• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). There are daily reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Measured differences between interconnect cables

GXAlan

Major Contributor
Forum Donor
Joined
Jan 15, 2020
Messages
1,370
Likes
2,170
Presented purely for interest.

TLDR:
1) Differences in cables exist and can be measured at the audio level and achieve statistically significant differences
2) The differences are vanishingly small, so it doesn't make sense when listening to audio
3) The differences may make a difference when measuring gear
4) These tests will be repeated when my E1DA Cosmos Grade-A unit arrives later this week.

I'm going to tell the whole story and the process I took.

Background, Korg DS-DAC-10R in loopback mode. This isn't a high performance DAC/ADC but it does have a clean power supply, easy to access RCA jacks and it's pretty consistent. It's consistently noisy and consistently distorted from run to run. My main goal was to run through a few different software options in preparation of my incoming E1DA Cosmos ADC.

Cables on test:
1) Silver plated copper, flat ribbon cable, unshielded. I thought it sounded pretty good and then I tried it where I had RF interference and ditched it. I actually wanted to send it to Amir a few years back, but cable testing was a low priority. This is the cable shown here:
and here is proof I thought they sounded different.
1663724054018.png


2) Monster Cable Ultra Series 800 "THX" Certified
Got these as a freebie with something I bought.

3) Straight Wire Virtuoso 3 (JBL Synthesis branded)
Got a bunch of these and I also thought it sounded noticeably better than I bought more. These came from eBay from a pulled system.

First Experiment
RMAA loopback.
I ran the cables A->B->C->C->B->A sequentially and got this:
1663724579880.png


No real difference but it was interesting that the noise and dynamic range of the silver cables were very different from the Straight Wires. If you just throw them as aggregate as unpaired student's t-test, it's p < 0.0001
1663724771833.png


Next I ran multitone tests, no real difference. I felt as if there were differences in the subjective SHAPE of the noise in the bass area
1663724827423.png

1663724856275.png

1663724878801.png



This is where it gets very interesting
1663724936711.png


1663724954661.png


1663725004749.png



1. I don't know why the Straightwire delivered an extra dB during one run. I didn't change anything.
2. Notice the humps at ~15 kHz and 20kHz.

I then decided to run a test tone at exactly 15kHz
1663725122825.png


1663725148823.png


1663725166549.png



These tests are run serially. That is, I ran the full set of 50 Hz test tones. Then ran the full set of 15 kHz test tones. (I changed the cables 6 times not 3 times).

With these sets of tests, it was pretty clear that the left and right cables were different with the straight wire. We're still looking at a difference that is vanishingly small, but you do see differences.


Reversing the cables shows that the humps follow the cable.

1663725277568.png



Since I had multiple StraightWire cables, I found some replacements and measured them again. Boom. Really clean results. This was about 1 hour later.
1663725354749.png


Immediately after getting these flat results with the newer cable, I went back to the broken cable 1 hour later after the original set of tests to ensure that it wasn't random noise in the system somewhere

1663725419531.png


If you compare the 15 kHz results between the Monster Cable and the working Straight Wire cables:
1663725649528.png
 

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
16,444
Likes
28,333
I don't think using the T-test like this is appropriate. Firstly too few samples. Yeah, the math works, but too few samples. Secondly this just isn't the use for such a thing.

The one Straightwire having that noise hump is interesting. Would be nice to figure out why? Can you remove the ends to see if a connection is poorly done or nearly touching the shield etc. ?
 
OP
G

GXAlan

Major Contributor
Forum Donor
Joined
Jan 15, 2020
Messages
1,370
Likes
2,170
I don't think using the T-test like this is appropriate. Firstly too few samples. Yeah, the math works, but too few samples. Secondly this just isn't the use for such a thing.

The one Straightwire having that noise hump is interesting. Would be nice to figure out why? Can you remove the ends to see if a connection is poorly done or nearly touching the shield etc. ?

Students t-test just requires the distributions to be normal, which we would expect and going with a two tailed comparison for continuous datasets is fine. The reason you get away with few samples is that the standard deviations are very small relative to the effect size.

A very nice way to look at this is the Vanderbilt University Power Analysis tool where you get to figure out how big of a simple you need to reject the null hypothesis at your desired p-value.

If you look at the L-08m review, I identify the humps as correlating to the use of ceiling LED lights. What’s interesting is that the cables are all shielded, and the lights were all on for this set of tests and there were no dimmers involved. It was a different room from the room I did the L-08m measurements though.

It may be some hole in the shield from old cables affecting the results
 

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
16,444
Likes
28,333
My old university, but I haven't used their Power analysis tools. I still don't see how this fits for such a statistical treatment in this kind of testing. The math works, but it isn't fit for this use.
 
OP
G

GXAlan

Major Contributor
Forum Donor
Joined
Jan 15, 2020
Messages
1,370
Likes
2,170
My old university, but I haven't used their Power analysis tools. I still don't see how this fits for such a statistical treatment in this kind of testing.
You are comparing two means and want to reject the null hypothesis that the two means are identical.

1) assumption is that the distribution follows a normal curve. Since we are looking at random noise, it’s pretty fair to assume that it is a normal distribution. This is where a PhD statistician can spend more time to prove a measurement is normally distributed.

2) are the tests paired? No. Each measurement is independent.

3) is it a two- or one-tailed test? Is there only one way the comparison could go? Or could one population have a higher mean or a lower mean? Two-tailed.

4) is the data categorical or continuous? Continuous.

5) then Student’s t-test is applicable.

I have shown that the two groups of measurements are statistically different at a confidence level of p=0.0001. There is a 1/100th of a percent that I detected a difference between the two cables when no difference actually exists.

But statistically significant is not clinically significant.

If I were to repeat this test another 99 times, I would expect 95 of those tests to show that the difference between the two cables is somewhere between 0.567 to 0.733 dB.

Here is the free program I mentioned. Very convenient and to the point.
 

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
39,499
Likes
179,991
Location
Seattle Area
5) then Student’s t-test is applicable.
As a practical matter though, it is not necessary. You can just provide standard deviation and mean to give confidence in the results. We use statistical analysis when the data itself is so random looking that we can't make inferences from it. With these measurements, that is not the case.
 

kongwee

Addicted to Fun and Learning
Joined
Jan 22, 2022
Messages
792
Likes
200
Nobody is doing a full setup and measured at the speaker side for the real SPL. Assuming that is no difference.
 
OP
G

GXAlan

Major Contributor
Forum Donor
Joined
Jan 15, 2020
Messages
1,370
Likes
2,170
Agreed.

100% cables are a waste of money for audiophile tweaks. :). This is a clear example of something measurable but not actually important.
 

dorakeg

Active Member
Joined
Jul 20, 2022
Messages
150
Likes
86
Nobody is doing a full setup and measured at the speaker side for the real SPL. Assuming that is no difference.

Actually this report from QED provides very good insight on cables. ITs very old but I would say still pretty good read.

 

kongwee

Addicted to Fun and Learning
Joined
Jan 22, 2022
Messages
792
Likes
200
Actually this report from QED provides very good insight on cables. ITs very old but I would say still pretty good read.

I really been through lots of these stuff. In the end, nobody plot the frequency response direct from full speaker setup.
 

MAB

Addicted to Fun and Learning
Forum Donor
Joined
Nov 15, 2021
Messages
703
Likes
1,343
Location
Portland, OR, USA
I really been through lots of these stuff. In the end, nobody plot the frequency response direct from full speaker setup.
Yeah, we all been through lots of these things. In the end, someone always says something like "but you didn't plot the frequency response direct from full speaker setup"... as if that nullifies the result. I hope you realize that is a totally different experiment than what the OP did, with different test method and different resolution.
I also hope you realize that the experiment you ask for actually has been done many times. For instance the impact of binding posts:
I post this one because the investigator actually constructed an experiment that by design was to measure specific parameters at moderately high resolution, then measure the response of a speaker. And one of the legs of the experiment was a hammer test using steel nails as binding posts (pun intended!;)) And did show very slight measured reduction in SPL at high frequencies, consistent with our understanding of how and circuits and speakers work. Details are in the link, don't want to distract from this thread though...
OP here is exploring a totally different realm:
2) The differences are vanishingly small, so it doesn't make sense when listening to audio
So, if you're working on vanishingly small, and work had already been done on really small, sorta small, small, and not so small, don't you think it's a bit absurd to raise the "but you didn't prove audibility" flag???
And, if I understand the first post, OP gonna show us the results again after some really nice new test hardware arrives!:)
So @kongwee , we are really interested here in how these measurements progress. I think the audibility argument is a bit misdirected!
 
