Glancing over the Toshiba notes again, I found a serious oversight that I made. You have to subtract .7-.8V from the Vgs (off) and use that voltage to bias the device This makes it MORE difficult to calculate. I will have to work on it, but it may be difficult to do with just any jfet, and get something useful. For example, a 2SK170 can never be TC zero, because its pinch off voltage is too low. Interesting problem.
Therefore for Zero TC, stick with a Zener equivalent reference. Everybody else, it doesn't matter that much, anyway.
Therefore for Zero TC, stick with a Zener equivalent reference. Everybody else, it doesn't matter that much, anyway.
And what did I say about this in the last few pages?
BTW, that's a simple quadratic equation that I routinely had to solve by hand, 30 years ago when I was a student.
Hi John,
I completely agree, and that is how I normally approach doing this if I wanted an average of the entire band. Those older (solid state!) HP AC voltmeters perform far better than most people might think. Both of mine reside with the other equipment that went with my shop when I sold it. Sadly, I don't think the new owners have any clue what they have there.
Why did I suggest using an averaging DVM with fast response? Your question did concern using the 3581A with a tight bandpass filter of 1 Hz. The output of which will obviously be all over the face of the meter. Using a wide band meter would average the entire band in with the frequency you wanted to investigate. That seems to be completely out of the parameters of your question. Anyway, allowing the newer meter to take several peak readings and average those might give you a higher average reading - or not. You'd have to run the experiment and see. Not "you" as in John Curl, the anyone who is interested "you".
I'm not really sure how a filter might actually indicate compared to a proper detect and average method.
I didn't think the method I suggested confused anything. It seemed to be straight forward to me. Building a filter to stick in front of the meter actually may be less certain than simply allowing the 3581A's filters (a known quantity and ready to go) to do the work. Either way, you have another box between the meter and circuit. Using the HP box rather than building another one would be my direction if we needed to filter anything. Then there is the question of input impedance and frequency response to deal with. Again, the 3581A is a known quantity. A constructed circuit has theoretical performance that may not be reality. It hasn't been characterized yet. No thanks!
-Chris
I completely agree, and that is how I normally approach doing this if I wanted an average of the entire band. Those older (solid state!) HP AC voltmeters perform far better than most people might think. Both of mine reside with the other equipment that went with my shop when I sold it. Sadly, I don't think the new owners have any clue what they have there.
Why did I suggest using an averaging DVM with fast response? Your question did concern using the 3581A with a tight bandpass filter of 1 Hz. The output of which will obviously be all over the face of the meter. Using a wide band meter would average the entire band in with the frequency you wanted to investigate. That seems to be completely out of the parameters of your question. Anyway, allowing the newer meter to take several peak readings and average those might give you a higher average reading - or not. You'd have to run the experiment and see. Not "you" as in John Curl, the anyone who is interested "you".
I'm not really sure how a filter might actually indicate compared to a proper detect and average method.I didn't think the method I suggested confused anything. It seemed to be straight forward to me. Building a filter to stick in front of the meter actually may be less certain than simply allowing the 3581A's filters (a known quantity and ready to go) to do the work. Either way, you have another box between the meter and circuit. Using the HP box rather than building another one would be my direction if we needed to filter anything. Then there is the question of input impedance and frequency response to deal with. Again, the 3581A is a known quantity. A constructed circuit has theoretical performance that may not be reality. It hasn't been characterized yet. No thanks!
Just as well, because I really hated calibrating chart recorders. What a pain! From a calibration technician's viewpoint. At least you end up with a hard copy to back up your numbers though.
-Chris
Well, I thought that it would be easier, but it is still not impossible. Toshiba has shown examples of Zero TC operation, with the 2SK30 for example that look good and easily doable. I haven't tried every device yet, to find an optimum, but it is a moot point anyway, because the follower buffer would have to be compensated at well. However, with a bipolar buffer follower, the Vbe would drop, and the beta would increase over a reasonable temperature range, so maybe a TC on the jfet is not such a bad idea, after all.
As far as low impedance drive, the LM329 is ONLY good for that, IF you don't add an RC noise reducing filter, so therefore you would have to live with 7-100uV from your source. I can't do that, myself. Perhaps others can, with their circuits. IC's sure are great for power supply rejection.
