It is probably easier and less prone to error if I just measure noise with a sensitive analog ac voltmeter or a spectrum analyzer.
With QA401, first - here is the noise floor of the monitor output when no input signal processing. Ref 1v = 0dbv
And here is monitor noise but with processing (analyzer mode) turned on.... no signal from DUT;
Both just wide band random noise.
THx-RNMarsh
With QA401, first - here is the noise floor of the monitor output when no input signal processing. Ref 1v = 0dbv
And here is monitor noise but with processing (analyzer mode) turned on.... no signal from DUT;
Both just wide band random noise.
THx-RNMarsh
Last edited:
If I understand you correctly, divide the 96kHz sample rate by 96 and phase lock to that.
Where do we find such tools that give us complete access to the complex FFT? Other than writing our own code.
Windows accepts any integer sample rate between the popular audio sample rates. We don't have to be stuck with software that only handles sound card sample rates.
Octave, Python. In fact any data acquisition card with a trigger input like a scope could be used to average as much as you want. Just let enough fundamental leak through your TT to get a good trigger and average as long as you want just like averaging a noisy signal on a digital scope. Averaging in the time domain has the same effect as the vector averaging in the frequency domain.
It's not the sample rate that matters but the FFT length making them the same puts integer frequencies at each bin and makes the bins 1Hz wide so it reduces cals/confusion. A sound card in full duplex should work fine as long as the same crystal drives the DAC and A/D. You could use ARTA and export to Python or Octave, they both have arbitrary length FFT's built in. I put this in my digital RIAA article, there is nothing magic about power of two FFT's and all common sampling frequencies are factorable down to small prime numbers so the FFT is about as fast as it can get. If you take a large 7 digit prime number and try an FFT of that length you will suddenly see the FPU brought to its knees.
Last edited:
The only differences in noise level I have seen comes from how the wide band random noise is measured.... average, rms, peak/pk to pk detection and processed etc. You can get small to significant difference in noise levels resulting. My HP Spectrum analyzer shows 8-10dB greater noise than the 725/401 combo. But both show same amplitude of applied signal.
THx-RNMarsh
THx-RNMarsh
Last edited:
Scott, Do you still have this available available?
This is what I have from AP. Am I on the same page as everyone else?
FFT Window Scaling Factors:
Blackman Harris 3-term 1.73
Blackman Harris 4-term 2.00435
Equiripple 2.63191
Flat-top 3.82211
Gaussian 2.21535
Hanning 1.5
Hamming 1.36283
Rectangular (No Window) 1
Rife-Vincent 4-term 2.31
Rife-Vincent 5-term 2.62653
There never was a Mr. Hanning. The guy's name was
Julius von Hann, Austrian meteorologist.
< https://en.wikipedia.org/wiki/Julius_von_Hann >
Julius von Hann, Austrian meteorologist.
< https://en.wikipedia.org/wiki/Julius_von_Hann >
https://holometer.fnal.gov/GH_FFT.pdf
Spectrum and spectral density estimation by the Discrete Fourier
transform (DFT), including a comprehensive list of window
functions and some new flat-top windows.
A fine tutorial on what we've been discussing. Just so the point is not lost, without some kind of synchronization of sampling to input, windowing and power averaging are used because a fixed tone not exactly at a FFT bin center would rotate in phase so simple vector averaging would not work. This is why Audacity, ARTA, etc. use the FFT magnitude (power) so noise averages to it's mean power and not to zero, as a random +- value would when averaged as a vector. By vector I mean the one formed by the real and imaginary part of the FFT (same as talking about magnitude and phase of the voltage).
The whole point of these flat-top windows is to give accurate amplitudes for tones not exactly in a bin. Some of the other windows have rounded tops in the frequency domain and you can still get what is called "scalloping" loss usually small but an important detail.
Last edited:
You need a more better 1kHz for the comparison, or you need to measure David's generator on more better equipment.
Do you know, how may be the residual of your equipment now?
Victor.
There is an easy test for this. The distortion on the low pass output of the gen is 6dB lower than on the band pass output. The outputs correspond to the slopes of the SVF. LP, BP and HP outputs. 6dB slope for the HP and BP and 12dB for the LP.
I can measure this 6dB different in distortion between the two outputs which means the multiplier distortion is not the dominate source of the oscillator distortion. If you can't measure the difference then what are seeing is the analyzer's residual.
I can measure this 6dB different in distortion between the two outputs which means the multiplier distortion is not the dominate source of the oscillator distortion. If you can't measure the difference then what are seeing is the analyzer's residual.
Last edited:
You need a more better 1kHz for the comparison, or you need to measure David's generator on more better equipment.
Do you know, how may be the residual of your equipment now?
Victor.
Well, about 3 years ago, on another forum, I said I could measure to -160db below 1v. At that time i was using an HP spectrum analyzer on the monitor output. With the QA400/01 on the monitor output, the results are the same....maybe a little better...... -165-170. Limited by the noise floor at the monitor output when on -100dB FS range.
THx-RNMarsh
Last edited:
But did you ever see any oscillator, which has no harmonics down to these low -165dB... -170dB levels, when it measured with your equipment?
Victor.
I have only measured and improved a few ultra-low distortion gen/osc. None-- including AP's gen in the 2722 - is so low I can not see it.
But you can also see the harmonics from its own gen/osc with the analyzer in the AP 2722. But, the ShibaSoku is more accurate than the AP on the 2H harmonic.
I cannot easily use the monitor on the AP2722 or the Panasonic VP 7722 because it isnt calibrated --- but it could be done with extra hardware as I did on the ShibaSoku 725D for an accuracy check. So, for practical purposes, the 725D can be utilized with FFT on its calibrated scale from the monitor, which allows us to easily see further down for harmonics.
THx-RNMarsh
Last edited:
I only trust Samuel G. to verify these things, and yes most emphatically not myself either. He understands the details, I'm not about to invest the time and equipment.
BTW what is the answer -141.32, -145, -165. Elimination of confounders (fortuitous cancellations, etc.) requires a repeatable answer from everyone.
Last edited:
In my opinion, until we can't get stable and repeatable real results down to expected levels, we can't trust these results. This way is step by step improvements in oscillators and in the measurement equipment. If we wish to know, that the harmonics what we see is really from the DUT, previously we must see more better and repeatable results from the another one.
Every buffer, every filter in the equipment is a potentially distortions source. In my opinion, simplest and more trustful way is to use passive twin T at the input of the measurement equipment. Then we need extremely good parts only in this simple device - four identical resistors and four capacitors. I understand, that the active twin T is more effective, but an opamp in this filter is the part with unknown (for me) effect to the residual distortions.
Victor.
Last edited: