Yes I know. I was being silly. The emoticons are not working for me. Never have.
Cheers,
Lowering the noise-
This can be a clear manageable task. The first step would be to figure out where the noise is circuit limited. Usually that issue would be at the first stage but it may be further down the chain. In the distortion analyzer there is an input amp (discrete in the Panasonic and ShibaSoku). It would be worth figuring its noise floor in nV/rtHz since you will not get below that threshold. Looking at Pchi's reverse engineered schematics of the 725C http://www.diyaudio.com/forums/equi...utomatic-distortion-analyzer.html#post3402699 most of the circuit is high impedance so a noise floor of 5 nV/rtHz won't limit the performance. Something like an LT1468 would be a fine option if the rest of the circuit is compatible. However it may show no real noise improvement. A TL072 is 18 nV/rtHz or equivalent to a 22K Ohm resistor. Some stages are lower Z than that so it could make a difference. However the input bias current could be a problem. An OPA
134 could be an alternative with 8 nV/rtHz.
This can be a clear manageable task. The first step would be to figure out where the noise is circuit limited. Usually that issue would be at the first stage but it may be further down the chain. In the distortion analyzer there is an input amp (discrete in the Panasonic and ShibaSoku). It would be worth figuring its noise floor in nV/rtHz since you will not get below that threshold. Looking at Pchi's reverse engineered schematics of the 725C http://www.diyaudio.com/forums/equi...utomatic-distortion-analyzer.html#post3402699 most of the circuit is high impedance so a noise floor of 5 nV/rtHz won't limit the performance. Something like an LT1468 would be a fine option if the rest of the circuit is compatible. However it may show no real noise improvement. A TL072 is 18 nV/rtHz or equivalent to a 22K Ohm resistor. Some stages are lower Z than that so it could make a difference. However the input bias current could be a problem. An OPA
134 could be an alternative with 8 nV/rtHz.
This is true of every osc/analyzer circuit I have. Thus, the opamp replacements in key places with the most modern, SOTA, low noise, low distortion opamps have been so beneficial in upgrading them.
Thx-Richard
I hope you realize that you are saying exactly the opposite thing. My experience is the same as Demian's, it is VERY difficult (maybe impossible) to maintain and equivalent noise of even 1K Ohm through an entire oscillator signal chain. The simulation/calculation with perfect noiseless amps is easy enough to do.
I am not disagreeing with Demian nor you. I believe you guys. But the proof is in the results. using other opamps to replace the ones originally used has had a benficial affect in both noise and harmonic distortion levels. IMO Demian just says he finds it fruitless to try to go lower than a certain level. I have not said I wanted noise lower than a 1K resistor. Just lower the noise that is contributable by the opamp used in that particular environment. THAT has been accomplished already -- past tense.
Thx-RNMarsh
Thx-RNMarsh
I'm saying from the schematics the noise floor is totally dominated by the passives with fairly mundane op-amps, Demian suggests LT1458 the use of expensive 4-5 nV FET amps will not further reduce the noise. The ultra-low noise bi-polar amps are likely to make it worse due to current noise.
Yes it applies to the 339a too. I couldn't get the noise down past the the stated residual and that's because of the passives.
Ok lets fine tune the subject again --> There is an oscillator and there is an analyzer. The 339a has both. The 725 does not have an oscillator.
I have two 339a here... one is stock and one is modified. I can do real time comparisons. I get a 2-3dB improvement in rms broad band noise levels just with opamp changes. But a much greater reduction in THD.... which was/is the primary goal.
I think the comments of noise in passives compared to the older opamps may not be including the low freq noise levels which are a lot higher in the older opamps than the mid freq levels. Since both the 339a and the 725d have meter displays, you can also see the effect of quieter oscillator circuit at low freq by a steadier meter needle..... its doesnt vibrate and kick around as much with the newer opamps. (nice when tuning for nulls).
