Bonsai, I know this thread is chaotic and I'm not exactly helping, but I really wish you the best and I am grateful you have decided to do this.
In this exemple you are downgrading the amp because of the speakers eventual poor quality.
Beside the point was about a 20W/channel amplifier, not 5-10W, that s a big difference since in apartments environments 40W
total power is overkill, the old german DIN 45500 standard was on point when it did set the minimal RMS power at 2 x 6W for a hifi reproduction.
The "cost per watt" seems unnecessary. It does not really apply to DIY amps and owners of expensive amps are not running these tests anyhow (I hope).
Regarding speaker cables and RFI/EMI, the discussion has brought back a memory from some years ago: A 100R non-inductive resistor across the speaker terminals can be helpful. IIRC, it was @1audio who mentioned using a particular Caddock resistor for that purpose.
Also reminded me of @1audio speaking to the issue of measuring nonstationary distortion: https://www.diyaudio.com/community/...eapon-to-fight-the-jitter.192465/post-6346129
Seems to me he also at some point discussed certain possible problems with amplifiers that would be difficult or practically impossible to measure. Have to see if I can find that one.
Also reminded me of @1audio speaking to the issue of measuring nonstationary distortion: https://www.diyaudio.com/community/...eapon-to-fight-the-jitter.192465/post-6346129
Seems to me he also at some point discussed certain possible problems with amplifiers that would be difficult or practically impossible to measure. Have to see if I can find that one.
I still have not had time to read through all the feedback - dogs to vets today, etc
However, as a suggestion, I would like to see us finalize the basic assessment in the next few days and take the next step.
With respect to distortion, everyone agrees that there is not a good correlation between distortion figures below a certain level and perceived sound quality ON DOUBLE BLIND TESTS. We also know that above a certain level, no matter what the type of distortion, it is clearly audible. Is that acceptable?
I'd like to propose we set up an 'independent technical assessment group' of 5 (no more) well-qualified individuals who we all agree can bring something serious to the table to go away and then come back with some recommendations on how we are going to assess distortion in the proposed revision of the document. I will not participate in that working group because they need to do it independently and most importantly, there are better qualified people here that can do it.
So, put some proposals forward as to who can do this, and let's give the team 4-6 weeks to investigate it and come back. Here are the guidelines: -
1. Decide what levels and types of distortion are acceptable given that below a certain level distortion levels are indistinguishable from each other
2. Specify a test in some detail. For example e.g. 'run an FFT, extract all distortion harmonics between xHz and yHz and weight them as follows, etc'. The test must specify the amplifier output power levels and so forth.
3. The test result should be extractable with a Python program and/or a script running on something like a QA401/2/3 from an FFT data dump so it can quickly be turned into a working standard
4. Where possible, links to supporting scholarly articles should be included.
Anyway, I hope you like the suggestion. I will take in the feedback received so far and update the doc in post 1 today and tomorrow
🙂
However, as a suggestion, I would like to see us finalize the basic assessment in the next few days and take the next step.
With respect to distortion, everyone agrees that there is not a good correlation between distortion figures below a certain level and perceived sound quality ON DOUBLE BLIND TESTS. We also know that above a certain level, no matter what the type of distortion, it is clearly audible. Is that acceptable?
I'd like to propose we set up an 'independent technical assessment group' of 5 (no more) well-qualified individuals who we all agree can bring something serious to the table to go away and then come back with some recommendations on how we are going to assess distortion in the proposed revision of the document. I will not participate in that working group because they need to do it independently and most importantly, there are better qualified people here that can do it.
So, put some proposals forward as to who can do this, and let's give the team 4-6 weeks to investigate it and come back. Here are the guidelines: -
1. Decide what levels and types of distortion are acceptable given that below a certain level distortion levels are indistinguishable from each other
2. Specify a test in some detail. For example e.g. 'run an FFT, extract all distortion harmonics between xHz and yHz and weight them as follows, etc'. The test must specify the amplifier output power levels and so forth.
3. The test result should be extractable with a Python program and/or a script running on something like a QA401/2/3 from an FFT data dump so it can quickly be turned into a working standard
4. Where possible, links to supporting scholarly articles should be included.
Anyway, I hope you like the suggestion. I will take in the feedback received so far and update the doc in post 1 today and tomorrow
🙂
keantoken,
a quick response to your points (and thanks for the feedback btw) with some clarification.
1. A low impedance load usually degrades loop stability sub-optimally compensated amps. However, I will concede that in most modern amps using EF2 or EF3, the phase margin degradation between 2 and 8 ohms is minimal. The problem with the inductor in series with the load resistor is it forms a damped tank circuit with the load cap, so you get the expected response peaking. This does not expose the amplifier stability with capacitive load.
2. Agree. A Bowes test is a good idea to catch this. A low frequency (say 100 Hz) has a small 1us rise/fall time square wave of 10 kHz or so superimposed on it. The LF signal pushes and pulls the device parameters around, while the square wave signal provides information on the loop stability. On a well-engineered amp, the square wave is well-behaved at all points on the LF signal.
3. Agree. that's why the test should be done at ~3V pk~pk.
4. Agree - you have to separate the ringing due to amp output L and load capacitance - all amps ring with this to some extent. If you change the amp load cap and look at the frequency, it should be pretty easy to deduce the amplifier output L - be that parasitic or because the designer has used an output coupling inductor. The ringing we are looking for here is that due to instability because of incorrect compensation i.e. inadequate gain and phase margins, so it's important that the test document makes that explicit.
Generaly speakin ; every sign of output signal oscillation is the sure sign that amplifier is on his way and want to become oscillator or transmitter .
