How low can you go?

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
I slowly start to realize that our difference of opinion may be largely semantic. To me a spec is not just a parameter indicating some aspect of the performance, but a constraint on such a parameter set by a customer, marketer or system designer. If you don't meet it, in the best case you have to ask the system designer to redistribute error budgets and in the worst case your company gets sued by a dissatisfied customer.

Anyway, maybe you get more replies when you (have a moderator) change the title of the thread. When I first saw the title, I thought it was either something about subwoofers or an off-topic thread about limbo dancing.
 
Hi Guys
MrMagic: I do not like numbers that begin with a decimal or zero, so choose to use perfectly good and functional prefixes that already exist. To me ppb (parts-per-billion0 sounds a lot better than fractional-ppm. It's also impossible to lose the decimal when copying.

I don't think there are many insensitive ES head phones. Head phones in general have very high sensitivty and the reference level is 1mW. This is about the effective power for the ESL at 100Vrms across the stators.

I know lots of designs that aim for highest output, using +/-400V and higher. When I investigated my project and looked at the capabilities of the head phones themselves, they can output 130dB ! That is totally ridiculous. The problem is that to gain another 10dB of effective output requires about a third more drive signal than the previous step. IXYS has 4kV mosfets and these can certainly be used for the drive you have, although I prefer a true complementary circuit.

Most of the very high output ES drivers are single-ended inasmuch as there is active pull-down by an "n" device (tube, mosfet, BJT) and passive pull-up using a resistor. Since the drive is balanced, there is a mix of passive/active drive going on throughout the signal cycle, but like yourself, I prefer active pull-up/down.

The first time you get six zeroes for THD20 in LTspice you are then ruined forever. Seeing anything else is a bit of a let down - haha. So, I aim for all-zeroes for the signal range I expect to use and will settle for <ppm or single-ppm for much higher levels.

For example, Using the same front-end as for my ES driver, I expanded the design as a PA for dynamic speakers. With 100V rails and 90Vpk output, it is 6ppm at 2R at 20kHz, and about 4ppm at 8R. At 1kHz, THD is about half. Reducing output to 30Vpk, THD20 is about 1ppm at 2R and a feww-hundred ppb at 8R; THD1 is about 100ppb at 2R and <10ppb at 8R. Going down to 3Vpk out, everything is six-zeroes (<10ppb). This is a big amp with eight pairs of output devices in one form and sixteen pairs in another. A made a hybrid class-G version with three rails and its performance is a little bit worse at highest output and lowest z-load.

For the big amp, I figured the goal would be to bridge them as a super-amp, so low-z performance is important. It was more of a design exercise than something to build since 1kW+ is way ridiculous. In my living space with 90dB/1W/1m speakers, I cannot stay in the room at anywhere close to 90dB SPL. I've measured the power I need and it is <1W or so for normal use, split between two subs and two sats. It is easy to get good performance at these levels, which is good for me, since I'm not that smart... haha
 
I slowly start to realize that our difference of opinion may be largely semantic. To me a spec is not just a parameter indicating some aspect of the performance, but a constraint on such a parameter set by a customer, marketer or system designer. If you don't meet it, in the best case you have to ask the system designer to redistribute error budgets and in the worst case your company gets sued by a dissatisfied customer.

Try to see specs as challenges instead. A company might decide to design a high performance product, because they consider it a worthwhile challenge, they like creating products that come closer to perfection, and they are proud of them. Those companies who aim profit by any means, like by cheating the customer, are second-class companies with "hit-and-run" marketing policies, who will only invest in the minimum research and development possible.

That said, the hype nowadays is that high-performance is not needed, therefore impressive numbers are not needed, numbers do not represent performance, so research and development is not needed either, and that suits all companies that get away with mediocre products, using poetic descriptions to attract the customers instead, who coincidentally lack the paying capabilities of the past.

