Something seriously wrong there with Square output -- don't know sweep rate, so is that RF on the bottom, or is that line-frequency stuff? The symmetry is OK, so that pot is right. I can't figure out how that stuff on the bottom got on the bottom -- that circuit pretty much goes to ground on the negative transition -- is your scope input inverted?
I took more scope pictures and made a delayed time base to blow up that noise (as soon as I get my SD card to mount!)... Not sure of the spectrum that's at. What I already posted is at 0.2ms & 0.2V/div.
In any case, after experimenting with this more, over 40kHz the output starts becoming unstable. Usually, there are visible, but small, "drops" of the needle (IG-18 meter), indicating that the voltage of the sine drops from full 10 scale. From about 70kHz up the needle starts pulsating.
These are two independent - but I guess may also happen concurrently - behaviors. The drops can apparently be cured by carefully coming up in frequency towards the areas of trouble. This obviously messes with the distortion numbers, which seem to go a magnitude higher when this behavior occurs. It messed most of my weekend up as I was coming up with apparently random distortion figures and was not sure why (part of my 334a learning curve is also involved here...). But I think this stability (or lack thereof) is what was causing this.
What would you suggest doing to deal with this? Ideally, I would be absolutely fine with a lower output if the 100kHz sine is clean and stays this way. Hopefully, this will also further lower distortion at midband. My hunch is that it has to do with the adjustments - which I have to redo anyway to make this work with square section off.
In any case, after experimenting with this more, over 40kHz the output starts becoming unstable. Usually, there are visible, but small, "drops" of the needle (IG-18 meter), indicating that the voltage of the sine drops from full 10 scale. From about 70kHz up the needle starts pulsating.
These are two independent - but I guess may also happen concurrently - behaviors. The drops can apparently be cured by carefully coming up in frequency towards the areas of trouble. This obviously messes with the distortion numbers, which seem to go a magnitude higher when this behavior occurs. It messed most of my weekend up as I was coming up with apparently random distortion figures and was not sure why (part of my 334a learning curve is also involved here...). But I think this stability (or lack thereof) is what was causing this.
What would you suggest doing to deal with this? Ideally, I would be absolutely fine with a lower output if the 100kHz sine is clean and stays this way. Hopefully, this will also further lower distortion at midband. My hunch is that it has to do with the adjustments - which I have to redo anyway to make this work with square section off.
Last edited:
Excessive gain in the first or second stage can compromise bandwidth and result in poor HF response. In addition, tuning capacitor mismatch can affect the loop gain, which also leads to instability. But the crazy signal on the square wave needs more investigation and may also help explain why disconnecting the sq. wave causes upset with the sine wave.
Thank you for your clarity.
I assume adjusting the feedback/bias (but especially the former) per lowest distortion will decrease gain, so that's on my to do list. A couple of thoughts here - I am not looking for full 10V out, I can't imagine needing it in my testing / measurements life. I am far more concerned with the distortion figures. And secondly - you recommend, I think, not getting the highest of B+ voltage to the circuit, but instead being a bit conservative with it. I have about 41V (can't recall the exact number), so that may be a sweet spot I hope, but obviously limiting its output swing capabilities.
Also on my to do list - check caps values with DMM (Fluke 189). Hopefully, its precision at this task is enough to get the job done.
But before the two items above - I will get to the investigation as soon as my time (and six months old) will allow. In case any of the transistors in that section is busted, my prime (and available) replacement part is 2N2222A (original is 2N2369) - would you concur?
I assume adjusting the feedback/bias (but especially the former) per lowest distortion will decrease gain, so that's on my to do list. A couple of thoughts here - I am not looking for full 10V out, I can't imagine needing it in my testing / measurements life. I am far more concerned with the distortion figures. And secondly - you recommend, I think, not getting the highest of B+ voltage to the circuit, but instead being a bit conservative with it. I have about 41V (can't recall the exact number), so that may be a sweet spot I hope, but obviously limiting its output swing capabilities.
Also on my to do list - check caps values with DMM (Fluke 189). Hopefully, its precision at this task is enough to get the job done.
But before the two items above - I will get to the investigation as soon as my time (and six months old) will allow. In case any of the transistors in that section is busted, my prime (and available) replacement part is 2N2222A (original is 2N2369) - would you concur?
Last edited:
The smallest tuning cap, approx 500pF, cabe trimmed with a small parallel adjustable cap if it is a little small, or replaced with another compound cap including a trimmer. The process for me was to measure the 5uF cap, and it turned out to be the largest to three digits, of all five caps -- I was lucky. So I used a cap meter to match the 500nF caps three digits to the 5uF caps three digits by paralleling, and so on down the line. But when you get to the 500pF cap, you have to account for circuit C including stray C. The easiest way to do this is to check the osc. Freq. at the top end point of each range -- they should match to 2-1/2 digits if not to three, except at 110kHz -- here, you'll want to use a counter and trim the 500pF cap value so that the frequency is correct. I hope this makes sense for you.
But you really have to find out what is wrong withe square wave first.
The 2N2222 is a fine choice all the way around, so use it any place you need a small signal NPN device.
But you really have to find out what is wrong withe square wave first.
The 2N2222 is a fine choice all the way around, so use it any place you need a small signal NPN device.
I got as far as checking all voltages on the T/S last night against the schematic:
1. All voltages for symmetry fully CW are off by more than 10%, but not by much (they're within 15-20%). The ones for fully CCW are OK.
2. Voltages for Q8 are far more off and potentially symptomatic. On collector, there is basically no change in voltage when running the symmetry control CCW to CW (it should go from 4.3V to 30V; it stays put at about 25V). On its base, variations are from 3.03V (CCW) to 3.25V (CW). It should go from 0V to 2.5V. I suspect Q8 and possibly D8. Thoughts?
I also just briefly managed to readjust with T/S off and run some measurements. The distortion results at lower frequencies were quite stunning (close to HP 334A floor level) at 0.014% (100Hz), 0.012% (1kHz), but it skyrockets from there (0.25% @ 10kHz; 0.3% at 100kHz). With the T/S on, these values are less stellar, but much better across the band (0.085% @ 1kHz to 0.068% at 40kHz). Stability remains very frail with T/S off, though, and settling time seems to visibly go up. If I could get this to work reliably with T/S on (wide range of output without motorboating, stable and clean output over 40kHz), I think I'd call it the day, though I love the project and really want to get the most out of it (and need a signal generator on my bench!).
You mention reducing gain. If I try to do this by using the feedback control (reducing output), I often encountered difficulties maintaining the output stable. This has always been an issue during adjustment per manual instructions - I can't really do the 6-8 on meter scale adjustment they specify and measure with DMM because the voltage bounces all the way between 6 and 8, if I can even get it there with a very sensitive feedback control. If I get it to 10, it's solid and I can continue adjustments. Luckily, bias control is also adjusted through the scope part of the measurements, otherwise it's had to make it when the needle is already at full scale from the first step. Any other ways to reduce gain, other than taking out the CCS from Q5?
Argh! What a humpty-dumpty this has been... I feel I should just order all Qs you mention on IG18#1 and rebuilt this from scratch with new parts.
Radu.
1. All voltages for symmetry fully CW are off by more than 10%, but not by much (they're within 15-20%). The ones for fully CCW are OK.
2. Voltages for Q8 are far more off and potentially symptomatic. On collector, there is basically no change in voltage when running the symmetry control CCW to CW (it should go from 4.3V to 30V; it stays put at about 25V). On its base, variations are from 3.03V (CCW) to 3.25V (CW). It should go from 0V to 2.5V. I suspect Q8 and possibly D8. Thoughts?
I also just briefly managed to readjust with T/S off and run some measurements. The distortion results at lower frequencies were quite stunning (close to HP 334A floor level) at 0.014% (100Hz), 0.012% (1kHz), but it skyrockets from there (0.25% @ 10kHz; 0.3% at 100kHz). With the T/S on, these values are less stellar, but much better across the band (0.085% @ 1kHz to 0.068% at 40kHz). Stability remains very frail with T/S off, though, and settling time seems to visibly go up. If I could get this to work reliably with T/S on (wide range of output without motorboating, stable and clean output over 40kHz), I think I'd call it the day, though I love the project and really want to get the most out of it (and need a signal generator on my bench!).
You mention reducing gain. If I try to do this by using the feedback control (reducing output), I often encountered difficulties maintaining the output stable. This has always been an issue during adjustment per manual instructions - I can't really do the 6-8 on meter scale adjustment they specify and measure with DMM because the voltage bounces all the way between 6 and 8, if I can even get it there with a very sensitive feedback control. If I get it to 10, it's solid and I can continue adjustments. Luckily, bias control is also adjusted through the scope part of the measurements, otherwise it's had to make it when the needle is already at full scale from the first step. Any other ways to reduce gain, other than taking out the CCS from Q5?
Argh! What a humpty-dumpty this has been... I feel I should just order all Qs you mention on IG18#1 and rebuilt this from scratch with new parts.
Radu.
Last edited:
I've been stressing how wrong your square wave signal is and how vital it is to figure out why disconnecting the square wave section is messing up the sine generator. Get all of that sorted out first. After that is right, other factors can be dealt with. It does seem like you have a transistor problem -- replace the apparently bad one.
The IG-18 wants to put out around 10VRMS and trying to get it to do less without making some other changes is a lost cause. Set bias and feedback for a clean waveform at 1kHz or 10kHz, then fine tune feedback for lowest distortion using the 334 at 10kHz. That should get you to the optimum point. This oscillator will not produce really low THD at any frequency; we've just added a few easy steps to make it do better than the stock circuit. It can be roughly an order of magnitude improvement, which is nothing to sneeze at.
The IG-18 wants to put out around 10VRMS and trying to get it to do less without making some other changes is a lost cause. Set bias and feedback for a clean waveform at 1kHz or 10kHz, then fine tune feedback for lowest distortion using the 334 at 10kHz. That should get you to the optimum point. This oscillator will not produce really low THD at any frequency; we've just added a few easy steps to make it do better than the stock circuit. It can be roughly an order of magnitude improvement, which is nothing to sneeze at.
OK, now how about the upset of the sine when you disconnect the square -- is that fixed too? I raise this point because now, removing the square from operation will have a noticeable effect on sine distortion, especially at higher frequencies; but only if you have built the meter buffer stage, which I think I remember you did do...
Yes, it is fixed. I've switched the T/S on and off repeatedly and it doesn't seem like there is much upsetting of sine going on. A barely noticeable shift up and down the Y axis, but that's it.
Now, having that out of the way, here's how this thing is behaving at this point. Up to around 400Hz, I get pretty solid behavior, with an extending settling time as I approach 400Hz (which takes 15 seconds to do it). If I get to 500Hz, the settling becomes motor boating (or takes longer than 60 seconds, which is how long I've watched it on the scope). In other words, it never really settles down.
As a sample, 300Hz with the T/S off reads at 0.024% distortion. Pretty clean, with only the tiniest of spikes, barely noticeable on the 334A's meter. If I turn on the T/S, the read is about the same, but the spikes become noticeable and may go as high as 0.040%.
Now, having that out of the way, here's how this thing is behaving at this point. Up to around 400Hz, I get pretty solid behavior, with an extending settling time as I approach 400Hz (which takes 15 seconds to do it). If I get to 500Hz, the settling becomes motor boating (or takes longer than 60 seconds, which is how long I've watched it on the scope). In other words, it never really settles down.
As a sample, 300Hz with the T/S off reads at 0.024% distortion. Pretty clean, with only the tiniest of spikes, barely noticeable on the 334A's meter. If I turn on the T/S, the read is about the same, but the spikes become noticeable and may go as high as 0.040%.
Last edited:
OK. I think I got this.The key is to have all instruments hooked up and maneuver the feedback until distortion reads a minimum. The scope is essential for times when you've tried too much and the oscillator becomes unstable - it's by far the easiest way to get it back in the ballpark (stable sine).
As soon as I figured out how this best works, I've obtained a minimum at 10kHz (0.017%) and will leave the feedback at that spot. I may be able to go a bit lower, but it's so sensitive and can go way off so quickly (at which point one has to basically start over) I'm not sure I need the 0.001 improvement. At the same time, 0.017% may as well be the floor at 10kHz. Either that, or the HP's floor.
I'm planning to go across the spectrum and see how well it does at this operating point.
Thanks much, Dick.
As soon as I figured out how this best works, I've obtained a minimum at 10kHz (0.017%) and will leave the feedback at that spot. I may be able to go a bit lower, but it's so sensitive and can go way off so quickly (at which point one has to basically start over) I'm not sure I need the 0.001 improvement. At the same time, 0.017% may as well be the floor at 10kHz. Either that, or the HP's floor.
I'm planning to go across the spectrum and see how well it does at this operating point.
Thanks much, Dick.
Something may have been heating up or something, 'cause I can't really get back to 0.017%@10kHz (a few hours down the line). It's more like 0.02% and after fiddling more with the feedback pot it creeped even higher. But - I'm pretty sure I can get it back to 0.017% from a cold starting point.
Here are my plots: 100Hz: 0.015%; 1kHz: a little under 0.001% (HP's measuring threshold; take the value with a grain of salt); 10kHz: 0.017%; 100kHz: 0.082%.
