How to Test Objectively Test Tube Rolling?

[watchnerd]

I have a lot of 6922-family tubes.

Like 16 pairs, most of which are NOS. They range from cheapie modern production JJs to expensive collectible Siemens, Telefunkens, and Mullards.

When used in my Mjolnir 2 I *think* I hear a difference between some of them, which probably isn't completely unreasonable given the variations in specs and age and condition.

But the process of tube rolling is perilous because:

--It can't be blind. You have to take the tubes out and swap them.

--It's not quick. It can easily take 3-5 minutes to just physically swap the tubes (longer if they're hot).

--Time it takes tubes to warm up to full steady-state temperature

As a result, I'm thinking I need to devise a digital capture methodology, sample different tubes, and then put into a software ABX tester.

Currently thinking:

Raspberry Pi DAC -> Mjolnir 2 -> Pre-out -> Devialet line-in -> ADC -> USB out -> Audacity on Macbook

So assuming the chain is okay:

1. What protocol should I use for "tube readiness"? Should I just let them sit there for 5 min before testing? Play music for 30 min before sampling? Other?

2. Given the large number of tubes to sort through, the decision tree will be insanely complex if I have too many songs to listen through. So I'm convinced I need to pick one song. Any major problems with this?

3. Should the levels be matched? If one tube has more gain than another (quite likely) that is an attribute of the tube that could make it sound different. Should this be eliminated through normalization?

Feedback and thoughts welcome.

 

[dc655321]

Yes to the digital capture and consistent warmup period, no to the measure differences with your ears.

Check out @pkane's Delta Wave software.

 

[watchnerd]

Yes to the digital capture and consistent warmup period, no to the measure differences with your ears.

Check out @pkane's Delta Wave software.

Every time I've tried to use Delta Wave I don't feel like I learned anything because the tubes don't perfectly null, anyway.

 

[watchnerd]

The next question:

Should I use high or low gain?

My understanding of the Mjolnir 2 architecture is that the high gain setting is more directly coupled.

It's certainly noisier, with tubes, than low gain in my experience.

 

[pkane]

Every time I've tried to use Delta Wave I don't feel like I learned anything because the tubes don't perfectly null, anyway.

It's not the tubes that don't null. And you don't need a perfect null to be able to compare two tubes in a circuit. The difference is what you're after.

 

[watchnerd]

It's not the tubes that don't null. And you don't need a perfect null to be able to compare two tubes in a circuit. The difference is what you're after.

I can try it, but it just looked like a lot of random hash noise last time I did it.

 

[pkane]

I can try it, but it just looked like a lot of random hash noise last time I did it.

Or try a simpler test using REW RTA and its signal generator. Measure distortions, noise, etc. You can also do signal sweeps with REW to measure THD+N over frequency or level, like those Amir does.

 

[watchnerd]

Or try a simpler test using REW RTA and its signal generator. Measure distortions, noise, etc. You can also do signal sweeps with REW to measure THD+N over frequency or level, like those Amir does.

No way I'm going to get through that much work for 15-16 pairs of tubes.

I'd need to have some simpler process of elimination and maybe do that for the top few.

 

[pkane]

No way I'm going to get through that much work for 15-16 pairs of tubes.

Not much work once you figure out how to do it, takes a few minutes per test. Learning what to do and how to do it right will take more time, but is worth the effort, in my opinion.

 

[dc655321]

the tubes don't perfectly null, anyway.

If your interest is "what's the difference between tube A and tube B", then that's exactly what to look at.

 

[watchnerd]

If your interest is "what's the difference between tube A and tube B", then that's exactly what to look at.

Noise seems to be the dominant component every time I've done it in the past.

But we can try again.

 

[dc655321]

Noise seems to be the dominant component every time I've done it in the past.

But we can try again.

Multiple iterations and averaging are the enemy of noise

Maybe you could describe your methodology that lead to noisy results? Perhaps there was something systematic that could be corrected in future attempts...

 

[watchnerd]

Multiple iterations and averaging are the enemy of noise

Maybe you could describe your methodology that lead to noisy results? Perhaps there was something systematic that could be corrected in future attempts...

Honestly, I don't remember the methodology, other than it was using music and not test signals.

It was years ago.

The systematic part causing noise is probably the tubes themselves -- it's easily audible on many tubes if you crank up the volume with no music playing, and audibly varies from tube to tube.

