I have been pursueing two strategies -- one is to notch out the 2H or dominant harmonic. The other is to cancel it out. The first is the easiest and first to get done and demonstrated. The latter, is the best but more difficult. There are a lot of ideas in the amplifier design area using cancelation techniques that might provide insights for oscillator distortion reductions.
Thx-RNmarsh
Last edited:
I ran into this reference in the SRS DS360 manual http://www.thinksrs.com/downloads/PDFs/Manuals/DS360m.pdf ;
This gives a pretty good idea of how SRS implemented the idea of a tuned filter after a DAC. The DAC http://www.analog.com/static/imported-files/data_sheets_obsolete/209524523AD1862.pdf is listed as obsolete and is not in the current best tier of DAC's. A newer DAC might give a significant improvement over the already good performance of the instrument.
This gives a pretty good idea of how SRS implemented the idea of a tuned filter after a DAC. The DAC http://www.analog.com/static/imported-files/data_sheets_obsolete/209524523AD1862.pdf is listed as obsolete and is not in the current best tier of DAC's. A newer DAC might give a significant improvement over the already good performance of the instrument.
Nothing beyond second order state variable. If you go to page 139 of the PDF the parts values of the filters are listed. You may be able to reconstruct a filter from those and see what they are doing. My guess is a simple Butterworth would be enough since its not used for the other waveforms. They are not using a Sallen Key filter (probably because of the distortion) even though it would save a few opamps.
update on Shibasoku 590AR
I mentioned earlier that the Shibasoku was not consistant in its distortion. I verified the performance of the 725 with Viktors oscillator. I was able to get a second harmonic in the -140 dB range consistently. This gives me some confidence in the analyzer working right.
The 590AR has a limited number of adjustments. It has a level adjust for the main out, trims to adjust the reference voltage range (+/- 1 dB from the fine level control on the front panel) DC voltage adjustment and DC offset adjustments for three stages. No adjustments for distortion.
First I adjusted the DC offsets. The three adjustments all interact. With some fiddling I got all the offset test points dialed to within a few millivolts. The unit meets its distortion spec but the second harmonic is around -125 dB. Fiddling I discovered that the offset trims can drop the second harmonic by 20 dB. A further round of tweaking managed to get the output offset to less than 10 mV and the distortion to -140 dB or better. However the internal offsets were as much as 400 mV.
I'm concerned that the small offset (internal balance a pot between the input pair JFET sources) would have such a large effect. I have some alternate MC1495 analog multiplier chips to try and see how much they affect the distortion. I'm left wondering if I'm cancelling distortions or really removing some non-linearity.
I mentioned earlier that the Shibasoku was not consistant in its distortion. I verified the performance of the 725 with Viktors oscillator. I was able to get a second harmonic in the -140 dB range consistently. This gives me some confidence in the analyzer working right.
The 590AR has a limited number of adjustments. It has a level adjust for the main out, trims to adjust the reference voltage range (+/- 1 dB from the fine level control on the front panel) DC voltage adjustment and DC offset adjustments for three stages. No adjustments for distortion.
First I adjusted the DC offsets. The three adjustments all interact. With some fiddling I got all the offset test points dialed to within a few millivolts. The unit meets its distortion spec but the second harmonic is around -125 dB. Fiddling I discovered that the offset trims can drop the second harmonic by 20 dB. A further round of tweaking managed to get the output offset to less than 10 mV and the distortion to -140 dB or better. However the internal offsets were as much as 400 mV.
I'm concerned that the small offset (internal balance a pot between the input pair JFET sources) would have such a large effect. I have some alternate MC1495 analog multiplier chips to try and see how much they affect the distortion. I'm left wondering if I'm cancelling distortions or really removing some non-linearity.
Variable auto-null notch filter - Shows potential -
Well I finally got around to testing the idea of the HP-334A being a tunable filter if the input stages/buffers et al were cleaner.
Recall the disconnect of the meter makes a huge improvement .... yesrterday I just replaced A2R37 - a 3K resistor with 6mA flowing thru it -- with a selected jFET as a CCS. This is a gain stage so I wondered if the thd could be lowered with more feedback using this simple test.
