I do not see the analog side of the equation going away. The features needed by the analog side may be fewer, but the parameters must be as good as possible to achieve the best results and to move forward. No digital processing can fix randomly wrong data.
I believe if all non-output opamps in my soundcard were replaced with OPA1656s, the resolution threshold would improve substantially. The analog hardware does matter.
I believe if all non-output opamps in my soundcard were replaced with OPA1656s, the resolution threshold would improve substantially. The analog hardware does matter.
Just some thoughts here - forgive me if I’m being naive.
1. If you take a low noise oscillator and notch the fundamental out, you will have some harmonic content - down in tne noise if you’re lucky.
2. You them amplify the distortion harmonics (100-1000x) with a very linear amp (say AD797 or LM4562 level device maybe even configured as a composite amp to raise loop gain and further suppress distortion ).
3.You measure the distortion products and save them using a high performance A-D
4. Next measure the DUT and subtract the distortion products saved in step 3
5. You calibrate each measurement cycle like this.
I don’t know the math of ‘mixing’ the reference amp and the DUT distortion like this, but on the face of it it seems this hybrid approach (digital and analog) is the best way forward where you beverage the PPB distortion performance of modern opamps to look into the harmonics.
1. If you take a low noise oscillator and notch the fundamental out, you will have some harmonic content - down in tne noise if you’re lucky.
2. You them amplify the distortion harmonics (100-1000x) with a very linear amp (say AD797 or LM4562 level device maybe even configured as a composite amp to raise loop gain and further suppress distortion ).
3.You measure the distortion products and save them using a high performance A-D
4. Next measure the DUT and subtract the distortion products saved in step 3
5. You calibrate each measurement cycle like this.
I don’t know the math of ‘mixing’ the reference amp and the DUT distortion like this, but on the face of it it seems this hybrid approach (digital and analog) is the best way forward where you beverage the PPB distortion performance of modern opamps to look into the harmonics.
Honestly, I have no experience with analog sources, I am mostly a software guy. I understand the DAC output cannot be as clean as a good analog oscillator, in terms of noise and various artefacts. Probably not in terms of THD either.
But I do not understand how to use a freewheeling analog oscillator practically in ultra-high-resolution THD measurements with A/D sampling timed by an independent clock and trust the FFT results at the same time, without specifically making sure that signal at beginning and end of a single FFT round is not leaking to (or even has moved to) adjacent FFT bins and that every FFT round used in subsequent averaging has the fundamental in the same bin. Such check would not be difficult to program, even post-measurement, analyzing some file with recorded stream.
That is why I find of major importance to know in detail how proprietary analyzers actually work and what they do with the sampled data.
I fully agree with you. As I have posted here before, the AP 2722 has a 'windowing function' that isn't really a window function, and is called 'none, move to bin center'. The concept is clear: after acquiring the samples, analyze and resample as necessary to make the fundamental and harmonics fall in the center of the bin.
What we don't know (at least I don't know) is whether that is done at the very end of the acquision phase, or for each signal cycle, or what. That is important to know whether a source drift, or a source absolute frequency deviation, or both, are neutralised.
Jan
If leaking to adjacent bins occurs, the measured (or averaged) level of the fundamental should drop correspondingly.
Could a drop in the measured level of the fundamental for FFT 16k and 1M indicate the leak? For a nonleaking measurement the levels for different FFT lengths should be more or less identical, IMO.
Also levels of the bins adjacent to the fundamental should reveal the leak too, IIUC.
Could a drop in the measured level of the fundamental for FFT 16k and 1M indicate the leak? For a nonleaking measurement the levels for different FFT lengths should be more or less identical, IMO.
Also levels of the bins adjacent to the fundamental should reveal the leak too, IIUC.
This is a very good point, and I believe that nudging the frequency over such a small range (0.1% p-p) is very do-able at only a small hit to distortion performance. Changing the frequency of an SVO over such a small range can be done in a straightforward way with a very non-intrusive JFET circuit in the global feedback circuit of the SVO. It is especially helpful that the source of a JFET involved in this would be at the virtual ground potential of the SVO summing op amp.
Cheers,
Bob
These are good points. The kind of distortion performance we need of the oscillator is quite application-dependent, such as among THD testing, IM testing and frequency response testing. To first order, one might just be better off with different oscillators for these 3 different types of testing.