I still have to give another try to 10kHz to re-hit the sweet spot, but it's in really good shape either way at this point.
Here are my plots: 100Hz: 0.015%; 1kHz: a little under 0.001% (HP's measuring threshold; take the value with a grain of salt); 10kHz: 0.017%; 100kHz: 0.082%.
I still have to give another try to 10kHz to re-hit the sweet spot, but it's in really good shape either way at this point.
Thank YOU for your guidance through this. It's definitely been instrumental for my success in obtaining these results. I do appreciate it!
I totally agree with the CCS used as load for an LTP being pretty much ideal. In my modified Dynaco MKIIIs, which I am using with a Triode Electronics board, adding a CCS on the LTP has been a huge improvement of the operation of that stage. The CCSs I'm using with the IG-18 - a JFET cascode CCS - adds a tremendous improvement in the PSRR in the audio band (Walt Jung, Sources 101 P1), especially in low frequencies, where it goes below -130dB. Such a configuration is very close to the ideal CCS for audio applications, from my standpoint. And also, I had to use it because 50V JFETs are kind of hard to find... I had none at hand, and even though the optimal operation of the circuit would not expose the (R12) CCS to full B+ (but rather close to half of it), I started off with a very off circuit where I expected this particular CCS to see over 40V. None of the JFETs I had at hand would have taken that lightly. A cascade CCS sorted out that issue AND provided the performance I wanted.
Anyway, after things settled down these past few days and the lid(s) were screwed back to close everything off, below are my latest plot points. Obviously, as Murphy's laws dictate, I wasn't able to go down to 0.017% at 10kHz again. No sweat, these latest numbers probably represent a more reproducible performance and repeatability is essential for testing equipment.
A few notes before that, though. They are at roughly 10V out, which seems to be the most stable operating point for this thing - a bit over and a bit under, the sine generator would experience instability. I'm happy about this, though, because:
1. Distortion numbers are quite good and they are a good base line for what I am doing (mostly tube work);
2. I can maintain the usability of the panel meter. Not only that I want to be able to use the attenuator as designed, but it also comes handy to know when the oscillator reaches its operating plateau.
Here are the values:
10Hz 0.165%
20Hz 0.077%
100Hz 0.016%
1kHz 0.010% (my 0.001% from my previous post was obviously a typo as that certainly is not the HP's measurement floor...)
5kHz 0.022%
10kHz 0.021%
20kHz 0.024%
50kHz 0.034%
100kHz 0.058%
One note worth making is that the high frequency distortion seems to go down quite substantially if I press the BNC connector at the output of the IG-18. This may either be caused by a loose connection (though my cables and adapters are pretty reliable, I think), or maybe the frequency is high enough that the proximity of my hand influences the measurement (though, again, the RCA cables I'm using are double shielded). In any case, the 0.058% at 100kHz drops down to about 0.040% if I do this. This doesn't seem to happen at lower frequencies.
Another interesting fact is that lower frequencies distortion figures are quite high. Not sure how the ripple plays into this or what else is going on. I do understand that this cannot possibly get excellent distortion figures across the entire audio band and much beyond that.
But this also tells me that there is room for improvement and - as I can never rest with things like this - something tells me I'm going to revisit this project.
Until then... Cheers and enjoy!
Radu.
I totally agree with the CCS used as load for an LTP being pretty much ideal. In my modified Dynaco MKIIIs, which I am using with a Triode Electronics board, adding a CCS on the LTP has been a huge improvement of the operation of that stage. The CCSs I'm using with the IG-18 - a JFET cascode CCS - adds a tremendous improvement in the PSRR in the audio band (Walt Jung, Sources 101 P1), especially in low frequencies, where it goes below -130dB. Such a configuration is very close to the ideal CCS for audio applications, from my standpoint. And also, I had to use it because 50V JFETs are kind of hard to find... I had none at hand, and even though the optimal operation of the circuit would not expose the (R12) CCS to full B+ (but rather close to half of it), I started off with a very off circuit where I expected this particular CCS to see over 40V. None of the JFETs I had at hand would have taken that lightly. A cascade CCS sorted out that issue AND provided the performance I wanted.
Anyway, after things settled down these past few days and the lid(s) were screwed back to close everything off, below are my latest plot points. Obviously, as Murphy's laws dictate, I wasn't able to go down to 0.017% at 10kHz again. No sweat, these latest numbers probably represent a more reproducible performance and repeatability is essential for testing equipment.