Speaking of challenges, personally I'm motivated by two challenges: novelties, and optimization. Both can be extremely fun. Specs in the general notion as features and behaviors, might satisfy the former, and specs as numbers, the later, and usually that's the sequence in development too: first you achieve the main features, then you optimize to perfection -although several revising cycles might be needed until the design takes the final form, which is what "development" actually means.

In the end, an extraordinary design can only be expressed precisely with extraordinary specs -not colorful, emotional expressions. So, in that case, specs is the proof that something extraordinary has been achieved, the celebration of a successful development expressed with victorious features and numbers. ;)

I do not like numbers that begin with a decimal or zero
. . . . .

The first time you get six zeroes for THD20 in LTspice you are then ruined forever.
Seeing anything else is a bit of a let down - haha.
You contradict yourself, but I know exactly the feeling of beeing ruined by the number of zeroes :)

Imagine 4 zeroes at 1300V and 1300 gain -orders of magnitude higher voltage and gain, (that's where things get exponentially difficult) ;)
After that, I can only accept sub-ppm THD @20Khz to keep those 4 zeroes intact :D
I personally prefer zeroes, because one can visually and intuitively see the big difference with the best highly expensive amps in the *market. It is also an intuitive challenge during development, to push numbers as far as possible to the right until they ...vanish into a sea of zeroes, while competition struggles to count them :D

*I haven't found a single loudspeaker power amplifier with less than 0.001% distortion at up to 20Khz (and that is the Elektra HD2 7 Channel Power Amplifier), if anyone knows of a loudspeaker amplifier with less distortion at 20Khz, please let me know!

IXYS has 4kV mosfets and these can certainly be used for the drive you have, although I prefer a true complementary circuit.
Nope, they certainly can NOT be used for the same purpose. They can only be used in pulsed applications by design. If you look closer to their graphs and calculate the demands, you'll find out.

There are no semiconductors for complementary topologies at such voltages and cascoded topologies are too complex for my taste. I'd accept them only for very high voltages in order to drive ESLs directly (as there is no SS alternative).

Most of the very high output ES drivers are single-ended inasmuch as there is active pull-down by an "n" device (tube, mosfet, BJT) and passive pull-up using a resistor. Since the drive is balanced, there is a mix of passive/active drive going on throughout the signal cycle, but like yourself, I prefer active pull-up/down.
Single ended with a pull-up resistor, are highly asymmetrical causing too much distortion at high frequencies, and thus cannot be used when aiming low THD for the full audio band.

I don't think there are many insensitive ES head phones.
If there are no insensitive headphones, then there are insensitive ears that justify those HV amps in production :)

they can output 130dB ! That is totally ridiculous.
Agreed, and that is why I'll trade ridiculous high output for better low frequency response in my next custom-designed ES headphones.

Anyway, maybe you get more replies when you (have a moderator) change the title of the thread. When I first saw the title, I thought it was either something about subwoofers or an off-topic thread about limbo dancing
Initially I wanted to provoke a competitive spirit for new designs, or great forgotten designs, but apparently there is no interest whatsoever, nor to simply collect most projects in one thread.
More than 200 different members have seen the thread, so I bet that those who would be interested to include their project, have already seen it but they are not interested. I have started to realize that direct-comparison is terrifying to most people.

So, if philosophical spec-tology has become the thread's main subject, then so be it. :p
And it would be ironic to call it "Amplifier Project Database" after so much blah-blah anyway. Administrators won't let us change the name twice!

That said, bold members proud for their projects, are still welcome.
We are open and we are waiting for you!
:D

.
 
Last edited:
Member
Joined 2011
Paid Member
I don't think it would upset Nelson Pass, if a super enthusiastic and motivated person dug the measurement results out of his First Watt website, and included them in a table here. There are more than a dozen different amplifiers on that site, built and measured! Here's an example of the measurement data he has published.

_
 

Attachments

  • figure_1a.png
    figure_1a.png
    36.1 KB · Views: 212
  • figure_1b.png
    figure_1b.png
    28.8 KB · Views: 213
  • figure_2.png
    figure_2.png
    35.3 KB · Views: 208
  • figure_3.png
    figure_3.png
    81.5 KB · Views: 212
Last edited:
It was suggested to me that I throw my name in the hat here. I am measuring the following on my Modulus-686 prototype.

