Hi Bob,
It is too long ago to recall details exactly... but iirc, the shunt R (drain to gate) affected the 2H the most. A series R change would weakly affect the 3H the most. And, of course they interact.
I could get the 2H very low, as I said.... but not the 3H.... that annoyed me because if I could get lower 3H, I could get the total distortion very much lower.
BTW - at those levels, I was not able to read harmonics accurately with the QA400 and was using the A-P.
THx-RNMarsh
It is too long ago to recall details exactly... but iirc, the shunt R (drain to gate) affected the 2H the most. A series R change would weakly affect the 3H the most. And, of course they interact.
I could get the 2H very low, as I said.... but not the 3H.... that annoyed me because if I could get lower 3H, I could get the total distortion very much lower.
BTW - at those levels, I was not able to read harmonics accurately with the QA400 and was using the A-P.
THx-RNMarsh
Last edited:
Wow, I had no idea the 339A oscillator was that good. My 339A is certainly not that good, but is stock without mods.
Cheers,
Bob
Not only is the distortion of a tweaked 339A very low but also the noise.
A stunningly good oscillator considering the voltage level it operates at.
By scale it actually out performs my SVO. My SVO is set at 2.5Vrms output and the 339A
is 7.07Vrms. With no attenuation of course.
I'd like to try my multiplier and control loop design with a bridged T to see if there is a difference. This would confirm my position that the ripple in the output of the control loop is the a limiting factor of prior art.
I do have a white paper here that quantifies the effect of the ripple. The math is more than I care to take on.
The ripple in the output of the control loop is always an interesting matter. It often boils down to settling time at low frequencies.
In my oscillator, the approach was quite simple: a full-wave rectifier that was largely peak-detecting with a filter time constant that was optimized for each frequency decade. There was, however, back-to-back diode agc gain dependent on error amplitude. So amount of ripple getting through was much less under quiescent conditions.
With SVO, one has both in-phase and quadrature versions of the signal available, so one could do a quad-phase rectification using a full-wave rectifier on both the in-phase and quadrature signals. This doubles the ripple frequency. and reduces ripple amplitude substantially. In principle, one could synthesize 8 phases using different combinations of the in-phase and quadrature-phase signals, but it is probably not worth it.
I have also often wondered whether peak detection or average detection of the full-wave or quad-wave rectified signal would be better. Also, of course, there is the possibility of RMS detection.
I've never gotten myself excited enough to try sample-hold.
There are also some experiments that could be done that I don't recall having done, namely ones to find out whether agc ripple, JFET element, or op amp nonlinearity is the dominant contributor of distortion in my oscillator. If one doubles the signal voltage across the JFET element and distortion does not go up substantially, it is probably not the main contributor. Similarly, if one doubles the gain from the agc control signal (or halves the agc filter time constant) and overall distortion does not go up significantly, it is probably not the agc ripple that is dominating.
Cheers,
Bob
Hi Bob -- you've outlined some nice tests and I will move to try them soon. I have also been thinking about something other than the 1/2 wave rectification used in the 339.
I have shared before that I experienced a condition of metastable equilibrium with the 339 oscillator where the FET drain voltage went to zero and the oscillator continued to run stably for almost 10 seconds -- the distortion simply disappeared completely from view, so it is clear that the AGC is the source of the residual -- now I need to know which part is the main factor. I would like to try ripple cancellation rather than extended filtering. But first I have a few other things to do....
I have shared before that I experienced a condition of metastable equilibrium with the 339 oscillator where the FET drain voltage went to zero and the oscillator continued to run stably for almost 10 seconds -- the distortion simply disappeared completely from view, so it is clear that the AGC is the source of the residual -- now I need to know which part is the main factor. I would like to try ripple cancellation rather than extended filtering. But first I have a few other things to do....
I would say sample and hold is the way to go after trying just about everything else.
It certainly settles the fastest. There is the charge injection to contend with. I got around the charge injection by using an ADC driving an Mdac for a peak detector. The so called infinite sample and hold. It's a bit pricey though. But for DIY worth it.
The multiplier is made up of the same ADC and Mdac. The Mdac can take an AC signal as a reference which modulated by the Mdac's code and is inverted by an I/V convertor. The output from the I/V convertor is summed with the non inverted reference input signal through a pair of resistors in a 1:2 ratio. When the code is 50% the output of the summing network is null. Above and below center code the output is either non inverted or inverted proportional to the code. A digitally controlled analog multiplier. The sampled peak detector is ripple free. Only a component of 60Hz is present.
I haven't tried either the multiplier or sampled peak detector apart from one another. So I can't be sure which is ultimately responsible for the greatly reduced distortion. Apart from this the SVO is pretty much the same as your design upgraded by Dick.
We're having this discussion in the QA400 thread. This probably belongs in the oscillator thread.
It certainly settles the fastest. There is the charge injection to contend with. I got around the charge injection by using an ADC driving an Mdac for a peak detector. The so called infinite sample and hold. It's a bit pricey though. But for DIY worth it.
The multiplier is made up of the same ADC and Mdac. The Mdac can take an AC signal as a reference which modulated by the Mdac's code and is inverted by an I/V convertor. The output from the I/V convertor is summed with the non inverted reference input signal through a pair of resistors in a 1:2 ratio. When the code is 50% the output of the summing network is null. Above and below center code the output is either non inverted or inverted proportional to the code. A digitally controlled analog multiplier. The sampled peak detector is ripple free. Only a component of 60Hz is present.
I haven't tried either the multiplier or sampled peak detector apart from one another. So I can't be sure which is ultimately responsible for the greatly reduced distortion. Apart from this the SVO is pretty much the same as your design upgraded by Dick.
We're having this discussion in the QA400 thread. This probably belongs in the oscillator thread.
Andrew, if I under stand this correctly, then this is the proof to which
Davada was alluding, at least from the point of view of resistance.
In my case with the 350D and using the Leader Function Generator
(50 ohms), the 350D wants 600 ohms or nothing. At 50 ohms + 550 ohms
added resistor = 600 ohms.
The attenuation, is -2.9226 dB of the switching attenuator, also called
insertion loss?
Then from that point I can check the accuracy of the attenuator....
AND
With the HP instruments HP339A, Counters, FFTS, Shibasoku, etc if they are already at 600 ohms that's it. I just account for the -2.9266.
And you didn't even work with the log function on the math. : )
The price on the old HP Calcs continues to go up....
Reverse Polish Notation (RPN) is great it makes it hard to use
normal calculators...it's even worse when you get older
and more set in your ways.
HP339A Power Supply Mod
@RichEEM, Bob, David,
I'm following you on this and was thinking about it too.
In a prior life I used to work at Dallas Works (Western Electric,
Bell Labs, Lucent Technologies Power Systems) it was a great place
before they got sold and dismantled.
The key thing I came out of there was how important the
power supply is for all the stuff we do. However, it is not
sexy as changing opamps or caps or FETs or other circuit
or component changes or special cabling... etc, etc.
I had some other fun and games with my HP339a so I'm kind
of not wanting fool with it for a while, but in the back of my
head, I kept thinking what about changing the P/S.
Not only along the lines you are describing, what about using
something like Hexfreds or MURs feeding the regs and updating the
regs in the HP339a and getting that power supply fast as
lightning and smooth as glass?
Bob are those P/S configs discussed in your book?
As I'm trying to understand big concepts here would
the quadrature approach be similar to what ATT might
be doing with their Uverse signals? That is, using vectoring
to run the bit streams at different phases?
If we can do quadrature, why wouldn't octature do it better?
OR
for the effort, 16-ture.
If you think about it, would it be the quadrature squared.
Is there a method of squaring a signal, or squaring the rate?
Can it be applied to Power?
Are there power supply modules, sub components, diodes,
regulators, controllers, opamps, or ICs that might be available to test this?
Any part numbers we might be able to look at?
I'm asking these questions because I don't know, I'm trying to
understand it all.
Didn't Samuel Groner post a schematic of a Multiplier that
he was saying was really good and simple to implement?
It had plots and graphs for determining how to optimize.
Cheers,
@RichEEM, Bob, David,
I'm following you on this and was thinking about it too.
In a prior life I used to work at Dallas Works (Western Electric,
Bell Labs, Lucent Technologies Power Systems) it was a great place
before they got sold and dismantled.
The key thing I came out of there was how important the
power supply is for all the stuff we do. However, it is not
sexy as changing opamps or caps or FETs or other circuit
or component changes or special cabling... etc, etc.
I had some other fun and games with my HP339a so I'm kind
of not wanting fool with it for a while, but in the back of my
head, I kept thinking what about changing the P/S.
Not only along the lines you are describing, what about using
something like Hexfreds or MURs feeding the regs and updating the
regs in the HP339a and getting that power supply fast as
lightning and smooth as glass?
