Amplitude of sidebands at a specified delta f, very hard at 1kHz rather than 100MHz (possibly).
Not hard with fft but lots of uncertainty about the measurement. Analog measurement with a swept spectrum analyzer is limited by the filter by and skirt. One could measure the time interval variation for timing but that will not show am. If we sweep the output of a notch filter with the spectrum analyzer will it show meaningful noise info?
Sent from my SGH-M919 using Tapatalk
Sent from my SGH-M919 using Tapatalk
Noise sidebands are uncorrelated - I don't think FFT will work if it uses any averaging.
Cheers,
Bob
Hi Bob,
As Samuel pointed out the noise in an oscillator is not white. One method proposed is to stop the oscillator and use a notch filter tuned to the same frequency of the filter to reduce the peak gain to unity so a noise measurement can be treated as white.
The Shibasuko 725 reduces the noise with a digital process, synchronous averaging, then band pass filters the reconstructed harmonic waveform. This is done at the output of the analog analyzer so the fundamental has been removed. This technique allows for THD measurements. Additionally the first five harmonic can be isolated, selected and monitored.
The noise is the difference between THD+N and THD plus the analyzer's noise. If the analyzer's noise can be determined then it can be subtracted out from the measurement.
Hello,
Please also consider that measuring noise must always be refereed to bandwidth of interest. Giving two results without knowing precise set-up for how they are done is useless.
It is much interesting to speak about noise spectral density (even if it can change much at different frequencies). I say that because for example you will not achieve same noise level at 48k/96k or 192k sampling rate...(That said also that wider the bandwidth will be and less good the THD+N level will be).
Please also consider that measuring noise must always be refereed to bandwidth of interest. Giving two results without knowing precise set-up for how they are done is useless.
It is much interesting to speak about noise spectral density (even if it can change much at different frequencies). I say that because for example you will not achieve same noise level at 48k/96k or 192k sampling rate...(That said also that wider the bandwidth will be and less good the THD+N level will be).
Noise in an oscillator is not white, of course, but it is still uncorrelated. I want to be able to measure the noise of an oscillator while it is running, as if it were a black box.
Cheers,
Bob
Understood. Then how about the difference method. This was separate from shutting the oscillator down. It would only measure the wide band noise.
Maybe some helpful osc noise measurement info can be found here: http://www.qsl.net/va3iul/Phase noise in Oscillators.pdf
But just low noise design in general will get you a low noise osc..... low noise device, low noise stable PS, low noise grounding etc etc.
THx-RNMarsh
But just low noise design in general will get you a low noise osc..... low noise device, low noise stable PS, low noise grounding etc etc.
THx-RNMarsh
Last edited:
This is phase.
What noise are we concerned with measuring. Phase, AM, wide band white or all?
Low noise design is indeed important. However, there are sometimes tradeoffs to be made. Let me give you an example. We have the case of a JFET as the agc control element in an oscillator. The nonlinearity of the JFET is one of the usual contributors to oscillator distortion. We can reduce this source of distortion by making the voltage across the JFET arbitrarily low by design. However, we then must amplify the signal that has been controlled-attenuated by the JFET up to a voltage level that is enough, when injected into the oscillator loop, to provide adequate agc authority (range).
The more we have to amplify that controlled signal up, the more noise we inject into the oscillator. So someone who greatly attenuates the voltage on the JFET to get very low distortion at the expense of fairly high oscillator noise is sort of "cheating".
Obviously, mixed in with all of this, we must know which contributors to oscillator noise are dominant. Ditto contributors to distortion. Only then can we make the best engineering tradeoff.
Cheers,
Bob
From what i have seen the analog oscillators have a pretty broad "skirt" with noise reducing as you get further from the fundamental. The output of a low distortion DAC (AK4490) however is more like a single line with some small spreading at the bottom (which could be the window function). I can post some FFT's next weekend when i return from some travel.
If I use a peak detect function for the FFT will that catch the noise issue in question?
If I use a peak detect function for the FFT will that catch the noise issue in question?
At some level I don't see your point. The noise and distortion of an amplifier are to the most part separable. The oscillator's purpose in this context is to exercise the non-linearities of the amplifier. Can you propose a test facilitated by an oscillator with exemplary noise sidebands? As I see it they just pass through, the harmonics usually displaying the same level as the fundamental, the THD being the same in any case. The noise can simply be measured with no input.
The most effective way to measure the noise of an oscillator design requires two identical units. Phase lock them and then feed a differential input. Signal will drop and noise will increase by sqrt(2).
Noise increase in the region of the osc freq.... it should get amplified by the Q gain over a narrow BW (high Q) centered on the osc freq.
THx-RNMarsh
THx-RNMarsh
Does it matter to a THD measurement? Are we building oscillators to measure their own noise, an end in itself?
If the equiv topology of the osc/generator is of a series resonance, then at the res freq, the impedance is R or minimum. But if an equiv parallel circuit osc/gen then at the res freq (Q peak) the impedance (R) is maximum. So depending on the type of circuit, noise could be min or max at res freq.
??
THx-RNMarsh
??
THx-RNMarsh