You don't get entirely empty bins unless the waveforms (or an integer number of them) exactly fit in the FFT size, though. Spectral leakage will get into otherwise empty bins even far away (assuming enough dynamic range is the gear, of course). Windowing, particularly high performance ones like Blackman-Harris or Hodie, will minimize the leakage, though, good enough in most cases. To completely avoid it, integer numbers of the waveforms should fit in the FFT size, then you can FFT without having to use any windows.
That also needs to have the recording sample rate locked to the generating sample rate, though, otherwise through 128k+ samples even slight rate differences will cause endpoint mismatches again.
That also needs to have the recording sample rate locked to the generating sample rate, though, otherwise through 128k+ samples even slight rate differences will cause endpoint mismatches again.
You bet, overlooked detail. At least one USB sound card I have has a faulty clock divider (1 kHz = 1.006 kHz) total disaster mixing it with another card.
Different sound cards often run at slightly different clock rates. Some internal (PCI) sound cards can be configured to run off the system clock or their own clock. If the system clock, or some other external clock source, is supported then that might be a fix.
Skin Effect T&M
FYI
Using a test and measurement for skin effect (caused by eddy currents) sent to me by Davada over here, I did a skin affect test to determine the freq it becomes a player --
I said 8Khz. Other say >100KHz et al. So here is what I got from this setup:
Using 1mil of copper thickness (placed between the xmt and recv coils) =
-1dB at 4.5KHz and -3dB at 12Khz.
Using a double sided pcb (FR4) of 1mil copper =
-1dB at 1KHz
-4.5dB at 4.5KHz
-6dB at 8.0KHz
-9dB at 12KHz.
Using two such pcb back to back = -9dB at 8KHz.
Aluminum was spot checked and so was steel plate (from a chassis cover). Obviously, the steel came in at very low freq and at high atten./loss.
THx-RNMarsh
FYI
Using a test and measurement for skin effect (caused by eddy currents) sent to me by Davada over here, I did a skin affect test to determine the freq it becomes a player --
I said 8Khz. Other say >100KHz et al. So here is what I got from this setup:
Using 1mil of copper thickness (placed between the xmt and recv coils) =
-1dB at 4.5KHz and -3dB at 12Khz.
Using a double sided pcb (FR4) of 1mil copper =
-1dB at 1KHz
-4.5dB at 4.5KHz
-6dB at 8.0KHz
-9dB at 12KHz.
Using two such pcb back to back = -9dB at 8KHz.
Aluminum was spot checked and so was steel plate (from a chassis cover). Obviously, the steel came in at very low freq and at high atten./loss.
THx-RNMarsh
Last edited:
In this case I consider .6% a flaw and broken. IMO when you design a product the nominal performance is computed exactly, that is you can buy any crystal you want and design a PLL to give, for instance, exactly 44,100 Hz sampling frequency. This should be where you start, yes if you want to lock to an external clock there might be some numerical limits to the exact ratio that you can accommodate. This product only generated its own clocks.
Okay. But in the early days of multi-track digital recording, people tried using multiple sound cards at once for recording. Over the course of a song, 2-3 minutes, they found that by the end of the song the tracks were out of time with each other and the recording was therefore unusable. I'm not sure exactly at what point the cards should be considered broken, but .6% does seem like a lot.
However, it doesn't seem to be all that much in terms of cents of pitch. From Wikipedia: "In one study, changes in tone quality reduced student musicians' ability to recognize, as out-of-tune, pitches that deviated from their appropriate values by ±12 cents." However, I think other people have found that about 1 cent pitch error is hearable by some people.
Anyway, if a sound card manufacturer figures that humans can't hear small pitch errors, they may consider good crystal accuracy as overkill compared to what is actually needed for sound card clock accuracy.
FYI
Using a test and measurement for skin effect (caused by eddy currents) sent to me by Davada over here, I did a skin affect test to determine the freq it becomes a player --
THx-RNMarsh
Is the test sample smaller than the coils?
Yes I understand I think we're talking about two different things. Two 10 MHz frequency references that have a slow wander from each other and eventually get un-synced is different from a .6% by design error.
Two plays of the same LP superimposed is a fun revelation.
About the same size... 95%. Some >100%
Only one was 65%.... first one listed.
Alum foil (100% of coil size).... -3dB at 12KHz
-RNM
Last edited:
FYI
Using a test and measurement for skin effect (caused by eddy currents) sent to me by Davada over here, I did a skin affect test to determine the freq it becomes a player --
I said 8Khz. Other say >100KHz et al. So here is what I got from this setup:
View attachment 566983
Using 1mil of copper thickness (placed between the xmt and recv coils) =
-1dB at 4.5KHz and -3dB at 12Khz.
Using a double sided pcb (FR4) of 1mil copper =
-1dB at 1KHz
-4.5dB at 4.5KHz
-6dB at 8.0KHz
-9dB at 12KHz.
Using two such pcb back to back = -9dB at 8KHz.
Aluminum was spot checked and so was steel plate (from a chassis cover). Obviously, the steel came in at very low freq and at high atten./loss.
THx-RNMarsh
What is it without a plate. You have to account for the inductance of the coils.