Now then, the analyzer side is where i got the most noise reduction.... and the better performance as an analyzer. That is the part, as I have said before, that i want to look at reducing the noise in the other analyzers.... the 7722 and the 725. I have already showed the levels of noise via the analyzer monitor ports... that might also be improved and results in more harmonics being seen with the spectrum analyzer. Without affecting the notch filters, that leaves the output from the notch on towards the monitor port as the only areas to look at.
Thx-RNMarsh
I have two 339a here... one is stock and one is modified. I can do real time comparisons. I get a 2-3dB improvement in rms broad band noise levels just with opamp changes. But a much greater reduction in THD.... which was/is the primary goal.
I think the comments of noise in passives compared to the older opamps may not be including the low freq noise levels which are a lot higher in the older opamps than the mid freq levels. Since both the 339a and the 725d have meter displays, you can also see the effect of quieter oscillator circuit at low freq by a steadier meter needle..... its doesnt vibrate and kick around as much with the newer opamps. (nice when tuning for nulls).
Now then, the analyzer side is where i got the most noise reduction.... and the better performance as an analyzer. That is the part, as I have said before, that i want to look at reducing the noise in the other analyzers.... the 7722 and the 725. I have already showed the levels of noise via the analyzer monitor ports... that might also be improved and results in more harmonics being seen with the spectrum analyzer. Without affecting the notch filters, that leaves the output from the notch on towards the monitor port as the only areas to look at.
Thx-RNMarsh
Last edited:
I think it would be wise to inject a test tone for such comparisons just to make sure the gain is the same. And what's the sensitivity (dB/div) of the analyzer?
Indeed the frontend of the analyzer seems one of the places where there is some room for noise improvement if 70ties opamps are in use, at least for ranges where no input attenuator is used. At which range setting did you do the comparison, and how does this keep up at others?
Samuel
Indeed the frontend of the analyzer seems one of the places where there is some room for noise improvement if 70ties opamps are in use, at least for ranges where no input attenuator is used. At which range setting did you do the comparison, and how does this keep up at others?
Samuel
A stock 339a analyzer has a full power bandwidth of over 500KkHz. To get this using the old Harris op amps left the 339a on the edge of parasitic oscillation It doesn't take much for a stock 339a to be set off. Simply connecting the monitor output to even another HP gear like a HP3581C which is the metered version of your 3580A would cause this parasitic oscillation to start up. Floating earth grounds on the out board gear usually clears the oscillations up. I got the same thing with my scope ground.
Using a LT1468 properly compensated with the correct amount of Fb C eliminates the bursts of parasitic oscillation observed with the Harris part.
My experience with high frequency parasitic oscillations in the megahertz is that it raises the first few harmonics significantly and raises the measured noise floor. This effect appears on the 339a meter as well as the monitor port. The observed increased noise floor measured with FFT agrees and won't be any different using a analog spectrum analyzer.
The effect is not isolated to the 339a analyzer it a matter of fact for all analog circuitry.
I have observed the same with audio range oscillators and amplifiers in general.
With that said then it stands to reason we should be clear as to why the noise floor and distortion drops when changing old 1970's op amps out for newer ones. Changes in bandwidth with the new op amp effect the stability of the circuit. It's unfair to narrow the bandwidth of a wide band analyzer and then claim an improvement because the THD+N drops. That why I said it's necessary to calibrate as you go along with making these changes. If you do this then you will see if the bandwidth narrows and if that's a cause for the observed improvement. With out doing this then simply swapping out components is quite meaningless.
These parasitic oscillations mentioned are often at a very low levels making them difficult to measure. Some times bursts occur at a particular signal level and disappear when signal is removed. A vary small oscillation can be swamped by a large signal making it impossible to see on an oscilloscope but will show up as elevated distortion and noise.
Using a LT1468 properly compensated with the correct amount of Fb C eliminates the bursts of parasitic oscillation observed with the Harris part.
My experience with high frequency parasitic oscillations in the megahertz is that it raises the first few harmonics significantly and raises the measured noise floor. This effect appears on the 339a meter as well as the monitor port. The observed increased noise floor measured with FFT agrees and won't be any different using a analog spectrum analyzer.
