quan-Tech, rather than rely on his ears only. Sorry about that Scott, but you did what was necessary, and you proved out right about the Phillips devices. The data sheet showed nothing below 100KHz, as far as I remember, and you can't just go by specs, or even previous batches, IF you want to be accurate.
I selected input devices for noise below 1.2nV at 10 Hz, as measured by the Analyzer. Somewhat noisier devices I used for the second stage. Very high noise (up to 10nV at 10Hz) I sometimes used with the second gain module in place of the complementary mosfet that I originally used. (10 times lower distortion with this last change, and I don't completely know why---upgrade 2 )
Now why did I worry so much about the 1/f noise in these fets? When you think about it, the RIAA curve forces me to give more than 80dB of gain at 10Hz, which is 10 times the gain at 1KHz. This is audible when all the low frequency is much higher than normal, in fact, I have to put in a 400Hz hi pass filter when accurately measuring the midrange noise of the Vendetta, in order to remove the low noise contribution, even with all my pre-selecting. Today, bipolar input stages do a pretty good job in this area, but jfets still have a ways to go. This is why I insisted that Scott Wurcer measure the low frequency noise (even a few samples) of the Phillips fet that he introduced here with a Quan-Tech, rather than rely on his ears only. Sorry about that Scott, but you did what was necessary, and you proved out right about the Phillips devices. The data sheet showed nothing below 100KHz, as far as I remember, and you can't just go by specs, or even previous batches, IF you want to be accurate.
WRT the JFET discussion, Dimitri has measured a number of JFETS and published the results in EE (IIRC).
The BF882 came out absolutely tops and by a significant margin. It truly is a superb N channel JFET wrt noise - there isn't anything on the market that comes close (now or historically). No P channel complement, but at the levels you see in MM and MC, single ended still gets you low distortion.
The BF882 came out absolutely tops and by a significant margin. It truly is a superb N channel JFET wrt noise - there isn't anything on the market that comes close (now or historically). No P channel complement, but at the levels you see in MM and MC, single ended still gets you low distortion.
Measurements Rate SMT Low-Voltage n-JFETs Under Consistent Conditions | Power content from Electronic Design
The other amazing thing is the high gm/C ratio, which means that paralleling is eminently feasible for many applications and offers still lower e sub n.
The other amazing thing is the high gm/C ratio, which means that paralleling is eminently feasible for many applications and offers still lower e sub n.
Someone here actually found the reference design schematic (AM front end for car radios). Thanks to the fact that the emergency broadcast system will be AM probably forever there will be an ample supply.
- Status
- Not open for further replies.