As far as low impedance drive, the LM329 is ONLY good for that, IF you don't add an RC noise reducing filter, so therefore you would have to live with 7-100uV from your source. I can't do that, myself. Perhaps others can, with their circuits. IC's sure are great for power supply rejection.
Hi dimitri,
Hi John,
I don't think going to a 2SK117 or JAN # is going to ruin the quality of a CCS that much. It might be good to find some acceptable JAN parts to get away from dependency on Japanese suppliers. Normal experimenters have too much trouble getting these, and now I do too. My normal guy who bought directly from Japan retired several years ago, choking my supply of real devices off.
-Chris
EDIT:
Hi John,
Absolutely! And that is the reason why I put the plastic down for a 3585A. I'm still paying it off as a matter of fact, but as a tool - it's wonderful! Later on, a 3562A or equiv. would be a welcome addition.
Hi John,
It's been explained to me that a long channel device works better as a CCS. The noise created by a J-Fet will be attenuated anyway by your Vref. Assuming your CCS looks like a 350 K resistance (not unreasonable), the other part of the AC noise divider will be the dynamic impedance of the reference. A few ohms for an LED, 5 ~ 12 ohms for a zener depending on where it's operating at. That's a lot of noise attenuation. If you are looking at an active load, as in a load for a transistor collector, you get less attenuation unless your CCS is cascoded in order to get up into the several meg. ohm region.
I don't think going to a 2SK117 or JAN # is going to ruin the quality of a CCS that much. It might be good to find some acceptable JAN parts to get away from dependency on Japanese suppliers. Normal experimenters have too much trouble getting these, and now I do too. My normal guy who bought directly from Japan retired several years ago, choking my supply of real devices off.
-Chris
EDIT:
Hi John,
At low frequencies only. Everything goes to pot past 10 KHz typically.
Last edited:
Anatech, remember 'Bounce,bounce, bounce'? If I am not mistaken, you are stating what I was concerned about when using a wave analyzer with narrow pickup. Thanks for stating:"Your question did concern using the 3581A with a tight bandpass filter of 1 Hz. The output will obviously be all over the face of the meter." Please remember, I said it first, and this was my intent when I said: 'Bounce, bounce, bounce' because that what it can look like. Please understand that I was recommending wide band settings, not narrow, in order to reduce the measurement inaccuracy. I regret adding a passing comment to the person who recommended narrow bandwidth measurement.
True, and the 60dB gain has to be low noise. A low noise opamp (nomina odiosa here, I guess) will do, but the entire setup becomes complicated.
Your argument for erratic LM samples for noise, when combining a low impedance goal strengthens my curiosity about a string of 4 generic 1.8Vf Red or Yellow or Green color Leds well bypassed only with a high value low ESR cap, fed by low noise Jfet CCS, being the two trick pony in the end. We will know if syn08 will run a test for them too.
Edit: Make that a 3 trick pony. Resistant to feeder Jfet's Idss Tc related wandering VS a relatively high value resistor, too.
Hi John,
Not wanting to "get into it", but the 'Bounce,bounce, bounce' was not what I had a problem with. In fact, it should be obvious that I agree with you from my recent posts, and the fact that what you said was a truthful reality. One thing I am familiar with is using instrumentation to some degree.
I have respect for most everyone here, which includes both you and syn08. But everyone is different and personalities can clash. This is more likely behind a monitor and keyboard than in person.
I hope to met you in person one day John. At the very least, you would be interesting to talk to in person I think.
The line of discussion is fruitful right now, let's continue. There are no lasting issues worth worrying about.
BTW, 2SK30A might work well for CCS duty. They also have a 50 VDC breakdown, 10 V more than the 2SK170. I'd like to find a number commonly available from US inventory to use. Most have a breakdown in the 25 VDC region, some up to 40 or 50 VDC.
I wonder how noisy / quiet the current regulator diodes are. Too bad they are so expensive to use. Can't get much simpler though, and some of those have 100 VDC breakdown voltages. An example would be http://www.vishay.com/docs/70195/70195.pdf
The prices shown at Mouser will scare any adult!
