I have the need to measure sonics differences caused by different audio cables.
I reason that one method would be to record a (any) particular musical passage to hard drive via a high quality A/D convertor - I have access to a couple of high quality studio convertors.
I intend to record the same musical passage from the same source several times as wave files using only different audio interconnect leads.
I would like to then compare these files digitally, and then analyse any differences for changes in FR, phase, amplitude, spectrum etc.
Anybody have experience or knowlege of any of this, or know of suitable software to perform these tasks.
Thanks, Eric.
I reason that one method would be to record a (any) particular musical passage to hard drive via a high quality A/D convertor - I have access to a couple of high quality studio convertors.
I intend to record the same musical passage from the same source several times as wave files using only different audio interconnect leads.
I would like to then compare these files digitally, and then analyse any differences for changes in FR, phase, amplitude, spectrum etc.
Anybody have experience or knowlege of any of this, or know of suitable software to perform these tasks.
Thanks, Eric.
Keith Howard wrote a couple of articles on exactly this in HFN recently. He exploited a sound card that could be sync'd to an S/PDIF signal so that the samples would always be in the same place. IIRC he had some problems getting good correlations between recordings when there was no difference in the system.
See http://www.audiosignal.co.uk/freeware.html for some supporting software.
Paul
See http://www.audiosignal.co.uk/freeware.html for some supporting software.
Paul
the problem I see with this approach is that it requires precise timing on both recordings, and you can get timing differences from a lot of things, like mic placement, internal clocks, room sonic properties for example.
if you reconstruct the original waveform from the 44.1khz samplings, you create artifacts based on the algorithm you used, and the "articificial" differences in the original recordings (due to timing for example) would persist.
the alternative is to measure "characteristics" of the two waveforms. there, the law of large numbers kicks in, but you lose some resolution because of that.
You can do for example some time series analysis (SAS can do that), or FFT on the waveforms. Time series analysis is however sensitive to each data point so you need to make sure that both series line up exactly - hard to do.
there are lots of software packages that can do FFT and it is far more robust.
if you reconstruct the original waveform from the 44.1khz samplings, you create artifacts based on the algorithm you used, and the "articificial" differences in the original recordings (due to timing for example) would persist.
the alternative is to measure "characteristics" of the two waveforms. there, the law of large numbers kicks in, but you lose some resolution because of that.
You can do for example some time series analysis (SAS can do that), or FFT on the waveforms. Time series analysis is however sensitive to each data point so you need to make sure that both series line up exactly - hard to do.
there are lots of software packages that can do FFT and it is far more robust.
mrfeedback said:
I did exactly what you are describing using Wavelab from Steinberg. I did a test of a power cord. You can see it
Here
The trick is to trim all recordings to start and end at exactly the same sample. You have to zoom in in time and amplitude and be really careful about it. An offset of a single sample will give you bad results.
Again:
http://www.goldwave.com/
The demo version of Goldwave is fully functional and has lots of analysing features.
Also have a look at Sample Champion:
http://www.purebits.com/
Its RTA has lots of correlation analysing features. But it is not free.
Cheers
http://www.goldwave.com/
The demo version of Goldwave is fully functional and has lots of analysing features.
Also have a look at Sample Champion:
http://www.purebits.com/
Its RTA has lots of correlation analysing features. But it is not free.
Cheers
The setup used by Keith Howard uses the digital output of the CDP to clock the ADC, so the samples for each test pretty much line up. This allows you to test with a standard hifi setup.
He appends a level and synchronisation header onto the test material so that compensation for gain drift etc can be automated.
It's worth digging out his articles, it was far from trivial to get two measurements of the same setups showing no change. And, IIRC, when his CDP was powered off/on it would generate different output for the same CD, which is an unexpected artifact of its internal processing.
Paul
He appends a level and synchronisation header onto the test material so that compensation for gain drift etc can be automated.
It's worth digging out his articles, it was far from trivial to get two measurements of the same setups showing no change. And, IIRC, when his CDP was powered off/on it would generate different output for the same CD, which is an unexpected artifact of its internal processing.
Paul
Re: Re: Wave File Analysis Software....
I went through it and have to confess that I may not understnd every single thing of it.
it seems to me that other than the "difference" file, all others are pretty much identical between the two power cords. ie. your listening impression isn't supported by data.
Now, the difference file: a lot of things can cause the "difference" to be there: maybe the alignment isn't exactly correct, maybe the gain in two runs not exactly the same? etc.
One way to rule it out, and you can do this without the MPC cord, is to use the same method on the same cord multiple times. If you see "difference" from one run to the next, you know for sure that the methodology is flawed.