Gain: +20 dB
THD+N: 0.00026 % (140 W @ 8 Ω)
THD+N: 0.00029 % (260 W @ 4 Ω)
Onset of clipping: 0.0005 % THD+N @ 220 W (8 Ω)
Onset of clipping: 0.070 % THD+N @ 380 W (4 Ω)
Noise: 15 uV RMS (A-weighted, 20Hz-20kHz)

Build cost: $300/channel + power supply, connectors, heat sinks, etc.

I've attached the THD+N plots. I'll start a separate thread for it in the Vendor's Bazaar over the weekend, so if you have questions, please keep an eye out for it there.

You can find the numbers for my Modulus-86 here: Modulus-86 Rev. 2.2: Composite amplifier achieving 65 W (4 Ω) at 0.00006 % THD.
A brief summary:
0.000061 % THD (1 W, 8 Ω, 1 kHz).
0.000067 % THD (40 W, 8 Ω, 1 kHz).
0.00038 % THD+N (40 W, 8 Ω, 1 kHz).

Tom
 

Attachments

  • Modulus-686 PROTO_ THD+N vs Output Power (8 ohm, 20 dB gain, MW SE-600-36, 1 kHz, 20 kHz BW).png
    Modulus-686 PROTO_ THD+N vs Output Power (8 ohm, 20 dB gain, MW SE-600-36, 1 kHz, 20 kHz BW).png
    50.8 KB · Views: 231
  • Modulus-686 PROTO_ THD+N vs Output Power (4 ohm, 20 dB gain, MW SE-600-36, 1 kHz, 20 kHz BW).png
    Modulus-686 PROTO_ THD+N vs Output Power (4 ohm, 20 dB gain, MW SE-600-36, 1 kHz, 20 kHz BW).png
    50.9 KB · Views: 96
Last edited:
Try to see specs as challenges instead.

If you would try to see specs as constraints on performance figures set by customers or marketers, you might understand why I consider them neither necessary nor useful for hobby projects. You can optimize whatever parameters you like to optimize using all your ingenuity and all your common sense without having some external person forcing a set of constraints on them.

It's a bit similar to art. Artists often make their best works when they don't have to worry about whether a customer will like it.
 
Well, I have no "external persons" dictating specs to me, and I imagine that is the case with all DIYers, so that doesn't apply to us.

And if I had my own hardware company, no one else would dictate the products' specs than myself, and there are many companies like that.

What is required to make art, is inspiration, and the ability to use it in a creation, or part of it, nothing else.
Even if the specs were "dictated" to the designer, if the designer had an inspiration on how to overcome the problems and achieve the tough specs because he likes his job and considers tough specs a challenge, it would still be an inspired product, that would contribute with its inspired novelties (a case where inspiration and art are triggered by tough specs or constraints, as in many other cases).

And that is the real question, the actual quality of the specific product and its contribution to technology, not whether the boss worried about the consumers and the course of his company, or whether the specs were "dictated".
 
Well, I have no "external persons" dictating specs to me, and I imagine that is the case with all DIYers, so that doesn't apply to us.

Precisely, so according to the meaning of the word "spec" that I'm used to, you have no specs. You have ideas about what kind of performance you would like, but you can change them whenever that makes sense.

Hence my claim that worrying about specs sounds more like work than like hobby to me, which started the whole discussion between you and me. And hence my later remark that we largely disagree about semantics.
 
You are a moving target MarcelvdG and thus we are going in circles...
An externally hosted image should be here but it was not working when we last tested it.

Unless you become SPECific about what you really mean, we cannot arrive to a common conclusion. Even a discussion, needs to follow some specs actually. ;)

So, let's take the most negative meaning of specs: forced constraints.

You can't attempt anything worthwhile without forcing some constraints, ie specifying a minimum goal or specs. Even when you decide to design or play a game, you have to set some strict constraints that shouldn't be possible to violate, in order to make the game challenging and thus, fun. Without a challenge, there is simply NO fun in anything. Period.