-Chris
Not wanting to "get into it", but the 'Bounce,bounce, bounce' was not what I had a problem with. In fact, it should be obvious that I agree with you from my recent posts, and the fact that what you said was a truthful reality. One thing I am familiar with is using instrumentation to some degree.
I agree with that as well, but the bandwidth setting will be forced by what the focus of the measurement is. I'm not making a statement on the suitability of instrument settings. I'm only commenting that if that setting is required, there is a way to get meaningful numbers from it.
We both do, and that was the reason for my post.
I have respect for most everyone here, which includes both you and syn08. But everyone is different and personalities can clash. This is more likely behind a monitor and keyboard than in person.
I hope to met you in person one day John. At the very least, you would be interesting to talk to in person I think.
The line of discussion is fruitful right now, let's continue. There are no lasting issues worth worrying about.
BTW, 2SK30A might work well for CCS duty. They also have a 50 VDC breakdown, 10 V more than the 2SK170. I'd like to find a number commonly available from US inventory to use. Most have a breakdown in the 25 VDC region, some up to 40 or 50 VDC.
I wonder how noisy / quiet the current regulator diodes are. Too bad they are so expensive to use. Can't get much simpler though, and some of those have 100 VDC breakdown voltages. An example would be http://www.vishay.com/docs/70195/70195.pdf
The prices shown at Mouser will scare any adult!
-Chris
Hi Salas,
Why four diodes in series? Or is this the voltage you wanted to end up with?
Hi Steve,
I was merely hoping that Toshiba had scanned the book in. The easiest way would be to cut the spine off the book and use an auto-feeder on the scanner. No, I expected to pay something for a book, back in time. They simply did not seem to be available to us. One day, maybe.
Just imagine all the technical information that is sitting around, unused and possibly being thrown out.
Thank you for looking Steve,
-Chris
Why four diodes in series? Or is this the voltage you wanted to end up with?
Well, to filter the supply, you need to use a resistance in series with the reference before your circuit. If you don't, then all this will do is create a time delay as the capacitor slowly charges up. It will not really filter that much out when you compare it to the impedance of your Vref.
Hi Steve,
My gosh Steve! I wouldn't actually ask anyone to do that. Especially not you! That's asking way too much of anyone.
I was merely hoping that Toshiba had scanned the book in. The easiest way would be to cut the spine off the book and use an auto-feeder on the scanner. No, I expected to pay something for a book, back in time. They simply did not seem to be available to us. One day, maybe.
Just imagine all the technical information that is sitting around, unused and possibly being thrown out.
Thank you for looking Steve,
-Chris
Hi John,
The signature of a technician ...
I had a business partner that stupidly placed all my early books on a freshly painted bookshelf! He was the idiot that painted it! I think your imagination is filling in the blanks. What do you say to someone that dim? They are all gone, I got angry just seeing them like that. Cost me a lot to replace most.
-Chris
That only proves you use it.
The signature of a technician ...Yes, they really should. I have a library where I collect technical books (and read them). There is nothing like sitting down with a book, you can't get the same appreciation with a computer and PDF.
I had a business partner that stupidly placed all my early books on a freshly painted bookshelf! He was the idiot that painted it! I think your imagination is filling in the blanks. What do you say to someone that dim? They are all gone, I got angry just seeing them like that. Cost me a lot to replace most.
-Chris
Hi Salas,
Why four diodes in series? Or is this the voltage you wanted to end up with?
Well, to filter the supply, you need to use a resistance in series with the reference before your circuit. If you don't, then all this will do is create a time delay as the capacitor slowly charges up. It will not really filter that much out when you compare it to the impedance of your Vref.
-Chris
In the previous pages I was cooking for a DC stable practical to make with available in my severely limited common DIYer's drawer components, very low noise about 7V Vref for biasing the base of the cascode BJT for 4 parallel 2SK170BL low MC input stage, since I am looking to get a 0.25mV MC. So I want to boost my gain avoiding as much hiss as possible.
You suggest I feed from the voltage reference node through a 100R to a big shunt cap if I get you right?
Attachments
Ok John, now that you realized that it's not at all easy to bring a JFET at the zero tempco point (BTW, you didn't consider the large Idss variations and the large noise impact of a source resistor) perhaps it's a good time to address a few more things you didn't seem to fully understand.