I did get a modest improvement when viewing the harmonics via the QA400 on my new PC. I also got another improvement when I adjusted the amp bias.... however, it only lowered 3H harmonic... but I'll take it.
Together, it looks like it is possible to add a few more CCS in key gain stage areas and get enough improvement to be usable under .01%. That is good enough for many power amplifier tests. May be useful for some work as a flexible auto-null notch filter. But, replacing the amps with some modern opamps would make it very attractive. The way the circuitry/pcb are laid out it would be fairly easy to do.
Here is what I get out of the 334A's monitor distortion residual port with just those few changes mentioned already. A good beginning for someone. Fundamental is 1.5KHz. The panel meter disconnected. Shows some promise after all. Just dont expect a ShibaSoku level of performance nor a sys2 A-P. But still a usefull, low cost tool.
Thx-RNMarsh
Well I finally got around to testing the idea of the HP-334A being a tunable filter if the input stages/buffers et al were cleaner.
Recall the disconnect of the meter makes a huge improvement .... yesrterday I just replaced A2R37 - a 3K resistor with 6mA flowing thru it -- with a selected jFET as a CCS. This is a gain stage so I wondered if the thd could be lowered with more feedback using this simple test.
I did get a modest improvement when viewing the harmonics via the QA400 on my new PC. I also got another improvement when I adjusted the amp bias.... however, it only lowered 3H harmonic... but I'll take it.
Together, it looks like it is possible to add a few more CCS in key gain stage areas and get enough improvement to be usable under .01%. That is good enough for many power amplifier tests. May be useful for some work as a flexible auto-null notch filter. But, replacing the amps with some modern opamps would make it very attractive. The way the circuitry/pcb are laid out it would be fairly easy to do.
Here is what I get out of the 334A's monitor distortion residual port with just those few changes mentioned already. A good beginning for someone. Fundamental is 1.5KHz. The panel meter disconnected. Shows some promise after all. Just dont expect a ShibaSoku level of performance nor a sys2 A-P. But still a usefull, low cost tool.
Thx-RNMarsh
Last edited:
BTW - that is excellent results for your T&M system. Good news that it is verifiable, too. When ever i had offsets, they always affect the 2H as well. It really needs just better matched transistors to reduce the distortion and the offset. But, I too would go for the distortion reduction first a that is what we are after. I dont think the offsets hurt anything but make the next stage (if direct-coupled) have more distortion and then its offset adjustement range has to be greater.
-RM
I'm now struggling with it shifting when I move it. i moved the generator back on top of the analyzer and the 2nd went up. I think its sensitive so some local noise source or grounding issue. After much fiddling and verification I have it back to approx -140 I think. I am cross checking it using the B&K 1607 to attenuate the fundamental by 40 dB which leaves the signal at 100 mV, enough for the analyzer to lock and read. After much calibration verification I have the 2nd harmonic at -27dB on the -80 dB scale or -107 + the -40 dB of the fundamental attenuation or -147 corrected by +7 dB (loss from the 1607 at that difference from the fundamental, a separate calibration drill) yields -140 dB for the second harmonic.
Measured using FFT on the monitor out direct (no 1607) I get -127 for the second harmonic, same using the distortion monitor.
I need another way to verify these numbers. I suspect the input amp of the Shibasoku 725 could be limited to -127 second harmonic. Maybe I'll open it up and see if I can tweak it. But I need to retweak the generator with the 1607 so I'm not near the system limits.
This is very close to counting angels on the head of a pin How many angels can dance on the head of a pin? - Wikipedia, the free encyclopedia .
Measured using FFT on the monitor out direct (no 1607) I get -127 for the second harmonic, same using the distortion monitor.
I need another way to verify these numbers. I suspect the input amp of the Shibasoku 725 could be limited to -127 second harmonic. Maybe I'll open it up and see if I can tweak it. But I need to retweak the generator with the 1607 so I'm not near the system limits.