I am a very big fan of IM testing, especially 2-tone and 3-tone, largely because the distortion products of interest are usually in-band and because the THD of the multiple oscillators is not much of an issue as long as it is reasonable. IM testing also often does not require much flexibility in oscillator frequency selection, so cheap, fairly simple fixed-frequency oscillators that do better than 0.01% are usable. For example, 18.5 kHz and 19.5 kHz are pretty much all you need for CCIF IM. SMPTE IM, nominally at 60 Hz and 7000 Hz is also a good example. When I did my coherent IM analyzer, I did choose to have additional frequencies for the low-frequency tone, like 2 Hz, 5 Hz and 20 Hz, if I recall. The other nice thing about many IM tests is that you usually know the frequencies of the products you are looking for, so post-filtering of the products can greatly reduce ingress of hum and noise into the measurement.
Cheers,
Bob
Syncing a low distortion oscillator whose frequency can be controlled (nudged) is quite straightforward with a properly designed PLL. My question is, given that we are going to analyze the signal with an FFT device or instrument, where do we get the clock and its correspondingly divided-down frequency with which to lock the oscillator? For example, can we get access to the A/D clock and divide it in some appropriate way to get the, e.g., 1 kHz, clock to synchronize the oscillator to? If so, is that sampling clock easily tapped out of a soundcard or a QA401, for example?
Cheers,
Bob
The FFT has some length. Modern libraries offer any integer-value FFT length, that means e.g. any fixed integer Hz (viewed by the ADC clock) is always in the middle of the bins. I use 48kHz FFT for 48kHz samplerate in my project, to align integer-Hz frequencies.
Other projects such as REW allow adjusting the measurement frequency they generate for the DAC to fit the nearest bin of the chosen FFT length (being only powers of 2, IMO only for historical reasons, the java computing library used by REW allows any FFT length).
Other projects (presumably the AP does that) allow resampling the incoming stream to a new value to shift new frequency to a FFT bin center.
Diana does some resampling too.
I remember other members here taking about using sinc function interpolation between the bin results.
Other projects do not address the issue at all (perhaps Arta?).
Any stable frequency can be resampled to fit the existing bins (AP way). Any FFT length can be changed to fit the given fundamental frequency to the best-fitting bin. I calculate the best-ffiting FFT length in the range of 0.5x - 2x samplerate every distortion measurement cycle (250ms) for the non-integer mode (i.e. DAC and ADC have separate clocks, but crystal-stable frequency, no freewheeling analog oscillator).
Everything is possible with some software environment available. A proprietary solution is limited to what the vendor decided to implement.
IMO AP with its "fit bin to center" feature will accept any stable frequency, i.e. the oscillator could be PLLed to a signal from its DAC. I would assume the same be possible with QA. But IMO it makes sense to use directly the digitally-generated DAC signal then, with some conditioning if possible and needed.
Other projects such as REW allow adjusting the measurement frequency they generate for the DAC to fit the nearest bin of the chosen FFT length (being only powers of 2, IMO only for historical reasons, the java computing library used by REW allows any FFT length).
Other projects (presumably the AP does that) allow resampling the incoming stream to a new value to shift new frequency to a FFT bin center.
Diana does some resampling too.
I remember other members here taking about using sinc function interpolation between the bin results.
Other projects do not address the issue at all (perhaps Arta?).
Any stable frequency can be resampled to fit the existing bins (AP way). Any FFT length can be changed to fit the given fundamental frequency to the best-fitting bin. I calculate the best-ffiting FFT length in the range of 0.5x - 2x samplerate every distortion measurement cycle (250ms) for the non-integer mode (i.e. DAC and ADC have separate clocks, but crystal-stable frequency, no freewheeling analog oscillator).
Everything is possible with some software environment available. A proprietary solution is limited to what the vendor decided to implement.
IMO AP with its "fit bin to center" feature will accept any stable frequency, i.e. the oscillator could be PLLed to a signal from its DAC. I would assume the same be possible with QA. But IMO it makes sense to use directly the digitally-generated DAC signal then, with some conditioning if possible and needed.
Pavel, I should also mention that the AP has a digital generator but with higher distortion than the analog (or the external Vic in my case).
Someone mentioned low-passing the output of a D/A source, but would such a filter not phase shift the fundamental and harmonics, and thus make the synchronous FFT more complicated?
Jan
Someone mentioned low-passing the output of a D/A source, but would such a filter not phase shift the fundamental and harmonics, and thus make the synchronous FFT more complicated?
Jan
Jan- The frequency won't change with a low pass filter on the source and the phase relationship of the harmonics in the source do change but we are looking at the DUT not the source oscillator. If you are using a technique like Pavel's the correction would be worked out including the filter.