Bob are those P/S configs discussed in your book?
As I'm trying to understand big concepts here would
the quadrature approach be similar to what ATT might
be doing with their Uverse signals? That is, using vectoring
to run the bit streams at different phases?
If we can do quadrature, why wouldn't octature do it better?
OR
for the effort, 16-ture.
If you think about it, would it be the quadrature squared.
Is there a method of squaring a signal, or squaring the rate?
Can it be applied to Power?
Are there power supply modules, sub components, diodes,
regulators, controllers, opamps, or ICs that might be available to test this?
Any part numbers we might be able to look at?
I'm asking these questions because I don't know, I'm trying to
understand it all.
Didn't Samuel Groner post a schematic of a Multiplier that
he was saying was really good and simple to implement?
It had plots and graphs for determining how to optimize.
Cheers,
Last edited:
If you want to understand quadrature rectification, run LTSpice. Put 4 voltage sources on the capture. Run the voltage sources 90 degrees apart. Two of the Vs are simple just inverted copies of the 90 degree quadrature sine. What you will see is four sine all 90 degrees apart. If you sum these source with diodes you will have a quadrature rectification.
The voltage near the peaks never falls lower than 0.707 of the peaks. This arrangement produces less ripple when filtered. The frequency is four time the fundamental of the sine.
it's been tried and tested.The harmonics from the rectifier show up very clearly in the spectrum of the oscillator. With a bit of thought it is possible to double the rectification and have eight peaks at eight time the fundamental frequency.
If it's the Masters thesis you're referring to the peak detector is a track and hold sample and hold arrangement. They are more complex and require great care in getting it all right.
But by far is superior to all other methods. I took it a step further and used an ADC/Mdac pair which simplifies the design but at a dollar cost.
Power supply mods probably wouldn't do much for the performance of the 339A. The op amp have very good PSSR. What I found with my own oscillator is a supper regulator didn't make any difference to the performance nor was it any quieter, A rebuild of the power supply for the 339A would be to overcome the unavailability of original parts. I found the original 339A regulator ICs to be quite quiet. The power supply is sufficient.
@npn -- I don't know about the reference you made to MPAA. I've used a USB "soundcard" and ARTA software very successfully; I think the QA400 and its software are equally good. Your choice. But using spectrum analysis is vital to good results.
@Bob Cordell -- my 339 is stock except for the JFET gate feedback trimmer in the oscillator, and it is just very good. That old HA2625 is really a great opamp and has only recently been exceeded by anything else, at least for oscillators.
I also have an old Heathkit IG-18 rebuilt as a 339 oscillator that has been changed from the actual 339 design to one that has a 10:1 ratio for the tuning caps, instead of the 100:1 ratio of the HP design. If I use that one as a test bed, I'll change the tuning cap ratio back to 100:1, because it does in fact work better, mainly I think because the control voltage range needed at the FET is lower with the larger ratio -- the amp gain is higher.
@Bob Cordell -- my 339 is stock except for the JFET gate feedback trimmer in the oscillator, and it is just very good. That old HA2625 is really a great opamp and has only recently been exceeded by anything else, at least for oscillators.
I also have an old Heathkit IG-18 rebuilt as a 339 oscillator that has been changed from the actual 339 design to one that has a 10:1 ratio for the tuning caps, instead of the 100:1 ratio of the HP design. If I use that one as a test bed, I'll change the tuning cap ratio back to 100:1, because it does in fact work better, mainly I think because the control voltage range needed at the FET is lower with the larger ratio -- the amp gain is higher.
I think I de-lidded an HA2625 in the early 70's and traced out most of its circuitry to see what they were doing while I was working on an op amp for HF. I recall being very impressed, and I believe that they had dielectric isolation and maybe DI vertical PNPs.
Cheers,
Bob
My first "electronic" calculator was RPN.
It was great, very intuitive. The slide rule and log tables it replaced were put in the drawer and rarely came back out.
It took me a short while to learn the key strokes of it's replacement.
Now I forget how the RPN worked, it was way back in the early 1970s.
3*2+5*5 = 26.
Did not have to use brackets because the calculator software assumes that certain operators are to be used in a defined order. If the defined order is not what one requires one gets the wrong answer.
RPN was different and easier in that one did not need to remember the "defined order assumption"
Did not have to use brackets because the calculator software assumes that certain operators are to be used in a defined order. If the defined order is not what one requires one gets the wrong answer.
RPN was different and easier in that one did not need to remember the "defined order assumption"