A few notes before that, though. They are at roughly 10V out, which seems to be the most stable operating point for this thing - a bit over and a bit under, the sine generator would experience instability. I'm happy about this, though, because:
1. Distortion numbers are quite good and they are a good base line for what I am doing (mostly tube work);
2. I can maintain the usability of the panel meter. Not only that I want to be able to use the attenuator as designed, but it also comes handy to know when the oscillator reaches its operating plateau.
Here are the values:
10Hz 0.165%
20Hz 0.077%
100Hz 0.016%
1kHz 0.010% (my 0.001% from my previous post was obviously a typo as that certainly is not the HP's measurement floor...)
5kHz 0.022%
10kHz 0.021%
20kHz 0.024%
50kHz 0.034%
100kHz 0.058%
One note worth making is that the high frequency distortion seems to go down quite substantially if I press the BNC connector at the output of the IG-18. This may either be caused by a loose connection (though my cables and adapters are pretty reliable, I think), or maybe the frequency is high enough that the proximity of my hand influences the measurement (though, again, the RCA cables I'm using are double shielded). In any case, the 0.058% at 100kHz drops down to about 0.040% if I do this. This doesn't seem to happen at lower frequencies.
Another interesting fact is that lower frequencies distortion figures are quite high. Not sure how the ripple plays into this or what else is going on. I do understand that this cannot possibly get excellent distortion figures across the entire audio band and much beyond that.
But this also tells me that there is room for improvement and - as I can never rest with things like this - something tells me I'm going to revisit this project.
Until then... Cheers and enjoy!
Radu.
Last edited:
And just to give a bit of insight into this last step, based on my experience - the adjustments after the mods are done.
The main adjustment to work with is obviously the feedback control, and you may notice that it may feel much more sensitive than before all this was done. I've done this having a distortion analyzer hooked to output and trying to obtain a minimum. Running the pot through different positions, in my case, resulted in a very narrow range where the sine wave would be stable, and - not sure if luckily or by design - right around full 10Vac out. I spent a lot of time gently nudging the pot down in gain, and repeated may times when I pushed it out of the stability margins. Having a scope hooked up makes this much much easier to do - highly recommended. This is because the waveform allows spotting instability immediately and easily figuring out if it's up or down where you want to go.
When the feedback reaches a position of minimal distortion, adjusting the bias will bring the figures even lower. Not sure if going back and forth helps, but knowing how cranky the feedback control is, I preferred to leave it alone after reaching a minimum at - as richiem recommended - 10kHz. Spend the time with the feedback control - lots of it -, then just fine tune it with the bias control.
I hope this helps.
The main adjustment to work with is obviously the feedback control, and you may notice that it may feel much more sensitive than before all this was done. I've done this having a distortion analyzer hooked to output and trying to obtain a minimum. Running the pot through different positions, in my case, resulted in a very narrow range where the sine wave would be stable, and - not sure if luckily or by design - right around full 10Vac out. I spent a lot of time gently nudging the pot down in gain, and repeated may times when I pushed it out of the stability margins. Having a scope hooked up makes this much much easier to do - highly recommended. This is because the waveform allows spotting instability immediately and easily figuring out if it's up or down where you want to go.
When the feedback reaches a position of minimal distortion, adjusting the bias will bring the figures even lower. Not sure if going back and forth helps, but knowing how cranky the feedback control is, I preferred to leave it alone after reaching a minimum at - as richiem recommended - 10kHz. Spend the time with the feedback control - lots of it -, then just fine tune it with the bias control.
I hope this helps.
Last edited:
Hi Rax -- the LF distortion is worse due to the thermal time constant of the lamp -- it cant stay warm long enough to be stable and the LF signals modulate it -- this is the big limitation to lamps as feedback elements. The solution is to use lamps with more massive filaments that have a longer time constant, but then the oscillator is much more finicky about everything\, since these lamps are lower impedance and require more drive current...
RE the "hand" effect -- you'll find this with lots of gear. It will be less prominent or disappear if you shield the power transformer or make an external power dupply that is well removed from the oscillator. This is also where changing the 600ohm load switch to a floating or hard ground switch can be very helpful -- some gear you will drive will want it floating and some will want a hard ground.
You've done really well on this, congrats.
RE the "hand" effect -- you'll find this with lots of gear. It will be less prominent or disappear if you shield the power transformer or make an external power dupply that is well removed from the oscillator. This is also where changing the 600ohm load switch to a floating or hard ground switch can be very helpful -- some gear you will drive will want it floating and some will want a hard ground.
You've done really well on this, congrats.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.