The effect is not isolated to the 339a analyzer it a matter of fact for all analog circuitry.
I have observed the same with audio range oscillators and amplifiers in general.
With that said then it stands to reason we should be clear as to why the noise floor and distortion drops when changing old 1970's op amps out for newer ones. Changes in bandwidth with the new op amp effect the stability of the circuit. It's unfair to narrow the bandwidth of a wide band analyzer and then claim an improvement because the THD+N drops. That why I said it's necessary to calibrate as you go along with making these changes. If you do this then you will see if the bandwidth narrows and if that's a cause for the observed improvement. With out doing this then simply swapping out components is quite meaningless.
These parasitic oscillations mentioned are often at a very low levels making them difficult to measure. Some times bursts occur at a particular signal level and disappear when signal is removed. A vary small oscillation can be swamped by a large signal making it impossible to see on an oscilloscope but will show up as elevated distortion and noise.
We've been here before. I looked for oscillations as you had them. I checked with a wideband Spectrum analyzer: Instek GSP-810 (150KHz -1GHz). Did not see any evidence of oscillations. I also used my TEK TDS3032 DPO (300MHz w/FFT option module). Other instruments at my finger tips are an Anritzu MS420B Spectrum/Network Analyzer (10Hz to 30MHz); and an HP 8753D Network Analyzer (30KHz - 6Ghz).
My 339A BW is about 400KHz... the stock one is 590Khz (-3dB). Thats wide enough for me and stable.
BTW - I dont use any of the Harris opamps, now. I have a few new ones and used ones to anyone who needs some.
Thx-RNMarsh
My 339A BW is about 400KHz... the stock one is 590Khz (-3dB). Thats wide enough for me and stable.
BTW - I dont use any of the Harris opamps, now. I have a few new ones and used ones to anyone who needs some.
Thx-RNMarsh
Last edited:
I dont actually recall... was just flipping dials and comparing the two under same conditions.... I think it is probably 1db/div. Am looking for clues as to what and where any improvments could be made... not actually doing measurements..... but another condition showed the large variations in noise level when dialing in a lower freq range on the spec analyzer. This is where the most improvement is at. I attribute this to the low noise in the low freqs of the newer opamps. Since i dont recall all the settings of the time, I'll just delete this picture.
[I have Victor making me a 100Hz oscillator of super low thd at this time. So, I'll probably revisit the noise down there again.]
Thx-RNMarsh
Last edited:
Reading those paragraphs (right side of page 45), it behooves me to point out that the "full complex number" FFT gives both magnitude and phase angle for each bin. This info can be used to good effect. The fundamental's phase angle will be zero (or some constant, presuming the sampling clock and the sinewave generator are locked, as has been discussed). The harmonic amplitude and phase will be, as the article states, the vector sum of all sources (generator and analyzer). Putting a notch filter between the generator and analyzer at the harmonic frequency will remove the generator's component and reveal only the analyzer's component, and this can be vectorially subtracted from the harmonic amplitude and phase measured without the notch filter, resulting in the amplitude and phase of that harmonic from the generator alone. Repeat for every harmonic of interest, and Bob's your uncle.
Yes, this is a lot of work, and no doubt as others have suggested with doing similar things it's all finicky to get "good" readings at these low levels and stuff. It also depends on getting the raw sampled data into a computer where you can read it and do the appropriate complex FFTs, and vector subtraction of the results.
I own an older model dual domain A-P but dont know if it considers the phase as well so that the low level harmonics are considered accurate measurments of them.
Who would know? Besides A-P.
Interesting guy --- computer research scientist: Lomont.org
and --- lomont.org/software/misc/FFT/simplefft.pdf
Thx-RNMarsh
Who would know? Besides A-P.
Interesting guy --- computer research scientist: Lomont.org
and --- lomont.org/software/misc/FFT/simplefft.pdf
Thx-RNMarsh
Last edited: