Ah, I understand. The biggest differentiator is that some designs have the output filter inside their feedback loop, and some do not. Being an analog man, with trust in analog components, I prefer to select linear output filter components and only apply feedback around the A-D conversion in the chip itself. The two techniques will sound, and measure, differently.

I covered the issues of filter design topology, and component selection, in my web 'tutorial' http://trevormarshall.com/class-d-tutorial/ The best way to analyze filter performance is to use the circuit emulator, as I did. I give the design files on my webpage. Generally, the filters used in commercial Class D designs are far from optimal, too little thought has gone into them. Since my interest was also in Class E, 'the resonant filter', I understand the importance of filter design and component selection.

Each of these chipsets converts the incoming analog signal to a single-bit digital representation. That is then amplified and passed to the output filter to convert it back to analog. The TI chip technique, delta-sigma modulation, was what I have used in the past, and should be pretty good. I haven't listened to any of the TI designs, however. The IRS2092 conversion is excellent, I did test several of those boards. I selected the TDA7498 because they sounded good, were cheap on Ebay from SURE, pretty easily modified to bring them up to peak performance. Most important - they were small and didn't get warm. I needed to build a tiny amp and speakers that I could throw into a carry-on size case as I move from city to city. Portability was uppermost in my mind for both the amp and speakers.

I have attached the distortion vs power curves into 8 ohm (55W) and 4 ohms (100W)for the Tripath from HIFIMEDIY, FYI.

Do you mean the analog-digital conversion algorithms?

Vitaliy, I bought the "HIFIMEDIY AMT2050 V2," a marginally earlier version of this one: http://hifimediy.com/index.php?route=product/product&product_id=52 . I also bought a headphone amp using the TA2024 and a few boards with IRS2092 technology.

Of these the IRS2092 measured best, with lowest IMD and THD, but I wanted a single supply amp I could feed from a standard portable "power brick."

The Tripath sounded OK, although a lot of negative feedback has been used, and so it really didn't like having supersonic frequencies sent to it, or handling highly reactive loads (such as speakers). The "pop" at power-on (a relay stops this) was annoying, but my biggest concern was the lack of current limiting on the output. My new speakers include SEAS L16Rn woofers, which are 8 ohm and have a 250W short-term power handling (needing lots of voltage). Tripath dissipation was quite high, the heatsink remained warm even at idle. All these factors pushed me to the TDA7498, which is self contained, output-current limited at all frequencies, and can run OK off a higher voltage supply (I use 36V).

I have been watching the new TAS5613 TI chipset which has the higher output voltage (50V supply) capable of driving my woofers to distraction. Haven't tried this yet, however, as I still get 1cm woofer cone excursions with the TDA7498 :) I think the delta-sigma conversion of the TI chipset ought to be a better technology, I used it with great success in the 1970s. Maybe I will take the plunge and buy a board...

The biggest problem I have with most of the available Class D boards is that their inductors are just not capable of handling the maximum output power without saturating, and getting quite hot. That is one reason why I changed to a higher-current Bourns inductor in my final TDA7498 amps ( http://trevormarshall.com/class-d-tutorial/ ). The HIFIMEDIY do not suffer from this problem, nor do the IRS2092 boards.

Fidelity is greatly reduced when ceramic capacitors are used in the output circuit, due to the non-linear characteristics of the X7R dielectric. I documented in the above 'tutorial' the lowered distortion when I changed all the output capacitors to plastic film. Input caps need to also be plastic, for much the same reason. Ceramic and Aluminium seem OK for power bypass caps.

When you test these switching amps, make sure you use a load which is the equivalent circuit of your speakers. Generally this includes voice coil inductance. Because the output filters are very dependent on load impedance, the pulse (and IM) output into a speaker will not be the same as that into a purely resistive load.

Hope that helps :)

Second, I do not agree that "analog audio technology was pretty well fleshed out by the early '70s". As progress in capacitors, powerful transistors and opamps was after this period. Same is true for digital amplifier, even if academics somehow though otherwise.