Last edited:
OP
G

GXAlan

Major Contributor
Forum Donor
Joined
Jan 15, 2020
Messages
1,370
Likes
2,170
And, if I understand the first post, OP gonna show us the results again after some really nice new test hardware arrives!:)
So @kongwee , we are really interested here in how these measurements progress. I think the audibility argument is a bit misdirected!

What is interesting is that I got my E1DA Cosmos and identified that hump as correlating to the presence of LED lights or not.

Believe it or not, the silver cable literally broke with one of the strands de-laminating from the ribbon portion so I had to discard it.

In this first test, what I have shown is differences in shielding presumably. The broken Straight Wire cable may have a small enough hole in the shielding and when we're looking at numbers, it can be enough of a difference.

I don't think I need additional cable testing since there is other equipment to prioritize right now.
 

MAB

Addicted to Fun and Learning
Forum Donor
Joined
Nov 15, 2021
Messages
703
Likes
1,343
Location
Portland, OR, USA
What is interesting is that I got my E1DA Cosmos and identified that hump as correlating to the presence of LED lights or not.
That is interesting. What on earth is going on with the LED???
I don't think I need additional cable testing since there is other equipment to prioritize right now.
OK, yeah. Good to move on rather than dwell on cables.
 

DonH56

Master Contributor
Technical Expert
Forum Donor
Joined
Mar 15, 2016
Messages
6,716
Likes
13,225
Location
Monument, CO
LED lights include a switch-mode power supply and are fairly notorious for radiating noise (EMI/RFI), especially the cheaper ones.

Modern instant/fast-start fluorescent lights have a ballast that runs at about 60 kHz, another common noise source.
 
OP
G

GXAlan

Major Contributor
Forum Donor
Joined
Jan 15, 2020
Messages
1,370
Likes
2,170
LED lights include a switch-mode power supply and are fairly notorious for radiating noise (EMI/RFI), especially the cheaper ones.

Modern instant/fast-start fluorescent lights have a ballast that runs at about 60 kHz, another common noise source.

Yeah, these were contractor grade low end stuff. I haven’t had a chance to switch the ones in this test room with the nicer Halo recessed lights. Would be neat to compare :)
 

kongwee

Addicted to Fun and Learning
Joined
Jan 22, 2022
Messages
792
Likes
200
Yeah, we all been through lots of these things. In the end, someone always says something like "but you didn't plot the frequency response direct from full speaker setup"... as if that nullifies the result. I hope you realize that is a totally different experiment than what the OP did, with different test method and different resolution.
I also hope you realize that the experiment you ask for actually has been done many times. For instance the impact of binding posts:
I post this one because the investigator actually constructed an experiment that by design was to measure specific parameters at moderately high resolution, then measure the response of a speaker. And one of the legs of the experiment was a hammer test using steel nails as binding posts (pun intended!;)) And did show very slight measured reduction in SPL at high frequencies, consistent with our understanding of how and circuits and speakers work. Details are in the link, don't want to distract from this thread though...
OP here is exploring a totally different realm:

So, if you're working on vanishingly small, and work had already been done on really small, sorta small, small, and not so small, don't you think it's a bit absurd to raise the "but you didn't prove audibility" flag???
And, if I understand the first post, OP gonna show us the results again after some really nice new test hardware arrives!:)
So @kongwee , we are really interested here in how these measurements progress. I think the audibility argument is a bit misdirected!
Another link with partial setup testing the usual voltage dB measure. Nobody I said again done in SPL. Capture the real waveform from a complete setup.
 
Top Bottom