First, your statement here: http://www.diyaudio.com/forums/soli...rch-preamplifier-part-ii-189.html#post1961619
Let's see how this goes. Let's consider the simplified case of a JFET having the gate connected to the source, feeding a resistor to the ground. Obviously, the current through the resistor is Idss. Now, asume you need an output voltage of Vo volts. Therefore, the resistor has to be R=Vo/Idss.
So far so good. Now, the JFET is at Vgs=0 therefore the transconductance is Gm=2*Idss/Vt where Vt is the threshold/pinch voltage. How much is Au, the current source gain? Obviously Gm*R
And guess what's the final result, after a simple substitution above? The gain is Au=2*Vo/Vt. It does NOT depend on Idss or otherwise directly on the JFET Gm. The only chance to lower the current source gain is to use a JFET with large pinch voltage.
Let's plug in some numbers:
Assume Vo=19.5V (as in your Vendetta phono preamp) and a few JFET options:
K170BL with Vt=0.7V and Vn=1nV/rtHz
J201 with Vt=1V and Vn=6nV/rtHz
J203 vith Vt=2.5V and Vn=4nV/rtHz
Doing the math, from an output noise perspective, the values are:
K170: 38nV/rtHz
J201: 162nV/rtHz
J203: 43nV/rtHz
As you see, there is absolutely no direct relationship between the Gm and the output noise. J203 has the largest transconductance and still it's at par with the K170, noise wise. OTOH, J201 has the lowest transconductance and at the same time by far the worst noise.
Therefore, here's the bottom line: Given an required output voltage, use the JFET with the lowest noise and the largest pinch voltage. Determine the required resistor, based on the device Idss. The output noise will be: 2*Vn*Vo/Vt + the resistor noise, where Vn is the device noise.
Of course, a large Vt JFET will have a second order effect on the noise: the voltage across the JFET will be Vt and the power dissipation Vt*Idss. This may heat a little the JFET (in particular if it's a SOT323 SMD part), hence it's noise will increase. Of course, small Vt JFETs won't experience the same degree of heating.
To be continued...
First, your statement here: http://www.diyaudio.com/forums/soli...rch-preamplifier-part-ii-189.html#post1961619
Let's see how this goes. Let's consider the simplified case of a JFET having the gate connected to the source, feeding a resistor to the ground. Obviously, the current through the resistor is Idss. Now, asume you need an output voltage of Vo volts. Therefore, the resistor has to be R=Vo/Idss.
So far so good. Now, the JFET is at Vgs=0 therefore the transconductance is Gm=2*Idss/Vt where Vt is the threshold/pinch voltage. How much is Au, the current source gain? Obviously Gm*R
And guess what's the final result, after a simple substitution above? The gain is Au=2*Vo/Vt. It does NOT depend on Idss or otherwise directly on the JFET Gm. The only chance to lower the current source gain is to use a JFET with large pinch voltage.
Let's plug in some numbers:
Assume Vo=19.5V (as in your Vendetta phono preamp) and a few JFET options:
K170BL with Vt=0.7V and Vn=1nV/rtHz
J201 with Vt=1V and Vn=6nV/rtHz
J203 vith Vt=2.5V and Vn=4nV/rtHz
Doing the math, from an output noise perspective, the values are:
K170: 38nV/rtHz
J201: 162nV/rtHz
J203: 43nV/rtHz
As you see, there is absolutely no direct relationship between the Gm and the output noise. J203 has the largest transconductance and still it's at par with the K170, noise wise. OTOH, J201 has the lowest transconductance and at the same time by far the worst noise.
Therefore, here's the bottom line: Given an required output voltage, use the JFET with the lowest noise and the largest pinch voltage. Determine the required resistor, based on the device Idss. The output noise will be: 2*Vn*Vo/Vt + the resistor noise, where Vn is the device noise.
Of course, a large Vt JFET will have a second order effect on the noise: the voltage across the JFET will be Vt and the power dissipation Vt*Idss. This may heat a little the JFET (in particular if it's a SOT323 SMD part), hence it's noise will increase. Of course, small Vt JFETs won't experience the same degree of heating.
To be continued...
Last edited:
- Status
- Not open for further replies.