If some of my specs are "flexible", that's due to a strategy defined by other specs (you can consider it an algorithm) and like me, a company might follow the same strategy or alghrithm when attempting something novel, or extremely optimized via R&D way beyond what's available, so in that case, both me and them cannot pre-define all the final specs precisely -doing so while in uncharted waters, would be ridiculous.

For example, the THD <0.01% and 1400V I had set for my ES amp on the other thread, were definitely "specs" (constraints), as were the following:

-DC,
-Direct drive (which means a dual supply requirement),
-ultra low noise (the minimum I can achieve)
-deep optimization (perfectionism)
-high reliability
-reasonable cost
-competitive to the best consumer products

All are currently satisfied (with ultra low noise, and THD hitting a record) except voltage which I had to lower to 1300V due to component limitations, by giving priority to "high reliability" and "reasonable cost" specs / constraints.

Now my new goal is to make the prototype keep them, specifically THD < 1ppm and I've set some new real-life noise-related goals that require more R&D. That, aside the additional feature-specs (like embedding a DAC) that also require more work.

Bottom line: Strict constraining specs are required for anything worthwhile, even for fun, or hobby. Being flexible with some of the specs, might be a strategy, not an indication of being loose and having more fun. If you are too loose, you are not self-desciplined and that reduces your efficiency and your ability to accomplish a worthwhile goal, considerably, and finally long-term satisfaction and enjoyment.
 
Last edited:
I would love to give you some practical examples, but that would violate NDAs, so I have to stick with hypothetical cases.

Suppose you would have overconstrained one parameter while forgetting another one, one that's far more important during normal use. As a hobbyist you would probably find that out during the design phase and adjust your requirements accordingly. In a professional context, that's far more difficult.

A typical example are the audio DAC chips I mentioned before. Suppose a customer puts constraints on chip area and dynamic range, forgets headroom, and makes the constraints on chip area and dynamic range so tight that you can only just meet them without any headroom. The only way you can meet the customer's requirements is to make a DAC that has no headroom and is therefore not suitable for playing music. Of course you could try to convince the customer to reduce the dynamic range and add headroom, but their marketers would probably not accept that, because the customer's customers only look at dynamic range anyway.

In the end you would spend months to develop a crap product with impressive-looking datasheet numbers that's unsuitable for playing music, and probably get quite frustrated about it, especially if you like your job.
 
Marcel makes sense to me.
Sometimes specmanship is a bad thing. Audio circuits should aim to convince a human that they are hearing music. The very best measurement device is a human. Everything else is a proxy measure. A proxy is of questionable dependability so focusing exclusively on proxies is risky if not downright misleading.
For example, the THD proxy that is often top of the list is very poorly correlated. So why judge a design by it? The DAC spec examples illustrate how choosing the wrong spec can lead to designs with great specs and poor sound.

When I buy a piece of audio gear I don’t bring a laboratory to the showroom with me and let it decide what I should buy. Of course not. I audition. I take a lot of time. I compare using different music. Before I go to the shop I read magazine reviews to confirm a number of things but mainly the audition by the reviewer.

So which diyaudio project a person chooses may be driven by all sorts of things. For the particular criteria of “playing music convincingly” I don’t think a long list of proxy specs is going to be very helpful.
 
Last edited:
In the end you would spend months to develop a crap product with impressive-looking datasheet numbers that's unsuitable for playing music, and probably get quite frustrated about it, especially if you like your job.
Unless you are a DAC/amp supplier to Apple! :D
<rant>I just don't believe how poor music is on my 6S and its even worse on the external DAC dongle they provide for newer, jack-castrated phones. These are really expensive devices and why doesn't Apple ask someone who knows what they are talking about to sort this out? Instead, we get a $1000+ phone with a highly complex face recognition system that nobody wants and Beats quality audio.
<end rant>
 
Suppose you would have overconstrained one parameter while forgetting another one
. . . . . . . . . . . .
Suppose a customer puts constraints on chip area and dynamic range, forgets headroom, and makes the constraints on chip area and dynamic range so tight that you can only just meet them without any headroom.
That's a classical case of a logical leap, or fallacy.
You blame the scientific / technical institution of using specifications, and the companies and DIYers who take specs seriously, just because some "customers" and companies abuse that institution, trying to cheat people with various tricks.