This is very close to counting angels on the head of a pin How many angels can dance on the head of a pin? - Wikipedia, the free encyclopedia .
Tremendous effort, Demian. the input amp may have a similar 2H trim for balancing/cancelling... maybe the jFET pair on the input has one or needs to have one added there? With such verification, we can trust the results and also better know and understand the limits of sound cards, QA400 etc.
-RM
-RM
Last edited:
What you're observing is exactly the main difficulty with distortion cancellation--the distortion sources observed at these low levels we're concerned with are usually not stable enough for accurate cancellation. That makes IMHO distortion cancellation not the best (as suggested before in this thread), but rather the worst approach to achieve lower distortion. Removing the initial distortion source is far more dependable (and feasible, as I hope my results indicate).
Samuel
If we can get one more filter stage into an SVO with out breaking the bank....
Demian at one point I had grounded copper foil wrapped around the polypropylene caps in my SVO. This lower the distortion by about 7dB and cleaned up a lot of noise. This little experiment shows how much the caps act as antenna picking up fields off the switching circuits in the AGC. Mainly the zero crossing detector which is rich with harmonics. A bit of attention to things like this can go a long way. Trying to solve something like this is in another way is a source of frustration. Look for areas that might have large peak currents. Try a bit of shielding here and there and see if there is difference.
Demian at one point I had grounded copper foil wrapped around the polypropylene caps in my SVO. This lower the distortion by about 7dB and cleaned up a lot of noise. This little experiment shows how much the caps act as antenna picking up fields off the switching circuits in the AGC. Mainly the zero crossing detector which is rich with harmonics. A bit of attention to things like this can go a long way. Trying to solve something like this is in another way is a source of frustration. Look for areas that might have large peak currents. Try a bit of shielding here and there and see if there is difference.
Last edited:
I'm working with three separate oscillators all capable of -127 dB second harmonic per the 725. Since they all have the same limit I suspect that is the 725's internal limit. I'm understandably reluctant to muck around inside the 725. Its really dense and very well made. the attention to detail in its manufacturing is remarkable. Even the relay cans are all meticulously soldered to ground. There is a DC offset trimmer on its input stage and I may fiddle with it. However I'll switch to using the B&K 1607 in front of it for now. I think I need to build a small single frequency version that does what I need- specifically attenuating the fundamental about 20-40 dB so the harmonics are well above the internal limits of the analyzer.
The Shibasoku oscillator (590AR) has all of its tuning components including some of the caps for the AGC on the bottom side of the chassis with a metal plate separating them from the circuitry. The KH4400 doesn't.
My KH4400 is reliably getting to -127dB 2nd H on the 725 after some judicious tuning of the distortion adjust (otherwise as shipped). It has more 3-5 harmonic than the 590AR but they are all low. Viktor's oscillator has the lowest higher harmonics of the three. It also has lower noise than the KH4400, leading me to explore changing the opamps/drive circuitry in the KH4400.
The problem with ever more powerful microscopes is that you keep seeing new things. . .
The Shibasoku oscillator (590AR) has all of its tuning components including some of the caps for the AGC on the bottom side of the chassis with a metal plate separating them from the circuitry. The KH4400 doesn't.
My KH4400 is reliably getting to -127dB 2nd H on the 725 after some judicious tuning of the distortion adjust (otherwise as shipped). It has more 3-5 harmonic than the 590AR but they are all low. Viktor's oscillator has the lowest higher harmonics of the three. It also has lower noise than the KH4400, leading me to explore changing the opamps/drive circuitry in the KH4400.
The problem with ever more powerful microscopes is that you keep seeing new things. . .
When I made my 1 kHz low distortion oscillator I used the method described in this article:
PCB Layout Tricks?striped capacitors and more - The Signal - Archives - TI E2E Community
I have not tried to turn the capacitors the "wrong way" though, so I don't know how much influence it has on the performance.
PCB Layout Tricks?striped capacitors and more - The Signal - Archives - TI E2E Community
I have not tried to turn the capacitors the "wrong way" though, so I don't know how much influence it has on the performance.