Well, I was there and I remember it well. Indeed, my workshop is still flush with components from the 1970s. There has been little progress in capacitors since the 1970's. Tantalum was already prevalent, and low-ESR aluminium were also commonly available.

Powerful transistors? Heck, I was selling 2000 watt RMS amplifiers to the rock groups. The 2N3055 dates from the 1960's and was available in higher power and complementary versions by the early 1970s (see http://en.wikipedia.org/wiki/2N3055 ). I used to run them in parallel to get the higher power levels.

I have in front of me the "RCA Solid State Power Circuits Designer's Handbook" from 1971. You might like to chase down a copy in your local library, as it succinctly defines the powerful transistors available at that time. SPICE was already available to model electronic circuits, and SPICE variants still underpin the standard emulation tools of today.

Both bipolar and FET input op-amps were available in the 1970's and, although continual refinement is visible in today's products, their circuit diagrams still look essentially the same as they did in the 1970s.

There have been advances in Class D amplification in the nineties and noughties, but a lot of the newer designs, such as Tripath, have been promoted by slick marketing and "enthusiast personal preference" rather than necessarily being any better. I ended up using the lower-cost TDA7498 series for my recent high-end amp design, it measures well and sounds great. And yes, I have Tripath boards and have tested and listened to them extensively...

Even though PV is intended for scientific facts, a lot of what is pleasing to the ear actually measures poorly. As designers created "better" performing opamps in IC form, they lost the pleasing aspects of distortion and bandwidth limitations

Indeed, vacuum tube amplifiers measure very poorly, yet many enthusiasts (not me) swear that the sound they give is sweet and easy to take...

I measure everything before I use it. The measurements help me figure out where the weak points lie. Microphones, recorders, and you will see that I shoot ISO charts on every video recorder I use (I have posted many of those tests here). I agree with Vitaliy that measurements are critical (for the above-stated reason) and that there is just as much opinionated rubbish being posted in the HiFi forums these days as there was when the CD came along in 1982 and started to displace vinyl records. Abba's "The Visitor" was one of the first all-digital CD's and I remember the breath of fresh air I felt when I first played it. So I am pretty cycincal about folk who think vinyl sounds better. I have about 10 feet of vinyl records on my shelves, and the best moving coil cartridges and a Denon turntable, but I prefer the clarity of CD's. Unfortunately, many of them are not produced any better than the job Vanda and Young did on "Dirty Deeds Done Dirt Cheap" (http://en.wikipedia.org/wiki/Vanda_&_Young ) or the classical catalog of Deutsche Grammophon from the 60's and 70's :)

I think a lot of the esoteric designers consider negative feedback poor design practice due to the phase margins, nonlinearities that feedback can produce, and teetering on the edge of instabililty vs. performance.

Even though PV is intended for scientific facts, a lot of what is pleasing to the ear actually measures poorly. As designers created "better" performing opamps in IC form, they lost the pleasing aspects of distortion and bandwidth limitations.

Does a large, hot, high distortion and slow API2520 opamp outperform an NE5532? On paper, absolutely not, at least based on the clinical measurements that we would make. Does it sound better to most who objectively listen? Almost definitely. It's one of those times where science has actually steered us away from the direction we wanted to go, to the direction we thought we wanted to go.

Interesting dissertation on cables and distortions..

http://sound.westhost.com/cables-p3.htm

High-end audio was nearly non-existent in the 1970s short of a few companies like McIntosh and even those marketed their gear around objective criteria .

What a silly statement. Analog audio technology was pretty well fleshed out by the early '70s and by the late '70s academics like myself were exploring the limits of Class D and Class E digital amplification. Nakamichi is one analog brand that excelled - measurably better than than McIntosh ever was - I still use a Nakamichi amp as my TV audio power amp, I forget how many decades old it is. Early '70s saw the introduction of delta-modulation and logarithmic PCM signal processing - which are still used as the basis for today's computer audio technologies. Celestion speakers from the 1970s (used in the Marshall guitar amps) are the speakers most desired by many (if not most) of today's guitarists.

I disagree that any conclusions at all can be drawn about what happens with 'real music.' As somebody who has designed power amps (voice and guitar) for rock musicians (including Bon Scott) and also classical HiFi rigs, I know that none of my amps with bad IM and TIM ever sounded good. Mostly these are not subtle defects, either.

I agree 100% with the craziness about silver cables, etc - that was a trend which came along after the basic technology was fully developed. Early 1980s, if I remember correctly. While I also agree that swapping out op-amps for a (circa 1980) NE5532 doesn't make a lot of sense, some of the newer circuits, particularly from Analog Devices, do have more linear circuitry and wider bandwidth, and are usually what I choose when I need an audio op-amp. Fundamentally I am a discrete-component guy, however. The op-amps have buckets of gain trying to overcome their non-linearity with negative feedback. Discrete component audio design generally relies on fundamentally linear circuitry, with only a small amount of negative feedback.

@Vitaliy, I sure hope I have not tricked anyone; I do my own research, so I can stand behind my results.

I will simply point out a few flaws in the opamp tests, and you can reach your own conclusions. Maybe I'm wrong and those test were perfectly done, certainly a possibility.

Scientific methodology is not a hard and fast fact. There are different ways to do tests. This is just my opinion.

First of all, they cherry picked the tests. That is, they did not test a number of crucial items like zero cross distortion. Why didn't they test that? I really don't know, but I suspect, and again, this is just conjecture, that they didn't build the experiment from scratch but instead used off-the-shelf software, which of course has no controls built in. However, I didn't see why they tested some things and not others. Do I care? No. I only care what it sounds like.

Second, there are no controls. Where are the bias controls? No controls, no science. It's a simple as that. I didn't invent the scientific method. Where are the standard deviation figures?

Lastly, they did find that the more "expensive" chips, and I put that in quotes because $3 is not expensive, are less noisy. Now, you can argue that less noisy is not important, but I personally would probably go for the less noisy one than the more noisy one, especially if my own tests show that the less noisy one sounds better. But suppose the noisy one sounded better? Well, then you have to choose, and in this case that 3dB isn't going to make that much difference anyway.

But here's the kicker--I buy the one that sounds the best. Well, I buy all of them and sell the ones I don't like. So my Sennheiser MKH 40 has a self dB noise of 10dB. The Neumann TLM 103 I tested has a self noise of of 7, 3dB better. I use the Sennheiser as my go-to mic. Now maybe the reason I didn't like the sound of the Neumann (which is certainly an OK mic, BTW) was the IC. The IC is the cheap chip they put in to cut down on the labor costs. Who knows? Well, I know. I took the capsule out of the microphone, and I had a discrete preamp built for it, and tested it again. Sounded much better (though not as good as the Sennheiser). And, it was also slightly noisier. That noise reduction came at a price, and the price was the sound quality. Just like in cameras. However, in the opamp, the noise reduction did not come at a price. It was a better design. You can take my word for it, or you can try it yourself. It is an expensive experiment, I can tell you that. So why would I do that? Because I had to know, is it the chip, or is it the capsule? Turns out it is the chip, in this case. Maybe I'm wrong, but it is my test, my reality. I can live with that.

So basically, I simply would not buy a mic, a chip, an opamp etc based on the test. That's not tricking anyone; that's the way I work. And I do not know a single audio engineer who would either. I've met people who swear by specs, but I'm not one of them. Similarly there is no way I would by a solid silver speaker cable. And yet, people do, and they swear they sound better. Are they delusional? Or do they hear better than me. I have no idea.

Now it could be that you are right and it is unfair to compare it to tests of cameras. That seems like a reasonable analogy to me, but I'm not an expert on camera sensors. So I can't say. I can't say if DxO methodology is good or not.

But I wouldn't buy a camera based on the specs, even though I'm not an expert. I still have to live with whatever knowledge I have and work with it--that's all I have. And there are plenty of people on this site who know more than me.

PS Re the above video--Poppy is one of smartest people on the planet.

High-end audio was nearly non-existent in the 1970s short of a few companies like McIntosh and even those marketed their gear around objective criteria. If anyone tried to sell a $2000 power cord in 1980 they would have been laughed out of business but now there are dozens of companies selling them. The wine industry has followed a very similar path. In 1980 people tended to buy wine they thought tasted good. But, courtesy of decades of clever marketing, wine has been made out to be much more complex and important. An entire industry in the USA has grown up around wineries, wine tourism, wine bars, tastings, pairings, flights, expert reviews, etc. Consumers have been taught to doubt their own ability to judge wine and instead trust expert wine ratings with their point scores and elaborate tasting notes. And, not surprisingly, the experts tend to rate more expensive wines much higher. It’s scary how similar it is to high-end audio. Compare “the Warped Vine Cabernet was leathery with hints of green pollen” to “the UberDAC had a congested sense of pace and tarnished rhythms”. Check out the books by Robin Goldstein, Mike Veseth, Jonathan Nossiter, and the documentary Mondovino.

http://nwavguy.blogspot.ru/2012/04/what-we-hear.html

@thepalalias

We have topic about HP amps. :-)

However, I sold my GH1 and purchased a GH2. So basically, what this shows is that the science is only good up to a point

I think main point that it shows that you tricked people :-) First going from op-amp measurements to cameras, after this telling that (tsss! secret!) that normal audio tests does not exist with basis being that you know some bad ones (and it is just logic flaw). And concluding that your camera purchase is not based on DXOmark.

Thing is - all your points are present in this article. Op-amps play their specific role and only way to do it is proper design and measurement of result (that every manufacturer is doing). As for more costly chips and difference - article on the link has perfect explanation for all this.

I'm happy to test audio scientifically; I just wouldn't buy audio gear based on the scientific test. And I understand that cameras and audio gear are different, but I think it is illogical to buy audio gear based on science and video gear based on "look"

Having said that, there really is very little scientific testing of audio gear. What we have instead is people using widely available software packages and just plugging their gear into it. This is absolutely not science, this is "fun with gear". So there is nothing wrong with this, and a "straight wire" approach, where you loop the audio signal back and measure the noise, is interesting. But it isn't Scientific.

In order for to follow the scientific method, you need to set controls, set baselines, and run full tests that are repeatable. We don't see this.

In fact, one "science" website tested the FR2LE and published completely different results from two different tests.

There are no baselines or control data, and a lot of the tests are missing data, like zero cross distortion. If you are missing tests, then there is no complete picture.

In the case of op omps, I can't see using a cheap part that that tests worse and sounds worse using the non scientific method used in a lot of these tests, if it is a $3 part. However, that's a personal choice. I tested them myself, and I heard a difference in the sound. That's good enough for me.

There was a famous post on Headfi where one guy said the high end gear didn't sound better than his computer audio output. He got a lot of flac for it. He was railed upon. But the fact is, I have one or two computers that have really good, super good audio jacks. I also have some that are terrible.

The GH1 tests better, scientifically, than the GH2. In every category. http://www.dxomark.com/Cameras/Compare/Side-by-side/Panasonic-Lumix-DMC-GH3-versus-Panasonic-Lumix-DMC-GH2-versus-Panasonic-Lumix-DMC-GH1___842_677_630

However, I sold my GH1 and purchased a GH2. So basically, what this shows is that the science is only good up to a point.

Vitaliy, I test all the equipment I buy before I use it. Here is a distortion spectrum I took when I bought a Zoom H2 back in 2007. It is not bad, -51dB maximum distortion at an input level of -6dB.

I test microphones for overload using my sound level meter. Everything has to work well. And that is the problem. "Well" can only be defined by personal experience.