It's like considering science evil, or banning it altogether, just because some fraudsters use science to cheat the public with pseudo-scientific evidence.
It doesn't work like that, which is why I've insisted a couple of times in our debate-Odyssey that this is irrelevant, or OT.

Abusement, fraud and cheating, exists in every sector you can imagine. That doesn't mean anything, there will always be a percentage of parasitic and pro-chaotic behavior, like the noise you see on a FFT graph. No system is 100% perfect on this universe.

It's up to us to look for the fine print in datasheets or spec lists and check in depth what we read, in order to avoid being deceived, 99.9% of the time.

Personally, if I discover that a company systematically tries to cheat me, I'll ban that company completely, and most likely I will share my experience with other people in an emphatic way.

Such companies don't have a bright future, or no future at all.
It's on the fundamental principles of marketing the requirement of a great product, and a policy that respects the customer, to ensure a marketing success and start using marketing in the first place.


The very best measurement device is a human. Everything else is a proxy measure. A proxy is of questionable dependability so focusing exclusively on proxies is risky if not downright misleading.
The problem is that we humans are highly emotional beings, prone to various psychological effects like placebo, and extremely high sensor illusions (visual, acoustical, etc), let alone the differences in personal training, experience and knowledge, therefore our judgement is a mixture of too many unknown and unmeasurable parameters.

So in order to have a good judgement of performance using our vulnerable sensors and perceptions, highly sophisticated tests have to be designed and used. Just going into a shop is rather hopeless to judge subtle specs like those for an amplifier. You will have a better chance with loudspeakers and headphones though where the differences are huge.


For example, the THD proxy that is often top of the list is very poorly correlated. So why judge a design by it?
I would love to hear why "THD is very poorly correlated", assuming all else equal of course. I'm all ears.


<rant>I just don't believe how poor music is on my 6S and its even worse on the external DAC dongle they provide for newer, jack-castrated phones. These are really expensive devices and why doesn't Apple ask someone who knows what they are talking about to sort this out? Instead, we get a $1000+ phone with a highly complex face recognition system that nobody wants and Beats quality audio.
<end rant>
So you're missing Steve, right? :D
 
Last edited:
So in order to have a good judgement of performance using our vulnerable sensors and perceptions, highly sophisticated tests have to be designed and used.
Our auditory systems have evolved over millions of years. Have some self-respect.

I would love to hear why "THD is very poorly correlated", assuming all else equal of course. I'm all ears.
Speaker THD.

So you're missing Steve, right? :D
No. I'm missing decent sound quality in my phone.
 
So in order to have a good judgement of performance using our vulnerable sensors and perceptions, highly sophisticated tests have to be designed and used.
Our auditory systems have evolved over millions of years. Have some self-respect.
Our auditory systems have evolved, but our perception when doing audio tests, is the sum of several illusions, plus distortion is more apparent in specific sound combinations (content-dependent), which is why you need to listen with a proper test method in order to exclude all other conflicting factors, and focus on the most revealing parts.

For example, the THD proxy that is often top of the list is very poorly correlated. So why judge a design by it?
I would love to hear why "THD is very poorly correlated", assuming all else equal of course. I'm all ears.
Speaker THD.
We were talking about amplifiers, THD is not on the top of the list in loudspeaker specs, it hardly exists in most specs, and I've already stated previously that there is no problem with loudspeakers because the differences are huge.

saying that humans are not qualified to judge music is...well, bizarre.
Nobody said that, and we are not talking about judging music, we are talking about judging subtle audio reproduction differences. Whatever test you can imagine, needs a proper test environment and methodology. That's basic knowledge.
 
Last edited:
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.