Well, HpW-Works supports up to 32M FFT using a 64Bit MS OS (8 GBytes system memory) and has included the required high performance FFT Windows functions! The calculation is internal done by 64 & 80 bit float.
This exactly the point! In the old days good sound card as Turtle Beach Tahiti went down to DC and at this time the Sound Blasters using Codec's where no value. For measurement the source or converter should deal in a higher league!
Currently I have to time to deal with such HW... This HW should also deal with input & output gain / attenuation / load (Pete already asked
).There are also some new ADC/DAC on the horizon as Arda Technologies - Home...
HpW
I'm still confused about what AP model you're talking and what your exact setup is--the top-notch models System One, System Two and SYS-27xx surely have an analog notch filter which you can route ahead of the ADC. If you're feeding the ADC with the input signal directly don't blame anyone else for unreliable results... An ordinary ADC is good down to -100 dB THD, a recent high-performance one perhaps to -110 dB.
Samuel
A-Prec Sys one. The generator has lower thd/harmonics than is shown by its own FFT. I had not heard about that before. It doesnt look like a notch is used.... but attribute the problem to poor adc/dac of the era.
I see the data from the newest AP and it looks like a notch is used. I didnt get that with the system One (dual domain model). So, now, I want to see an FFT that gives more accurate results without buying a newer A-P ($$$). I need a better ADC. I was dissapointed that newest ADC's can not be used straight into them and not be a lot better than -100.... So, now am looking for a notch filter to make/buy/steal to go with a couple of adc i recently acquired.
Meanwhile, I bought a ShibaSoku AD725D distortion analyzer. That with an excellent sine wave source has made the whole ADC issue mute. Almost. I still want to get as low Harmonic measurement as I can using an ADC with a notch filter to see harmonics beyond the 5th measured by ShibaSoku. I should be able to hook up, as Damian says, to the instrument with any ADC and get accurate data on harmonics but I havent seen that happen yet. ??
Yes, it looks like the -100 is close to the limit straight in, as you say. And, then drop 5-10 dB from the max input level. So we have a real 90dB range to work with for accurate measurements use... which makes me think John Curl is correct... we (he) can design analog which is still better than what digital can do.
Thx for your input to my saga on this.... waiting on a pcb for a twin-t filter now. -RNMarsh
Last edited:
Dual-Domain System One surely can route the notch filter ahead of the ADC. I've just run a measurement on the one we have, and I see the harmonics at/below -120 dB (1 kHz/+10 dBu).
Setup as follows:
* Analog Analyzer Function Reading set to "THD+N Ratio"
* Digital Analyzer Source set to "Anlr Rdg Ratio"
Samuel
Setup as follows:
* Analog Analyzer Function Reading set to "THD+N Ratio"
* Digital Analyzer Source set to "Anlr Rdg Ratio"
Samuel
24 bit resistor ladder DAC, interesting they don't mention factory calibration and/or any trimming.
I agree with Samuel I have never seen an AP without a notch on the fundamental available.
Here's a simulator for you, Infinisim - Results. Probably beyond the budget.
I have to stand by what i saw.... took it to another with newer AP model and compared.... they saw same thing. Either mine has an problem or that is the way it is made.
But, -120 isnt good enough anyway. Marginal. I'm looking for at least -130. -140 would be more useful.
BTW --just got a note - Victor hyper tuned a 1KHz oscillator that takes the H2 down to -150 ! That pretty much takes it out of the picture for my use !
Thx-RNMarsh
But, -120 isnt good enough anyway. Marginal. I'm looking for at least -130. -140 would be more useful.
BTW --just got a note - Victor hyper tuned a 1KHz oscillator that takes the H2 down to -150 ! That pretty much takes it out of the picture for my use !
Thx-RNMarsh
Last edited:
THD+N is not a good figure of merit when the distortion is so low. The noise in the measurement bandwidth need to be figured out before anything else. If the source noise (equivalent source resistance) is too high you won't get useful info with that number. The low voltage devices used today make this more of an issue. 140 dB below 1V in a 20 KHz (or 30 KHz using the traditional measurement) is a pretty small resistor. A high res FFT confuses this since the measurement bandwidth gets quite small and the noise in that band is very low. To get to -140 dBV in a 20 KHz band your source + input needs to be less than 30 Ohms assuming its noise free etc. A 50 Ohm generator won't get there. High voltage ADC's are really history. Most run on 5V or 3.3V so the input may be 2V max.
I think measuring the individual harmonics is the only meaningful way forward if this actually will accomplish anything. I think -120 dB is probably 40 to 60 dB beyond audibility in practice. However complex IM would be a different issue.
I think measuring the individual harmonics is the only meaningful way forward if this actually will accomplish anything. I think -120 dB is probably 40 to 60 dB beyond audibility in practice. However complex IM would be a different issue.
The service manual is still available and it describes the operation of the fundamental notching in detail. Quite clever, they quadrature detect the residual fundamental and tune out both I and Q components. Maybe we are talking about different things, there was no A/D in 1985 that even remotely could do -100dB.
EDIT - I see, THD + N, I agree a useless metric.
Last edited:
Of course you can configure it such that the ADC sees the direct input signal, and you will then see excess distortion. If you post a screenshot of the settings you're using I might have a chance to point you in the right direction (some settings aren't obvious), or, as Scott suggests, consult the manual.
My sincere apologies for repeating myself every 20 pages or so in this thread, but quoting direct analog distortion measurements at the -150 dB level is simply not credible at the current state of the art (unless someone gives a detailed, repeatable proof how he verified measurement integrity to this level, which is pretty unlikely).
As noted several pages back I've designed a fixed frequency passive notch filter PCB which is currently probably your best DIY bet. It should get you to -130 dB, and perhaps even to -140 dB. But it is cumbersome to use.
Samuel
Perhaps we should do more in the digital domain. Borrow some of the techniques used in the ShibaSoku AD725D. The frequency of the input signal can be adjusted digitally to fall in the center of the bin and absolutely periodic. From what I understand if this condition is met windowing is not needed. Noise reduction can be done before it hits the FFT and all this can be done without lose from the original signal.
A modern ShibaSoku built from today's technology.
I spotted a few small boxes designed for EMI RF shielding which come with BNC connector
at each end. The size is quite limited. Do you suppose your notch filter can squeezed onto a small very board? The boxes are expensive but I think well suited for this purpose.
I what to do at least a couple of these for a reference notch filter.
Sure. Can we get the noise down or just the distortion up with out adding too much to it???
You can always reduce the FFT bin width by taking longer FFT's but if you want 140dB below 10dBV that's -130dBV or 316nV and even a 64K FFT is around 1Hz BW at 96K sampling IIRC. Even the passive notch penalty should not hurt.