Right now, you cannot tell conclusively if the "difference" is caused by the cables or something else.
Thunau said:
I went through it and have to confess that I may not understnd every single thing of it.
it seems to me that other than the "difference" file, all others are pretty much identical between the two power cords. ie. your listening impression isn't supported by data.
Now, the difference file: a lot of things can cause the "difference" to be there: maybe the alignment isn't exactly correct, maybe the gain in two runs not exactly the same? etc.
One way to rule it out, and you can do this without the MPC cord, is to use the same method on the same cord multiple times. If you see "difference" from one run to the next, you know for sure that the methodology is flawed.
Right now, you cannot tell conclusively if the "difference" is caused by the cables or something else.
Re: Re: Re: Wave File Analysis Software....
You are right, you do not understand every single part of the procedure.
The gains are fixed (no way of changing them without soldering), the master clock is the same in all captures.
I did take extreme care in trimming the samples, it's something I used to do for living. The starting point and the absolute number of samples is constant. I did a control recording (MPC twice) and that difference resuled in very low level noise- no musical information). The power cord does make audible difference. That is reflected in the MP3 file on the "Conclusions" page. Just listen to it.
millwood said:
You are right, you do not understand every single part of the procedure.
The gains are fixed (no way of changing them without soldering), the master clock is the same in all captures.
I did take extreme care in trimming the samples, it's something I used to do for living. The starting point and the absolute number of samples is constant. I did a control recording (MPC twice) and that difference resuled in very low level noise- no musical information). The power cord does make audible difference. That is reflected in the MP3 file on the "Conclusions" page. Just listen to it.
Correlating non-synchronised samples
it would be possible to 'line up' two samples by using correlation to determine the parametric-time difference. This can be done efficiently using a FFT (search for convolution, which is closely related to correlation). The resulting data *should* be sharply peaked, the peak indicating the parametric-time difference. This could then be used to time-offset one of the samples being compared.
Dave
it would be possible to 'line up' two samples by using correlation to determine the parametric-time difference. This can be done efficiently using a FFT (search for convolution, which is closely related to correlation). The resulting data *should* be sharply peaked, the peak indicating the parametric-time difference. This could then be used to time-offset one of the samples being compared.
Dave
there are two elements to the timing thing: when does sampling start? and how consistent is the system sampling?
On the gain front, similarly two issues: noise introduced by digitization and slight variations of gain from one run to the next.
then you will have to deal with potential variations in your procedures / methodology. running the test one time or a limited few times makes your "conclusion" questionable at best.
I am sure others will add a lot more to this, and all of which will be helpful in getting us to the correct answer, whatever that may be.
This isn't meant to say that your experiment sucked because it did not. It is a starting point in the right direction. However, it is not sufficient to conclude based on it.
I hope you will take this in a constructive manner.
hopefully, we all will realize how difficult it is to prove anything scientifically.
On the gain front, similarly two issues: noise introduced by digitization and slight variations of gain from one run to the next.
then you will have to deal with potential variations in your procedures / methodology. running the test one time or a limited few times makes your "conclusion" questionable at best.
I am sure others will add a lot more to this, and all of which will be helpful in getting us to the correct answer, whatever that may be.
This isn't meant to say that your experiment sucked because it did not. It is a starting point in the right direction. However, it is not sufficient to conclude based on it.
I hope you will take this in a constructive manner.
hopefully, we all will realize how difficult it is to prove anything scientifically.
Re: Re: Re: Re: Re: Wave File Analysis Software....
Sorry, it came out harsher then it was meant.
When your digitization noise or jitter levels consistently change with the power cord isn't it proof that the power cord changes the sound?
And like I said, the analog gain is the same (fixed resistors in the buffer stage). Even then, if you end up consistently with different levels by using a different power cord maybe it is due to the power cord?
I dont think that gain should be forcefuly matched from one power cord to another.
The power cord should be the only variable.
millwood said:
Sorry, it came out harsher then it was meant.
millwood said:
When your digitization noise or jitter levels consistently change with the power cord isn't it proof that the power cord changes the sound?
And like I said, the analog gain is the same (fixed resistors in the buffer stage). Even then, if you end up consistently with different levels by using a different power cord maybe it is due to the power cord?
I dont think that gain should be forcefuly matched from one power cord to another.
The power cord should be the only variable.
Re: Re: Re: Re: Re: Re: Wave File Analysis Software....
the gain may not be the same. for example, temperature may have changed between test so rather than a gain of 10x, could it be 10.00001x? will that show up in your tests?
I think we need to go beyand the current level of sophistication if we want to get to the kind of resolution that seems necessary to see the impact cables, especially power cords, may have on sound.
one way to partially get around that is to use some kind of "indexed" or "differentiated" data rather than the raw data for the test. for example, normalize all data points (to their mean?) and then take a differential. That way, you got rid of any linear effects on the data.
for example. The following two series of data, S1 and S2, contain exactly the same information: S2=5+3*S1
S1 S2 S3=S1-S2 S4=S1+S2
1 8 -7 9
2 11 -9 13
6 23 -17 29
98 299 -201 397
3 14 -11 17
5 20 -15 25
6 23 -17 29
7 26 -19 33
8 29 -21 37
so you will get a lot of variance in S3 or S4, leading you to believe that S1 and S2 aren't the same - a false conclusion.
sure, the real data series may not different as much as the ones illustrated here but the point is the same.
In stats, there are grouping analysis, and convolution coefficient is another approach to see, statistically, just how far apart two seemingly different series are.
Thunau said:
the gain may not be the same. for example, temperature may have changed between test so rather than a gain of 10x, could it be 10.00001x? will that show up in your tests?
I think we need to go beyand the current level of sophistication if we want to get to the kind of resolution that seems necessary to see the impact cables, especially power cords, may have on sound.
one way to partially get around that is to use some kind of "indexed" or "differentiated" data rather than the raw data for the test. for example, normalize all data points (to their mean?) and then take a differential. That way, you got rid of any linear effects on the data.
for example. The following two series of data, S1 and S2, contain exactly the same information: S2=5+3*S1
S1 S2 S3=S1-S2 S4=S1+S2
1 8 -7 9
2 11 -9 13
6 23 -17 29
98 299 -201 397
3 14 -11 17
5 20 -15 25
6 23 -17 29
7 26 -19 33
8 29 -21 37
so you will get a lot of variance in S3 or S4, leading you to believe that S1 and S2 aren't the same - a false conclusion.
sure, the real data series may not different as much as the ones illustrated here but the point is the same.
In stats, there are grouping analysis, and convolution coefficient is another approach to see, statistically, just how far apart two seemingly different series are.
I don't know much about statistics. I took some back in high school but that has been about 20 years ago. So I will not argue here. But, in your example S1 and S2 are different. S2=5+3*S1. Their change is linear but they are not the same.
If my power cord did something like this to the output voltage of my CD player everybody would agree that it makes an audible difference. And that is exactly what I proved for myself by doing and documenting my test. I gathered data, changed one variable (the power cord) and gathered more data, then compared the two sets of data. The difference was much bigger than two sets of data gathered without changing the variable.
It was more of a reassurance that what I hear is not just self suggestion. IOW when I change the power cord I get different audio on the outputs of my converter. I didn't analyse the data statistically- I don't have the tools. But, I did spectrum analysis on the difference file and control difference file. They were dramatically different.
I did admit in my conclusions that the test was not 100% scientific.
The one lesson I learned from this whole little project is to not dismiss people's claims that power cables make an audible difference. I know of at least two such cords that can make a difference.
I could test more gear in more controlled environment but I see no point. I leave that to "real" scientists. My own results will always suspect to others.
If my power cord did something like this to the output voltage of my CD player everybody would agree that it makes an audible difference. And that is exactly what I proved for myself by doing and documenting my test. I gathered data, changed one variable (the power cord) and gathered more data, then compared the two sets of data. The difference was much bigger than two sets of data gathered without changing the variable.
It was more of a reassurance that what I hear is not just self suggestion. IOW when I change the power cord I get different audio on the outputs of my converter. I didn't analyse the data statistically- I don't have the tools. But, I did spectrum analysis on the difference file and control difference file. They were dramatically different.
I did admit in my conclusions that the test was not 100% scientific.
The one lesson I learned from this whole little project is to not dismiss people's claims that power cables make an audible difference. I know of at least two such cords that can make a difference.
I could test more gear in more controlled environment but I see no point. I leave that to "real" scientists. My own results will always suspect to others.
Hello Jan,
I say good on you for going to this kind of effort to categorise the changes that you hear.
Even if the methodolgy is not perfect, you are it seems showing differences.
I too have noted very clear differences attributable to power cords and signal leads.
Thanks to everybody else for the links and info.
Eric.
I say good on you for going to this kind of effort to categorise the changes that you hear.
Even if the methodolgy is not perfect, you are it seems showing differences.
I too have noted very clear differences attributable to power cords and signal leads.
Thanks to everybody else for the links and info.
Eric.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.