Even with nominally exact (is that a weird concept
) sample rates you can get out of step for FFT analysis purposes. For instance, 50ppm is a pretty common frequency tolerance spec on a soundcard crystal, but for a test running over 2 minutes, that's already an error of 6 milliseconds or a 180 degree phase shift even by 83Hz. This turned out to be a big issue when I was messing with DiffMaker, and one of the reasons why tracks tested with it have to be kept pretty short.You can't assume the record and play sample clocks are locked to each other on soundcards, either. Some of the USB-based soundcards use an internal crystal for one function but phase lock to the host's 1kHz USB clock for the other). Makes for a mess when people try to do synchronous measurements and tests involving averaging.
I normalize to zero ref level at each freq. Then insert metal between coils and read atten dB's.
THx-RNMarsh
Last edited:
Even with nominally exact (is that a weird concept
You can't assume the record and play sample clocks are locked to each other on soundcards, either. Some of the USB-based soundcards use an internal crystal for one function but phase lock to the host's 1kHz USB clock for the other). Makes for a mess when people try to do synchronous measurements and tests involving averaging.
I believe the AES spec is +/-10 PPM and better can be had for a price. With multiple cards you really need to lock them to a common clock. Pro cards support this.
I just finished measuring a handful of USB devices and the 1 KHz varied all over the place. The adaptive devices were the worst with 1 KHz as much as 6 Hz off but changing all the time. Averaged measurements were ugly and useless.The ASYNC devices were much better but still not perfect.
If the source clock is not locked to the ADC clock we are back to the spectral leakage problem. If the source and capture are not perfect what is the best approach? I use large captures and high order windows which seems to work for me. Am I fooling myself? When I have cross checked with a swept analyzer (TEK 7L5) I get comparable results.
FYI
Using a test and measurement for skin effect (caused by eddy currents) sent to me by Davada over here, I did a skin affect test to determine the freq it becomes a player --
I said 8Khz. Other say >100KHz et al. So here is what I got from this setup:
View attachment 566983
Using 1mil of copper thickness (placed between the xmt and recv coils) =
-1dB at 4.5KHz and -3dB at 12Khz.
Using a double sided pcb (FR4) of 1mil copper =
-1dB at 1KHz
-4.5dB at 4.5KHz
-6dB at 8.0KHz
-9dB at 12KHz.
Using two such pcb back to back = -9dB at 8KHz.
Aluminum was spot checked and so was steel plate (from a chassis cover). Obviously, the steel came in at very low freq and at high atten./loss.
THx-RNMarsh
I can't figure out how you are using that test set-up. Do you place the test materiel between the coils?
A setup like that is sometimes used to check for cracks in metal samples, in which case a crack would change the normal eddy current effects. If you google "Non-Destructive Techniques Based on Eddy Current Testing" it finds a document with that name. The setup is shown in Figure 24, on page 2552. It's not clear to me how one could interpret the results of that test in terms of audio cable performance. It would seem easier and more direct to measure some actual speaker cable for differences in AC and DC resistive losses at various frequencies. Maybe I am missing something.
Edit: Here is a link http://www.mdpi.com/1424-8220/11/3/2525
Also, this type of test setup is also sometimes referred to as the "through-transmission method"
Last edited:
What is the effect of clock wander when running long term (hours) measurement of noise signals ?.
Dan.
FYI
Using a test and measurement for skin effect (caused by eddy currents) sent to me by Davada over here, I did a skin affect test to determine the freq it becomes a player --
I said 8Khz. Other say >100KHz et al. So here is what I got from this setup:
View attachment 566983
Using 1mil of copper thickness (placed between the xmt and recv coils) =
-1dB at 4.5KHz and -3dB at 12Khz.
Using a double sided pcb (FR4) of 1mil copper =
-1dB at 1KHz
-4.5dB at 4.5KHz
-6dB at 8.0KHz
-9dB at 12KHz.
Using two such pcb back to back = -9dB at 8KHz.
Aluminum was spot checked and so was steel plate (from a chassis cover). Obviously, the steel came in at very low freq and at high atten./loss.
THx-RNMarsh
That is not a test of skin effect per se. It is a measure of eddy current based flux exclusion. If you go to my gallery, you will find measurements showing inductance of coils in air as well as against copper clad.
You are seeing the loss of coupling because the metals are fighting the rate of change of magnetic field. It is NOT consistent with the coil wire diameter, nor is it consistent with any interconnect wire used in audio.
It is consistent with a conductor of the diameter of your occluding metal.
The plots in my gallery show several things. First, the increase of resistance caused by proximity effect losses within the conductor; it increases without bound. Second, it shows the inductance falloff as the current centroid shifts due to proximity; it is bounded as it cannot leave the conductor.
I am not able to view the test setup from where I am, so look forward to seeing it. It may be better poised to view the non linear mixing of two signals when one modulates the permeability of a steel enclosure using the wall as isolation between noisy currents and low level signals.
That said, your setup is not exposing skin effect on wires, so you cannot claim that 8 kHz threshold, as you are not measuring skinning per se.
Nice test though.
John
That's huge, Huge! Can you tell us how you tested and determined such a gross flaw? Did you keep a record of the spread?
- Status
- Not open for further replies.