FWIW my best distortion analyzer (Rediometer CLT-1) is all frequency domain analog circuitry with a 3rd harmonic floor of -170 dB. The design dates from the late 1960's so no digital capabilities. it uses passive filters to operate. A multipole low pass filter on the output of the oscillator and power amp for the fundamental and a high pass filter and a bandpass filter for the harmonic. It has some capabilities that aren't needed but conceptually a system could be built using similar concepts and a few frequencies with a distortion floor that's really low. The inductors would be the challenge but they demonstrated its possible using ferrite cores.
I have a scan of the full manual if anyone is interested.
FWIW my best distortion analyzer (Rediometer CLT-1) is all frequency domain analog circuitry with a 3rd harmonic floor of -170 dB. The design dates from the late 1960's so no digital capabilities. it uses passive filters to operate. A multipole low pass filter on the output of the oscillator and power amp for the fundamental and a high pass filter and a bandpass filter for the harmonic. It has some capabilities that aren't needed but conceptually a system could be built using similar concepts and a few frequencies with a distortion floor that's really low. The inductors would be the challenge but they demonstrated its possible using ferrite cores.
I have a scan of the full manual if anyone is interested.
The digital generator could be used for the PLL, perhaps.
There is always a time delay between DAC and ADC. FFT yields complex amplitude (i.e. real amplitude and real phase) for the given block of samples being analyzed. Shift the buffer by one sample and all the FFT'd phases will rotate by the sample time. In distortion compensation the amplitude and phase of fundamental and harmonics are of equal importance. But for measuring DUT performance only the amplitudes are interesting, unless people want to study dynamic parameters of their chain (i.e. why their harmonics are not rotated against the fundamental at pi/2 multiples which is valid only for static (polynomial transfer) distortion).
There is always a time delay between DAC and ADC. FFT yields complex amplitude (i.e. real amplitude and real phase) for the given block of samples being analyzed. Shift the buffer by one sample and all the FFT'd phases will rotate by the sample time. In distortion compensation the amplitude and phase of fundamental and harmonics are of equal importance. But for measuring DUT performance only the amplitudes are interesting, unless people want to study dynamic parameters of their chain (i.e. why their harmonics are not rotated against the fundamental at pi/2 multiples which is valid only for static (polynomial transfer) distortion).
I'm on the Time-Nuts list (where much of the discussion is above my pay grade) where an ongoing topic is keeping things temperature stable, either by waiting several hours for crystal ovens and such to stabilize after powering up, or just running things 24/7.> That's sooo 19th century ;-)
If you happen to have 150 Million € and what to buy one of the latest EUV scanners from a certain Dutch company,
they have stabilisation time of more than 12 hours.
And that is State of the Art. There ain't anything better.
Sometimes you cannot change the law of physics.
Patrick
For something that's basically an R-C oscillator (!), I think it would be easier to make the frequency slightly tunable than to keep it from drifting over the time frame of large FFTs.
Thinking another moment, one way to change frequency is to intentionally change the temperature of a few slightly temperature-sensitive components using resistors for heat and Peltier modules for cooling.
Hello all,
Does anyone need a power supply for one (or more) of Victor's oscillators?
I have designed a linear power supply to mount inside the aluminum housing from eBay along with Victor's oscillator. I bought a 1kHz and a 10kHz oscillator from him along with 2 aluminum cases from China on eBay and will be powering either one at a time with a 20VAC wall wart supplying power to a small linear power supply installed in the back of each case. The power supply circuit board is the same size as Victor's board.
The PCB layout is done and I will be sending the design out to OshPark soon for fabrication. Just wondering if anyone else is also in need a power supply before I place the order. OshPark provides boards in multiples of three.
I can post the schematic and/or PCB layout if there is any interest.
Mike
Does anyone need a power supply for one (or more) of Victor's oscillators?
I have designed a linear power supply to mount inside the aluminum housing from eBay along with Victor's oscillator. I bought a 1kHz and a 10kHz oscillator from him along with 2 aluminum cases from China on eBay and will be powering either one at a time with a 20VAC wall wart supplying power to a small linear power supply installed in the back of each case. The power supply circuit board is the same size as Victor's board.
The PCB layout is done and I will be sending the design out to OshPark soon for fabrication. Just wondering if anyone else is also in need a power supply before I place the order. OshPark provides boards in multiples of three.
I can post the schematic and/or PCB layout if there is any interest.
Mike