 

[Balle Clorin]

The best way is to measure the amplifier for gain, frequency response and distortion with the different tubes in. The is a valid result, the n for fun:
Then set the gain to get the same volume/ volt out. Compare the best and worst and hear if it is any difference,

 

[PierreV]

I tried it a couple of times. I got frequency response differences in REW + UMIK (edit: between SOVTEK and Golden Lion tubes).
But I feel it didn't really matter because

- the amp is not "Hi-Fi" by this site's definition (which is fine). I doubt any tube amp is anyway.
- the amp is not a high-end amp that would provide bias adjustments and current bias state. I expect it to be out of spec.
- I also expect the tubes to be out of spec, inconsistent from batch to batch.
- I expect "matched" pairs not to be necessarily matched (charging more for matched tube that aren't even matched would be only be a minor offense in the world of audiophilia nervosa compared to cables, ethernet switches and the like).

which is absolutely fine.

I am not anti-tube btw, I am listening to my tube amplifier right now, fed by a RP and a Khadas TB, on extremely cheap Tannoy Reveal 6P speakers which I picked up for EUR 120 a pair when they were discontinued.

The system is definitely smooth and liquid compared to the dry resolution of the Hypex driven Scala Utopias .
A different experience and an agreeable one afaiac.

 

[watchnerd]

The system is definitely smooth and liquid compared to the dry resolution of the Hypex driven Scala Utopias .

I thought at first you meant Klipsch La Scala driven by Hypex.

And my thought was: OMG LOUD

 

[don'ttrustauthority]

I have a lot of 6922-family tubes.

Like 16 pairs, most of which are NOS. They range from cheapie modern production JJs to expensive collectible Siemens, Telefunkens, and Mullards.

When used in my Mjolnir 2 I *think* I hear a difference between some of them, which probably isn't completely unreasonable given the variations in specs and age and condition.

But the process of tube rolling is perilous because:

--It can't be blind. You have to take the tubes out and swap them.

--It's not quick. It can easily take 3-5 minutes to just physically swap the tubes (longer if they're hot).

--Time it takes tubes to warm up to full steady-state temperature

As a result, I'm thinking I need to devise a digital capture methodology, sample different tubes, and then put into a software ABX tester.

Currently thinking:

Raspberry Pi DAC -> Mjolnir 2 -> Pre-out -> Devialet line-in -> ADC -> USB out -> Audacity on Macbook

So assuming the chain is okay:

1. What protocol should I use for "tube readiness"? Should I just let them sit there for 5 min before testing? Play music for 30 min before sampling? Other?

2. Given the large number of tubes to sort through, the decision tree will be insanely complex if I have too many songs to listen through. So I'm convinced I need to pick one song. Any major problems with this?

3. Should the levels be matched? If one tube has more gain than another (quite likely) that is an attribute of the tube that could make it sound different. Should this be eliminated through normalization?

Feedback and thoughts welcome.

I would lend someone my Vali 2+ for this. It'd be an easier way to swap quickly as it's one tube and the tube is higher up, though you could take the cover off the Mjolnir for this.

Get a friend to swap 2-4 tubes. Listen to some music over a few days and identify which tube is in. I think a Gold Lion sounds way different than a stock tube, more bass, more treble, easy to hear. I think if you correctly identified a tube that was swapped by someone else would be valid if you did it 3, 4 times out of 4, 5 tries.

Most of these tubes are close but some are clearly different sounding.

 

[watchnerd]

I would lend someone my Vali 2+ for this. It'd be an easier way to swap quickly as it's one tube and the tube is higher up, though you could take the cover off the Mjolnir for this.

Get a friend to swap 2-4 tubes. Listen to some music over a few days and identify which tube is in. I think a Gold Lion sounds way different than a stock tube, more bass, more treble, easy to hear. I think if you correctly identified a tube that was swapped by someone else would be valid if you did it 3, 4 times out of 4, 5 tries.

Most of these tubes are close but some are clearly different sounding.

You still have to wait for the tubes to warm up again.

That's going to be longer than your audio memory.

Plus, how do you swap tubes double blind?

FLAC via ABX software avoids all that.

 

[Chrispy]

Tube Rolling: Does it Make a Difference?

I recently reviewed the Mhdt Pagoda tube DAC. Owner was kind enough to send me three other tubes to test with it. Here are the set of tubes I received: Let's start again with the tube supplied with the unit (GE 5670) Now let's switch to Tesla 6CC42 (GA): Other than slightly more mains...

www.audiosciencereview.com

 

[watchnerd]

Tube Rolling: Does it Make a Difference?

I recently reviewed the Mhdt Pagoda tube DAC. Owner was kind enough to send me three other tubes to test with it. Here are the set of tubes I received: Let's start again with the tube supplied with the unit (GE 5670) Now let's switch to Tesla 6CC42 (GA): Other than slightly more mains...

www.audiosciencereview.com

Not a listening test.