Markw4, thanks for the ITU paper. Can you PM or email the other stuff to me please. It was an important issue to me in da previous Millenium but I've been a beach bum all of this one. I believe there is one IEC(?) recommendation for ABC tests which I'd very much like to see. The importance of ABC vs ABX is you get statistical significance a LOT quicker.
My involvement in DBLTs goes back to well before https://www.aes.org/e-lib/browse.cfm?elib=3798 though the 'final' form was very different from what's described there and in our other published papers. Da false Harman prophets do quote some of our stuff.
You'll note I distinguish very carefully between discrimination and preference. No point asking for preference if there is no discrimination. But IF there is reliable discrimination, I WILL ask for preference.
And YES. The preference is always for the 'correct' (eg less distorted for THD tests) presentation EXCEPT for da JLH Class A test. I've spent a long time trying to get its THD profile in something that isn't also a space heater 🙂
And YES. If you do a LOT of DBLTs, you quickly learn how to cheat 😱 But doing these tests is so expensive (eg GedLee on minimum $$$$) that you NEVER fudge a test. I've had to abandon a huge amount of $$$ data in the past .. eg when I realised my panel, under certain circumstances, could detect frequency response anomalies of 0.2dB
One (there are loads) big thing I don't agree with da Harman false prophets is the need for 'trained' listeners. When there is discrimination and preference, da untrained listener agrees with Bat Ears ... IF you let the listener select his or her OWN music ... and other stuff I do which Harman bla bla etc don't do. We were quite surprised when we found this.
Last edited:
PM sent.
Regarding the need for trained listeners, it depends on what the research question is. If you want to know what naive listeners off the street will notice, then it wouldn't make sense to train them. If you want to know the limits of human performance, you have to train them. Its like wanting to see what the best basketball players in the world can do on a good day. You don't want people picked randomly off the street filling in as untrained substitutes.
Does this mean you were sorting them out using a DAW and EVIL software rather than your Golden Pinnae 😱
Absolutely. But IF there was general discrimination, there was also general preference. That was surprising when we first realised this.
The other surprising (perhaps not with hindsignt) result was that reviewers were significantly less discriminating than da man (and more particularly da woman) in the street.
This is really sad as these days, you almost NEVER get to hear stuff in a shop. You just read da reviews on da WWW 🙁
Oh! And da wannabe Golden Pinnae are ALWAYS deaf (give random results in DBLTs)
Last edited:
Bonsai, at the risk of going off topic in yet another direction, do you know of any 'good' amps that are unconditionally stable with load without an output inductor? I don't think this is theoretically possible.
I do know of one practical amp that was but it was lacking in other respects for listening. We used it as the main amp in our anechoic to replace the B&K one.
I do know of one practical amp that was but it was lacking in other respects for listening. We used it as the main amp in our anechoic to replace the B&K one.
This was measured with 42V PSU. When pushed more hard (higher output power), both even and odd harmonics rise, but still with quite quickly damped profile. Heat and low power of this amp is an issue, and also mediocre PSRR. I have played with it in 2012 (again after 35 years), but after initial enthusiasm I was disappointed and the amp finished in the dust bin.
It would seem trying to tell what makes an amp sound good is a very wide spread of characteristics, maybe defining what makes an amp sound bad maybe easier to discern and look for a lack of those facturrs.
I still had to listen, but I could chose to loop short sections I thought were different. Also I could do quick A/B switching in the looped section. Its was enough to be able to bubble sort them.
BTW, it turns out I saved those files. After all these years I just now listened to two of them on my laptop with $20 amazon headphones. They still sound different to me, although not exactly like what I would normally think of as HD/THD. Just some slight differences between some of the cymbals. Anyone interested in giving them a try?
This test is not to identify response problems, it is to expose oscillation bursts. The tank peak caused by the parallel resonance is not the cause of oscillation, and is fully shunted by the amp's zobel network. Please distinguish FULLY between ringing and oscillation because ringing does not ever confirm instability. Cables ring, speakers ring, inductors ring, but passive components NEVER oscillate on their own. I NEVER suggest that just because there is ringing in the response it is unstable because this is simply not true. Oscillation is always a malfunction whereas ringing is just an extended form of overshoot.
When you add a capacitor to the amp output, the response becomes a function of the wire inductance which has nothing to do with the amp quality. What concerns us is if the amp oscillates which is something the inductance is clearly not responsible for.
An amp with 10 degrees of phase margin is viable if the phase margin never drops by 10 degrees during operation. An amp with 10 degrees phase margin may not even have ringing or overshoot. But an amp will not oscillate unless phase margin goes to zero. You will never get rid of ringing unless you test only resistive loads.
If you are looking for ringing, look for it with resistive loads for the test below. Flat response into capacitors is an amp FEATURE, and is an unreasonable requirement to have for amps in the current market.
I've used this test for a while, but I didn't know it had a name. This is an excellent test to do, but not on loads which are EXPECTED to ring like capacitors. If signal generators that can do this are everywhere, then please add this test to the document. But when you add a capacitor to the output of an amplifier, it rings with the wire inductance which has nothing to do with the feedback loop quality.
This test might expose misbehavior of some sort, but it doesn't replace any other test. It does resemble very high phase load stress where voltage is low but current is high.
Ringing due to instability has a negative decay, IE it becomes an oscillation. If the ringing has a positive decay it says nothing about instability and you can't distinguish ringing due to poor phase margin from parasitic ringing without a full blown analysis. Poor phase margin doesn't even guarantee you will have ringing. Using ringing to hunt for instability is a Moby Dick type activity. That's why, 40Hz C||LR load gives plenty of stress and allows oscillation to build and doesn't give false positives (ringing with positive decay is ALWAYS a false positive for categorical instabilty).